run 2.5.1 test without waiting for 1st e2e

.axolotl-complete.bash

@@ -1,41 +0,0 @@
-#!/bin/bash
-
-_axolotl_completions() {
-    local cur prev
-    COMPREPLY=()
-    cur="${COMP_WORDS[COMP_CWORD]}"
-    prev="${COMP_WORDS[COMP_CWORD-1]}"
-
-    # If we're completing the first argument (the command)
-    if [[ $COMP_CWORD -eq 1 ]]; then
-        mapfile -t COMPREPLY < <(compgen -W "delinearize-llama4 fetch lm-eval merge-sharded-fsdp-weights quantize vllm-serve evaluate inference merge-lora preprocess train" -- "$cur")
-        return 0
-    fi
-
-    # Commands that should complete with directories and YAML files
-    local -a yaml_commands=("merge-sharded-fsdp-weights" "quantize" "vllm-serve" "evaluate" "inference" "merge-lora" "preprocess" "train")
-
-    # Check if previous word is in our list
-    if [[ " ${yaml_commands[*]} " =~ (^|[[:space:]])$prev($|[[:space:]]) ]]; then
-        # Use filename completion which handles directories properly
-        compopt -o filenames
-        mapfile -t COMPREPLY < <(compgen -f -- "$cur")
-
-        # Filter to only include directories and YAML files
-        local -a filtered=()
-        for item in "${COMPREPLY[@]}"; do
-            if [[ -d "$item" ]] || [[ "$item" == *.yaml ]] || [[ "$item" == *.yml ]]; then
-                filtered+=("$item")
-            fi
-        done
-        COMPREPLY=("${filtered[@]}")
-
-        return 0
-    fi
-
-    # Default: no completion
-    return 0
-}
-
-# Remove the -o nospace option - let filenames handle it
-complete -F _axolotl_completions axolotl

.bandit

@@ -1,3 +1,3 @@
 [bandit]
 exclude = tests
-skips = B101,B615,B102,B110
+skips = B101

.coderabbit.yaml

@@ -1,17 +0,0 @@
-# yaml-language-server: $schema=https://coderabbit.ai/integrations/schema.v2.json
-language: "en-US"
-early_access: false
-reviews:
-  profile: "chill"
-  request_changes_workflow: false
-  high_level_summary: true
-  review_status: true
-  collapse_walkthrough: true
-  poem: false
-  sequence_diagrams: false
-  auto_review:
-    enabled: true
-    drafts: false
-    auto_incremental_review: false
-chat:
-  auto_reply: true

.coveragerc

@@ -1,14 +0,0 @@
-[run]
-source = axolotl
-omit =
-    */tests/*
-    setup.py
-
-[report]
-exclude_lines =
-    pragma: no cover
-    def __repr__
-    raise NotImplementedError
-    if __name__ == .__main__.:
-    pass
-    raise ImportError

.flake8

@@ -0,0 +1,5 @@
+[flake8]
+max-line-length = 88
+
+select = C,E,F,W,B,B950
+extend-ignore = E203, E501, W503

.github/CONTRIBUTING.md

@@ -15,7 +15,7 @@ First of all, thank you for your interest in contributing to axolotl! We appreci
   - [Commit Messages](#commit-messages)
 - [Additional Resources](#additional-resources)
 
-## Code of Conduct
+## Code of Conductcode
 
 All contributors are expected to adhere to our [Code of Conduct](CODE_OF_CONDUCT.md). Please read it before participating in the axolotl community.
 
@@ -31,11 +31,7 @@ PRs are **greatly welcome**!
 
 Please run below to setup env
 ```bash
-# Install axolotl + dev and test dependencies
-export UV_TORCH_BACKEND=cu128  # or cu130
-uv venv --no-project --relocatable
-source .venv/bin/activate
-uv pip install --no-build-isolation -e '.[deepspeed]' --group dev --group test
+pip3 install -r requirements-dev.txt -r requirements-tests.txt
 pre-commit install
 
 # test
@@ -61,23 +57,11 @@ We welcome ideas for improvements and new features. To suggest an enhancement, o
 5. Push your branch to your fork on GitHub.
 6. Open a new pull request against the `main` branch of the axolotl repository. Include a clear and concise description of your changes, referencing any related issues.
 
-#### Skipping CI Checks
-
-You can skip certain CI checks by including specific keywords in your commit messages:
-
-- `[skip ci]` or `skip ci` - Skips all CI checks for that commit
-- `[skip-e2e]` or `skip-e2e` - Skips only end-to-end tests while running other CI checks. You may also include this in the title of your PR to disable end-to-end tests for the entire PR.
-
 ## Style Guidelines
 
 ### Code Style
 
-axolotl uses [Ruff](https://docs.astral.sh/ruff/) as its code style guide. Please ensure that your code follows these guidelines.
-
-Use the pre-commit linter to ensure that your code is formatted consistently.
-```bash
-pre-commit run --all-files
-```
+axolotl uses [{codestyle}]({URLofCodestyle}) as its code style guide. Please ensure that your code follows these guidelines.
 
 ### Commit Messages
 
@@ -87,6 +71,6 @@ Write clear and concise commit messages that briefly describe the changes made i
 
 - [GitHub Help](https://help.github.com/)
 - [GitHub Pull Request Documentation](https://docs.github.com/en/github/collaborating-with-issues-and-pull-requests)
-- [Ruff](https://docs.astral.sh/ruff/)
+- [{codestyle}]({URLofCodestyle})
 
 Thank you once again for your interest in contributing to axolotl. We look forward to collaborating with you and creating an even better project together!

.github/FUNDING.yml

@@ -1,13 +1,13 @@
 # These are supported funding model platforms
 
-github: # Replace with up to 4 GitHub Sponsors-enabled usernames e.g., [user1, user2]
+github: [winglian, OpenAccess-AI-Collective] # Replace with up to 4 GitHub Sponsors-enabled usernames e.g., [user1, user2]
 patreon: # Replace with a single Patreon username
 open_collective: # Replace with a single Open Collective username
-ko_fi: # Replace with a single Ko-fi username
+ko_fi: axolotl_ai # Replace with a single Ko-fi username
 tidelift: # Replace with a single Tidelift platform-name/package-name e.g., npm/babel
 community_bridge: # Replace with a single Community Bridge project-name e.g., cloud-foundry
 liberapay: # Replace with a single Liberapay username
 issuehunt: # Replace with a single IssueHunt username
 otechie: # Replace with a single Otechie username
 lfx_crowdfunding: # Replace with a single LFX Crowdfunding project-name e.g., cloud-foundry
-custom: # Replace with up to 4 custom sponsorship URLs e.g., ['link1', 'link2']
+custom: ['https://quickchart.io/qr?text=bitcoin%3Abc1qxlgwlqwfea5s2cxm42xqsfmwjct0rj8w8ea5np&size=480&centerImageUrl=https%3A%2F%2Fupload.wikimedia.org%2Fwikipedia%2Fcommons%2Fthumb%2F4%2F46%2FBitcoin.svg%2F64px-Bitcoin.svg.png'] # Replace with up to 4 custom sponsorship URLs e.g., ['link1', 'link2']

.github/PULL_REQUEST_TEMPLATE.md

@@ -15,11 +15,6 @@
 <!--- Include details of your testing environment, tests ran to see how -->
 <!--- your change affects other areas of the code, etc. -->
 
-## AI Usage Disclaimer
-
-<!--- Was AI (e.g., ChatGPT, Claude, Copilot) used to generate or assist with this PR? -->
-<!--- Please indicate: No / Yes (specify which tool and to what extent) -->
-
 ## Screenshots (if appropriate)
 
 ## Types of changes

.github/workflows/base.yml

@@ -5,102 +5,55 @@ on:
     branches:
       - "main"
     paths:
-      - 'docker/Dockerfile-base'
-      - 'docker/Dockerfile-uv-base'
+      - 'Dockerfile-base'
       - '.github/workflows/base.yml'
   pull_request:
     paths:
-      - 'docker/Dockerfile-base'
-      - 'docker/Dockerfile-uv-base'
+      - 'Dockerfile-base'
       - '.github/workflows/base.yml'
   workflow_dispatch:
 
-permissions:
-  contents: read
-
 jobs:
   build-base:
-    if: ${{ github.repository_owner == 'axolotl-ai-cloud' && (github.event_name != 'pull_request' || !github.event.pull_request.draft) }}
-    timeout-minutes: 480
+    if: github.repository_owner == 'axolotl-ai-cloud'
     # this job needs to be run on self-hosted GPU runners...
-    runs-on: ubuntu-latest-m
-    env:
-      HAS_DOCKERHUB_CREDS: ${{ secrets.DOCKERHUB_USERNAME != '' && secrets.DOCKERHUB_TOKEN != '' }}
+    runs-on: axolotl-gpu-runner
     strategy:
       fail-fast: false
       matrix:
         include:
-          - cuda: "128"
-            cuda_version: 12.8.1
-            cudnn_version: ""
-            python_version: "3.11"
-            pytorch: 2.9.1
-            torch_cuda_arch_list: "7.0 7.5 8.0 8.6 8.7 8.9 9.0+PTX"
-            dockerfile: "Dockerfile-base"
-            platforms: "linux/amd64,linux/arm64"
-          - cuda: "128"
-            cuda_version: 12.8.1
-            cudnn_version: ""
-            python_version: "3.11"
-            pytorch: 2.10.0
-            torch_cuda_arch_list: "7.0 7.5 8.0 8.6 8.7 8.9 9.0+PTX"
-            dockerfile: "Dockerfile-base"
-            platforms: "linux/amd64,linux/arm64"
-          - cuda: "128"
-            cuda_version: 12.8.1
-            cudnn_version: ""
-            python_version: "3.12"
-            pytorch: 2.10.0
-            torch_cuda_arch_list: "7.0 7.5 8.0 8.6 8.7 8.9 9.0+PTX"
-            dockerfile: "Dockerfile-base"
-            platforms: "linux/amd64,linux/arm64"
-#          - cuda: "129"
-#            cuda_version: 12.9.1
-#            cudnn_version: ""
-#            python_version: "3.12"
-#            pytorch: 2.9.1
+#          - cuda: "121"
+#            cuda_version: 12.1.1
+#            cudnn_version: 8
+#            python_version: "3.10"
+#            pytorch: 2.3.1
 #            torch_cuda_arch_list: "7.0 7.5 8.0 8.6 8.7 8.9 9.0+PTX"
-#            dockerfile: "Dockerfile-base"
-#            platforms: "linux/amd64,linux/arm64"
-          - cuda: "130"
-            cuda_version: 13.0.0
-            cudnn_version: ""
-            python_version: "3.11"
-            pytorch: 2.9.1
-            torch_cuda_arch_list: "9.0+PTX"
-            dockerfile: "Dockerfile-base"
-            platforms: "linux/amd64,linux/arm64"
-          - cuda: "130"
-            cuda_version: 13.0.0
-            cudnn_version: ""
-            python_version: "3.12"
-            pytorch: 2.9.1
-            torch_cuda_arch_list: "9.0+PTX"
-            dockerfile: "Dockerfile-base"
-            platforms: "linux/amd64,linux/arm64"
-          - cuda: "130"
-            cuda_version: 13.0.0
-            cudnn_version: ""
-            python_version: "3.12"
-            pytorch: 2.10.0
-            torch_cuda_arch_list: "9.0+PTX"
-            dockerfile: "Dockerfile-base"
-            platforms: "linux/amd64,linux/arm64"
-#          - cuda: "128"
-#            cuda_version: 12.8.1
+#            from_base_img: ""
+#            from_base_tag: ""
+#          - cuda: "121"
+#            cuda_version: 12.1.1
+#            cudnn_version: 8
+#            python_version: "3.11"
+#            pytorch: 2.3.1
+#            torch_cuda_arch_list: "7.0 7.5 8.0 8.6 8.7 8.9 9.0+PTX"
+#            from_base_img: ""
+#            from_base_tag: ""
+#          - cuda: "124"
+#            cuda_version: 12.4.1
 #            cudnn_version: ""
 #            python_version: "3.11"
-#            pytorch: nightly
+#            pytorch: 2.4.1
 #            torch_cuda_arch_list: "7.0 7.5 8.0 8.6 8.7 8.9 9.0+PTX"
-#            dockerfile: "Dockerfile-base-nightly"
-#          # "next" is for release candidates of pytorch
-#          - cuda: "128"
-#            cuda_version: 12.8.1
-#            cudnn_version: ""
-#            python_version: "3.11"
-#            pytorch: next
-#            torch_cuda_arch_list: "7.0 7.5 8.0 8.6 8.7 8.9 9.0+PTX"
-#            dockerfile: "Dockerfile-base-next"
+#            from_base_img: ""
+#            from_base_tag: ""
+          - cuda: "124"
+            cuda_version: 12.4.1
+            cudnn_version: ""
+            python_version: "3.11"
+            pytorch: 2.5.1
+            torch_cuda_arch_list: "7.0 7.5 8.0 8.6 8.7 8.9 9.0+PTX"
+            from_base_img: nvcr.io/nvidia/pytorch
+            from_base_tag: 24.10-py3
     steps:
       - name: Checkout
         uses: actions/checkout@v4
@@ -109,129 +62,22 @@ jobs:
         uses: docker/metadata-action@v5
         with:
           images: |
-            axolotlai/axolotl-base
+            winglian/axolotl-base
+#            axolotlai/axolotl-base
       - name: Login to Docker Hub
-        uses: docker/login-action@v3
-        if: ${{ github.event_name != 'pull_request' && env.HAS_DOCKERHUB_CREDS == 'true' }}
+        uses: docker/login-action@v2
         with:
           username: ${{ secrets.DOCKERHUB_USERNAME }}
           password: ${{ secrets.DOCKERHUB_TOKEN }}
       - name: Set up Docker Buildx
         uses: docker/setup-buildx-action@v3
       - name: Build
-        uses: docker/build-push-action@v5
+        uses: docker/build-push-action@v4
         with:
           context: .
-          file: ./docker/${{ matrix.dockerfile }}
-          platforms: ${{ matrix.platforms }}
-          push: ${{ github.event_name != 'pull_request' }}
-          tags: ${{ steps.metadata.outputs.tags }}-base-py${{ matrix.python_version }}-cu${{ matrix.cuda }}-${{ matrix.pytorch }}${{ matrix.axolotl_extras != '' && '-' || '' }}${{ matrix.axolotl_extras }}
-          labels: ${{ steps.metadata.outputs.labels }}
-          build-args: |
-            CUDA_VERSION=${{ matrix.cuda_version }}
-            CUDNN_VERSION=${{ matrix.cudnn_version }}
-            CUDA=${{ matrix.cuda }}
-            PYTHON_VERSION=${{ matrix.python_version }}
-            PYTORCH_VERSION=${{ matrix.pytorch }}
-            TORCH_CUDA_ARCH_LIST=${{ matrix.torch_cuda_arch_list }}
-  build-base-uv:
-    if: ${{ github.repository_owner == 'axolotl-ai-cloud' && (github.event_name != 'pull_request' || !github.event.pull_request.draft) }}
-    timeout-minutes: 480
-    runs-on: ubuntu-latest-m
-    env:
-      HAS_DOCKERHUB_CREDS: ${{ secrets.DOCKERHUB_USERNAME != '' && secrets.DOCKERHUB_TOKEN != '' }}
-    strategy:
-      fail-fast: false
-      matrix:
-        include:
-          - cuda: "128"
-            cuda_version: 12.8.1
-            cudnn_version: ""
-            python_version: "3.11"
-            pytorch: 2.9.1
-            torch_cuda_arch_list: "7.0 7.5 8.0 8.6 8.7 8.9 9.0+PTX"
-            dockerfile: "Dockerfile-uv-base"
-            platforms: "linux/amd64,linux/arm64"
-          - cuda: "128"
-            cuda_version: 12.8.1
-            cudnn_version: ""
-            python_version: "3.12"
-            pytorch: 2.9.1
-            torch_cuda_arch_list: "7.0 7.5 8.0 8.6 8.7 8.9 9.0+PTX"
-            dockerfile: "Dockerfile-uv-base"
-            platforms: "linux/amd64,linux/arm64"
-          - cuda: "128"
-            cuda_version: 12.8.1
-            cudnn_version: ""
-            python_version: "3.11"
-            pytorch: 2.10.0
-            torch_cuda_arch_list: "7.0 7.5 8.0 8.6 8.7 8.9 9.0+PTX"
-            dockerfile: "Dockerfile-uv-base"
-            platforms: "linux/amd64,linux/arm64"
-          - cuda: "128"
-            cuda_version: 12.8.1
-            cudnn_version: ""
-            python_version: "3.12"
-            pytorch: 2.10.0
-            torch_cuda_arch_list: "7.0 7.5 8.0 8.6 8.7 8.9 9.0+PTX"
-            dockerfile: "Dockerfile-uv-base"
-            platforms: "linux/amd64,linux/arm64"
-#          - cuda: "129"
-#            cuda_version: 12.9.1
-#            cudnn_version: ""
-#            python_version: "3.12"
-#            pytorch: 2.9.1
-#            torch_cuda_arch_list: "7.0 7.5 8.0 8.6 8.7 8.9 9.0+PTX"
-#            dockerfile: "Dockerfile-uv-base"
-#            platforms: "linux/amd64,linux/arm64"
-          - cuda: "130"
-            cuda_version: 13.0.0
-            cudnn_version: ""
-            python_version: "3.11"
-            pytorch: 2.9.1
-            torch_cuda_arch_list: "9.0+PTX"
-            dockerfile: "Dockerfile-uv-base"
-            platforms: "linux/amd64,linux/arm64"
-          - cuda: "130"
-            cuda_version: 13.0.0
-            cudnn_version: ""
-            python_version: "3.12"
-            pytorch: 2.9.1
-            torch_cuda_arch_list: "9.0+PTX"
-            dockerfile: "Dockerfile-uv-base"
-            platforms: "linux/amd64,linux/arm64"
-          - cuda: "130"
-            cuda_version: 13.0.0
-            cudnn_version: ""
-            python_version: "3.12"
-            pytorch: 2.10.0
-            torch_cuda_arch_list: "9.0+PTX"
-            dockerfile: "Dockerfile-uv-base"
-            platforms: "linux/amd64,linux/arm64"
-    steps:
-      - name: Checkout
-        uses: actions/checkout@v4
-      - name: Docker metadata
-        id: metadata
-        uses: docker/metadata-action@v5
-        with:
-          images: |
-            axolotlai/axolotl-base-uv
-      - name: Login to Docker Hub
-        uses: docker/login-action@v3
-        if: ${{ github.event_name != 'pull_request' && env.HAS_DOCKERHUB_CREDS == 'true' }}
-        with:
-          username: ${{ secrets.DOCKERHUB_USERNAME }}
-          password: ${{ secrets.DOCKERHUB_TOKEN }}
-      - name: Set up Docker Buildx
-        uses: docker/setup-buildx-action@v3
-      - name: Build
-        uses: docker/build-push-action@v5
-        with:
-          context: .
-          file: ./docker/${{ matrix.dockerfile }}
-          platforms: ${{ matrix.platforms }}
-          push: ${{ github.event_name != 'pull_request' }}
+          file: ./docker/Dockerfile-base
+          push: true
+#          push: ${{ github.event_name != 'pull_request' }}
           tags: ${{ steps.metadata.outputs.tags }}-base-py${{ matrix.python_version }}-cu${{ matrix.cuda }}-${{ matrix.pytorch }}${{ matrix.axolotl_extras != '' && '-' || '' }}${{ matrix.axolotl_extras }}
           labels: ${{ steps.metadata.outputs.labels }}
           build-args: |
@@ -241,3 +87,5 @@ jobs:
             PYTHON_VERSION=${{ matrix.python_version }}
             PYTORCH_VERSION=${{ matrix.pytorch }}
             TORCH_CUDA_ARCH_LIST=${{ matrix.torch_cuda_arch_list }}
+            BASE_IMAGE=${{ matrix.from_base_img || '' }}
+            BASE_TAG=${{ matrix.from_base_tag || '' }}

.github/workflows/docs.yml

@@ -12,9 +12,6 @@ jobs:
     build-deploy:
         runs-on: ubuntu-latest
         steps:
-        - name: cleanup node
-          run: |
-            sudo rm -rf /usr/share/dotnet /usr/local/lib/android /opt/ghc /opt/hostedtoolcache/CodeQL
         - name: Check out repository
           uses: actions/checkout@v4
         - name: Set up Quarto
@@ -22,13 +19,10 @@ jobs:
         - name: Setup Python
           uses: actions/setup-python@v5
           with:
-            python-version: '3.11'
-        - name: Install dependencies
+            python-version: '3.10'
+        - name: install dependencies
           run: |
-            python3 -m pip install jupyter quartodoc
-            python3 -m pip install -e .
-        - name: Build autodoc
-          run: quartodoc build
+            python3 -m pip install jupyter
         - name: Publish to GitHub Pages (and render)
           uses: quarto-dev/quarto-actions/publish@v2
           with:

.github/workflows/lint.yml

@@ -1,30 +1,23 @@
 name: lint
 on:
   # check on PRs, and manual triggers
-  merge_group:
   pull_request:
-      types: [opened, synchronize, reopened, ready_for_review]
       paths:
        - '**.py'
-       - 'pyproject.toml'
+       - 'requirements.txt'
        - '.github/workflows/*.yml'
        - "*.[q]md"
        - "examples/**/*.y[a]?ml"
-       - ".pre-commit-config.yaml"
   workflow_dispatch:
 
-permissions:
-  contents: read
-
 jobs:
   pre-commit:
     name: pre-commit
     runs-on: ubuntu-latest
-    if: ${{ !github.event.pull_request.draft }}
     steps:
       - uses: actions/checkout@v4
       - uses: actions/setup-python@v5
         with:
-          python-version: "3.11"
+          python-version: "3.10"
           cache: 'pip' # caching pip dependencies
       - uses: pre-commit/action@v3.0.1

.github/workflows/main.yml

@@ -8,9 +8,6 @@ on:
       - "v*"
   workflow_dispatch:
 
-permissions:
-  contents: read
-
 jobs:
   build-axolotl:
     if: ${{ ! contains(github.event.commits[0].message, '[skip docker]') && github.repository_owner == 'axolotl-ai-cloud' }}
@@ -18,37 +15,27 @@ jobs:
       fail-fast: false
       matrix:
         include:
-          - cuda: 128
-            cuda_version: 12.8.1
+          - cuda: 121
+            cuda_version: 12.1.1
+            python_version: "3.10"
+            pytorch: 2.3.1
+            axolotl_extras: mamba-ssm
+          - cuda: 121
+            cuda_version: 12.1.1
             python_version: "3.11"
-            pytorch: 2.9.1
-            axolotl_extras:
-            platforms: "linux/amd64,linux/arm64"
+            pytorch: 2.3.1
+            axolotl_extras: mamba-ssm
             is_latest: true
-          - cuda: 128
-            cuda_version: 12.8.1
-            python_version: "3.12"
-            pytorch: 2.10.0
-            axolotl_extras:
-            platforms: "linux/amd64,linux/arm64"
-#          - cuda: 129
-#            cuda_version: 12.9.1
-#            python_version: "3.12"
-#            pytorch: 2.9.1
-#            axolotl_extras:
-#            platforms: "linux/amd64,linux/arm64"
-          - cuda: 130
-            cuda_version: 13.0.0
+          - cuda: 124
+            cuda_version: 12.4.1
             python_version: "3.11"
-            pytorch: 2.9.1
+            pytorch: 2.4.1
             axolotl_extras:
-            platforms: "linux/amd64,linux/arm64"
-          - cuda: 130
-            cuda_version: 13.0.0
-            python_version: "3.12"
-            pytorch: 2.10.0
+          - cuda: 124
+            cuda_version: 12.4.1
+            python_version: "3.11"
+            pytorch: 2.5.1
             axolotl_extras:
-            platforms: "linux/amd64,linux/arm64"
     runs-on: axolotl-gpu-runner
     steps:
       - name: Checkout
@@ -58,6 +45,7 @@ jobs:
         uses: docker/metadata-action@v5
         with:
           images: |
+            winglian/axolotl
             axolotlai/axolotl
           tags: |
             type=ref,event=branch
@@ -74,13 +62,11 @@ jobs:
         uses: docker/build-push-action@v5
         with:
           context: .
-          platforms: ${{ matrix.platforms }}
           build-args: |
             BASE_TAG=${{ github.ref_type == 'tag' && 'main' || github.ref_name }}-base-py${{ matrix.python_version }}-cu${{ matrix.cuda }}-${{ matrix.pytorch }}
             CUDA=${{ matrix.cuda }}
             PYTORCH_VERSION=${{ matrix.pytorch }}
             AXOLOTL_ARGS=${{ matrix.axolotl_args }}
-            AXOLOTL_EXTRAS=${{ matrix.axolotl_extras}}
           file: ./docker/Dockerfile
           push: ${{ github.event_name != 'pull_request' }}
           tags: |
@@ -89,122 +75,34 @@ jobs:
             ${{ (matrix.is_latest) && format('{0}-latest', steps.metadata.outputs.tags) || '' }}
           labels: ${{ steps.metadata.outputs.labels }}
 
-  build-axolotl-uv:
-    if: ${{ ! contains(github.event.commits[0].message, '[skip docker]') && github.repository_owner == 'axolotl-ai-cloud' }}
-    strategy:
-      fail-fast: false
-      matrix:
-        include:
-          - cuda: 128
-            cuda_version: 12.8.1
-            python_version: "3.11"
-            pytorch: 2.9.1
-            axolotl_extras:
-            platforms: "linux/amd64,linux/arm64"
-          - cuda: 128
-            cuda_version: 12.8.1
-            python_version: "3.12"
-            pytorch: 2.9.1
-            axolotl_extras:
-            platforms: "linux/amd64,linux/arm64"
-            is_latest: true
-          - cuda: 128
-            cuda_version: 12.8.1
-            python_version: "3.12"
-            pytorch: 2.10.0
-            axolotl_extras:
-            platforms: "linux/amd64,linux/arm64"
-          - cuda: 130
-            cuda_version: 13.0.0
-            python_version: "3.11"
-            pytorch: 2.9.1
-            axolotl_extras:
-            platforms: "linux/amd64,linux/arm64"
-          - cuda: 130
-            cuda_version: 13.0.0
-            python_version: "3.12"
-            pytorch: 2.10.0
-            axolotl_extras:
-            platforms: "linux/amd64,linux/arm64"
-    runs-on: axolotl-gpu-runner
-    steps:
-      - name: Checkout
-        uses: actions/checkout@v4
-      - name: Docker metadata
-        id: metadata
-        uses: docker/metadata-action@v5
-        with:
-          images: |
-            axolotlai/axolotl-uv
-          tags: |
-            type=ref,event=branch
-            type=pep440,pattern={{version}}
-      - name: Set up Docker Buildx
-        uses: docker/setup-buildx-action@v3
-      - name: Login to Docker Hub
-        uses: docker/login-action@v3
-        with:
-          username: ${{ secrets.DOCKERHUB_USERNAME }}
-          password: ${{ secrets.DOCKERHUB_TOKEN }}
-      # guidance for testing before pushing: https://docs.docker.com/build/ci/github-actions/test-before-push/
-      - name: Build and export to Docker
-        uses: docker/build-push-action@v5
-        with:
-          context: .
-          platforms: ${{ matrix.platforms }}
-          build-args: |
-            BASE_TAG=${{ github.ref_type == 'tag' && 'main' || github.ref_name }}-base-py${{ matrix.python_version }}-cu${{ matrix.cuda }}-${{ matrix.pytorch }}
-            CUDA=${{ matrix.cuda }}
-            PYTORCH_VERSION=${{ matrix.pytorch }}
-            AXOLOTL_ARGS=${{ matrix.axolotl_args }}
-            AXOLOTL_EXTRAS=${{ matrix.axolotl_extras}}
-          file: ./docker/Dockerfile-uv
-          push: ${{ github.event_name != 'pull_request' }}
-          tags: |
-            ${{ steps.metadata.outputs.tags }}-py${{ matrix.python_version }}-cu${{ matrix.cuda }}-${{ matrix.pytorch }}${{ matrix.axolotl_extras != '' && '-' || '' }}${{ matrix.axolotl_extras }}
-            ${{ steps.metadata.outputs.tags }}-py${{ matrix.python_version }}-cu${{ matrix.cuda }}-${{ matrix.pytorch }}
-            ${{ (matrix.is_latest) && format('{0}-latest', steps.metadata.outputs.tags) || '' }}
-          labels: ${{ steps.metadata.outputs.labels }}
-
   build-axolotl-cloud:
     needs: build-axolotl
     if: ${{ ! contains(github.event.commits[0].message, '[skip docker]') && github.repository_owner == 'axolotl-ai-cloud' }}
     # this job needs to be run on self-hosted GPU runners...
     strategy:
-      fail-fast: false
       matrix:
         include:
-          - cuda: 128
-            cuda_version: 12.8.1
+          - cuda: 121
+            cuda_version: 12.1.1
+            python_version: "3.10"
+            pytorch: 2.3.1
+            axolotl_extras:
+          - cuda: 121
+            cuda_version: 12.1.1
             python_version: "3.11"
-            pytorch: 2.9.1
+            pytorch: 2.3.1
             axolotl_extras:
             is_latest: true
-            platforms: "linux/amd64,linux/arm64"
-          - cuda: 128
-            cuda_version: 12.8.1
-            python_version: "3.12"
-            pytorch: 2.10.0
-            axolotl_extras:
-            platforms: "linux/amd64,linux/arm64"
-#          - cuda: 129
-#            cuda_version: 12.9.1
-#            python_version: "3.12"
-#            pytorch: 2.9.1
-#            axolotl_extras:
-#            platforms: "linux/amd64,linux/arm64"
-          - cuda: 130
-            cuda_version: 13.0.0
+          - cuda: 124
+            cuda_version: 12.4.1
             python_version: "3.11"
-            pytorch: 2.9.1
+            pytorch: 2.4.1
             axolotl_extras:
-            platforms: "linux/amd64,linux/arm64"
-          - cuda: 130
-            cuda_version: 13.0.0
-            python_version: "3.12"
-            pytorch: 2.10.0
+          - cuda: 124
+            cuda_version: 12.4.1
+            python_version: "3.11"
+            pytorch: 2.5.1
             axolotl_extras:
-            platforms: "linux/amd64,linux/arm64"
     runs-on: axolotl-gpu-runner
     steps:
       - name: Checkout
@@ -214,6 +112,7 @@ jobs:
         uses: docker/metadata-action@v5
         with:
           images: |
+            winglian/axolotl-cloud
             axolotlai/axolotl-cloud
           tags: |
             type=ref,event=branch
@@ -229,7 +128,6 @@ jobs:
         uses: docker/build-push-action@v5
         with:
           context: .
-          platforms: ${{ matrix.platforms }}
           build-args: |
             BASE_TAG=${{ github.ref_type == 'tag' && 'main' || github.ref_name }}-py${{ matrix.python_version }}-cu${{ matrix.cuda }}-${{ matrix.pytorch }}${{ matrix.axolotl_extras != '' && '-' || '' }}${{ matrix.axolotl_extras }}
             CUDA=${{ matrix.cuda }}
@@ -240,100 +138,18 @@ jobs:
              ${{ (matrix.is_latest) && format('{0}-latest', steps.metadata.outputs.tags) || '' }}
           labels: ${{ steps.metadata.outputs.labels }}
 
-  build-axolotl-cloud-uv:
-    needs: build-axolotl-uv
-    if: ${{ ! contains(github.event.commits[0].message, '[skip docker]') && github.repository_owner == 'axolotl-ai-cloud' }}
-    # this job needs to be run on self-hosted GPU runners...
-    strategy:
-      fail-fast: false
-      matrix:
-        include:
-          - cuda: 128
-            cuda_version: 12.8.1
-            python_version: "3.11"
-            pytorch: 2.9.1
-            axolotl_extras:
-            platforms: "linux/amd64,linux/arm64"
-          - cuda: 128
-            cuda_version: 12.8.1
-            python_version: "3.12"
-            pytorch: 2.9.1
-            axolotl_extras:
-            is_latest: true
-            platforms: "linux/amd64,linux/arm64"
-          - cuda: 128
-            cuda_version: 12.8.1
-            python_version: "3.12"
-            pytorch: 2.10.0
-            axolotl_extras:
-            platforms: "linux/amd64,linux/arm64"
-          - cuda: 130
-            cuda_version: 13.0.0
-            python_version: "3.11"
-            pytorch: 2.9.1
-            axolotl_extras:
-            platforms: "linux/amd64,linux/arm64"
-          - cuda: 130
-            cuda_version: 13.0.0
-            python_version: "3.12"
-            pytorch: 2.10.0
-            axolotl_extras:
-            platforms: "linux/amd64,linux/arm64"
-    runs-on: axolotl-gpu-runner
-    steps:
-      - name: Checkout
-        uses: actions/checkout@v4
-      - name: Docker metadata
-        id: metadata
-        uses: docker/metadata-action@v5
-        with:
-          images: |
-            axolotlai/axolotl-cloud-uv
-          tags: |
-            type=ref,event=branch
-            type=pep440,pattern={{version}}
-      - name: Login to Docker Hub
-        uses: docker/login-action@v3
-        with:
-          username: ${{ secrets.DOCKERHUB_USERNAME }}
-          password: ${{ secrets.DOCKERHUB_TOKEN }}
-      - name: Set up Docker Buildx
-        uses: docker/setup-buildx-action@v3
-      - name: Build
-        uses: docker/build-push-action@v5
-        with:
-          context: .
-          platforms: ${{ matrix.platforms }}
-          build-args: |
-            BASE_TAG=${{ github.ref_type == 'tag' && 'main' || github.ref_name }}-py${{ matrix.python_version }}-cu${{ matrix.cuda }}-${{ matrix.pytorch }}${{ matrix.axolotl_extras != '' && '-' || '' }}${{ matrix.axolotl_extras }}
-            CUDA=${{ matrix.cuda }}
-          file: ./docker/Dockerfile-cloud-uv
-          push: ${{ github.event_name != 'pull_request' }}
-          tags: |
-             ${{ steps.metadata.outputs.tags }}-py${{ matrix.python_version }}-cu${{ matrix.cuda }}-${{ matrix.pytorch }}${{ matrix.axolotl_extras != '' && '-' || '' }}${{ matrix.axolotl_extras }}
-             ${{ (matrix.is_latest) && format('{0}-latest', steps.metadata.outputs.tags) || '' }}
-          labels: ${{ steps.metadata.outputs.labels }}
-
   build-axolotl-cloud-no-tmux:
     needs: build-axolotl
     if: ${{ ! contains(github.event.commits[0].message, '[skip docker]') && github.repository_owner == 'axolotl-ai-cloud' }}
     # this job needs to be run on self-hosted GPU runners...
     strategy:
-      fail-fast: false
       matrix:
         include:
-          - cuda: 128
-            cuda_version: 12.8.1
+          - cuda: 121
+            cuda_version: 12.1.1
             python_version: "3.11"
-            pytorch: 2.9.1
+            pytorch: 2.3.1
             axolotl_extras:
-            is_latest: true
-          - cuda: 130
-            cuda_version: 13.0.0
-            python_version: "3.11"
-            pytorch: 2.9.1
-            axolotl_extras:
-            is_latest:
     runs-on: axolotl-gpu-runner
     steps:
       - name: Checkout
@@ -343,6 +159,7 @@ jobs:
         uses: docker/metadata-action@v5
         with:
           images: |
+            winglian/axolotl-cloud-term
             axolotlai/axolotl-cloud-term
           tags: |
             type=ref,event=branch
@@ -358,7 +175,6 @@ jobs:
         uses: docker/build-push-action@v5
         with:
           context: .
-          platforms: linux/amd64,linux/arm64
           build-args: |
             BASE_TAG=${{ github.ref_type == 'tag' && 'main' || github.ref_name }}-py${{ matrix.python_version }}-cu${{ matrix.cuda }}-${{ matrix.pytorch }}${{ matrix.axolotl_extras != '' && '-' || '' }}${{ matrix.axolotl_extras }}
             CUDA=${{ matrix.cuda }}

.github/workflows/multi-gpu-e2e.yml

@@ -3,54 +3,43 @@ name: docker-multigpu-tests-biweekly
 on:
   pull_request:
     paths:
-      - "tests/e2e/multigpu/**.py"
-      - "pyproject.toml"
-      - ".github/workflows/multi-gpu-e2e.yml"
-      - "scripts/cutcrossentropy_install.py"
-      - "src/axolotl/core/trainers/mixins/sequence_parallel.py"
-      - "src/axolotl/utils/distributed.py"
+      - 'tests/e2e/multigpu/*.py'
   workflow_dispatch:
   schedule:
-    - cron: "0 0 * * 1,4" # Runs at 00:00 UTC every monday & thursday
+    - cron: '0 0 * * 1,4'  # Runs at 00:00 UTC every monday & thursday
 
 # Cancel jobs on the same ref if a new one is triggered
 concurrency:
   group: ${{ github.workflow }}-${{ github.ref }}
   cancel-in-progress: ${{ github.ref != 'refs/heads/main' }}
 
-permissions:
-  contents: read
-
-env:
-  MODAL_IMAGE_BUILDER_VERSION: "2025.06"
-
 jobs:
   test-axolotl-multigpu:
-    if: ${{ ! contains(github.event.commits[0].message, '[skip e2e]') && github.repository_owner == 'axolotl-ai-cloud' && (github.event_name != 'pull_request' || !github.event.pull_request.draft) }}
+    if: ${{ ! contains(github.event.commits[0].message, '[skip e2e]') && github.repository_owner == 'axolotl-ai-cloud' }}
     strategy:
       fail-fast: false
       matrix:
         include:
-          #          - cuda: 129
-          #            cuda_version: 12.9.1
-          #            python_version: "3.12"
-          #            pytorch: 2.9.1
-          #            axolotl_extras: "fbgemm-gpu"
-          #            num_gpus: 2
-          #            dockerfile: "Dockerfile-uv.jinja"
-          - cuda: 130
-            cuda_version: 13.0.0
+          - cuda: 121
+            cuda_version: 12.1.1
             python_version: "3.11"
-            pytorch: 2.9.1
+            pytorch: 2.3.1
             axolotl_extras:
-            #            axolotl_extras: fbgemm-gpu
             num_gpus: 2
-          - cuda: 128
-            cuda_version: 12.8.1
+          - cuda: 124
+            cuda_version: 12.4.1
             python_version: "3.11"
-            pytorch: 2.10.0
-            axolotl_extras: "fbgemm-gpu"
+            pytorch: 2.4.1
+            axolotl_extras:
             num_gpus: 2
+            nightly_build: "true"
+          - cuda: 124
+            cuda_version: 12.4.1
+            python_version: "3.11"
+            pytorch: 2.5.1
+            axolotl_extras:
+            num_gpus: 2
+            nightly_build: "true"
     runs-on: [self-hosted, modal]
     timeout-minutes: 120
     steps:
@@ -59,11 +48,11 @@ jobs:
       - name: Install Python
         uses: actions/setup-python@v5
         with:
-          python-version: "3.11"
+          python-version: "3.10"
       - name: Install Modal
         run: |
           python -m pip install --upgrade pip
-          pip install modal==1.3.0.post1 jinja2
+          pip install modal==0.63.64 jinja2
       - name: Update env vars
         run: |
           echo "BASE_TAG=main-base-py${{ matrix.python_version }}-cu${{ matrix.cuda }}-${{ matrix.pytorch }}" >> $GITHUB_ENV
@@ -72,9 +61,7 @@ jobs:
           echo "AXOLOTL_EXTRAS=${{ matrix.axolotl_extras}}" >> $GITHUB_ENV
           echo "CUDA=${{ matrix.cuda }}" >> $GITHUB_ENV
           echo "N_GPUS=${{ matrix.num_gpus }}" >> $GITHUB_ENV
-          echo "E2E_DOCKERFILE=${{ matrix.dockerfile || 'Dockerfile-uv.jinja'}}" >> $GITHUB_ENV
+          echo "NIGHTLY_BUILD=${{ matrix.nightly_build }}" >> $GITHUB_ENV
       - name: Run tests job on Modal
-        env:
-          CODECOV_TOKEN: ${{ secrets.CODECOV_TOKEN }}
         run: |
-          modal run -m cicd.multigpu
+          modal run cicd.multigpu

.github/workflows/nightlies.yml

@@ -5,9 +5,6 @@ on:
   schedule:
     - cron: '0 0 * * *'  # Runs at 00:00 UTC every day
 
-permissions:
-  contents: read
-
 jobs:
   build-axolotl:
     if: ${{ ! contains(github.event.commits[0].message, '[skip docker]') && github.repository_owner == 'axolotl-ai-cloud' }}
@@ -15,10 +12,26 @@ jobs:
       fail-fast: false
       matrix:
         include:
-          - cuda: 128
-            cuda_version: 12.8.1
+          - cuda: 121
+            cuda_version: 12.1.1
+            python_version: "3.10"
+            pytorch: 2.3.1
+            axolotl_extras:
+          - cuda: 121
+            cuda_version: 12.1.1
             python_version: "3.11"
-            pytorch: 2.9.1
+            pytorch: 2.3.1
+            axolotl_extras:
+            is_latest: true
+          - cuda: 124
+            cuda_version: 12.4.1
+            python_version: "3.11"
+            pytorch: 2.4.1
+            axolotl_extras:
+          - cuda: 124
+            cuda_version: 12.4.1
+            python_version: "3.11"
+            pytorch: 2.5.1
             axolotl_extras:
     runs-on: axolotl-gpu-runner
     steps:
@@ -29,6 +42,7 @@ jobs:
         uses: docker/metadata-action@v5
         with:
           images: |
+            winglian/axolotl
             axolotlai/axolotl
           tags: |
             type=raw,value={{ branch }}-{{ date 'YYYYMMDD' }}
@@ -62,10 +76,26 @@ jobs:
     strategy:
       matrix:
         include:
-          - cuda: 128
-            cuda_version: 12.8.1
+          - cuda: 121
+            cuda_version: 12.1.1
+            python_version: "3.10"
+            pytorch: 2.3.1
+            axolotl_extras:
+          - cuda: 121
+            cuda_version: 12.1.1
             python_version: "3.11"
-            pytorch: 2.9.1
+            pytorch: 2.3.1
+            axolotl_extras:
+            is_latest: true
+          - cuda: 124
+            cuda_version: 12.4.1
+            python_version: "3.11"
+            pytorch: 2.4.1
+            axolotl_extras:
+          - cuda: 124
+            cuda_version: 12.4.1
+            python_version: "3.11"
+            pytorch: 2.5.1
             axolotl_extras:
     runs-on: axolotl-gpu-runner
     steps:
@@ -76,6 +106,7 @@ jobs:
         uses: docker/metadata-action@v5
         with:
           images: |
+            winglian/axolotl-cloud
             axolotlai/axolotl-cloud
           tags: |
             type=raw,value={{ branch }}-{{ date 'YYYYMMDD' }}

.github/workflows/precommit-autoupdate.yml

@@ -1,42 +0,0 @@
-name: Pre-commit auto-update
-
-on:
-  schedule:
-    - cron: '0 0 1 * *'  # Run monthly
-  workflow_dispatch:  # Manual kickoff
-
-permissions: {}
-
-jobs:
-  auto-update:
-    runs-on: ubuntu-latest
-    permissions:
-      contents: write
-      pull-requests: write
-    steps:
-      - uses: actions/checkout@v4
-
-      - uses: actions/setup-python@v5
-        with:
-          python-version: '3.11'
-
-      - name: Update pre-commit hooks
-        id: update
-        run: |
-          pip install pre-commit
-          pre-commit autoupdate
-          if [[ -n $(git status --porcelain) ]]; then
-            echo "changes=true" >> $GITHUB_OUTPUT
-          fi
-
-      - name: Create Pull Request
-        if: steps.update.outputs.changes == 'true'
-        uses: peter-evans/create-pull-request@v6
-        with:
-          token: ${{ secrets.GITHUB_TOKEN }}
-          branch: update/pre-commit-hooks
-          delete-branch: true
-          title: "chore: update pre-commit hooks"
-          commit-message: "chore: update pre-commit hooks"
-          body: |
-            Automated PR to update pre-commit hooks to their latest versions.

.github/workflows/preview-docs.yml

@@ -1,77 +0,0 @@
-name: Preview
-on:
-  workflow_dispatch:
-  pull_request:
-    types: [opened, synchronize, reopened, ready_for_review]
-
-    # Run the workflow only when one of these files changes
-    paths:
-      - '**/*.md'      # any Markdown file
-      - '**/*.qmd'     # any Quarto file
-      - '_quarto.yml'
-      - docs/scripts/generate_config_docs.py
-      - src/axolotl/utils/schemas/**.py
-      - .github/workflows/preview-docs.yml
-
-permissions:
-  contents: read
-  pull-requests: write
-
-jobs:
-  preview:
-    runs-on: ubuntu-latest
-    if: ${{ !github.event.pull_request.draft }}
-    steps:
-      - name: cleanup node
-        run: |
-          sudo rm -rf /usr/share/dotnet /usr/local/lib/android /opt/ghc /opt/hostedtoolcache/CodeQL
-
-      - name: Check out repository
-        uses: actions/checkout@v4
-        with:
-          ref: ${{ github.event.pull_request.head.sha }}
-
-      - name: Set up Quarto
-        uses: quarto-dev/quarto-actions/setup@v2
-
-      - name: Setup Python
-        uses: actions/setup-python@v5
-        with:
-          python-version: '3.11'
-
-      - name: Install dependencies
-        run: |
-          python3 -m pip install jupyter quartodoc
-          python3 -m pip install -e .
-
-      - name: Build autodoc
-        run: quartodoc build
-
-      - name: Quarto render
-        run: quarto render
-
-      - name: Netlify Publish
-        uses: nwtgck/actions-netlify@v3.0
-        if: ${{ github.event.pull_request.head.repo.full_name == github.repository }}
-        id: netlify
-        with:
-          publish-dir: './_site'
-          enable-pull-request-comment: false
-          enable-github-deployment: false
-          github-token: ${{ secrets.GITHUB_TOKEN }}
-          deploy-message: "Deployed On Netlify"
-          github-deployment-environment: 'preview'
-          github-deployment-description: 'Preview Deployment'
-        env:
-          NETLIFY_AUTH_TOKEN: ${{ secrets.NETLIFY_AUTH_TOKEN }}
-          NETLIFY_SITE_ID: ${{ secrets.NETLIFY_SITE_ID }}
-
-      - name: Update PR with preview link
-        if: ${{ steps.netlify.outcome == 'success' }}
-        uses: marocchino/sticky-pull-request-comment@v2
-        with:
-          GITHUB_TOKEN: ${{ secrets.GITHUB_TOKEN }}
-          message: |
-            📖 **Documentation Preview**: ${{ steps.netlify.outputs.deploy-url }}
-
-            Deployed on Netlify from commit ${{ github.event.pull_request.head.sha }}

.github/workflows/pypi.yml

@@ -3,14 +3,9 @@ name: publish pypi
 on:
   push:
     tags:
-      - "v*"
+      - 'v*'
   workflow_dispatch:
 
-permissions: {}
-
-env:
-  UV_SYSTEM_PYTHON: "1"
-
 jobs:
   setup_release:
     name: Create Release
@@ -18,13 +13,10 @@ jobs:
     permissions:
       contents: write
     steps:
-      - name: Checkout code
-        uses: actions/checkout@v4
-
       - name: Create release
         env:
           GITHUB_TOKEN: ${{ secrets.GITHUB_TOKEN }}
-        run: gh release create "$GITHUB_REF_NAME" --generate-notes
+        run: gh release create "$GITHUB_REF_NAME" # GITHUB_REF_NAME is the tag name in `on.push.tags` workflows
   pypi-publish:
     name: Upload release to PyPI
     runs-on: ubuntu-latest
@@ -33,8 +25,7 @@ jobs:
       name: pypi
       url: https://pypi.org/p/axolotl
     permissions:
-      contents: read
-      id-token: write # IMPORTANT: this permission is mandatory for trusted publishing
+      id-token: write  # IMPORTANT: this permission is mandatory for trusted publishing
     steps:
       - name: Check out repository code
         uses: actions/checkout@v4
@@ -42,25 +33,21 @@ jobs:
       - name: Setup Python
         uses: actions/setup-python@v5
         with:
-          python-version: "3.11"
-
-      - name: Install uv
-        uses: astral-sh/setup-uv@v7
+          python-version: "3.10"
 
       - name: Install dependencies
         run: |
-          uv pip install wheel packaging
-          uv pip install --no-build-isolation -e .
-          uv pip install black mypy pre-commit types-requests quartodoc jupyter blobfile tiktoken \
-            codecov codecov-cli pytest pytest-cov pytest-retry pytest-sugar pytest-xdist tbparse
+          pip3 install wheel packaging
+          pip3 install -e .
+          pip3 install -r requirements-dev.txt -r requirements-tests.txt
 
       - name: Extract tag name
         id: tag
-        run: echo "TAG_NAME=$(echo $GITHUB_REF | cut -d / -f 3)" >> "$GITHUB_OUTPUT"
+        run: echo ::set-output name=TAG_NAME::$(echo $GITHUB_REF | cut -d / -f 3)
 
-      - name: Update version in VERSION file
+      - name: Update version in setup.py
         run: |
-          echo "${{ steps.tag.outputs.TAG_NAME }}" | sed 's/^v//' > VERSION
+          sed -i -E 's/version="([0-9.]+)",/version="${{ steps.tag.outputs.TAG_NAME }}",/g' setup.py
 
       - name: Build a source dist
         run: |

.github/workflows/tests-nightly.yml

@@ -2,17 +2,7 @@ name: Tests Nightly against upstream main
 on:
   workflow_dispatch:
   schedule:
-    - cron: "0 0 * * *" # Runs at 00:00 UTC every day
-  pull_request:
-    types: [opened, synchronize, reopened, ready_for_review]
-    paths:
-      - ".github/workflows/tests-nightly.yml"
-
-permissions:
-  contents: read
-
-env:
-  UV_SYSTEM_PYTHON: "1"
+    - cron: '0 0 * * *'  # Runs at 00:00 UTC every day
 
 jobs:
   pre-commit:
@@ -22,76 +12,58 @@ jobs:
       - uses: actions/checkout@v4
       - uses: actions/setup-python@v5
         with:
-          python-version: "3.11"
-          cache: "pip" # caching pip dependencies
+          python-version: "3.10"
+          cache: 'pip' # caching pip dependencies
       - uses: pre-commit/action@v3.0.1
         env:
           SKIP: no-commit-to-branch
 
-  prime-cdn-s3-cache:
-    name: Prefetch S3 once to prime the CDN cache
-    runs-on: ubuntu-latest
-    if: ${{ !github.event.pull_request.draft }}
-    timeout-minutes: 10
-    steps:
-      - name: Restore Cache from S3
-        id: hf-cache-restore-s3
-        run: |
-          curl -v -H "Range: bytes=0-1023" -L https://axolotl-ci.b-cdn.net/hf-cache.tar.zst > /dev/null
-
   pytest:
     name: PyTest
     runs-on: ubuntu-latest
-    needs: [prime-cdn-s3-cache]
     strategy:
       fail-fast: false
+      max-parallel: 2
       matrix:
-        python_version: ["3.12"] # TODO include py3.14 once https://github.com/mistralai/mistral-common/pull/194 is merged
-        pytorch_version: ["2.9.1", "2.10.0"]
+        python_version: ["3.10", "3.11"]
+        pytorch_version: ["2.3.1", "2.4.1", "2.5.1"]
+        exclude:
+          - python_version: "3.10"
+            pytorch_version: "2.4.1"
+          - python_version: "3.10"
+            pytorch_version: "2.5.1"
     timeout-minutes: 20
 
     steps:
       - name: Check out repository code
         uses: actions/checkout@v4
 
-      - name: Restore Cache from S3
-        id: hf-cache-restore-s3
-        run: |
-          mkdir -p /home/runner/.cache/huggingface/hub
-          curl -L https://axolotl-ci.b-cdn.net/hf-cache.tar.zst | tar -xf - -C /home/runner/.cache/huggingface/hub/  --use-compress-program unzstd
-
       - name: Setup Python
         uses: actions/setup-python@v5
         with:
           python-version: ${{ matrix.python_version }}
-
-      - name: Install uv
-        uses: astral-sh/setup-uv@v7
+          cache: 'pip' # caching pip dependencies
 
       - name: Install PyTorch
         run: |
-          uv pip install torch==${{ matrix.pytorch_version }} torchvision
-          uv pip freeze | grep -E "^(torch|torchvision)==" > /tmp/torch-pin.txt
+          pip3 install torch==${{ matrix.pytorch_version }} --index-url https://download.pytorch.org/whl/cpu
+
+      - name: Update requirements.txt
+        run: |
+          sed -i 's#^transformers.*#transformers @ git+https://github.com/huggingface/transformers.git@main#' requirements.txt
+          sed -i 's#^peft.*#peft @ git+https://github.com/huggingface/peft.git@main#' requirements.txt
+          sed -i 's#^accelerate.*#accelerate @ git+https://github.com/huggingface/accelerate.git@main#' requirements.txt
+          sed -i 's#^trl.*#trl @ git+https://github.com/huggingface/trl.git@main#' requirements.txt
+          sed -i 's#^datasets.*#datasets @ git+https://github.com/huggingface/datasets.git@main#' requirements.txt
 
       - name: Install dependencies
         run: |
-          uv pip install --no-build-isolation -e . --override /tmp/torch-pin.txt
-          python scripts/cutcrossentropy_install.py --uv | sh
-          uv pip install black mypy pre-commit types-requests quartodoc jupyter blobfile tiktoken \
-            codecov codecov-cli pytest pytest-cov pytest-retry pytest-sugar pytest-xdist tbparse
-
-      - name: Override with nightly HF packages
-        run: |
-          uv pip install --no-deps \
-            "transformers @ git+https://github.com/huggingface/transformers.git@main" \
-            "peft @ git+https://github.com/huggingface/peft.git@main" \
-            "accelerate @ git+https://github.com/huggingface/accelerate.git@main" \
-            "trl @ git+https://github.com/huggingface/trl.git@main" \
-            "datasets @ git+https://github.com/huggingface/datasets.git@main"
-
-      - name: Make sure PyTorch version wasn't clobbered
-        run: |
-          python -c "import torch; assert '${{ matrix.pytorch_version }}' in torch.__version__, f'Expected torch ${{ matrix.pytorch_version }} but got {torch.__version__}'"
+          pip3 install --upgrade pip
+          pip3 install --upgrade packaging
+          pip3 install -U -e .
+          python scripts/unsloth_install.py | sh
+          python scripts/cutcrossentropy_install.py | sh
+          pip3 install -r requirements-dev.txt -r requirements-tests.txt
 
       - name: Ensure axolotl CLI was installed
         run: |
@@ -99,39 +71,42 @@ jobs:
 
       - name: Run tests
         run: |
-          pytest -v --durations=10 -n8 --dist loadfile --ignore=tests/e2e/ --ignore=tests/patched/ --ignore=tests/cli/ tests/
-          pytest -v --durations=10 tests/patched/
-          pytest -v --durations=10 tests/cli/
+          pytest -n8 --dist loadfile --ignore=tests/e2e/ --ignore=tests/patched/ tests/
+          pytest tests/patched/
 
+      - name: cleanup pip cache
+        run: |
+          find "$(pip cache dir)/http-v2" -type f -mtime +14 -exec rm {} \;
 
   docker-e2e-tests:
     if: github.repository_owner == 'axolotl-ai-cloud'
     # this job needs to be run on self-hosted GPU runners...
     runs-on: [self-hosted, modal]
-    timeout-minutes: 120
+    timeout-minutes: 60
     needs: [pre-commit, pytest]
 
     strategy:
       fail-fast: false
       matrix:
         include:
-          - cuda: 128
-            cuda_version: 12.8.1
+          - cuda: 121
+            cuda_version: 12.1.1
+            python_version: "3.10"
+            pytorch: 2.3.1
+            num_gpus: 1
+            axolotl_extras: mamba-ssm
+            nightly_build: "true"
+          - cuda: 124
+            cuda_version: 12.4.1
             python_version: "3.11"
-            pytorch: 2.9.1
+            pytorch: 2.4.1
             num_gpus: 1
             axolotl_extras:
             nightly_build: "true"
-          - cuda: 128
-            cuda_version: 12.8.1
+          - cuda: 124
+            cuda_version: 12.4.1
             python_version: "3.11"
-            pytorch: 2.10.0
-            num_gpus: 1
-            axolotl_extras:
-          - cuda: 130
-            cuda_version: 13.0.0
-            python_version: "3.12"
-            pytorch: 2.9.1
+            pytorch: 2.5.1
             num_gpus: 1
             axolotl_extras:
             nightly_build: "true"
@@ -141,55 +116,11 @@ jobs:
       - name: Install Python
         uses: actions/setup-python@v5
         with:
-          python-version: "3.11"
+          python-version: "3.10"
       - name: Install Modal
         run: |
           python -m pip install --upgrade pip
-          pip install modal==1.3.0.post1 jinja2
-      - name: Update env vars
-        run: |
-          echo "BASE_TAG=main-base-py${{ matrix.python_version }}-cu${{ matrix.cuda }}-${{ matrix.pytorch }}" >> $GITHUB_ENV
-          echo "PYTORCH_VERSION=${{ matrix.pytorch}}" >> $GITHUB_ENV
-          echo "AXOLOTL_ARGS=${{ matrix.axolotl_args}}" >> $GITHUB_ENV
-          echo "AXOLOTL_EXTRAS=${{ matrix.axolotl_extras}}" >> $GITHUB_ENV
-          echo "CUDA=${{ matrix.cuda }}" >> $GITHUB_ENV
-          echo "N_GPUS=${{ matrix.num_gpus }}" >> $GITHUB_ENV
-          echo "E2E_DOCKERFILE=${{ matrix.dockerfile || 'Dockerfile-uv.jinja'}}" >> $GITHUB_ENV
-          echo "NIGHTLY_BUILD=${{ matrix.nightly_build }}" >> $GITHUB_ENV
-      - name: Run tests job on Modal
-        env:
-          CODECOV_TOKEN: ${{ secrets.CODECOV_TOKEN }}
-        run: |
-          modal run cicd.e2e_tests
-  docker-e2e-multigpu-tests:
-    if: github.repository_owner == 'axolotl-ai-cloud'
-    # this job needs to be run on self-hosted GPU runners...
-    runs-on: [self-hosted, modal]
-    timeout-minutes: 120
-    needs: [pre-commit, pytest, docker-e2e-tests]
-
-    strategy:
-      fail-fast: false
-      matrix:
-        include:
-          - cuda: 128
-            cuda_version: 12.8.1
-            python_version: "3.11"
-            pytorch: 2.9.1
-            num_gpus: 2
-            axolotl_extras:
-            nightly_build: "true"
-    steps:
-      - name: Checkout
-        uses: actions/checkout@v4
-      - name: Install Python
-        uses: actions/setup-python@v5
-        with:
-          python-version: "3.11"
-      - name: Install Modal
-        run: |
-          python -m pip install --upgrade pip
-          pip install modal==1.3.0.post1 jinja2
+          pip install modal==0.63.64 jinja2
       - name: Update env vars
         run: |
           echo "BASE_TAG=main-base-py${{ matrix.python_version }}-cu${{ matrix.cuda }}-${{ matrix.pytorch }}" >> $GITHUB_ENV
@@ -200,7 +131,5 @@ jobs:
           echo "N_GPUS=${{ matrix.num_gpus }}" >> $GITHUB_ENV
           echo "NIGHTLY_BUILD=${{ matrix.nightly_build }}" >> $GITHUB_ENV
       - name: Run tests job on Modal
-        env:
-          CODECOV_TOKEN: ${{ secrets.CODECOV_TOKEN }}
         run: |
-          modal run cicd.multigpu
+          modal run cicd.tests

.github/workflows/tests.yml

@@ -1,24 +1,24 @@
 name: Tests
 on:
   # check on push/merge to main, PRs, and manual triggers
-  merge_group:
   push:
     branches:
       - "main"
     paths:
-      - "**.py"
-      - "pyproject.toml"
-      - ".github/workflows/*.yml"
-      - "cicd/cicd.sh"
-      - "cicd/Dockerfile-uv.jinja"
+      - '**.py'
+      - 'requirements.txt'
+      - '.github/workflows/*.yml'
+      - 'requirements-tests.txt'
+      - 'cicd/cicd.sh'
+      - 'cicd/Dockerfile.jinja'
   pull_request:
-    types: [opened, synchronize, reopened, ready_for_review]
-    paths:
-      - "**.py"
-      - "pyproject.toml"
-      - ".github/workflows/*.yml"
-      - "cicd/cicd.sh"
-      - "cicd/Dockerfile-uv.jinja"
+      paths:
+       - '**.py'
+       - 'requirements.txt'
+       - '.github/workflows/*.yml'
+       - 'requirements-tests.txt'
+       - 'cicd/cicd.sh'
+       - 'cicd/Dockerfile.jinja'
   workflow_dispatch:
 
 # Cancel jobs on the same ref if a new one is triggered
@@ -26,255 +26,190 @@ concurrency:
   group: ${{ github.workflow }}-${{ github.ref }}
   cancel-in-progress: ${{ github.ref != 'refs/heads/main' }}
 
-permissions:
-  contents: read
-
-env:
-  TRANSFORMERS_IS_CI: "yes"
-  UV_SYSTEM_PYTHON: "1"
-
 jobs:
   pre-commit:
     name: pre-commit
     runs-on: ubuntu-latest
-    if: ${{ !github.event.pull_request.draft }}
     steps:
       - uses: actions/checkout@v4
       - uses: actions/setup-python@v5
         with:
-          python-version: "3.11"
-          cache: "pip" # caching pip dependencies
+          python-version: "3.10"
+          cache: 'pip' # caching pip dependencies
       - uses: pre-commit/action@v3.0.1
         env:
           SKIP: no-commit-to-branch
 
-  prime-cdn-s3-cache:
-    name: Prefetch S3 once to prime the CDN cache
-    runs-on: ubuntu-latest
-    if: ${{ !github.event.pull_request.draft }}
-    timeout-minutes: 10
-    steps:
-      - name: Restore Cache from S3
-        id: hf-cache-restore-s3
-        run: |
-          curl -v -H "Range: bytes=0-1023" -L https://axolotl-ci.b-cdn.net/hf-cache.tar.zst > /dev/null
-
   pytest:
     name: PyTest
     runs-on: ubuntu-latest
-    if: ${{ !github.event.pull_request.draft }}
-    needs: [prime-cdn-s3-cache]
     strategy:
       fail-fast: false
+      max-parallel: 2
       matrix:
-        python_version: ["3.12", "3.14"]
-        pytorch_version: ["2.9.1", "2.10.0"]
+        python_version: ["3.10", "3.11"]
+        pytorch_version: ["2.3.1", "2.4.1", "2.5.1"]
         exclude:
-          - python_version: "3.14"
-            pytorch_version: "2.9.1"
-    timeout-minutes: 25
+          - python_version: "3.10"
+            pytorch_version: "2.4.1"
+          - python_version: "3.10"
+            pytorch_version: "2.5.1"
+    timeout-minutes: 20
 
     steps:
-      - name: cleanup node
-        run: |
-          sudo rm -rf /usr/share/dotnet /usr/local/lib/android /opt/ghc /opt/hostedtoolcache/CodeQL
-
       - name: Check out repository code
         uses: actions/checkout@v4
 
-      - name: Restore Cache from S3
-        id: hf-cache-restore-s3
-        run: |
-          mkdir -p ~/.cache/huggingface/hub
-          curl -L https://axolotl-ci.b-cdn.net/hf-cache.tar.zst | tar -xpf - -C ~/.cache/huggingface/hub/  --use-compress-program unzstd --strip-components=1
-          ls -ltr ~/.cache/huggingface/hub/
-
       - name: Setup Python
         uses: actions/setup-python@v5
         with:
           python-version: ${{ matrix.python_version }}
+          cache: 'pip' # caching pip dependencies
 
-      - name: Install uv
-        uses: astral-sh/setup-uv@v7
+      - name: upgrade pip
+        run: |
+          pip3 install --upgrade pip
+          pip3 install --upgrade packaging setuptools wheel
 
       - name: Install PyTorch
         run: |
-          uv pip install torch==${{ matrix.pytorch_version }} torchvision
-          uv pip freeze | grep -E "^(torch|torchvision)==" > /tmp/torch-pin.txt
+          pip3 install torch==${{ matrix.pytorch_version }}
 
       - name: Install dependencies
         run: |
-          uv pip install --no-build-isolation -e . --override /tmp/torch-pin.txt
-          python scripts/cutcrossentropy_install.py --uv | sh
-          uv pip install black mypy pre-commit types-requests quartodoc jupyter blobfile tiktoken \
-            codecov codecov-cli pytest pytest-cov pytest-retry pytest-sugar pytest-xdist tbparse
-
-      - name: Make sure PyTorch version wasn't clobbered
-        run: |
-          python -c "import torch; assert '${{ matrix.pytorch_version }}' in torch.__version__, f'Expected torch ${{ matrix.pytorch_version }} but got {torch.__version__}'"
+          pip3 show torch
+          pip3 install -U -e .
+          python scripts/unsloth_install.py | sh
+          python scripts/cutcrossentropy_install.py | sh
+          pip3 install -r requirements-dev.txt -r requirements-tests.txt
 
       - name: Ensure axolotl CLI was installed
         run: |
           axolotl --help
 
-      - name: Pre-Download dataset fixture
-        run: |
-          hf download --repo-type=dataset axolotl-ai-internal/axolotl-oss-dataset-fixtures
-
-      - name: Show HF cache
-        run: hf cache ls
-
       - name: Run tests
         run: |
-          df -h
-          pytest -v --durations=10 -n4 --dist loadfile --ignore=tests/e2e/ --ignore=tests/patched/ --ignore=tests/cli/ --ignore=tests/monkeypatch/ tests/ --cov=axolotl --cov-report=xml
-          df -h
-          pytest -v --durations=10 tests/monkeypatch/ --cov=axolotl --cov-append --cov-report=xml
-          df -h
-          pytest -v --durations=10 tests/patched/ --cov=axolotl --cov-append --cov-report=xml
-          df -h
-          pytest -v --durations=10 tests/cli/ --cov=axolotl --cov-append --cov-report=xml
+          pytest -n8 --dist loadfile --ignore=tests/e2e/ --ignore=tests/patched/ tests/
+          pytest tests/patched/
 
-      - name: Show HF cache
-        run: hf cache ls
-
-      - name: Upload coverage to Codecov
-        uses: codecov/codecov-action@v5
-        with:
-          token: ${{ secrets.CODECOV_TOKEN }}
-          files: ./coverage.xml
-          flags: unittests,pytorch-${{ matrix.pytorch_version }}
-          fail_ci_if_error: false
+      - name: cleanup pip cache
+        run: |
+          find "$(pip cache dir)/http-v2" -type f -mtime +14 -exec rm {} \;
 
   pytest-sdist:
     name: PyTest from Source Dist
     runs-on: ubuntu-latest
-    if: ${{ !github.event.pull_request.draft }}
-    needs: [prime-cdn-s3-cache]
     strategy:
       fail-fast: false
+      max-parallel: 1
       matrix:
-        python_version: ["3.12", "3.14"]
-        pytorch_version: ["2.9.1", "2.10.0"]
-        exclude:
-          - python_version: "3.14"
-            pytorch_version: "2.9.1"
-    timeout-minutes: 30
+        python_version: ["3.11"]
+        pytorch_version: ["2.4.1", "2.5.1"]
+    timeout-minutes: 20
 
     steps:
-      - name: cleanup node
-        run: |
-          sudo rm -rf /usr/share/dotnet /usr/local/lib/android /opt/ghc /opt/hostedtoolcache/CodeQL
-
       - name: Check out repository code
         uses: actions/checkout@v4
 
-      - name: Restore Cache from S3
-        id: hf-cache-restore-s3
-        run: |
-          mkdir -p ~/.cache/huggingface/hub
-          curl -L https://axolotl-ci.b-cdn.net/hf-cache.tar.zst | tar -xpf - -C ~/.cache/huggingface/hub/  --use-compress-program unzstd --strip-components=1
-          ls -ltr ~/.cache/huggingface/hub/
-
       - name: Setup Python
         uses: actions/setup-python@v5
         with:
           python-version: ${{ matrix.python_version }}
+          cache: 'pip' # caching pip dependencies
 
-      - name: Install uv
-        uses: astral-sh/setup-uv@v7
+      - name: upgrade pip
+        run: |
+          pip3 install --upgrade pip
+          pip3 install --upgrade packaging setuptools wheel
 
       - name: Install PyTorch
         run: |
-          uv pip install torch==${{ matrix.pytorch_version }} torchvision
-          uv pip freeze | grep -E "^(torch|torchvision)==" > /tmp/torch-pin.txt
+          pip3 install torch==${{ matrix.pytorch_version }}
 
       - name: Install dependencies
         run: |
-          uv pip install packaging setuptools_scm build wheel psutil
-          python -m build --no-isolation --sdist
-          uv pip install --no-build-isolation dist/axolotl*.tar.gz --override /tmp/torch-pin.txt
-          python scripts/cutcrossentropy_install.py --uv | sh
-          uv pip install black mypy pre-commit types-requests quartodoc jupyter blobfile tiktoken \
-            codecov codecov-cli pytest pytest-cov pytest-retry pytest-sugar pytest-xdist tbparse
-
-      - name: Make sure PyTorch version wasn't clobbered
-        run: |
-          python -c "import torch; assert '${{ matrix.pytorch_version }}' in torch.__version__, f'Expected torch ${{ matrix.pytorch_version }} but got {torch.__version__}'"
+          pip3 show torch
+          python3 setup.py sdist
+          pip3 install dist/axolotl*.tar.gz
+          python scripts/unsloth_install.py | sh
+          python scripts/cutcrossentropy_install.py | sh
+          pip3 install -r requirements-dev.txt -r requirements-tests.txt
 
       - name: Ensure axolotl CLI was installed
         run: |
           axolotl --help
 
-      - name: Verify agent docs are discoverable
-        run: |
-          # Agent docs live in docs/agents/ (source of truth) and are resolved
-          # at runtime from the repo checkout or via `axolotl fetch docs`
-          axolotl agent-docs --list
-          axolotl agent-docs | grep -q "Fine-tuning framework"
-          axolotl agent-docs grpo | grep -q "GRPO"
-          axolotl agent-docs sft | grep -q "SFT"
-          python -c "from axolotl.cli.agent_docs import get_doc, list_topics; assert len(list_topics()) >= 5; assert 'GRPO' in get_doc('grpo')"
-
-      - name: Show HF cache
-        run: hf cache ls
-
       - name: Run tests
         run: |
-          pytest -v --durations=10 -n4 --dist loadfile --ignore=tests/e2e/ --ignore=tests/patched/ --ignore=tests/cli/ --ignore=tests/monkeypatch/ tests/ --cov=axolotl --cov-report=xml
-          pytest -v --durations=10 tests/monkeypatch/ --cov=axolotl --cov-append --cov-report=xml
-          pytest -v --durations=10 tests/cli/
+          pytest -n8 --dist loadfile --ignore=tests/e2e/ --ignore=tests/patched/ tests/
+          pytest tests/patched/
 
-      - name: Show HF cache
-        run: hf cache ls
+      - name: cleanup pip cache
+        run: |
+          find "$(pip cache dir)/http-v2" -type f -mtime +14 -exec rm {} \;
 
-  gate-skip-e2e:
-    needs: [pre-commit]
-    runs-on: ubuntu-latest
-    outputs:
-      skip: ${{ steps.compute.outputs.skip }}
-    steps:
-      - uses: actions/github-script@v7
-        id: compute
-        with:
-          script: |
-            const token = /\[skip-e2e\]/i;
-            let msg = '';
-            if (context.eventName === 'push') {
-              msg = context.payload.head_commit?.message || '';
-            } else if (context.eventName === 'pull_request') {
-              const { owner, repo } = context.repo;
-              const prNumber = context.payload.pull_request.number;
-              const commits = await github.paginate(
-                github.rest.pulls.listCommits,
-                { owner, repo, pull_number: prNumber, per_page: 100 }
-              );
-              msg = commits.at(-1)?.commit?.message || '';
-            }
-            const title = context.payload.pull_request?.title || '';
-            const body  = context.payload.pull_request?.body  || '';
-            const skip = token.test(msg) || token.test(title) || token.test(body);
-            core.setOutput('skip', String(skip));
+#  docker-e2e-tests-1st:
+#    if: ${{ ! contains(github.event.commits[0].message, '[skip e2e]') && github.repository_owner == 'axolotl-ai-cloud' }}
+#    # this job needs to be run on self-hosted GPU runners...
+#    runs-on: [self-hosted, modal]
+#    timeout-minutes: 90
+#    needs: [pre-commit, pytest, pytest-sdist]
+#
+#    strategy:
+#      fail-fast: false
+#      matrix:
+#        include:
+#          - cuda: 124
+#            cuda_version: 12.4.1
+#            python_version: "3.11"
+#            pytorch: 2.4.1
+#            num_gpus: 1
+#            axolotl_extras:
+#    steps:
+#      - name: Checkout
+#        uses: actions/checkout@v4
+#      - name: Install Python
+#        uses: actions/setup-python@v5
+#        with:
+#          python-version: "3.10"
+#      - name: Install Modal
+#        run: |
+#          python -m pip install --upgrade pip
+#          pip install modal==0.63.64 jinja2
+#      - name: Update env vars
+#        run: |
+#          echo "BASE_TAG=main-base-py${{ matrix.python_version }}-cu${{ matrix.cuda }}-${{ matrix.pytorch }}" >> $GITHUB_ENV
+#          echo "PYTORCH_VERSION=${{ matrix.pytorch}}" >> $GITHUB_ENV
+#          echo "AXOLOTL_ARGS=${{ matrix.axolotl_args}}" >> $GITHUB_ENV
+#          echo "AXOLOTL_EXTRAS=${{ matrix.axolotl_extras}}" >> $GITHUB_ENV
+#          echo "CUDA=${{ matrix.cuda }}" >> $GITHUB_ENV
+#          echo "N_GPUS=${{ matrix.num_gpus }}" >> $GITHUB_ENV
+#      - name: Run tests job on Modal
+#        run: |
+#          modal run cicd.tests
 
-  docker-e2e-tests-1st:
-    # Run this job first as a gate for running the remainder of the test matrix
-    if: >
-      github.repository_owner == 'axolotl-ai-cloud' &&
-      (github.event_name != 'pull_request' || !github.event.pull_request.draft) &&
-      needs.gate-skip-e2e.outputs.skip != 'true'
+  docker-e2e-tests:
+    if: github.repository_owner == 'axolotl-ai-cloud'
     # this job needs to be run on self-hosted GPU runners...
     runs-on: [self-hosted, modal]
-    timeout-minutes: 120
+    timeout-minutes: 90
+#    needs: [pre-commit, pytest, docker-e2e-tests-1st]
     needs: [pre-commit, pytest]
 
     strategy:
       fail-fast: false
       matrix:
         include:
-          - cuda: 130
-            cuda_version: 13.0.0
-            python_version: "3.12"
-            pytorch: 2.9.1
+#          - cuda: 121
+#            cuda_version: 12.1.1
+#            python_version: "3.10"
+#            pytorch: 2.3.1
+#            num_gpus: 1
+#            axolotl_extras: mamba-ssm
+          - cuda: 124
+            cuda_version: 12.4.1
+            python_version: "3.11"
+            pytorch: 2.5.1
             num_gpus: 1
             axolotl_extras:
     steps:
@@ -283,125 +218,19 @@ jobs:
       - name: Install Python
         uses: actions/setup-python@v5
         with:
-          python-version: "3.11"
+          python-version: "3.10"
       - name: Install Modal
         run: |
           python -m pip install --upgrade pip
-          pip install modal==1.3.0.post1 jinja2
+          pip install modal==0.63.64 jinja2
       - name: Update env vars
         run: |
-          echo "BASE_TAG=main-base-py${{ matrix.python_version }}-cu${{ matrix.cuda }}-${{ matrix.pytorch }}" >> $GITHUB_ENV
+          echo "BASE_TAG=pr-2139-base-py${{ matrix.python_version }}-cu${{ matrix.cuda }}-${{ matrix.pytorch }}" >> $GITHUB_ENV
           echo "PYTORCH_VERSION=${{ matrix.pytorch}}" >> $GITHUB_ENV
           echo "AXOLOTL_ARGS=${{ matrix.axolotl_args}}" >> $GITHUB_ENV
           echo "AXOLOTL_EXTRAS=${{ matrix.axolotl_extras}}" >> $GITHUB_ENV
           echo "CUDA=${{ matrix.cuda }}" >> $GITHUB_ENV
-          echo "MODAL_IMAGE_BUILDER_VERSION=2024.10" >> $GITHUB_ENV
-          echo "N_GPUS=${{ matrix.num_gpus }}" >> $GITHUB_ENV
-          echo "E2E_DOCKERFILE=${{ matrix.dockerfile || 'Dockerfile-uv.jinja'}}" >> $GITHUB_ENV
-      - name: Run tests job on Modal
-        env:
-          CODECOV_TOKEN: ${{ secrets.CODECOV_TOKEN }}
-        run: |
-          modal run cicd.e2e_tests
-
-  docker-e2e-tests:
-    if: >
-      github.repository_owner == 'axolotl-ai-cloud' &&
-      (github.event_name != 'pull_request' || !github.event.pull_request.draft) &&
-      needs.gate-skip-e2e.outputs.skip != 'true'
-    # this job needs to be run on self-hosted GPU runners...
-    runs-on: [self-hosted, modal]
-    timeout-minutes: 120
-    # Only run the remainder of the matrix if the first e2e check passed;
-    # this is to save on wasted compute costs for known failures that get caught in the first run
-    needs: [pre-commit, pytest, gate-skip-e2e, docker-e2e-tests-1st]
-
-    strategy:
-      fail-fast: false
-      matrix:
-        include:
-          - cuda: 128
-            cuda_version: 12.8.1
-            python_version: "3.11"
-            pytorch: 2.9.1
-            num_gpus: 1
-            axolotl_extras:
-          - cuda: 128
-            cuda_version: 12.8.1
-            python_version: "3.11"
-            pytorch: 2.10.0
-            num_gpus: 1
-            axolotl_extras:
-          - cuda: 130
-            cuda_version: 13.0.0
-            python_version: "3.11"
-            pytorch: 2.9.1
-            num_gpus: 1
-            axolotl_extras:
-    steps:
-      - name: Checkout
-        uses: actions/checkout@v4
-      - name: Install Python
-        uses: actions/setup-python@v5
-        with:
-          python-version: "3.11"
-      - name: Install Modal
-        run: |
-          python -m pip install --upgrade pip
-          pip install modal==1.3.0.post1 jinja2
-      - name: Update env vars
-        run: |
-          echo "BASE_TAG=main-base-py${{ matrix.python_version }}-cu${{ matrix.cuda }}-${{ matrix.pytorch }}" >> $GITHUB_ENV
-          echo "PYTORCH_VERSION=${{ matrix.pytorch}}" >> $GITHUB_ENV
-          echo "AXOLOTL_ARGS=${{ matrix.axolotl_args}}" >> $GITHUB_ENV
-          echo "AXOLOTL_EXTRAS=${{ matrix.axolotl_extras}}" >> $GITHUB_ENV
-          echo "CUDA=${{ matrix.cuda }}" >> $GITHUB_ENV
-          echo "MODAL_IMAGE_BUILDER_VERSION=2024.10" >> $GITHUB_ENV
-          echo "N_GPUS=${{ matrix.num_gpus }}" >> $GITHUB_ENV
-          echo "GPU_TYPE=${{ matrix.gpu_type || 'L40S'}}" >> $GITHUB_ENV
-          echo "E2E_DOCKERFILE=${{ matrix.dockerfile || 'Dockerfile-uv.jinja'}}" >> $GITHUB_ENV
-      - name: Run tests job on Modal
-        env:
-          CODECOV_TOKEN: ${{ secrets.CODECOV_TOKEN }}
-        run: |
-          modal run cicd.e2e_tests
-
-  docker-e2e-cleanup:
-    runs-on: [self-hosted, modal]
-    timeout-minutes: 90
-    needs: [docker-e2e-tests]
-    if: ${{ !github.event.pull_request.draft }}
-
-    strategy:
-      fail-fast: false
-      matrix:
-        include:
-          - cuda: 128
-            cuda_version: 12.8.1
-            python_version: "3.11"
-            pytorch: 2.9.1
-            num_gpus: 1
-            axolotl_extras:
-    steps:
-      - name: Checkout
-        uses: actions/checkout@v4
-      - name: Install Python
-        uses: actions/setup-python@v5
-        with:
-          python-version: "3.11"
-      - name: Install Modal
-        run: |
-          python -m pip install --upgrade pip
-          pip install modal==1.3.0.post1 jinja2
-      - name: Update env vars
-        run: |
-          echo "BASE_TAG=main-base-py${{ matrix.python_version }}-cu${{ matrix.cuda }}-${{ matrix.pytorch }}" >> $GITHUB_ENV
-          echo "PYTORCH_VERSION=${{ matrix.pytorch}}" >> $GITHUB_ENV
-          echo "AXOLOTL_ARGS=${{ matrix.axolotl_args}}" >> $GITHUB_ENV
-          echo "AXOLOTL_EXTRAS=${{ matrix.axolotl_extras}}" >> $GITHUB_ENV
-          echo "CUDA=${{ matrix.cuda }}" >> $GITHUB_ENV
-          echo "MODAL_IMAGE_BUILDER_VERSION=2024.10" >> $GITHUB_ENV
           echo "N_GPUS=${{ matrix.num_gpus }}" >> $GITHUB_ENV
       - name: Run tests job on Modal
         run: |
-          modal run cicd.cleanup
+          modal run cicd.tests

.gitignore

@@ -1,7 +1,6 @@
 **/axolotl.egg-info
 configs
 last_run_prepared/
-outputs
 .vscode
 _site/
 
@@ -181,15 +180,8 @@ prepared-datasets/
 submit.sh
 *.out*
 
-# Quartodoc generated files
-objects.json
-site_libs/
-
 typings/
 out/
 
 # vim
 *.swp
-
-# scm auto-versioning
-src/axolotl/_version.py

.isort.cfg

@@ -0,0 +1,3 @@
+[settings]
+profile=black
+known_third_party=wandb,comet_ml

.pre-commit-config.yaml

@@ -3,21 +3,31 @@ default_language_version:
 
 repos:
 -   repo: https://github.com/pre-commit/pre-commit-hooks
-    rev: v6.0.0
+    rev: v4.4.0
     hooks:
     -   id: check-yaml
     -   id: end-of-file-fixer
     -   id: trailing-whitespace
     -   id: no-commit-to-branch
         args: ['--branch', 'main']
--   repo: https://github.com/astral-sh/ruff-pre-commit
-    rev: v0.15.8
+-   repo: https://github.com/psf/black
+    rev: 23.3.0
     hooks:
-    -   id: ruff
-        args: [--fix]
-    -   id: ruff-format
+    -   id: black
+-   repo: https://github.com/pycqa/isort
+    rev: 5.12.0
+    hooks:
+      - id: isort
+-   repo: https://github.com/PyCQA/flake8
+    rev: 6.0.0
+    hooks:
+    - id: flake8
+-   repo: https://github.com/PyCQA/pylint
+    rev: v2.17.4
+    hooks:
+    - id: pylint
 -   repo: https://github.com/pre-commit/mirrors-mypy
-    rev: v1.19.1
+    rev: v1.3.0
     hooks:
     - id: mypy
       additional_dependencies:
@@ -26,7 +36,7 @@ repos:
             'pydantic>=2.5.3',
         ]
 -   repo: https://github.com/PyCQA/bandit
-    rev: 1.9.4
+    rev: 1.7.5
     hooks:
     -   id: bandit
         args: [

.pylintrc

@@ -0,0 +1,14 @@
+[MASTER]
+init-hook="from pylint.config import find_pylintrc; import os, sys; sys.path.append(os.path.dirname(find_pylintrc()))"
+
+[TYPECHECK]
+
+# List of members which are set dynamically and missed by Pylint inference
+# system, and so shouldn't trigger E1101 when accessed.
+generated-members=numpy.*, torch.*
+
+
+[pylint.messages_control]
+disable=missing-function-docstring, line-too-long, import-error,
+    too-many-arguments, too-many-locals, too-many-statements, too-many-branches, too-few-public-methods,
+    too-many-instance-attributes, fixme, import-outside-toplevel, logging-fstring-interpolation,

.runpod/.gitignore

@@ -1,161 +0,0 @@
-# Byte-compiled / optimized / DLL files
-__pycache__/
-*.py[cod]
-*$py.class
-
-# C extensions
-*.so
-
-# Distribution / packaging
-.Python
-build/
-develop-eggs/
-dist/
-downloads/
-eggs/
-.eggs/
-lib/
-lib64/
-parts/
-sdist/
-var/
-wheels/
-share/python-wheels/
-*.egg-info/
-.installed.cfg
-*.egg
-MANIFEST
-
-# PyInstaller
-#  Usually these files are written by a python script from a template
-#  before PyInstaller builds the exe, so as to inject date/other infos into it.
-*.manifest
-*.spec
-
-# Installer logs
-pip-log.txt
-pip-delete-this-directory.txt
-
-# Unit test / coverage reports
-htmlcov/
-.tox/
-.nox/
-.coverage
-.coverage.*
-.cache
-nosetests.xml
-coverage.xml
-*.cover
-*.py,cover
-.hypothesis/
-.pytest_cache/
-cover/
-
-# Translations
-*.mo
-*.pot
-
-# Django stuff:
-*.log
-local_settings.py
-db.sqlite3
-db.sqlite3-journal
-
-# Flask stuff:
-instance/
-.webassets-cache
-
-# Scrapy stuff:
-.scrapy
-
-# Sphinx documentation
-docs/_build/
-
-# PyBuilder
-.pybuilder/
-target/
-
-# Jupyter Notebook
-.ipynb_checkpoints
-
-# IPython
-profile_default/
-ipython_config.py
-
-# pyenv
-#   For a library or package, you might want to ignore these files since the code is
-#   intended to run in multiple environments; otherwise, check them in:
-# .python-version
-
-# pipenv
-#   According to pypa/pipenv#598, it is recommended to include Pipfile.lock in version control.
-#   However, in case of collaboration, if having platform-specific dependencies or dependencies
-#   having no cross-platform support, pipenv may install dependencies that don't work, or not
-#   install all needed dependencies.
-#Pipfile.lock
-
-# poetry
-#   Similar to Pipfile.lock, it is generally recommended to include poetry.lock in version control.
-#   This is especially recommended for binary packages to ensure reproducibility, and is more
-#   commonly ignored for libraries.
-#   https://python-poetry.org/docs/basic-usage/#commit-your-poetrylock-file-to-version-control
-#poetry.lock
-
-# pdm
-#   Similar to Pipfile.lock, it is generally recommended to include pdm.lock in version control.
-#pdm.lock
-#   pdm stores project-wide configurations in .pdm.toml, but it is recommended to not include it
-#   in version control.
-#   https://pdm.fming.dev/#use-with-ide
-.pdm.toml
-
-# PEP 582; used by e.g. github.com/David-OConnor/pyflow and github.com/pdm-project/pdm
-__pypackages__/
-
-# Celery stuff
-celerybeat-schedule
-celerybeat.pid
-
-# SageMath parsed files
-*.sage.py
-
-# Environments
-.env
-.venv
-env/
-venv/
-ENV/
-env.bak/
-venv.bak/
-
-# Spyder project settings
-.spyderproject
-.spyproject
-
-# Rope project settings
-.ropeproject
-
-# mkdocs documentation
-/site
-
-# mypy
-.mypy_cache/
-.dmypy.json
-dmypy.json
-
-# Pyre type checker
-.pyre/
-
-# pytype static type analyzer
-.pytype/
-
-# Cython debug symbols
-cython_debug/
-
-# PyCharm
-#  JetBrains specific template is maintained in a separate JetBrains.gitignore that can
-#  be found at https://github.com/github/gitignore/blob/main/Global/JetBrains.gitignore
-#  and can be added to the global gitignore or merged into this file.  For a more nuclear
-#  option (not recommended) you can uncomment the following to ignore the entire idea folder.
-#.idea/
-pod/scripts/config.yaml

.runpod/Dockerfile

@@ -1,19 +0,0 @@
-FROM axolotlai/axolotl-cloud:main-py3.11-cu124-2.6.0
-
-COPY .runpod/requirements.txt /requirements.txt
-RUN --mount=type=cache,target=/root/.cache/pip \
-    python3 -m pip install --upgrade pip && \
-    python3 -m pip install --upgrade -r /requirements.txt
-
-# Environment settings
-ARG BASE_VOLUME="/runpod-volume"
-ENV BASE_VOLUME=$BASE_VOLUME
-ENV HF_DATASETS_CACHE="${BASE_VOLUME}/huggingface-cache/datasets"
-ENV HUGGINGFACE_HUB_CACHE="${BASE_VOLUME}/huggingface-cache/hub"
-ENV HF_HUB_CACHE="${BASE_VOLUME}/huggingface-cache/hub"
-ENV TRANSFORMERS_CACHE="${BASE_VOLUME}/huggingface-cache/hub"
-
-COPY .runpod/src /src
-
-WORKDIR /src
-CMD ["python3", "/src/handler.py"]

.runpod/README.md

@@ -1,335 +0,0 @@
-<h1>LLM Post Training- Full fine-tune, LoRA, QLoRa etc. Llama/Mistral/Gemma and more</h1>
-
-# Configuration Options
-
-This document outlines all available configuration options for training models. The configuration can be provided as a JSON request.
-
-## Usage
-
-You can use these configuration Options:
-
-1. As a JSON request body:
-
-```json
-{
-  "input": {
-    "user_id": "user",
-    "model_id": "model-name",
-    "run_id": "run-id",
-    "credentials": {
-      "wandb_api_key": "", # add your Weights & biases key. TODO:  you will be able to set this in Enviornment variables.
-      "hf_token": "", # add your HF_token. TODO:  you will be able to set this in Enviornment variables.
-    },
-    "args": {
-      "base_model": "NousResearch/Llama-3.2-1B",
-      // ... other options
-    }
-  }
-}
-```
-
-## Configuration Options
-
-### Model Configuration
-
-| Option              | Description                                                                                   | Default              |
-| ------------------- | --------------------------------------------------------------------------------------------- | -------------------- |
-| `base_model`        | Path to the base model (local or HuggingFace)                                                 | Required             |
-| `base_model_config` | Configuration path for the base model                                                         | Same as base_model   |
-| `revision_of_model` | Specific model revision from HuggingFace hub                                                  | Latest               |
-| `tokenizer_config`  | Custom tokenizer configuration path                                                           | Optional             |
-| `model_type`        | Type of model to load                                                                         | AutoModelForCausalLM |
-| `tokenizer_type`    | Type of tokenizer to use                                                                      | AutoTokenizer        |
-| `hub_model_id`      | Repository ID where the model will be pushed on Hugging Face Hub (format: username/repo-name) | Optional             |
-
-## Model Family Identification
-
-| Option                     | Default | Description                    |
-| -------------------------- | ------- | ------------------------------ |
-| `is_falcon_derived_model`  | `false` | Whether model is Falcon-based  |
-| `is_llama_derived_model`   | `false` | Whether model is LLaMA-based   |
-| `is_qwen_derived_model`    | `false` | Whether model is Qwen-based    |
-| `is_mistral_derived_model` | `false` | Whether model is Mistral-based |
-
-## Model Configuration Overrides
-
-| Option                                          | Default    | Description                        |
-| ----------------------------------------------- | ---------- | ---------------------------------- |
-| `overrides_of_model_config.rope_scaling.type`   | `"linear"` | RoPE scaling type (linear/dynamic) |
-| `overrides_of_model_config.rope_scaling.factor` | `1.0`      | RoPE scaling factor                |
-
-### Model Loading Options
-
-| Option         | Description                   | Default |
-| -------------- | ----------------------------- | ------- |
-| `load_in_8bit` | Load model in 8-bit precision | false   |
-| `load_in_4bit` | Load model in 4-bit precision | false   |
-| `bf16`         | Use bfloat16 precision        | false   |
-| `fp16`         | Use float16 precision         | false   |
-| `tf32`         | Use tensor float 32 precision | false   |
-
-## Memory and Device Settings
-
-| Option             | Default   | Description             |
-| ------------------ | --------- | ----------------------- |
-| `gpu_memory_limit` | `"20GiB"` | GPU memory limit        |
-| `lora_on_cpu`      | `false`   | Load LoRA on CPU        |
-| `device_map`       | `"auto"`  | Device mapping strategy |
-| `max_memory`       | `null`    | Max memory per device   |
-
-## Training Hyperparameters
-
-| Option                        | Default   | Description                 |
-| ----------------------------- | --------- | --------------------------- |
-| `gradient_accumulation_steps` | `1`       | Gradient accumulation steps |
-| `micro_batch_size`            | `2`       | Batch size per GPU          |
-| `eval_batch_size`             | `null`    | Evaluation batch size       |
-| `num_epochs`                  | `4`       | Number of training epochs   |
-| `warmup_steps`                | `100`     | Warmup steps                |
-| `warmup_ratio`                | `0.05`    | Warmup ratio                |
-| `learning_rate`               | `0.00003` | Learning rate               |
-| `lr_quadratic_warmup`         | `false`   | Quadratic warmup            |
-| `logging_steps`               | `null`    | Logging frequency           |
-| `eval_steps`                  | `null`    | Evaluation frequency        |
-| `evals_per_epoch`             | `null`    | Evaluations per epoch       |
-| `save_strategy`               | `"epoch"` | Checkpoint saving strategy  |
-| `save_steps`                  | `null`    | Saving frequency            |
-| `saves_per_epoch`             | `null`    | Saves per epoch             |
-| `save_total_limit`            | `null`    | Maximum checkpoints to keep |
-| `max_steps`                   | `null`    | Maximum training steps      |
-
-### Dataset Configuration
-
-```yaml
-datasets:
-  - path: vicgalle/alpaca-gpt4 # HuggingFace dataset or TODO: You will be able to add the local path.
-    type: alpaca # Format type (alpaca, gpteacher, oasst, etc.)
-    ds_type: json # Dataset type
-    data_files: path/to/data # Source data files
-    train_on_split: train # Dataset split to use
-```
-
-## Chat Template Settings
-
-| Option                   | Default                          | Description            |
-| ------------------------ | -------------------------------- | ---------------------- |
-| `chat_template`          | `"tokenizer_default"`            | Chat template type     |
-| `chat_template_jinja`    | `null`                           | Custom Jinja template  |
-| `default_system_message` | `"You are a helpful assistant."` | Default system message |
-
-## Dataset Processing
-
-| Option                            | Default                    | Description                         |
-| --------------------------------- | -------------------------- | ----------------------------------- |
-| `dataset_prepared_path`           | `"data/last_run_prepared"` | Path for prepared dataset           |
-| `push_dataset_to_hub`             | `""`                       | Push dataset to HF hub              |
-| `dataset_num_proc`                | `4`                        | Number of preprocessing processes   |
-| `dataset_keep_in_memory`          | `false`                    | Keep dataset in memory              |
-| `shuffle_merged_datasets`         | `true`                     | Shuffle merged datasets             |
-| `shuffle_before_merging_datasets` | `false`                    | Shuffle each dataset before merging |
-| `dataset_exact_deduplication`     | `true`                     | Deduplicate datasets                |
-
-## LoRA Configuration
-
-| Option                     | Default                | Description                    |
-| -------------------------- | ---------------------- | ------------------------------ |
-| `adapter`                  | `"lora"`               | Adapter type (lora/qlora)      |
-| `lora_model_dir`           | `""`                   | Directory with pretrained LoRA |
-| `lora_r`                   | `8`                    | LoRA attention dimension       |
-| `lora_alpha`               | `16`                   | LoRA alpha parameter           |
-| `lora_dropout`             | `0.05`                 | LoRA dropout                   |
-| `lora_target_modules`      | `["q_proj", "v_proj"]` | Modules to apply LoRA          |
-| `lora_target_linear`       | `false`                | Target all linear modules      |
-| `peft_layers_to_transform` | `[]`                   | Layers to transform            |
-| `lora_modules_to_save`     | `[]`                   | Modules to save                |
-| `lora_fan_in_fan_out`      | `false`                | Fan in/out structure           |
-
-## Optimization Settings
-
-| Option                    | Default | Description                |
-| ------------------------- | ------- | -------------------------- |
-| `train_on_inputs`         | `false` | Train on input prompts     |
-| `group_by_length`         | `false` | Group by sequence length   |
-| `gradient_checkpointing`  | `false` | Use gradient checkpointing |
-| `early_stopping_patience` | `3`     | Early stopping patience    |
-
-## Learning Rate Scheduling
-
-| Option                     | Default    | Description          |
-| -------------------------- | ---------- | -------------------- |
-| `lr_scheduler`             | `"cosine"` | Scheduler type       |
-| `lr_scheduler_kwargs`      | `{}`       | Scheduler parameters |
-| `cosine_min_lr_ratio`      | `null`     | Minimum LR ratio     |
-| `cosine_constant_lr_ratio` | `null`     | Constant LR ratio    |
-| `lr_div_factor`            | `null`     | LR division factor   |
-
-## Optimizer Settings
-
-| Option                 | Default      | Description         |
-| ---------------------- | ------------ | ------------------- |
-| `optimizer`            | `"adamw_hf"` | Optimizer choice    |
-| `optim_args`           | `{}`         | Optimizer arguments |
-| `optim_target_modules` | `[]`         | Target modules      |
-| `weight_decay`         | `null`       | Weight decay        |
-| `adam_beta1`           | `null`       | Adam beta1          |
-| `adam_beta2`           | `null`       | Adam beta2          |
-| `adam_epsilon`         | `null`       | Adam epsilon        |
-| `max_grad_norm`        | `null`       | Gradient clipping   |
-
-## Attention Implementations
-
-| Option                     | Default | Description                   |
-| -------------------------- | ------- | ----------------------------- |
-| `flash_optimum`            | `false` | Use better transformers       |
-| `xformers_attention`       | `false` | Use xformers                  |
-| `flash_attention`          | `false` | Use flash attention           |
-| `flash_attn_cross_entropy` | `false` | Flash attention cross entropy |
-| `flash_attn_rms_norm`      | `false` | Flash attention RMS norm      |
-| `flash_attn_fuse_mlp`      | `false` | Fuse MLP operations           |
-| `sdp_attention`            | `false` | Use scaled dot product        |
-| `s2_attention`             | `false` | Use shifted sparse attention  |
-
-## Tokenizer Modifications
-
-| Option           | Default | Description                  |
-| ---------------- | ------- | ---------------------------- |
-| `special_tokens` | -       | Special tokens to add/modify |
-| `tokens`         | `[]`    | Additional tokens            |
-
-## Distributed Training
-
-| Option                  | Default | Description           |
-| ----------------------- | ------- | --------------------- |
-| `fsdp`                  | `null`  | FSDP configuration    |
-| `fsdp_config`           | `null`  | FSDP config options   |
-| `deepspeed`             | `null`  | Deepspeed config path |
-| `ddp_timeout`           | `null`  | DDP timeout           |
-| `ddp_bucket_cap_mb`     | `null`  | DDP bucket capacity   |
-| `ddp_broadcast_buffers` | `null`  | DDP broadcast buffers |
-
-<details>
-<summary><h3>Example Configuration Request:</h3></summary>
-
-Here's a complete example for fine-tuning a LLaMA model using LoRA:
-
-```json
-{
-  "input": {
-    "user_id": "user",
-    "model_id": "llama-test",
-    "run_id": "test-run",
-    "credentials": {
-      "wandb_api_key": "",
-      "hf_token": ""
-    },
-    "args": {
-      "base_model": "NousResearch/Llama-3.2-1B",
-      "load_in_8bit": false,
-      "load_in_4bit": false,
-      "strict": false,
-      "datasets": [
-        {
-          "path": "teknium/GPT4-LLM-Cleaned",
-          "type": "alpaca"
-        }
-      ],
-      "dataset_prepared_path": "last_run_prepared",
-      "val_set_size": 0.1,
-      "output_dir": "./outputs/lora-out",
-      "adapter": "lora",
-      "sequence_len": 2048,
-      "sample_packing": true,
-      "eval_sample_packing": true,
-      "pad_to_sequence_len": true,
-      "lora_r": 16,
-      "lora_alpha": 32,
-      "lora_dropout": 0.05,
-      "lora_target_modules": [
-        "gate_proj",
-        "down_proj",
-        "up_proj",
-        "q_proj",
-        "v_proj",
-        "k_proj",
-        "o_proj"
-      ],
-      "gradient_accumulation_steps": 2,
-      "micro_batch_size": 2,
-      "num_epochs": 1,
-      "optimizer": "adamw_8bit",
-      "lr_scheduler": "cosine",
-      "learning_rate": 0.0002,
-      "train_on_inputs": false,
-      "group_by_length": false,
-      "bf16": "auto",
-      "tf32": false,
-      "gradient_checkpointing": true,
-      "logging_steps": 1,
-      "flash_attention": true,
-      "loss_watchdog_threshold": 5,
-      "loss_watchdog_patience": 3,
-      "warmup_steps": 10,
-      "evals_per_epoch": 4,
-      "saves_per_epoch": 1,
-      "weight_decay": 0,
-      "hub_model_id": "runpod/llama-fr-lora",
-      "wandb_name": "test-run-1",
-      "wandb_project": "test-run-1",
-      "wandb_entity": "axo-test",
-      "special_tokens": {
-        "pad_token": "<|end_of_text|>"
-      }
-    }
-  }
-}
-```
-
-</details>
-
-### Advanced Features
-
-#### Wandb Integration
-
-- `wandb_project`: Project name for Weights & Biases
-- `wandb_entity`: Team name in W&B
-- `wandb_watch`: Monitor model with W&B
-- `wandb_name`: Name of the W&B run
-- `wandb_run_id`: ID for the W&B run
-
-#### Performance Optimization
-
-- `sample_packing`: Enable efficient sequence packing
-- `eval_sample_packing`: Use sequence packing during evaluation
-- `torch_compile`: Enable PyTorch 2.0 compilation
-- `flash_attention`: Use Flash Attention implementation
-- `xformers_attention`: Use xFormers attention implementation
-
-### Available Optimizers
-
-The following optimizers are supported:
-
-- `adamw_hf`: HuggingFace's AdamW implementation
-- `adamw_torch`: PyTorch's AdamW
-- `adamw_torch_fused`: Fused AdamW implementation
-- `adamw_torch_xla`: XLA-optimized AdamW
-- `adamw_apex_fused`: NVIDIA Apex fused AdamW
-- `adafactor`: Adafactor optimizer
-- `adamw_anyprecision`: Anyprecision AdamW
-- `adamw_bnb_8bit`: 8-bit AdamW from bitsandbytes
-- `lion_8bit`: 8-bit Lion optimizer
-- `lion_32bit`: 32-bit Lion optimizer
-- `sgd`: Stochastic Gradient Descent
-- `adagrad`: Adagrad optimizer
-
-## Notes
-
-- Set `load_in_8bit: true` or `load_in_4bit: true` for memory-efficient training
-- Enable `flash_attention: true` for faster training on modern GPUs
-- Use `gradient_checkpointing: true` to reduce memory usage
-- Adjust `micro_batch_size` and `gradient_accumulation_steps` based on your GPU memory
-
-For more detailed information, please refer to the [documentation](https://axolotl-ai-cloud.github.io/axolotl/docs/config-reference.html).
-
-### Errors:
-
-- if you face any issues with the Flash Attention-2, Delete yoor worker and Re-start.

.runpod/hub.json

@@ -1,93 +0,0 @@
-{
-  "title": "Axolotl Fine-Tuning",
-  "description": "Serverless fine-tuning of open-source LLMs with Axolotl. Supports LoRA, QLoRA, DPO, and more using Hugging Face models and datasets.",
-  "type": "serverless",
-  "category": "language",
-  "iconUrl": "https://avatars.githubusercontent.com/u/167502477",
-  "config": {
-    "runsOn": "GPU",
-    "containerDiskInGb": 200,
-    "gpuCount": 1,
-    "allowedCudaVersions": [
-      "12.8",
-      "12.7",
-      "12.6",
-      "12.5",
-      "12.4"
-    ],
-    "presets": [],
-    "env": [
-      {
-        "key": "TOKENIZER",
-        "input": {
-          "name": "Tokenizer",
-          "type": "string",
-          "description": "Name or path of the Hugging Face tokenizer to use.",
-          "default": "",
-          "advanced": true
-        }
-      },
-      {
-        "key": "MAX_NUM_SEQS",
-        "input": {
-          "name": "Max Num Seqs",
-          "type": "number",
-          "description": "Maximum number of sequences per iteration.",
-          "default": 256,
-          "advanced": true
-        }
-      },
-      {
-        "key": "DISABLE_LOG_STATS",
-        "input": {
-          "name": "Disable Log Stats",
-          "type": "boolean",
-          "description": "Disable logging statistics.",
-          "default": false,
-          "trueValue": "true",
-          "falseValue": "false"
-        }
-      },
-      {
-        "key": "LOAD_FORMAT",
-        "input": {
-          "name": "Load Format",
-          "type": "string",
-          "description": "The format of the model weights to load.",
-          "default": "auto",
-          "options": [
-            {
-              "label": "auto",
-              "value": "auto"
-            },
-            {
-              "label": "pt",
-              "value": "pt"
-            },
-            {
-              "label": "safetensors",
-              "value": "safetensors"
-            },
-            {
-              "label": "npcache",
-              "value": "npcache"
-            },
-            {
-              "label": "dummy",
-              "value": "dummy"
-            },
-            {
-              "label": "tensorizer",
-              "value": "tensorizer"
-            },
-            {
-              "label": "bitsandbytes",
-              "value": "bitsandbytes"
-            }
-          ],
-          "advanced": true
-        }
-      }
-    ]
-  }
-}

.runpod/requirements.txt

@@ -1,7 +0,0 @@
-# Required Python packages get listed here, one per line.
-# Reccomended to lock the version number to avoid unexpected changes.
-
-# You can also install packages from a git repository, e.g.:
-# git+https://github.com/runpod/runpod-python.git
-# To learn more, see https://pip.pypa.io/en/stable/reference/requirements-file-format/
-runpod~=1.7.0

.runpod/src/config/config.yaml

@@ -1,564 +0,0 @@
-# # This is the huggingface model that contains *.pt, *.safetensors, or *.bin files
-# # This can also be a relative path to a model on disk
-# base_model: ./llama-7b-hf
-# # You can specify an ignore pattern if the model repo contains more than 1 model type (*.pt, etc)
-# base_model_ignore_patterns:
-# # If the base_model repo on hf hub doesn't include configuration .json files,
-# # You can set that here, or leave this empty to default to base_model
-# base_model_config: ./llama-7b-hf
-# # You can specify to choose a specific model revision from huggingface hub
-# model_revision:
-# # Optional tokenizer configuration override in case you want to use a different tokenizer
-# # than the one defined in the base model
-# tokenizer_config:
-# # If you want to specify the type of model to load, AutoModelForCausalLM is a good choice too
-# model_type: AutoModelForCausalLM
-# # Corresponding tokenizer for the model AutoTokenizer is a good choice
-# tokenizer_type: AutoTokenizer
-# # Trust remote code for untrusted source
-# trust_remote_code:
-# # use_fast option for tokenizer loading from_pretrained, default to True
-# tokenizer_use_fast:
-# # Whether to use the legacy tokenizer setting, defaults to True
-# tokenizer_legacy:
-# # Resize the model embeddings when new tokens are added to multiples of 32
-# # This is reported to improve training speed on some models
-# resize_token_embeddings_to_32x:
-
-# # Used to identify which the model is based on
-# is_falcon_derived_model:
-# is_llama_derived_model:
-# # Please note that if you set this to true, `padding_side` will be set to "left" by default
-# is_mistral_derived_model:
-# is_qwen_derived_model:
-
-# # optional overrides to the base model configuration
-# model_config:
-#   # RoPE Scaling https://github.com/huggingface/transformers/pull/24653
-#   rope_scaling:
-#     type: # linear | dynamic
-#     factor: # float
-
-# # Whether you are training a 4-bit GPTQ quantized model
-# gptq: true
-# gptq_groupsize: 128 # group size
-# gptq_model_v1: false # v1 or v2
-
-# # This will attempt to quantize the model down to 8 bits and use adam 8 bit optimizer
-# load_in_8bit: true
-# # Use bitsandbytes 4 bit
-# load_in_4bit:
-
-# # Use CUDA bf16
-# bf16: true # bool or 'full' for `bf16_full_eval`. require >=ampere
-# # Use CUDA fp16
-# fp16: true
-# # Use CUDA tf32
-# tf32: true # require >=ampere
-
-# # No AMP (automatic mixed precision)
-# bfloat16: true # require >=ampere
-# float16: true
-
-# # A list of one or more datasets to finetune the model with
-# datasets:
-#   # HuggingFace dataset repo | s3://,gs:// path | "json" for local dataset, make sure to fill data_files
-#   - path: vicgalle/alpaca-gpt4
-#   # The type of prompt to use for training. [alpaca, sharegpt, gpteacher, oasst, reflection]
-#     type: alpaca # format | format:<prompt_style> (chat/instruct) | <prompt_strategies>.load_<load_fn>
-#     ds_type: # Optional[str] (json|arrow|parquet|text|csv) defines the datatype when path is a file
-#     data_files: # Optional[str] path to source data files
-#     shards: # Optional[int] number of shards to split data into
-#     name: # Optional[str] name of dataset configuration to load
-#     train_on_split: train # Optional[str] name of dataset split to load from
-
-#     # Optional[str] fastchat conversation type, only used with type: sharegpt
-#     conversation:  # Options (see Conversation 'name'): https://github.com/lm-sys/FastChat/blob/main/fastchat/conversation.py
-#     field_human: # Optional[str]. Human key to use for conversation.
-#     field_model: # Optional[str]. Assistant key to use for conversation.
-
-#   # Custom user prompt
-#   - path: repo
-#     type:
-#       # The below are defaults. only set what's needed.
-#       system_prompt: ""
-#       system_format: "{system}"
-#       field_system: system
-#       field_instruction: instruction
-#       field_input: input
-#       field_output: output
-
-#       # Customizable to be single line or multi-line
-#       # 'format' can include {input}
-#       format: |-
-#         User: {instruction} {input}
-#         Assistant:
-#       # 'no_input_format' cannot include {input}
-#       no_input_format: "{instruction} "
-
-#       # For `completion` datasets only, uses the provided field instead of `text` column
-#       field:
-
-# # Axolotl attempts to save the dataset as an arrow after packing the data together so
-# # subsequent training attempts load faster, relative path
-# dataset_prepared_path: data/last_run_prepared
-# # Push prepared dataset to hub
-# push_dataset_to_hub: # repo path
-# # The maximum number of processes to use while preprocessing your input dataset. This defaults to `os.cpu_count()`
-# # if not set.
-# dataset_num_proc: # defaults to os.cpu_count() if not set
-# # push checkpoints to hub
-# hub_model_id: # repo path to push finetuned model
-# # how to push checkpoints to hub
-# # https://huggingface.co/docs/transformers/v4.31.0/en/main_classes/trainer#transformers.TrainingArguments.hub_strategy
-# hub_strategy:
-# # Whether to use hf `use_auth_token` for loading datasets. Useful for fetching private datasets
-# # Required to be true when used in combination with `push_dataset_to_hub`
-# hf_use_auth_token: # boolean
-# # How much of the dataset to set aside as evaluation. 1 = 100%, 0.50 = 50%, etc. 0 for no eval.
-# val_set_size: 0.04
-# # Num shards for whole dataset
-# dataset_shard_num:
-# # Index of shard to use for whole dataset
-# dataset_shard_idx:
-
-# # The maximum length of an input to train with, this should typically be less than 2048
-# # as most models have a token/context limit of 2048
-# sequence_len: 2048
-# # Pad inputs so each step uses constant sized buffers
-# # This will reduce memory fragmentation and may prevent OOMs, by re-using memory more efficiently
-# pad_to_sequence_len:
-# # Max sequence length to concatenate training samples together up to
-# # Inspired by StackLLaMA. see https://huggingface.co/blog/stackllama#supervised-fine-tuning
-# # FutureWarning: This will soon be DEPRECATED
-# max_packed_sequence_len: 1024
-# # Use efficient multi-packing with block diagonal attention and per sequence position_ids. Recommend set to 'true'
-# sample_packing:
-# # Set to 'false' if getting errors during eval with sample_packing on.
-# eval_sample_packing:
-# # You can set these packing optimizations AFTER starting a training at least once.
-# # The trainer will provide recommended values for these values.
-# sample_packing_eff_est:
-# total_num_tokens:
-
-# # If you want to use 'lora' or 'qlora' or leave blank to train all parameters in original model
-# adapter: lora
-# # If you already have a lora model trained that you want to load, put that here.
-# # This means after training, if you want to test the model, you should set this to the value of `lora_out_dir`.
-# lora_model_dir:
-
-# # LoRA hyperparameters
-# # For more details about the following options, see:
-# # https://www.anyscale.com/blog/fine-tuning-llms-lora-or-full-parameter-an-in-depth-analysis-with-llama-2
-# lora_r: 8
-# lora_alpha: 16
-# lora_dropout: 0.05
-# lora_target_modules:
-#   - q_proj
-#   - v_proj
-# #  - k_proj
-# #  - o_proj
-# #  - gate_proj
-# #  - down_proj
-# #  - up_proj
-# lora_target_linear: # If true, will target all linear layers
-
-# # If you added new tokens to the tokenizer, you may need to save some LoRA modules because they need to know the new tokens.
-# # For LLaMA and Mistral, you need to save `embed_tokens` and `lm_head`. It may vary for other models.
-# # `embed_tokens` converts tokens to embeddings, and `lm_head` converts embeddings to token probabilities.
-# # https://github.com/huggingface/peft/issues/334#issuecomment-1561727994
-# lora_modules_to_save:
-# #  - embed_tokens
-# #  - lm_head
-
-# # Once you complete training, the model will be saved to the following directory.
-# # If you merge the adapter to the base model, a subdirectory `merged` will be created under this directory.
-# # Make sure `lora_model_dir` points to this directory if you want to use the trained model.
-# lora_out_dir:
-# lora_fan_in_fan_out: false
-
-# # ReLoRA configuration
-# # Must use either 'lora' or 'qlora' adapter, and does not support fsdp or deepspeed
-# relora_steps: # Number of steps per ReLoRA restart
-# relora_warmup_steps: # Number of per-restart warmup steps
-# relora_cpu_offload: # True to perform lora weight merges on cpu during restarts, for modest gpu memory savings
-
-# # wandb configuration if you're using it
-# wandb_mode: # "offline" to save run metadata locally and not sync to the server, "disabled" to turn off wandb
-# wandb_project: # Your wandb project name
-# wandb_entity: # A wandb Team name if using a Team
-# wandb_watch:
-# wandb_run_id: # Set the name of your wandb run
-# wandb_log_model: # "checkpoint" to log model to wandb Artifacts every `save_steps` or "end" to log only at the end of training
-
-# # Where to save the full-finetuned model to
-# output_dir: ./completed-model
-
-# # Whether to use torch.compile and which backend to use
-# torch_compile:  # bool
-# torch_compile_backend:  # Optional[str]
-
-# # Training hyperparameters
-
-# # If greater than 1, backpropagation will be skipped and the gradients will be accumulated for the given number of steps.
-# gradient_accumulation_steps: 1
-# # The number of samples to include in each batch. This is the number of samples sent to each GPU.
-# micro_batch_size: 2
-# eval_batch_size:
-# num_epochs: 4
-# warmup_steps: 100  # cannot use with warmup_ratio
-# warmup_ratio: 0.05  # cannot use with warmup_steps
-# learning_rate: 0.00003
-# lr_quadratic_warmup:
-# logging_steps:
-# save_strategy: # Set to `no` to skip checkpoint saves
-# save_steps: # Leave empty to save at each epoch
-# eval_steps: # Leave empty to eval at each epoch, integers for every N steps. decimal for fraction of total steps
-# save_total_limit: # Checkpoints saved at a time
-# # Maximum number of iterations to train for. It precedes num_epochs which means that
-# # if both are set, num_epochs will not be guaranteed.
-# # e.g., when 1 epoch is 1000 steps => `num_epochs: 2` and `max_steps: 100` will train for 100 steps
-# max_steps:
-
-# eval_table_size: # Approximate number of predictions sent to wandb depending on batch size. Enabled above 0. Default is 0
-# eval_table_max_new_tokens: # Total number of tokens generated for predictions sent to wandb. Default is 128
-
-# # Whether to mask out or include the human's prompt from the training labels
-# train_on_inputs: false
-# # Group similarly sized data to minimize padding.
-# # May be slower to start, as it must download and sort the entire dataset.
-# # Note that training loss may have an oscillating pattern with this enabled.
-# group_by_length: false
-
-# # Whether to use gradient checkpointing https://huggingface.co/docs/transformers/v4.18.0/en/performance#gradient-checkpointing
-# gradient_checkpointing: false
-
-# # Stop training after this many evaluation losses have increased in a row
-# # https://huggingface.co/transformers/v4.2.2/_modules/transformers/trainer_callback.html#EarlyStoppingCallback
-# early_stopping_patience: 3
-
-# # Specify a scheduler and kwargs to use with the optimizer
-# lr_scheduler: # 'one_cycle' | empty for cosine
-# lr_scheduler_kwargs:
-
-# # For one_cycle optim
-# lr_div_factor: # Learning rate div factor
-
-# # Specify optimizer
-# # Valid values are driven by the Transformers OptimizerNames class, see:
-# # https://github.com/huggingface/transformers/blob/95b374952dc27d8511541d6f5a4e22c9ec11fb24/src/transformers/training_args.py#L134
-# #
-# # Note that not all optimizers may be available in your environment, ex: 'adamw_anyprecision' is part of
-# # torchdistx, 'adamw_bnb_8bit' is part of bnb.optim.Adam8bit, etc. When in doubt, it is recommended to start with the optimizer used
-# # in the examples/ for your model and fine-tuning use case.
-# #
-# # Valid values for 'optimizer' include:
-# # - adamw_hf
-# # - adamw_torch
-# # - adamw_torch_fused
-# # - adamw_torch_xla
-# # - adamw_apex_fused
-# # - adafactor
-# # - adamw_anyprecision
-# # - sgd
-# # - adagrad
-# # - adamw_bnb_8bit
-# # - lion_8bit
-# # - lion_32bit
-# # - paged_adamw_32bit
-# # - paged_adamw_8bit
-# # - paged_lion_32bit
-# # - paged_lion_8bit
-# optimizer:
-# # Specify weight decay
-# weight_decay:
-# # adamw hyperparams
-# adam_beta1:
-# adam_beta2:
-# adam_epsilon:
-# # Gradient clipping max norm
-# max_grad_norm:
-
-# # Augmentation techniques
-# # NEFT https://arxiv.org/abs/2310.05914, set this to a number (paper default is 5) to add noise to embeddings
-# # currently only supported on Llama and Mistral
-# noisy_embedding_alpha:
-
-# # Whether to bettertransformers
-# flash_optimum:
-# # Whether to use xformers attention patch https://github.com/facebookresearch/xformers:
-# xformers_attention:
-# # Whether to use flash attention patch https://github.com/Dao-AILab/flash-attention:
-# flash_attention:
-# flash_attn_cross_entropy:  # Whether to use flash-attention cross entropy implementation - advanced use only
-# flash_attn_rms_norm:  # Whether to use flash-attention rms norm implementation - advanced use only
-# flash_attn_fuse_mlp: # Whether to fuse part of the MLP into a single operation
-# # Whether to use scaled-dot-product attention
-# # https://pytorch.org/docs/stable/generated/torch.nn.functional.scaled_dot_product_attention.html
-# sdp_attention:
-# # Landmark attention (only llama)
-# landmark_attention:
-# # xpos RoPE see https://github.com/kaiokendev/cutoff-len-is-context-len/blob/main/util/xpos_rope_llama_monkey_patch.py
-# # LLaMA only
-# xpos_rope:
-
-# # Resume from a specific checkpoint dir
-# resume_from_checkpoint:
-# # If resume_from_checkpoint isn't set and you simply want it to start where it left off.
-# # Be careful with this being turned on between different models.
-# auto_resume_from_checkpoints: false
-
-# # Don't mess with this, it's here for accelerate and torchrun
-# local_rank:
-
-# # Add or change special tokens.
-# # If you add tokens here, you don't need to add them to the `tokens` list.
-# special_tokens:
-#   # bos_token: "<s>"
-#   # eos_token: "</s>"
-#   # unk_token: "<unk>"
-
-# # Add extra tokens.
-# tokens:
-
-# # FSDP
-# fsdp:
-# fsdp_config:
-
-# # Deepspeed config path. e.g., deepspeed/zero3.json
-# deepspeed:
-
-# # Advanced DDP Arguments
-# ddp_timeout:
-# ddp_bucket_cap_mb:
-# ddp_broadcast_buffers:
-
-# # Path to torch distx for optim 'adamw_anyprecision'
-# torchdistx_path:
-
-# # Set to HF dataset for type: 'completion' for streaming instead of pre-tokenize
-# pretraining_dataset:
-
-# # Debug mode
-# debug:
-
-# # Seed
-# seed:
-
-# # Allow overwrite yml config using from cli
-# strict:
-
-base_model: ${BASE_MODEL}
-base_model_ignore_patterns: ${BASE_MODEL_IGNORE_PATTERNS}
-base_model_config: ${BASE_MODEL_CONFIG}
-revision_of_model: ${REVISION_OF_MODEL}
-tokenizer_config: ${TOKENIZER_CONFIG}
-model_type: ${MODEL_TYPE}
-tokenizer_type: ${TOKENIZER_TYPE}
-trust_remote_code: ${TRUST_REMOTE_CODE}
-tokenizer_use_fast: ${TOKENIZER_USE_FAST}
-tokenizer_legacy: ${TOKENIZER_LEGACY}
-resize_token_embeddings_to_32x: ${RESIZE_TOKEN_EMBEDDINGS_TO_32X}
-
-is_falcon_derived_model: ${IS_FALCON_DERIVED_MODEL}
-is_llama_derived_model: ${IS_LLAMA_DERIVED_MODEL}
-is_qwen_derived_model: ${IS_QWEN_DERIVED_MODEL}
-is_mistral_derived_model: ${IS_MISTRAL_DERIVED_MODEL}
-
-overrides_of_model_config:
-  rope_scaling:
-    type: ${ROPE_SCALING_TYPE}
-    factor: ${ROPE_SCALING_FACTOR}
-
-bnb_config_kwargs:
-  llm_int8_has_fp16_weight: ${BNB_LLM_INT8_HAS_FP16_WEIGHT}
-  bnb_4bit_quant_type: ${BNB_4BIT_QUANT_TYPE}
-  bnb_4bit_use_double_quant: ${BNB_4BIT_USE_DOUBLE_QUANT}
-
-gptq: ${GPTQ}
-load_in_8bit: ${LOAD_IN_8BIT}
-load_in_4bit: ${LOAD_IN_4BIT}
-bf16: ${BF16}
-fp16: ${FP16}
-tf32: ${TF32}
-bfloat16: ${BFLOAT16}
-float16: ${FLOAT16}
-
-gpu_memory_limit: ${GPU_MEMORY_LIMIT}
-lora_on_cpu: ${LORA_ON_CPU}
-
-datasets:
-  - path: ${DATASET_PATH}
-    type: ${DATASET_TYPE}
-    ds_type: ${DATASET_DS_TYPE}
-    data_files: ${DATASET_DATA_FILES}
-    shards: ${DATASET_SHARDS}
-    name: ${DATASET_NAME}
-    train_on_split: ${DATASET_TRAIN_ON_SPLIT}
-    revision: ${DATASET_REVISION}
-    trust_remote_code: ${DATASET_TRUST_REMOTE_CODE}
-
-rl: ${RL}
-dpo_use_weighting: ${DPO_USE_WEIGHTING}
-
-chat_template: ${CHAT_TEMPLATE}
-chat_template_jinja: ${CHAT_TEMPLATE_JINJA}
-default_system_message: ${DEFAULT_SYSTEM_MESSAGE}
-dataset_prepared_path: ${DATASET_PREPARED_PATH}
-push_dataset_to_hub: ${PUSH_DATASET_TO_HUB}
-dataset_num_proc: ${DATASET_NUM_PROC}
-dataset_keep_in_memory: ${DATASET_KEEP_IN_MEMORY}
-hub_model_id: ${HUB_MODEL_ID}
-hub_strategy: ${HUB_STRATEGY}
-hf_use_auth_token: ${HF_USE_AUTH_TOKEN}
-val_set_size: ${VAL_SET_SIZE}
-dataset_shard_num: ${DATASET_SHARD_NUM}
-dataset_shard_idx: ${DATASET_SHARD_IDX}
-
-sequence_len: ${SEQUENCE_LEN}
-pad_to_sequence_len: ${PAD_TO_SEQUENCE_LEN}
-sample_packing: ${SAMPLE_PACKING}
-eval_sample_packing: ${EVAL_SAMPLE_PACKING}
-sample_packing_eff_est: ${SAMPLE_PACKING_EFF_EST}
-total_num_tokens: ${TOTAL_NUM_TOKENS}
-sample_packing_group_size: ${SAMPLE_PACKING_GROUP_SIZE}
-sample_packing_bin_size: ${SAMPLE_PACKING_BIN_SIZE}
-
-batch_flattening: ${BATCH_FLATTENING}
-device_map: ${DEVICE_MAP}
-max_memory: ${MAX_MEMORY}
-
-adapter: ${ADAPTER}
-lora_model_dir: ${LORA_MODEL_DIR}
-
-lora_r: ${LORA_R}
-lora_alpha: ${LORA_ALPHA}
-lora_dropout: ${LORA_DROPOUT}
-lora_target_modules:
-  - ${LORA_TARGET_MODULES}
-lora_target_linear: ${LORA_TARGET_LINEAR}
-peft_layers_to_transform: ${PEFT_LAYERS_TO_TRANSFORM}
-lora_modules_to_save: ${LORA_MODULES_TO_SAVE}
-lora_fan_in_fan_out: ${LORA_FAN_IN_FAN_OUT}
-
-loraplus_lr_ratio: ${LORAPLUS_LR_RATIO}
-loraplus_lr_embedding: ${LORAPLUS_LR_EMBEDDING}
-
-peft:
-  loftq_config:
-    loftq_bits: ${LOFTQ_BITS}
-
-relora_steps: ${RELORA_STEPS}
-relora_warmup_steps: ${RELORA_WARMUP_STEPS}
-relora_anneal_steps: ${RELORA_ANNEAL_STEPS}
-relora_prune_ratio: ${RELORA_PRUNE_RATIO}
-relora_cpu_offload: ${RELORA_CPU_OFFLOAD}
-
-wandb_mode: ${WANDB_MODE}
-wandb_project: ${WANDB_PROJECT}
-wandb_entity: ${WANDB_ENTITY}
-wandb_watch: ${WANDB_WATCH}
-wandb_name: ${WANDB_NAME}
-wandb_run_id: ${WANDB_RUN_ID}
-wandb_log_model: ${WANDB_LOG_MODEL}
-
-mlflow_tracking_uri: ${MLFLOW_TRACKING_URI}
-mlflow_experiment_name: ${MLFLOW_EXPERIMENT_NAME}
-mlflow_run_name: ${MLFLOW_RUN_NAME}
-hf_mlflow_log_artifacts: ${HF_MLFLOW_LOG_ARTIFACTS}
-
-use_comet: ${USE_COMET}
-comet_api_key: ${COMET_API_KEY}
-comet_workspace: ${COMET_WORKSPACE}
-comet_project_name: ${COMET_PROJECT_NAME}
-comet_experiment_key: ${COMET_EXPERIMENT_KEY}
-comet_mode: ${COMET_MODE}
-comet_online: ${COMET_ONLINE}
-comet_experiment_config: ${COMET_EXPERIMENT_CONFIG}
-
-output_dir: ${OUTPUT_DIR}
-
-torch_compile: ${TORCH_COMPILE}
-torch_compile_backend: ${TORCH_COMPILE_BACKEND}
-
-gradient_accumulation_steps: ${GRADIENT_ACCUMULATION_STEPS}
-micro_batch_size: ${MICRO_BATCH_SIZE}
-eval_batch_size: ${EVAL_BATCH_SIZE}
-num_epochs: ${NUM_EPOCHS}
-warmup_steps: ${WARMUP_STEPS}
-warmup_ratio: ${WARMUP_RATIO}
-learning_rate: ${LEARNING_RATE}
-lr_quadratic_warmup: ${LR_QUADRATIC_WARMUP}
-logging_steps: ${LOGGING_STEPS}
-eval_steps: ${EVAL_STEPS}
-evals_per_epoch: ${EVALS_PER_EPOCH}
-save_strategy: ${SAVE_STRATEGY}
-save_steps: ${SAVE_STEPS}
-saves_per_epoch: ${SAVES_PER_EPOCH}
-save_total_limit: ${SAVE_TOTAL_LIMIT}
-max_steps: ${MAX_STEPS}
-
-eval_table_size: ${EVAL_TABLE_SIZE}
-eval_max_new_tokens: ${EVAL_MAX_NEW_TOKENS}
-eval_causal_lm_metrics: ${EVAL_CAUSAL_LM_METRICS}
-
-profiler_steps: ${PROFILER_STEPS}
-loss_watchdog_threshold: ${LOSS_WATCHDOG_THRESHOLD}
-loss_watchdog_patience: ${LOSS_WATCHDOG_PATIENCE}
-
-train_on_inputs: ${TRAIN_ON_INPUTS}
-group_by_length: ${GROUP_BY_LENGTH}
-gradient_checkpointing: ${GRADIENT_CHECKPOINTING}
-early_stopping_patience: ${EARLY_STOPPING_PATIENCE}
-
-lr_scheduler: ${LR_SCHEDULER}
-lr_scheduler_kwargs: ${LR_SCHEDULER_KWARGS}
-cosine_min_lr_ratio: ${COSINE_MIN_LR_RATIO}
-cosine_constant_lr_ratio: ${COSINE_CONSTANT_LR_RATIO}
-lr_div_factor: ${LR_DIV_FACTOR}
-
-optimizer: ${OPTIMIZER}
-optim_args: ${OPTIM_ARGS}
-optim_target_modules: ${OPTIM_TARGET_MODULES}
-weight_decay: ${WEIGHT_DECAY}
-adam_beta1: ${ADAM_BETA1}
-adam_beta2: ${ADAM_BETA2}
-adam_epsilon: ${ADAM_EPSILON}
-max_grad_norm: ${MAX_GRAD_NORM}
-
-neftune_noise_alpha: ${NEFTUNE_NOISE_ALPHA}
-
-flash_optimum: ${FLASH_OPTIMUM}
-xformers_attention: ${XFORMERS_ATTENTION}
-flash_attention: ${FLASH_ATTENTION}
-flash_attn_cross_entropy: ${FLASH_ATTN_CROSS_ENTROPY}
-flash_attn_rms_norm: ${FLASH_ATTN_RMS_NORM}
-flash_attn_fuse_mlp: ${FLASH_ATTN_FUSE_MLP}
-sdp_attention: ${SDP_ATTENTION}
-s2_attention: ${S2_ATTENTION}
-resume_from_checkpoint: ${RESUME_FROM_CHECKPOINT}
-auto_resume_from_checkpoints: ${AUTO_RESUME_FROM_CHECKPOINTS}
-
-local_rank: ${LOCAL_RANK}
-
-special_tokens:
-  bos_token: ${SPECIAL_TOKEN_BOS}
-  eos_token: ${SPECIAL_TOKEN_EOS}
-  unk_token: ${SPECIAL_TOKEN_UNK}
-  pad_token: ${SPECIAL_TOKEN_PAD}
-
-tokens: ${TOKENS}
-
-fsdp: ${FSDP}
-fsdp_config: ${FSDP_CONFIG}
-deepspeed: ${DEEPSPEED}
-
-ddp_timeout: ${DDP_TIMEOUT}
-ddp_bucket_cap_mb: ${DDP_BUCKET_CAP_MB}
-ddp_broadcast_buffers: ${DDP_BROADCAST_BUFFERS}
-
-torchdistx_path: ${TORCHDISTX_PATH}
-pretraining_dataset: ${PRETRAINING_DATASET}
-debug: ${DEBUG}
-seed: ${SEED}
-strict: ${STRICT}

.runpod/src/handler.py

@@ -1,66 +0,0 @@
-"""
-Runpod serverless entrypoint handler
-"""
-
-import os
-
-import runpod
-import yaml
-from huggingface_hub._login import login
-from train import train
-from utils import get_output_dir
-
-BASE_VOLUME = os.environ.get("BASE_VOLUME", "/runpod-volume")
-if not os.path.exists(BASE_VOLUME):
-    os.makedirs(BASE_VOLUME)
-
-logger = runpod.RunPodLogger()
-
-
-async def handler(job):
-    runpod_job_id = job["id"]
-    inputs = job["input"]
-    run_id = inputs.get("run_id", "default_run_id")
-    args = inputs.get("args", {})
-
-    # Set output directory
-    output_dir = os.path.join(BASE_VOLUME, get_output_dir(run_id))
-    args["output_dir"] = output_dir
-
-    # First save args to a temporary config file
-    config_path = "/workspace/test_config.yaml"
-
-    # Add run_name and job_id to args before saving
-    args["run_name"] = run_id
-    args["runpod_job_id"] = runpod_job_id
-
-    yaml_data = yaml.dump(args, default_flow_style=False)
-    with open(config_path, "w", encoding="utf-8") as file:
-        file.write(yaml_data)
-
-    # Handle credentials
-    credentials = inputs.get("credentials", {})
-
-    if "wandb_api_key" in credentials:
-        os.environ["WANDB_API_KEY"] = credentials["wandb_api_key"]
-    if "hf_token" in credentials:
-        os.environ["HF_TOKEN"] = credentials["hf_token"]
-
-    if os.environ.get("HF_TOKEN"):
-        login(token=os.environ["HF_TOKEN"])
-    else:
-        logger.info("No HF_TOKEN provided. Skipping login.")
-
-    logger.info("Starting Training.")
-    async for result in train(config_path):  # Pass the config path instead of args
-        logger.info(result)
-    logger.info("Training Complete.")
-
-    # Cleanup
-    if "WANDB_API_KEY" in os.environ:
-        del os.environ["WANDB_API_KEY"]
-    if "HF_TOKEN" in os.environ:
-        del os.environ["HF_TOKEN"]
-
-
-runpod.serverless.start({"handler": handler, "return_aggregate_stream": True})

.runpod/src/test_input.json

@@ -1,61 +0,0 @@
-{
-  "input": {
-    "user_id": "user",
-    "model_id": "llama-test",
-    "run_id": "llama-test",
-    "credentials": {
-      "wandb_api_key": "",
-      "hf_token": ""
-    },
-    "args": {
-      "base_model": "NousResearch/Meta-Llama-3-8B",
-      "model_type": "LlamaForCausalLM",
-      "tokenizer_type": "AutoTokenizer",
-      "load_in_8bit": true,
-      "load_in_4bit": false,
-      "strict": false,
-      "datasets": [
-        {
-          "path": "mhenrichsen/alpaca_2k_test",
-          "type": "alpaca"
-        }
-      ],
-      "val_set_size": 0.05,
-      "output_dir": "./outputs/lora-out",
-      "sequence_len": 4096,
-      "sample_packing": true,
-      "eval_sample_packing": false,
-      "pad_to_sequence_len": true,
-      "adapter": "lora",
-      "lora_r": 32,
-      "lora_alpha": 16,
-      "lora_dropout": 0.05,
-      "lora_target_linear": true,
-      "lora_modules_to_save": [
-        "embed_tokens",
-        "lm_head"
-      ],
-      "gradient_accumulation_steps": 4,
-      "micro_batch_size": 2,
-      "num_epochs": 1,
-      "optimizer": "adamw_bnb_8bit",
-      "lr_scheduler": "cosine",
-      "learning_rate": 0.0002,
-      "train_on_inputs": false,
-      "group_by_length": false,
-      "bf16": "auto",
-      "tf32": false,
-      "gradient_checkpointing": true,
-      "logging_steps": 1,
-      "flash_attention": true,
-      "warmup_steps": 1,
-      "evals_per_epoch": 1,
-      "eval_max_new_tokens": 128,
-      "saves_per_epoch": 1,
-      "weight_decay": 0.0,
-      "special_tokens": {
-        "pad_token": "<|end_of_text|>"
-      }
-    }
-  }
-}

.runpod/src/train.py

@@ -1,45 +0,0 @@
-"""
-Runpod train entrypoint
-"""
-
-import asyncio
-
-
-async def train(config_path: str, gpu_id: str = "0", preprocess: bool = True):
-    """
-    Run preprocessing (if enabled) and training with the given config file
-    :param config_path: Path to the YAML config file
-    :param gpu_id: GPU ID to use (default: "0")
-    :param preprocess: Whether to run preprocessing (default: True)
-
-    """
-    # First check if preprocessing is needed
-    if preprocess:
-        # Preprocess command
-        preprocess_cmd = (
-            f"CUDA_VISIBLE_DEVICES={gpu_id} axolotl preprocess {config_path}"
-        )
-        process = await asyncio.create_subprocess_shell(
-            preprocess_cmd,
-            stdout=asyncio.subprocess.PIPE,
-            stderr=asyncio.subprocess.STDOUT,
-        )
-
-        if process.stdout is not None:
-            async for line in process.stdout:
-                yield f"Preprocessing: {line.decode().strip()}"
-        await process.wait()
-        yield "Preprocessing completed."
-    else:
-        yield "Skipping preprocessing step."
-
-    # Training command
-    train_cmd = f"axolotl train {config_path}"
-    process = await asyncio.create_subprocess_shell(
-        train_cmd, stdout=asyncio.subprocess.PIPE, stderr=asyncio.subprocess.STDOUT
-    )
-
-    if process.stdout is not None:
-        async for line in process.stdout:
-            yield f"Training: {line.decode().strip()}"
-    await process.wait()

.runpod/src/utils.py

@@ -1,89 +0,0 @@
-"""
-Runpod launcher utils
-"""
-
-import os
-
-import yaml
-
-
-def get_output_dir(run_id):
-    path = f"fine-tuning/{run_id}"
-    return path
-
-
-def make_valid_config(input_args):
-    """
-    Creates and saves updated config file, returns the path to the new config
-    :param input_args: dict of input args
-    :return: str, path to the updated config file
-    """
-    # Load default config
-    with open("config/config.yaml", "r", encoding="utf-8") as fin:
-        all_args = yaml.safe_load(fin)
-
-    if not input_args:
-        print("No args provided, using defaults")
-    else:
-        all_args.update(input_args)
-
-    # Create updated config path
-    updated_config_path = "config/updated_config.yaml"
-
-    # Save updated config to new file
-    with open(updated_config_path, "w", encoding="utf-8") as f:
-        yaml.dump(all_args, f)
-
-    return updated_config_path
-
-
-def set_config_env_vars(args: dict):
-    """
-    Convert API arguments into environment variables.
-    Handles nested dictionaries, lists, and special values.
-
-    Args:
-        args (dict): The arguments dictionary from the API request
-    """
-
-    def process_value(value):
-        """Convert Python values to string format for environment variables"""
-        if value is None:
-            return ""
-        if isinstance(value, bool):
-            return str(value).lower()
-        if isinstance(value, (list, dict)):
-            return str(value)
-        return str(value)
-
-    def set_env_vars(data, prefix=""):
-        """Recursively set environment variables from nested dictionary"""
-        for key, value in data.items():
-            env_key = prefix + key.upper()
-
-            # Handle special cases
-            if isinstance(value, dict):
-                # For nested dictionaries (like special_tokens)
-                set_env_vars(value, f"{env_key}_")
-            elif isinstance(value, list):
-                # Handle list of dictionaries (like datasets)
-                if value and isinstance(value[0], dict):
-                    for i, item in enumerate(value):
-                        set_env_vars(item, f"{env_key}_{i}_")
-                else:
-                    # For simple lists (like lora_target_modules)
-                    os.environ[env_key] = process_value(value)
-            else:
-                # Handle all other cases
-                os.environ[env_key] = process_value(value)
-
-    # Clear any existing related environment variables
-    # This prevents old values from persisting
-    for key in list(os.environ.keys()):
-        if key.startswith(
-            ("BASE_MODEL", "MODEL_TYPE", "TOKENIZER_TYPE", "DATASET", "LORA_", "WANDB_")
-        ):
-            del os.environ[key]
-
-    # Set new environment variables
-    set_env_vars(args)

.runpod/test-input.json

@@ -1,86 +0,0 @@
-{
-  "input": {
-    "name": "quick_smoke_test_sft",
-    "user_id": "user",
-    "model_id": "llama-test",
-    "run_id": "llama-test",
-    "credentials": {
-      "wandb_api_key": "",
-      "hf_token": ""
-    },
-    "args": {
-      "base_model": "HuggingFaceTB/SmolLM2-135M",
-      "model_type": "AutoModelForCausalLM",
-      "tokenizer_type": "AutoTokenizer",
-      "load_in_4bit": true,
-      "strict": false,
-      "datasets": [
-        {
-          "path": "mhenrichsen/alpaca_2k_test",
-          "type": "alpaca",
-          "split": "train[:10%]"
-        }
-      ],
-      "val_set_size": 0.02,
-      "output_dir": "./outputs/lora-out",
-      "sequence_len": 4096,
-      "sample_packing": true,
-      "eval_sample_packing": false,
-      "pad_to_sequence_len": true,
-      "adapter": "qlora",
-      "lora_r": 32,
-      "lora_alpha": 64,
-      "lora_dropout": 0.05,
-      "lora_target_linear": true,
-      "lora_modules_to_save": [
-        "embed_tokens",
-        "lm_head"
-      ],
-      "gradient_accumulation_steps": 2,
-      "micro_batch_size": 1,
-      "num_epochs": 1,
-      "optimizer": "adamw_torch_fused",
-      "lr_scheduler": "cosine",
-      "learning_rate": 0.0002,
-      "train_on_inputs": false,
-      "group_by_length": false,
-      "bf16": "auto",
-      "tf32": true,
-      "gradient_checkpointing": true,
-      "logging_steps": 1,
-      "flash_attention": true,
-      "warmup_steps": 1,
-      "evals_per_epoch": 1,
-      "eval_max_new_tokens": 128,
-      "saves_per_epoch": 1,
-      "weight_decay": 0.0,
-      "special_tokens": {
-        "pad_token": "<|endoftext|>"
-      },
-      "max_steps": 20
-    },
-    "timeout": 100000
-  },
-  "config": {
-    "gpuTypeId": "NVIDIA GeForce RTX 4090",
-    "gpuCount": 1,
-    "containerDiskInGb": 200,
-    "env": [
-      {
-        "key": "TOKENIZER",
-        "value": ""
-      },
-      {
-        "key": "DISABLE_LOG_STATS",
-        "value": "true"
-      }
-    ],
-    "allowedCudaVersions": [
-      "12.8",
-      "12.7",
-      "12.6",
-      "12.5",
-      "12.4"
-    ]
-  }
-}

.runpod/tests.json

@@ -1,90 +0,0 @@
-{
-  "tests": [
-    {
-      "name": "quick_smoke_test_sft",
-      "input": {
-        "user_id": "user",
-        "model_id": "llama-test",
-        "run_id": "llama-test",
-        "credentials": {
-          "wandb_api_key": "",
-          "hf_token": ""
-        },
-        "args": {
-          "base_model": "HuggingFaceTB/SmolLM2-135M",
-          "model_type": "AutoModelForCausalLM",
-          "tokenizer_type": "AutoTokenizer",
-          "load_in_4bit": true,
-          "strict": false,
-          "datasets": [
-            {
-              "path": "mhenrichsen/alpaca_2k_test",
-              "type": "alpaca",
-              "split": "train[:10%]"
-            }
-          ],
-          "val_set_size": 0.02,
-          "output_dir": "./outputs/lora-out",
-          "sequence_len": 4096,
-          "sample_packing": true,
-          "eval_sample_packing": false,
-          "pad_to_sequence_len": true,
-          "adapter": "qlora",
-          "lora_r": 32,
-          "lora_alpha": 64,
-          "lora_dropout": 0.05,
-          "lora_target_linear": true,
-          "lora_modules_to_save": [
-            "embed_tokens",
-            "lm_head"
-          ],
-          "gradient_accumulation_steps": 2,
-          "micro_batch_size": 1,
-          "num_epochs": 1,
-          "optimizer": "adamw_torch_fused",
-          "lr_scheduler": "cosine",
-          "learning_rate": 0.0002,
-          "train_on_inputs": false,
-          "group_by_length": false,
-          "bf16": "auto",
-          "tf32": true,
-          "gradient_checkpointing": true,
-          "logging_steps": 1,
-          "flash_attention": true,
-          "warmup_steps": 1,
-          "evals_per_epoch": 1,
-          "eval_max_new_tokens": 128,
-          "saves_per_epoch": 1,
-          "weight_decay": 0.0,
-          "special_tokens": {
-            "pad_token": "<|endoftext|>"
-          },
-          "max_steps": 20
-        }
-      },
-      "timeout": 100000
-    }
-  ],
-  "config": {
-    "gpuTypeId": "NVIDIA GeForce RTX 4090",
-    "gpuCount": 1,
-    "containerDiskInGb": 200,
-    "env": [
-      {
-        "key": "TOKENIZER",
-        "value": ""
-      },
-      {
-        "key": "DISABLE_LOG_STATS",
-        "value": "true"
-      }
-    ],
-    "allowedCudaVersions": [
-      "12.8",
-      "12.7",
-      "12.6",
-      "12.5",
-      "12.4"
-    ]
-  }
-}

AGENTS.md

@@ -1,99 +0,0 @@
-# Axolotl
-
-Fine-tuning framework for LLMs. Config-driven: every training run is defined by a single YAML file.
-
-## Tech Stack
-
-Python, PyTorch, HuggingFace Transformers, TRL, PEFT (LoRA/QLoRA), DeepSpeed, FSDP, vLLM (for GRPO generation).
-
-## Commands
-
-```bash
-axolotl train config.yaml              # Train (single or multi-GPU, auto-detected)
-axolotl preprocess config.yaml         # Tokenize dataset and validate config
-axolotl preprocess config.yaml --debug # Inspect tokenized samples and label masking
-axolotl inference config.yaml          # Interactive inference
-axolotl merge-lora config.yaml         # Merge LoRA adapter into base model
-axolotl vllm-serve config.yaml         # Start vLLM server for GRPO/EBFT training
-axolotl fetch examples                 # Download example configs
-axolotl agent-docs                     # Show agent-optimized docs (bundled with pip package)
-axolotl agent-docs grpo                # Topic-specific agent reference
-axolotl config-schema                  # Dump config JSON schema
-```
-
-## Training Methods
-
-| Method | Config Key | When to Use |
-|--------|-----------|-------------|
-| SFT | *(default)* | Input-output pairs, instruction tuning |
-| DPO/IPO | `rl: dpo` / `rl: dpo, dpo_loss_type: ["ipo"]` | Paired preference data (chosen vs rejected) |
-| KTO | `rl: kto` | Unpaired binary preference labels |
-| ORPO | `rl: orpo` | Single-stage alignment, no ref model |
-| GRPO | `rl: grpo` | RL with verifiable reward functions (math, code) |
-| EBFT | `rl: ebft` | Feature-matching rewards from internal representations |
-
-Agent-specific references:
-- [docs/agents/sft.md](docs/agents/sft.md) — supervised fine-tuning
-- [docs/agents/preference_tuning.md](docs/agents/preference_tuning.md) — DPO, IPO, KTO, ORPO, SimPO
-- [docs/agents/grpo.md](docs/agents/grpo.md) — GRPO online RL with reward functions
-- [docs/agents/reward_modelling.md](docs/agents/reward_modelling.md) — outcome and process reward models
-- [docs/agents/pretraining.md](docs/agents/pretraining.md) — continual pretraining
-- [docs/agents/model_architectures.md](docs/agents/model_architectures.md) — model-specific quirks (Gemma4, Qwen3.5 MoE, etc.)
-- [docs/agents/new_model_support.md](docs/agents/new_model_support.md) — debugging and adding support for new model architectures
-
-## Config Pattern
-
-All training is config-driven. A YAML file specifies model, adapter, dataset(s), and hyperparameters:
-
-```yaml
-base_model: meta-llama/Llama-3.1-8B-Instruct
-adapter: lora                    # or qlora, or omit for full fine-tune
-datasets:
-  - path: my_dataset
-    type: chat_template          # prompt strategy (see docs/dataset-formats/)
-output_dir: ./outputs/lora-out
-```
-
-Config schema: `src/axolotl/utils/schemas/config.py` (AxolotlInputConfig).
-
-## Project Structure
-
-```
-src/axolotl/
-  cli/                           # CLI entry points (train, preprocess, inference, merge_lora, vllm_serve)
-  core/
-    builders/                    # TrainerBuilder classes (causal.py for SFT, rl.py for RLHF)
-    trainers/                    # Trainer classes, mixins (optimizer, scheduler, packing)
-      dpo/                       # DPO trainer and config
-      grpo/                      # GRPO trainer and sampler
-  loaders/                       # Model, tokenizer, adapter, processor loading
-  prompt_strategies/             # Dataset format handlers (chat_template, alpaca, dpo/, kto/, orpo/)
-  utils/schemas/                 # Pydantic config schemas (config, model, training, peft, trl, fsdp)
-  integrations/                  # Plugins (liger, cut_cross_entropy, swanlab, nemo_gym)
-  monkeypatch/                   # Runtime patches for HF transformers
-
-examples/                        # Example YAML configs by model (llama-3/, qwen2/, mistral/, ebft/)
-deepspeed_configs/               # DeepSpeed JSON configs (zero2, zero3)
-docs/                            # Quarto documentation site
-```
-
-## Code Conventions
-
-- Config-driven: features are toggled via YAML, not code changes
-- Prompt strategies: `src/axolotl/prompt_strategies/` — each `type:` value maps to a function
-- Plugin system: `plugins:` list in config loads integration modules
-- Trainer mixins: `core/trainers/mixins/` for composable trainer behaviors
-- Schemas: all config validation via Pydantic in `utils/schemas/`
-
-## Key Documentation
-
-- [Getting Started](docs/getting-started.qmd) — quickstart tutorial
-- [Choosing a Method](docs/choosing_method.qmd) — SFT vs DPO vs GRPO decision guide
-- [Config Reference](docs/config-reference.qmd) — all config options
-- [Dataset Formats](docs/dataset-formats/) — chat_template, alpaca, input_output, completion
-- [RLHF](docs/rlhf.qmd) — DPO, KTO, ORPO, GRPO, EBFT configs and dataset formats
-- [GRPO Deep Dive](docs/grpo.qmd) — async training, custom rewards, scaling
-- [vLLM Serving](docs/vllm_serving.qmd) — vLLM setup for GRPO/EBFT
-- [Multi-GPU](docs/multi-gpu.qmd) — FSDP and DeepSpeed
-- [Training Stability](docs/training_stability.qmd) — debugging loss, NaN, OOM
-- [Debugging](docs/debugging.qmd) — VSCode setup, Docker debugging

CITATION.cff

@@ -1,10 +0,0 @@
-cff-version: 1.2.0
-type: software
-title: "Axolotl: Open Source LLM Post-Training"
-message: "If you use this software, please cite it as below."
-authors:
-  - name: "Axolotl maintainers and contributors"
-repository-code: "https://github.com/axolotl-ai-cloud/axolotl"
-url: "https://axolotl.ai/"
-license: Apache-2.0
-date-released: "2023-05-30"

CNAME

@@ -1 +0,0 @@
-docs.axolotl.ai

MANIFEST.in

@@ -1,7 +1,4 @@
+include requirements.txt
 include README.md
 include LICENSE
-include VERSION
-include src/axolotl/utils/chat_templates/templates/*.jinja
-include AGENTS.md
-recursive-include docs/agents *.md
 recursive-include axolotl *.py

README.md

@@ -1,232 +1,768 @@
 <p align="center">
     <picture>
-        <source media="(prefers-color-scheme: dark)" srcset="https://raw.githubusercontent.com/axolotl-ai-cloud/axolotl/887513285d98132142bf5db2a74eb5e0928787f1/image/axolotl_logo_digital_white.svg">
-        <source media="(prefers-color-scheme: light)" srcset="https://raw.githubusercontent.com/axolotl-ai-cloud/axolotl/887513285d98132142bf5db2a74eb5e0928787f1/image/axolotl_logo_digital_black.svg">
-        <img alt="Axolotl" src="https://raw.githubusercontent.com/axolotl-ai-cloud/axolotl/887513285d98132142bf5db2a74eb5e0928787f1/image/axolotl_logo_digital_black.svg" width="400" height="104" style="max-width: 100%;">
+        <source media="(prefers-color-scheme: dark)" srcset="image/axolotl_logo_digital_white.svg">
+        <source media="(prefers-color-scheme: light)" srcset="image/axolotl_logo_digital_black.svg">
+        <img alt="Axolotl" src="image/axolotl_logo_digital_black.svg" width="400" height="104" style="max-width: 100%;">
     </picture>
 </p>
-  <p align="center">
-      <strong>A Free and Open Source LLM Fine-tuning Framework</strong><br>
-  </p>
 
 <p align="center">
     <img src="https://img.shields.io/github/license/axolotl-ai-cloud/axolotl.svg?color=blue" alt="GitHub License">
     <img src="https://github.com/axolotl-ai-cloud/axolotl/actions/workflows/tests.yml/badge.svg" alt="tests">
-    <a href="https://codecov.io/gh/axolotl-ai-cloud/axolotl"><img src="https://codecov.io/gh/axolotl-ai-cloud/axolotl/branch/main/graph/badge.svg" alt="codecov"></a>
     <a href="https://github.com/axolotl-ai-cloud/axolotl/releases"><img src="https://img.shields.io/github/release/axolotl-ai-cloud/axolotl.svg" alt="Releases"></a>
-    <br/>
-    <a href="https://github.com/axolotl-ai-cloud/axolotl/graphs/contributors"><img src="https://img.shields.io/github/contributors-anon/axolotl-ai-cloud/axolotl?color=yellow&style=flat-square" alt="contributors" style="height: 20px;"></a>
     <img src="https://img.shields.io/github/stars/axolotl-ai-cloud/axolotl" alt="GitHub Repo stars">
-    <br/>
-    <a href="https://discord.com/invite/HhrNrHJPRb"><img src="https://img.shields.io/badge/discord-7289da.svg?style=flat-square&logo=discord" alt="discord" style="height: 20px;"></a>
-    <a href="https://twitter.com/axolotl_ai"><img src="https://img.shields.io/twitter/follow/axolotl_ai?style=social" alt="twitter" style="height: 20px;"></a>
-    <a href="https://colab.research.google.com/github/axolotl-ai-cloud/axolotl/blob/main/examples/colab-notebooks/colab-axolotl-example.ipynb"><img src="https://colab.research.google.com/assets/colab-badge.svg" alt="google-colab" style="height: 20px;"></a>
-    <br/>
+</p>
+<p align="center">
     <img src="https://github.com/axolotl-ai-cloud/axolotl/actions/workflows/tests-nightly.yml/badge.svg" alt="tests-nightly">
     <img src="https://github.com/axolotl-ai-cloud/axolotl/actions/workflows/multi-gpu-e2e.yml/badge.svg" alt="multigpu-semi-weekly tests">
 </p>
 
-
-## 🎉 Latest Updates
-
-- 2026/04:
-  - New model support has been added in Axolotl for [Mistral Medium 3.5](https://github.com/axolotl-ai-cloud/axolotl/tree/main/examples/mistral-medium-3_5) and [Gemma 4](https://github.com/axolotl-ai-cloud/axolotl/tree/main/examples/gemma4).
-  - Axolotl is now [uv-first](https://github.com/axolotl-ai-cloud/axolotl/pull/3545) and has [SonicMoE fused LoRA](https://github.com/axolotl-ai-cloud/axolotl/pull/3519) support.
-- 2026/03:
-  - New model support has been added in Axolotl for [Mistral Small 4](https://github.com/axolotl-ai-cloud/axolotl/tree/main/examples/mistral4), [Qwen3.5, Qwen3.5 MoE](https://github.com/axolotl-ai-cloud/axolotl/tree/main/examples/qwen3.5), [GLM-4.7-Flash](https://github.com/axolotl-ai-cloud/axolotl/tree/main/examples/glm47-flash), [GLM-4.6V](https://github.com/axolotl-ai-cloud/axolotl/tree/main/examples/glm46v), and [GLM-4.5-Air](https://github.com/axolotl-ai-cloud/axolotl/tree/main/examples/glm45).
-  - [MoE expert quantization](https://docs.axolotl.ai/docs/expert_quantization.html) support (via `quantize_moe_experts: true`) greatly reduces VRAM when training MoE models (FSDP2 compat).
-- 2026/02:
-  - [ScatterMoE LoRA](https://github.com/axolotl-ai-cloud/axolotl/pull/3410) support. LoRA fine-tuning directly on MoE expert weights using custom Triton kernels.
-  - Axolotl now has support for [SageAttention](https://github.com/axolotl-ai-cloud/axolotl/pull/2823) and [GDPO](https://github.com/axolotl-ai-cloud/axolotl/pull/3353) (Generalized DPO).
-- 2026/01:
-  - New integration for [EAFT](https://github.com/axolotl-ai-cloud/axolotl/pull/3366) (Entropy-Aware Focal Training), weights loss by entropy of the top-k logit distribution, and [Scalable Softmax](https://github.com/axolotl-ai-cloud/axolotl/pull/3338), improves long context in attention.
-- 2025/12:
-  - Axolotl now includes support for [Kimi-Linear](https://docs.axolotl.ai/docs/models/kimi-linear.html), [Plano-Orchestrator](https://docs.axolotl.ai/docs/models/plano.html), [MiMo](https://docs.axolotl.ai/docs/models/mimo.html), [InternVL 3.5](https://docs.axolotl.ai/docs/models/internvl3_5.html), [Olmo3](https://docs.axolotl.ai/docs/models/olmo3.html), [Trinity](https://docs.axolotl.ai/docs/models/trinity.html), and [Ministral3](https://docs.axolotl.ai/docs/models/ministral3.html).
-  - [Distributed Muon Optimizer](https://github.com/axolotl-ai-cloud/axolotl/pull/3264) support has been added for FSDP2 pretraining.
-- 2025/10: New model support has been added in Axolotl for: [Qwen3 Next](https://docs.axolotl.ai/docs/models/qwen3-next.html), [Qwen2.5-vl, Qwen3-vl](https://github.com/axolotl-ai-cloud/axolotl/tree/main/examples/qwen2_5-vl), [Qwen3, Qwen3MoE](https://docs.axolotl.ai/docs/models/qwen3.html), [Granite 4](https://docs.axolotl.ai/docs/models/granite4.html), [HunYuan](https://docs.axolotl.ai/docs/models/hunyuan.html), [Magistral 2509](https://docs.axolotl.ai/docs/models/magistral/vision.html), [Apertus](https://docs.axolotl.ai/docs/models/apertus.html), and [Seed-OSS](https://docs.axolotl.ai/docs/models/seed-oss.html).
-
-<details>
-
-<summary>Expand older updates</summary>
-
-- 2025/09: Axolotl now has text diffusion training. Read more [here](https://github.com/axolotl-ai-cloud/axolotl/tree/main/src/axolotl/integrations/diffusion).
-- 2025/08: QAT has been updated to include NVFP4 support. See [PR](https://github.com/axolotl-ai-cloud/axolotl/pull/3107).
-- 2025/07:
-  - ND Parallelism support has been added into Axolotl. Compose Context Parallelism (CP), Tensor Parallelism (TP), and Fully Sharded Data Parallelism (FSDP) within a single node and across multiple nodes. Check out the [blog post](https://huggingface.co/blog/accelerate-nd-parallel) for more info.
-  - Axolotl adds more models: [GPT-OSS](https://docs.axolotl.ai/docs/models/gpt-oss.html), [Gemma 3n](https://docs.axolotl.ai/docs/models/gemma3n.html), [Liquid Foundation Model 2 (LFM2)](https://docs.axolotl.ai/docs/models/LiquidAI.html), and [Arcee Foundation Models (AFM)](https://docs.axolotl.ai/docs/models/arcee.html).
-  - FP8 finetuning with fp8 gather op is now possible in Axolotl via `torchao`. Get started [here](https://docs.axolotl.ai/docs/mixed_precision.html#sec-fp8)!
-  - [Voxtral](https://docs.axolotl.ai/docs/models/voxtral.html), [Magistral 1.1](https://docs.axolotl.ai/docs/models/magistral.html), and [Devstral](https://docs.axolotl.ai/docs/models/devstral.html) with mistral-common tokenizer support has been integrated in Axolotl!
-  - TiledMLP support for single-GPU to multi-GPU training with DDP, DeepSpeed and FSDP support has been added to support Arctic Long Sequence Training. (ALST). See [examples](https://github.com/axolotl-ai-cloud/axolotl/tree/main/examples/alst) for using ALST with Axolotl!
-- 2025/06: Magistral with mistral-common tokenizer support has been added to Axolotl. See [docs](https://docs.axolotl.ai/docs/models/magistral.html) to start training your own Magistral models with Axolotl!
-- 2025/05: Quantization Aware Training (QAT) support has been added to Axolotl. Explore the [docs](https://docs.axolotl.ai/docs/qat.html) to learn more!
-- 2025/04: Llama 4 support has been added in Axolotl. See [docs](https://docs.axolotl.ai/docs/models/llama-4.html) to start training your own Llama 4 models with Axolotl's linearized version!
-- 2025/03: Axolotl has implemented Sequence Parallelism (SP) support. Read the [blog](https://huggingface.co/blog/axolotl-ai-co/long-context-with-sequence-parallelism-in-axolotl) and [docs](https://docs.axolotl.ai/docs/sequence_parallelism.html) to learn how to scale your context length when fine-tuning.
-- 2025/03: (Beta) Fine-tuning Multimodal models is now supported in Axolotl. Check out the [docs](https://docs.axolotl.ai/docs/multimodal.html) to fine-tune your own!
-- 2025/02: Axolotl has added LoRA optimizations to reduce memory usage and improve training speed for LoRA and QLoRA in single GPU and multi-GPU training (DDP and DeepSpeed). Jump into the [docs](https://docs.axolotl.ai/docs/lora_optims.html) to give it a try.
-- 2025/02: Axolotl has added GRPO support. Dive into our [blog](https://huggingface.co/blog/axolotl-ai-co/training-llms-w-interpreter-feedback-wasm) and [GRPO example](https://github.com/axolotl-ai-cloud/grpo_code) and have some fun!
-- 2025/01: Axolotl has added Reward Modelling / Process Reward Modelling fine-tuning support. See [docs](https://docs.axolotl.ai/docs/reward_modelling.html).
-
-</details>
-
-## ✨ Overview
-
-Axolotl is a free and open-source tool designed to streamline post-training and fine-tuning for the latest large language models (LLMs).
+Axolotl is a tool designed to streamline the fine-tuning of various AI models, offering support for multiple configurations and architectures.
 
 Features:
+- Train various Huggingface models such as llama, pythia, falcon, mpt
+- Supports fullfinetune, lora, qlora, relora, and gptq
+- Customize configurations using a simple yaml file or CLI overwrite
+- Load different dataset formats, use custom formats, or bring your own tokenized datasets
+- Integrated with xformer, flash attention, [liger kernel](https://github.com/linkedin/Liger-Kernel), rope scaling, and multipacking
+- Works with single GPU or multiple GPUs via FSDP or Deepspeed
+- Easily run with Docker locally or on the cloud
+- Log results and optionally checkpoints to wandb, mlflow or Comet
+- And more!
 
-- **Multiple Model Support**: Train various models like GPT-OSS, LLaMA, Mistral, Mixtral, Pythia, and many more models available on the Hugging Face Hub.
-- **Multimodal Training**: Fine-tune vision-language models (VLMs) including LLaMA-Vision, Qwen2-VL, Pixtral, LLaVA, SmolVLM2, GLM-4.6V, InternVL 3.5, Gemma 3n, and audio models like Voxtral with image, video, and audio support.
-- **Training Methods**: Full fine-tuning, LoRA, QLoRA, GPTQ, QAT, Preference Tuning (DPO, IPO, KTO, ORPO), RL (GRPO, GDPO), and Reward Modelling (RM) / Process Reward Modelling (PRM).
-- **Easy Configuration**: Re-use a single YAML configuration file across the full fine-tuning pipeline: dataset preprocessing, training, evaluation, quantization, and inference.
-- **Performance Optimizations**: [Multipacking](https://docs.axolotl.ai/docs/multipack.html), [Flash Attention 2/3/4](https://docs.axolotl.ai/docs/attention.html#flash-attention), [Xformers](https://docs.axolotl.ai/docs/attention.html#xformers), [Flex Attention](https://docs.axolotl.ai/docs/attention.html#flex-attention), [SageAttention](https://docs.axolotl.ai/docs/attention.html#sageattention), [Liger Kernel](https://docs.axolotl.ai/docs/custom_integrations.html#liger-kernels), [Cut Cross Entropy](https://docs.axolotl.ai/docs/custom_integrations.html#cut-cross-entropy), [ScatterMoE](https://docs.axolotl.ai/docs/custom_integrations.html#kernels-integration), [Sequence Parallelism (SP)](https://docs.axolotl.ai/docs/sequence_parallelism.html), [LoRA optimizations](https://docs.axolotl.ai/docs/lora_optims.html), [Multi-GPU training (FSDP1, FSDP2, DeepSpeed)](https://docs.axolotl.ai/docs/multi-gpu.html), [Multi-node training (Torchrun, Ray)](https://docs.axolotl.ai/docs/multi-node.html), and many more!
-- **Flexible Dataset Handling**: Load from local, HuggingFace, and cloud (S3, Azure, GCP, OCI) datasets.
-- **Cloud Ready**: We ship [Docker images](https://hub.docker.com/u/axolotlai) and also [PyPI packages](https://pypi.org/project/axolotl/) for use on cloud platforms and local hardware.
+<a href="https://www.phorm.ai/query?projectId=e315ba4a-4e14-421f-ab05-38a1f9076f25">
+  <img alt="phorm.ai" src="https://img.shields.io/badge/Phorm-Ask_AI-%23F2777A.svg?&logo=data:image/svg+xml;base64,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">
+</a>
 
+<table>
+<tr>
+<td>
 
+## Table of Contents
+- [Axolotl](#axolotl)
+  - [Table of Contents](#table-of-contents)
+  - [Quickstart ⚡](#quickstart-)
+    - [Usage](#usage)
+  - [Badge ❤🏷️](#badge-️)
+  - [Contributing 🤝](#contributing-)
+  - [Sponsors 🤝❤](#sponsors-)
+  - [Axolotl supports](#axolotl-supports)
+  - [Advanced Setup](#advanced-setup)
+    - [Environment](#environment)
+      - [Docker](#docker)
+      - [Conda/Pip venv](#condapip-venv)
+      - [Cloud GPU](#cloud-gpu)
+      - [Bare Metal Cloud GPU](#bare-metal-cloud-gpu)
+        - [LambdaLabs](#lambdalabs)
+        - [GCP](#gcp)
+      - [Windows](#windows)
+      - [Mac](#mac)
+      - [Google Colab](#google-colab)
+      - [Launching on public clouds via SkyPilot](#launching-on-public-clouds-via-skypilot)
+      - [Launching on public clouds via dstack](#launching-on-public-clouds-via-dstack)
+    - [Dataset](#dataset)
+    - [Config](#config)
+      - [All Config Options](#all-config-options)
+    - [Train](#train)
+      - [Preprocess dataset](#preprocess-dataset)
+      - [Multi-GPU](#multi-gpu)
+        - [DeepSpeed](#deepspeed)
+        - [FSDP](#fsdp)
+        - [FSDP + QLoRA](#fsdp--qlora)
+        - [Weights \& Biases Logging](#weights--biases-logging)
+        - [Special Tokens](#special-tokens)
+      - [Liger Kernel](#liger-kernel)
+    - [Inference Playground](#inference-playground)
+    - [Merge LORA to base](#merge-lora-to-base)
+  - [Common Errors 🧰](#common-errors-)
+    - [Tokenization Mismatch b/w Inference \& Training](#tokenization-mismatch-bw-inference--training)
+  - [Debugging Axolotl](#debugging-axolotl)
+  - [Need help? 🙋](#need-help-)
 
-## 🚀 Quick Start - LLM Fine-tuning in Minutes
+</td>
+<td>
 
-**Requirements**:
+<div align="center">
+  <img src="image/axolotl_symbol_digital_white.svg" alt="axolotl" width="160">
+  <div>
+    <p>
+      <b>Axolotl provides a unified repository for fine-tuning <br />a variety of AI models with ease</b>
+    </p>
+    <p>
+      Go ahead and Axolotl questions!!
+    </p>
+    <img src="https://github.com/axolotl-ai-cloud/axolotl/actions/workflows/pre-commit.yml/badge.svg?branch=main" alt="pre-commit">
+    <img alt="PyTest Status" src="https://github.com/axolotl-ai-cloud/axolotl/actions/workflows/tests.yml/badge.svg?branch=main">
+  </div>
+</div>
 
-- NVIDIA GPU (Ampere or newer for `bf16` and Flash Attention) or AMD GPU
-- Python >=3.11 (3.12 recommended)
-- PyTorch ≥2.9.1
+</td>
+</tr>
+</table>
 
-### Google Colab
+## Quickstart ⚡
 
-[![Open In Colab](https://colab.research.google.com/assets/colab-badge.svg)](https://colab.research.google.com/github/axolotl-ai-cloud/axolotl/blob/main/examples/colab-notebooks/colab-axolotl-example.ipynb#scrollTo=msOCO4NRmRLa)
+Get started with Axolotl in just a few steps! This quickstart guide will walk you through setting up and running a basic fine-tuning task.
 
-### Installation
+**Requirements**: *Nvidia* GPU (Ampere architecture or newer for `bf16` and Flash Attention) or *AMD* GPU, Python >=3.10 and PyTorch >=2.3.1.
 
 ```bash
-# install uv if you don't already have it installed (restart shell after)
-curl -LsSf https://astral.sh/uv/install.sh | sh
+git clone https://github.com/axolotl-ai-cloud/axolotl
+cd axolotl
 
-# change depending on system
-export UV_TORCH_BACKEND=cu128
+pip3 install packaging ninja
+pip3 install -e '.[flash-attn,deepspeed]'
+```
 
-# create a new virtual environment
-uv venv --python 3.12
-source .venv/bin/activate
+### Usage
+```bash
+# preprocess datasets - optional but recommended
+CUDA_VISIBLE_DEVICES="0" python -m axolotl.cli.preprocess examples/openllama-3b/lora.yml
 
-uv pip install torch==2.10.0 torchvision
-uv pip install --no-build-isolation axolotl[deepspeed]
+# finetune lora
+accelerate launch -m axolotl.cli.train examples/openllama-3b/lora.yml
 
-# Download example axolotl configs, deepspeed configs
+# inference
+accelerate launch -m axolotl.cli.inference examples/openllama-3b/lora.yml \
+    --lora_model_dir="./outputs/lora-out"
+
+# gradio
+accelerate launch -m axolotl.cli.inference examples/openllama-3b/lora.yml \
+    --lora_model_dir="./outputs/lora-out" --gradio
+
+# remote yaml files - the yaml config can be hosted on a public URL
+# Note: the yaml config must directly link to the **raw** yaml
+accelerate launch -m axolotl.cli.train https://raw.githubusercontent.com/axolotl-ai-cloud/axolotl/main/examples/openllama-3b/lora.yml
+```
+
+### Axolotl CLI
+
+If you've installed this package using `pip` from source, we now support a new, more
+streamlined CLI using [click](https://click.palletsprojects.com/en/stable/). Rewriting
+the above commands:
+
+```bash
+# preprocess datasets - optional but recommended
+CUDA_VISIBLE_DEVICES="0" axolotl preprocess examples/openllama-3b/lora.yml
+
+# finetune lora
+axolotl train examples/openllama-3b/lora.yml
+
+# inference
+axolotl inference examples/openllama-3b/lora.yml \
+    --lora-model-dir="./outputs/lora-out"
+
+# gradio
+axolotl inference examples/openllama-3b/lora.yml \
+    --lora-model-dir="./outputs/lora-out" --gradio
+
+# remote yaml files - the yaml config can be hosted on a public URL
+# Note: the yaml config must directly link to the **raw** yaml
+axolotl train https://raw.githubusercontent.com/axolotl-ai-cloud/axolotl/main/examples/openllama-3b/lora.yml
+```
+
+We've also added a new command for fetching `examples` and `deepspeed_configs` to your
+local machine. This will come in handy when installing `axolotl` from PyPI.
+
+```bash
+# Fetch example YAML files (stores in "examples/" folder)
 axolotl fetch examples
-axolotl fetch deepspeed_configs  # OPTIONAL
+
+# Fetch deepspeed config files (stores in "deepspeed_configs/" folder)
+axolotl fetch deepspeed_configs
+
+# Optionally, specify a destination folder
+axolotl fetch examples --dest path/to/folder
 ```
 
-#### Using Docker
+## Badge ❤🏷️
 
-Installing with Docker can be less error prone than installing in your own environment.
+Building something cool with Axolotl? Consider adding a badge to your model card.
+
+```markdown
+[<img src="https://raw.githubusercontent.com/axolotl-ai-cloud/axolotl/main/image/axolotl-badge-web.png" alt="Built with Axolotl" width="200" height="32"/>](https://github.com/axolotl-ai-cloud/axolotl)
+```
+
+[<img src="https://raw.githubusercontent.com/axolotl-ai-cloud/axolotl/main/image/axolotl-badge-web.png" alt="Built with Axolotl" width="200" height="32"/>](https://github.com/axolotl-ai-cloud/axolotl)
+
+## Sponsors 🤝❤
+
+If you love axolotl, consider sponsoring the project by reaching out directly to [wing@axolotl.ai](mailto:wing@axolotl.ai).
+
+---
+
+- [Modal](https://modal.com/) Modal lets you run data/AI jobs in the cloud, by just writing a few lines of Python. Customers use Modal to deploy Gen AI models at large scale, fine-tune LLM models, run protein folding simulations, and much more.
+
+---
+
+## Contributing 🤝
+
+Please read the [contributing guide](./.github/CONTRIBUTING.md)
+
+Bugs? Please check the [open issues](https://github.com/axolotl-ai-cloud/axolotl/issues/bug) else create a new Issue.
+
+PRs are **greatly welcome**!
+
+Please run the quickstart instructions followed by the below to setup env:
 ```bash
-docker run --gpus '"all"' --ipc=host --rm -it axolotlai/axolotl:main-latest
+pip3 install -r requirements-dev.txt -r requirements-tests.txt
+pre-commit install
+
+# test
+pytest tests/
+
+# optional: run against all files
+pre-commit run --all-files
 ```
 
-Other installation approaches are described [here](https://docs.axolotl.ai/docs/installation.html).
+Thanks to all of our contributors to date. Help drive open source AI progress forward by contributing to Axolotl.
 
-#### Cloud Providers
+<a href="https://github.com/axolotl-ai-cloud/axolotl/graphs/contributors">
+  <img src="https://contrib.rocks/image?repo=openaccess-ai-collective/axolotl" alt="contributor chart by https://contrib.rocks"/>
+</a>
+
+## Axolotl supports
+
+|             | fp16/fp32 | lora | qlora | gptq | gptq w/flash attn | flash attn | xformers attn |
+|-------------|:----------|:-----|-------|------|-------------------|------------|--------------|
+| llama       | ✅         | ✅    | ✅     | ✅             | ✅                 | ✅          | ✅            |
+| Mistral     | ✅         | ✅    | ✅     | ✅             | ✅                 | ✅          | ✅            |
+| Mixtral-MoE | ✅         | ✅    | ✅     | ❓             | ❓                 | ❓          | ❓            |
+| Mixtral8X22 | ✅         | ✅    | ✅     | ❓             | ❓                 | ❓          | ❓            |
+| Pythia      | ✅         | ✅    | ✅     | ❌             | ❌                 | ❌          | ❓            |
+| cerebras    | ✅         | ✅    | ✅     | ❌             | ❌                 | ❌          | ❓            |
+| btlm        | ✅         | ✅    | ✅     | ❌             | ❌                 | ❌          | ❓            |
+| mpt         | ✅         | ❌    | ❓     | ❌             | ❌                 | ❌          | ❓            |
+| falcon      | ✅         | ✅    | ✅     | ❌             | ❌                 | ❌          | ❓            |
+| gpt-j       | ✅         | ✅    | ✅     | ❌             | ❌                 | ❓          | ❓            |
+| XGen        | ✅         | ❓    | ✅     | ❓             | ❓                 | ❓          | ✅            |
+| phi         | ✅         | ✅    | ✅     | ❓             | ❓                 | ❓          | ❓            |
+| RWKV        | ✅         | ❓    | ❓     | ❓             | ❓                 | ❓          | ❓            |
+| Qwen        | ✅         | ✅    | ✅     | ❓             | ❓                 | ❓          | ❓            |
+| Gemma       | ✅         | ✅    | ✅     | ❓             | ❓                 | ✅          | ❓            |
+| Jamba       | ✅         | ✅    | ✅     | ❓             | ❓                 | ✅          | ❓            |
+
+✅: supported
+❌: not supported
+❓: untested
+
+## Advanced Setup
+
+### Environment
+
+#### Docker
+
+  ```bash
+  docker run --gpus '"all"' --rm -it axolotlai/axolotl:main-latest
+  ```
+
+  Or run on the current files for development:
+
+  ```sh
+  docker compose up -d
+  ```
+
+>[!Tip]
+> If you want to debug axolotl or prefer to use Docker as your development environment, see the [debugging guide's section on Docker](docs/debugging.qmd#debugging-with-docker).
+
+  <details>
+
+  <summary>Docker advanced</summary>
+
+  A more powerful Docker command to run would be this:
+
+  ```bash
+docker run --privileged --gpus '"all"' --shm-size 10g --rm -it --name axolotl --ipc=host --ulimit memlock=-1 --ulimit stack=67108864 --mount type=bind,src="${PWD}",target=/workspace/axolotl -v ${HOME}/.cache/huggingface:/root/.cache/huggingface axolotlai/axolotl:main-latest
+  ```
+
+  It additionally:
+  * Prevents memory issues when running e.g. deepspeed (e.g. you could hit SIGBUS/signal 7 error) through `--ipc` and `--ulimit` args.
+  * Persists the downloaded HF data (models etc.) and your modifications to axolotl code through `--mount`/`-v` args.
+  * The `--name` argument simply makes it easier to refer to the container in vscode (`Dev Containers: Attach to Running Container...`) or in your terminal.
+  * The `--privileged` flag gives all capabilities to the container.
+  * The `--shm-size 10g` argument increases the shared memory size. Use this if you see `exitcode: -7` errors using deepspeed.
+
+  [More information on nvidia website](https://docs.nvidia.com/deeplearning/frameworks/user-guide/index.html#setincshmem)
+
+  </details>
+
+#### Conda/Pip venv
+  1. Install python >=**3.10**
+
+  2. Install pytorch stable https://pytorch.org/get-started/locally/
+
+  3. Install Axolotl along with python dependencies
+        ```bash
+        pip3 install packaging
+        pip3 install -e '.[flash-attn,deepspeed]'
+        ```
+  4. (Optional) Login to Huggingface to use gated models/datasets.
+        ```bash
+        huggingface-cli login
+        ```
+        Get the token at huggingface.co/settings/tokens
+
+#### Cloud GPU
+
+For cloud GPU providers that support docker images, use [`axolotlai/axolotl-cloud:main-latest`](https://hub.docker.com/r/axolotlai/axolotl-cloud/tags)
+
+- on Latitude.sh use this [direct link](https://latitude.sh/blueprint/989e0e79-3bf6-41ea-a46b-1f246e309d5c)
+- on JarvisLabs.ai use this [direct link](https://jarvislabs.ai/templates/axolotl)
+- on RunPod use this [direct link](https://runpod.io/gsc?template=v2ickqhz9s&ref=6i7fkpdz)
+
+#### Bare Metal Cloud GPU
+
+##### LambdaLabs
+
+  <details>
+
+  <summary>Click to Expand</summary>
+
+  1. Install python
+  ```bash
+  sudo apt update
+  sudo apt install -y python3.10
+
+  sudo update-alternatives --install /usr/bin/python python /usr/bin/python3.10 1
+  sudo update-alternatives --config python # pick 3.10 if given option
+  python -V # should be 3.10
+
+  ```
+
+  2. Install pip
+  ```bash
+  wget https://bootstrap.pypa.io/get-pip.py
+  python get-pip.py
+  ```
+
+  3. Install Pytorch https://pytorch.org/get-started/locally/
+
+  4. Follow instructions on quickstart.
+
+  5. Run
+  ```bash
+  pip3 install protobuf==3.20.3
+  pip3 install -U --ignore-installed requests Pillow psutil scipy
+  ```
+
+  6. Set path
+  ```bash
+  export LD_LIBRARY_PATH=/usr/lib/x86_64-linux-gnu:$LD_LIBRARY_PATH
+  ```
+  </details>
+
+##### GCP
 
 <details>
 
-- [RunPod](https://runpod.io/gsc?template=v2ickqhz9s&ref=6i7fkpdz)
-- [Vast.ai](https://cloud.vast.ai?ref_id=62897&template_id=bdd4a49fa8bce926defc99471864cace&utm_source=github&utm_medium=developer_community&utm_campaign=template_launch_axolotl&utm_content=readme)
-- [PRIME Intellect](https://app.primeintellect.ai/dashboard/create-cluster?image=axolotl&location=Cheapest&security=Cheapest&show_spot=true)
-- [Modal](https://www.modal.com?utm_source=github&utm_medium=github&utm_campaign=axolotl)
-- [Novita](https://novita.ai/gpus-console?templateId=311)
-- [JarvisLabs.ai](https://jarvislabs.ai/templates/axolotl)
-- [Latitude.sh](https://latitude.sh/blueprint/989e0e79-3bf6-41ea-a46b-1f246e309d5c)
+<summary>Click to Expand</summary>
+
+Use a Deeplearning linux OS with cuda and pytorch installed. Then follow instructions on quickstart.
+
+Make sure to run the below to uninstall xla.
+```bash
+pip uninstall -y torch_xla[tpu]
+```
 
 </details>
 
-### Your First Fine-tune
+#### Windows
+Please use WSL or Docker!
+
+#### Mac
+
+Use the below instead of the install method in QuickStart.
+```
+pip3 install -e '.'
+```
+More info: [mac.md](/docs/mac.qmd)
+
+#### Google Colab
+
+Please use this example [notebook](examples/colab-notebooks/colab-axolotl-example.ipynb).
+
+#### Launching on public clouds via SkyPilot
+To launch on GPU instances (both on-demand and spot instances) on 7+ clouds (GCP, AWS, Azure, OCI, and more), you can use [SkyPilot](https://skypilot.readthedocs.io/en/latest/index.html):
 
 ```bash
-# Fetch axolotl examples
-axolotl fetch examples
-
-# Or, specify a custom path
-axolotl fetch examples --dest path/to/folder
-
-# Train a model using LoRA
-axolotl train examples/llama-3/lora-1b.yml
+pip install "skypilot-nightly[gcp,aws,azure,oci,lambda,kubernetes,ibm,scp]"  # choose your clouds
+sky check
 ```
 
-That's it! Check out our [Getting Started Guide](https://docs.axolotl.ai/docs/getting-started.html) for a more detailed walkthrough.
+Get the [example YAMLs](https://github.com/skypilot-org/skypilot/tree/master/llm/axolotl) of using Axolotl to finetune `mistralai/Mistral-7B-v0.1`:
+```
+git clone https://github.com/skypilot-org/skypilot.git
+cd skypilot/llm/axolotl
+```
 
+Use one command to launch:
+```bash
+# On-demand
+HF_TOKEN=xx sky launch axolotl.yaml --env HF_TOKEN
 
-## 📚 Documentation
+# Managed spot (auto-recovery on preemption)
+HF_TOKEN=xx BUCKET=<unique-name> sky spot launch axolotl-spot.yaml --env HF_TOKEN --env BUCKET
+```
 
-- [Installation Options](https://docs.axolotl.ai/docs/installation.html) - Detailed setup instructions for different environments
-- [Configuration Guide](https://docs.axolotl.ai/docs/config-reference.html) - Full configuration options and examples
-- [Dataset Loading](https://docs.axolotl.ai/docs/dataset_loading.html) - Loading datasets from various sources
-- [Dataset Guide](https://docs.axolotl.ai/docs/dataset-formats/) - Supported formats and how to use them
-- [Multi-GPU Training](https://docs.axolotl.ai/docs/multi-gpu.html)
-- [Multi-Node Training](https://docs.axolotl.ai/docs/multi-node.html)
-- [Multipacking](https://docs.axolotl.ai/docs/multipack.html)
-- [API Reference](https://docs.axolotl.ai/docs/api/) - Auto-generated code documentation
-- [FAQ](https://docs.axolotl.ai/docs/faq.html) - Frequently asked questions
+#### Launching on public clouds via dstack
+To launch on GPU instance (both on-demand and spot instances) on public clouds (GCP, AWS, Azure, Lambda Labs, TensorDock, Vast.ai, and CUDO), you can use [dstack](https://dstack.ai/).
 
-## AI Agent Support
+Write a job description in YAML as below:
 
-Axolotl ships with built-in documentation optimized for AI coding agents (Claude Code, Cursor, Copilot, etc.). These docs are bundled with the pip package — no repo clone needed.
+```yaml
+# dstack.yaml
+type: task
+
+image: axolotlai/axolotl-cloud:main-latest
+
+env:
+  - HUGGING_FACE_HUB_TOKEN
+  - WANDB_API_KEY
+
+commands:
+  - accelerate launch -m axolotl.cli.train config.yaml
+
+ports:
+  - 6006
+
+resources:
+  gpu:
+    memory: 24GB..
+    count: 2
+```
+
+then, simply run the job with `dstack run` command. Append `--spot` option if you want spot instance. `dstack run` command will show you the instance with cheapest price across multi cloud services:
 
 ```bash
-# Show overview and available training methods
-axolotl agent-docs
-
-# Topic-specific references
-axolotl agent-docs sft                 # supervised fine-tuning
-axolotl agent-docs grpo                # GRPO online RL
-axolotl agent-docs preference_tuning   # DPO, KTO, ORPO, SimPO
-axolotl agent-docs reward_modelling    # outcome and process reward models
-axolotl agent-docs pretraining         # continual pretraining
-axolotl agent-docs --list              # list all topics
-
-# Dump config schema for programmatic use
-axolotl config-schema
-axolotl config-schema --field adapter
+pip install dstack
+HUGGING_FACE_HUB_TOKEN=xxx WANDB_API_KEY=xxx dstack run . -f dstack.yaml # --spot
 ```
 
-If you're working with the source repo, agent docs are also available at `docs/agents/` and the project overview is in `AGENTS.md`.
+For further and fine-grained use cases, please refer to the official [dstack documents](https://dstack.ai/docs/) and the detailed description of [axolotl example](https://github.com/dstackai/dstack/tree/master/examples/fine-tuning/axolotl) on the official repository.
 
-## 🤝 Getting Help
+### Dataset
 
-- Join our [Discord community](https://discord.gg/HhrNrHJPRb) for support
-- Check out our [Examples](https://github.com/axolotl-ai-cloud/axolotl/tree/main/examples/) directory
-- Read our [Debugging Guide](https://docs.axolotl.ai/docs/debugging.html)
-- Need dedicated support? Please contact [✉️wing@axolotl.ai](mailto:wing@axolotl.ai) for options
+Axolotl supports a variety of dataset formats.  It is recommended to use a JSONL.  The schema of the JSONL depends upon the task and the prompt template you wish to use.  Instead of a JSONL, you can also use a HuggingFace dataset with columns for each JSONL field.
 
-## 🌟 Contributing
+See [the documentation](https://axolotl-ai-cloud.github.io/axolotl/docs/dataset-formats/) for more information on how to use different dataset formats.
 
-Contributions are welcome! Please see our [Contributing Guide](https://github.com/axolotl-ai-cloud/axolotl/blob/main/.github/CONTRIBUTING.md) for details.
+### Config
 
-## 📈 Telemetry
+See [examples](examples) for quick start. It is recommended to duplicate and modify to your needs. The most important options are:
 
-Axolotl has opt-out telemetry that helps us understand how the project is being used
-and prioritize improvements. We collect basic system information, model types, and
-error rates—never personal data or file paths. Telemetry is enabled by default. To
-disable it, set AXOLOTL_DO_NOT_TRACK=1. For more details, see our [telemetry documentation](https://docs.axolotl.ai/docs/telemetry.html).
+- model
+  ```yaml
+  base_model: ./llama-7b-hf # local or huggingface repo
+  ```
+  Note: The code will load the right architecture.
 
-## ❤️ Sponsors
+- dataset
+  ```yaml
+  datasets:
+      # huggingface repo
+    - path: vicgalle/alpaca-gpt4
+      type: alpaca
 
-Interested in sponsoring? Contact us at [wing@axolotl.ai](mailto:wing@axolotl.ai)
+      # huggingface repo with specific configuration/subset
+    - path: EleutherAI/pile
+      name: enron_emails
+      type: completion # format from earlier
+      field: text # Optional[str] default: text, field to use for completion data
 
-## 📝 Citing Axolotl
+      # huggingface repo with multiple named configurations/subsets
+    - path: bigcode/commitpackft
+      name:
+        - ruby
+        - python
+        - typescript
+      type: ... # unimplemented custom format
 
-If you use Axolotl in your research or projects, please cite it as follows:
+      # chat_template https://axolotl-ai-cloud.github.io/axolotl/docs/dataset-formats/conversation.html#chat_template
+    - path: ...
+      type: chat_template
+      chat_template: chatml # defaults to tokenizer's chat_template
 
-```bibtex
-@software{axolotl,
-  title = {Axolotl: Open Source LLM Post-Training},
-  author = {{Axolotl maintainers and contributors}},
-  url = {https://github.com/axolotl-ai-cloud/axolotl},
-  license = {Apache-2.0},
-  year = {2023}
-}
+      # local
+    - path: data.jsonl # or json
+      ds_type: json # see other options below
+      type: alpaca
+
+      # dataset with splits, but no train split
+    - path: knowrohit07/know_sql
+      type: context_qa.load_v2
+      train_on_split: validation
+
+      # loading from s3 or gcs
+      # s3 creds will be loaded from the system default and gcs only supports public access
+    - path: s3://path_to_ds # Accepts folder with arrow/parquet or file path like above. Supports s3, gcs.
+      ...
+
+      # Loading Data From a Public URL
+      # - The file format is `json` (which includes `jsonl`) by default. For different formats, adjust the `ds_type` option accordingly.
+    - path: https://some.url.com/yourdata.jsonl # The URL should be a direct link to the file you wish to load. URLs must use HTTPS protocol, not HTTP.
+      ds_type: json # this is the default, see other options below.
+  ```
+
+- loading
+  ```yaml
+  load_in_4bit: true
+  load_in_8bit: true
+
+  bf16: auto # require >=ampere, auto will detect if your GPU supports this and choose automatically.
+  fp16: # leave empty to use fp16 when bf16 is 'auto'. set to false if you want to fallback to fp32
+  tf32: true # require >=ampere
+
+  bfloat16: true # require >=ampere, use instead of bf16 when you don't want AMP (automatic mixed precision)
+  float16: true # use instead of fp16 when you don't want AMP
+  ```
+  Note: Repo does not do 4-bit quantization.
+
+- lora
+  ```yaml
+  adapter: lora # 'qlora' or leave blank for full finetune
+  lora_r: 8
+  lora_alpha: 16
+  lora_dropout: 0.05
+  lora_target_modules:
+    - q_proj
+    - v_proj
+  ```
+
+#### All Config Options
+
+See [these docs](docs/config.qmd) for all config options.
+
+### Train
+
+Run
+```bash
+accelerate launch -m axolotl.cli.train your_config.yml
 ```
 
-## 📜 License
+> [!TIP]
+> You can also reference a config file that is hosted on a public URL, for example `accelerate launch -m axolotl.cli.train https://yourdomain.com/your_config.yml`
 
-This project is licensed under the Apache 2.0 License - see the [LICENSE](LICENSE) file for details.
+#### Preprocess dataset
+
+You can optionally pre-tokenize dataset with the following before finetuning.
+This is recommended for large datasets.
+
+- Set `dataset_prepared_path:` to a local folder for saving and loading pre-tokenized dataset.
+- (Optional): Set `push_dataset_to_hub: hf_user/repo` to push it to Huggingface.
+- (Optional): Use `--debug` to see preprocessed examples.
+
+```bash
+python -m axolotl.cli.preprocess your_config.yml
+```
+
+#### Multi-GPU
+
+Below are the options available in axolotl for training with multiple GPUs. Note that DeepSpeed
+is the recommended multi-GPU option currently because FSDP may experience
+[loss instability](https://github.com/huggingface/transformers/issues/26498).
+
+##### DeepSpeed
+
+Deepspeed is an optimization suite for multi-gpu systems allowing you to train much larger models than you
+might typically be able to fit into your GPU's VRAM. More information about the various optimization types
+for deepspeed is available at https://huggingface.co/docs/accelerate/main/en/usage_guides/deepspeed#what-is-integrated
+
+We provide several default deepspeed JSON configurations for ZeRO stage 1, 2, and 3.
+
+```yaml
+deepspeed: deepspeed_configs/zero1.json
+```
+
+```shell
+accelerate launch -m axolotl.cli.train examples/llama-2/config.yml --deepspeed deepspeed_configs/zero1.json
+```
+
+##### FSDP
+
+- llama FSDP
+```yaml
+fsdp:
+  - full_shard
+  - auto_wrap
+fsdp_config:
+  fsdp_offload_params: true
+  fsdp_state_dict_type: FULL_STATE_DICT
+  fsdp_transformer_layer_cls_to_wrap: LlamaDecoderLayer
+```
+
+##### FSDP + QLoRA
+
+Axolotl supports training with FSDP and QLoRA, see [these docs](docs/fsdp_qlora.qmd) for more information.
+
+##### Weights & Biases Logging
+
+Make sure your `WANDB_API_KEY` environment variable is set (recommended) or you login to wandb with `wandb login`.
+
+- wandb options
+```yaml
+wandb_mode:
+wandb_project:
+wandb_entity:
+wandb_watch:
+wandb_name:
+wandb_log_model:
+```
+
+##### Comet Logging
+
+Make sure your `COMET_API_KEY` environment variable is set (recommended) or you login to wandb with `comet login`.
+
+- wandb options
+```yaml
+use_comet:
+comet_api_key:
+comet_workspace:
+comet_project_name:
+comet_experiment_key:
+comet_mode:
+comet_online:
+comet_experiment_config:
+```
+
+##### Special Tokens
+
+It is important to have special tokens like delimiters, end-of-sequence, beginning-of-sequence in your tokenizer's vocabulary.  This will help you avoid tokenization issues and help your model train better.  You can do this in axolotl like this:
+
+```yml
+special_tokens:
+  bos_token: "<s>"
+  eos_token: "</s>"
+  unk_token: "<unk>"
+tokens: # these are delimiters
+  - "<|im_start|>"
+  - "<|im_end|>"
+```
+
+When you include these tokens in your axolotl config, axolotl adds these tokens to the tokenizer's vocabulary.
+
+##### Liger Kernel
+
+Liger Kernel: Efficient Triton Kernels for LLM Training
+
+https://github.com/linkedin/Liger-Kernel
+
+Liger (LinkedIn GPU Efficient Runtime) Kernel is a collection of Triton kernels designed specifically for LLM training.
+It can effectively increase multi-GPU training throughput by 20% and reduces memory usage by 60%. The Liger Kernel
+composes well and is compatible with both FSDP and Deepspeed.
+
+```yaml
+plugins:
+  - axolotl.integrations.liger.LigerPlugin
+liger_rope: true
+liger_rms_norm: true
+liger_glu_activation: true
+liger_layer_norm: true
+liger_fused_linear_cross_entropy: true
+```
+
+### Inference Playground
+
+Axolotl allows you to load your model in an interactive terminal playground for quick experimentation.
+The config file is the same config file used for training.
+
+Pass the appropriate flag to the inference command, depending upon what kind of model was trained:
+
+- Pretrained LORA:
+  ```bash
+  python -m axolotl.cli.inference examples/your_config.yml --lora_model_dir="./lora-output-dir"
+  ```
+- Full weights finetune:
+  ```bash
+  python -m axolotl.cli.inference examples/your_config.yml --base_model="./completed-model"
+  ```
+- Full weights finetune w/ a prompt from a text file:
+  ```bash
+  cat /tmp/prompt.txt | python -m axolotl.cli.inference examples/your_config.yml \
+    --base_model="./completed-model" --prompter=None --load_in_8bit=True
+  ```
+-- With gradio hosting
+  ```bash
+  python -m axolotl.cli.inference examples/your_config.yml --gradio
+  ```
+
+Please use `--sample_packing False` if you have it on and receive the error similar to below:
+
+> RuntimeError: stack expects each tensor to be equal size, but got [1, 32, 1, 128] at entry 0 and [1, 32, 8, 128] at entry 1
+
+### Merge LORA to base
+
+The following command will merge your LORA adapater with your base model. You can optionally pass the argument `--lora_model_dir` to specify the directory where your LORA adapter was saved, otherwhise, this will be inferred from `output_dir` in your axolotl config file.  The merged model is saved in the sub-directory `{lora_model_dir}/merged`.
+
+```bash
+python3 -m axolotl.cli.merge_lora your_config.yml --lora_model_dir="./completed-model"
+```
+
+You may need to use the `gpu_memory_limit` and/or `lora_on_cpu` config options to avoid running out of memory. If you still run out of CUDA memory, you can try to merge in system RAM with
+
+```bash
+CUDA_VISIBLE_DEVICES="" python3 -m axolotl.cli.merge_lora ...
+```
+
+although this will be very slow, and using the config options above are recommended instead.
+
+## Common Errors 🧰
+
+See also the [FAQ's](./docs/faq.qmd) and [debugging guide](docs/debugging.qmd).
+
+> If you encounter a 'Cuda out of memory' error, it means your GPU ran out of memory during the training process. Here's how to resolve it:
+
+Please reduce any below
+  - `micro_batch_size`
+  - `eval_batch_size`
+  - `gradient_accumulation_steps`
+  - `sequence_len`
+
+If it does not help, try running without deepspeed and without accelerate (replace "accelerate launch" with "python") in the command.
+
+Using adamw_bnb_8bit might also save you some memory.
+
+> `failed (exitcode: -9)`
+
+Usually means your system has run out of system memory.
+Similarly, you should consider reducing the same settings as when you run out of VRAM.
+Additionally, look into upgrading your system RAM which should be simpler than GPU upgrades.
+
+> RuntimeError: expected scalar type Float but found Half
+
+Try set `fp16: true`
+
+> NotImplementedError: No operator found for `memory_efficient_attention_forward` ...
+
+Try to turn off xformers.
+
+> accelerate config missing
+
+It's safe to ignore it.
+
+> NCCL Timeouts during training
+
+See the [NCCL](docs/nccl.qmd) guide.
+
+
+### Tokenization Mismatch b/w Inference & Training
+
+For many formats, Axolotl constructs prompts by concatenating token ids _after_ tokenizing strings.  The reason for concatenating token ids rather than operating on strings is to maintain precise accounting for attention masks.
+
+If you decode a prompt constructed by axolotl, you might see spaces between tokens (or lack thereof) that you do not expect, especially around delimiters and special tokens.  When you are starting out with a new format, you should always do the following:
+
+1. Materialize some data using `python -m axolotl.cli.preprocess your_config.yml --debug`, and then decode the first few rows with your model's tokenizer.
+2. During inference, right before you pass a tensor of token ids to your model, decode these tokens back into a string.
+3. Make sure the inference string from #2 looks **exactly** like the data you fine tuned on from #1, including spaces and new lines.  If they aren't the same, adjust your inference server accordingly.
+4. As an additional troubleshooting step, you can look at the token ids between 1 and 2 to make sure they are identical.
+
+Having misalignment between your prompts during training and inference can cause models to perform very poorly, so it is worth checking this.  See [this blog post](https://hamel.dev/notes/llm/finetuning/05_tokenizer_gotchas.html) for a concrete example.
+
+## Debugging Axolotl
+
+See [this debugging guide](docs/debugging.qmd) for tips on debugging Axolotl, along with an example configuration for debugging with VSCode.
+
+## Need help? 🙋
+
+Join our [Discord server](https://discord.gg/HhrNrHJPRb) where our community members can help you.
+
+Need dedicated support? Please contact us at [✉️wing@axolotl.ai](ailto:wing@axolotl.ai) for dedicated support options.

SETUP_MIAAI.md

@@ -1,178 +0,0 @@
-# Axolotl Setup — miaai (RTX 5080, CUDA 13.2)
-
-## System Info
-- GPU: NVIDIA RTX 5080 (16GB VRAM, sm_120 / Blackwell)
-- Driver: 580.126.09 — max CUDA 13.0 shown by nvidia-smi, but nvcc from conda is 13.2
-- OS: Ubuntu 25.10 (Python 3.13 system — do NOT use system Python for ML)
-- Axolotl branch: `activeblue/main`
-
----
-
-## One-time Setup
-
-### 1. Install Miniconda
-```bash
-wget https://repo.anaconda.com/miniconda/Miniconda3-latest-Linux-x86_64.sh -O miniconda.sh
-bash miniconda.sh -b -p /opt/miniconda3
-/opt/miniconda3/bin/conda init bash
-source ~/.bashrc
-```
-
-### 2. Create Python 3.11 environment
-```bash
-conda create -n axolotl python=3.11 -y
-conda activate axolotl
-```
-
-### 3. Clone and sync repo with upstream
-```bash
-git clone https://git.activeblue.net/tocmo0nlord/axolotl.git
-cd axolotl
-git remote add upstream https://github.com/axolotl-ai-cloud/axolotl.git
-git fetch upstream
-git rebase upstream/main        # keeps activeblue patches on top
-git push origin activeblue/main --force-with-lease
-```
-
-### 4. Install CUDA toolkit (needed to compile flash-attn and bitsandbytes)
-```bash
-conda install -y -c "nvidia/label/cuda-12.8.0" cuda-toolkit
-export CUDA_HOME=$CONDA_PREFIX
-export PATH=$CUDA_HOME/bin:$PATH
-```
-
-> NOTE: Despite installing from the cuda-12.8.0 channel, conda resolves nvcc to **13.2.78**.
-> This is fine — use cu132 everywhere to match.
-
-### 5. Install PyTorch — use cu132 (matches nvcc from conda)
-> NOTE: torchaudio has no cu132 wheel — skip it, not needed for LLM training
-```bash
-pip install torch torchvision --index-url https://download.pytorch.org/whl/cu132
-python -c "import torch; print('CUDA:', torch.version.cuda); print('GPU:', torch.cuda.get_device_name(0))"
-```
-
-### 6. Install Axolotl
-```bash
-pip install -e "."
-```
-
-> **flash-attn compiles CUDA kernels from source — takes 15–25 min on 10 cores of i7-14700K.**
-> Always set `MAX_JOBS` to the number of available CPU cores:
-```bash
-MAX_JOBS=10 pip install flash-attn --no-build-isolation
-```
-
-### 7. Compile bitsandbytes from source for sm_120 (RTX 5080 / Blackwell)
-
-The prebuilt bitsandbytes wheels do not include sm_120 support and CUDA 13.2 dropped sm_50–53.
-You must compile from source with a patched CMakeLists.txt.
-
-```bash
-# Clone bitsandbytes v0.49.1
-git clone --branch v0.49.1 --depth 1 https://github.com/bitsandbytes-foundation/bitsandbytes.git /tmp/bnb_0491
-cd /tmp/bnb_0491
-
-# Patch CMakeLists.txt: override arch list to sm_120 only, just before the foreach loop
-# (cmake >= 3.23.0 skips the manual arch block and uses its own built-in list which lacks sm_120)
-sed -i '/    foreach(capability \${CMAKE_CUDA_ARCHITECTURES_ALL})/i\    # RTX 5080 sm_120 patch: override before capability list is built\n    set(CMAKE_CUDA_ARCHITECTURES_ALL 120)' CMakeLists.txt
-
-# Verify the patch landed at the right line
-grep -n "ARCHITECTURES_ALL\|foreach" CMakeLists.txt | tail -5
-# Should show: set(CMAKE_CUDA_ARCHITECTURES_ALL 120) immediately before the foreach line
-
-# Configure — must point cmake at conda's nvcc
-cmake \
-  -DCMAKE_CUDA_COMPILER=/opt/miniconda3/envs/axolotl/bin/nvcc \
-  -DCOMPUTE_BACKEND=cuda \
-  -S /tmp/bnb_0491 \
-  -B /tmp/bnb_0491/build 2>&1 | grep -E "(Capabilit|CUDA Ver|Error)"
-# Expected: "CUDA Capabilities Selected: 120"
-
-# Build (j10 uses 10 cores — adjust to your CPU)
-cmake --build /tmp/bnb_0491/build -j10
-
-# Install into conda site-packages
-SITE_PKG=/opt/miniconda3/envs/axolotl/lib/python3.11/site-packages
-cp -r /tmp/bnb_0491/bitsandbytes "$SITE_PKG/"
-
-# Verify
-python3 -c "
-import torch, bitsandbytes as bnb
-x = torch.randn(64, 64, device='cuda')
-l = bnb.nn.Linear8bitLt(64, 64).cuda()
-print('bitsandbytes CUDA OK:', l(x).shape)
-"
-```
-
----
-
-## Every Session (after first-time setup)
-```bash
-export PATH="/opt/miniconda3/bin:$PATH"
-conda activate axolotl
-export CUDA_HOME=$CONDA_PREFIX
-export PATH=$CUDA_HOME/bin:$PATH
-export PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True
-cd /home/tocmo0nlord/axolotl
-```
-
----
-
-## Training Config — human_chat_qlora.yml
-
-Key settings that work on RTX 5080 (16GB):
-
-| Setting | Value | Notes |
-|---|---|---|
-| `sequence_len` | `2048` | 4096 causes OOM during loss computation (logits x 128k vocab) |
-| `micro_batch_size` | `1` | Keep low; effective batch = micro x grad_accum |
-| `gradient_accumulation_steps` | `8` | Effective batch = 8 |
-| `adapter` | `qlora` | QLoRA 4-bit via bitsandbytes |
-| `attn_implementation` | `flash_attention_2` | Not the deprecated `flash_attention: true` |
-| `type` (datasets) | `chat_template` | Not the deprecated `sharegpt` |
-
-Dataset fields for SlimOrca / OpenHermes-2.5 (sharegpt-format with different field names):
-```yaml
-datasets:
-  - path: Open-Orca/SlimOrca
-    type: chat_template
-    field_messages: conversations
-    message_field_role: from
-    message_field_content: value
-    split: "train[:3%]"
-```
-
-## Run Training
-```bash
-export PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True
-axolotl train ~/human_chat_qlora.yml
-```
-
-Expected startup sequence:
-1. Config validation + capability detection (shows `sm_120`)
-2. Dataset tokenization (~65k samples, ~30 seconds)
-3. `Loading weights: 100% 291/291`
-4. `trainable params: 167,772,160 || all params: 8,198,033,408 || trainable%: 2.05`
-5. Initial eval: loss ~0.81, perplexity ~2.25, VRAM ~8.5GB
-6. Training steps at ~2.6 it/s, VRAM ~9-10GB
-
----
-
-## Common Pitfalls Encountered
-
-| Problem | Cause | Fix |
-|---|---|---|
-| `externally-managed-environment` | System Python 3.13 blocks pip | Use conda env, never system pip |
-| `No module named torch` (flash-attn) | pip builds in isolated env | Use `--no-build-isolation` |
-| `CUDA_HOME not set` | CUDA toolkit not installed | `conda install cuda-toolkit` from nvidia channel |
-| `CUDA version mismatch 13.2 vs 12.8` | Conda nvcc is 13.2, torch was cu128 | Reinstall torch with `--index-url .../cu132` |
-| `torchaudio` not found for cu132 | No cu132 wheel exists | Skip torchaudio — not needed |
-| `src refspec main does not match` | Fork default branch is `activeblue/main` | `git push origin activeblue/main` |
-| flash-attn compile is slow | Single-threaded by default | Set `MAX_JOBS=<cpu_count>` before pip install |
-| `nvcc fatal: Unsupported gpu architecture 'compute_50'` | bitsandbytes CMakeLists.txt hardcodes sm_50; CUDA 13.2 dropped it | Patch CMakeLists.txt (see step 7 above) |
-| `CUDA Capabilities Selected: 50;52;...` (ignores sm_120) | cmake >= 3.23 built-in arch list lacks sm_120 | Add `set(CMAKE_CUDA_ARCHITECTURES_ALL 120)` before foreach loop |
-| `BackendUnavailable: scikit_build_core` | pip install of bnb tries to rebuild | Copy .so directly to site-packages instead |
-| `torch.OutOfMemoryError` during eval | logits tensor (batch x 4096 x 128k vocab) too large | Set `sequence_len: 2048`, `micro_batch_size: 1` |
-| `type: sharegpt` deprecation warning | axolotl removed sharegpt type | Use `type: chat_template` with field mappings |
-| `flash_attention: true` deprecation | Old config key removed | Use `attn_implementation: flash_attention_2` |
-| Capybara dataset `field_messages null` | Capybara uses input/output format, not conversations | Switch to SlimOrca or OpenHermes-2.5 |

TODO.md

@@ -0,0 +1,10 @@
+# todo list
+
+- [] Validation of parameters for combinations that won't work
+
+
+
+## things that are known not to work
+
+- FSDP offload and gradient_checkpointing - https://github.com/pytorch/pytorch/issues/82203
+- adamw_bnb_8bit doesn't play well with FSDP offload

VERSION

@@ -1 +0,0 @@
-0.16.2.dev0

_quarto.yml

@@ -1,220 +1,12 @@
 project:
   type: website
-  pre-render:
-   - docs/scripts/generate_config_docs.py
-   - docs/scripts/generate_examples_docs.py
 
-quartodoc:
-  dir: docs/api
-  package: axolotl
-  title: API Reference
-  parser: google
-
-  sections:
-    - title: Core
-      desc: Core functionality for training
-      contents:
-        - train
-        - evaluate
-        - datasets
-        - convert
-        - prompt_tokenizers
-        - logging_config
-        - core.builders.base
-        - core.builders.causal
-        - core.builders.rl
-        - core.training_args
-        - core.chat.messages
-        - core.chat.format.chatml
-        - core.chat.format.llama3x
-        - core.chat.format.shared
-        - core.datasets.chat
-        - core.datasets.transforms.chat_builder
-    - title: CLI
-      desc: Command-line interface
-      contents:
-        - cli.main
-        - cli.train
-        - cli.evaluate
-        - cli.args
-        - cli.art
-        - cli.checks
-        - cli.config
-        - cli.delinearize_llama4
-        - cli.inference
-        - cli.merge_lora
-        - cli.merge_sharded_fsdp_weights
-        - cli.preprocess
-        - cli.quantize
-        - cli.vllm_serve
-        - cli.cloud.base
-        - cli.cloud.modal_
-        - cli.utils
-        - cli.utils.args
-        - cli.utils.fetch
-        - cli.utils.load
-        - cli.utils.sweeps
-        - cli.utils.train
-    - title: Trainers
-      desc: Training implementations
-      contents:
-        - core.trainers.base
-        - core.trainers.trl
-        - core.trainers.mamba
-        - core.trainers.dpo.trainer
-        - core.trainers.grpo.trainer
-        - core.trainers.grpo.sampler
-        - core.trainers.utils
-    - title: Model Loading
-      desc: Functionality for loading and patching models, tokenizers, etc.
-      contents:
-        - loaders.model
-        - loaders.tokenizer
-        - loaders.processor
-        - loaders.adapter
-        - loaders.patch_manager
-        - loaders.constants
-    - title: Mixins
-      desc: Mixin classes for augmenting trainers
-      contents:
-        - core.trainers.mixins.optimizer
-        - core.trainers.mixins.rng_state_loader
-        - core.trainers.mixins.scheduler
-    - title: Context Managers
-      desc: Context managers for altering trainer behaviors
-      contents:
-        - utils.ctx_managers.sequence_parallel
-    - title: Prompt Strategies
-      desc: Prompt formatting strategies
-      contents:
-        - prompt_strategies.base
-        - prompt_strategies.chat_template
-        - prompt_strategies.alpaca_chat
-        - prompt_strategies.alpaca_instruct
-        - prompt_strategies.alpaca_w_system
-        - prompt_strategies.user_defined
-        - prompt_strategies.llama2_chat
-        - prompt_strategies.completion
-        - prompt_strategies.input_output
-        - prompt_strategies.stepwise_supervised
-        - prompt_strategies.metharme
-        - prompt_strategies.orcamini
-        - prompt_strategies.pygmalion
-        - prompt_strategies.messages.chat
-        - prompt_strategies.dpo.chat_template
-        - prompt_strategies.dpo.llama3
-        - prompt_strategies.dpo.chatml
-        - prompt_strategies.dpo.zephyr
-        - prompt_strategies.dpo.user_defined
-        - prompt_strategies.dpo.passthrough
-        - prompt_strategies.kto.llama3
-        - prompt_strategies.kto.chatml
-        - prompt_strategies.kto.user_defined
-        - prompt_strategies.orpo.chat_template
-        - prompt_strategies.bradley_terry.llama3
-    - title: Kernels
-      desc: Low-level performance optimizations
-      contents:
-        - kernels.lora
-        - kernels.geglu
-        - kernels.swiglu
-        - kernels.quantize
-        - kernels.utils
-    - title: Monkey Patches
-      desc: Runtime patches for model optimizations
-      contents:
-        - monkeypatch.llama_attn_hijack_flash
-        - monkeypatch.llama_attn_hijack_xformers
-        - monkeypatch.mistral_attn_hijack_flash
-        - monkeypatch.multipack
-        - monkeypatch.relora
-        - monkeypatch.lora_kernels
-        - monkeypatch.utils
-        - monkeypatch.btlm_attn_hijack_flash
-        - monkeypatch.stablelm_attn_hijack_flash
-        - monkeypatch.trainer_fsdp_optim
-        - monkeypatch.transformers_fa_utils
-        - monkeypatch.data.batch_dataset_fetcher
-        - monkeypatch.mixtral
-        - monkeypatch.gradient_checkpointing.offload_cpu
-        - monkeypatch.gradient_checkpointing.offload_disk
-    - title: Utils
-      desc: Utility functions
-      contents:
-        - utils.tokenization
-        - utils.chat_templates
-        - utils.lora
-        - utils.model_shard_quant
-        - utils.bench
-        - utils.freeze
-        - utils.trainer
-        - utils.schedulers
-        - utils.distributed
-        - utils.dict
-        - utils.optimizers.adopt
-        - utils.data.streaming
-        - utils.data.sft
-        - utils.quantization
-    - title: Schemas
-      desc: Pydantic data models for Axolotl config
-      contents:
-        - utils.schemas.config
-        - utils.schemas.model
-        - utils.schemas.training
-        - utils.schemas.datasets
-        - utils.schemas.peft
-        - utils.schemas.trl
-        - utils.schemas.multimodal
-        - utils.schemas.integrations
-        - utils.schemas.enums
-        - utils.schemas.utils
-    - title: Integrations
-      desc: Third-party integrations and extensions
-      contents:
-        - integrations.base
-        - integrations.cut_cross_entropy.args
-        - integrations.grokfast.optimizer
-        - integrations.kd.trainer
-        - integrations.liger.args
-        - integrations.lm_eval.args
-        - integrations.spectrum.args
-    - title: Common
-      desc: Common utilities and shared functionality
-      contents:
-        - common.architectures
-        - common.const
-        - common.datasets
-    - title: Models
-      desc: Custom model implementations
-      contents:
-        - models.mamba.modeling_mamba
-    - title: Data Processing
-      desc: Data processing utilities
-      contents:
-        - utils.collators.core
-        - utils.collators.batching
-        - utils.collators.mamba
-        - utils.collators.mm_chat
-        - utils.samplers.multipack
-    - title: Callbacks
-      desc: Training callbacks
-      contents:
-        - utils.callbacks.perplexity
-        - utils.callbacks.profiler
-        - utils.callbacks.lisa
-        - utils.callbacks.mlflow_
-        - utils.callbacks.comet_
-        - utils.callbacks.qat
 website:
   title: "Axolotl"
-  description: "We make fine-tuning accessible, scalable, and fun"
+  description: "Fine-tuning"
   favicon: favicon.jpg
-
-  google-analytics: "G-9KYCVJBNMQ"
-
   navbar:
-    logo: image/axolotl_logo_digital_white.svg
-    title: false
+    title: Axolotl
     background: dark
     pinned: false
     collapse: false
@@ -233,135 +25,29 @@ website:
       contents:
         - text: Home
           href: index.qmd
-
-        - section: "Getting Started"
+        - section: "How-To Guides"
           contents:
-            - docs/getting-started.qmd
-            - docs/choosing_method.qmd
-            - docs/installation.qmd
-            - docs/inference.qmd
-            - section: "Model Guides"
-              contents:
-                - docs/models/kimi-linear.qmd
-                - docs/models/plano.qmd
-                - docs/models/mimo.qmd
-                - docs/models/internvl3_5.qmd
-                - docs/models/olmo3.qmd
-                - docs/models/trinity.qmd
-                - docs/models/arcee.qmd
-                - section: "Ministral3"
-                  contents:
-                    - docs/models/ministral3.qmd
-                    - docs/models/ministral3/think.qmd
-                    - docs/models/ministral3/vision.qmd
-                - section: "Magistral"
-                  contents:
-                    - docs/models/magistral.qmd
-                    - docs/models/magistral/think.qmd
-                    - docs/models/magistral/vision.qmd
-                - docs/models/ministral.qmd
-                - docs/models/mistral-small.qmd
-                - docs/models/voxtral.qmd
-                - docs/models/devstral.qmd
-                - docs/models/mistral.qmd
-                - docs/models/llama-4.qmd
-                - docs/models/llama-2.qmd
-                - docs/models/qwen3-next.qmd
-                - docs/models/qwen3.qmd
-                - docs/models/gemma3n.qmd
-                - docs/models/apertus.qmd
-                - docs/models/gpt-oss.qmd
-                - docs/models/seed-oss.qmd
-                - docs/models/phi.qmd
-                - docs/models/smolvlm2.qmd
-                - docs/models/granite4.qmd
-                - docs/models/LiquidAI.qmd
-                - docs/models/hunyuan.qmd
-                - docs/models/jamba.qmd
-                - docs/models/orpheus.qmd
-
-            - docs/cli.qmd
-            - docs/telemetry.qmd
-            - docs/config-reference.qmd
-            - text: "API Reference"
-              href: docs/api
-
+          # TODO Edit folder structure after we have more docs.
+            - docs/debugging.qmd
+            - docs/multipack.qmd
+            - docs/fsdp_qlora.qmd
+            - docs/input_output.qmd
+            - docs/rlhf.qmd
+            - docs/nccl.qmd
+            - docs/mac.qmd
+            - docs/multi-node.qmd
+            - docs/unsloth.qmd
+            - docs/amd_hpc.qmd
         - section: "Dataset Formats"
           contents: docs/dataset-formats/*
-
-        - section: "Deployments"
+        - section: "Reference"
           contents:
-            - docs/docker.qmd
-            - docs/multi-gpu.qmd
-            - docs/multi-node.qmd
-            - docs/ray-integration.qmd
-            - docs/amd_hpc.qmd
-            - docs/mac.qmd
+            - docs/config.qmd
+        - docs/faq.qmd
 
-        - section: "How To Guides"
-          contents:
-            - docs/multimodal.qmd
-            - docs/rlhf.qmd
-            - docs/grpo.qmd
-            - docs/ebft.qmd
-            - docs/vllm_serving.qmd
-            - docs/reward_modelling.qmd
-            - docs/lr_groups.qmd
-            - docs/lora_optims.qmd
-            - docs/dataset_loading.qmd
-            - docs/qat.qmd
-            - docs/quantize.qmd
-            - docs/1_58bit_finetuning.qmd
-            - docs/optimizations.qmd
-
-        - section: "Core Concepts"
-          contents:
-            - docs/batch_vs_grad.qmd
-            - docs/dataset_preprocessing.qmd
-            - docs/streaming.qmd
-            - docs/multipack.qmd
-            - docs/mixed_precision.qmd
-            - docs/optimizers.qmd
-            - docs/attention.qmd
-
-        - section: "Advanced Features"
-          contents:
-            - docs/fsdp_qlora.qmd
-            - docs/torchao.qmd
-            - docs/custom_integrations.qmd
-            - docs/sequence_parallelism.qmd
-            - docs/gradient_checkpointing.qmd
-            - docs/nd_parallelism.qmd
-            - docs/expert_quantization.qmd
-
-        - section: "Troubleshooting"
-          contents:
-            - docs/faq.qmd
-            - docs/training_stability.qmd
-            - docs/debugging.qmd
-            - docs/nccl.qmd
 
 format:
   html:
-    theme: darkly
+    theme: materia
     css: styles.css
     toc: true
-    # Enable better handling of line breaks in markdown
-    preserve-tabs: true
-    html-math-method: mathjax
-    # Improved markdown processing options
-    md-extensions:
-      - markdown_it
-      - def_list
-      - attr_list
-      - fenced_divs
-      - tables
-      - html_admonition
-      - lineblocks
-      - fancy_lists
-    # Control whitespace handling
-    whitespace: preserve
-    # Process newlines in paragraphs
-    wrap: preserve
-    # Better line break handling
-    preserve-linebreaks: true

benchmarks/bench_entropy.py

@@ -1,208 +0,0 @@
-"""Benchmark for entropy_from_logits Triton kernel vs original chunked implementation.
-
-Usage: CUDA_VISIBLE_DEVICES=0 python benchmarks/bench_entropy.py
-"""
-
-import gc
-import statistics
-
-import torch
-import torch.nn.functional as F
-
-from axolotl.monkeypatch.trainer.utils import entropy_from_logits
-
-V = 151936  # Qwen vocab
-WARMUP = 5
-BENCH_ITERS = 20
-MEM_ITERS = 10
-
-
-def entropy_from_logits_original(logits: torch.Tensor, chunk_size: int = 128):
-    """Original chunked implementation (reference)."""
-    original_shape = logits.shape[:-1]
-    num_classes = logits.shape[-1]
-    flat_logits = logits.reshape(-1, num_classes)
-    entropies = []
-    for chunk in flat_logits.split(chunk_size, dim=0):
-        logps = F.log_softmax(chunk, dim=-1)
-        chunk_entropy = -(torch.exp(logps) * logps).sum(-1)
-        entropies.append(chunk_entropy)
-    return torch.cat(entropies, dim=0).reshape(original_shape)
-
-
-def _clean_gpu():
-    gc.collect()
-    torch.cuda.empty_cache()
-    torch.cuda.reset_peak_memory_stats()
-    torch.cuda.reset_accumulated_memory_stats()
-    torch.cuda.synchronize()
-
-
-def profile_time(fn, logits, n_iters=BENCH_ITERS):
-    for _ in range(WARMUP):
-        out = fn(logits, chunk_size=128)
-        del out
-    torch.cuda.synchronize()
-
-    times = []
-    for _ in range(n_iters):
-        s = torch.cuda.Event(enable_timing=True)
-        e = torch.cuda.Event(enable_timing=True)
-        s.record()
-        out = fn(logits, chunk_size=128)
-        e.record()
-        torch.cuda.synchronize()
-        times.append(s.elapsed_time(e))
-        del out
-    return times
-
-
-def profile_memory(fn, logits, n_iters=MEM_ITERS):
-    for _ in range(WARMUP):
-        out = fn(logits, chunk_size=128)
-        del out
-    torch.cuda.synchronize()
-
-    peaks = []
-    for _ in range(n_iters):
-        _clean_gpu()
-        base = torch.cuda.max_memory_allocated()
-        out = fn(logits, chunk_size=128)
-        torch.cuda.synchronize()
-        peaks.append(torch.cuda.max_memory_allocated() - base)
-        del out
-    return [p / 1e6 for p in peaks]
-
-
-def fmt(values, unit=""):
-    mean = statistics.mean(values)
-    std = statistics.stdev(values) if len(values) > 1 else 0.0
-    return f"{mean:8.2f} ± {std:5.2f} {unit}  [min={min(values):.2f}, max={max(values):.2f}]"
-
-
-def benchmark_contiguous():
-    print("=" * 60)
-    print(
-        f"CONTIGUOUS BENCHMARK  (warmup={WARMUP}, time={BENCH_ITERS}, mem={MEM_ITERS})"
-    )
-    print("=" * 60)
-
-    configs = [
-        (1, 2048),
-        (1, 8192),
-        (1, 16384),
-        (4, 4096),
-        (8, 2048),
-        (16, 2048),
-        (16, 4096),
-    ]
-
-    for B, L in configs:
-        mem_gb = B * L * V * 2 / 1e9
-        if mem_gb > 28:
-            print(f"\n  skip B={B}, L={L} ({mem_gb:.1f} GB)")
-            continue
-
-        N = B * L
-        print(f"\n{'─' * 60}")
-        print(f"B={B:2d}, L={L:5d}  ({N:6d} rows, logits {mem_gb:.2f} GB)")
-        print(f"{'─' * 60}")
-
-        torch.manual_seed(42)
-        logits = torch.randn(B, L, V, device="cuda", dtype=torch.bfloat16)
-
-        t_orig = profile_time(entropy_from_logits_original, logits)
-        t_triton = profile_time(entropy_from_logits, logits)
-        orig_mean = statistics.mean(t_orig)
-        triton_mean = statistics.mean(t_triton)
-
-        print("  TIME (ms):")
-        print(f"    original: {fmt(t_orig, 'ms')}")
-        print(f"    triton:   {fmt(t_triton, 'ms')}")
-        print(f"    speedup:  {orig_mean / triton_mean:.2f}x")
-
-        m_orig = profile_memory(entropy_from_logits_original, logits)
-        m_triton = profile_memory(entropy_from_logits, logits)
-        orig_peak = statistics.mean(m_orig)
-        triton_peak = statistics.mean(m_triton)
-
-        print("  MEMORY (peak overhead):")
-        print(f"    original: {fmt(m_orig, 'MB')}")
-        print(f"    triton:   {fmt(m_triton, 'MB')}")
-        print(f"    saved:    {orig_peak - triton_peak:.1f} MB")
-
-        del logits
-        _clean_gpu()
-
-
-def benchmark_noncontiguous():
-    print("\n" + "=" * 60)
-    print(
-        f"NON-CONTIGUOUS BENCHMARK  (warmup={WARMUP}, time={BENCH_ITERS}, mem={MEM_ITERS})"
-    )
-    print("=" * 60)
-
-    configs = [
-        (4, 2048, "transpose"),
-        (4, 8192, "transpose"),
-        (8, 2048, "transpose"),
-        (4, 4096, "slice_batch"),
-    ]
-
-    for B, L, method in configs:
-        torch.manual_seed(42)
-
-        if method == "transpose":
-            raw = torch.randn(L, B, V, device="cuda", dtype=torch.bfloat16)
-            logits_nc = raw.transpose(0, 1)
-            raw_gb = L * B * V * 2 / 1e9
-        elif method == "slice_batch":
-            raw = torch.randn(B * 2, L, V, device="cuda", dtype=torch.bfloat16)
-            logits_nc = raw[::2]
-            raw_gb = B * 2 * L * V * 2 / 1e9
-        else:
-            continue
-
-        if raw_gb > 28:
-            print(f"\n  skip B={B}, L={L}, {method} ({raw_gb:.1f} GB)")
-            del raw, logits_nc
-            torch.cuda.empty_cache()
-            continue
-
-        N = B * L
-        print(f"\n{'─' * 60}")
-        print(f"B={B}, L={L}  {method}  ({N} rows, raw {raw_gb:.2f} GB)")
-        print(f"{'─' * 60}")
-
-        def original_with_copy(logits, chunk_size=128):
-            return entropy_from_logits_original(
-                logits.contiguous(), chunk_size=chunk_size
-            )
-
-        t_orig = profile_time(original_with_copy, logits_nc)
-        t_triton = profile_time(entropy_from_logits, logits_nc)
-        orig_mean = statistics.mean(t_orig)
-        triton_mean = statistics.mean(t_triton)
-
-        print("  TIME (ms):")
-        print(f"    orig+copy:     {fmt(t_orig, 'ms')}")
-        print(f"    triton-strided:{fmt(t_triton, 'ms')}")
-        print(f"    speedup:       {orig_mean / triton_mean:.2f}x")
-
-        m_orig = profile_memory(original_with_copy, logits_nc)
-        m_triton = profile_memory(entropy_from_logits, logits_nc)
-        orig_peak = statistics.mean(m_orig)
-        triton_peak = statistics.mean(m_triton)
-
-        print("  MEMORY (peak overhead):")
-        print(f"    orig+copy:     {fmt(m_orig, 'MB')}")
-        print(f"    triton-strided:{fmt(m_triton, 'MB')}")
-        print(f"    saved:         {orig_peak - triton_peak:.1f} MB")
-
-        del raw, logits_nc
-        _clean_gpu()
-
-
-if __name__ == "__main__":
-    benchmark_contiguous()
-    benchmark_noncontiguous()

benchmarks/bench_scattermoe_lora.py

@@ -1,284 +0,0 @@
-"""Benchmark for ScatterMoE LoRA Triton kernels.
-
-Measures forward, backward dX, and backward dA/dB kernels at common MoE
-model shapes. Reports per-kernel timings, LoRA overhead vs base scatter2scatter,
-and full fwd+bwd autograd throughput.
-
-Usage:
-  CUDA_VISIBLE_DEVICES=0 python benchmarks/bench_scattermoe_lora.py
-  CUDA_VISIBLE_DEVICES=0 python benchmarks/bench_scattermoe_lora.py --ranks 16 64
-  CUDA_VISIBLE_DEVICES=0 python benchmarks/bench_scattermoe_lora.py --models Qwen/Qwen3.5-35B-A3B
-"""
-
-import argparse
-import gc
-import time
-from functools import partial
-
-import torch
-
-from axolotl.integrations.kernels.libs.scattermoe_lora.kernels import (
-    lora_ops,
-    ops as base_ops,
-)
-from axolotl.integrations.kernels.libs.scattermoe_lora.parallel_experts import (
-    flatten_sort_count,
-)
-from axolotl.integrations.kernels.libs.scattermoe_lora.parallel_linear_lora import (
-    ScatterMoELoRA,
-)
-
-DEVICE = "cuda"
-DTYPE = torch.bfloat16
-WARMUP = 5
-ITERS = 20
-
-# ─── Model configs ──────────────────────────────────────────────────────────
-
-BUILTIN_CONFIGS = {
-    "Qwen3.5-35B-A3B": (256, 2048, 512, 8),  # E, H, I, k
-    "Qwen3-30B-A3B": (128, 2048, 768, 8),
-    "OLMoE-1B-7B": (64, 2048, 1024, 8),
-    "Mixtral-8x7B": (8, 4096, 14336, 2),
-}
-
-
-def _resolve_config(spec):
-    """Resolve a model spec to (E, H, I, k). Accepts builtin names or HF IDs."""
-    key = spec.lower().replace("/", "-")
-    for name, cfg in BUILTIN_CONFIGS.items():
-        if key in name.lower() or name.lower() in key:
-            return name, cfg
-
-    from transformers import AutoConfig
-
-    hf_cfg = AutoConfig.from_pretrained(spec, trust_remote_code=True)
-    if callable(getattr(hf_cfg, "get_text_config", None)):
-        tc = hf_cfg.get_text_config()
-        if hasattr(tc, "model_type") and tc.model_type != hf_cfg.model_type:
-            hf_cfg = tc
-    hidden = hf_cfg.hidden_size
-    inter = getattr(hf_cfg, "moe_intermediate_size", None) or hf_cfg.intermediate_size
-    experts = (
-        getattr(hf_cfg, "num_experts", None)
-        or getattr(hf_cfg, "num_local_experts", None)
-        or getattr(hf_cfg, "n_routed_experts", None)
-    )
-    top_k = (
-        getattr(hf_cfg, "num_experts_per_tok", None)
-        or getattr(hf_cfg, "num_experts_per_token", None)
-        or 2
-    )
-    name = spec.split("/")[-1]
-    return name, (experts, hidden, inter, top_k)
-
-
-# ─── Benchmark helpers ──────────────────────────────────────────────────────
-
-
-def _clean():
-    gc.collect()
-    torch.cuda.empty_cache()
-    torch.cuda.synchronize()
-
-
-def _bench(fn, warmup=WARMUP, iters=ITERS):
-    for _ in range(warmup):
-        fn()
-    torch.cuda.synchronize()
-    times = []
-    for _ in range(iters):
-        torch.cuda.synchronize()
-        t0 = time.perf_counter()
-        fn()
-        torch.cuda.synchronize()
-        times.append((time.perf_counter() - t0) * 1000)
-    times.sort()
-    return times[len(times) // 2]
-
-
-def _setup(num_experts, K, N, T, top_k, R):
-    torch.manual_seed(42)
-    x = torch.randn(T, K, device=DEVICE, dtype=DTYPE)
-    W = torch.randn(num_experts, K, N, device=DEVICE, dtype=DTYPE) * 0.02
-    lora_A = torch.randn(R * num_experts, K, device=DEVICE, dtype=DTYPE) * 0.01
-    lora_B = torch.randn(N, R * num_experts, device=DEVICE, dtype=DTYPE) * 0.01
-    logits = torch.randn(T, num_experts, device=DEVICE)
-    _, top_idx = torch.topk(torch.softmax(logits, dim=-1), top_k, dim=-1)
-    sei, ssi, eo = flatten_sort_count(top_idx, num_experts)
-    gx = base_ops.group(x, ssi, fan_out=top_k)
-    dy = torch.randn(gx.size(0), N, device=DEVICE, dtype=DTYPE)
-    return x, W, lora_A, lora_B, sei, ssi, eo, gx, dy
-
-
-# ─── Kernel wrappers (avoid B023 loop-variable capture) ──────────────────────
-
-
-def _call_fwd(x, W, sei, ssi, top_k, lA, lB):
-    return lora_ops.scatter2scatter_lora(
-        X=x,
-        W=W,
-        sorted_expert_idxs=sei,
-        sorted_scattered_idxs=ssi,
-        k=top_k,
-        lora_A=lA,
-        lora_B=lB,
-        scaling=2.0,
-    )
-
-
-def _call_base(x, W, sei, ssi, top_k):
-    return base_ops.scatter2scatter(
-        X=x,
-        W=W,
-        sorted_expert_idxs=sei,
-        sorted_scattered_idxs=ssi,
-        k=top_k,
-    )
-
-
-def _call_dx(dy, W, sei, ssi, lA, lB):
-    return lora_ops.scatter2scatter_lora_dX(
-        DY=dy,
-        W=W,
-        sorted_expert_idxs=sei,
-        sorted_scattered_idxs=ssi,
-        k=1,
-        lora_A=lA,
-        lora_B=lB,
-        scaling=2.0,
-        dy_grouped=True,
-        dx_grouped=False,
-    )
-
-
-def _call_bwd(dy, gx, lA, lB, eo, num_experts):
-    return lora_ops.group_bwd_lora(
-        DY=dy,
-        X=gx,
-        lora_A=lA,
-        lora_B=lB,
-        expert_offsets=eo,
-        E=num_experts,
-        scaling=2.0,
-    )
-
-
-# ─── Main ────────────────────────────────────────────────────────────────────
-
-
-def main():
-    parser = argparse.ArgumentParser(description="ScatterMoE LoRA kernel benchmark")
-    parser.add_argument(
-        "--models",
-        "-m",
-        nargs="+",
-        help="Model names or HF IDs (default: all builtins)",
-    )
-    parser.add_argument("--ranks", "-r", nargs="+", type=int, default=[16, 32, 64])
-    parser.add_argument("--seq-len", "-T", type=int, default=2048)
-    args = parser.parse_args()
-
-    T = args.seq_len
-    print(f"GPU: {torch.cuda.get_device_name()}")
-    print(f"T={T}, ranks={args.ranks}\n")
-
-    if args.models:
-        configs = [_resolve_config(m) for m in args.models]
-    else:
-        configs = list(BUILTIN_CONFIGS.items())
-
-    for model_name, (num_experts, hidden, inter, top_k) in configs:
-        print(f"{'=' * 70}")
-        print(f"  {model_name}: E={num_experts}, H={hidden}, I={inter}, k={top_k}")
-        print(f"{'=' * 70}")
-
-        for R in args.ranks:
-            for proj, K, N in [("gate_up", hidden, 2 * inter), ("down", inter, hidden)]:
-                _clean()
-                x, W, lA, lB, sei, ssi, eo, gx, dy = _setup(
-                    num_experts, K, N, T, top_k, R
-                )
-
-                # Forward with LoRA (auto-dispatched: fused or split)
-                dispatch = (
-                    "split"
-                    if (
-                        num_experts <= lora_ops._SPLIT_LORA_FWD_MAX_EXPERTS
-                        and K * N >= lora_ops._SPLIT_LORA_FWD_THRESHOLD
-                    )
-                    else "fused"
-                )
-                t_fwd = _bench(partial(_call_fwd, x, W, sei, ssi, top_k, lA, lB))
-                t_base = _bench(partial(_call_base, x, W, sei, ssi, top_k))
-                t_dx = _bench(partial(_call_dx, dy, W, sei, ssi, lA, lB))
-                t_bwd = _bench(partial(_call_bwd, dy, gx, lA, lB, eo, num_experts))
-
-                total = t_fwd + t_dx + t_bwd
-                overhead = t_fwd / t_base - 1 if t_base > 0 else 0
-
-                print(
-                    f"  R={R:>2} {proj:<8}  "
-                    f"fwd={t_fwd:>6.2f}ms [{dispatch}]  "
-                    f"base={t_base:>6.2f}ms "
-                    f"(+{overhead * 100:.0f}%)  "
-                    f"dx={t_dx:>6.2f}ms  bwd={t_bwd:>6.2f}ms  "
-                    f"total={total:>6.2f}ms"
-                )
-
-                # Full autograd fwd+bwd with memory measurement
-                x_ag = x.clone().requires_grad_(True)
-                lA_ag = lA.clone().requires_grad_(True)
-                lB_ag = lB.clone().requires_grad_(True)
-
-                def _run_autograd(
-                    _x=x_ag,
-                    _W=W,
-                    _k=top_k,
-                    _sei=sei,
-                    _ssi=ssi,
-                    _eo=eo,
-                    _lA=lA_ag,
-                    _lB=lB_ag,
-                ):
-                    out = ScatterMoELoRA.apply(
-                        _x,
-                        _W,
-                        _k,
-                        _sei,
-                        _ssi,
-                        _eo,
-                        _lA,
-                        _lB,
-                        2.0,
-                        None,
-                        None,
-                        False,
-                        False,
-                        True,
-                        False,
-                    )
-                    out.sum().backward()
-                    _x.grad = None
-                    _lA.grad = None
-                    _lB.grad = None
-
-                t_full = _bench(_run_autograd)
-
-                _clean()
-                torch.cuda.reset_peak_memory_stats()
-                mem_before = torch.cuda.memory_allocated()
-                _run_autograd()
-                torch.cuda.synchronize()
-                mem_peak = torch.cuda.max_memory_allocated() - mem_before
-
-                print(
-                    f"         full_fwd_bwd={t_full:>6.2f}ms  "
-                    f"peak_delta={mem_peak / 1e6:>6.1f}MB"
-                )
-
-        print()
-
-
-if __name__ == "__main__":
-    main()

benchmarks/bench_selective_logsoftmax.py

@@ -1,191 +0,0 @@
-"""Benchmark for selective_log_softmax Triton kernel vs original implementation.
-
-Usage: CUDA_VISIBLE_DEVICES=0 python benchmarks/bench_selective_logsoftmax.py
-"""
-
-import gc
-import statistics
-
-import torch
-
-from axolotl.monkeypatch.trainer.utils import (
-    selective_log_softmax,
-    selective_log_softmax_original,
-)
-
-V = 151936  # Qwen vocab
-WARMUP = 5
-BENCH_ITERS = 20
-MEM_ITERS = 10
-
-
-def _clean_gpu():
-    gc.collect()
-    torch.cuda.empty_cache()
-    torch.cuda.reset_peak_memory_stats()
-    torch.cuda.reset_accumulated_memory_stats()
-    torch.cuda.synchronize()
-
-
-def profile_time(fn, args, n_iters=BENCH_ITERS):
-    for _ in range(WARMUP):
-        fn(*args)
-    torch.cuda.synchronize()
-
-    times = []
-    for _ in range(n_iters):
-        s = torch.cuda.Event(enable_timing=True)
-        e = torch.cuda.Event(enable_timing=True)
-        s.record()
-        fn(*args)
-        e.record()
-        torch.cuda.synchronize()
-        times.append(s.elapsed_time(e))
-    return times
-
-
-def profile_memory(fn, args, n_iters=MEM_ITERS):
-    for _ in range(WARMUP):
-        out = fn(*args)
-        del out
-    torch.cuda.synchronize()
-
-    peaks = []
-    for _ in range(n_iters):
-        _clean_gpu()
-        base = torch.cuda.max_memory_allocated()
-        out = fn(*args)
-        torch.cuda.synchronize()
-        peaks.append(torch.cuda.max_memory_allocated() - base)
-        del out
-    return [p / 1e6 for p in peaks]
-
-
-def fmt(values, unit=""):
-    mean = statistics.mean(values)
-    std = statistics.stdev(values) if len(values) > 1 else 0.0
-    return f"{mean:8.2f} ± {std:5.2f} {unit}  [min={min(values):.2f}, max={max(values):.2f}]"
-
-
-def benchmark_forward():
-    print("=" * 60)
-    print(f"FORWARD BENCHMARK  (warmup={WARMUP}, time={BENCH_ITERS}, mem={MEM_ITERS})")
-    print("=" * 60)
-
-    configs = [
-        (1, 2048),
-        (1, 8192),
-        (4, 4096),
-        (8, 2048),
-        (16, 2048),
-        (16, 4096),
-    ]
-
-    for B, L in configs:
-        mem_gb = B * L * V * 2 / 1e9
-        if mem_gb > 28:
-            print(f"\n  skip B={B}, L={L} ({mem_gb:.1f} GB)")
-            continue
-
-        N = B * L
-        print(f"\n{'─' * 60}")
-        print(f"B={B:2d}, L={L:5d}  ({N:6d} rows, logits {mem_gb:.2f} GB)")
-        print(f"{'─' * 60}")
-
-        torch.manual_seed(42)
-        logits = torch.randn(B, L, V, device="cuda", dtype=torch.bfloat16)
-        index = torch.randint(0, V, (B, L), device="cuda")
-
-        t_orig = profile_time(selective_log_softmax_original, (logits, index))
-        t_triton = profile_time(selective_log_softmax, (logits, index))
-        orig_mean = statistics.mean(t_orig)
-        triton_mean = statistics.mean(t_triton)
-
-        print("  TIME (ms):")
-        print(f"    original: {fmt(t_orig, 'ms')}")
-        print(f"    triton:   {fmt(t_triton, 'ms')}")
-        print(f"    speedup:  {orig_mean / triton_mean:.2f}x")
-
-        m_orig = profile_memory(selective_log_softmax_original, (logits, index))
-        m_triton = profile_memory(selective_log_softmax, (logits, index))
-        orig_peak = statistics.mean(m_orig)
-        triton_peak = statistics.mean(m_triton)
-
-        print("  MEMORY (peak overhead):")
-        print(f"    original: {fmt(m_orig, 'MB')}")
-        print(f"    triton:   {fmt(m_triton, 'MB')}")
-        print(f"    saved:    {orig_peak - triton_peak:.1f} MB")
-
-        del logits, index
-        _clean_gpu()
-
-
-def benchmark_backward():
-    print("\n" + "=" * 60)
-    print(f"FWD+BWD BENCHMARK  (warmup={WARMUP}, time={BENCH_ITERS}, mem={MEM_ITERS})")
-    print("=" * 60)
-
-    configs = [
-        (1, 2048),
-        (1, 8192),
-        (4, 4096),
-        (8, 2048),
-        (16, 2048),
-        (16, 4096),
-    ]
-
-    def fwd_bwd_original(logits, index):
-        logits.grad = None
-        out = selective_log_softmax_original(logits, index)
-        out.sum().backward()
-
-    def fwd_bwd_triton(logits, index):
-        logits.grad = None
-        out = selective_log_softmax(logits, index)
-        out.sum().backward()
-
-    for B, L in configs:
-        mem_gb = B * L * V * 2 / 1e9
-        if mem_gb > 20:
-            print(f"\n  skip B={B}, L={L} ({mem_gb:.1f} GB, need room for grads)")
-            continue
-
-        N = B * L
-        print(f"\n{'─' * 60}")
-        print(f"B={B:2d}, L={L:5d}  ({N:6d} rows, logits {mem_gb:.2f} GB)")
-        print(f"{'─' * 60}")
-
-        torch.manual_seed(42)
-        logits_orig = torch.randn(
-            B, L, V, device="cuda", dtype=torch.bfloat16, requires_grad=True
-        )
-        logits_tri = logits_orig.detach().clone().requires_grad_(True)
-        index = torch.randint(0, V, (B, L), device="cuda")
-
-        t_orig = profile_time(fwd_bwd_original, (logits_orig, index))
-        t_triton = profile_time(fwd_bwd_triton, (logits_tri, index))
-        orig_mean = statistics.mean(t_orig)
-        triton_mean = statistics.mean(t_triton)
-
-        print("  FWD+BWD TIME (ms):")
-        print(f"    original: {fmt(t_orig, 'ms')}")
-        print(f"    triton:   {fmt(t_triton, 'ms')}")
-        print(f"    speedup:  {orig_mean / triton_mean:.2f}x")
-
-        m_orig = profile_memory(fwd_bwd_original, (logits_orig, index))
-        m_triton = profile_memory(fwd_bwd_triton, (logits_tri, index))
-        orig_peak = statistics.mean(m_orig)
-        triton_peak = statistics.mean(m_triton)
-
-        print("  FWD+BWD MEMORY (peak overhead):")
-        print(f"    original: {fmt(m_orig, 'MB')}")
-        print(f"    triton:   {fmt(m_triton, 'MB')}")
-        print(f"    saved:    {orig_peak - triton_peak:.1f} MB")
-
-        del logits_orig, logits_tri, index
-        _clean_gpu()
-
-
-if __name__ == "__main__":
-    benchmark_forward()
-    benchmark_backward()

cicd/Dockerfile-uv.jinja

@@ -1,55 +0,0 @@
-FROM axolotlai/axolotl-base-uv:{{ BASE_TAG }}
-
-ENV TORCH_CUDA_ARCH_LIST="7.0 7.5 8.0 8.6 9.0+PTX"
-ENV AXOLOTL_EXTRAS="{{ AXOLOTL_EXTRAS }}"
-ENV AXOLOTL_ARGS="{{ AXOLOTL_ARGS }}"
-ENV CUDA="{{ CUDA }}"
-ENV PYTORCH_VERSION="{{ PYTORCH_VERSION }}"
-ENV GITHUB_REF="{{ GITHUB_REF }}"
-ENV GITHUB_SHA="{{ GITHUB_SHA }}"
-ENV NIGHTLY_BUILD="{{ NIGHTLY_BUILD }}"
-ENV HF_HOME="{{ HF_HOME }}"
-
-RUN apt-get update && \
-    apt-get install -y --allow-change-held-packages vim curl nano zstd libnccl2 libnccl-dev ibverbs-providers ibverbs-utils infiniband-diags librdmacm-dev librdmacm1 rdmacm-utils slurm-wlm
-
-WORKDIR /workspace
-
-RUN git clone --depth=1 https://github.com/axolotl-ai-cloud/axolotl.git
-
-WORKDIR /workspace/axolotl
-
-RUN git fetch origin +$GITHUB_REF && \
-    git checkout FETCH_HEAD
-
-RUN uv pip install packaging==26.0 setuptools==78.1.1
-RUN uv pip install torchvision
-RUN uv pip uninstall causal_conv1d
-RUN if [ "$AXOLOTL_EXTRAS" != "" ] ; then \
-        uv pip install --no-build-isolation -e .[deepspeed,flash-attn,ring-flash-attn,optimizers,ray,$AXOLOTL_EXTRAS] $AXOLOTL_ARGS; \
-    else \
-        uv pip install --no-build-isolation -e .[deepspeed,flash-attn,ring-flash-attn,optimizers,ray] $AXOLOTL_ARGS; \
-    fi
-
-# Override with nightly HF packages for nightly builds
-RUN if [ "$NIGHTLY_BUILD" = "true" ] ; then \
-        uv pip install --no-deps \
-            "transformers @ git+https://github.com/huggingface/transformers.git@main" \
-            "peft @ git+https://github.com/huggingface/peft.git@main" \
-            "accelerate @ git+https://github.com/huggingface/accelerate.git@main" \
-            "trl @ git+https://github.com/huggingface/trl.git@main" \
-            "datasets @ git+https://github.com/huggingface/datasets.git@main"; \
-    fi
-
-RUN python scripts/cutcrossentropy_install.py --uv | sh
-
-# So we can test the Docker image
-RUN uv pip install black mypy pre-commit types-requests quartodoc jupyter blobfile tiktoken \
-    codecov codecov-cli pytest pytest-cov pytest-retry pytest-sugar pytest-xdist tbparse
-
-# fix so that git fetch/pull from remote works
-RUN git config remote.origin.fetch "+refs/heads/*:refs/remotes/origin/*" && \
-    git config --get remote.origin.fetch
-
-# helper for huggingface-login cli
-RUN git config --global credential.helper store

cicd/Dockerfile.jinja

@@ -0,0 +1,50 @@
+FROM winglian/axolotl-base:{{ BASE_TAG }}
+
+ENV TORCH_CUDA_ARCH_LIST="7.0 7.5 8.0 8.6+PTX"
+ENV AXOLOTL_EXTRAS="{{ AXOLOTL_EXTRAS }}"
+ENV AXOLOTL_ARGS="{{ AXOLOTL_ARGS }}"
+ENV CUDA="{{ CUDA }}"
+ENV PYTORCH_VERSION="{{ PYTORCH_VERSION }}"
+ENV GITHUB_REF="{{ GITHUB_REF }}"
+ENV GITHUB_SHA="{{ GITHUB_SHA }}"
+ENV NIGHTLY_BUILD="{{ NIGHTLY_BUILD }}"
+
+RUN apt-get update && \
+    apt-get install -y --allow-change-held-packages vim curl nano libnccl2 libnccl-dev
+
+WORKDIR /workspace
+
+RUN git clone --depth=1 https://github.com/axolotl-ai-cloud/axolotl.git
+
+WORKDIR /workspace/axolotl
+
+RUN git fetch origin +$GITHUB_REF && \
+    git checkout FETCH_HEAD
+
+# If AXOLOTL_EXTRAS is set, append it in brackets
+RUN if [ "$NIGHTLY_BUILD" = "true" ] ; then \
+        sed -i 's#^transformers.*#transformers @ git+https://github.com/huggingface/transformers.git@main#' requirements.txt; \
+        sed -i 's#^peft.*#peft @ git+https://github.com/huggingface/peft.git@main#' requirements.txt; \
+        sed -i 's#^accelerate.*#accelerate @ git+https://github.com/huggingface/accelerate.git@main#' requirements.txt; \
+        sed -i 's#^trl.*#trl @ git+https://github.com/huggingface/trl.git@main#' requirements.txt; \
+        sed -i 's#^datasets.*#datasets @ git+https://github.com/huggingface/datasets.git@main#' requirements.txt; \
+    fi
+
+RUN if [ "$AXOLOTL_EXTRAS" != "" ] ; then \
+        pip install -e .[deepspeed,flash-attn,optimizers,$AXOLOTL_EXTRAS] $AXOLOTL_ARGS; \
+    else \
+        pip install -e .[deepspeed,flash-attn,optimizers] $AXOLOTL_ARGS; \
+    fi
+
+RUN python scripts/unsloth_install.py | sh
+RUN python scripts/cutcrossentropy_install.py | sh
+
+# So we can test the Docker image
+RUN pip install -r requirements-dev.txt -r requirements-tests.txt
+
+# fix so that git fetch/pull from remote works
+RUN git config remote.origin.fetch "+refs/heads/*:refs/remotes/origin/*" && \
+    git config --get remote.origin.fetch
+
+# helper for huggingface-login cli
+RUN git config --global credential.helper store

cicd/cicd.sh

@@ -1,73 +1,7 @@
 #!/bin/bash
 set -e
 
-python -c "import torch; assert '$PYTORCH_VERSION' in torch.__version__, f'Expected torch $PYTORCH_VERSION but got {torch.__version__}'"
-
-set -o pipefail
-for i in 1 2 3; do
-  if curl --silent --show-error --fail -L \
-    https://axolotl-ci.b-cdn.net/hf-cache.tar.zst \
-    | tar -xpf - -C "${HF_HOME}/hub/" --use-compress-program unzstd --strip-components=1; then
-    echo "HF cache extracted successfully"
-    break
-  fi
-  echo "Attempt $i failed, cleaning up and retrying in 15s..."
-  rm -rf "${HF_HOME}/hub/"*
-  sleep 15
-done
-# hf download "NousResearch/Meta-Llama-3-8B"
-# hf download "NousResearch/Meta-Llama-3-8B-Instruct"
-# hf download "microsoft/Phi-4-reasoning"
-# hf download "microsoft/Phi-3.5-mini-instruct"
-# hf download "microsoft/Phi-3-medium-128k-instruct"
-
-# Run unit tests with initial coverage report
-pytest -v --durations=10 -n8 \
-  --ignore=tests/e2e/ \
-  --ignore=tests/patched/ \
-  --ignore=tests/cli \
-  /workspace/axolotl/tests/ \
-  --cov=axolotl
-
-# Run lora kernels tests with coverage append
-pytest -v --durations=10 \
-  /workspace/axolotl/tests/e2e/patched/lora_kernels \
-  --cov=axolotl \
-  --cov-append
-
-# Run patched tests excluding lora kernels with coverage append
-pytest --full-trace -vvv --durations=10 \
-  --ignore=tests/e2e/patched/lora_kernels \
-  /workspace/axolotl/tests/e2e/patched \
-  --cov=axolotl \
-  --cov-append
-
-# Run solo tests with coverage append
-pytest -v --durations=10 -n1 \
-  /workspace/axolotl/tests/e2e/solo/ \
-  --cov=axolotl \
-  --cov-append
-
-# Run integration tests with coverage append
-pytest -v --durations=10 \
-  /workspace/axolotl/tests/e2e/integrations/ \
-  --cov=axolotl \
-  --cov-append
-
-pytest -v --durations=10 /workspace/axolotl/tests/cli \
-  --cov=axolotl \
-  --cov-append
-
-# Run remaining e2e tests with coverage append and final report
-pytest -v --durations=10 \
-  --ignore=tests/e2e/solo/ \
-  --ignore=tests/e2e/patched/ \
-  --ignore=tests/e2e/multigpu/ \
-  --ignore=tests/e2e/integrations/ \
-  --ignore=tests/cli \
-  /workspace/axolotl/tests/e2e/ \
-  --cov=axolotl \
-  --cov-append \
-  --cov-report=xml:e2e-coverage.xml
-
-codecov upload-process -t $CODECOV_TOKEN -f e2e-coverage.xml -F e2e,pytorch-${PYTORCH_VERSION} || true
+pytest -v --durations=10 -n8 --ignore=tests/e2e/ --ignore=tests/patched/ /workspace/axolotl/tests/
+pytest -v --durations=10 -n1 --dist loadfile /workspace/axolotl/tests/patched/
+pytest -v --durations=10 -n1 --dist loadfile /workspace/axolotl/tests/e2e/patched/ /workspace/axolotl/tests/e2e/integrations/
+pytest -v --durations=10 --ignore=tests/e2e/patched/ --ignore=tests/e2e/multigpu/ --ignore=tests/e2e/integrations/ /workspace/axolotl/tests/e2e/

cicd/cleanup.py

@@ -1,19 +0,0 @@
-"""Modal app to run axolotl GPU cleanup"""
-
-from .single_gpu import VOLUME_CONFIG, app, cicd_image, run_cmd
-
-
-@app.function(
-    image=cicd_image,
-    timeout=60 * 60,
-    cpu=8.0,
-    memory=131072,
-    volumes=VOLUME_CONFIG,
-)
-def cleanup():
-    run_cmd("./cicd/cleanup.sh", "/workspace/axolotl")
-
-
-@app.local_entrypoint()
-def main():
-    cleanup.remote()

cicd/cleanup.sh

@@ -1,6 +0,0 @@
-#!/bin/bash
-set -e
-
-# cleanup old cache files for datasets processing and intermediate mappings
-find /workspace/data/huggingface-cache/hub/datasets -name "cache-*" -type f -mtime +1 -exec rm {} \;
-find /workspace/data/huggingface-cache/hub/datasets -name "*.lock" -type f -mtime +1 -exec rm {} \;

cicd/e2e_tests.py

@@ -1,20 +0,0 @@
-"""Modal app to run axolotl GPU tests"""
-
-from .single_gpu import GPU_CONFIG, VOLUME_CONFIG, app, cicd_image, run_cmd
-
-
-@app.function(
-    image=cicd_image,
-    gpu=GPU_CONFIG,
-    timeout=120 * 60,  # 90 min
-    cpu=8.0,
-    memory=131072,
-    volumes=VOLUME_CONFIG,
-)
-def cicd_pytest():
-    run_cmd("./cicd/cicd.sh", "/workspace/axolotl")
-
-
-@app.local_entrypoint()
-def main():
-    cicd_pytest.remote()

cicd/multigpu.py

@@ -1,6 +1,7 @@
 """
-modal application to run axolotl gpu tests in Modal
-"""
+ modal application to run axolotl gpu tests in Modal
+ """
+# pylint: disable=duplicate-code
 
 import os
 import pathlib
@@ -17,22 +18,16 @@ template_loader = jinja2.FileSystemLoader(searchpath=cicd_path)
 template_env = jinja2.Environment(
     loader=template_loader, autoescape=select_autoescape()
 )
-dockerfile = os.environ.get("E2E_DOCKERFILE", "Dockerfile-uv.jinja")
-df_template = template_env.get_template(dockerfile)
+df_template = template_env.get_template("Dockerfile.jinja")
 
 df_args = {
     "AXOLOTL_EXTRAS": os.environ.get("AXOLOTL_EXTRAS", ""),
     "AXOLOTL_ARGS": os.environ.get("AXOLOTL_ARGS", ""),
-    "PYTORCH_VERSION": os.environ.get("PYTORCH_VERSION", "2.6.0"),
-    "BASE_TAG": os.environ.get("BASE_TAG", "main-base-py3.11-cu126-2.6.0"),
-    "CUDA": os.environ.get("CUDA", "126"),
+    "PYTORCH_VERSION": os.environ.get("PYTORCH_VERSION", "2.3.1"),
+    "BASE_TAG": os.environ.get("BASE_TAG", "main-base-py3.11-cu121-2.3.1"),
+    "CUDA": os.environ.get("CUDA", "121"),
     "GITHUB_REF": os.environ.get("GITHUB_REF", "refs/heads/main"),
     "GITHUB_SHA": os.environ.get("GITHUB_SHA", ""),
-    "NIGHTLY_BUILD": os.environ.get("NIGHTLY_BUILD", ""),
-    "CODECOV_TOKEN": os.environ.get("CODECOV_TOKEN", ""),
-    "HF_HOME": "/workspace/data/huggingface-cache/hub",
-    "PYTHONUNBUFFERED": os.environ.get("PYTHONUNBUFFERED", "1"),
-    "DEEPSPEED_LOG_LEVEL": os.environ.get("DEEPSPEED_LOG_LEVEL", "WARNING"),
 }
 
 dockerfile_contents = df_template.render(**df_args)
@@ -41,23 +36,21 @@ temp_dir = tempfile.mkdtemp()
 with open(pathlib.Path(temp_dir) / "Dockerfile", "w", encoding="utf-8") as f:
     f.write(dockerfile_contents)
 
-cicd_image = Image.from_dockerfile(
-    pathlib.Path(temp_dir) / "Dockerfile",
-    force_build=True,
-    gpu="A10G",
-).env(df_args)
+cicd_image = (
+    Image.from_dockerfile(
+        pathlib.Path(temp_dir) / "Dockerfile",
+        force_build=True,
+        gpu="A10G",
+    )
+    .env(df_args)
+    .pip_install("fastapi==0.110.0", "pydantic==2.6.3")
+)
 
 app = App("Axolotl CI/CD", secrets=[])
 
-hf_cache_volume = modal.Volume.from_name(
-    "axolotl-ci-hf-hub-cache", create_if_missing=True
-)
-VOLUME_CONFIG = {
-    "/workspace/data/huggingface-cache/hub": hf_cache_volume,
-}
 
 N_GPUS = int(os.environ.get("N_GPUS", 2))
-GPU_CONFIG = f"H100:{N_GPUS}"
+GPU_CONFIG = modal.gpu.H100(count=N_GPUS)
 
 
 def run_cmd(cmd: str, run_folder: str):
@@ -65,16 +58,15 @@ def run_cmd(cmd: str, run_folder: str):
 
     # Propagate errors from subprocess.
     if exit_code := subprocess.call(cmd.split(), cwd=run_folder):  # nosec
-        exit(exit_code)
+        exit(exit_code)  # pylint: disable=consider-using-sys-exit
 
 
 @app.function(
     image=cicd_image,
     gpu=GPU_CONFIG,
-    timeout=120 * 60,
-    cpu=16.0,
+    timeout=60 * 60,
+    cpu=8.0,
     memory=131072 * N_GPUS,
-    volumes=VOLUME_CONFIG,
 )
 def cicd_pytest():
     run_cmd("./cicd/multigpu.sh", "/workspace/axolotl")

cicd/multigpu.sh

@@ -1,25 +1,5 @@
 #!/bin/bash
 set -e
 
-# Only run two tests at a time to avoid OOM on GPU (with coverage collection)
-pytest -v --durations=10 -n2 --maxfail=3 \
-  --ignore=/workspace/axolotl/tests/e2e/multigpu/solo/ \
-  --ignore=/workspace/axolotl/tests/e2e/multigpu/patched/ \
-  /workspace/axolotl/tests/e2e/multigpu/ \
-  --cov=axolotl
-
-# Run solo tests with coverage append
-pytest -v --durations=10 -n1 \
-  /workspace/axolotl/tests/e2e/multigpu/solo/ \
-  --cov=axolotl \
-  --cov-append
-
-pytest -v  --durations=10 -n1 /workspace/axolotl/tests/e2e/multigpu/patched/ \
-  --cov=axolotl \
-  --cov-append \
-  --cov-report=xml:multigpu-coverage.xml
-
-# Upload coverage to Codecov if CODECOV_TOKEN is available
-if [ -n "$CODECOV_TOKEN" ]; then
-  codecov upload-process -t "${CODECOV_TOKEN}" -f multigpu-coverage.xml -F multigpu,docker-tests,pytorch-${PYTORCH_VERSION} || true
-fi
+# only run one test at a time so as not to OOM the GPU
+pytest -v -n2 /workspace/axolotl/tests/e2e/multigpu/

cicd/single_gpu.py

@@ -1,73 +0,0 @@
-"""Modal app to run axolotl GPU tests"""
-
-import os
-import pathlib
-import tempfile
-
-import jinja2
-import modal
-import modal.experimental
-from jinja2 import select_autoescape
-from modal import App
-
-cicd_path = pathlib.Path(__file__).parent.resolve()
-
-template_loader = jinja2.FileSystemLoader(searchpath=cicd_path)
-template_env = jinja2.Environment(
-    loader=template_loader, autoescape=select_autoescape()
-)
-dockerfile = os.environ.get("E2E_DOCKERFILE", "Dockerfile-uv.jinja")
-df_template = template_env.get_template(dockerfile)
-
-df_args = {
-    "AXOLOTL_EXTRAS": os.environ.get("AXOLOTL_EXTRAS", ""),
-    "AXOLOTL_ARGS": os.environ.get("AXOLOTL_ARGS", ""),
-    "PYTORCH_VERSION": os.environ.get("PYTORCH_VERSION", "2.6.0"),
-    "BASE_TAG": os.environ.get("BASE_TAG", "main-base-py3.11-cu126-2.6.0"),
-    "CUDA": os.environ.get("CUDA", "126"),
-    "GITHUB_REF": os.environ.get("GITHUB_REF", "refs/heads/main"),
-    "GITHUB_SHA": os.environ.get("GITHUB_SHA", ""),
-    "NIGHTLY_BUILD": os.environ.get("NIGHTLY_BUILD", ""),
-    "CODECOV_TOKEN": os.environ.get("CODECOV_TOKEN", ""),
-    "HF_HOME": "/workspace/data/huggingface-cache/hub",
-    "PYTHONUNBUFFERED": os.environ.get("PYTHONUNBUFFERED", "1"),
-    "DEEPSPEED_LOG_LEVEL": os.environ.get("DEEPSPEED_LOG_LEVEL", "WARNING"),
-}
-
-dockerfile_contents = df_template.render(**df_args)
-
-temp_dir = tempfile.mkdtemp()
-with open(pathlib.Path(temp_dir) / "Dockerfile", "w", encoding="utf-8") as f:
-    f.write(dockerfile_contents)
-
-cicd_image = modal.experimental.raw_dockerfile_image(
-    pathlib.Path(temp_dir) / "Dockerfile",
-    # context_mount=None,
-    force_build=True,
-    # gpu="A10G",
-).env(df_args)
-
-app = App("Axolotl CI/CD", secrets=[])
-
-hf_cache_volume = modal.Volume.from_name(
-    "axolotl-ci-hf-hub-cache", create_if_missing=True
-)
-VOLUME_CONFIG = {
-    "/workspace/data/huggingface-cache/hub": hf_cache_volume,
-}
-
-N_GPUS = int(os.environ.get("N_GPUS", 1))
-GPU_TYPE = os.environ.get("GPU_TYPE", "L40S")
-GPU_CONFIG = f"{GPU_TYPE}:{N_GPUS}"
-
-
-def run_cmd(cmd: str, run_folder: str):
-    import subprocess  # nosec
-
-    sp_env = os.environ.copy()
-    sp_env["AXOLOTL_DATASET_NUM_PROC"] = "8"
-
-    # Propagate errors from subprocess.
-    exit_code = subprocess.call(cmd.split(), cwd=run_folder, env=sp_env)  # nosec
-    if exit_code:
-        raise RuntimeError(f"Command '{cmd}' failed with exit code {exit_code}")

cicd/tests.py

@@ -0,0 +1,79 @@
+"""
+ modal application to run axolotl gpu tests in Modal
+ """
+# pylint: disable=duplicate-code
+
+import os
+import pathlib
+import tempfile
+
+import jinja2
+import modal
+from jinja2 import select_autoescape
+from modal import App, Image
+
+cicd_path = pathlib.Path(__file__).parent.resolve()
+
+template_loader = jinja2.FileSystemLoader(searchpath=cicd_path)
+template_env = jinja2.Environment(
+    loader=template_loader, autoescape=select_autoescape()
+)
+df_template = template_env.get_template("Dockerfile.jinja")
+
+df_args = {
+    "AXOLOTL_EXTRAS": os.environ.get("AXOLOTL_EXTRAS", ""),
+    "AXOLOTL_ARGS": os.environ.get("AXOLOTL_ARGS", ""),
+    "PYTORCH_VERSION": os.environ.get("PYTORCH_VERSION", "2.3.1"),
+    "BASE_TAG": os.environ.get("BASE_TAG", "main-base-py3.11-cu121-2.3.1"),
+    "CUDA": os.environ.get("CUDA", "121"),
+    "GITHUB_REF": os.environ.get("GITHUB_REF", "refs/heads/main"),
+    "GITHUB_SHA": os.environ.get("GITHUB_SHA", ""),
+    "NIGHTLY_BUILD": os.environ.get("NIGHTLY_BUILD", ""),
+}
+
+dockerfile_contents = df_template.render(**df_args)
+
+temp_dir = tempfile.mkdtemp()
+with open(pathlib.Path(temp_dir) / "Dockerfile", "w", encoding="utf-8") as f:
+    f.write(dockerfile_contents)
+
+cicd_image = (
+    Image.from_dockerfile(
+        pathlib.Path(temp_dir) / "Dockerfile",
+        context_mount=None,
+        force_build=True,
+        gpu="A10G",
+    )
+    .env(df_args)
+    .pip_install("fastapi==0.110.0", "pydantic==2.6.3")
+)
+
+app = App("Axolotl CI/CD", secrets=[])
+
+
+N_GPUS = int(os.environ.get("N_GPUS", 1))
+GPU_CONFIG = modal.gpu.A10G(count=N_GPUS)
+
+
+def run_cmd(cmd: str, run_folder: str):
+    import subprocess  # nosec
+
+    # Propagate errors from subprocess.
+    if exit_code := subprocess.call(cmd.split(), cwd=run_folder):  # nosec
+        exit(exit_code)  # pylint: disable=consider-using-sys-exit
+
+
+@app.function(
+    image=cicd_image,
+    gpu=GPU_CONFIG,
+    timeout=60 * 60,
+    cpu=8.0,
+    memory=131072,
+)
+def cicd_pytest():
+    run_cmd("./cicd/cicd.sh", "/workspace/axolotl")
+
+
+@app.local_entrypoint()
+def main():
+    cicd_pytest.remote()

codecov.yml

@@ -1,58 +0,0 @@
-codecov:
-  require_ci_to_pass: yes
-  notify:
-    wait_for_ci: true
-
-coverage:
-  precision: 2
-  round: down
-  range: "70...100"
-  status:
-    project:
-      default:
-        # basic
-        target: auto
-        threshold: 1%
-        base: auto
-        # advanced
-        branches: null
-        if_no_uploads: error
-        if_not_found: success
-        if_ci_failed: error
-        only_pulls: true
-        flags: null
-        paths: null
-        informational: true
-    patch:
-      default:
-        # basic
-        target: auto
-        threshold: 1%
-        base: auto
-        # advanced
-        branches: null
-        if_no_uploads: error
-        if_not_found: success
-        if_ci_failed: error
-        only_pulls: false
-        flags: null
-        paths: null
-        informational: true
-
-parsers:
-  gcov:
-    branch_detection:
-      conditional: yes
-      loop: yes
-      method: no
-      macro: no
-
-comment:
-  layout: "reach,diff,flags,files,footer"
-  behavior: default
-  require_changes: no
-  require_base: no
-  require_head: yes
-
-github_checks:
-  annotations: false

deepspeed_configs/zero1_torch_compile.json

@@ -1,27 +0,0 @@
-{
-  "zero_optimization": {
-    "stage": 1,
-    "overlap_comm": true
-  },
-  "bf16": {
-    "enabled": "auto"
-  },
-  "fp16": {
-    "enabled": "auto",
-    "auto_cast": false,
-    "loss_scale": 0,
-    "initial_scale_power": 32,
-    "loss_scale_window": 1000,
-    "hysteresis": 2,
-    "min_loss_scale": 1
-  },
-  "compile": {
-    "disable": false,
-    "backend": "inductor"
-  },
-  "gradient_accumulation_steps": "auto",
-  "gradient_clipping": "auto",
-  "train_batch_size": "auto",
-  "train_micro_batch_size_per_gpu": "auto",
-  "wall_clock_breakdown": false
-}

deepspeed_configs/zero2_torch_compile.json

@@ -1,31 +0,0 @@
-{
-  "compile": {
-    "disable": false,
-    "backend": "inductor"
-  },
-  "zero_optimization": {
-    "stage": 2,
-    "offload_optimizer": {
-      "device": "cpu"
-    },
-    "contiguous_gradients": true,
-    "overlap_comm": true
-  },
-  "bf16": {
-    "enabled": "auto"
-  },
-  "fp16": {
-    "enabled": "auto",
-    "auto_cast": false,
-    "loss_scale": 0,
-    "initial_scale_power": 32,
-    "loss_scale_window": 1000,
-    "hysteresis": 2,
-    "min_loss_scale": 1
-  },
-  "gradient_accumulation_steps": "auto",
-  "gradient_clipping": "auto",
-  "train_batch_size": "auto",
-  "train_micro_batch_size_per_gpu": "auto",
-  "wall_clock_breakdown": false
-}

deepspeed_configs/zero3.json

@@ -7,9 +7,9 @@
     "reduce_bucket_size": "auto",
     "stage3_prefetch_bucket_size": "auto",
     "stage3_param_persistence_threshold": "auto",
-    "max_live_parameters": 0,
-    "max_reuse_distance": 0,
-    "gather_16bit_weights_on_model_save": true
+    "stage3_max_live_parameters": 0,
+    "stage3_max_reuse_distance": 0,
+    "stage3_gather_16bit_weights_on_model_save": true
   },
   "bf16": {
     "enabled": "auto"

deepspeed_configs/zero3_bf16.json

@@ -7,9 +7,9 @@
     "reduce_bucket_size": "auto",
     "stage3_prefetch_bucket_size": "auto",
     "stage3_param_persistence_threshold": "auto",
-    "max_live_parameters": 0,
-    "max_reuse_distance": 0,
-    "gather_16bit_weights_on_model_save": true
+    "stage3_max_live_parameters": 0,
+    "stage3_max_reuse_distance": 0,
+    "stage3_gather_16bit_weights_on_model_save": true
   },
   "bf16": {
     "enabled": true

deepspeed_configs/zero3_bf16_cpuoffload_all.json

@@ -17,9 +17,9 @@
     "reduce_bucket_size": "auto",
     "stage3_prefetch_bucket_size": "auto",
     "stage3_param_persistence_threshold": "auto",
-    "max_live_parameters": 0,
-    "max_reuse_distance": 0,
-    "gather_16bit_weights_on_model_save": true
+    "stage3_max_live_parameters": 0,
+    "stage3_max_reuse_distance": 0,
+    "stage3_gather_16bit_weights_on_model_save": true
   },
   "bf16": {
     "enabled": true

deepspeed_configs/zero3_bf16_cpuoffload_params.json

@@ -13,9 +13,9 @@
     "reduce_bucket_size": "auto",
     "stage3_prefetch_bucket_size": "auto",
     "stage3_param_persistence_threshold": "auto",
-    "max_live_parameters": 0,
-    "max_reuse_distance": 0,
-    "gather_16bit_weights_on_model_save": true
+    "stage3_max_live_parameters": 0,
+    "stage3_max_reuse_distance": 0,
+    "stage3_gather_16bit_weights_on_model_save": true
   },
   "bf16": {
     "enabled": true

devtools/dev_chat_template.yml

@@ -13,7 +13,7 @@ datasets:
 val_set_size: 0
 output_dir: temp_debug/axolotl_outputs/model
 dataset_prepared_path: temp_debug/axolotl_outputs/data
-dataset_num_proc: 1
+dataset_processes: 1
 
 sequence_len: 4096
 sample_packing: false

docker/Dockerfile

@@ -1,19 +1,17 @@
+ARG BASE_IMAGE=axolotlai/axolotl-base
 ARG BASE_TAG=main-base
-FROM axolotlai/axolotl-base:$BASE_TAG
+FROM $BASE_IMAGE:$BASE_TAG
 
 ARG TORCH_CUDA_ARCH_LIST="7.0 7.5 8.0 8.6+PTX"
 ARG AXOLOTL_EXTRAS=""
 ARG AXOLOTL_ARGS=""
 ARG CUDA="118"
 ARG PYTORCH_VERSION="2.1.2"
-ARG TARGETARCH
 
 ENV PYTORCH_VERSION=$PYTORCH_VERSION
 
 RUN apt-get update && \
-    apt-get install -y --allow-change-held-packages vim curl nano libnccl2 libnccl-dev rsync s3fs && \
-    rm -rf /var/cache/apt/archives && \
-    rm -rf /var/lib/apt/lists/*
+    apt-get install -y --allow-change-held-packages vim curl nano libnccl2 libnccl-dev rsync s3fs
 
 WORKDIR /workspace
 
@@ -21,27 +19,22 @@ RUN git clone --depth=1 https://github.com/axolotl-ai-cloud/axolotl.git
 
 WORKDIR /workspace/axolotl
 
-# If AXOLOTL_EXTRAS is set, append it in brackets; don't install deepspeed with arm64
-RUN pip uninstall -y causal_conv1d
-RUN if [ "$TARGETARCH" = "arm64" ]; then \
-        BASE_EXTRAS="optimizers,ray"; \
+# If AXOLOTL_EXTRAS is set, append it in brackets
+RUN if [ "$AXOLOTL_EXTRAS" != "" ] ; then \
+        pip install -e .[deepspeed,flash-attn,optimizers,$AXOLOTL_EXTRAS] $AXOLOTL_ARGS; \
     else \
-        BASE_EXTRAS="deepspeed,optimizers,ray"; \
-    fi && \
-    if [ "$AXOLOTL_EXTRAS" != "" ]; then \
-        pip install --no-build-isolation -e .[$BASE_EXTRAS,$AXOLOTL_EXTRAS] $AXOLOTL_ARGS; \
-    else \
-        pip install --no-build-isolation -e .[$BASE_EXTRAS] $AXOLOTL_ARGS; \
-    fi && \
-    python scripts/cutcrossentropy_install.py | sh && \
-    pip install pytest && \
-    pip cache purge
+        pip install -e .[deepspeed,flash-attn,optimizers] $AXOLOTL_ARGS; \
+    fi
 
-# fix so that git fetch/pull from remote works with shallow clone
+RUN python scripts/unsloth_install.py | sh
+RUN python scripts/cutcrossentropy_install.py | sh
+
+# So we can test the Docker image
+RUN pip install pytest
+
+# fix so that git fetch/pull from remote works
 RUN git config remote.origin.fetch "+refs/heads/*:refs/remotes/origin/*" && \
-    git config --get remote.origin.fetch && \
-    git config --global credential.helper store
+    git config --get remote.origin.fetch
 
-COPY .axolotl-complete.bash /root/.axolotl-complete.bash
-RUN chmod +x /root/.axolotl-complete.bash && \
-    echo 'source /root/.axolotl-complete.bash' >> ~/.bashrc
+# helper for huggingface-login cli
+RUN git config --global credential.helper store

docker/Dockerfile-base

@@ -2,59 +2,41 @@ ARG CUDA_VERSION="11.8.0"
 ARG CUDNN_VERSION="8"
 ARG UBUNTU_VERSION="22.04"
 ARG MAX_JOBS=4
-ARG TARGETARCH
 
-FROM nvidia/cuda:$CUDA_VERSION-cudnn$CUDNN_VERSION-devel-ubuntu$UBUNTU_VERSION AS base-builder
+ARG BASE_IMAGE=nvidia/cuda
+ARG DEFAULT_TAG=${CUDA_VERSION}-cudnn${CUDNN_VERSION}-devel-ubuntu${UBUNTU_VERSION}
+ARG BASE_TAG=""
+FROM ${BASE_IMAGE:-nvidia/cuda}:${BASE_TAG:-${DEFAULT_TAG}} AS base-builder
 
 ENV PATH="/root/miniconda3/bin:${PATH}"
 
-ARG TARGETARCH
-ARG PYTHON_VERSION="3.11"
+ARG PYTHON_VERSION="3.10"
 ARG PYTORCH_VERSION="2.1.2"
-ARG CUDA="128"
+ARG CUDA="118"
 ARG TORCH_CUDA_ARCH_LIST="7.0 7.5 8.0 8.6 9.0+PTX"
 
 ENV PYTHON_VERSION=$PYTHON_VERSION
 ENV TORCH_CUDA_ARCH_LIST=$TORCH_CUDA_ARCH_LIST
 
 RUN apt-get update \
-    && apt-get install -y --no-install-recommends \
-        wget git build-essential ninja-build git-lfs libaio-dev pkg-config \
-        ibverbs-providers ibverbs-utils infiniband-diags  \
-        librdmacm-dev librdmacm1 rdmacm-utils slurm-wlm \
-    && rm -rf /var/cache/apt/archives \
-    && rm -rf /var/lib/apt/lists/* \
-    && if [ "$TARGETARCH" = "amd64" ]; then \
-        MINICONDA_ARCH="x86_64"; \
-    elif [ "$TARGETARCH" = "arm64" ]; then \
-        MINICONDA_ARCH="aarch64"; \
-    else \
-        echo "Unsupported architecture: $TARGETARCH"; exit 1; \
-    fi \
-    && wget https://repo.anaconda.com/miniconda/Miniconda3-latest-Linux-${MINICONDA_ARCH}.sh \
+    && apt-get install -y wget git build-essential ninja-build git-lfs libaio-dev && rm -rf /var/lib/apt/lists/* \
+    && wget \
+    https://repo.anaconda.com/miniconda/Miniconda3-latest-Linux-x86_64.sh \
     && mkdir /root/.conda \
-    && bash Miniconda3-latest-Linux-${MINICONDA_ARCH}.sh -b \
-    && rm -f Miniconda3-latest-Linux-${MINICONDA_ARCH}.sh \
-    && conda tos accept --override-channels --channel https://repo.anaconda.com/pkgs/main \
-    && conda tos accept --override-channels --channel https://repo.anaconda.com/pkgs/r \
+    && bash Miniconda3-latest-Linux-x86_64.sh -b \
+    && rm -f Miniconda3-latest-Linux-x86_64.sh \
     && conda create -n "py${PYTHON_VERSION}" python="${PYTHON_VERSION}"
 
 ENV PATH="/root/miniconda3/envs/py${PYTHON_VERSION}/bin:${PATH}"
 
 WORKDIR /workspace
 
-RUN python3 -m pip install --upgrade pip && pip3 install -U packaging==26.0 setuptools==75.8.0 wheel psutil && \
-    python3 -m pip install --no-cache-dir -U torch==${PYTORCH_VERSION}+cu${CUDA} torchvision --extra-index-url https://download.pytorch.org/whl/cu$CUDA && \
-    python3 -m pip cache purge
-
-RUN if [ "$CUDA" != "130" ] ; then \
-        CAUSAL_CONV1D_FORCE_CXX11_ABI=TRUE CAUSAL_CONV1D_FORCE_BUILD=TRUE python3 -m pip install --no-cache-dir "causal_conv1d @ git+https://github.com/Dao-AILab/causal-conv1d.git@v1.5.4"; \
-        python3 -m pip install --no-cache-dir "mamba_ssm @ git+https://github.com/state-spaces/mamba.git@main"; \
-        python3 -m pip cache purge; \
-    fi
+RUN python3 -m pip install --upgrade pip && pip3 install packaging && \
+    python3 -m pip install --no-cache-dir -U torch==${PYTORCH_VERSION}+cu${CUDA} --extra-index-url https://download.pytorch.org/whl/cu$CUDA && \
+    python3 -m pip install --no-cache-dir "causal_conv1d @ git+https://github.com/Dao-AILab/causal-conv1d.git@main" && \
+    python3 -m pip install --no-cache-dir "mamba_ssm @ git+https://github.com/state-spaces/mamba.git@main"
 
 RUN git lfs install --skip-repo && \
     pip3 install awscli && \
     # The base image ships with `pydantic==1.8.2` which is not working
-    pip3 install -U --no-cache-dir pydantic==1.10.10 && \
-    pip3 cache purge
+    pip3 install -U --no-cache-dir pydantic==1.10.10

docker/Dockerfile-base-next

@@ -1,37 +0,0 @@
-ARG CUDA_VERSION="12.8.2"
-ARG UBUNTU_VERSION="22.04"
-ARG MAX_JOBS=4
-
-FROM nvidia/cuda:12.8.2-devel-ubuntu22.04 AS base-builder
-
-ENV PATH="/root/miniforge3/bin:${PATH}"
-
-ARG PYTHON_VERSION="3.11"
-ARG PYTORCH_VERSION="next"
-ARG CUDA="128"
-ARG TORCH_CUDA_ARCH_LIST="7.0 7.5 8.0 8.6 9.0 12.0+PTX"
-
-ENV PYTHON_VERSION=$PYTHON_VERSION
-ENV TORCH_CUDA_ARCH_LIST=$TORCH_CUDA_ARCH_LIST
-
-RUN apt-get update \
-    && apt-get install -y wget git build-essential ninja-build git-lfs libaio-dev pkg-config && rm -rf /var/lib/apt/lists/* \
-    && wget \
-    https://github.com/conda-forge/miniforge/releases/latest/download/Miniforge3-Linux-x86_64.sh \
-    && mkdir /root/.conda \
-    && bash Miniforge3-Linux-x86_64.sh -b \
-    && rm -f Miniforge3-Linux-x86_64.sh \
-    && /root/miniforge3/bin/conda create -n "py${PYTHON_VERSION}" python="${PYTHON_VERSION}"
-
-ENV PATH="/root/miniforge3/envs/py${PYTHON_VERSION}/bin:${PATH}"
-
-WORKDIR /workspace
-
-RUN python3 -m pip install --upgrade pip && pip3 install packaging && \
-    python3 -m pip install --no-cache-dir -U torch==2.7.1 --extra-index-url https://download.pytorch.org/whl/test/cu$CUDA && \
-    python3 -m pip install --no-cache-dir "causal_conv1d @ git+https://github.com/Dao-AILab/causal-conv1d.git@main" && \
-    python3 -m pip install --no-cache-dir "mamba_ssm @ git+https://github.com/state-spaces/mamba.git@main"
-
-RUN git lfs install --skip-repo && \
-    pip3 install awscli && \
-    pip3 install -U --no-cache-dir pydantic==2.10.6

docker/Dockerfile-base-nightly

@@ -1,43 +0,0 @@
-ARG CUDA_VERSION="12.8.1"
-ARG CUDNN_VERSION="8"
-ARG UBUNTU_VERSION="22.04"
-ARG MAX_JOBS=4
-
-FROM nvidia/cuda:$CUDA_VERSION-cudnn$CUDNN_VERSION-devel-ubuntu$UBUNTU_VERSION AS base-builder
-
-ENV PATH="/root/miniconda3/bin:${PATH}"
-
-ARG PYTHON_VERSION="3.11"
-ARG PYTORCH_VERSION="nightly"
-ARG CUDA="128"
-ARG TORCH_CUDA_ARCH_LIST="7.0 7.5 8.0 8.6 9.0+PTX"
-
-ENV PYTHON_VERSION=$PYTHON_VERSION
-ENV TORCH_CUDA_ARCH_LIST=$TORCH_CUDA_ARCH_LIST
-
-RUN apt-get update \
-    && apt-get install -y wget git build-essential ninja-build git-lfs libaio-dev pkg-config && rm -rf /var/lib/apt/lists/* \
-    && wget \
-    https://repo.anaconda.com/miniconda/Miniconda3-latest-Linux-x86_64.sh \
-    && mkdir /root/.conda \
-    && bash Miniconda3-latest-Linux-x86_64.sh -b \
-    && rm -f Miniconda3-latest-Linux-x86_64.sh \
-    && conda tos accept --override-channels --channel https://repo.anaconda.com/pkgs/main \
-    && conda tos accept --override-channels --channel https://repo.anaconda.com/pkgs/r \
-    && conda create -n "py${PYTHON_VERSION}" python="${PYTHON_VERSION}"
-
-ENV PATH="/root/miniconda3/envs/py${PYTHON_VERSION}/bin:${PATH}"
-
-WORKDIR /workspace
-
-RUN python3 -m pip install --upgrade pip && pip3 install -U packaging==26.0 setuptools==75.8.0 wheel && \
-    python3 -m pip install --no-cache-dir -U torch --extra-index-url https://download.pytorch.org/whl/nightly/cu$CUDA && \
-    python3 -m pip install --no-cache-dir "causal_conv1d @ git+https://github.com/Dao-AILab/causal-conv1d.git@main" && \
-    python3 -m pip install --no-cache-dir "mamba_ssm @ git+https://github.com/state-spaces/mamba.git@main" && \
-    python3 -m pip cache purge
-
-RUN git lfs install --skip-repo && \
-    pip3 install awscli && \
-    # The base image ships with `pydantic==1.8.2` which is not working
-    pip3 install -U --no-cache-dir pydantic==1.10.10 && \
-    pip3 cache purge

docker/Dockerfile-cloud

@@ -14,17 +14,13 @@ COPY scripts/motd /etc/motd
 
 RUN pip install jupyterlab notebook ipywidgets && \
     jupyter lab clean
-RUN apt update && \
-    apt install --yes --no-install-recommends openssh-server tmux iproute2 nvtop && \
-    rm -rf /var/cache/apt/archives && \
-    rm -rf /var/lib/apt/lists/* && \
+RUN apt install --yes --no-install-recommends openssh-server tmux && \
     mkdir -p ~/.ssh && \
     chmod 700 ~/.ssh && \
     printf "\n[[ -z \"\$TMUX\"  ]] && { tmux attach-session -t ssh_tmux || tmux new-session -s ssh_tmux; exit; }\n" >> ~/.bashrc && \
     printf "[ ! -z \"\$TERM\" -a -r /etc/motd ] && cat /etc/motd\n" >> ~/.bashrc && \
     chmod +x /workspace/axolotl/scripts/cloud-entrypoint.sh && \
-    chmod +x /root/cloud-entrypoint.sh && \
-    echo 'set-option -g history-limit 5000' >> ~/.tmux.conf
+    chmod +x /root/cloud-entrypoint.sh
 
 ENTRYPOINT ["/root/cloud-entrypoint.sh"]
 CMD ["sleep", "infinity"]

docker/Dockerfile-cloud-no-tmux

@@ -9,15 +9,13 @@ ENV HF_HUB_ENABLE_HF_TRANSFER="1"
 EXPOSE 8888
 EXPOSE 22
 
-COPY scripts/cloud-entrypoint.sh /root/cloud-entrypoint.sh
+COPY scripts/cloud-entrypoint-term.sh /root/cloud-entrypoint.sh
 COPY scripts/motd /etc/motd
 
 RUN pip install jupyterlab notebook ipywidgets && \
     jupyter lab clean
-RUN apt update && \
-    apt install --yes --no-install-recommends openssh-server tmux iproute2 nvtop ibverbs-providers ibverbs-utils infiniband-diags librdmacm-dev librdmacm1 rdmacm-utils slurm-wlm && \
-    rm -rf /var/cache/apt/archives && \
-    rm -rf /var/lib/apt/lists/* && \
+RUN apt install --yes --no-install-recommends openssh-server tmux sudo && \
+    pip3 install -U --no-cache-dir grpcio ray[default]==2.9.3 && \
     mkdir -p ~/.ssh && \
     chmod 700 ~/.ssh && \
     printf "[ ! -z \"\$TERM\" -a -r /etc/motd ] && cat /etc/motd\n" >> ~/.bashrc && \

docker/Dockerfile-cloud-uv

@@ -1,31 +0,0 @@
-ARG BASE_TAG=main
-FROM axolotlai/axolotl-uv:$BASE_TAG
-
-ENV HF_DATASETS_CACHE="/workspace/data/huggingface-cache/datasets"
-ENV HF_HUB_CACHE="/workspace/data/huggingface-cache/hub"
-ENV HF_HOME="/workspace/data/huggingface-cache/hub"
-ENV HF_HUB_ENABLE_HF_TRANSFER="1"
-
-EXPOSE 8888
-EXPOSE 22
-
-COPY scripts/cloud-entrypoint.sh /root/cloud-entrypoint.sh
-COPY scripts/motd /etc/motd
-
-RUN uv pip install jupyterlab notebook ipywidgets && \
-    jupyter lab clean
-RUN apt update && \
-    apt install --yes --no-install-recommends openssh-server tmux iproute2 nvtop && \
-    rm -rf /var/cache/apt/archives && \
-    rm -rf /var/lib/apt/lists/* && \
-    mkdir -p ~/.ssh && \
-    chmod 700 ~/.ssh && \
-    printf "\n[[ -z \"\$TMUX\"  ]] && { tmux attach-session -t ssh_tmux || tmux new-session -s ssh_tmux; exit; }\n" >> ~/.bashrc && \
-    printf "[ ! -z \"\$TERM\" -a -r /etc/motd ] && cat /etc/motd\n" >> ~/.bashrc && \
-    printf "source /workspace/axolotl-venv/bin/activate\n" >> ~/.bashrc && \
-    chmod +x /workspace/axolotl/scripts/cloud-entrypoint.sh && \
-    chmod +x /root/cloud-entrypoint.sh && \
-    echo 'set-option -g history-limit 5000' >> ~/.tmux.conf
-
-ENTRYPOINT ["/root/cloud-entrypoint.sh"]
-CMD ["sleep", "infinity"]

docker/Dockerfile-tests

@@ -24,9 +24,9 @@ RUN git fetch origin +$GITHUB_REF && \
 
 # If AXOLOTL_EXTRAS is set, append it in brackets
 RUN if [ "$AXOLOTL_EXTRAS" != "" ] ; then \
-        pip install --no-build-isolation -e .[deepspeed,mamba-ssm,$AXOLOTL_EXTRAS] $AXOLOTL_ARGS; \
+        pip install -e .[deepspeed,flash-attn,mamba-ssm,$AXOLOTL_EXTRAS] $AXOLOTL_ARGS; \
     else \
-        pip install --no-build-isolation -e .[deepspeed,mamba-ssm] $AXOLOTL_ARGS; \
+        pip install -e .[deepspeed,flash-attn,mamba-ssm] $AXOLOTL_ARGS; \
     fi
 
 # So we can test the Docker image

docker/Dockerfile-uv

@@ -1,47 +0,0 @@
-ARG BASE_TAG=main-base
-FROM axolotlai/axolotl-base-uv:$BASE_TAG
-
-ARG TORCH_CUDA_ARCH_LIST="7.0 7.5 8.0 8.6+PTX"
-ARG AXOLOTL_EXTRAS=""
-ARG AXOLOTL_ARGS=""
-ARG CUDA="118"
-ARG PYTORCH_VERSION="2.1.2"
-ARG TARGETARCH
-
-ENV PYTORCH_VERSION=$PYTORCH_VERSION
-
-RUN apt-get update && \
-    apt-get install -y --allow-change-held-packages vim curl nano libnccl2 libnccl-dev rsync s3fs && \
-    rm -rf /var/cache/apt/archives && \
-    rm -rf /var/lib/apt/lists/*
-
-WORKDIR /workspace
-
-RUN git clone --depth=1 https://github.com/axolotl-ai-cloud/axolotl.git
-
-WORKDIR /workspace/axolotl
-
-# If AXOLOTL_EXTRAS is set, append it in brackets; don't install deepspeed with arm64
-RUN uv pip uninstall causal_conv1d
-RUN if [ "$TARGETARCH" = "arm64" ]; then \
-        BASE_EXTRAS="optimizers,ray"; \
-    else \
-        BASE_EXTRAS="deepspeed,optimizers,ray"; \
-    fi && \
-    if [ "$AXOLOTL_EXTRAS" != "" ]; then \
-        uv pip install --no-build-isolation -e .[$BASE_EXTRAS,$AXOLOTL_EXTRAS] $AXOLOTL_ARGS; \
-    else \
-        uv pip install --no-build-isolation -e .[$BASE_EXTRAS] $AXOLOTL_ARGS; \
-    fi && \
-    python scripts/cutcrossentropy_install.py --uv | sh && \
-    uv pip install pytest && \
-    uv cache clean
-
-# fix so that git fetch/pull from remote works with shallow clone
-RUN git config remote.origin.fetch "+refs/heads/*:refs/remotes/origin/*" && \
-    git config --get remote.origin.fetch && \
-    git config --global credential.helper store
-
-COPY .axolotl-complete.bash /root/.axolotl-complete.bash
-RUN chmod +x /root/.axolotl-complete.bash && \
-    echo 'source /root/.axolotl-complete.bash' >> ~/.bashrc

docker/Dockerfile-uv-base

@@ -1,40 +0,0 @@
-ARG CUDA_VERSION="12.6.3"
-ARG CUDNN_VERSION=""
-ARG UBUNTU_VERSION="22.04"
-ARG MAX_JOBS=4
-ARG TARGETARCH
-
-FROM nvidia/cuda:$CUDA_VERSION-cudnn$CUDNN_VERSION-devel-ubuntu$UBUNTU_VERSION AS base-builder
-
-ARG TARGETARCH
-ARG PYTHON_VERSION="3.11"
-ARG PYTORCH_VERSION="2.6.0"
-ARG CUDA="126"
-ARG TORCH_CUDA_ARCH_LIST="7.0 7.5 8.0 8.6 9.0+PTX"
-
-ENV PYTHON_VERSION=$PYTHON_VERSION
-ENV TORCH_CUDA_ARCH_LIST=$TORCH_CUDA_ARCH_LIST
-ENV UV_TORCH_BACKEND="cu${CUDA}"
-
-RUN apt-get update \
-    && apt-get install -y wget git build-essential ninja-build git-lfs libaio-dev pkg-config curl && rm -rf /var/lib/apt/lists/* \
-    && git lfs install --skip-repo \
-    && curl -LsSf https://astral.sh/uv/install.sh | sh
-
-ENV PATH="/root/.local/bin:${PATH}"
-
-RUN uv python install ${PYTHON_VERSION}
-
-WORKDIR /workspace
-
-RUN uv venv --no-project --relocatable axolotl-venv
-
-ENV PATH="/workspace/axolotl-venv/bin:${PATH}"
-
-RUN uv pip install packaging setuptools wheel psutil \
-    && uv pip install torch==${PYTORCH_VERSION} torchvision \
-    && uv pip install awscli pydantic
-
-RUN if [ "$TARGETARCH" = "amd64" ]; then \
-        MAMBA_SKIP_CUDA_BUILD=TRUE CAUSAL_CONV1D_SKIP_CUDA_BUILD=TRUE uv pip install --no-build-isolation mamba_ssm causal_conv1d; \
-    fi

docs/.gitignore

@@ -1,7 +1,2 @@
 /.quarto/
 _site/
-/api/*.qmd
-/api/*.html
-config-reference.qmd
-models/**/*.qmd
-models/**/*.html

docs/1_58bit_finetuning.qmd

@@ -1,70 +0,0 @@
----
-title: "1.58-bit Finetuning"
-back-to-top-navigation: true
-toc: true
-toc-expand: 2
-toc-depth: 4
----
-
-## Overview
-
-1.58-bit finetuning allows you to finetune BitNet models when their prequantized weights are provided. In theory, it will be possible to fine-tune any LLM in 1.58bit format but the performance degradation will be dramatic.
-
-Axolotl supports 1.58-bit finetuning via the [`onebitllms`](https://github.com/tiiuae/onebitllms) library, which replaces standard linear layers with BitNet-compatible counterparts ready to use for training.
-
-::: {.callout-note}
-LoRA is not supported for BitNet models
-:::
-
-## Installation
-
-Install the `onebitllms` package before using this feature:
-
-```bash
-uv pip install onebitllms
-```
-
-Or from source:
-
-```bash
-uv pip install git+https://github.com/tiiuae/onebitllms
-```
-
-## Supported models
-
-For now, only `Falcon-E` series of models are supported. Make sure to use their `-prequantized` version:
-
-```bash
-tiiuae/Falcon-E-3B-Base-prequantized
-tiiuae/Falcon-E-1B-Base-prequantized
-```
-
-In theory, any other model would 'work' but the performance degradation will be huge. This remains an area of exploration.
-
-## Configuration
-
-To enable 1.58-bit finetuning, set the following in your configuration file:
-
-```yaml
-base_model: tiiuae/Falcon-E-3B-Base-prequantized  # A BitNet-compatible model
-
-use_onebitllms: true
-```
-
-::: {.callout-note}
-For BitNet models, it is recommended to use a higher learning rate than classic models (usually in the order of magnitude of 10x).
-:::
-
-## Considerations after training
-
-Once your model has been trained with 1.58bit fine-tuning, you can convert the trained model in ternary format using the `onebitllms` CLI:
-
-```bash
-onebitllms quantize_to_1bit INPUT_PATH OUTPUT_PATH
-```
-
-After that, you can use supported packages such as `llama.cpp` or Apple MLX package to run the trained model.
-
-## Example Configuration
-
-You can find example configurations in `examples/falcon-e` which contain one configuration for SFT and one configuration for DPO.

docs/agents/grpo.md

@@ -1,71 +0,0 @@
-# GRPO — Agent Reference
-
-Online RL with verifiable reward functions. For full config reference, async features, and scaling, see [grpo.qmd](../grpo.qmd). For vLLM setup, see [vllm_serving.qmd](../vllm_serving.qmd).
-
-## Architecture
-
-```
-Terminal 1 (GPU 0)                    Terminal 2 (GPU 1)
-┌──────────────────────┐              ┌──────────────────────────────────┐
-│  vLLM Server         │   HTTP       │  Trainer                         │
-│  Serves base model   │◄────────────►│  1. Send prompts to vLLM         │
-│  + LoRA adapter      │  /generate   │  2. Score completions (rewards)  │
-│                      │  /set_lora   │  3. Compute advantages           │
-│  Punica kernels for  │              │  4. PPO-clip gradient update     │
-│  LoRA inference      │              │  5. Sync LoRA weights to vLLM    │
-└──────────────────────┘              └──────────────────────────────────┘
-```
-
-## Components Required
-
-1. A YAML config with `rl: grpo`
-2. A reward module (Python file with reward functions)
-3. A running vLLM server (`axolotl vllm-serve config.yaml`)
-
-## Reward Function Signature
-
-```python
-def my_reward(completions, **kwargs) -> list[float]:
-    # completions[i][0]["content"] = text of i-th completion
-    # **kwargs contains dataset columns not removed by transform
-    return [score_for_each_completion]
-```
-
-Multiple rewards: `reward_funcs: [r1, r2]` with `reward_weights: [1.0, 0.5]`.
-
-## Key Async Features
-
-| Feature | Config | Purpose |
-|---------|--------|---------|
-| Async prefetch | `async_prefetch: true` | Overlap generation with training |
-| LoRA sync | `vllm_lora_sync: true` | Fast adapter sync via filesystem |
-| Streaming scoring | `streaming_partial_batch: true` | Score one group at a time |
-| Zero-adv skip | `skip_zero_advantage_batches: true` | Skip batches with no learning signal |
-| Replay buffer | `replay_buffer_size: 100` | Cache high-signal groups |
-| IS correction | `vllm_importance_sampling_correction: true` | Fix off-policy distribution shift |
-
-## Health Checks
-
-- `rewards/*/mean` > 0.15 within 20 steps (else: test reward function standalone)
-- `reward_std` > 0 on most steps (else: no learning signal)
-- `entropy` 0.05-0.5 (< 0.01 = mode collapse)
-- `grad_norm` 0.001-1.0 (> 10 = unstable, 0.0 = zero-advantage skip)
-
-See [training_stability.qmd](../training_stability.qmd) for detailed diagnostics.
-
-## File Map
-
-```
-src/axolotl/
-  cli/train.py                     # Entry point
-  cli/vllm_serve.py                # Entry point for vLLM server
-  core/trainers/grpo/
-    trainer.py                     # AxolotlGRPOTrainer
-    sampler.py                     # Sampling utilities
-  core/builders/rl.py              # HFRLTrainerBuilder — routes rl type → trainer
-  scripts/vllm_serve_lora.py       # vLLM serve script with LoRA sync support
-  utils/schemas/trl.py             # TRL config schema (all trl: options)
-
-docs/grpo.qmd                     # Full user docs: async, rewards, scaling, config reference
-docs/vllm_serving.qmd             # vLLM server modes, LoRA sync, weight sync
-```

docs/agents/model_architectures.md

@@ -1,198 +0,0 @@
-# Model Architectures — Agent Reference
-
-Model-specific quirks, required settings, and known issues. Check this before debugging training failures on specific model families.
-
-## VLM (Vision Language Model) Quick Start
-
-All VLM configs require these four lines:
-```yaml
-processor_type: AutoProcessor
-skip_prepare_dataset: true
-remove_unused_columns: false
-sample_packing: false
-```
-
-Decision tree for VLM config:
-```text
-Is the model multimodal (has vision/audio encoder)?
-  ├─ YES: Add `freeze_mm_modules: true` if training text only
-  │       Add `chat_template: <model_template>` (e.g. gemma4, qwen3_5, gemma3)
-  │       LoRA: use regex `lora_target_modules` to restrict to language model
-  └─ NO: Train as a regular text model
-
-Is the model MoE (e.g. Gemma4 26B-A4B, Qwen3.5 35B-A3B)?
-  ├─ YES: Add `lora_target_parameters` for expert LoRA
-  │       Consider ScatterMoE kernels (see Plugins section)
-  └─ NO: Standard LoRA config
-```
-
-## Plugins & Optimizations
-
-### Cut Cross Entropy (CCE)
-
-Computes loss from hidden states + lm_head weight without materializing the full logits tensor, saving significant VRAM. Install if not already present:
-
-```bash
-uv pip install "cut-cross-entropy[transformers] @ git+https://github.com/axolotl-ai-cloud/ml-cross-entropy.git@main"
-```
-
-```yaml
-plugins:
-  - axolotl.integrations.cut_cross_entropy.CutCrossEntropyPlugin
-```
-
-### ScatterMoE Kernels
-
-Fuses expert + LoRA computation into a single kernel for MoE models. Significant speedup for models with many experts.
-
-```yaml
-plugins:
-  - axolotl.integrations.kernels.KernelsPlugin
-use_kernels: true
-use_scattermoe: true
-experts_implementation: scattermoe
-
-# Expert LoRA targets (3D parameter tensors, not nn.Linear):
-lora_target_parameters:
-  - experts.gate_up_proj
-  - experts.down_proj
-```
-
-Supported: Gemma4 (`gemma4_text`), Mixtral, Qwen MoE variants. The plugin auto-detects model type and routing function. Without ScatterMoE, expert LoRA still works but runs base expert matmul and LoRA as separate operations.
-
-## Gemma 4
-
-**Models**: `google/gemma-4-26B-A4B` (MoE), `google/gemma-4-31B` (dense), `google/gemma-4-E2B`, `google/gemma-4-E4B`
-
-**Architecture**: Multimodal wrapper (`Gemma4ForConditionalGeneration`) over a text backbone (`Gemma4TextModel`), with optional vision/audio encoders. All Gemma4 HF repos have `model_type: "gemma4"` — even text-only variants load as multimodal with a vision tower.
-
-### Required settings
-
-```yaml
-# Always needed for Gemma4:
-freeze_mm_modules: true          # Freeze vision/audio encoders for text-only training
-gradient_checkpointing_kwargs:
-  use_reentrant: false           # Shared per-layer norms cause "marked ready twice" with reentrant
-
-# LoRA target — restrict to language model only (DO NOT use lora_target_linear: true):
-lora_target_modules: 'model.language_model.layers.[\d]+.(_checkpoint_wrapped_module.)?(mlp|self_attn).(up|down|gate|q|k|v|o)_proj'
-```
-
-### Auto-detection
-
-Axolotl auto-detects Gemma4 and applies:
-- `use_reentrant: false` for gradient checkpointing
-- `ddp_find_unused_parameters: true` for DDP (skipped when `activation_offloading: true`)
-
-### Multi-GPU
-
-| Strategy | Works? | Notes |
-|----------|--------|-------|
-| DDP | Yes | Auto-sets `ddp_find_unused_parameters=True` |
-| DDP + activation_offloading | Yes | `find_unused_parameters` is skipped (conflicts with checkpoint wrappers) |
-| FSDP1 | No | OOM during dequantization/sharding with QLoRA |
-| FSDP2 | Yes | Use `Gemma4TextDecoderLayer` (not `Gemma4DecoderLayer`) as wrap class |
-| FSDP2 + activation_offloading | Yes | Lowest VRAM (~26 GiB/GPU for 26B-A4B) |
-
-FSDP2 config:
-```yaml
-fsdp:
-  - full_shard
-  - auto_wrap
-fsdp_config:
-  fsdp_version: 2
-  fsdp_auto_wrap_policy: TRANSFORMER_BASED_WRAP
-  fsdp_transformer_layer_cls_to_wrap: Gemma4TextDecoderLayer
-```
-
-### MoE (26B-A4B)
-
-- `enable_moe_block: true`, 256 experts, top-k routing
-- No separate `SparseMoeBlock` — MoE is embedded in each decoder layer
-- Expert LoRA targets 3D parameter tensors:
-  ```yaml
-  lora_target_parameters:
-    - experts.gate_up_proj
-    - experts.down_proj
-  ```
-- ScatterMoE kernel acceleration:
-  ```yaml
-  plugins:
-    - axolotl.integrations.kernels.KernelsPlugin
-  use_kernels: true
-  use_scattermoe: true
-  experts_implementation: scattermoe
-  ```
-
-### VLM (Vision) Training
-
-All Gemma4 models load as `Gemma4ForConditionalGeneration` with a vision tower. No custom `ProcessingStrategy` needed — the base class auto-detects the image token.
-
-```yaml
-base_model: google/gemma-4-E2B-it   # or E4B-it, 26B-A4B
-processor_type: AutoProcessor
-freeze_mm_modules: true
-chat_template: gemma4
-
-skip_prepare_dataset: true
-remove_unused_columns: false
-sample_packing: false
-```
-
-A starting VLM loss of ~8-15 is typical. In most runs, loss converges below 1.0 within ~30-50 steps, though results may vary across configurations.
-
-For the 26B-A4B MoE variant with ScatterMoE + expert LoRA + CCE, add:
-```yaml
-plugins:
-  - axolotl.integrations.cut_cross_entropy.CutCrossEntropyPlugin
-  - axolotl.integrations.kernels.KernelsPlugin
-use_kernels: true
-use_scattermoe: true
-experts_implementation: scattermoe
-lora_target_parameters:
-  - experts.gate_up_proj
-  - experts.down_proj
-```
-
-### Common issues
-
-| Symptom | Cause | Fix |
-|---------|-------|-----|
-| `mm_token_type_ids is required` in DDP | `model.config` not accessible through DDP wrapper | Already fixed — `unwrap_model()` in `compute_loss` and `prediction_step` |
-| `marked a variable ready twice` in DDP | `ddp_find_unused_parameters=True` + activation_offloading checkpoint wrappers | Auto-handled — `find_unused_parameters` is skipped when `activation_offloading: true` |
-| Loss ~12 instead of ~0.5 | Using `lora_target_linear: true` (applies LoRA to vision/audio modules) | Use the regex `lora_target_modules` pattern instead |
-| FSDP2 `Could not find Gemma4AudioLayer` | Auto-wrap detects `_no_split_modules` including audio layers that don't exist | Explicitly set `fsdp_transformer_layer_cls_to_wrap: Gemma4TextDecoderLayer` |
-| `Gemma4ClippableLinear not supported` by PEFT | Vision tower uses a non-standard linear wrapper | Axolotl patches this automatically via `_patch_peft_clippable_linear()` |
-
-### E2B/E4B dense models
-
-These have `hidden_size_per_layer_input: 256` (per-layer input embeddings) and `attention_k_eq_v: False`. Known issue: loss starts higher than expected (~12 vs ~0.5 for 26B). Root cause under investigation — may be related to the per-layer input mechanism or the `Gemma4ForConditionalGeneration` loss computation.
-
-## Gemma 3
-
-**Models**: `google/gemma-3-*`
-
-- `ddp_find_unused_parameters: true` needed (multimodal unused params)
-- `use_reentrant: false` recommended
-- Attention mask must be dropped for sample packing (handled automatically)
-- Multi-GPU test currently skipped (`tests/e2e/multigpu/test_gemma3.py`)
-
-## Qwen 3.5 MoE
-
-**Models**: `Qwen/Qwen3.5-35B-A3B`
-
-- Hybrid architecture: DeltaNet linear attention (30 layers) + full attention (10 layers)
-- 256 experts, 8 active per token
-- Known weight scale drift in late DeltaNet layers (36-38) due to AdamW + rare expert interaction
-- Fix: `normalize_weight_scales` config to detect and rescale outliers:
-  ```yaml
-  normalize_weight_scales:
-    - name_pattern: 'linear_attn\.conv1d\.weight'
-      threshold: 1.3
-  ```
-
-## General MoE Notes
-
-- `lora_target_linear: true` with multimodal MoE models will apply LoRA to ALL linear modules including vision/audio encoders — use regex `lora_target_modules` to restrict to language model only
-- Rare experts get larger effective learning rate from AdamW (small second-moment estimates) — can cause weight drift in recurrent/SSM components. Use `normalize_weight_scales` with `dry_run: true` to detect.
-- For ScatterMoE kernel support, set `experts_implementation: scattermoe` and add the KernelsPlugin

docs/agents/new_model_support.md

@@ -1,181 +0,0 @@
-# New Model Support — Agent Reference
-
-Guide for debugging and adding support for new model architectures in axolotl. Based on lessons learned from Gemma4, Gemma3, Qwen2-VL, and other multimodal/MoE models.
-
-## Quick Validation Checklist
-
-When testing a new model, run through these checks in order:
-
-1. **Does the model load?** `axolotl preprocess config.yaml` — catches config schema errors
-2. **Does LoRA apply?** Check for "Unsupported layer type" warnings from PEFT
-3. **Is the initial loss sane?** First-step loss for a pretrained model should be 0.5–2.0 for SFT
-4. **Does sample packing work?** Compare loss with `sample_packing: true` vs `false` — should be similar
-5. **Is CCE active?** Check for "Applying Cut Cross Entropy" log and verify peak VRAM is lower
-
-## Loss Debugging
-
-### Expected initial loss
-A pretrained model doing SFT should start with loss roughly in the 0.5–2.0 range. If loss starts above 3.0, something is wrong. If it's near `log(vocab_size)` (≈ 12 for 262K vocab), the model is predicting at random — attention masking or model weights are broken.
-
-### Direct comparison technique
-The fastest way to isolate a loss issue — bypass the trainer entirely:
-
-```python
-# Load model via axolotl's pipeline (applies all patches)
-from axolotl.cli.config import load_cfg
-from axolotl.utils.config import normalize_config, prepare_plugins
-from axolotl.loaders.tokenizer import load_tokenizer
-from axolotl.loaders.model import ModelLoader
-
-cfg = load_cfg("your_config.yaml")
-normalize_config(cfg)
-prepare_plugins(cfg)
-tokenizer = load_tokenizer(cfg)
-model, _ = ModelLoader(cfg, tokenizer).load()
-
-# Forward pass on preprocessed data
-model.train()
-out = model(input_ids, labels=labels)
-print(f"Direct loss: {out.loss.item()}")  # Compare to trainer's reported loss
-```
-
-If direct loss is correct (~1.0) but trainer reports 3–4x higher, check `model_accepts_loss_kwargs` (see below).
-
-### `model_accepts_loss_kwargs` inflation
-HF Trainer checks if the model's `forward()` has `**kwargs` and sets `model_accepts_loss_kwargs=True`. This changes loss normalization: the trainer does NOT divide loss by `gradient_accumulation_steps` before logging. The gradient is correct — only the logged loss is inflated.
-
-**Symptom**: Logged loss ≈ actual_loss × gradient_accumulation_steps.
-
-**Which models are affected**: Any model with `**kwargs` in forward (common in multimodal models for extra inputs like `mm_token_type_ids`, `pixel_values`, etc.).
-
-**Fix location**: `src/axolotl/core/trainers/base.py` `__init__()` — after `super().__init__()`, check if the unwrapped model actually has `num_items_in_batch` in its forward signature. If not, set `self.model_accepts_loss_kwargs = False`.
-
-## Multimodal Models (ForConditionalGeneration)
-
-Many recent models use `ForConditionalGeneration` as the top-level class, not `ForCausalLM`:
-- Gemma3 → `Gemma3ForConditionalGeneration`
-- Gemma4 → `Gemma4ForConditionalGeneration`
-- Qwen2-VL → `Qwen2VLForConditionalGeneration`
-- LLaVA → `LlavaForConditionalGeneration`
-
-### Why this matters
-
-| Component | Targets `ForCausalLM` | Needs `ForConditionalGeneration` |
-|-----------|----------------------|--------------------------------|
-| CCE patches | ✅ (default) | ❌ silently inactive if not patched |
-| PEFT LoRA | ✅ | May fail on custom layer types |
-| HF Trainer label handling | ✅ | May need extra inputs |
-
-### Required extra inputs
-Multimodal models require special inputs during training even for text-only data:
-
-| Model | Required Input | Value for Text-Only |
-|-------|---------------|-------------------|
-| Gemma4 | `mm_token_type_ids` | `torch.zeros_like(input_ids)` |
-| Gemma3 | `token_type_ids` | `torch.zeros_like(input_ids)` |
-
-Auto-inject in `compute_loss()` when not provided by the data collator. See `core/trainers/base.py`.
-
-### Custom layer types and PEFT
-Vision towers often use custom module wrappers that PEFT doesn't support:
-
-| Model | Custom Layer | Wraps | Fix |
-|-------|-------------|-------|-----|
-| Gemma4 | `Gemma4ClippableLinear` | `nn.Linear` | Redirect to `.linear` child |
-
-Fix location: `src/axolotl/loaders/adapter.py` `_patch_peft_clippable_linear()`.
-
-## Sample Packing
-
-### How packed sequence detection works (transformers ≥ 5.x)
-`transformers.masking_utils._preprocess_mask_arguments()` detects packed sequences from `position_ids` resets. But **only when `attention_mask is None`**:
-
-```python
-# From masking_utils.py:
-if position_ids is not None and attention_mask is None and past_key_values is None:
-    packed_sequence_mask = find_packed_sequence_indices(position_ids)
-```
-
-If the collator provides an all-ones `attention_mask`, packing detection is **skipped** and the model builds a single causal mask spanning all packed sequences → cross-sequence attention leakage → very high loss.
-
-### Fix for models using `create_causal_mask_mapping`
-For Gemma3, Gemma4, and similar models that use the new transformers masking system, remove `attention_mask` from inputs when sample packing is active:
-
-```python
-# In compute_loss():
-if (
-    self.args.sample_packing
-    and model_type in ("gemma4", "gemma3")
-    and "attention_mask" in inputs
-    and "position_ids" in inputs
-):
-    del inputs["attention_mask"]
-```
-
-Fix location: `src/axolotl/core/trainers/base.py` `compute_loss()`.
-
-### Models that DON'T need this fix
-Older models that use `_prepare_4d_causal_attention_mask` (Llama, Mistral, Qwen2, etc.) handle sample packing via axolotl's multipack attention monkeypatch instead. Only models using the new `create_causal_mask_mapping` / `create_causal_mask` masking system need the `attention_mask` removal.
-
-## Attention Backend Selection
-
-| Backend | Config | head_dim limit | torch_compile | Notes |
-|---------|--------|---------------|---------------|-------|
-| FA2 | `attn_implementation: flash_attention_2` | 256 | ✅ | Fastest when supported |
-| FA4 | auto with `attn_implementation: flash_attention_2` | 256 (SM90+) | ✅ | Auto-detected on H100+ |
-| SDPA | `attn_implementation: sdpa` | None | ✅ | Universal fallback |
-| flex | `attn_implementation: flex_attention` | None | ⚠️ Triton OOM for large head_dim | Good for variable head dims |
-| eager | `attn_implementation: eager` | None | ✅ | Slowest, always works |
-
-**Check model support**: Look at `_supports_flash_attn_2`, `_supports_flex_attn`, `_supports_sdpa` attributes on the model class.
-
-**head_dim gotcha**: The 256 limit is specific to flash-attn CUDA kernels, NOT PyTorch-level. SDPA and flex_attention both handle arbitrary head_dim. Models with `global_head_dim > 256` (Gemma4: 512) must use SDPA or flex.
-
-**flex + compile gotcha**: `torch_compile` with flex_attention can hit Triton shared memory OOM for large head_dim. Falls back to eager per-function (not a crash, but slower). Unsloth disables flex for Gemma4 for this reason.
-
-## Cut Cross Entropy (CCE)
-
-### How CCE patches work
-CCE replaces the model's `forward()` with a fused version that computes loss from hidden states + lm_head weight without materializing the full logits tensor. This saves ~`batch × seq_len × vocab_size × dtype_bytes` of VRAM.
-
-### Adding CCE for a new model
-1. Check if the model type is in `cut_cross_entropy.transformers.patch.PATCH_FNS`
-2. If not, axolotl's generic fallback (`integrations/cut_cross_entropy/__init__.py` `patch_llama_like()`) patches `{Prefix}ForCausalLM.forward` with `cce_forward`
-3. For multimodal models (`ForConditionalGeneration`), a model-specific patch is needed in `ml-cross-entropy` repo
-4. The multimodal `cce_forward` must accept all extra kwargs (pixel_values, mm_token_type_ids, etc.) and pop any that would conflict before calling `self.model()`
-
-### Common CCE pitfall
-If CCE appears active (log says "Applying Cut Cross Entropy") but peak VRAM doesn't decrease, check which class was patched. If the model loads as `ForConditionalGeneration` but CCE patched `ForCausalLM`, the patch is silently inactive.
-
-## MoE Models
-
-### Dense MLP vs MoE experts
-Some MoE models (e.g., Gemma4) have BOTH dense MLP layers and MoE expert layers at every decoder layer:
-- `gate_proj/up_proj/down_proj` → targets the **dense MLP** (`Gemma4TextMLP`)
-- `experts.gate_up_proj/experts.down_proj` → targets the **MoE experts** (`Gemma4TextExperts`)
-
-LoRA on the dense MLP works normally. Expert LoRA via `lora_target_parameters` requires PEFT support for the specific expert module type (may warn "Unsupported layer type").
-
-### ScatterMoE kernels
-`use_scattermoe: true` with `experts_implementation: scattermoe` registers fused expert kernels via transformers' `ExpertsInterface`. Significant speedup for MoE models. Requires the kernels plugin:
-```yaml
-plugins:
-  - axolotl.integrations.kernels.KernelsPlugin
-use_kernels: true
-use_scattermoe: true
-experts_implementation: scattermoe
-```
-
-## Where to Add Model-Specific Fixes
-
-| What | Where | Example |
-|------|-------|---------|
-| Missing forward inputs | `core/trainers/base.py` `compute_loss()` | mm_token_type_ids injection |
-| Attention mask fixes | `core/trainers/base.py` `compute_loss()` | Sample packing mask removal |
-| Loss logging fixes | `core/trainers/base.py` `__init__()` | model_accepts_loss_kwargs override |
-| PEFT/LoRA patches | `loaders/adapter.py` | ClippableLinear redirect |
-| Attention patches | `monkeypatch/attention/` | FA4 tuple fix |
-| Model-specific patches | `loaders/patch_manager.py` `_apply_model_specific_patches()` | Llama4, Kimi, NemotronH |
-| CCE patches | `ml-cross-entropy` repo `transformers/` | Per-model cce_forward |
-| Example configs | `examples/<model>/` | Validated YAML |
-| Config validation | `utils/schemas/validation.py` | Compatibility checks |

docs/agents/preference_tuning.md

@@ -1,121 +0,0 @@
-# Preference Learning (RLHF) — Agent Reference
-
-Reference for DPO, IPO, KTO, ORPO, and SimPO. For config templates and dataset format examples, see [rlhf.qmd](../rlhf.qmd). For GRPO, see [grpo.qmd](../grpo.qmd). For EBFT, see [ebft.qmd](../ebft.qmd).
-
-## Method Overview
-
-| Method | Data Requirement | Key Idea | Best For |
-|--------|-----------------|----------|----------|
-| **DPO** | Paired (chosen + rejected) | Implicit reward via preference pairs | General alignment, most common |
-| **IPO** | Paired (chosen + rejected) | DPO with different loss (avoids overfitting) | When DPO overfits |
-| **KTO** | Unpaired (completion + binary label) | Kahneman-Tversky loss, no pairs needed | When you only have thumbs-up/down |
-| **ORPO** | Paired (chosen + rejected) | Combined SFT + preference, no ref model | Single-stage alignment, saves VRAM |
-| **SimPO** | Paired (chosen + rejected) | Length-normalized, no ref model | Simple setup, length-robust |
-
-Default: start with DPO. All methods require `sample_packing: false`.
-
-## Architecture
-
-```
-┌──────────────┐   ┌───────────────┐   ┌───────────────┐
-│ Policy Model │   │ Reference     │   │ Preference    │
-│ (trainable)  │   │ Model (frozen)│   │ Dataset       │
-└──────┬───────┘   └──────┬────────┘   └──────┬────────┘
-       └──────────┬───────┘                    │
-                  v                            │
-       Forward pass on chosen + rejected <─────┘
-                  │
-       Preference Loss (DPO/IPO/KTO/...)
-                  │
-       Backprop + Update
-
-Exception: ORPO and SimPO do NOT use a reference model (~50% less VRAM).
-```
-
-No vLLM server needed (unlike GRPO). Offline RL with pre-collected preference data.
-
-## Method Selection
-
-1. Paired preference data (chosen + rejected)?
-   - Default → `rl: dpo`
-   - Overfitting → `rl: dpo, dpo_loss_type: ["ipo"]`
-   - VRAM-limited → `rl: orpo` (no ref model)
-   - Length-sensitive → `rl: simpo` (no ref model)
-2. Only binary labels (good/bad)? → `rl: kto`
-3. Single-stage training (no separate SFT)? → `rl: orpo`
-
-| | DPO | IPO | KTO | ORPO | SimPO |
-|---|---|---|---|---|---|
-| **Reference model** | Yes | Yes | Yes | No | No |
-| **VRAM overhead** | ~2x model | ~2x model | ~2x model | ~1x model | ~1x model |
-| **TRL trainer class** | DPOTrainer | DPOTrainer | KTOTrainer | ORPOTrainer | CPOTrainer |
-
-## Prompt Strategy Resolution
-
-The `type` field resolves to a Python function:
-
-```
-type: "chatml.intel"
-  → axolotl.prompt_strategies.dpo.chatml.intel(cfg, **kwargs)
-  → returns transform_fn(sample) → {"prompt", "chosen", "rejected"}
-
-type: "chat_template.default"
-  → axolotl.prompt_strategies.dpo.chat_template.default(cfg, dataset_idx, **kwargs)
-
-type: {"field_prompt": "prompt", ...}   (dict)
-  → axolotl.prompt_strategies.dpo.user_defined.default(...)
-```
-
-Module base: `axolotl.prompt_strategies.{rl_method}` — replace `dpo` with `kto` or `orpo`.
-
-## Healthy Training Indicators
-
-| Metric | Healthy Range | Problem |
-|--------|--------------|---------|
-| `train/loss` | Decreasing, 0.3-0.7 | Flat or increasing = broken data or too high LR |
-| `rewards/chosen` | Increasing | Flat = model not learning preferences |
-| `rewards/rejected` | Decreasing | Increasing = model prefers wrong responses |
-| `rewards/margins` | Positive and increasing | Negative = prefers rejected over chosen |
-| `rewards/accuracies` | > 0.5, toward 0.7+ | < 0.5 = worse than random |
-| `logps/rejected` | Decreasing | Increasing = reward hacking |
-| `grad_norm` | 0.01 - 10.0 | > 100 = exploding gradients |
-
-Method-specific: DPO/IPO watch `rewards/margins`; KTO loss is noisier; ORPO monitor SFT + odds ratio components; SimPO check length-normalized reward separation.
-
-## Known Issues
-
-| Issue | Fix |
-|-------|-----|
-| Sample packing crash | Set `sample_packing: false` (required for all preference methods) |
-| KTO `KeyError: 'label'` | Ensure dataset has boolean `label` column |
-| ORPO/KTO `KeyError` during tokenization | Add `remove_unused_columns: false` |
-| ORPO template not applied | ORPO requires explicit `chat_template` setting |
-| OOM with ref model (DPO/IPO/KTO) | Use LoRA/QLoRA, or switch to ORPO/SimPO (no ref model) |
-| IPO + label_smoothing | Do not set `dpo_label_smoothing` when `rl: ipo` |
-
-Full troubleshooting: [training_stability.qmd](../training_stability.qmd)
-
-## File Map
-
-```
-src/axolotl/
-  core/trainers/dpo/              # DPO trainer, args, strategy
-  core/builders/rl.py             # HFRLTrainerBuilder — routes rl type → trainer class
-  core/training_args.py           # AxolotlKTOConfig, AxolotlORPOConfig, AxolotlCPOConfig
-  prompt_strategies/
-    dpo/                          # DPO/IPO/SimPO dataset strategies
-      chat_template.py            # chat_template.default, chat_template.argilla_chat
-      chatml.py                   # chatml.default/intel/icr/argilla_chat/prompt_pairs/ultra
-      llama3.py                   # llama3 variants (same subtypes as chatml)
-      user_defined.py             # Custom field mapping
-      passthrough.py              # No transform
-    kto/                          # KTO dataset strategies (chatml, llama3, user_defined)
-    orpo/                         # ORPO dataset strategies (chat_template.argilla)
-  utils/schemas/enums.py          # RLType enum (dpo, ipo, kto, orpo, simpo, grpo, gdpo, ebft)
-  utils/schemas/config.py         # All rl/dpo/kto/orpo/simpo config fields
-
-docs/rlhf.qmd                    # Full user docs: all dataset formats, config templates
-docs/choosing_method.qmd          # SFT vs DPO vs GRPO decision guide
-examples/qwen2/dpo.yaml           # DPO example
-examples/llama-3/qlora-1b-kto.yaml  # KTO example
-```

docs/agents/pretraining.md

@@ -1,75 +0,0 @@
-# Pretraining / Continual Pretraining — Agent Reference
-
-Train on raw text with no input masking. Two approaches depending on dataset size.
-
-## When to Use
-
-- Continual pretraining on domain-specific corpora
-- Adapting a base model to a new language or domain before fine-tuning
-- Pretraining-style data where the entire text is the training signal
-
-## Choosing an Approach
-
-| | Non-streaming (`type: completion`) | Streaming (`pretraining_dataset`) |
-|---|---|---|
-| **Dataset size** | Fits in memory | Too large to fit in memory |
-| **Tokenization** | Pre-tokenized before training | On-demand during training |
-| **Config key** | `datasets:` | `pretraining_dataset:` |
-| **Long text handling** | Splits texts exceeding `sequence_len` | Concatenates into fixed-length sequences |
-| **Benefit** | Can preprocess on CPU, transfer to GPU | Start training immediately, no preprocessing |
-
-## Non-Streaming: `type: completion`
-
-For smaller datasets that fit in memory. Pre-tokenizes the entire dataset.
-
-```yaml
-datasets:
-  - path: my_corpus
-    type: completion
-    # field: text              # Column name (default: "text")
-```
-
-## Streaming: `pretraining_dataset`
-
-For large corpora. Streams data on-demand without loading everything into memory.
-
-```yaml
-pretraining_dataset:
-  - path: HuggingFaceFW/fineweb-edu
-    type: pretrain
-    text_column: text
-    split: train
-
-max_steps: 1000                          # Required — axolotl can't infer dataset size
-streaming_multipack_buffer_size: 10000   # Buffer for sample packing
-pretrain_multipack_attn: true            # Prevent cross-attention between packed samples
-```
-
-`max_steps` is required for streaming — one step = `sequence_len * micro_batch_size * gradient_accumulation_steps * num_gpus` tokens.
-
-Full streaming docs: [streaming.qmd](../streaming.qmd)
-
-## Dataset Format
-
-```json
-{"text": "The complete document text goes here."}
-```
-
-## Key Settings
-
-- `sample_packing: true` + `pad_to_sequence_len: true` — pack documents into fixed-length sequences
-- `flash_attention: true` — required for sample packing
-- No adapter — typically full fine-tune for pretraining
-- `train_on_inputs: true` — default for completion (all tokens trained on)
-
-## File Map
-
-```
-src/axolotl/
-  prompt_strategies/completion.py    # Non-streaming: completion prompt strategy (no masking)
-  utils/data/sft.py                  # Non-streaming: dataset loading and processing
-  utils/data/streaming.py            # Streaming: encode_streaming(), wrap_streaming_dataset()
-  utils/schemas/config.py            # Config fields: pretraining_dataset, pretrain_multipack_attn, etc.
-
-examples/streaming/pretrain.yaml     # Full streaming pretraining example config
-```

docs/agents/reward_modelling.md

@@ -1,48 +0,0 @@
-# Reward Modelling — Agent Reference
-
-Train models to score responses for use as reward signals in RL. For full docs, see [reward_modelling.qmd](../reward_modelling.qmd).
-
-## Types
-
-### Outcome Reward Models (ORM)
-
-Train a classifier to predict preference over entire interactions. Uses `AutoModelForSequenceClassification`.
-
-```yaml
-base_model: google/gemma-2-2b
-model_type: AutoModelForSequenceClassification
-num_labels: 1
-reward_model: true
-chat_template: gemma
-datasets:
-  - path: argilla/distilabel-intel-orca-dpo-pairs
-    type: bradley_terry.chat_template
-```
-
-Dataset format: `{"system": "...", "input": "...", "chosen": "...", "rejected": "..."}`
-
-### Process Reward Models (PRM)
-
-Train a token classifier to score each reasoning step. Uses `AutoModelForTokenClassification`.
-
-```yaml
-base_model: Qwen/Qwen2.5-3B
-model_type: AutoModelForTokenClassification
-num_labels: 2
-process_reward_model: true
-datasets:
-  - path: trl-lib/math_shepherd
-    type: stepwise_supervised
-```
-
-Dataset format: see [stepwise_supervised.qmd](../dataset-formats/stepwise_supervised.qmd).
-
-## File Map
-
-```
-src/axolotl/
-  core/builders/causal.py                    # Handles reward_model flag in trainer builder
-  prompt_strategies/bradley_terry/           # Bradley-Terry prompt strategies
-  prompt_strategies/stepwise_supervised.py   # PRM dataset strategy
-  utils/schemas/config.py                    # reward_model, process_reward_model config fields
-```

docs/agents/sft.md

@@ -1,139 +0,0 @@
-# SFT — Agent Reference
-
-Supervised fine-tuning pipeline reference. For config templates and dataset format examples, see [getting-started.qmd](../getting-started.qmd) and [dataset-formats/](../dataset-formats/).
-
-## Architecture
-
-```
-YAML Config → axolotl train config.yaml
-
-  1. Load base model (+ quantization if QLoRA/8-bit)
-  2. Apply adapter layers (LoRA/QLoRA) if configured
-  3. Load + tokenize dataset(s)
-     - Apply prompt template (chat_template / alpaca / custom)
-     - Mask inputs (train_on_inputs: false)
-     - Pack samples into sequences (sample_packing: true)
-  4. Training loop (HuggingFace Trainer)
-     - forward → loss → backward → optimizer step → lr scheduler step
-  5. Save model / adapter weights + tokenizer
-
-Multi-GPU: FSDP or DeepSpeed shards model across GPUs automatically.
-```
-
-## Components Required
-
-1. A YAML config — model, dataset(s), adapter settings, hyperparameters
-2. A dataset — HuggingFace Hub, local JSONL/JSON/Parquet, or S3/GCS path
-3. (Optional) A custom prompt strategy — for non-standard dataset formats
-
-No external server processes needed (unlike GRPO which requires vLLM).
-
-## Dataset Format Decision Tree
-
-```
-Is your data in chat/message format?
-  ├─ YES: OpenAI message format (role/content)?
-  │   ├─ YES ──────────────────────> type: chat_template  (recommended)
-  │   └─ NO (custom field names) ──> type: chat_template + message_property_mappings
-  └─ NO: Instruction/response pairs?
-      ├─ YES ──> type: alpaca       (instruction, input, output)
-      └─ NO: Raw text?
-          ├─ YES with segments ─────> type: input_output  (template-free masking)
-          └─ YES continuous ────────> type: completion     (pretraining-style)
-```
-
-Full format specs: [dataset-formats/](../dataset-formats/)
-
-## Model Size to Adapter Choice
-
-| Model Size | LoRA | QLoRA (4-bit) | Full Fine-Tune | VRAM (approx) |
-|-----------|------|---------------|----------------|---------------|
-| 1-3B | Preferred | Low-budget option | Single GPU OK | 8-16 GB (LoRA) |
-| 7-8B | Preferred | Good balance | Needs multi-GPU | 16-24 GB (LoRA) |
-| 13-14B | Preferred | Good balance | Multi-GPU required | 24-40 GB (LoRA) |
-| 30-70B | LoRA or QLoRA | Preferred for single GPU | Multi-node | 40-80 GB (QLoRA) |
-
-## Hyperparameter Ranges
-
-| Parameter | LoRA | QLoRA | Full FT |
-|-----------|------|-------|---------|
-| `learning_rate` | 1e-4 to 3e-4 | 1e-4 to 3e-4 | 1e-5 to 5e-5 |
-| `lora_r` | 16-64 | 16-64 | N/A |
-| `lora_alpha` | 1-2x `lora_r` | 1-2x `lora_r` | N/A |
-| `micro_batch_size` | 2-8 | 2-4 | 1-2 |
-| `gradient_accumulation_steps` | 2-8 | 4-16 | 4-16 |
-| `num_epochs` | 1-3 | 1-3 | 1-3 |
-| `optimizer` | `adamw_8bit` | `adamw_bnb_8bit` | `adamw_torch_fused` |
-
-Effective batch = micro_batch * grad_accum * num_gpus. Lower LR for larger models.
-
-## Healthy Training Indicators
-
-| Metric | Healthy | Problem |
-|--------|---------|---------|
-| `train_loss` | Decreasing, starting ~2-4 for chat models | Flat or increasing from step 1 — data or LR issue |
-| `eval_loss` | Decreasing, tracks train_loss | Increasing while train_loss decreases — overfitting |
-| `grad_norm` | 0.1-10, relatively stable | Spikes >100 — instability. 0.0 — frozen weights |
-| `learning_rate` | Follows scheduler curve | Flat or NaN — config issue |
-
-Watch for: loss never decreasing (check `train_on_inputs`, dataset, LR), loss goes to 0 quickly (overfitting), eval_loss diverging (reduce epochs, add regularization). See [training_stability.qmd](../training_stability.qmd).
-
-## Known Issues
-
-| Issue | Fix |
-|-------|-----|
-| OOM during training | Reduce `micro_batch_size`, enable `gradient_checkpointing`, reduce `sequence_len` |
-| `sample_packing` + SDPA + bf16 = 0.0 loss | Use `attn_implementation: flash_attention_2` or disable `sample_packing` |
-| Missing chat template error | Set `chat_template: chatml` explicitly |
-| Label masking wrong | Run `axolotl preprocess config.yaml --debug` and inspect labels |
-| Loss NaN | Use `bf16: auto`, lower LR, check data for empty samples |
-| Tokenizer pad token / infinite loss | Set `special_tokens: pad_token: "<\|end_of_text\|>"` |
-| FSDP save hangs | Use `fsdp_state_dict_type: FULL_STATE_DICT` |
-| DeepSpeed CheckpointError | Set `use_reentrant: true` in `gradient_checkpointing_kwargs` |
-
-## Profiling
-
-To profile training and identify optimization opportunities:
-
-```yaml
-# Profile steps 3-7 (after warmup/autotuning settles)
-profiler_steps_start: 3
-profiler_steps: 5
-```
-
-This produces `profiler_trace.json` (Chrome trace) and `snapshot.pickle` (memory snapshot) in `output_dir`.
-View the Chrome trace at `chrome://tracing`.
-
-To programmatically inspect the trace:
-```bash
-python scripts/analyze_profile.py output_dir/
-```
-
-The trace shows per-kernel CUDA times, memory allocations, and operator-level breakdown. Look for:
-- **Large matmul kernels**: candidates for fusion or quantization
-- **Memory copies (H2D/D2H)**: unnecessary data movement
-- **Small frequent kernels**: candidates for kernel fusion
-- **Gaps between kernels**: pipeline bubbles from CPU overhead
-
-Full troubleshooting: [training_stability.qmd](../training_stability.qmd), [debugging.qmd](../debugging.qmd)
-
-## File Map
-
-```
-src/axolotl/
-  cli/train.py                     # Entry point for `axolotl train`
-  cli/preprocess.py                # Entry point for `axolotl preprocess`
-  core/builders/causal.py          # HFCausalTrainerBuilder — wires config → SFT trainer
-  core/trainers/base.py            # AxolotlTrainer — base trainer class
-  core/trainers/mixins/            # Packing, optimizer, scheduler, checkpoints
-  prompt_strategies/               # Format handlers: chat_template, alpaca, completion, input_output
-  utils/schemas/config.py          # AxolotlInputConfig — main config schema
-  utils/schemas/datasets.py        # SFTDataset, DatasetConfig
-  utils/schemas/peft.py            # LoraConfig — LoRA parameters
-  integrations/liger/              # Liger kernel plugin
-
-examples/llama-3/                  # LoRA, QLoRA, full FT example configs
-docs/getting-started.qmd           # Quickstart with config templates
-docs/optimizations.qmd             # Flash attention, gradient checkpointing, sample packing
-docs/multi-gpu.qmd                 # FSDP and DeepSpeed setup
-```

docs/amd_hpc.qmd

@@ -1,5 +1,5 @@
 ---
-title: AMD GPUs on HPC Systems
+title: Training with AMD GPUs on HPC Systems
 description: A comprehensive guide for using Axolotl on distributed systems with AMD GPUs
 ---
 
@@ -52,7 +52,7 @@ export GPU_ARCHS="gfx90a"
 cd flash-attention
 export PYTHON_SITE_PACKAGES=$(python -c 'import site; print(site.getsitepackages()[0])')
 patch "${PYTHON_SITE_PACKAGES}/torch/utils/hipify/hipify_python.py" hipify_patch.patch
-pip install --no-build-isolation .
+pip install .
 ```
 
 ### 6. Install Axolotl
@@ -63,7 +63,7 @@ Clone and install Axolotl:
 git clone https://github.com/axolotl-ai-cloud/axolotl
 cd axolotl
 pip install packaging ninja
-pip install --no-build-isolation -e .
+pip install -e .
 ```
 
 ### 7. Apply xformers Workaround
@@ -86,7 +86,7 @@ export HF_DATASETS_OFFLINE=1
 Download a base model using the Hugging Face CLI:
 
 ```bash
-hf download meta-llama/Meta-Llama-3.1-8B --local-dir ~/hfdata/llama3.1-8B
+huggingface-cli download meta-llama/Meta-Llama-3.1-8B --local-dir ~/hfdata/llama3.1-8B
 ```
 
 ### 10. Create Axolotl Configuration

docs/attention.qmd

@@ -1,242 +0,0 @@
----
-title: Attention
-description: Supported attention modules in Axolotl
----
-
-Axolotl routes attention via a single config field:
-
-```yaml
-attn_implementation: <backend>
-```
-
-`attn_implementation` is passed through to `transformers` verbatim (via
-`model.config._attn_implementation`). Accepted values are the HF-native
-backends, axolotl-registered backends, or a hub-kernel path.
-
-## Backends
-
-| `attn_implementation` | Description |
-|---|---|
-| `eager` | Plain PyTorch attention. No packing support. |
-| `sdpa` | PyTorch `scaled_dot_product_attention`. No packing support. |
-| `flash_attention_2` | Dao-AILab Flash Attention 2. |
-| `flash_attention_3` | Dao-AILab Flash Attention 3 (Hopper+). |
-| `flex_attention` | Torch Flex Attention (requires torch ≥ 2.6). |
-| `xformers` | xFormers memory-efficient attention. |
-| `sage` | SageAttention (QK int8 / PV fp16). |
-| `s2` | Shifted-Sparse Attention (LLaMA only, FA2 under the hood). |
-| `fp8` | torchao FP8 low-precision attention (requires SM90+, torch ≥ 2.11). Loaded as SDPA and patched post-load. |
-| `kernels-community/flash-attn3` | HF hub FA3 kernel. |
-| `kernels-community/sage-attention` | HF hub SageAttention kernel. |
-| Other `<org>/<name>` path | Any hub-kernel path supported by `transformers`. |
-
-Short-form aliases (`flash`, `fa2`, `flex`, `sdp`, etc.) are **not accepted** —
-set the canonical name above.
-
-### Capability flags
-
-Axolotl derives three boolean capability flags from `attn_implementation` and
-exposes them on the validated config:
-
-- `cfg.attn_supports_packing` — backend supports varlen sample packing via
-  `position_ids`. Gates multipack patches and `sample_packing_drop_attention_mask`.
-- `cfg.attn_uses_flash_lib` — backend needs the `flash_attn` (Dao-AILab)
-  monkeypatches (FA4 auto, LLaMA flash hijack, ring-FA).
-- `cfg.attn_needs_dtype_cast` — backend requires fp16/bf16 embeddings
-  (everything except `eager` and `sdpa`).
-
-These are **computed** — they cannot be overridden from YAML.
-
-## Per-backend notes
-
-### SDPA
-
-Default PyTorch attention. See
-[PyTorch docs](https://docs.pytorch.org/docs/stable/generated/torch.nn.functional.scaled_dot_product_attention.html).
-
-```yaml
-attn_implementation: sdpa
-```
-
-### Flash Attention
-
-Axolotl supports FA2, FA3, and FA4. The best available version is used
-automatically based on your installed packages and GPU.
-
-```yaml
-attn_implementation: flash_attention_2  # or flash_attention_3
-```
-
-#### Flash Attention 2
-
-Requirements: Ampere, Ada, or Hopper GPUs (Turing or lower not supported)
-
-```bash
-pip install flash-attn --no-build-isolation
-```
-
-::: {.callout-tip}
-
-If you get `undefined symbol` while training, ensure you installed PyTorch prior to Axolotl.
-Alternatively, try reinstall or downgrade a version.
-
-:::
-
-#### Flash Attention 3
-
-Requirements: Hopper only and CUDA 12.8 (recommended)
-
-```bash
-git clone https://github.com/Dao-AILab/flash-attention.git
-cd flash-attention/hopper
-python setup.py install
-```
-
-#### Flash Attention 4
-
-Requirements: Hopper or Blackwell GPUs. Auto-applied when `attn_uses_flash_lib`
-is true and FA4 is importable.
-
-FA4 is still a pre-release on PyPI, so `--pre` is required:
-
-```bash
-pip install --pre flash-attn-4
-```
-
-Or from source:
-
-```bash
-git clone https://github.com/Dao-AILab/flash-attention.git
-cd flash-attention/flash_attn/cute
-pip install -e .
-
-# FA2's flash_attn package includes a cute/ stub that shadows FA4.
-# Remove it so Python can find the real FA4 module:
-rm -r $(python -c "import flash_attn; print(flash_attn.__path__[0])")/cute
-```
-
-::: {.callout-note}
-
-**Hopper (SM90) users**: The backward kernel is not yet included in the pip package. To use FA4
-for training on Hopper, install from source using the instructions above.
-
-:::
-
-::: {.callout-warning}
-
-FA4 only supports head dimensions up to 128 (`d ≤ 128`). The DeepSeek shape `(192, 128)` is
-also supported but only on Blackwell. Axolotl automatically detects incompatible head dimensions
-and falls back to FA2/3.
-
-:::
-
-### AMD
-
-Requirements: ROCm 6.0 and above. See
-[Flash Attention AMD docs](https://github.com/Dao-AILab/flash-attention/tree/main?tab=readme-ov-file#amd-rocm-support).
-
-### Flex Attention
-
-```yaml
-attn_implementation: flex_attention
-torch_compile: true  # recommended
-```
-
-Requires torch ≥ 2.6. See [PyTorch docs](https://pytorch.org/blog/flexattention/).
-
-### SageAttention
-
-Requirements: Ampere, Ada, or Hopper GPUs.
-
-```yaml
-attn_implementation: sage
-```
-
-```bash
-pip install sageattention==2.2.0 --no-build-isolation
-```
-
-::: {.callout-warning}
-
-Only LoRA/QLoRA recommended. Full finetuning has been observed to drop loss to 0. See
-[GitHub Issue](https://github.com/thu-ml/SageAttention/issues/198).
-
-:::
-
-For more details: [Sage Attention](https://github.com/thu-ml/SageAttention).
-
-### xFormers
-
-```yaml
-attn_implementation: xformers
-```
-
-::: {.callout-tip}
-
-Recommended for Turing GPUs or below (e.g. Colab T4).
-
-:::
-
-### Shifted Sparse Attention
-
-::: {.callout-warning}
-
-Planned for deprecation. Prefer one of the backends above.
-
-:::
-
-Requirements: LLaMA model architecture. Loaded as FA2 under the hood and
-patched to implement shifted-sparse attention. Does not support sample packing.
-
-```yaml
-attn_implementation: s2
-```
-
-### FP8
-
-torchao low-precision attention. Loaded as SDPA and patched post-load.
-
-Requirements: SM90+ (Hopper/Blackwell), PyTorch ≥ 2.11, torchao ≥ 0.17,
-flash-attn with FA3. KV caching must be disabled.
-
-```yaml
-attn_implementation: fp8
-```
-
-### Hub kernels
-
-```yaml
-attn_implementation: kernels-community/flash-attn3
-```
-
-Passed through to `transformers`; axolotl does not install the kernel itself.
-For recognized hub paths the capability flags are set automatically; for
-arbitrary paths axolotl uses conservative defaults (`attn_supports_packing=False`,
-`attn_uses_flash_lib=False`).
-
-## Migrating from legacy boolean flags
-
-The following legacy config fields are **deprecated** and will be removed in a
-future release. Each emits a `DeprecationWarning` when set and is stripped from
-the validated config.
-
-| Legacy | Canonical |
-|---|---|
-| `flash_attention: true` | `attn_implementation: flash_attention_2` |
-| `sdp_attention: true` | `attn_implementation: sdpa` |
-| `xformers_attention: true` | `attn_implementation: xformers` |
-| `flex_attention: true` | `attn_implementation: flex_attention` |
-| `sage_attention: true` | `attn_implementation: sage` |
-| `s2_attention: true` | `attn_implementation: s2` |
-| `eager_attention: true` | `attn_implementation: eager` |
-
-Combining `attn_implementation` with a legacy flag (e.g. `attn_implementation:
-flash_attention_2` **and** `flash_attention: true`) raises — pick one.
-
-::: {.callout-note}
-
-Existing example configs under `examples/` still use the legacy flags. They
-continue to work with a deprecation warning; they will be migrated in a
-follow-up pass.
-
-:::

docs/checkpoint_saving.qmd

@@ -1,86 +0,0 @@
----
-title: "Checkpoint Saving"
-format:
-  html:
-    toc: true
-    toc-depth: 2
-    number-sections: true
-execute:
-  enabled: false
----
-
-## Overview
-
-Axolotl supports on-demand checkpoint saving during training. You can trigger checkpoints via file-based triggers (for programmatic control) or Control+C (for interactive use).
-
-## File-Based Checkpoint Trigger
-
-### Configuration
-
-Enable in your config:
-
-```yaml
-dynamic_checkpoint:
-  enabled: true
-  check_interval: 100  # Optional: check every N steps (default: 100)
-  trigger_file_path: "axolotl_checkpoint.save"  # Optional: custom filename
-```
-
-**Options:**
-- `enabled`: `true` to enable (required)
-- `check_interval`: Steps between file checks. Default: 100. Lower = faster response, higher I/O overhead.
-- `trigger_file_path`: Custom trigger filename. Default: `axolotl_checkpoint.save`
-
-### How It Works
-
-1. Rank 0 checks for trigger file every `check_interval` steps in `output_dir`
-2. When detected, file is deleted and checkpoint is saved
-3. In distributed training, rank 0 broadcasts to synchronize all ranks
-
-### Usage
-
-**Command line:**
-```bash
-touch /path/to/output_dir/axolotl_checkpoint.save
-```
-
-**Programmatic:**
-```python
-from pathlib import Path
-Path("/path/to/output_dir/axolotl_checkpoint.save").touch()
-```
-
-Checkpoint saves within the next `check_interval` steps. The trigger file is auto-deleted after detection, so you can create it multiple times.
-
-**Custom filename:**
-```yaml
-dynamic_checkpoint:
-  enabled: true
-  trigger_file_path: "my_trigger.save"
-```
-```bash
-touch /path/to/output_dir/my_trigger.save
-```
-
-## Control+C (SIGINT) Checkpoint
-
-Pressing `Ctrl+C` during training saves the model state and exits gracefully. **Note:** This saves only the model weights, not optimizer state. For resumable checkpoints, use the file-based trigger.
-
-## Best Practices
-
-- **Check interval**: Lower values (10-50) for fast training, default 100 for slower training
-- **Distributed training**: Create trigger file once; rank 0 handles synchronization
-- **Resume**: Dynamic checkpoints can be resumed like regular checkpoints via `resume_from_checkpoint`
-
-## Example
-
-```yaml
-output_dir: ./outputs/lora-out
-save_steps: 500  # Scheduled checkpoints
-
-dynamic_checkpoint:
-  enabled: true
-  check_interval: 50
-```
-
-This enables scheduled checkpoints every 500 steps plus on-demand saves via file trigger (checked every 50 steps).

docs/choosing_method.qmd

@@ -1,206 +0,0 @@
----
-title: "Which Fine-Tuning Method Should I Use?"
-description: "A decision guide for choosing the right fine-tuning method, adapter, and hardware configuration in Axolotl."
-format:
-  html:
-    toc: true
-    toc-depth: 3
-    number-sections: true
-execute:
-  enabled: false
----
-
-## Overview {#sec-overview}
-
-Axolotl supports four broad categories of fine-tuning, each suited to different data types, objectives, and resource constraints.
-
-| Method | What It Does | Data You Need |
-|--------|-------------|---------------|
-| **Supervised Fine-Tuning (SFT)** | Teaches the model to produce specific outputs given inputs | Input-output pairs (instructions, conversations, completions) |
-| **Preference Learning (DPO/KTO/ORPO)** | Steers the model toward preferred outputs and away from dispreferred ones | Chosen/rejected response pairs (DPO, ORPO) or binary labels (KTO) |
-| **Reinforcement Learning (GRPO)** | Optimizes the model against a reward signal through online generation | A reward function (code or model-based) and a prompt dataset |
-| **Reward Modeling** | Trains a model to score responses, for use as a reward signal in RL | Preference pairs ranked by quality |
-
-Each method is configured through a YAML file with `rl: <method>` (or omitted for SFT). All methods support LoRA, QLoRA, and full fine-tuning unless otherwise noted.
-
-## Decision Tree {#sec-decision-tree}
-
-Use the following flowchart to choose your method. Start at the top and follow the path that matches your situation.
-
-```
-Do you have a reward function (code-based or model-based)?
-├── YES
-│   └── Use GRPO (rl: grpo)
-│       The model generates its own completions and learns from reward scores.
-│       Best for: math, code, reasoning, tasks with verifiable answers.
-│       See: rlhf.qmd#grpo
-│
-└── NO
-    │
-    Do you have preference pairs (chosen vs. rejected responses)?
-    ├── YES
-    │   │
-    │   Are they paired (same prompt, one chosen, one rejected)?
-    │   ├── YES → Use DPO (rl: dpo)
-    │   │         Direct optimization without a separate reward model.
-    │   │         See: rlhf.qmd#dpo
-    │   │
-    │   └── NO (only binary good/bad labels)
-    │       └── Use KTO (rl: kto)
-    │           Works with unpaired preference data.
-    │           See: rlhf.qmd#kto
-    │
-    └── NO
-        │
-        Do you have input-output examples?
-        ├── YES → Use SFT
-        │         The simplest and most common method.
-        │         See: getting-started.qmd
-        │
-        └── NO
-            └── You need to create training data first.
-                Consider generating preference pairs with an LLM judge,
-                or writing a reward function for GRPO.
-```
-
-::: {.callout-tip}
-**When in doubt, start with SFT.** It is the most straightforward method and works well for most tasks. You can always move to preference learning or RL later to further refine behavior.
-:::
-
-### Method Comparison at a Glance
-
-| Criterion | SFT | DPO | KTO | GRPO |
-|-----------|-----|-----|-----|------|
-| Data complexity | Low (input-output pairs) | Medium (preference pairs) | Medium (binary labels) | Low (prompts + reward code) |
-| Compute cost | Low | Medium | Medium | High (requires vLLM server) |
-| Learning signal | Supervised | Contrastive | Contrastive | Online reward |
-| Online generation | No | No | No | Yes |
-| Reward model needed | No | No | No | No (uses reward functions) |
-| Best for | Task adaptation, instruction following | Safety, style alignment | Unpaired preference data | Reasoning, math, code |
-
-::: {.callout-note}
-**ORPO** is an alternative to DPO that combines SFT and preference optimization in a single training stage, removing the need for a separate SFT step. Configure with `rl: orpo`. See [rlhf.qmd](rlhf.qmd) for details.
-:::
-
-## Adapter Selection {#sec-adapter-selection}
-
-Once you have chosen a method, decide how to apply the parameter updates. The three main options trade off VRAM usage against model quality.
-
-### QLoRA
-
-- **How it works**: The base model is loaded in 4-bit (NF4) quantization. Small low-rank adapter matrices are trained in higher precision on top.
-- **VRAM savings**: Roughly 4x reduction in model memory compared to full fine-tuning.
-- **Quality**: Slight degradation due to quantization noise, but often negligible for task-specific fine-tuning.
-- **When to use**: When your GPU cannot fit the model in full precision, or when you want fast experimentation.
-
-```yaml
-adapter: qlora
-load_in_4bit: true
-lora_r: 32
-lora_alpha: 64
-lora_target_linear: true
-```
-
-### LoRA
-
-- **How it works**: The base model is loaded at full precision (or 8-bit). Low-rank adapter matrices are trained alongside.
-- **VRAM savings**: Roughly 2-3x reduction compared to full fine-tuning (model weights are frozen, only adapters + optimizer states for adapters are stored).
-- **Quality**: Very close to full fine-tuning for most tasks, especially with higher rank values.
-- **When to use**: When you have enough VRAM for the base model but not for full optimizer states.
-
-```yaml
-adapter: lora
-lora_r: 32
-lora_alpha: 64
-lora_target_linear: true
-```
-
-::: {.callout-tip}
-For GRPO training, LoRA is strongly recommended. The vLLM server needs to sync weights from the trainer, and LoRA sync (`trl.vllm_lora_sync: true`) is far more efficient than syncing full merged weights. See [vLLM Serving](vllm_serving.qmd) for details.
-:::
-
-### Full Fine-Tuning
-
-- **How it works**: All model parameters are updated during training. No adapters.
-- **VRAM savings**: None. Requires memory for model weights, gradients, and optimizer states (roughly 4x model size in bf16 with AdamW).
-- **Quality**: Highest potential quality, especially for large distribution shifts.
-- **When to use**: When you have ample GPU memory or multi-GPU setups, and need maximum performance. Also required for pre-training.
-
-```yaml
-# No adapter or load_in_* lines needed
-micro_batch_size: 1
-gradient_accumulation_steps: 16
-```
-
-### Quick Comparison
-
-| | QLoRA | LoRA | Full |
-|---|---|---|---|
-| Trainable params | ~0.1-1% | ~0.1-1% | 100% |
-| Model memory | ~25% of full | ~50-100% of full | 100% |
-| Optimizer memory | Tiny (adapters only) | Tiny (adapters only) | 2x model size (AdamW) |
-| Training speed | Slower (dequantization overhead) | Baseline | Faster per-step (no adapter overhead) |
-| Inference | Merge or serve with adapter | Merge or serve with adapter | Direct |
-| Multi-GPU required? | Rarely | For 13B+ models | For 7B+ models |
-
-## Hardware Mapping {#sec-hardware-mapping}
-
-The tables below provide approximate GPU memory requirements. Actual usage depends on context length, batch size, and optimizer choice.
-
-### SFT / Preference Learning
-
-| Model Size | QLoRA (4-bit) | LoRA (bf16) | Full (bf16 + AdamW) |
-|------------|--------------|-------------|---------------------|
-| 1-3B | 6-8 GB | 8-12 GB | 24-32 GB |
-| 7-8B | 10-14 GB | 16-24 GB | 60-80 GB |
-| 13-14B | 16-20 GB | 28-40 GB | 120+ GB |
-| 30-34B | 24-32 GB | 64-80 GB | 2-4x 80 GB |
-| 70-72B | 40-48 GB | 2x 80 GB | 4-8x 80 GB |
-
-::: {.callout-important}
-These estimates assume a short context length (512-2048 tokens) and micro_batch_size of 1-2. Longer sequences and larger batches increase memory significantly due to activations. Use [gradient checkpointing](gradient_checkpointing.qmd) to reduce activation memory at the cost of ~30% slower training.
-:::
-
-### GRPO (RL Training)
-
-GRPO requires additional GPU(s) for the vLLM generation server. Plan for at least two GPUs: one for training, one for vLLM.
-
-| Model Size | Training GPU (LoRA, bf16) | vLLM GPU | Total GPUs |
-|------------|--------------------------|----------|------------|
-| 0.5-3B | 1x 24 GB | 1x 24 GB | 2x 24 GB |
-| 7-8B | 1x 80 GB | 1x 80 GB | 2x 80 GB |
-| 13-14B | 1-2x 80 GB | 1-2x 80 GB | 2-4x 80 GB |
-| 30-72B | 2-4x 80 GB (FSDP/DeepSpeed) | 2-4x 80 GB (tensor parallel) | 4-8x 80 GB |
-
-::: {.callout-tip}
-For single-GPU GRPO, use `vllm_mode: colocate` with `vllm_enable_sleep_mode: true`. The vLLM engine shares the GPU and offloads VRAM when not generating. This works for smaller models (up to ~3B on a 24 GB GPU) but is slower than the two-GPU server mode.
-:::
-
-### Multi-GPU Threshold
-
-You need multi-GPU training when:
-
-- **Full fine-tuning** of models 7B+ (use FSDP or DeepSpeed ZeRO)
-- **LoRA** of models 30B+ (or 13B+ with long contexts)
-- **GRPO** almost always (separate vLLM server), unless using colocate mode
-
-See [Multi-GPU Training](multi-gpu.qmd) for FSDP and DeepSpeed configuration.
-
-## Quick Links {#sec-quick-links}
-
-| Method | Config Key | Documentation | Example Config |
-|--------|-----------|---------------|----------------|
-| SFT | *(default, no `rl:` key)* | [Getting Started](getting-started.qmd) | `examples/llama-3/lora-1b.yml` |
-| DPO | `rl: dpo` | [RLHF - DPO](rlhf.qmd#dpo) | See rlhf.qmd |
-| KTO | `rl: kto` | [RLHF - KTO](rlhf.qmd#kto) | See rlhf.qmd |
-| ORPO | `rl: orpo` | [RLHF - ORPO](rlhf.qmd#orpo) | See rlhf.qmd |
-| GRPO | `rl: grpo` | [RLHF - GRPO](rlhf.qmd#grpo), [vLLM Serving](vllm_serving.qmd) | See rlhf.qmd |
-| Reward Modeling | `rl: reward_trainer` | [Reward Modelling](reward_modelling.qmd) | See reward_modelling.qmd |
-
-### Related Guides
-
-- [Configuration Reference](config-reference.qmd) -- Full list of all config options
-- [Dataset Formats](dataset-formats) -- How to structure your training data
-- [Optimizations](optimizations.qmd) -- Flash attention, gradient checkpointing, mixed precision
-- [Multi-GPU Training](multi-gpu.qmd) -- FSDP and DeepSpeed setup
-- [vLLM Serving](vllm_serving.qmd) -- Setting up vLLM for GRPO training

docs/cli.qmd

@@ -1,343 +0,0 @@
----
-title: "Command Line Interface (CLI)"
-format:
-  html:
-    toc: true
-    toc-expand: 2
-    toc-depth: 3
-execute:
-  enabled: false
----
-
-The Axolotl CLI provides a streamlined interface for training and fine-tuning large language models. This guide covers
-the CLI commands, their usage, and common examples.
-
-
-## Basic Commands
-
-All Axolotl commands follow this general structure:
-
-```bash
-axolotl <command> [config.yml] [options]
-```
-
-The config file can be local or a URL to a raw YAML file.
-
-### Launcher Arguments
-
-For commands that support multi-GPU (`train`, `evaluate`, ...), you can pass launcher-specific arguments using the `--` separator:
-
-```bash
-# Pass torchrun arguments
-axolotl train config.yml --launcher torchrun -- --nproc_per_node=2 --nnodes=1
-
-# Pass accelerate arguments
-axolotl train config.yml --launcher accelerate -- --config_file=accelerate_config.yml --num_processes=4
-```
-
-Arguments after `--` are passed directly to the launcher (torchrun, accelerate launch, etc.).
-
-## Command Reference
-
-### fetch
-
-Downloads example configurations and deepspeed configs to your local machine.
-
-```bash
-# Get example YAML files
-axolotl fetch examples
-
-# Get deepspeed config files
-axolotl fetch deepspeed_configs
-
-# Specify custom destination
-axolotl fetch examples --dest path/to/folder
-```
-
-### preprocess
-
-Preprocesses and tokenizes your dataset before training. This is recommended for large datasets.
-
-```bash
-# Basic preprocessing
-axolotl preprocess config.yml
-
-# Preprocessing with one GPU
-CUDA_VISIBLE_DEVICES="0" axolotl preprocess config.yml
-
-# Debug mode to see processed examples
-axolotl preprocess config.yml --debug
-
-# Debug with limited examples
-axolotl preprocess config.yml --debug --debug-num-examples 5
-```
-
-Configuration options:
-
-```yaml
-dataset_prepared_path: Local folder for saving preprocessed data
-push_dataset_to_hub: HuggingFace repo to push preprocessed data (optional)
-```
-
-### train
-
-Trains or fine-tunes a model using the configuration specified in your YAML file.
-
-```bash
-# Basic training
-axolotl train config.yml
-
-# Train and set/override specific options
-axolotl train config.yml \
-    --learning-rate 1e-4 \
-    --micro-batch-size 2 \
-    --num-epochs 3
-
-# Training without accelerate
-axolotl train config.yml --launcher python
-
-# Pass launcher-specific arguments using -- separator
-axolotl train config.yml --launcher torchrun -- --nproc_per_node=2 --nnodes=1
-axolotl train config.yml --launcher accelerate -- --config_file=accelerate_config.yml
-
-# Resume training from checkpoint
-axolotl train config.yml --resume-from-checkpoint path/to/checkpoint
-```
-
-It is possible to run sweeps over multiple hyperparameters by passing in a sweeps config.
-
-```bash
-# Basic training with sweeps
-axolotl train config.yml --sweep path/to/sweep.yaml
-```
-
-Example sweep config:
-```yaml
-_:
-  # This section is for dependent variables we need to fix
-  - load_in_8bit: false
-    load_in_4bit: false
-    adapter: lora
-  - load_in_8bit: true
-    load_in_4bit: false
-    adapter: lora
-
-# These are independent variables
-learning_rate: [0.0003, 0.0006]
-lora_r:
-  - 16
-  - 32
-lora_alpha:
-  - 16
-  - 32
-  - 64
-```
-
-
-
-### inference
-
-Runs inference using your trained model in either CLI or Gradio interface mode.
-
-```bash
-# CLI inference with LoRA
-axolotl inference config.yml --lora-model-dir="./outputs/lora-out"
-
-# CLI inference with full model
-axolotl inference config.yml --base-model="./completed-model"
-
-# Gradio web interface
-axolotl inference config.yml --gradio \
-    --lora-model-dir="./outputs/lora-out"
-
-# Inference with input from file
-cat prompt.txt | axolotl inference config.yml \
-    --base-model="./completed-model"
-```
-
-### merge-lora
-
-Merges trained LoRA adapters into the base model.
-
-```bash
-# Basic merge
-axolotl merge-lora config.yml
-
-# Specify LoRA directory (usually used with checkpoints)
-axolotl merge-lora config.yml --lora-model-dir="./lora-output/checkpoint-100"
-
-# Merge using CPU (if out of GPU memory)
-CUDA_VISIBLE_DEVICES="" axolotl merge-lora config.yml
-```
-
-Configuration options:
-
-```yaml
-gpu_memory_limit: Limit GPU memory usage
-lora_on_cpu: Load LoRA weights on CPU
-```
-
-### merge-sharded-fsdp-weights
-
-Merges sharded FSDP model checkpoints into a single combined checkpoint.
-
-```bash
-# Basic merge
-axolotl merge-sharded-fsdp-weights config.yml
-```
-
-### evaluate
-
-Evaluates a model's performance (loss etc) on the train and eval datasets.
-
-```bash
-# Basic evaluation
-axolotl evaluate config.yml
-
-# Evaluation with launcher arguments
-axolotl evaluate config.yml --launcher torchrun -- --nproc_per_node=2
-```
-
-### lm-eval
-
-Runs LM Evaluation Harness on your model.
-
-```bash
-# Basic evaluation
-axolotl lm-eval config.yml
-```
-
-Configuration options:
-
-```yaml
-lm_eval_model: # model to evaluate (local or hf path)
-
-# List of tasks to evaluate
-lm_eval_tasks:
-  - arc_challenge
-  - hellaswag
-lm_eval_batch_size: # Batch size for evaluation
-output_dir: # Directory to save evaluation results
-```
-
-See [LM Eval Harness integration docs](https://docs.axolotl.ai/docs/custom_integrations.html#language-model-evaluation-harness-lm-eval) for full configuration details.
-
-### delinearize-llama4
-
-Delinearizes a Llama 4 linearized model into a regular HuggingFace Llama 4 model. This only works with the non-quantized linearized model.
-
-```bash
-axolotl delinearize-llama4 --model path/to/model_dir --output path/to/output_dir
-```
-
-This would be necessary to use with other frameworks. If you have an adapter, merge it with the non-quantized linearized model before delinearizing.
-
-### quantize
-
-Quantizes a model using the quantization configuration specified in your YAML file.
-
-```bash
-axolotl quantize config.yml
-```
-
-See [Quantization](./quantize.qmd) for more details.
-
-
-## Legacy CLI Usage
-
-While the new Click-based CLI is preferred, Axolotl still supports the legacy module-based CLI:
-
-```bash
-# Preprocess
-python -m axolotl.cli.preprocess config.yml
-
-# Train
-accelerate launch -m axolotl.cli.train config.yml
-
-# Inference
-accelerate launch -m axolotl.cli.inference config.yml \
-    --lora_model_dir="./outputs/lora-out"
-
-# Gradio interface
-accelerate launch -m axolotl.cli.inference config.yml \
-    --lora_model_dir="./outputs/lora-out" --gradio
-```
-
-::: {.callout-important}
-When overriding CLI parameters in the legacy CLI, use same notation as in yaml file (e.g., `--lora_model_dir`).
-
-**Note:** This differs from the new Click-based CLI, which uses dash notation (e.g., `--lora-model-dir`). Keep this in mind if you're referencing newer documentation or switching between CLI versions.
-:::
-
-## Remote Compute with Modal Cloud
-
-Axolotl supports running training and inference workloads on Modal cloud infrastructure. This is configured using a
-cloud YAML file alongside your regular Axolotl config.
-
-### Cloud Configuration
-
-Create a cloud config YAML with your Modal settings:
-
-```yaml
-# cloud_config.yml
-provider: modal
-gpu: a100       # Supported: l40s, a100-40gb, a100-80gb, a10g, h100, t4, l4
-gpu_count: 1    # Number of GPUs to use
-timeout: 86400  # Maximum runtime in seconds (24 hours)
-branch: main    # Git branch to use (optional)
-
-volumes:        # Persistent storage volumes
-  - name: axolotl-cache
-    mount: /workspace/cache
-  - name: axolotl-data
-    mount: /workspace/data
-  - name: axolotl-artifacts
-    mount: /workspace/artifacts
-
-secrets:        # Secrets to inject
-  - WANDB_API_KEY
-  - HF_TOKEN
-```
-
-### Running on Modal Cloud
-
-Commands that support the --cloud flag:
-
-```bash
-# Preprocess on cloud
-axolotl preprocess config.yml --cloud cloud_config.yml
-
-# Train on cloud
-axolotl train config.yml --cloud cloud_config.yml
-
-# Run lm-eval on cloud
-axolotl lm-eval config.yml --cloud cloud_config.yml
-```
-
-### Cloud Configuration Options
-
-```yaml
-provider:    # compute provider, currently only `modal` is supported
-gpu:         # GPU type to use
-gpu_count:   # Number of GPUs (default: 1)
-memory:      # RAM in GB (default: 128)
-timeout:     # Maximum runtime in seconds
-timeout_preprocess: # Preprocessing timeout
-branch:      # Git branch to use
-docker_tag:  # Custom Docker image tag
-volumes:     # List of persistent storage volumes
-
-# Environment variables to pass. Can be specified in two ways:
-# 1. As a string: Will load the value from the host computer's environment variables
-# 2. As a key-value pair: Will use the specified value directly
-# Example:
-# env:
-#   - CUSTOM_VAR  # Loads from host's $CUSTOM_VAR
-#   - {CUSTOM_VAR: "value"}  # Uses "value" directly
-env:
-
-# Secrets to inject. Same input format as `env` but for sensitive data.
-secrets:
-  # - HF_TOKEN
-  # - WANDB_API_KEY
-```

docs/config.qmd

@@ -0,0 +1,519 @@
+---
+title: Config options
+description: A complete list of all configuration options.
+---
+
+```yaml
+# This is the huggingface model that contains *.pt, *.safetensors, or *.bin files
+# This can also be a relative path to a model on disk
+base_model: ./llama-7b-hf
+# You can specify an ignore pattern if the model repo contains more than 1 model type (*.pt, etc)
+base_model_ignore_patterns:
+# If the base_model repo on hf hub doesn't include configuration .json files,
+# You can set that here, or leave this empty to default to base_model
+base_model_config: ./llama-7b-hf
+# You can specify to choose a specific model revision from huggingface hub
+revision_of_model:
+# Optional tokenizer configuration path in case you want to use a different tokenizer
+# than the one defined in the base model
+tokenizer_config:
+# If you want to specify the type of model to load, AutoModelForCausalLM is a good choice too
+model_type: AutoModelForCausalLM
+# Corresponding tokenizer for the model AutoTokenizer is a good choice
+tokenizer_type: AutoTokenizer
+# Trust remote code for untrusted source
+trust_remote_code:
+# use_fast option for tokenizer loading from_pretrained, default to True
+tokenizer_use_fast:
+# Whether to use the legacy tokenizer setting, defaults to True
+tokenizer_legacy:
+# Resize the model embeddings when new tokens are added to multiples of 32
+# This is reported to improve training speed on some models
+resize_token_embeddings_to_32x:
+
+# (Internal use only)
+# Used to identify which the model is based on
+is_falcon_derived_model:
+is_llama_derived_model:
+is_qwen_derived_model:
+# Please note that if you set this to true, `padding_side` will be set to "left" by default
+is_mistral_derived_model:
+
+# optional overrides to the base model configuration
+overrides_of_model_config:
+  # RoPE Scaling https://github.com/huggingface/transformers/pull/24653
+  rope_scaling:
+    type: # linear | dynamic
+    factor: # float
+
+# optional overrides to the bnb 4bit quantization configuration
+# https://huggingface.co/docs/transformers/main/main_classes/quantization#transformers.BitsAndBytesConfig
+bnb_config_kwargs:
+  # These are default values
+  llm_int8_has_fp16_weight: false
+  bnb_4bit_quant_type: nf4
+  bnb_4bit_use_double_quant: true
+
+
+# Whether you are training a 4-bit GPTQ quantized model
+gptq: true
+
+# This will attempt to quantize the model down to 8 bits and use adam 8 bit optimizer
+load_in_8bit: true
+# Use bitsandbytes 4 bit
+load_in_4bit:
+
+# Use CUDA bf16
+bf16: true # bool or 'full' for `bf16_full_eval`. require >=ampere
+# Use CUDA fp16
+fp16: true
+# Use CUDA tf32
+tf32: true # require >=ampere
+
+# No AMP (automatic mixed precision)
+bfloat16: true # require >=ampere
+float16: true
+
+# Limit the memory for all available GPUs to this amount (if an integer, expressed in gigabytes); default: unset
+gpu_memory_limit: 20GiB
+# Do the LoRA/PEFT loading on CPU -- this is required if the base model is so large it takes up most or all of the available GPU VRAM, e.g. during a model and LoRA merge
+lora_on_cpu: true
+
+# A list of one or more datasets to finetune the model with
+datasets:
+  # HuggingFace dataset repo | s3://,gs:// path | "json" for local dataset, make sure to fill data_files
+  - path: vicgalle/alpaca-gpt4
+    # The type of prompt to use for training. [alpaca, gpteacher, oasst, reflection]
+    type: alpaca # format | format:<prompt_style> (chat/instruct) | <prompt_strategies>.load_<load_fn>
+    ds_type: # Optional[str] (json|arrow|parquet|text|csv) defines the datatype when path is a file
+    data_files: # Optional[str] path to source data files
+    shards: # Optional[int] number of shards to split data into
+    name: # Optional[str] name of dataset configuration to load
+    train_on_split: train # Optional[str] name of dataset split to load from
+    revision: # Optional[str] The specific revision of the dataset to use when loading from the Hugging Face Hub. This can be a commit hash, tag, or branch name. If not specified, the latest version will be used. This parameter is ignored for local datasets.
+    trust_remote_code: # Optional[bool] Trust remote code for untrusted source
+
+  # Custom user instruction prompt
+  - path: repo
+    type:
+      # The below are defaults. only set what's needed if you use a different column name.
+      system_prompt: ""
+      system_format: "{system}"
+      field_system: system
+      field_instruction: instruction
+      field_input: input
+      field_output: output
+
+      # Customizable to be single line or multi-line
+      # Use {instruction}/{input} as key to be replaced
+      # 'format' can include {input}
+      format: |-
+        User: {instruction} {input}
+        Assistant:
+      # 'no_input_format' cannot include {input}
+      no_input_format: "{instruction} "
+
+      # For `completion` datsets only, uses the provided field instead of `text` column
+      field:
+
+  # Using chat template
+  - path: ...
+    # Set type to `chat_template` to use this strategy
+    type: chat_template
+    # Specify the name of the chat template to use
+    # The name of the chat template to use for training, following values are supported:
+    # - tokenizer_default: Uses the chat template that is available in the tokenizer_config.json. If the chat template is not available in the tokenizer, it will raise an error. This is the default.
+    # - alpaca/inst/chatml/gemma/cohere/llama3/phi_3/deepseek_v2/jamba: These chat templates are available in the axolotl codebase at src/axolotl/utils/chat_templates.py
+    # - tokenizer_default_fallback_*: where * is the name of the chat template to fallback to if the tokenizer does not have a chat template else default to tokenizer. E.g. tokenizer_default_fallback_chatml.
+    # - jinja: Uses a custom jinja template for the chat template. The custom jinja template should be provided in the chat_template_jinja field.
+    chat_template: tokenizer_default
+    # Custom jinja template for chat template. This will be only used if `chat_template` is set to `jinja` or empty (in which case chat_template is automatically set to `jinja`).
+    chat_template_jinja:
+    # The key in the data example that contains the messages. Default is "messages".
+    field_messages: messages
+    # The key in the message turn that contains the role. Default is "role".
+    message_field_role: role
+    # The key in the message turn that contains the content. Default is "content".
+    message_field_content: content
+    # Optional[Dict[str, List]]. Roles mapping for the messages.
+    roles:
+      user: ["human", "user"]
+      assistant: ["gpt", "assistant", "ai"]
+      system: ["system"]
+
+    ## NOTE: Leaving the below empty will default to using the simple legacy tokenization strategy where only last message is trained on.
+
+    # Optional[List[str]]. Roles to train on. The tokens from these roles will be considered for the loss.
+    roles_to_train: ["gpt", "assistant"]
+    # Optional[str]. Which EOS tokens to train on in the conversation. Possible values are:
+    # - all: train on all EOS tokens
+    # - turn: train on the EOS token at the end of each trainable turn
+    # - last: train on the last EOS token in the conversation
+    train_on_eos: last
+    # The key in the message turn that indicates via boolean whether tokens of a turn should be considered for training. Useful to selectively train on certain turns besides the `roles_to_train`.
+    message_field_training: training
+    # The key in the message turn that contains the training details. Useful to selectively train on certain tokens in a turn.
+    # The value of the key is a List[Dict] containing `begin_offset` (start character index in content), `end_offset` (end character index in content), and `train` (boolean whether to train).
+    # See example at `docs/dataset-formats/conversation.qmd`
+    message_field_training_detail: train_detail
+
+
+# If false, the datasets will not be shuffled and will keep their original order in `datasets`.
+# The same applies to the `test_datasets` option and the `pretraining_dataset` option. Default is true.
+shuffle_merged_datasets: true
+
+Deduplicates datasets and test_datasets with identical entries.
+dataset_exact_deduplication: true
+
+# A list of one or more datasets to eval the model with.
+# You can use either test_datasets, or val_set_size, but not both.
+test_datasets:
+  - path: /workspace/data/eval.jsonl
+    ds_type: json
+    # You need to specify a split. For "json" datasets the default split is called "train".
+    split: train
+    type: completion
+    data_files:
+      - /workspace/data/eval.jsonl
+
+# use RL training: 'dpo', 'ipo', 'kto'
+rl:
+# whether to perform weighting if doing DPO training. Boolean.
+dpo_use_weighting:
+
+# The name of the chat template to use for training, following values are supported:
+# - tokenizer_default: Uses the chat template that is available in the tokenizer_config.json. If the chat template is not available in the tokenizer, it will raise an error. This is the default value.
+# - alpaca/inst/chatml/gemma/cohere/llama3/phi_3/deepseek_v2/jamba: These chat templates are available in the axolotl codebase at src/axolotl/utils/chat_templates.py
+# - tokenizer_default_fallback_*: where * is the name of the chat template to fallback to. E.g. tokenizer_default_fallback_chatml. This is useful when the chat template is not available in the tokenizer.
+# - jinja: Uses a custom jinja template for the chat template. The custom jinja template should be provided in the chat_template_jinja field.
+# The selected chat template will be saved to the tokenizer_config.json for easier inferencing
+# Note: It is recommended to set train_on_inputs to true when using a chat template that is different from the model's default chat template.
+chat_template: tokenizer_default
+# custom jinja template for chat template. This will be only used if chat_template is set to `jinja` or `null` (in which case chat_template is automatically set to `jinja`). Default is null.
+chat_template_jinja: null
+# Changes the default system message
+default_system_message: You are a helpful assistant. Please give a long and detailed answer. # Currently only supports chatml.
+# Axolotl attempts to save the dataset as an arrow after packing the data together so
+# subsequent training attempts load faster, relative path
+dataset_prepared_path: data/last_run_prepared
+# Push prepared dataset to hub
+push_dataset_to_hub: # repo path
+# The maximum number of processes to use while preprocessing your input dataset. This defaults to `os.cpu_count()`
+# if not set.
+dataset_processes: # defaults to os.cpu_count() if not set
+# Keep dataset in memory while preprocessing
+# Only needed if cached dataset is taking too much storage
+dataset_keep_in_memory:
+# push checkpoints to hub
+hub_model_id: # private repo path to push finetuned model
+# how to push checkpoints to hub
+# https://huggingface.co/docs/transformers/v4.31.0/en/main_classes/trainer#transformers.TrainingArguments.hub_strategy
+hub_strategy:
+# Whether to use hf `use_auth_token` for loading datasets. Useful for fetching private datasets
+# Required to be true when used in combination with `push_dataset_to_hub`
+hf_use_auth_token: # boolean
+# How much of the dataset to set aside as evaluation. 1 = 100%, 0.50 = 50%, etc. 0 for no eval.
+val_set_size: 0.04
+# Num shards for whole dataset
+dataset_shard_num:
+# Index of shard to use for whole dataset
+dataset_shard_idx:
+
+# The maximum length of an input to train with, this should typically be less than 2048
+# as most models have a token/context limit of 2048
+sequence_len: 2048
+# Pad inputs so each step uses constant sized buffers
+# This will reduce memory fragmentation and may prevent OOMs, by re-using memory more efficiently
+pad_to_sequence_len:
+# Use efficient multi-packing with block diagonal attention and per sequence position_ids. Recommend set to 'true'
+sample_packing:
+# Set to 'false' if getting errors during eval with sample_packing on.
+eval_sample_packing:
+# You can set these packing optimizations AFTER starting a training at least once.
+# The trainer will provide recommended values for these values.
+sample_packing_eff_est:
+total_num_tokens:
+# Increasing the following values helps with packing, but usually only slightly (<%1.)
+# The number of samples packed at a time.
+sample_packing_group_size: 100000
+# The number of samples which can be packed into one sequence. Increase if using a large sequence_len with many short samples.
+sample_packing_bin_size: 200
+
+# Passed through to transformers when loading the model when launched without accelerate
+# Use `sequential` when training w/ model parallelism to limit memory
+device_map:
+# Defines the max memory usage per gpu on the system. Passed through to transformers when loading the model.
+max_memory:
+
+# If you want to use 'lora' or 'qlora' or leave blank to train all parameters in original model
+adapter: lora
+# If you already have a lora model trained that you want to load, put that here.
+# This means after training, if you want to test the model, you should set this to the value of `output_dir`.
+# Note that if you merge an adapter to the base model, a new subdirectory `merged` will be created under the `output_dir`.
+lora_model_dir:
+
+# LoRA hyperparameters
+# For more details about the following options, see:
+# https://www.anyscale.com/blog/fine-tuning-llms-lora-or-full-parameter-an-in-depth-analysis-with-llama-2
+lora_r: 8
+lora_alpha: 16
+lora_dropout: 0.05
+lora_target_modules:
+  - q_proj
+  - v_proj
+#  - k_proj
+#  - o_proj
+#  - gate_proj
+#  - down_proj
+#  - up_proj
+lora_target_linear: # If true, will target all linear modules
+peft_layers_to_transform: # The layer indices to transform, otherwise, apply to all layers
+
+# If you added new tokens to the tokenizer, you may need to save some LoRA modules because they need to know the new tokens.
+# For LLaMA and Mistral, you need to save `embed_tokens` and `lm_head`. It may vary for other models.
+# `embed_tokens` converts tokens to embeddings, and `lm_head` converts embeddings to token probabilities.
+# https://github.com/huggingface/peft/issues/334#issuecomment-1561727994
+lora_modules_to_save:
+#  - embed_tokens
+#  - lm_head
+
+lora_fan_in_fan_out: false
+
+# LoRA+ hyperparameters
+# For more details about the following options, see:
+# https://arxiv.org/abs/2402.12354  and `src/axolotl/core/train_builder.py`
+loraplus_lr_ratio: # loraplus learning rate ratio lr_B / lr_A. Recommended value is 2^4.
+loraplus_lr_embedding: #  loraplus learning rate for lora embedding layers. Default value is 1e-6.
+
+peft:
+  # Configuration options for loftq initialization for LoRA
+  # https://huggingface.co/docs/peft/developer_guides/quantization#loftq-initialization
+  loftq_config:
+    loftq_bits:  # typically 4 bits
+
+# ReLoRA configuration
+# Must use either 'lora' or 'qlora' adapter, and does not support fsdp or deepspeed
+relora_steps: # Number of steps per ReLoRA restart
+relora_warmup_steps: # Number of per-restart warmup steps
+relora_anneal_steps: # Number of anneal steps for each relora cycle
+relora_prune_ratio: # threshold for optimizer magnitude when pruning
+relora_cpu_offload: # True to perform lora weight merges on cpu during restarts, for modest gpu memory savings
+
+# wandb configuration if you're using it
+# Make sure your `WANDB_API_KEY` environment variable is set (recommended) or you login to wandb with `wandb login`.
+wandb_mode: # "offline" to save run metadata locally and not sync to the server, "disabled" to turn off wandb
+wandb_project: # Your wandb project name
+wandb_entity: # A wandb Team name if using a Team
+wandb_watch:
+wandb_name: # Set the name of your wandb run
+wandb_run_id: # Set the ID of your wandb run
+wandb_log_model: # "checkpoint" to log model to wandb Artifacts every `save_steps` or "end" to log only at the end of training
+
+# mlflow configuration if you're using it
+mlflow_tracking_uri: # URI to mlflow
+mlflow_experiment_name: # Your experiment name
+mlflow_run_name: # Your run name
+hf_mlflow_log_artifacts:  # set to true to copy each saved checkpoint on each save to mlflow artifact registry
+
+# Comet configuration if you're using it
+# Make sure your `COMET_API_KEY` environment variable is set (recommended) or you login to Comet with `comet login`.
+# Check out our documentation for more details https://www.comet.com/docs/v2/api-and-sdk/python-sdk/reference/Experiment-Creation/#comet_ml.start
+use_comet: # Enable or disable Comet integration.
+comet_api_key: # API key for Comet. Recommended to set via `comet login`.
+comet_workspace: # Workspace name in Comet. Defaults to the user's default workspace.
+comet_project_name: # Project name in Comet. Defaults to Uncategorized.
+comet_experiment_key: # Identifier for the experiment. Used to append data to an existing experiment or control the key of new experiments. Default to a random key.
+comet_mode: # Create a new experiment ("create") or log to an existing one ("get"). Default ("get_or_create") auto-selects based on configuration.
+comet_online: # Set to True to log data to Comet server, or False for offline storage. Default is True.
+comet_experiment_config: # Dictionary for additional configuration settings, see the doc for more details.
+
+# Where to save the full-finetuned model to
+output_dir: ./completed-model
+
+# Whether to use torch.compile and which backend to use
+torch_compile:  # bool
+torch_compile_backend:  # Optional[str]
+
+# Training hyperparameters
+
+# If greater than 1, backpropagation will be skipped and the gradients will be accumulated for the given number of steps.
+gradient_accumulation_steps: 1
+# The number of samples to include in each batch. This is the number of samples sent to each GPU.
+# Batch size per gpu = micro_batch_size * gradient_accumulation_steps
+micro_batch_size: 2
+eval_batch_size:
+num_epochs: 4
+warmup_steps: 100  # cannot use with warmup_ratio
+warmup_ratio: 0.05  # cannot use with warmup_steps
+learning_rate: 0.00003
+lr_quadratic_warmup:
+logging_steps:
+eval_steps: # Leave empty to eval at each epoch, integers for every N steps. decimal for fraction of total steps
+evals_per_epoch: # number of times per epoch to run evals, mutually exclusive with eval_steps
+save_strategy: # Set to `"no"` to skip checkpoint saves
+save_steps: # Leave empty to save at each epoch
+saves_per_epoch: # number of times per epoch to save a checkpoint, mutually exclusive with save_steps
+save_total_limit: # Checkpoints saved at a time
+# Maximum number of iterations to train for. It precedes num_epochs which means that
+# if both are set, num_epochs will not be guaranteed.
+# e.g., when 1 epoch is 1000 steps => `num_epochs: 2` and `max_steps: 100` will train for 100 steps
+max_steps:
+
+eval_table_size: # Approximate number of predictions sent to wandb depending on batch size. Enabled above 0. Default is 0
+eval_max_new_tokens: # Total number of tokens generated for predictions sent to wandb. Default is 128
+eval_causal_lm_metrics: # HF evaluate metrics used during evaluation. Default is ["sacrebleu", "comet", "ter", "chrf", "perplexity"]
+
+loss_watchdog_threshold: # High loss value, indicating the learning has broken down (a good estimate is ~2 times the loss at the start of training)
+loss_watchdog_patience: # Number of high-loss steps in a row before the trainer aborts (default: 3)
+
+# Save model as safetensors (require safetensors package)
+save_safetensors:
+
+# Whether to mask out or include the human's prompt from the training labels
+train_on_inputs: false
+# Group similarly sized data to minimize padding.
+# May be slower to start, as it must download and sort the entire dataset.
+# Note that training loss may have an oscillating pattern with this enabled.
+group_by_length: false
+
+# Whether to use gradient checkpointing https://huggingface.co/docs/transformers/v4.18.0/en/performance#gradient-checkpointing
+gradient_checkpointing: false
+# additional kwargs to pass to the trainer for gradient checkpointing
+# gradient_checkpointing_kwargs:
+#   use_reentrant: true
+
+# Stop training after this many evaluation losses have increased in a row
+# https://huggingface.co/transformers/v4.2.2/_modules/transformers/trainer_callback.html#EarlyStoppingCallback
+early_stopping_patience: 3
+
+# Specify a scheduler and kwargs to use with the optimizer
+lr_scheduler: # 'one_cycle' | 'log_sweep' | empty for cosine
+lr_scheduler_kwargs:
+cosine_min_lr_ratio: # decay lr to some percentage of the peak lr, e.g. cosine_min_lr_ratio=0.1 for 10% of peak lr
+cosine_constant_lr_ratio: # freeze lr at some percentage of the step, e.g. cosine_constant_lr_ratio=0.8 means start cosine_min_lr at 80% of training step (https://arxiv.org/pdf/2308.04014.pdf)
+
+# For one_cycle optim
+lr_div_factor: # Learning rate div factor
+
+# Specify optimizer
+# Valid values are driven by the Transformers OptimizerNames class, see:
+# https://github.com/huggingface/transformers/blob/95b374952dc27d8511541d6f5a4e22c9ec11fb24/src/transformers/training_args.py#L134
+#
+# Note that not all optimizers may be available in your environment, ex: 'adamw_anyprecision' is part of
+# torchdistx, 'adamw_bnb_8bit' is part of bnb.optim.Adam8bit, etc. When in doubt, it is recommended to start with the optimizer used
+# in the examples/ for your model and fine-tuning use case.
+#
+# Valid values for 'optimizer' include:
+# - adamw_hf
+# - adamw_torch
+# - adamw_torch_fused
+# - adamw_torch_xla
+# - adamw_apex_fused
+# - adopt_adamw (an EXPERIMENTAL optimizer, only for torch version >= 2.5.1)
+# - adafactor
+# - adamw_anyprecision
+# - sgd
+# - adagrad
+# - adamw_bnb_8bit
+# - lion_8bit
+# - lion_32bit
+# - paged_adamw_32bit
+# - paged_adamw_8bit
+# - paged_lion_32bit
+# - paged_lion_8bit
+# - galore_adamw
+# - galore_adamw_8bit
+# - galore_adafactor
+# - galore_adamw_layerwise
+# - galore_adamw_8bit_layerwise
+# - galore_adafactor_layerwise
+optimizer:
+# Dictionary of arguments to pass to the optimizer
+optim_args:
+# For Galore Optimizers the following optim_args are available
+# rank:  # type: int
+# update_proj_gap  # type: int
+# scale  # type: float
+# proj_type:  # type: str, default = std
+
+# The target modules to optimize, i.e. the module names that you would like to train, right now this is used only for GaLore algorithm
+optim_target_modules:
+# - self_attn  # for llama
+# - mlp
+
+# Specify weight decay
+weight_decay:
+# adamw hyperparams
+adam_beta1:
+adam_beta2:
+adam_epsilon:
+# Gradient clipping max norm
+max_grad_norm:
+
+# Augmentation techniques
+# NEFT https://arxiv.org/abs/2310.05914, set this to a number (paper default is 5) to add noise to embeddings
+# currently only supported on Llama and Mistral
+neftune_noise_alpha:
+
+# Whether to bettertransformers
+flash_optimum:
+# Whether to use xformers attention patch https://github.com/facebookresearch/xformers:
+xformers_attention:
+# Whether to use flash attention patch https://github.com/Dao-AILab/flash-attention:
+flash_attention:
+flash_attn_cross_entropy:  # Whether to use flash-attention cross entropy implementation - advanced use only
+flash_attn_rms_norm:  # Whether to use flash-attention rms norm implementation - advanced use only
+flash_attn_fuse_qkv: # Whether to fuse QKV into a single operation
+flash_attn_fuse_mlp: # Whether to fuse part of the MLP into a single operation
+# Whether to use scaled-dot-product attention
+# https://pytorch.org/docs/stable/generated/torch.nn.functional.scaled_dot_product_attention.html
+sdp_attention:
+# Shifted-sparse attention (only llama) - https://arxiv.org/pdf/2309.12307.pdf
+s2_attention:
+# Resume from a specific checkpoint dir
+resume_from_checkpoint:
+# If resume_from_checkpoint isn't set and you simply want it to start where it left off.
+# Be careful with this being turned on between different models.
+auto_resume_from_checkpoints: false
+
+# Don't mess with this, it's here for accelerate and torchrun
+local_rank:
+
+# Add or change special tokens.
+# If you add tokens here, you don't need to add them to the `tokens` list.
+special_tokens:
+  # bos_token: "<s>"
+  # eos_token: "</s>"
+  # unk_token: "<unk>"
+  # pad_token: "[PAD]"
+
+# Add extra tokens.
+tokens:
+
+# FSDP
+fsdp:
+fsdp_config:
+
+# Deepspeed config path. e.g., deepspeed_configs/zero3.json
+deepspeed:
+
+# Advanced DDP Arguments
+ddp_timeout:
+ddp_bucket_cap_mb:
+ddp_broadcast_buffers:
+
+# Path to torch distx for optim 'adamw_anyprecision'
+torchdistx_path:
+
+# Set to HF dataset for type: 'completion' for streaming instead of pre-tokenize
+pretraining_dataset:
+
+# Debug mode
+debug:
+
+# Seed
+seed:
+
+# Allow overwrite yml config using from cli
+strict:
+```

docs/custom_integrations.qmd

@@ -1,121 +0,0 @@
----
-title: Custom Integrations
-toc: true
-toc-depth: 3
----
-
-```{python}
-#| echo: false
-
-import os
-import re
-
-def process_readme(integration_name):
-    try:
-        path = f'../src/axolotl/integrations/{integration_name}/README.md'
-        with open(path, 'r') as f:
-            txt = f.read()
-            # Remove h1 headings
-            txt = re.sub(r'^# .*\n?', '', txt, flags=re.MULTILINE)
-            # Convert h2 to h3
-            txt = re.sub(r'^## ', '### ', txt, flags=re.MULTILINE)
-            return txt
-    except FileNotFoundError:
-        return None
-
-def print_section(name, folder_name):
-    output = f"\n## {name}\n"
-    content = process_readme(folder_name)
-    if content:
-        output += content
-    output += f"\nPlease see reference [here](https://github.com/axolotl-ai-cloud/axolotl/tree/main/src/axolotl/integrations/{folder_name})\n"
-    return output
-```
-
-```{python}
-#| output: asis
-#| echo: false
-
-# Introduction text
-print("""
-Axolotl adds custom features through `integrations`. They are located within the `src/axolotl/integrations` directory.
-
-To enable them, please check the respective documentations.
-""")
-
-# Sections
-sections = [
-    ("Cut Cross Entropy", "cut_cross_entropy"),
-    ("Grokfast", "grokfast"),
-    ("Knowledge Distillation (KD)", "kd"),
-    ("Liger Kernels", "liger"),
-    ("Language Model Evaluation Harness (LM Eval)", "lm_eval"),
-    ("Spectrum", "spectrum"),
-    ("LLMCompressor", "llm_compressor")
-]
-
-for folder_name in os.listdir("../src/axolotl/integrations/"):
-    if folder_name in [path for name, path in sections]:
-        # skip if already in sections
-        continue
-    if os.path.exists(f"../src/axolotl/integrations/{folder_name}/README.md"):
-        # grab the first heading in README.md as the section name
-        with open(f"../src/axolotl/integrations/{folder_name}/README.md", "r") as f:
-            txt = f.read()
-            matches = re.search(r'^# (.*)\n?', txt, flags=re.MULTILINE)
-            if matches:
-                name = matches.group(1)
-            else:
-                continue
-            sections.append((name, folder_name))
-
-# sort sections by name
-sections = sorted(sections, key=lambda x: x[0])
-
-for section_name, folder_name in sections:
-    print(print_section(section_name, folder_name))
-```
-
-## Adding a new integration
-
-Plugins can be used to customize the behavior of the training pipeline through [hooks](https://en.wikipedia.org/wiki/Hooking). See [`axolotl.integrations.BasePlugin`](https://github.com/axolotl-ai-cloud/axolotl/blob/main/src/axolotl/integrations/base.py) for the possible hooks.
-
-To add a new integration, please follow these steps:
-
-1. Create a new folder in the `src/axolotl/integrations` directory.
-2. Add any relevant files (`LICENSE`, `README.md`, `ACKNOWLEDGEMENTS.md`, etc.) to the new folder.
-3. Add `__init__.py` and `args.py` files to the new folder.
-  - `__init__.py` should import the integration and hook into the appropriate functions.
-  - `args.py` should define the arguments for the integration.
-4. (If applicable) Add CPU tests under `tests/integrations` or GPU tests under `tests/e2e/integrations`.
-
-::: {.callout-tip}
-
-See [src/axolotl/integrations/cut_cross_entropy](https://github.com/axolotl-ai-cloud/axolotl/tree/main/src/axolotl/integrations/cut_cross_entropy) for a minimal integration example.
-
-:::
-
-::: {.callout-warning}
-
-If you could not load your integration, please ensure you are pip installing in editable mode.
-
-```bash
-pip install -e .
-```
-
-and correctly spelled the integration name in the config file.
-
-```yaml
-plugins:
-  - axolotl.integrations.your_integration_name.YourIntegrationPlugin
-```
-
-:::
-
-::: {.callout-note}
-
-It is not necessary to place your integration in the `integrations` folder. It can be in any location, so long as it's installed in a package in your python env.
-
-See this repo for an example: [https://github.com/axolotl-ai-cloud/diff-transformer](https://github.com/axolotl-ai-cloud/diff-transformer)
-
-:::

docs/dataset-formats/conversation.qmd

@@ -4,15 +4,27 @@ description: Conversation format for supervised fine-tuning.
 order: 3
 ---
 
+## sharegpt
+
+IMPORTANT: ShareGPT is deprecated!. Please see `chat_template` section below.
+
+
+## pygmalion
+
+```{.json filename="data.jsonl"}
+{"conversations": [{"role": "...", "value": "..."}]}
+```
+
+
 ## chat_template
 
 Chat Template strategy uses a jinja2 template that converts a list of messages into a prompt. Support using tokenizer's template, a supported template, or custom jinja2.
 
 ```{.json filename="data.jsonl"}
-{"messages": [{"role": "...", "content": "..."}, {"role": "...", "content": "..."}, ...]}
+{"conversations": [{"role": "...", "content": "..."}]}
 ```
 
-See [configs](../config-reference.qmd) for full configs and supported templates.
+See `config.qmd` for full configs and supported templates.
 
 ### Migrating from sharegpt
 
@@ -32,9 +44,8 @@ datasets:
     type: chat_template
 
     field_messages: conversations
-    message_property_mappings:
-      role: from
-      content: value
+    message_field_role: from
+    message_field_content: value
 
 # new (if setting a new chat_template like chatml, gemma, etc)
 chat_template: chatml
@@ -43,61 +54,43 @@ datasets:
     type: chat_template
 
     field_messages: conversations
-    message_property_mappings:
-      role: from
-      content: value
+    message_field_role: from
+    message_field_content: value
 ```
 
 We recommend checking the below examples for other usecases.
 
 ### Examples
 
-#### Training on last message
-
-(Legacy) Using the default chat template in the tokenizer_config.json on OpenAI messages format, training on only last message.
+1. Using the default chat template in the tokenizer_config.json on OpenAI messages format, training on only last message.
 
 ```yaml
 datasets:
   - path: ...
     type: chat_template
-    roles_to_train:
-    train_on_eos:
 ```
 
-::: {.callout-tip}
-If you receive an error like "`chat_template` choice is `tokenizer_default` but tokenizer's `chat_template` is null.", it means the tokenizer does not have a default `chat_template`. Follow the examples below instead to set a custom `chat_template`.
-:::
-
-#### Overriding default chat template
-
-Using the `gemma` chat template to override the tokenizer_config.json's chat template on OpenAI messages format, training on all assistant messages.
+2. Using the `gemma` chat template to override the tokenizer_config.json's chat template on OpenAI messages format, training on all assistant messages.
 
 ```yaml
 chat_template: gemma # this overwrites the tokenizer's chat_template
 datasets:
   - path: ...
     type: chat_template
-    roles_to_train: ["assistant"]  # default value
+    roles_to_train: ["assistant"]
 ```
 
-::: {.callout-note}
-If you want to use built-in chat_template, use `chat_template: tokenizer_default` (this is set by default).
-:::
-
-#### Using default chat template with fallback
-
-Using the tokenizer_config.json's chat template or `chatml` as fallback if the former's chat template does not exist, on OpenAI messages format, training on all assistant messages.
+3. Using the tokenizer_config.json's chat template or `chatml` as fallback if the former's chat template does not exist, on OpenAI messages format, training on all assistant messages.
 
 ```yaml
 chat_template: tokenizer_default_fallback_chatml # this overwrites the tokenizer's chat_template
 datasets:
   - path: ...
     type: chat_template
+    roles_to_train: ["assistant"]
 ```
 
-#### Custom Jinja template
-
-Using a custom jinja template on OpenAI messages format, training on all assistant messages.
+4. Using a custom jinja template on OpenAI messages format, training on all assistant messages.
 
 ```yaml
 # chat_template: jinja # `jinja` will be implied if the `chat_template_jinja` is set and this field is empty
@@ -106,152 +99,10 @@ chat_template_jinja: "{{ bos_token }}{% for message in messages %}{% if (message
 datasets:
   - path: ...
     type: chat_template
+    roles_to_train: ["assistant"]
 ```
 
-::: {.callout-tip}
-`chat_template_jinja` also accepts a file path to a `.jinja2` file instead of an inline string:
-
-```yaml
-chat_template_jinja: ./path/to/my_template.jinja2
-```
-:::
-
-::: {.callout-important}
-Please make sure that your `tokenizer.eos_token` is same as EOS (End-of-Sequence) token in template. Otherwise, set `eos_token` under `special_tokens: `.
-:::
-
-#### Using template with different token for EOT and EOS
-
-- If you are using a template that has a different EOT (End-of-Turn) token from EOS token or multiple EOT tokens (like Mistral V7 Tekken), set the `eot_tokens: ` config. The handling of EOT tokens follows `train_on_eos: ` which defaults to turn.
-
-```yaml
-eot_tokens:
-  - "[/INST]"
-  # - "[/SYSTEM_PROMPT]"
-
-datasets:
-  - path: ...
-    type: chat_template
-
-    # optional
-    train_on_eot: turn  # defaults read from train_on_eos (which defaults to turn)
-```
-
-::: {.callout-tip}
-See [config documentation](../config-reference.qmd) for detailed explanations of "turn", "last", and "all" options for training on tokens.
-:::
-
-::: {.callout-note}
-Using `eot_tokens` requires each token that exists in `chat_template` to be a single token in the tokenizer. Otherwise, the tokenizer will split the token and cause unexpected behavior.
-
-You can add those tokens as new tokens under `tokens: ` or (recommended) override unused added_tokens via `added_tokens_overrides: `. See [config](../config-reference.qmd) for more details.
-:::
-
-- Continuing from the previous example, if you want to train on all EOT token trainable turns but only last EOS token, set `train_on_eos: last`.
-
-```yaml
-eot_tokens:
-  - "[/INST]"
-  # ...
-
-datasets:
-  - path: ...
-    type: chat_template
-
-    train_on_eos: last
-    train_on_eot: turn
-```
-
-::: {.callout-tip}
-If EOS token only appears at the end of a prompt, `train_on_eos: last` is equivalent to `train_on_eos: turn`. Therefore, generally, you can leave them to their defaults and omit them.
-:::
-
-
-#### Using tool use
-
-Instead of passing `tools` via the system prompt, an alternative method would be to have the `tools` in a separate column and loaded via `chat_template` to let the template dynamically build it.
-
-```json
-{
-    "tools": [
-        {
-            "type": "...",
-            "function": {
-                "name": "...",
-                "description": "...",
-                "parameters": {
-                    "type": "...",
-                    "properties": {
-                        // ...
-                    },
-                    "required": ["..."],
-                },
-            },
-        },
-    ],
-    "messages": [
-        // ...
-        {
-            "role": "assistant", // call the function via assistant
-            "tool_calls": [
-                {
-                    "id": "...",  // required only for mistral
-                    "type": "function",
-                    "function": {
-                        "name": "...",
-                        "arguments": {
-                            "...": "...",
-                        }
-                    }
-                }
-            ]
-        },
-        {
-            "role": "tool",
-            "tool_call_id": "...",  // required only for mistral
-            "name": "...",
-            "content": "..."
-        },
-    ],
-}
-```
-
-::: {.callout-note}
-Tools need to follow [JSON schema](https://json-schema.org/learn/getting-started-step-by-step).
-:::
-
-::: {.callout-warning}
-If you have tool arguments with same name but different dtypes (like `"time": string` and `"time": number`), please save `arguments: ` as JSON string to prevent `datasets` from having casting issues.
-
-```
-"arguments": "{\"...\": \"...\"}"
-```
-
-The same is applicable for tool parameters.
-
-```
-"parameters": "{\"...\": \"...\"}"
-```
-
-:::
-
-Example config for Llama4:
-```yaml
-chat_template: llama4
-datasets:
-  - path: Nanobit/text-tools-2k-test
-    type: chat_template
-    # field_tools: tools # default is `tools`
-```
-
-::: {.callout-tip}
-Look into the `chat_template` you are using to see if it supports `tools` and what the expected role is for the tool answer. In the example above, the tool answer is expected to be in the `tool` or `ipython` role for `llama4` template.
-:::
-
-
-#### Using fine-grained control over token masking
-
-(Advanced) Using fine-grained control over tokens and turns to train in a conversation
+5. (Advanced) Using fine-grained control over tokens and turns to train in a conversation
 
 For a data sample that looks like:
 
@@ -289,164 +140,12 @@ datasets:
     type: chat_template
     chat_template: tokenizer_default
     field_messages: conversations
-    message_property_mappings:
-      role: from
-      content: value
+    message_field_role: from
+    message_field_content: value
     roles_to_train: []
     train_on_eos: turn
     message_field_training: train
     message_field_training_detail: train_detail
 ```
 
-::: {.callout-tip}
-It is not necessary to set both `message_field_training` and `message_field_training_detail` at once.
-:::
-
-#### Content parts with per-part training control
-
-Instead of using character offsets with `train_detail`, you can split a message's content into a list of parts, each with its own training flag. This is useful when you want to mask specific sections of a response (e.g., mask reasoning but train on the answer).
-
-```{.json filename="data.jsonl"}
-{
-  "messages": [
-    {"role": "user", "content": [{"type": "text", "text": "What is 2+2?"}]},
-    {
-      "role": "assistant",
-      "content": [
-        {"type": "text", "text": "Let me think step by step...", "train": false},
-        {"type": "text", "text": " The answer is 4.", "train": true}
-      ]
-    }
-  ]
-}
-```
-
-The configuration is the same as standard `chat_template` — no extra fields needed:
-
-```yaml
-datasets:
-  - path: ...
-    type: chat_template
-    roles_to_train: ["assistant"]
-```
-
-Each content part supports:
-
-- `type`: `"text"` (required)
-- `text`: the text value (also accepts `content` or `value` as the key)
-- `train`: `true`/`false` (optional) — whether to train on this part
-- `weight`: `0`/`1` (optional) — alternative to `train`
-
-If a part has no `train` or `weight` flag, it inherits the turn-level training decision (from `roles_to_train`, `message_field_training`, or `train_on_inputs`).
-
-::: {.callout-warning title="Whitespace at part boundaries"}
-BPE tokenizers (used by Llama, Qwen, Mistral, GPT, etc.) prepend spaces to word tokens. For example, `" answer"` is a single token — the space is part of it. This means **where you place whitespace between content parts matters**:
-
-**Split BEFORE spaces** (space goes with the next part):
-
-```json
-[
-  {"type": "text", "text": "Let me think...", "train": false},
-  {"type": "text", "text": " The answer is 4.", "train": true}
-]
-```
-
-**DON'T put trailing spaces** on a part (the space merges with the next word into one token that straddles the boundary, and straddling tokens are masked):
-
-```json
-[
-  {"type": "text", "text": "Let me think... ", "train": false},
-  {"type": "text", "text": "The answer is 4.", "train": true}
-]
-```
-
-In the bad example, `" The"` becomes a single token that spans both parts. Because it straddles the boundary, it is conservatively **masked** (not trained) — even though the second part has `train: true`.
-
-**Newlines** typically merge with preceding punctuation (e.g., `":\n"` is one token). Keep newlines with the preceding part:
-
-```json
-[
-  {"type": "text", "text": "Thinking:\n", "train": false},
-  {"type": "text", "text": "The answer is 4.", "train": true}
-]
-```
-
-Axolotl will log a warning if it detects trailing whitespace at a boundary between parts with different training flags.
-:::
-
-::: {.callout-note}
-When all content parts in a message are strings, they are concatenated before being passed to the chat template. This means content parts work with **any** Jinja template — the template sees a plain string, and the per-part training flags are applied during tokenization.
-:::
-
-##### Per-part training on reasoning_content
-
-For templates that support a separate `reasoning_content` field (e.g., `qwen3`), the same content-parts format works on `reasoning_content`. This is useful for masking incorrect reasoning steps while training on self-corrections:
-
-```{.json filename="data.jsonl"}
-{
-  "messages": [
-    {"role": "user", "content": [{"type": "text", "text": "What is 2+2?"}]},
-    {
-      "role": "assistant",
-      "reasoning_content": [
-        {"type": "text", "text": "Hmm maybe 2+2=5.", "train": false},
-        {"type": "text", "text": " Wait no, 2+2=4.", "train": true}
-      ],
-      "content": [
-        {"type": "text", "text": "The answer is 4.", "train": true}
-      ]
-    }
-  ]
-}
-```
-
-The `reasoning_content` and `content` fields are handled independently — each has its own token boundaries and per-part masking. No additional configuration is needed beyond what the template already requires.
-
-::: {.callout-tip}
-When `reasoning_content` is provided as a separate field, `split_thinking` is not needed — the reasoning is already separated from the content in the data.
-:::
-
-The same whitespace rules apply to `reasoning_content` parts as to `content` parts — split before spaces, keep newlines with the preceding part.
-
-
-#### Reasoning split
-
-(For Qwen3 template only) Enable reasoning split, where the reasoning is split from the content and passed as a separate field into the template.
-
-```yaml
-datasets:
-  - path: ...
-    type: chat_template
-    chat_template: qwen3
-    split_thinking: true
-```
-
-For example, a content can look like:
-
-```json
-{
-  "content": "<think>Some thinking outputs</think>Output after thinking."
-}
-```
-
-After split, it will look like:
-
-```json
-{
-  "reasoning_content": "Some thinking outputs",
-  "content": "Output after thinking..."
-}
-```
-
-
-## sharegpt
-
-::: {.callout-important}
-ShareGPT is deprecated!. Please see [chat_template](#chat_template) section.
-:::
-
-## pygmalion
-
-```{.json filename="data.jsonl"}
-{"conversations": [{"role": "...", "value": "..."}]}
-```
+Tip: It is not necessary to use both `message_field_training` and `message_field_training_detail` at a time.

docs/dataset-formats/index.qmd

@@ -1,452 +1,14 @@
 ---
 title: Dataset Formats
-description: Guide to Dataset Formats in Axolotl
-back-to-top-navigation: true
-toc: true
-toc-depth: 5
+description: Supported dataset formats.
+listing:
+  fields: [title, description]
+  type: table
+  sort-ui: false
+  filter-ui: false
+  max-description-length: 250
 ---
 
+Axolotl supports a variety of dataset formats.  It is recommended to use a JSONL format.  The schema of the JSONL depends upon the task and the prompt template you wish to use. Instead of a JSONL, you can also use a HuggingFace dataset with columns for each JSONL field.
 
-Axolotl is a training framework that aims to make the process convenient yet flexible to users by simply passing a config yaml file.
-
-As there are a lot of available options in Axolotl, this guide aims to provide an simplify the user experience to choosing the proper choice.
-
-Axolotl supports 3 kinds of training methods: pre-training, supervised fine-tuning, and preference-based post-training (e.g. DPO, ORPO, PRMs). Each method has their own dataset format which are described below.
-
-::: {.callout-tip}
-
-This guide will mainly use JSONL as an introduction. Please refer to the [dataset loading docs](../dataset_loading.qmd) to understand how to load datasets from other sources.
-
-For `pretraining_dataset:` specifically, please refer to the [Pre-training section](#pre-training).
-:::
-
-## Pre-training
-
-Pre-training trains on raw text corpora with no input masking. The dataset format is simple:
-
-```json
-{"text": "first row"}
-{"text": "second row"}
-```
-
-Axolotl supports two approaches:
-
-### Streaming (large datasets)
-
-For large corpora that don't fit in memory, use `pretraining_dataset` with [streaming](../streaming.qmd). Data is tokenized on-demand during training.
-
-```yaml
-pretraining_dataset:
-  - path: HuggingFaceFW/fineweb-edu
-    type: pretrain
-    text_column: text
-    split: train
-```
-
-::: {.callout-important}
-Streaming requires `max_steps` in your config — Axolotl cannot infer the dataset size. One step = `sequence_len * micro_batch_size * gradient_accumulation_steps * num_gpus` tokens.
-:::
-
-See [Streaming Datasets](../streaming.qmd) for full configuration details.
-
-### Non-streaming (smaller datasets)
-
-For datasets that fit in memory, use `type: completion` under `datasets:`. The entire dataset is pre-tokenized before training, which can be done on a CPU-only machine.
-
-```yaml
-datasets:
-  - path: my_corpus
-    type: completion
-```
-
-::: {.callout-note}
-With `completion`, texts exceeding `sequence_len` are split into multiple samples automatically.
-:::
-
-## Supervised fine-tuning (SFT)
-
-Supervised fine-tuning is the process of training models to respond to an instruction or chat input.
-
-As there are a wide variety of dataset formats, Axolotl tries to support a majority of the formats available in public datasets.
-
-Axolotl provides four approaches for loading datasets, however, it's easier to work backwards from the dataset you have available to figure out which approach to use.
-
-A flow chart is as follows:
-
-1. Do you already have the dataset tokenized? If yes, check [Pre-Tokenized Dataset](#pre-tokenized-dataset).
-
-2. Do you want to format the dataset yourself and manually choose each section to mask? If yes, check [Template Free Dataset](#template-free-dataset)
-
-3. Is your dataset in a "conversation" format, containing a `list[messages]`? If yes, check [Conversation Dataset](#conversation-dataset)
-
-4. Is your dataset in an "instruct" format, containing `{ instruction, response }`? If yes, check [Instruction Dataset](#instruction-dataset)
-
-If you went through the flow chart and did not find one that matches, it is recommended to preprocess your dataset into one of the above or create a thread on Github Discussion.
-
-::: {.callout-tip}
-You can mix and match within each approach or across approaches to train a model on a variety of datasets.
-:::
-
-### Pre-Tokenized Dataset
-
-We suggest this approach when you want to bring your own tokenized dataset.
-
-Axolotl expects the dataset to have three keys:
-
-- `input_ids`: from tokenizing formatted prompt
-- `attention_mask`: for masking padding. If you don't add padding, it would be equal to `len(input_ids) * [1]`
-- `labels`: this is the same as `input_ids`, however, if you want to mask certain tokens, you would set those indices to `-100`.
-
-::: {.callout-tip}
-Make sure to add BOS/EOS tokens to your prompt and mask it appropriately.
-:::
-
-A config for this would look like:
-
-```yaml
-datasets:
-  - path: A.jsonl
-    type:
-```
-
-::: {.callout-note}
-`type: ` is empty!
-:::
-
-Reference: [Pre-Tokenized Dataset Documentation](tokenized.qmd).
-
-### Template Free Dataset
-
-We reccomend this approach when you want granular control over the prompt formatting, special tokens, and masking, whilst letting Axolotl handle the tokenization. This is very useful if your dataset has unique prompts that differ across samples and where one single general template wouldn't suffice.
-
-In the example below, you could see that there is no proper structure. At the same time, it's very flexible as there are no constraints on how your prompt can look.
-
-```json
-{
-    "segments": [
-        {
-            "label": true,
-            "text": "<s>Hello\n"
-        },
-        {
-            "label": true,
-            "text": "hi there!. "
-        },
-        {
-            "label": false,
-            "text": "goodbye "
-        },
-        {
-            "label": true,
-            "text": "farewell</s>"
-        }
-    ]
-}
-```
-
-Each prompt must be have a key called `segments` which is a list of `{ text, label }`.
-
-```yaml
-datasets:
-  - path: A.jsonl
-    type: input_output
-```
-
-Reference: [Template Free Documentation](template_free.qmd).
-
-### Conversation Dataset
-
-`conversation` messages are a list of messages which usually contain a `role` and `content` key.
-
-::: {.callout-tip}
-Fun fact: Axolotl synonymously refers to "chat" messages as `conversation` messages due to how FastChat initially used this term to build a widely used [fastchat conversation](https://github.com/lm-sys/FastChat/blob/main/fastchat/conversation.py) method for formatting chat messages prior to the creation of `chat_templates`.
-:::
-
-#### What are `chat_templates`?
-
-The current most popular and convenient method for inference is to use `chat_templates` for formatting prompts. Axolotl supports using `chat_templates` for training to ensure that the model performs in the same environment as in inference.
-
-Here's a quick rundown on `chat_template`: A `chat_template` is a Jinja2 template which formats a list of messages into a prompt.
-
-An example of a prompt formatted into a popular template called ChatML can be seen below:
-
-Single prompt (pretty-printed):
-```json
-{
-    "messages": [
-        {
-            "role": "user",
-            "content": "Hi"
-        },
-        {
-            "role": "assistant",
-            "content": "How can I help you?"
-        },
-        {
-            "role": "user",
-            "content": "Can you add 3+5?"
-        },
-        {
-            "role": "assistant",
-            "content": "The answer is 8."
-        }
-    ]
-}
-```
-
-The ChatML template is as follows:
-```jinja2
-{% if not add_generation_prompt is defined %}{% set add_generation_prompt = false %}{% endif %}{% for message in messages %}{{'<|im_start|>' + message['role'] + '\n' + message['content'] + '<|im_end|>' + '\n'}}{% endfor %}{% if add_generation_prompt %}{{ '<|im_start|>assistant\n' }}{% endif %}
-```
-
-The above prompt formatted into this template will result in:
-
-```
-<|im_start|>user
-Hi<|im_end|>
-<|im_start|>assistant
-How can I help you?<|im_end|>
-<|im_start|>user
-Can you add 3+5?<|im_end|>
-<|im_start|>assistant
-The answer is 8.<|im_end|>
-```
-
-By using delimiters (`<|im_start|>` and `<|im_end|>`), a prompt separates different speakers which helps the model identify which portion belongs to whom.
-
-#### Common Conversation Dataset formats
-
-Older conversation datasets with the following format are colloquially called `sharegpt` datasets.
-
-```json
-{"conversations": [{"from": "...", "value": "..."}]}
-```
-
-Newer conversation datasets usually follow the OpenAI format.
-
-```json
-{"messages": [{"role": "...", "content": "..."}]}
-```
-
-Axolotl supports both as well as allowing customization of any kind of key.
-
-#### Chat Template Usage
-
-To properly use this method, it is important to identify three things:
-
-1. Which `chat_template` would you use?
-
-2. What are the keys in your dataset, and what are the possible roles? For example, in OpenAI format, the keys would be `messages`, `role`, and `content`, respectively, whereas the possible roles are `system`, `user`, and `assistant`.
-
-3. What do you want to mask? For instance, only assistant messages, only last message, or nothing.
-
-##### Choosing a `chat_template`
-
-There are a lot of `chat_templates` out there. Axolotl supports the common ones: [supported chat templates](https://github.com/axolotl-ai-cloud/axolotl/blob/860609392184cf62a7e0ca676658b170e059ce6c/src/axolotl/utils/chat_templates.py#L17). For example, to use ChatML, it would be `chat_template: chatml`.
-
-However, it is also possible to use the already configured template within the tokenizer by specifying `chat_template: tokenizer_default`. If you want a fallback (in case some tokenizer does not have it pre-configured), you can do `chat_template: tokenizer_default_fallback_chatml` to fallback to the ChatML template if a tokenizer template was not found.
-
-One last but powerful approach is to bring your own template. This can be set via:
-
-```yaml
-chat_template_jinja: # your template
-```
-
-##### Setting `chat_template` dataset keys
-
-We currently default to OpenAI format for dataset keys, so if that's your current dataset format, there's nothing to do here.
-
-If your dataset format is different, here are the keys you should check (with their defaults):
-
-```yaml
-datasets:
-    ...
-    field_messages: messages  # this should point to the key containing the list of conversations
-    message_property_mappings:  # this is a mapping from keys in your dataset to keys in chat_template
-      role: role
-      content: content
-```
-
-In some `chat_templates` (e.g. [Gemma](https://huggingface.co/google/gemma-2b-it/blob/main/tokenizer_config.json#L1507)), the roles are hardcoded to `user` and `assistant`. Consequently, you may find it necessary to map the roles in your dataset to these above. We currently have some defaults that should work for common datasets, but if you get a `KeyError`, it would be necessary to add mapping for your roles. Here is an example of how it would look like:
-
-```yaml
-datasets:
-    ...
-    roles:
-      assistant:
-        - gpt
-        - model
-      user:
-        - human
-```
-
-In the example above, all `gpt` and `model` values are converted to `assistant`. All `human` values are converted to `user.`
-
-##### Handling masking
-
-The common use case for `chat_template` is for chat messages, therefore, it is common to mask all non-assistant messages. Assistant messages refer to the bot messages that you want the model to learn on.
-
-To train on all `assistant` messages, you would set the following configs.
-
-```yaml
-datasets:
-    ...
-    roles_to_train: ["assistant"]
-    train_on_eos: "turn"
-```
-
-The `train_on_eos` config means that it would mask all EOS tokens for turns that aren't assistant-turns. The other options are: `all` and `last` to choose which EOS to train on.
-
-Perhaps, you want to train on `assistant` and `narrator` roles, you can simply add `narrator` to the list of `roles_to_train`. You would also need to add it to the mapping of `roles` above.
-
-```yaml
-datasets:
-    ...
-    roles_to_train: ["assistant", "narrator"]
-    roles:
-      assistant:
-        - gpt
-        - model
-      user:
-        - human
-      narrator: ["narrator"]
-```
-
-::: {.callout-tip}
-As chat_templates may use hardcoded EOS/EOT tokens that are different from the tokenizer's EOS, it is highly recommended to set them. For example, `ChatML` uses `<|im_end|>` to end turns.
-
-```yaml
-special_tokens:
-  eos_token: <|im_end|>
-```
-
-:::
-
-##### Applying `chat_template`
-
-Once all the above steps are completed, you could combine all these configs together to form a bespoke configuration for your custom dataset.
-
-```yaml
-datasets:
-  - path: A.jsonl
-    type: chat_template
-
-    # step 1
-    chat_template: chatml
-
-    # step 2
-    field_messages: messages
-    message_property_mappings:
-      role: role
-      content: content
-
-    roles:
-      assistant:
-        - gpt
-        - model
-        - assistant
-      user:
-        - human
-        - user
-
-    # step 3
-    roles_to_train: ["assistant"]
-    train_on_eos: "turn"
-
-special_tokens:
-  eos_token: <|im_end|>
-```
-
-If this config were to be applied to the sample dataset above, the output would look as such (which can be retrieved via `axolotl preprocess config.yaml --debug`):
-
-```
-<|im_start|>(-100, 128256) user(-100, 882)
-(-100, 198) Hi(-100, 13347) <|im_end|>(-100, 128257)
-(-100, 198) <|im_start|>(-100, 128256) assistant(-100, 78191)
-(-100, 198) How(4438, 4438)  can(649, 649)  I(358, 358)  help(1520, 1520)  you(499, 499) ?(30, 30) <|im_end|>(128257, 128257)
-(-100, 198) <|im_start|>(-100, 128256) user(-100, 882)
-(-100, 198) Can(-100, 6854)  you(-100, 499)  add(-100, 923)  (-100, 220) 3(-100, 18) +(-100, 10) 5(-100, 20) ?(-100, 30) <|im_end|>(-100, 128257)
-(-100, 198) <|im_start|>(-100, 128256) assistant(-100, 78191)
-(-100, 198) The(791, 791)  answer(4320, 4320)  is(374, 374)  (220, 220) 8(23, 23) .(13, 13) <|im_end|>(128257, 128257)
-(-100, 198)
-```
-
-The first number refers to the label, the second refers to the `token_id`. For example, `-100` labels appear on non-assistant portions, meaning that they are masked during. For assistant portions, the label is the same as the `token_id`.
-
-::: {.callout-note}
-
-If during `preprocess`, there are a lot of warnings of `Could not find content __ boundary`, please check the FAQ section for [chat_templates](../faq.qmd#chat-templates).
-
-:::
-
-#### Reference
-
-Please see docs [here](conversation.qmd).
-
-### Instruction Dataset
-
-Instruction datasets are used to train instruction-following models and comprise a prompt, containing an instruction, and a single response. In contrast to chat datasets which may be multi-turn, instruct datasets are typically single-turn.
-
-An example is of a common format called Alpaca:
-```json
-{"instruction": "...", "input": "...", "output": "..."}
-```
-
-Using those keys, a prompt can be built based on it.
-```
-Below is an instruction that describes a task, paired with an input that provides further context. Write a response that appropriately completes the request.
-
-### Instruction:
-{instruction}
-
-### Input:
-{input}
-
-### Response:
-{output}
-```
-
-This can be configured as such:
-```yaml
-datasets:
-  - path: A.jsonl
-    type: alpaca
-```
-
-Axolotl supports many kinds of instruction dataset. All of them can be found in the [Instruction Dataset Documentation](inst_tune.qmd) with their respective type and sample row format.
-
-#### Custom Instruct Prompt Format
-
-Due to the myriad possibilities of instruction formats, Axolotl allows customizing your own instruction format without having to dive into the code directly.
-
-In the example below, a sample row is used to output in `mistral_v1` format.
-```json
-{"input": "...", "output": "..."}
-```
-
-```yaml
-datasets:
-  - path: repo
-    type:
-      system_prompt: ""
-
-      field_system:
-      field_instruction: input
-      field_input:
-      field_output: output
-
-      # multi-line example with input
-      format: |-
-        [INST] {instruction} {input} [/INST]
-
-      # single-line example without input
-      no_input_format: "[INST] {instruction} [/INST]"
-```
-
-The config sets that the `field_instruction` is actually named `input`, and the `field_input` is empty as we don't have an `input` in this sample. Generally, `instruction` can be thought as the question to the model, and `input` as the additional information with `output` being the response. It is not necessary to have an `input` nor `system`. In the end, the most important part is to understand what format you want it to look like and how you can customize this to your use case.
-
-Reference: [Custom Instruct Prompt Format Documentation](inst_tune.qmd#how-to-add-custom-prompt-format).
-
-## Reinforcement Learning from Human Feedback (RLHF)
-
-As there are multiple RLHF methods with their own dataset requirements. Please see [RLHF documentation](../rlhf.qmd) for more detail.
+Below are these various formats organized by task:

docs/dataset-formats/inst_tune.qmd

@@ -186,4 +186,4 @@ datasets:
       no_input_format: "[INST] {instruction} [/INST]"
 ```
 
-See full config options under [here](../config-reference.qmd).
+See full config options under [here](../config.qmd).

docs/dataset-formats/pretraining.qmd

@@ -4,9 +4,23 @@ description: Data format for a pre-training completion task.
 order: 1
 ---
 
-::: {.callout-note}
-Pre-training documentation has been consolidated:
+For pretraining, there is no prompt template or roles.  The only required field is `text`:
+
+```{.json filename="data.jsonl"}
+{"text": "first row"}
+{"text": "second row"}
+...
+```
+
+:::{.callout-note}
+
+### Streaming is recommended for large datasets
+
+Axolotl usually loads the entire dataset into memory. This will be challenging for large datasets. Use the following config to enable streaming:
+
+```{.yaml filename="config.yaml"}
+pretraining_dataset: # hf path only
+...
+```
 
-- **Streaming pretraining** (large datasets): See [Streaming Datasets](../streaming.qmd#pretraining-with-streaming)
-- **Non-streaming pretraining** (`type: completion`): See [Dataset Formats](index.qmd#pre-training)
 :::

docs/dataset-formats/stepwise_supervised.qmd

@@ -1,26 +0,0 @@
----
-title: Stepwise Supervised Format
-description: Format for datasets with stepwise completions and labels
-order: 3
----
-
-## Stepwise Supervised
-
-The stepwise supervised format is designed for chain-of-thought (COT) reasoning
-datasets where each example contains multiple completion steps and a preference label
-for each step.
-
-### Example
-
-Here's a simple example of a stepwise supervised dataset entry:
-
-```json
-{
-  "prompt": "Which number is larger, 9.8 or 9.11?",
-  "completions": [
-    "The fractional part of 9.8 is 0.8, while the fractional part of 9.11 is 0.11.",
-    "Since 0.11 is greater than 0.8, the number 9.11 is larger than 9.8."
-  ],
-  "labels": [true, false]
-}
-```

docs/dataset-formats/template_free.qmd

@@ -1,239 +1,7 @@
 ---
 title: Template-Free
 description: Construct prompts without a template.
-toc: true
-toc-depth: 3
 order: 4
 ---
 
-## Background {#sec-background}
-
-### Masking Inputs {#masking-inputs}
-
-One of the most popular features of
-[axolotl](https://github.com/axolotl-ai-cloud/axolotl) is
-setting the following configuration value:
-
-
-```yaml
-train_on_inputs: false
-```
-
-If you declare a [dataset formats](https://github.com/axolotl-ai-cloud/axolotl?tab=readme-ov-file#dataset)
-such as `alpaca` or `chatml`, axolotl knows what is an input
-(i.e. human) vs. an output (i.e. the assistant) and masks the input
-labels so that your model can focus on predicting the outputs only.
-
-### You may not want prompt templates {#sec-you-may-not-want-prompt-templates}
-
-However, there are many situations where you don't want to use one of
-these formats or templates. This is because they can:
-
--   Add unnecessary boilerplate to your prompts.
--   Create artifacts like special delimiters `<|im_start|>` that can
-    quickly become footguns if you don't include them correctly at
-    inference time.
--   Enforce a *chat* interface when you do not want one. Sometimes you
-    just want to fine-tune a model to a very specific task and do NOT
-    want multi-turn conversations, roles, etc.
--   Limit you to only certain roles that the template allows.
-
-### The `input_output` format {#sec-the-inputoutput-format}
-
-You can construct your prompts without a template by using the
-`input_output` format, by setting `type: input_output` in your
-configuration file like this:
-
-**config.yml**
-
-```yaml
-train_on_inputs: false # Mask segments of your data
-datasets:
-  - path: output.jsonl
-    type: input_output  # use template free prompt construction
-```
-
-Unlike `type: completion`, which is also template-free,
-`type: input_output` allows you to mask segments of your text. More
-details on how this works are described below.
-
-## Usage {#sec-usage}
-
-This is how you can use the `input_output` format:
-
-### 1. Prepare Data {#sec-1-prepare-data}
-
-To use the `input_output` format, collect your data in the following
-format into a jsonl file (below is the first row from the file
-`output`.jsonl` pretty printed):
-
-```bash
-$ head -n1 output.jsonl | python -m json.tool
-```
-
-:::{.cell-output .cell-output-stdout}
-    {
-        "segments": [
-            {
-                "label": true,
-                "text": "<s>Hello\n"
-            },
-            {
-                "label": true,
-                "text": "hi there!. "
-            },
-            {
-                "label": false,
-                "text": "goodbye "
-            },
-            {
-                "label": true,
-                "text": "farewell</s>"
-            }
-        ]
-    }
-:::
-
-Set `label:false` when you want to mask a segment of text so that the
-model isn't trained on it. Some things to keep in mind:
-
-> [!IMPORTANT]
-> 1.  **EOS, BOS, spaces, newlines etc. are entirely up to you. Axolotl
-    concatenates all the segments as-is.** The tokenizer doesn't add
-    anything additional. Notice how I added spaces, newlines, `<s>`
-    (BOS), and `</s>` (EOS) myself.
-> 2.  Make sure you check the materialized output to validate that the
-    prompt is getting assembled how you like.
-
-### 2. Use `type: input_output` {#sec-2-use-type-inputoutput}
-
-Let's materialize data with our `output.jsonl` file by setting
-`type: input_output` in our axolotl config:
-
-```yaml
-# training_config.yaml
-base_model: mistralai/Mistral-7B-v0.1
-data_seed: 49
-seed: 49
-
-datasets:
-  - path: output.jsonl
-    type: input_output
-val_set_size: 0.1
-
-sequence_len: 896
-sample_packing: false
-
-micro_batch_size: 2
-gradient_accumulation_steps: 3
-eval_batch_size: 2
-num_epochs: 1
-learning_rate: 0.0002
-
-train_on_inputs: false
-special_tokens:
-  bos_token: "<s>"
-  eos_token: "</s>"
-  unk_token: "<unk>"
-```
-
-You can use the following command to materialize your data. The
-`--debug` flag will print the tokens, along with the labels so you can
-verify that the correct items are being ignored:
-
-```bash
-axolotl preprocess training_config.yaml --debug
-
-...
-[2024-03-05 23:36:46,969] [INFO] [axolotl.check_example_labels:35] [PID:607731] [RANK:0] <s>(1, 1) Hello(22557, 22557)
-(13, 13) hi(12014, 12014) there(736, 736) !(28808, 28808) .(28723, 28723) (28705, 28705) good(-100, 1179) bye(-100, 17664) (-100, 28705) fare(19111, 19111) well(5458, 5458) </s>(2, 2)
-
-```
-
-The format is `decoded_token`(`label`, `token_id`), for example,
-`<s>(1, 1)` means that the token is `<s>`, the label is `1` and the
-token_id is `1`. When the label is `-100` then that token is ignored for
-training.
-
-### 3. Check the prompts {#sec-3-check-the-prompts}
-
-Here is another way to check the materialized output:
-
-```python
-from transformers import AutoTokenizer
-from datasets import load_from_disk
-import yaml
-
-directory = !ls last_run_prepared/
-with open('training_config.yaml', 'r') as f:
-    cfg = yaml.safe_load(f)
-model_id = cfg['base_model']
-tok = AutoTokenizer.from_pretrained(model_id)
-ds = load_from_disk(f'last_run_prepared/{directory[0]}/')
-```
-
-```python
->>> row = ds[0]
->>> print(tok.decode(row['input_ids']))
-<s> Hello
-    hi there!.  goodbye  farewell</s>
-```
-
-We can check that the right tokens are ignored by comparing the labels
-to each token:
-
-```python
-import pandas as pd
-pd.DataFrame([{'token': tok.decode(i), 'label': l, 'id':i} for i,l in
-              zip(row['input_ids'], row['labels'])])
-```
-
-| token | label | id    |
-|-------|-------|-------|
-| 0     | \<s\> | 1     |
-| 1     | Hello | 22557 |
-| 2     | \\n   | 13    |
-| 3     | hi    | 12014 |
-| 4     | there | 736   |
-| 5     | !     | 28808 |
-| 6     | .     | 28723 |
-| 7     |       | 28705 |
-| 8     | good  | -100  |
-| 9     | bye   | -100  |
-| 10    |       | -100  |
-| 11    | fare  | 19111 |
-| 12    | well  | 5458  |
-| 13    | \</s\>| 2     |
-
-
-
-If we look at the input data, the above table seems correct! (The jsonl
-version is repeated below for reference):
-
-
-```bash
-$ head -n1 output.jsonl | python -m json.tool
-```
-
-:::{.cell-output .cell-output-stdout}
-    {
-        "segments": [
-            {
-                "label": true,
-                "text": "<s>Hello\n"
-            },
-            {
-                "label": true,
-                "text": "hi there!. "
-            },
-            {
-                "label": false,
-                "text": "goodbye "
-            },
-            {
-                "label": true,
-                "text": "farewell</s>"
-            }
-        ]
-    }
-:::
+See [these docs](../input_output.qmd).

docs/dataset_loading.qmd

@@ -1,268 +0,0 @@
----
-title: Dataset Loading
-description: Understanding how to load datasets from different sources
-back-to-top-navigation: true
-toc: true
-toc-depth: 5
----
-
-## Overview
-
-Datasets can be loaded in a number of different ways depending on the how it is saved (the extension of the file) and where it is stored.
-
-## Loading Datasets
-
-We use the `datasets` library to load datasets and a mix of `load_dataset` and `load_from_disk` to load them.
-
-You may recognize the similar named configs between `load_dataset` and the `datasets` section of the config file.
-
-```yaml
-datasets:
-  - path:
-    name:
-    data_files:
-    split:
-    revision:
-    trust_remote_code:
-```
-
-::: {.callout-tip}
-
-Do not feel overwhelmed by the number of options here. A lot of them are optional. In fact, the most common config to use would be `path` and sometimes `data_files`.
-
-:::
-
-This matches the API of [`datasets.load_dataset`](https://github.com/huggingface/datasets/blob/0b5998ac62f08e358f8dcc17ec6e2f2a5e9450b6/src/datasets/load.py#L1838-L1858), so if you're familiar with that, you will feel right at home.
-
-For HuggingFace's guide to load different dataset types, see [here](https://huggingface.co/docs/datasets/loading).
-
-For full details on the config, see [config-reference.qmd](config-reference.qmd).
-
-::: {.callout-note}
-
-You can set multiple datasets in the config file by more than one entry under `datasets`.
-
-```yaml
-datasets:
-  - path: /path/to/your/dataset
-  - path: /path/to/your/other/dataset
-```
-
-:::
-
-### Local dataset
-
-#### Files
-
-To load a JSON file, you would do something like this:
-
-```python
-from datasets import load_dataset
-
-dataset = load_dataset("json", data_files="data.json")
-```
-
-Which translates to the following config:
-
-```yaml
-datasets:
-  - path: data.json
-    ds_type: json
-```
-
-In the example above, it can be seen that we can just point the `path` to the file or directory along with the `ds_type` to load the dataset.
-
-This works for CSV, JSON, Parquet, and Arrow files.
-
-::: {.callout-tip}
-
-If `path` points to a file and `ds_type` is not specified, we will automatically infer the dataset type from the file extension, so you could omit `ds_type` if you'd like.
-
-:::
-
-#### Directory
-
-If you're loading a directory, you can point the `path` to the directory.
-
-Then, you have two options:
-
-##### Loading entire directory
-
-You do not need any additional configs.
-
-We will attempt to load in the following order:
-- datasets saved with `datasets.save_to_disk`
-- loading entire directory of files (such as with parquet/arrow files)
-
-```yaml
-datasets:
-  - path: /path/to/your/directory
-```
-
-##### Loading specific files in directory
-
-Provide `data_files` with a list of files to load.
-
-```yaml
-datasets:
-    # single file
-  - path: /path/to/your/directory
-    ds_type: csv
-    data_files: file1.csv
-
-    # multiple files
-  - path: /path/to/your/directory
-    ds_type: json
-    data_files:
-      - file1.jsonl
-      - file2.jsonl
-
-    # multiple files for parquet
-  - path: /path/to/your/directory
-    ds_type: parquet
-    data_files:
-      - file1.parquet
-      - file2.parquet
-
-```
-
-### HuggingFace Hub
-
-The method you use to load the dataset depends on how the dataset was created, whether a folder was uploaded directly or a HuggingFace Dataset was pushed.
-
-::: {.callout-note}
-
-If you're using a private dataset, you will need to enable the `hf_use_auth_token` flag in the root-level of the config file.
-
-:::
-
-#### Folder uploaded
-
-This would mean that the dataset is a single file or file(s) uploaded to the Hub.
-
-```yaml
-datasets:
-  - path: org/dataset-name
-    data_files:
-      - file1.jsonl
-      - file2.jsonl
-```
-
-#### HuggingFace Dataset
-
-This means that the dataset is created as a HuggingFace Dataset and pushed to the Hub via `datasets.push_to_hub`.
-
-```yaml
-datasets:
-  - path: org/dataset-name
-```
-
-::: {.callout-note}
-
-There are some other configs which may be required like `name`, `split`, `revision`, `trust_remote_code`, etc depending on the dataset.
-
-:::
-
-### Remote Filesystems
-
-Via the `storage_options` config under `load_dataset`, you can load datasets from remote filesystems like S3, GCS, Azure, and OCI.
-
-::: {.callout-warning}
-
-This is currently experimental. Please let us know if you run into any issues!
-
-:::
-
-The only difference between the providers is that you need to prepend the path with the respective protocols.
-
-```yaml
-datasets:
-    # Single file
-  - path: s3://bucket-name/path/to/your/file.jsonl
-
-    # Directory
-  - path: s3://bucket-name/path/to/your/directory
-```
-
-For directory, we load via `load_from_disk`.
-
-#### S3
-
-Prepend the path with `s3://`.
-
-The credentials are pulled in the following order:
-
-- `AWS_ACCESS_KEY_ID`, `AWS_SECRET_ACCESS_KEY`, and `AWS_SESSION_TOKEN` environment variables
-- from the `~/.aws/credentials` file
-- for nodes on EC2, the IAM metadata provider
-
-::: {.callout-note}
-
-We assume you have credentials setup and not using anonymous access. If you want to use anonymous access, let us know! We may have to open a config option for this.
-
-:::
-
-Other environment variables that can be set can be found in [boto3 docs](https://boto3.amazonaws.com/v1/documentation/api/latest/guide/configuration.html#using-environment-variables)
-
-#### GCS
-
-Prepend the path with `gs://` or `gcs://`.
-
-The credentials are loaded in the following order:
-
-- gcloud credentials
-- for nodes on GCP, the google metadata service
-- anonymous access
-
-#### Azure
-
-##### Gen 1
-
-Prepend the path with `adl://`.
-
-Ensure you have the following environment variables set:
-
-- `AZURE_STORAGE_TENANT_ID`
-- `AZURE_STORAGE_CLIENT_ID`
-- `AZURE_STORAGE_CLIENT_SECRET`
-
-##### Gen 2
-
-Prepend the path with `abfs://` or `az://`.
-
-Ensure you have the following environment variables set:
-
-- `AZURE_STORAGE_ACCOUNT_NAME`
-- `AZURE_STORAGE_ACCOUNT_KEY`
-
-Other environment variables that can be set can be found in [adlfs docs](https://github.com/fsspec/adlfs?tab=readme-ov-file#setting-credentials)
-
-#### OCI
-
-Prepend the path with `oci://`.
-
-It would attempt to read in the following order:
-
-- `OCIFS_IAM_TYPE`, `OCIFS_CONFIG_LOCATION`, and `OCIFS_CONFIG_PROFILE` environment variables
-- when on OCI resource, resource principal
-
-Other environment variables:
-
-- `OCI_REGION_METADATA`
-
-Please see the [ocifs docs](https://ocifs.readthedocs.io/en/latest/getting-connected.html#Using-Environment-Variables).
-
-### HTTPS
-
-The path should start with `https://`.
-
-```yaml
-datasets:
-  - path: https://path/to/your/dataset/file.jsonl
-```
-
-This must be publically accessible.
-
-## Next steps
-
-Now that you know how to load datasets, you can learn more on how to load your specific dataset format into your target output format [dataset formats docs](dataset-formats).

docs/dataset_preprocessing.qmd

@@ -3,11 +3,8 @@ title: Dataset Preprocessing
 description: How datasets are processed
 ---
 
-## Overview
-
 Dataset pre-processing is the step where Axolotl takes each dataset you've configured alongside
-the [dataset format](dataset-formats) and prompt strategies to:
-
+the (dataset format)[../dataset-formats/] and prompt strategies to:
  - parse the dataset based on the *dataset format*
  - transform the dataset to how you would interact with the model based on the *prompt strategy*
  - tokenize the dataset based on the configured model & tokenizer
@@ -15,12 +12,10 @@ the [dataset format](dataset-formats) and prompt strategies to:
 
 The processing of the datasets can happen one of two ways:
 
-1. Before kicking off training by calling `axolotl preprocess config.yaml --debug`
+1. Before kicking off training by calling `python -m axolotl.cli.preprocess /path/to/your.yaml --debug`
 2. When training is started
 
-### What are the benefits of pre-processing?
-
-When training interactively or for sweeps
+What are the benefits of pre-processing? When training interactively or for sweeps
 (e.g. you are restarting the trainer often), processing the datasets can oftentimes be frustratingly
 slow. Pre-processing will cache the tokenized/formatted datasets according to a hash of dependent
 training parameters so that it will intelligently pull from its cache when possible.
@@ -33,12 +28,8 @@ default path of `./last_run_prepared/`, but will ignore anything already cached
 setting `dataset_prepared_path: ./last_run_prepared`, the trainer will use whatever pre-processed
 data is in the cache.
 
-### What are the edge cases?
-
-Let's say you are writing a custom prompt strategy or using a user-defined
+What are the edge cases? Let's say you are writing a custom prompt strategy or using a user-defined
 prompt template. Because the trainer cannot readily detect these changes, we cannot change the
-calculated hash value for the pre-processed dataset.
-
-If you have `dataset_prepared_path: ...` set
+calculated hash value for the pre-processed dataset. If you have `dataset_prepared_path: ...` set
 and change your prompt templating logic, it may not pick up the changes you made and you will be
 training over the old prompt.

docs/debugging.qmd

@@ -6,10 +6,6 @@ description: How to debug Axolotl
 
 This document provides some tips and tricks for debugging Axolotl.  It also provides an example configuration for debugging with VSCode.  A good debugging setup is essential to understanding how Axolotl code works behind the scenes.
 
-::: {.callout-tip}
-For training-specific debugging (loss spikes, NaN gradients, OOM errors, RL training stability), see [Training Stability & Debugging](training_stability.qmd).
-:::
-
 ## Table of Contents
 
 - [General Tips](#general-tips)
@@ -33,15 +29,13 @@ While debugging it's helpful to simplify your test scenario as much as possible.
 1. **Make sure you are using the latest version of axolotl**:  This project changes often and bugs get fixed fast.  Check your git branch and make sure you have pulled the latest changes from `main`.
 1. **Eliminate concurrency**: Restrict the number of processes to 1 for both training and data preprocessing:
     - Set `CUDA_VISIBLE_DEVICES` to a single GPU, ex: `export CUDA_VISIBLE_DEVICES=0`.
-    - Set `dataset_num_proc: 1` in your axolotl config or run the training command with `--dataset_num_proc=1`.
+    - Set `dataset_processes: 1` in your axolotl config or run the training command with `--dataset_processes=1`.
 2. **Use a small dataset**: Construct or use a small dataset from HF Hub. When using a small dataset, you will often have to make sure `sample_packing: False` and `eval_sample_packing: False` to avoid errors.  If you are in a pinch and don't have time to construct a small dataset but want to use from the HF Hub, you can shard the data (this will still tokenize the entire dataset, but will only use a fraction of the data for training.  For example, to shard the dataset into 20 pieces, add the following to your axolotl config):
-
     ```yaml
-    datasets:
+    dataset:
         ...
         shards: 20
     ```
-
 3. **Use a small model**: A good example of a small model is [TinyLlama/TinyLlama-1.1B-Chat-v1.0](https://huggingface.co/TinyLlama/TinyLlama-1.1B-Chat-v1.0).
 4. **Minimize iteration time**: Make sure the training loop finishes as fast as possible, with these settings.
     - `micro_batch_size: 1`
@@ -76,10 +70,8 @@ datasets:
 Make sure you have an [editable install](https://setuptools.pypa.io/en/latest/userguide/development_mode.html) of Axolotl, which ensures that changes you make to the code are reflected at runtime.  Run the following commands from the root of this project:
 
 ```bash
-export UV_TORCH_BACKEND=cu128  # or cu130
-uv venv --no-project --relocatable
-source .venv/bin/activate
-uv pip install --no-build-isolation -e '.[deepspeed]' --group dev --group test
+pip3 install packaging
+pip3 install -e '.[flash-attn,deepspeed]'
 ```
 
 #### Remote Hosts
@@ -91,9 +83,9 @@ If you developing on a remote host, you can easily use VSCode to debug remotely.
 
 The easiest way to get started is to modify the [.vscode/launch.json](../.vscode/launch.json) file in this project.  This is just an example configuration, so you may need to modify or copy it to suit your needs.
 
-For example, to mimic the command `cd devtools && CUDA_VISIBLE_DEVICES=0 axolotl train dev_chat_template.yml`, you would use the below configuration[^1].  Note that we add additional flags that override the axolotl config and incorporate the tips above (see the comments). We also set the working directory to `devtools` and set the `env` variable `HF_HOME` to a temporary folder that is later partially deleted.  This is because we want to delete the HF dataset cache before each run in order to ensure that the data preprocessing code is run from scratch.
+For example, to mimic the command `cd devtools && CUDA_VISIBLE_DEVICES=0 accelerate launch -m axolotl.cli.train dev_chat_template.yml`, you would use the below configuration[^1].  Note that we add additional flags that override the axolotl config and incorporate the tips above (see the comments). We also set the working directory to `devtools` and set the `env` variable `HF_HOME` to a temporary folder that is later partially deleted.  This is because we want to delete the HF dataset cache before each run in order to ensure that the data preprocessing code is run from scratch.
 
-```json
+```jsonc
 // .vscode/launch.json
 {
     "version": "0.2.0",
@@ -107,7 +99,7 @@ For example, to mimic the command `cd devtools && CUDA_VISIBLE_DEVICES=0 axolotl
                 "-m", "axolotl.cli.train", "dev_chat_template.yml",
                 // The flags below simplify debugging by overriding the axolotl config
                 // with the debugging tips above.  Modify as needed.
-                "--dataset_num_proc=1",      // limits data preprocessing to one process
+                "--dataset_processes=1",      // limits data preprocessing to one process
                 "--max_steps=1",              // limits training to just one step
                 "--batch_size=1",             // minimizes batch size
                 "--micro_batch_size=1",       // minimizes batch size
@@ -140,7 +132,7 @@ For example, to mimic the command `cd devtools && CUDA_VISIBLE_DEVICES=0 axolotl
 
 Below is the [./vscode/tasks.json](../.vscode/tasks.json) file that defines the `cleanup-for-dataprep` task.  This task is run before each debugging session when you use the above configuration.  Note how there are two tasks that delete the two folders mentioned above.  The third task `cleanup-for-dataprep` is a composite task that combines the two tasks.  A composite task is necessary because VSCode does not allow you to specify multiple tasks in the `preLaunchTask` argument of the `launch.json` file.
 
-```json
+```jsonc
 // .vscode/tasks.json
 // this file is used by launch.json
 {
@@ -210,18 +202,17 @@ cd axolotl
 Next, run the desired docker image and mount the current directory. Below is a docker command you can run to do this:[^2]
 
 ```bash
-docker run --privileged --gpus '"all"' --shm-size 10g --rm -it --name axolotl --ipc=host --ulimit memlock=-1 --ulimit stack=67108864 --mount type=bind,src="${PWD}",target=/workspace/axolotl -v ${HOME}/.cache/huggingface:/root/.cache/huggingface axolotlai/axolotl-uv:main-latest
+docker run --privileged --gpus '"all"' --shm-size 10g --rm -it --name axolotl --ipc=host --ulimit memlock=-1 --ulimit stack=67108864 --mount type=bind,src="${PWD}",target=/workspace/axolotl -v ${HOME}/.cache/huggingface:/root/.cache/huggingface axolotlai/axolotl:main-py3.10-cu118-2.0.1
 ```
 
 >[!Tip]
 > To understand which containers are available, see the [Docker section of the README](../README.md#docker) and the [DockerHub repo](https://hub.docker.com/r/axolotlai/axolotl/tags).  For details of how the Docker containers are built, see axolotl's [Docker CI builds](../.github/workflows/main.yml).
 
-You will now be in the container.  Next, install Axolotl with dev dependencies:
+You will now be in the container.  Next, perform an editable install of Axolotl:
 
 ```bash
-uv venv --no-project --relocatable
-source .venv/bin/activate
-uv pip install --no-build-isolation -e '.[deepspeed]' --group dev --group test
+pip3 install packaging
+pip3 install -e '.[flash-attn,deepspeed]'
 ```
 
 ### Attach To Container
@@ -249,6 +240,6 @@ style="border-radius: 10px; display: block; margin: auto;" width="560" height="3
 </div>
 <br>
 
-[^1]: The VSCode config uses `accelerate.commands.launch` as the Python module entry point, which is what `axolotl train` invokes under the hood.
+[^1]: The config actually mimics the command `CUDA_VISIBLE_DEVICES=0 python -m accelerate.commands.launch -m axolotl.cli.train devtools/chat_template.yml`, but this is the same thing.
 
 [^2]: Many of the below flags are recommended best practices by Nvidia when using nvidia-container-toolkit.  You can read more about these flags [here](https://docs.nvidia.com/deeplearning/frameworks/user-guide/index.html).