offload activations to disk instead of CPU RAM

2025-05-11 14:19:49 -04:00
766 changed files with 19385 additions and 51482 deletions
--- a/.axolotl-complete.bash
+++ b/.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
--- a/.bandit
+++ b/.bandit
@@ -1,3 +1,3 @@
 [bandit]
 exclude = tests
-skips = B101,B615,B102,B110
+skips = B101
--- a/.coderabbit.yaml
+++ b/.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
--- a/.flake8
+++ b/.flake8
@@ -0,0 +1,5 @@
+[flake8]
+max-line-length = 88
+
+select = C,E,F,W,B,B950
+extend-ignore = E203, E501, W503
--- a/.github/CONTRIBUTING.md
+++ b/.github/CONTRIBUTING.md
@@ -57,13 +57,6 @@ 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
--- a/.github/workflows/base.yml
+++ b/.github/workflows/base.yml
@@ -5,76 +5,65 @@ 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:

 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
+    runs-on: axolotl-gpu-runner
    strategy:
      fail-fast: false
      matrix:
        include:
+          - 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"
+          - cuda: "124"
+            cuda_version: 12.4.1
+            cudnn_version: ""
+            python_version: "3.11"
+            pytorch: 2.6.0
+            torch_cuda_arch_list: "7.0 7.5 8.0 8.6 8.7 8.9 9.0+PTX"
+          - cuda: "126"
+            cuda_version: 12.6.3
+            cudnn_version: ""
+            python_version: "3.11"
+            pytorch: 2.6.0
+            torch_cuda_arch_list: "7.0 7.5 8.0 8.6 8.7 8.9 9.0+PTX"
          - cuda: "126"
            cuda_version: 12.6.3
            cudnn_version: ""
            python_version: "3.11"
            pytorch: 2.7.0
            torch_cuda_arch_list: "7.0 7.5 8.0 8.6 8.7 8.9 9.0+PTX"
-            dockerfile: "Dockerfile-base"
-          - cuda: "126"
+          - cuda: "128"
            cuda_version: 12.6.3
            cudnn_version: ""
            python_version: "3.11"
-            pytorch: 2.7.1
+            pytorch: 2.7.0
            torch_cuda_arch_list: "7.0 7.5 8.0 8.6 8.7 8.9 9.0+PTX"
-            dockerfile: "Dockerfile-base"
          - cuda: "128"
            cuda_version: 12.8.1
            cudnn_version: ""
            python_version: "3.11"
-            pytorch: 2.7.1
+            pytorch: nightly
            torch_cuda_arch_list: "7.0 7.5 8.0 8.6 8.7 8.9 9.0+PTX"
-            dockerfile: "Dockerfile-base"
          - cuda: "128"
            cuda_version: 12.8.1
            cudnn_version: ""
            python_version: "3.11"
-            pytorch: 2.8.0
+            pytorch: next
            torch_cuda_arch_list: "7.0 7.5 8.0 8.6 8.7 8.9 9.0+PTX"
-            dockerfile: "Dockerfile-base"
-          - cuda: "128"
-            cuda_version: 12.8.1
-            cudnn_version: ""
-            python_version: "3.11"
-            pytorch: 2.9.0
-            torch_cuda_arch_list: "7.0 7.5 8.0 8.6 8.7 8.9 9.0+PTX"
-            dockerfile: "Dockerfile-base"
-#          - cuda: "128"
-#            cuda_version: 12.8.1
-#            cudnn_version: ""
-#            python_version: "3.11"
-#            pytorch: nightly
-#            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"
    steps:
      - name: Checkout
        uses: actions/checkout@v4
@@ -96,74 +85,7 @@ jobs:
        uses: docker/build-push-action@v4
        with:
          context: .
-          file: ./docker/${{ matrix.dockerfile }}
-          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
-    strategy:
-      fail-fast: false
-      matrix:
-        include:
-          - cuda: "126"
-            cuda_version: 12.6.3
-            cudnn_version: ""
-            python_version: "3.11"
-            pytorch: 2.7.1
-            torch_cuda_arch_list: "7.0 7.5 8.0 8.6 8.7 8.9 9.0+PTX"
-            dockerfile: "Dockerfile-uv-base"
-          - cuda: "128"
-            cuda_version: 12.8.1
-            cudnn_version: ""
-            python_version: "3.11"
-            pytorch: 2.7.1
-            torch_cuda_arch_list: "7.0 7.5 8.0 8.6 8.7 8.9 9.0+PTX"
-            dockerfile: "Dockerfile-uv-base"
-          - cuda: "128"
-            cuda_version: 12.8.1
-            cudnn_version: ""
-            python_version: "3.11"
-            pytorch: 2.8.0
-            torch_cuda_arch_list: "7.0 7.5 8.0 8.6 8.7 8.9 9.0+PTX"
-            dockerfile: "Dockerfile-uv-base"
-          - cuda: "128"
-            cuda_version: 12.8.1
-            cudnn_version: ""
-            python_version: "3.11"
-            pytorch: 2.9.0
-            torch_cuda_arch_list: "7.0 7.5 8.0 8.6 8.7 8.9 9.0+PTX"
-            dockerfile: "Dockerfile-uv-base"
-    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@v2
-        with:
-          username: ${{ secrets.DOCKERHUB_USERNAME }}
-          password: ${{ secrets.DOCKERHUB_TOKEN }}
-      - name: Set up Docker Buildx
-        uses: docker/setup-buildx-action@v3
-      - name: Build
-        uses: docker/build-push-action@v4
-        with:
-          context: .
-          file: ./docker/${{ matrix.dockerfile }}
+          file: ${{ matrix.pytorch == 'nightly' && './docker/Dockerfile-base-nightly' || matrix.pytorch == 'next' && './docker/Dockerfile-base-next' || './docker/Dockerfile-base' }}
          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 }}
--- a/.github/workflows/docs.yml
+++ b/.github/workflows/docs.yml
@@ -23,7 +23,7 @@ jobs:
        - name: Install dependencies
          run: |
            python3 -m pip install jupyter quartodoc
-            python3 -m pip install -e .
+            python3 -m pip install -e . --no-deps
        - name: Build autodoc
          run: quartodoc build
        - name: Publish to GitHub Pages (and render)
--- a/.github/workflows/lint.yml
+++ b/.github/workflows/lint.yml
@@ -3,21 +3,18 @@ on:
  # check on PRs, and manual triggers
  merge_group:
  pull_request:
-      types: [opened, synchronize, reopened, ready_for_review]
      paths:
       - '**.py'
       - 'requirements.txt'
       - '.github/workflows/*.yml'
       - "*.[q]md"
       - "examples/**/*.y[a]?ml"
-       - ".pre-commit-config.yaml"
  workflow_dispatch:

 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
--- a/.github/workflows/main.yml
+++ b/.github/workflows/main.yml
@@ -15,27 +15,22 @@ jobs:
      fail-fast: false
      matrix:
        include:
+          - cuda: 124
+            cuda_version: 12.4.1
+            python_version: "3.11"
+            pytorch: 2.5.1
+            axolotl_extras:
+          - cuda: 124
+            cuda_version: 12.4.1
+            python_version: "3.11"
+            pytorch: 2.6.0
+            axolotl_extras: vllm
+            is_latest: true
          - cuda: 126
            cuda_version: 12.6.3
            python_version: "3.11"
            pytorch: 2.7.0
            axolotl_extras:
-          - cuda: 126
-            cuda_version: 12.6.3
-            python_version: "3.11"
-            pytorch: 2.7.1
-            axolotl_extras: vllm
-            is_latest: true
-          - cuda: 128
-            cuda_version: 12.8.1
-            python_version: "3.11"
-            pytorch: 2.7.1
-            axolotl_extras:
-          - cuda: 128
-            cuda_version: 12.8.1
-            python_version: "3.11"
-            pytorch: 2.8.0
-            axolotl_extras:
    runs-on: axolotl-gpu-runner
    steps:
      - name: Checkout
@@ -83,33 +78,22 @@ jobs:
    strategy:
      matrix:
        include:
+          - cuda: 124
+            cuda_version: 12.4.1
+            python_version: "3.11"
+            pytorch: 2.5.1
+            axolotl_extras:
+          - cuda: 124
+            cuda_version: 12.4.1
+            python_version: "3.11"
+            pytorch: 2.6.0
+            axolotl_extras:
+            is_latest: true
          - cuda: 126
            cuda_version: 12.6.3
            python_version: "3.11"
            pytorch: 2.7.0
            axolotl_extras:
-          - cuda: 126
-            cuda_version: 12.6.3
-            python_version: "3.11"
-            pytorch: 2.7.1
-            axolotl_extras:
-            is_latest:
-          - cuda: 126
-            cuda_version: 12.6.3
-            python_version: "3.11"
-            pytorch: 2.7.1
-            axolotl_extras: vllm
-            is_latest: true
-          - cuda: 128
-            cuda_version: 12.8.1
-            python_version: "3.11"
-            pytorch: 2.7.1
-            axolotl_extras:
-          - cuda: 128
-            cuda_version: 12.8.1
-            python_version: "3.11"
-            pytorch: 2.8.0
-            axolotl_extras:
    runs-on: axolotl-gpu-runner
    steps:
      - name: Checkout
@@ -152,24 +136,11 @@ jobs:
    strategy:
      matrix:
        include:
-          - cuda: 126
-            cuda_version: 12.6.3
+          - cuda: 124
+            cuda_version: 12.4.1
            python_version: "3.11"
-            pytorch: 2.7.1
+            pytorch: 2.6.0
            axolotl_extras:
-            is_latest:
-          - cuda: 126
-            cuda_version: 12.6.3
-            python_version: "3.11"
-            pytorch: 2.7.1
-            axolotl_extras: vllm
-            is_latest: true
-          - cuda: 128
-            cuda_version: 12.8.1
-            python_version: "3.11"
-            pytorch: 2.8.0
-            axolotl_extras:
-            is_latest:
    runs-on: axolotl-gpu-runner
    steps:
      - name: Checkout
--- a/.github/workflows/multi-gpu-e2e.yml
+++ b/.github/workflows/multi-gpu-e2e.yml
@@ -3,7 +3,7 @@ name: docker-multigpu-tests-biweekly
 on:
  pull_request:
    paths:
-      - 'tests/e2e/multigpu/**.py'
+      - 'tests/e2e/multigpu/*.py'
      - 'requirements.txt'
      - 'setup.py'
      - 'pyproject.toml'
@@ -21,23 +21,30 @@ concurrency:

 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: ${{ ! contains(github.event.commits[0].message, '[skip e2e]') && github.repository_owner == 'axolotl-ai-cloud' }}
    strategy:
      fail-fast: false
      matrix:
        include:
-          - cuda: 126
-            cuda_version: 12.6.3
+          - cuda: 124
+            cuda_version: 12.4.1
            python_version: "3.11"
-            pytorch: 2.7.1
+            pytorch: 2.6.0
            axolotl_extras: vllm
            num_gpus: 2
            nightly_build: "true"
-          - cuda: 128
-            cuda_version: 12.8.1
+          - cuda: 124
+            cuda_version: 12.4.1
            python_version: "3.11"
-            pytorch: 2.8.0
-            axolotl_extras: fbgemm-gpu
+            pytorch: 2.5.1
+            axolotl_extras:
+            num_gpus: 2
+            nightly_build: "true"
+          - cuda: 126
+            cuda_version: 12.6.3
+            python_version: "3.11"
+            pytorch: 2.7.0
+            axolotl_extras:
            num_gpus: 2
            nightly_build: "true"
    runs-on: [self-hosted, modal]
@@ -52,7 +59,7 @@ jobs:
      - name: Install Modal
        run: |
          python -m pip install --upgrade pip
-          pip install modal==1.0.2 jinja2
+          pip install modal==0.71.8 jinja2
      - name: Update env vars
        run: |
          echo "BASE_TAG=main-base-py${{ matrix.python_version }}-cu${{ matrix.cuda }}-${{ matrix.pytorch }}" >> $GITHUB_ENV
--- a/.github/workflows/nightlies.yml
+++ b/.github/workflows/nightlies.yml
@@ -12,15 +12,15 @@ jobs:
      fail-fast: false
      matrix:
        include:
-          - cuda: 126
-            cuda_version: 12.6.3
+          - cuda: 124
+            cuda_version: 12.4.1
            python_version: "3.11"
-            pytorch: 2.7.1
+            pytorch: 2.5.1
            axolotl_extras:
-          - cuda: 128
-            cuda_version: 12.8.1
+          - cuda: 124
+            cuda_version: 12.4.1
            python_version: "3.11"
-            pytorch: 2.8.0
+            pytorch: 2.6.0
            axolotl_extras:
    runs-on: axolotl-gpu-runner
    steps:
@@ -65,15 +65,15 @@ jobs:
    strategy:
      matrix:
        include:
-          - cuda: 126
-            cuda_version: 12.6.3
+          - cuda: 124
+            cuda_version: 12.4.1
            python_version: "3.11"
-            pytorch: 2.7.1
+            pytorch: 2.5.1
            axolotl_extras:
-          - cuda: 128
-            cuda_version: 12.8.1
+          - cuda: 124
+            cuda_version: 12.4.1
            python_version: "3.11"
-            pytorch: 2.8.0
+            pytorch: 2.6.0
            axolotl_extras:
    runs-on: axolotl-gpu-runner
    steps:
--- a/.github/workflows/precommit-autoupdate.yml
+++ b/.github/workflows/precommit-autoupdate.yml
@@ -25,6 +25,7 @@ jobs:
          pre-commit autoupdate
          if [[ -n $(git status --porcelain) ]]; then
            echo "changes=true" >> $GITHUB_OUTPUT
+            git diff .pre-commit-config.yaml > pre-commit-update.diff
          fi

      - name: Create Pull Request
@@ -38,3 +39,11 @@ jobs:
          commit-message: "chore: update pre-commit hooks"
          body: |
            Automated PR to update pre-commit hooks to their latest versions.
+
+            <details>
+            <summary>Changes:</summary>
+
+            ```diff
+            ${{ steps.update.outputs.diff }}
+            ```
+            </details>
--- a/.github/workflows/preview-docs.yml
+++ b/.github/workflows/preview-docs.yml
@@ -2,15 +2,13 @@ name: Preview
 on:
  workflow_dispatch:
  pull_request:
-    types: [opened, synchronize, reopened, ready_for_review]
+    types: [opened, synchronize, reopened]

    # 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
+      - '_quarto.yaml'

 permissions:
  checks: write
@@ -25,12 +23,9 @@ permissions:
 jobs:
  preview:
    runs-on: ubuntu-latest
-    if: ${{ !github.event.pull_request.draft }}
    steps:
      - 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
@@ -43,7 +38,7 @@ jobs:
      - name: Install dependencies
        run: |
          python3 -m pip install jupyter quartodoc
-          python3 -m pip install -e .
+          python3 -m pip install -e . --no-deps

      - name: Build autodoc
        run: quartodoc build
@@ -53,12 +48,10 @@ jobs:

      - 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
+          enable-pull-request-comment: true
+          enable-github-deployment: true
          github-token: ${{ secrets.GITHUB_TOKEN }}
          deploy-message: "Deployed On Netlify"
          github-deployment-environment: 'preview'
@@ -66,13 +59,3 @@ jobs:
        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 }}
--- a/.github/workflows/tests-nightly.yml
+++ b/.github/workflows/tests-nightly.yml
@@ -18,26 +18,31 @@ jobs:
        env:
          SKIP: no-commit-to-branch

-  pytest:
-    name: PyTest
+  preload-cache:
+    name: Preload HF cache
    runs-on: ubuntu-latest
    strategy:
      fail-fast: false
-      max-parallel: 2
      matrix:
        python_version: ["3.11"]
-        pytorch_version: ["2.7.1", "2.8.0"]
+        pytorch_version: ["2.6.0"]
    timeout-minutes: 20

+    env:
+      AXOLOTL_IS_CI_CACHE_PRELOAD: "1"
+
    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://d1dttdx32dkk5p.cloudfront.net/hf-cache.tar.zst | tar -xf - -C /home/runner/.cache/huggingface/hub/  --use-compress-program unzstd
+      - name: Restore HF cache
+        id: hf-cache-restore
+        uses: actions/cache/restore@v4
+        with:
+          path: |
+            /home/runner/.cache/huggingface/hub/datasets--*
+            /home/runner/.cache/huggingface/hub/models--*
+          key: ${{ runner.os }}-hf-hub-cache-v2

      - name: Setup Python
        uses: actions/setup-python@v5
@@ -52,7 +57,92 @@ jobs:

      - name: Install PyTorch
        run: |
-          pip3 install torch==${{ matrix.pytorch_version }} torchvision
+          pip3 install torch==${{ matrix.pytorch_version }}
+
+      - name: Install dependencies
+        run: |
+          pip3 show torch
+          pip3 install --no-build-isolation -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: Make sure PyTorch version wasn't clobbered
+        run: |
+          python -c "import torch; assert '${{ matrix.pytorch_version }}' in torch.__version__"
+
+      - name: Ensure axolotl CLI was installed
+        run: |
+          axolotl --help
+
+      - name: Pre-Download dataset fixture
+        run: |
+          huggingface-cli download --repo-type=dataset axolotl-ai-internal/axolotl-oss-dataset-fixtures
+
+      - name: Run tests
+        run: |
+          pytest -v tests/conftest.py
+
+      - 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 {} \;
+
+      - name: Save HF cache
+        id: hf-cache
+        uses: actions/cache/save@v4
+        with:
+          path: |
+            /home/runner/.cache/huggingface/hub/datasets--*
+            /home/runner/.cache/huggingface/hub/models--*
+          key: ${{ steps.hf-cache-restore.outputs.cache-primary-key }}
+
+  pytest:
+    name: PyTest
+    runs-on: ubuntu-latest
+    needs: [preload-cache]
+    strategy:
+      fail-fast: false
+      max-parallel: 2
+      matrix:
+        python_version: ["3.11"]
+        pytorch_version: ["2.5.1", "2.6.0", "2.7.0"]
+    timeout-minutes: 20
+
+    steps:
+      - name: Check out repository code
+        uses: actions/checkout@v4
+
+      - name: Restore HF cache
+        id: hf-cache-restore
+        uses: actions/cache/restore@v4
+        with:
+          path: |
+            /home/runner/.cache/huggingface/hub/datasets--*
+            /home/runner/.cache/huggingface/hub/models--*
+          key: ${{ runner.os }}-hf-hub-cache-v2
+
+      - name: Setup Python
+        uses: actions/setup-python@v5
+        with:
+          python-version: ${{ matrix.python_version }}
+          cache: 'pip' # caching pip dependencies
+
+      - name: upgrade pip
+        run: |
+          pip3 install --upgrade pip
+          pip3 install --upgrade packaging==23.2 setuptools==75.8.0 wheel
+
+      - name: Install PyTorch
+        run: |
+          pip3 install torch==${{ matrix.pytorch_version }}

      - name: Update requirements.txt
        run: |
@@ -78,11 +168,15 @@ jobs:
        run: |
          axolotl --help

+      - name: Pre-Download dataset fixture
+        run: |
+          huggingface-cli download --repo-type=dataset axolotl-ai-internal/axolotl-oss-dataset-fixtures
+
      - 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 -v -n8 --dist loadfile --ignore=tests/e2e/ --ignore=tests/patched/ --ignore=tests/cli/ tests/
+          pytest -v tests/patched/
+          pytest -v tests/cli/

      - name: cleanup pip cache
        run: |
@@ -92,24 +186,24 @@ jobs:
    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: 126
-            cuda_version: 12.6.3
+          - cuda: 124
+            cuda_version: 12.4.1
            python_version: "3.11"
-            pytorch: 2.7.1
+            pytorch: 2.5.1
            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: 2.8.0
+            pytorch: 2.6.0
            num_gpus: 1
            axolotl_extras:
            nightly_build: "true"
@@ -123,7 +217,7 @@ jobs:
      - name: Install Modal
        run: |
          python -m pip install --upgrade pip
-          pip install modal==1.0.2 jinja2
+          pip install modal==0.71.8 jinja2
      - name: Update env vars
        run: |
          echo "BASE_TAG=main-base-py${{ matrix.python_version }}-cu${{ matrix.cuda }}-${{ matrix.pytorch }}" >> $GITHUB_ENV
@@ -137,45 +231,3 @@ jobs:
      - name: Run tests job on Modal
        run: |
          modal run cicd.e2e_tests
-  docker-e2e-multigpu-tests:
-    if: github.repository_owner == 'axolotl-ai-cloud'
-    # this job needs to be run on self-hosted GPU runners...
-    runs-on: [self-hosted, modal]
-    timeout-minutes: 120
-    needs: [pre-commit, pytest, docker-e2e-tests]
-
-    strategy:
-      fail-fast: false
-      matrix:
-        include:
-          - cuda: 126
-            cuda_version: 12.6.3
-            python_version: "3.11"
-            pytorch: 2.7.1
-            num_gpus: 2
-            axolotl_extras:
-            nightly_build: "true"
-    steps:
-      - name: Checkout
-        uses: actions/checkout@v4
-      - name: Install Python
-        uses: actions/setup-python@v5
-        with:
-          python-version: "3.11"
-      - name: Install Modal
-        run: |
-          python -m pip install --upgrade pip
-          pip install modal==1.0.2 jinja2
-      - name: Update env vars
-        run: |
-          echo "BASE_TAG=main-base-py${{ 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 }}" >> $GITHUB_ENV
-          echo "NIGHTLY_BUILD=${{ matrix.nightly_build }}" >> $GITHUB_ENV
-          echo "CODECOV_TOKEN=${{ secrets.CODECOV_TOKEN }}" >> $GITHUB_ENV
-      - name: Run tests job on Modal
-        run: |
-          modal run cicd.multigpu
--- a/.github/workflows/tests.yml
+++ b/.github/workflows/tests.yml
@@ -13,7 +13,6 @@ on:
      - 'cicd/cicd.sh'
      - 'cicd/Dockerfile.jinja'
  pull_request:
-      types: [opened, synchronize, reopened, ready_for_review]
      paths:
       - '**.py'
       - 'requirements.txt'
@@ -35,7 +34,6 @@ 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
@@ -46,27 +44,31 @@ jobs:
        env:
          SKIP: no-commit-to-branch

-  pytest:
-    name: PyTest
+  preload-cache:
+    name: Preload HF cache
    runs-on: ubuntu-latest
-    if: ${{ !github.event.pull_request.draft }}
-#    needs: [preload-cache]
    strategy:
      fail-fast: false
      matrix:
        python_version: ["3.11"]
-        pytorch_version: ["2.7.1", "2.8.0"]
+        pytorch_version: ["2.6.0"]
    timeout-minutes: 20

+    env:
+      AXOLOTL_IS_CI_CACHE_PRELOAD: "1"
+
    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://d1dttdx32dkk5p.cloudfront.net/hf-cache.tar.zst | tar -xf - -C /home/runner/.cache/huggingface/hub/  --use-compress-program unzstd
+      - name: Restore HF cache
+        id: hf-cache-restore
+        uses: actions/cache/restore@v4
+        with:
+          path: |
+            /home/runner/.cache/huggingface/hub/datasets--*
+            /home/runner/.cache/huggingface/hub/models--*
+          key: ${{ runner.os }}-hf-hub-cache-v2

      - name: Setup Python
        uses: actions/setup-python@v5
@@ -81,12 +83,12 @@ jobs:

      - name: Install PyTorch
        run: |
-          pip3 install --no-cache-dir torch==${{ matrix.pytorch_version }} torchvision
+          pip3 install torch==${{ matrix.pytorch_version }}

      - name: Install dependencies
        run: |
          pip3 show torch
-          pip3 install --no-cache-dir --no-build-isolation -U -e .
+          pip3 install --no-build-isolation -U -e .
          python scripts/unsloth_install.py | sh
          python scripts/cutcrossentropy_install.py | sh
          pip3 install -r requirements-dev.txt -r requirements-tests.txt
@@ -105,10 +107,93 @@ jobs:

      - name: Run tests
        run: |
-          pytest -v --durations=10 -n8 --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/patched/ --cov=axolotl --cov-append --cov-report=xml
-          pytest -v --durations=10 tests/cli/ --cov=axolotl --cov-append --cov-report=xml
+          pytest -v tests/conftest.py
+
+      - 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 {} \;
+
+      - name: Save HF cache
+        id: hf-cache
+        uses: actions/cache/save@v4
+        with:
+          path: |
+            /home/runner/.cache/huggingface/hub/datasets--*
+            /home/runner/.cache/huggingface/hub/models--*
+          key: ${{ steps.hf-cache-restore.outputs.cache-primary-key }}
+
+  pytest:
+    name: PyTest
+    runs-on: ubuntu-latest
+    needs: [preload-cache]
+    strategy:
+      fail-fast: false
+      matrix:
+        python_version: ["3.11"]
+        pytorch_version: ["2.5.1", "2.6.0", "2.7.0"]
+    timeout-minutes: 20
+
+    steps:
+      - name: Check out repository code
+        uses: actions/checkout@v4
+
+      - name: Restore HF cache
+        id: hf-cache-restore
+        uses: actions/cache/restore@v4
+        with:
+          path: |
+            /home/runner/.cache/huggingface/hub/datasets--*
+            /home/runner/.cache/huggingface/hub/models--*
+          key: ${{ runner.os }}-hf-hub-cache-v2
+
+      - name: Setup Python
+        uses: actions/setup-python@v5
+        with:
+          python-version: ${{ matrix.python_version }}
+          cache: 'pip' # caching pip dependencies
+
+      - name: upgrade pip
+        run: |
+          pip3 install --upgrade pip
+          pip3 install --upgrade packaging==23.2 setuptools==75.8.0 wheel
+
+      - name: Install PyTorch
+        run: |
+          pip3 install torch==${{ matrix.pytorch_version }}
+
+      - name: Install dependencies
+        run: |
+          pip3 show torch
+          pip3 install --no-build-isolation -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: Make sure PyTorch version wasn't clobbered
+        run: |
+          python -c "import torch; assert '${{ matrix.pytorch_version }}' in torch.__version__"
+
+      - name: Ensure axolotl CLI was installed
+        run: |
+          axolotl --help
+
+      - name: Pre-Download dataset fixture
+        run: |
+          huggingface-cli download --repo-type=dataset axolotl-ai-internal/axolotl-oss-dataset-fixtures
+
+      - name: Run tests
+        run: |
+          pytest -v -n8 --dist loadfile --ignore=tests/e2e/ --ignore=tests/patched/ --ignore=tests/cli/ tests/ --cov=axolotl --cov-report=xml
+          pytest -v tests/patched/ --cov=axolotl --cov-append --cov-report=xml
+          pytest -v tests/cli/ --cov=axolotl --cov-append --cov-report=xml

      - name: Upload coverage to Codecov
        uses: codecov/codecov-action@v5
@@ -125,23 +210,26 @@ jobs:
  pytest-sdist:
    name: PyTest from Source Dist
    runs-on: ubuntu-latest
-    if: ${{ !github.event.pull_request.draft }}
+    needs: [preload-cache]
    strategy:
      fail-fast: false
      matrix:
        python_version: ["3.11"]
-        pytorch_version: ["2.7.1", "2.8.0"]
+        pytorch_version: ["2.5.1", "2.6.0", "2.7.0"]
    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://d1dttdx32dkk5p.cloudfront.net/hf-cache.tar.zst | tar -xf - -C /home/runner/.cache/huggingface/hub/  --use-compress-program unzstd
+      - name: Restore HF cache
+        id: hf-cache-restore
+        uses: actions/cache/restore@v4
+        with:
+          path: |
+            /home/runner/.cache/huggingface/hub/datasets--*
+            /home/runner/.cache/huggingface/hub/models--*
+          key: ${{ runner.os }}-hf-hub-cache-v2

      - name: Setup Python
        uses: actions/setup-python@v5
@@ -156,13 +244,13 @@ jobs:

      - name: Install PyTorch
        run: |
-          pip3 install --no-cache-dir torch==${{ matrix.pytorch_version }} torchvision
+          pip3 install torch==${{ matrix.pytorch_version }}

      - name: Install dependencies
        run: |
          pip3 show torch
          python -m build --no-isolation --sdist
-          pip3 install --no-cache-dir --no-build-isolation dist/axolotl*.tar.gz
+          pip3 install --no-build-isolation 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
@@ -180,70 +268,31 @@ jobs:

      - name: Run tests
        run: |
-          pytest -v --durations=10 -n8 --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 -v -n8 --dist loadfile --ignore=tests/e2e/ --ignore=tests/patched/ --ignore=tests/cli/ tests/
+          pytest -v tests/patched/
+          pytest -v tests/cli/

      - name: cleanup pip cache
        run: |
          find "$(pip cache dir)/http-v2" -type f -mtime +14 -exec rm {} \;

-  gate-skip-e2e:
-    needs: [pre-commit, pytest, pytest-sdist]
-    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:
-    # 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'
+    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: 120
-    needs: [pre-commit, pytest, pytest-sdist, gate-skip-e2e]
+    timeout-minutes: 90
+    needs: [pre-commit, pytest, pytest-sdist]

    strategy:
      fail-fast: false
      matrix:
        include:
-          - cuda: 126
-            cuda_version: 12.6.3
+          - cuda: 124
+            cuda_version: 12.4.1
            python_version: "3.11"
-            pytorch: 2.7.1
+            pytorch: 2.6.0
            num_gpus: 1
-            axolotl_extras:
-          - cuda: 126
-            cuda_version: 12.6.3
-            python_version: "3.11"
-            pytorch: 2.7.1
-            num_gpus: 1
-            axolotl_extras:
-            dockerfile: "Dockerfile-uv.jinja"
+            axolotl_extras: vllm
    steps:
      - name: Checkout
        uses: actions/checkout@v4
@@ -254,7 +303,7 @@ jobs:
      - name: Install Modal
        run: |
          python -m pip install --upgrade pip
-          pip install modal==1.0.2 jinja2
+          pip install modal==0.71.8 jinja2
      - name: Update env vars
        run: |
          echo "BASE_TAG=main-base-py${{ matrix.python_version }}-cu${{ matrix.cuda }}-${{ matrix.pytorch }}" >> $GITHUB_ENV
@@ -265,81 +314,43 @@ jobs:
          echo "MODAL_IMAGE_BUILDER_VERSION=2024.10" >> $GITHUB_ENV
          echo "N_GPUS=${{ matrix.num_gpus }}" >> $GITHUB_ENV
          echo "CODECOV_TOKEN=${{ secrets.CODECOV_TOKEN }}" >> $GITHUB_ENV
-          echo "E2E_DOCKERFILE=${{ matrix.dockerfile || 'Dockerfile.jinja'}}" >> $GITHUB_ENV
      - name: Run tests job on Modal
        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'
+    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
-    # 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]
+    timeout-minutes: 90
+    needs: [pre-commit, pytest, docker-e2e-tests-1st]

    strategy:
      fail-fast: false
      matrix:
        include:
-          - cuda: 128
-            cuda_version: 12.8.1
+          - cuda: 124
+            cuda_version: 12.4.1
            python_version: "3.11"
-            pytorch: 2.7.1
+            pytorch: 2.6.0
+            num_gpus: 1
+            axolotl_extras: llmcompressor
+          - cuda: 124
+            cuda_version: 12.4.1
+            python_version: "3.11"
+            pytorch: 2.4.1
            num_gpus: 1
            axolotl_extras:
-          - cuda: 128
-            cuda_version: 12.8.1
+          - cuda: 124
+            cuda_version: 12.4.1
            python_version: "3.11"
-            pytorch: 2.8.0
+            pytorch: 2.5.1
            num_gpus: 1
-            gpu_type: "B200"
-            axolotl_extras: fbgemm-gpu
-    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==1.0.2 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'}}" >> $GITHUB_ENV
-          echo "CODECOV_TOKEN=${{ secrets.CODECOV_TOKEN }}" >> $GITHUB_ENV
-          echo "E2E_DOCKERFILE=${{ matrix.dockerfile || 'Dockerfile.jinja'}}" >> $GITHUB_ENV
-      - name: Run tests job on Modal
-        run: |
-          modal run cicd.e2e_tests
-
-  docker-e2e-cleanup:
-    runs-on: [self-hosted, modal]
-    timeout-minutes: 90
-    needs: [docker-e2e-tests]
-    if: ${{ !github.event.pull_request.draft }}
-
-    strategy:
-      fail-fast: false
-      matrix:
-        include:
+            axolotl_extras:
          - cuda: 126
            cuda_version: 12.6.3
            python_version: "3.11"
-            pytorch: 2.7.1
+            pytorch: 2.7.0
            num_gpus: 1
            axolotl_extras:
    steps:
@@ -352,7 +363,7 @@ jobs:
      - name: Install Modal
        run: |
          python -m pip install --upgrade pip
-          pip install modal==1.0.2 jinja2
+          pip install modal==0.71.8 jinja2
      - name: Update env vars
        run: |
          echo "BASE_TAG=main-base-py${{ matrix.python_version }}-cu${{ matrix.cuda }}-${{ matrix.pytorch }}" >> $GITHUB_ENV
@@ -365,4 +376,4 @@ jobs:
          echo "CODECOV_TOKEN=${{ secrets.CODECOV_TOKEN }}" >> $GITHUB_ENV
      - name: Run tests job on Modal
        run: |
-          modal run cicd.cleanup
+          modal run cicd.e2e_tests
--- a/.gitignore
+++ b/.gitignore
@@ -190,6 +190,3 @@ out/

 # vim
 *.swp
-
-# scm auto-versioning
-src/axolotl/_version.py
--- a/.isort.cfg
+++ b/.isort.cfg
@@ -0,0 +1,4 @@
+[settings]
+profile=black
+known_third_party=wandb,comet_ml
+known_local_folder=src,tests
--- a/.pre-commit-config.yaml
+++ b/.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: v5.0.0
    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: v0.14.0
+-   repo: https://github.com/psf/black
+    rev: 25.1.0
    hooks:
-    -   id: ruff
-        args: [--fix]
-    -   id: ruff-format
+    -   id: black
+-   repo: https://github.com/pycqa/isort
+    rev: 6.0.1
+    hooks:
+      - id: isort
+-   repo: https://github.com/PyCQA/flake8
+    rev: 7.1.2
+    hooks:
+    - id: flake8
+-   repo: https://github.com/pylint-dev/pylint
+    rev: v3.3.6
+    hooks:
+    - id: pylint
 -   repo: https://github.com/pre-commit/mirrors-mypy
-    rev: v1.18.2
+    rev: v1.15.0
    hooks:
    - id: mypy
      additional_dependencies:
@@ -26,7 +36,7 @@ repos:
            'pydantic>=2.5.3',
        ]
 -   repo: https://github.com/PyCQA/bandit
-    rev: 1.8.6
+    rev: 1.8.3
    hooks:
    -   id: bandit
        args: [
--- a/.pylintrc
+++ b/.pylintrc
@@ -0,0 +1,15 @@
+[MASTER]
+init-hook="from pylint.config import find_default_config_files; import sys; sys.path.append(next(find_default_config_files()).parent.as_posix())"
+
+[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,
+    too-many-positional-arguments, possibly-used-before-assignment
--- a/.runpod/README.md
+++ b/.runpod/README.md
@@ -119,15 +119,14 @@ datasets:

 ## 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_processes`               | `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                |
+| Option                        | Default                    | Description                       |
+| ----------------------------- | -------------------------- | --------------------------------- |
+| `dataset_prepared_path`       | `"data/last_run_prepared"` | Path for prepared dataset         |
+| `push_dataset_to_hub`         | `""`                       | Push dataset to HF hub            |
+| `dataset_processes`           | `4`                        | Number of preprocessing processes |
+| `dataset_keep_in_memory`      | `false`                    | Keep dataset in memory            |
+| `shuffle_merged_datasets`     | `true`                     | Shuffle merged datasets           |
+| `dataset_exact_deduplication` | `true`                     | Deduplicate datasets              |

 ## LoRA Configuration

@@ -185,6 +184,7 @@ datasets:
 | `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_qkv`      | `false` | Fuse QKV operations           |
 | `flash_attn_fuse_mlp`      | `false` | Fuse MLP operations           |
 | `sdp_attention`            | `false` | Use scaled dot product        |
 | `s2_attention`             | `false` | Use shifted sparse attention  |
@@ -328,7 +328,7 @@ The following optimizers are supported:
 - 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).
+For more detailed information, please refer to the [documentation](https://axolotl-ai-cloud.github.io/axolotl/docs/config.html).

 ### Errors:

--- a/.runpod/src/config/config.yaml
+++ b/.runpod/src/config/config.yaml
@@ -97,7 +97,7 @@
 #       # 'no_input_format' cannot include {input}
 #       no_input_format: "{instruction} "

-#       # For `completion` datasets only, uses the provided field instead of `text` column
+#       # For `completion` datsets 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
@@ -242,12 +242,16 @@
 # early_stopping_patience: 3

 # # Specify a scheduler and kwargs to use with the optimizer
-# lr_scheduler: # 'one_cycle' | empty for cosine
+# lr_scheduler: # 'one_cycle' | 'log_sweep' | empty for cosine
 # lr_scheduler_kwargs:

 # # For one_cycle optim
 # lr_div_factor: # Learning rate div factor

+# # For log_sweep optim
+# log_sweep_min_lr:
+# log_sweep_max_lr:
+
 # # 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
@@ -296,6 +300,7 @@
 # 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
@@ -540,6 +545,7 @@ 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_qkv: ${FLASH_ATTN_FUSE_QKV}
 flash_attn_fuse_mlp: ${FLASH_ATTN_FUSE_MLP}
 sdp_attention: ${SDP_ATTENTION}
 s2_attention: ${S2_ATTENTION}
--- a/.runpod/src/handler.py
+++ b/.runpod/src/handler.py
@@ -57,10 +57,8 @@ async def handler(job):
    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"]
+    del os.environ["WANDB_API_KEY"]
+    del os.environ["HF_TOKEN"]


 runpod.serverless.start({"handler": handler, "return_aggregate_stream": True})
--- a/CITATION.cff
+++ b/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"
--- a/MANIFEST.in
+++ b/MANIFEST.in
@@ -2,5 +2,4 @@ include requirements.txt
 include README.md
 include LICENSE
 include src/setuptools_axolotl_dynamic_dependencies.py
-include src/axolotl/utils/chat_templates/templates/*.jinja
 recursive-include axolotl *.py
--- a/README.md
+++ b/README.md
@@ -5,9 +5,6 @@
        <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%;">
    </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">
@@ -20,69 +17,44 @@
    <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/>
    <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>

+Axolotl is a tool designed to streamline post-training for various AI models.
+Post-training refers to any modifications or additional training performed on
+pre-trained models - including full model fine-tuning, parameter-efficient tuning (like
+LoRA and QLoRA), supervised fine-tuning (SFT), instruction tuning, and alignment
+techniques. With support for multiple model architectures and training configurations,
+Axolotl makes it easy to get started with these techniques.

-## 🎉 Latest Updates
-
- 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://github.com/axolotl-ai-cloud/axolotl/tree/main/examples/gpt-oss), [Gemma 3n](https://github.com/axolotl-ai-cloud/axolotl/tree/main/examples/gemma3n), [Liquid Foundation Model 2 (LFM2)](https://github.com/axolotl-ai-cloud/axolotl/tree/main/examples/lfm2), and [Arcee Foundation Models (AFM)](https://github.com/axolotl-ai-cloud/axolotl/tree/main/examples/afm).
-  - 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://github.com/axolotl-ai-cloud/axolotl/tree/main/examples/voxtral), [Magistral 1.1](https://github.com/axolotl-ai-cloud/axolotl/tree/main/examples/magistral), and [Devstral](https://github.com/axolotl-ai-cloud/axolotl/tree/main/examples/devstral) 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!
- 2025/05: Quantization Aware Training (QAT) support has been added to Axolotl. Explore the [docs](https://docs.axolotl.ai/docs/qat.html) to learn more!
- 2025/03: 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.
-
-<details>
-
-<summary>Expand older updates</summary>
-
- 2025/06: Magistral with mistral-common tokenizer support has been added to Axolotl. See [examples](https://github.com/axolotl-ai-cloud/axolotl/tree/main/examples/magistral) to start training your own Magistral models with Axolotl!
- 2025/04: Llama 4 support has been added in Axolotl. See [examples](https://github.com/axolotl-ai-cloud/axolotl/tree/main/examples/llama-4) to start training your own Llama 4 models with Axolotl's linearized version!
- 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 designed to work with YAML config files that contain everything you need to
+preprocess a dataset, train or fine-tune a model, run model inference or evaluation,
+and much more.

 Features:

- **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, 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), 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](https://github.com/Dao-AILab/flash-attention), [Xformers](https://github.com/facebookresearch/xformers), [Flex Attention](https://pytorch.org/blog/flexattention/), [Liger Kernel](https://github.com/linkedin/Liger-Kernel), [Cut Cross Entropy](https://github.com/apple/ml-cross-entropy/tree/main), [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.
+- 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 [xformers](https://github.com/facebookresearch/xformers), 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!

-
-
-## 🚀 Quick Start - LLM Fine-tuning in Minutes
+## 🚀 Quick Start

 **Requirements**:

 - NVIDIA GPU (Ampere or newer for `bf16` and Flash Attention) or AMD GPU
 - Python 3.11
- PyTorch ≥2.7.1
-
-### Google Colab
-
-[![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)
+- PyTorch ≥2.4.1

 ### Installation

-#### Using pip
-
 ```bash
 pip3 install -U packaging==23.2 setuptools==75.8.0 wheel ninja
 pip3 install --no-build-isolation axolotl[flash-attn,deepspeed]
@@ -92,29 +64,8 @@ axolotl fetch examples
 axolotl fetch deepspeed_configs  # OPTIONAL
 ```

-#### Using Docker
-
-Installing with Docker can be less error prone than installing in your own environment.
-```bash
-docker run --gpus '"all"' --rm -it axolotlai/axolotl:main-latest
-```
-
 Other installation approaches are described [here](https://docs.axolotl.ai/docs/installation.html).

-#### Cloud Providers
-
-<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)
-
-</details>
-
 ### Your First Fine-tune

 ```bash
@@ -130,12 +81,19 @@ axolotl train examples/llama-3/lora-1b.yml

 That's it! Check out our [Getting Started Guide](https://docs.axolotl.ai/docs/getting-started.html) for a more detailed walkthrough.

+## ✨ Key Features
+
+- **Multiple Model Support**: Train various models like LLaMA, Mistral, Mixtral, Pythia, and more
+- **Training Methods**: Full fine-tuning, LoRA, QLoRA, and more
+- **Easy Configuration**: Simple YAML files to control your training setup
+- **Performance Optimizations**: Flash Attention, xformers, multi-GPU training
+- **Flexible Dataset Handling**: Use various formats and custom datasets
+- **Cloud Ready**: Run on cloud platforms or local hardware

 ## 📚 Documentation

 - [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
+- [Configuration Guide](https://docs.axolotl.ai/docs/config.html) - Full configuration options and examples
 - [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)
@@ -154,24 +112,41 @@ That's it! Check out our [Getting Started Guide](https://docs.axolotl.ai/docs/ge

 Contributions are welcome! Please see our [Contributing Guide](https://github.com/axolotl-ai-cloud/axolotl/blob/main/.github/CONTRIBUTING.md) for details.

+## Supported Models
+
+|             | 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
+
 ## ❤️ Sponsors

+Thank you to our sponsors who help make Axolotl possible:
+
+- [Modal](https://www.modal.com?utm_source=github&utm_medium=github&utm_campaign=axolotl) - Modal lets you run
+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 large language models, run protein folding simulations, and much more.
+
 Interested in sponsoring? Contact us at [wing@axolotl.ai](mailto:wing@axolotl.ai)

-## 📝 Citing Axolotl
-
-If you use Axolotl in your research or projects, please cite it as follows:
-
-```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}
-}
-```
-
 ## 📜 License

 This project is licensed under the Apache 2.0 License - see the [LICENSE](LICENSE) file for details.
--- a/TODO.md
+++ b/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
--- a/_quarto.yml
+++ b/_quarto.yml
@@ -1,6 +1,5 @@
 project:
  type: website
-  pre-render: docs/scripts/generate_config_docs.py

 quartodoc:
  dir: docs/api
@@ -18,9 +17,7 @@ quartodoc:
        - convert
        - prompt_tokenizers
        - logging_config
-        - core.builders.base
-        - core.builders.causal
-        - core.builders.rl
+        - core.trainer_builder
        - core.training_args
        - core.chat.messages
        - core.chat.format.chatml
@@ -35,53 +32,24 @@ quartodoc:
        - 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.sweeps
+        - cli.utils
        - 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:
@@ -118,7 +86,7 @@ quartodoc:
        - kernels.swiglu
        - kernels.quantize
        - kernels.utils
-    - title: Monkey Patches
+    - title: MonkeyPatches
      desc: Runtime patches for model optimizations
      contents:
        - monkeypatch.llama_attn_hijack_flash
@@ -135,16 +103,17 @@ quartodoc:
        - monkeypatch.trainer_fsdp_optim
        - monkeypatch.transformers_fa_utils
        - monkeypatch.unsloth_
+        - monkeypatch.attention.mllama
        - monkeypatch.data.batch_dataset_fetcher
        - monkeypatch.mixtral
-        - monkeypatch.gradient_checkpointing.offload_cpu
-        - monkeypatch.gradient_checkpointing.offload_disk
    - title: Utils
      desc: Utility functions
      contents:
+        - utils.models
        - utils.tokenization
        - utils.chat_templates
        - utils.lora
+        - utils.lora_embeddings
        - utils.model_shard_quant
        - utils.bench
        - utils.freeze
@@ -153,9 +122,9 @@ quartodoc:
        - utils.distributed
        - utils.dict
        - utils.optimizers.adopt
-        - utils.data.streaming
+        - utils.data.pretraining
        - utils.data.sft
-        - utils.quantization
+        - utils.gradient_checkpointing.unsloth
    - title: Schemas
      desc: Pydantic data models for Axolotl config
      contents:
@@ -205,14 +174,12 @@ quartodoc:
        - utils.callbacks.lisa
        - utils.callbacks.mlflow_
        - utils.callbacks.comet_
-        - utils.callbacks.qat
+
 website:
  title: "Axolotl"
  description: "We make fine-tuning accessible, scalable, and fun"
  favicon: favicon.jpg

-  google-analytics: "G-9KYCVJBNMQ"
-
  navbar:
    logo: image/axolotl_logo_digital_white.svg
    title: false
@@ -241,7 +208,7 @@ website:
            - docs/installation.qmd
            - docs/inference.qmd
            - docs/cli.qmd
-            - docs/config-reference.qmd
+            - docs/config.qmd
            - text: "API Reference"
              href: docs/api

@@ -265,18 +232,12 @@ website:
            - docs/lr_groups.qmd
            - docs/lora_optims.qmd
            - docs/dataset_loading.qmd
-            - docs/qat.qmd
-            - docs/quantize.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

        - section: "Advanced Features"
          contents:
@@ -285,8 +246,6 @@ website:
            - docs/torchao.qmd
            - docs/custom_integrations.qmd
            - docs/sequence_parallelism.qmd
-            - docs/gradient_checkpointing.qmd
-            - docs/nd_parallelism.qmd

        - section: "Troubleshooting"
          contents:
--- a/cicd/Dockerfile-uv.jinja
+++ b/cicd/Dockerfile-uv.jinja
@@ -1,53 +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 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 +$GITHUB_REF && \
-    git checkout FETCH_HEAD
-
-# If AXOLOTL_EXTRAS is set, append it in brackets
-RUN if [ "$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 uv pip install packaging==23.2 setuptools==75.8.0
-RUN uv pip install torchvision
-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
-
-RUN python scripts/unsloth_install.py --uv | sh
-RUN python scripts/cutcrossentropy_install.py --uv | sh
-
-# So we can test the Docker image
-RUN uv 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
--- a/cicd/Dockerfile.jinja
+++ b/cicd/Dockerfile.jinja
@@ -1,6 +1,6 @@
 FROM axolotlai/axolotl-base:{{ BASE_TAG }}

-ENV TORCH_CUDA_ARCH_LIST="7.0 7.5 8.0 8.6 8.7 8.9 9.0+PTX"
+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 }}"
@@ -9,10 +9,9 @@ ENV GITHUB_REF="{{ GITHUB_REF }}"
 ENV GITHUB_SHA="{{ GITHUB_SHA }}"
 ENV NIGHTLY_BUILD="{{ NIGHTLY_BUILD }}"
 ENV HF_HOME="{{ HF_HOME }}"
-ENV AXOLOTL_DATASET_NUM_PROC="8"

 RUN apt-get update && \
-    apt-get install -y --allow-change-held-packages vim curl nano libnccl2 libnccl-dev ibverbs-providers ibverbs-utils infiniband-diags librdmacm-dev librdmacm1 rdmacm-utils slurm-wlm
+    apt-get install -y --allow-change-held-packages vim curl nano libnccl2 libnccl-dev

 WORKDIR /workspace

--- a/cicd/cicd.sh
+++ b/cicd/cicd.sh
@@ -18,7 +18,7 @@ pytest -v --durations=10 \
  --cov-append

 # Run patched tests excluding lora kernels with coverage append
-pytest --full-trace -vvv --durations=10 \
+pytest -v --durations=10 \
  --ignore=tests/e2e/patched/lora_kernels \
  /workspace/axolotl/tests/e2e/patched \
  --cov=axolotl \
--- a/cicd/cleanup.py
+++ b/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()
--- a/cicd/cleanup.sh
+++ b/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 {} \;
--- a/cicd/e2e_tests.py
+++ b/cicd/e2e_tests.py
@@ -1,12 +1,75 @@
 """Modal app to run axolotl GPU tests"""

-from .single_gpu import GPU_CONFIG, VOLUME_CONFIG, app, cicd_image, run_cmd
+# 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", "2.4.1"),
+    "BASE_TAG": os.environ.get("BASE_TAG", "main-base-py3.11-cu121-2.4.1"),
+    "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",
+}
+
+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)
+
+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_CONFIG = modal.gpu.L40S(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=120 * 60,  # 90 min
+    timeout=60 * 60,
    cpu=8.0,
    memory=131072,
    volumes=VOLUME_CONFIG,
--- a/cicd/multigpu.py
+++ b/cicd/multigpu.py
@@ -2,6 +2,8 @@
 modal application to run axolotl gpu tests in Modal
 """

+# pylint: disable=duplicate-code
+
 import os
 import pathlib
 import tempfile
@@ -22,9 +24,9 @@ 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", "2.4.1"),
+    "BASE_TAG": os.environ.get("BASE_TAG", "main-base-py3.11-cu121-2.4.1"),
+    "CUDA": os.environ.get("CUDA", "121"),
    "GITHUB_REF": os.environ.get("GITHUB_REF", "refs/heads/main"),
    "GITHUB_SHA": os.environ.get("GITHUB_SHA", ""),
    "CODECOV_TOKEN": os.environ.get("CODECOV_TOKEN", ""),
@@ -53,7 +55,7 @@ VOLUME_CONFIG = {
 }

 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):
@@ -61,14 +63,14 @@ 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=90 * 60,
+    cpu=8.0,
    memory=131072 * N_GPUS,
    volumes=VOLUME_CONFIG,
 )
--- a/cicd/multigpu.sh
+++ b/cicd/multigpu.sh
@@ -2,7 +2,7 @@
 set -e

 # Only run two tests at a time to avoid OOM on GPU (with coverage collection)
-pytest -v --durations=10 -n2 \
+pytest -v -n2 \
  --ignore=/workspace/axolotl/tests/e2e/multigpu/solo/ \
  --ignore=/workspace/axolotl/tests/e2e/multigpu/patched/ \
  /workspace/axolotl/tests/e2e/multigpu/ \
@@ -19,7 +19,5 @@ pytest -v  --durations=10 -n1 /workspace/axolotl/tests/e2e/multigpu/patched/ \
  --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-${PYTORCH_VERSION} || true
-fi
+# Upload coverage to Codecov
+codecov upload-process -t "${CODECOV_TOKEN}" -f multigpu-coverage.xml -F multigpu,docker-tests,pytorch-${PYTORCH_VERSION} || true
--- a/cicd/single_gpu.py
+++ b/cicd/single_gpu.py
@@ -1,77 +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.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.
-    try:
-        exit_code = subprocess.call(cmd.split(), cwd=run_folder, env=sp_env)  # nosec
-        if exit_code:
-            print(f"Command '{cmd}' failed with exit code {exit_code}")
-            return exit_code
-    except Exception as e:  # pylint: disable=broad-except
-        print(f"Command '{cmd}' failed with exception {e}")
--- a/codecov.yml
+++ b/codecov.yml
@@ -12,22 +12,21 @@ coverage:
      default:
        # basic
        target: auto
-        threshold: 1%
+        threshold: 0%
        base: auto
        # advanced
        branches: null
        if_no_uploads: error
        if_not_found: success
        if_ci_failed: error
-        only_pulls: true
+        only_pulls: false
        flags: null
        paths: null
-        informational: true
    patch:
      default:
        # basic
        target: auto
-        threshold: 1%
+        threshold: 0%
        base: auto
        # advanced
        branches: null
--- a/deepspeed_configs/zero2_torch_compile.json
+++ b/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
-}
--- a/deepspeed_configs/zero3.json
+++ b/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"
--- a/deepspeed_configs/zero3_bf16.json
+++ b/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
--- a/deepspeed_configs/zero3_bf16_cpuoffload_all.json
+++ b/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
--- a/deepspeed_configs/zero3_bf16_cpuoffload_params.json
+++ b/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
--- a/devtools/dev_chat_template.yml
+++ b/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
--- a/docker/Dockerfile
+++ b/docker/Dockerfile
@@ -10,9 +10,7 @@ ARG PYTORCH_VERSION="2.1.2"
 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

@@ -25,17 +23,17 @@ RUN if [ "$AXOLOTL_EXTRAS" != "" ] ; then \
        pip install --no-build-isolation -e .[deepspeed,flash-attn,ring-flash-attn,optimizers,ray,$AXOLOTL_EXTRAS] $AXOLOTL_ARGS; \
    else \
        pip install --no-build-isolation -e .[deepspeed,flash-attn,ring-flash-attn,optimizers,ray] $AXOLOTL_ARGS; \
-    fi && \
-    python scripts/unsloth_install.py | sh && \
-    python scripts/cutcrossentropy_install.py | sh && \
-    pip install pytest && \
-    pip cache purge
+    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
--- a/docker/Dockerfile-base
+++ b/docker/Dockerfile-base
@@ -16,19 +16,12 @@ 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/* \
+    && 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}"
@@ -37,18 +30,14 @@ WORKDIR /workspace

 RUN python3 -m pip install --upgrade pip && pip3 install -U packaging==23.2 setuptools==75.8.0 wheel && \
    python3 -m pip install --no-cache-dir -U torch==${PYTORCH_VERSION}+cu${CUDA} torchvision --extra-index-url https://download.pytorch.org/whl/cu$CUDA && \
-    CAUSAL_CONV1D_FORCE_CXX11_ABI=TRUE CAUSAL_CONV1D_FORCE_BUILD=TRUE python3 -m pip install --no-cache-dir causal_conv1d==1.5.2 && \
-    python3 -m pip install --no-cache-dir "mamba_ssm @ git+https://github.com/state-spaces/mamba.git@main" && \
-    python3 -m pip cache purge
+    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

-RUN if [ "$PYTORCH_VERSION" = "2.9.0" ] && [ "$CUDA" = "128" ] ; then \
-        wget https://github.com/mjun0812/flash-attention-prebuild-wheels/releases/download/v0.4.17/flash_attn-2.8.3+cu128torch2.9-cp311-cp311-linux_x86_64.whl; \
-        pip3 install --no-cache-dir flash_attn-2.8.3+cu128torch2.9-cp311-cp311-linux_x86_64.whl; \
-        rm flash_attn-2.8.3+cu128torch2.9-cp311-cp311-linux_x86_64.whl; \
+RUN if [ "$PYTORCH_VERSION" = "2.7.0" ] ; then \
+        pip3 install flash-attn==2.7.4.post1; \
    fi
--- a/docker/Dockerfile-base-next
+++ b/docker/Dockerfile-base-next
@@ -29,7 +29,7 @@ ENV PATH="/root/miniconda3/envs/py${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$CUDA && \
+    python3 -m pip install --no-cache-dir -U torch==2.7.0 --extra-index-url https://download.pytorch.org/whl/test/cu$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"

--- a/docker/Dockerfile-base-nightly
+++ b/docker/Dockerfile-base-nightly
@@ -22,22 +22,18 @@ RUN apt-get update \
    && 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==23.2 setuptools==75.8.0 wheel && \
+RUN python3 -m pip install --upgrade pip && pip3 install packaging && \
    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
+    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
--- a/docker/Dockerfile-cloud
+++ b/docker/Dockerfile-cloud
@@ -14,10 +14,7 @@ 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 iproute2 nvtop && \
    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 && \
--- a/docker/Dockerfile-cloud-no-tmux
+++ b/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 && \
--- a/docker/Dockerfile-uv-base
+++ b/docker/Dockerfile-uv-base
@@ -1,42 +0,0 @@
-ARG CUDA_VERSION="12.6.3"
-ARG CUDNN_VERSION=""
-ARG UBUNTU_VERSION="22.04"
-ARG MAX_JOBS=4
-
-FROM nvidia/cuda:$CUDA_VERSION-cudnn$CUDNN_VERSION-devel-ubuntu$UBUNTU_VERSION AS base-builder
-
-ARG PYTHON_VERSION="3.11"
-ARG PYTORCH_VERSION="2.6.0"
-ARG CUDA="126"
-ARG TORCH_CUDA_ARCH_LIST="7.0 7.5 8.0 8.6 9.0+PTX"
-
-ENV PYTHON_VERSION=$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==${PYTORCH_VERSION} torchvision \
-    && uv pip install --no-build-isolation "causal_conv1d @ git+https://github.com/Dao-AILab/causal-conv1d.git@main" \
-    && uv pip install "mamba_ssm @ git+https://github.com/state-spaces/mamba.git@main" \
-    && uv pip install awscli pydantic
-
-RUN if [ "$PYTORCH_VERSION" = "2.9.0" ] && [ "$CUDA" = "128" ] ; then \
-        wget https://github.com/mjun0812/flash-attention-prebuild-wheels/releases/download/v0.4.17/flash_attn-2.8.3+cu128torch2.9-cp311-cp311-linux_x86_64.whl; \
-        uv pip install --no-cache-dir flash_attn-2.8.3+cu128torch2.9-cp311-cp311-linux_x86_64.whl; \
-        rm flash_attn-2.8.3+cu128torch2.9-cp311-cp311-linux_x86_64.whl; \
-    fi
--- a/docs/.gitignore
+++ b/docs/.gitignore
@@ -2,4 +2,3 @@
 _site/
 /api/*.qmd
 /api/*.html
-config-reference.qmd
--- a/docs/cli.qmd
+++ b/docs/cli.qmd
@@ -23,20 +23,6 @@ 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
@@ -94,11 +80,7 @@ axolotl train config.yml \
    --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
+axolotl train config.yml --no-accelerate

 # Resume training from checkpoint
 axolotl train config.yml --resume-from-checkpoint path/to/checkpoint
@@ -193,9 +175,6 @@ 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
@@ -230,16 +209,6 @@ 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

@@ -308,6 +277,9 @@ axolotl preprocess config.yml --cloud cloud_config.yml
 # Train on cloud
 axolotl train config.yml --cloud cloud_config.yml

+# Train without accelerate on cloud
+axolotl train config.yml --cloud cloud_config.yml --no-accelerate
+
 # Run lm-eval on cloud
 axolotl lm-eval config.yml --cloud cloud_config.yml
 ```
--- a/docs/config.qmd
+++ b/docs/config.qmd
@@ -0,0 +1,743 @@
+---
+title: Config Reference
+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:
+# Optional[bool] Whether to shrink the embeddings to len(tokenizer). By default, we won't shrink.
+shrink_embeddings:
+# Optional[bool] Don't upcast the embeddings to float32 when using PEFT. Useful for low-VRAM GPUs
+embeddings_skip_upcast:
+# Whether to load the model with randomly initialized weights. Useful for
+# pre-training a model from scratch or debugging purposes.
+random_init_weights:
+
+# (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 the base model loading from_pretrained
+overrides_of_model_kwargs:
+  # use_cache: False
+
+# 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`, or 'auto' for automatic detection. require >=ampere
+# Use CUDA fp16
+fp16: true
+# Use CUDA tf32
+tf32: true # require >=ampere
+# Note: if bf16 is set to 'auto', and fp16 is set to true, we will prefer the explict fp16 setting
+
+# 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
+
+# List[str]. Add plugins to extend the pipeline.
+# See `src/axolotl/integrations` for the available plugins or doc below for more details.
+# https://docs.axolotl.ai/docs/custom_integrations.html
+plugins:
+  # - axolotl.integrations.cut_cross_entropy.CutCrossEntropyPlugin
+
+# 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] split dataset into N pieces (use with shards_idx)
+    shards_idx: # Optional[int] = 0 the index of sharded dataset to use
+
+    preprocess_shards: # Optional[int] process dataset in N sequential chunks for memory efficiency (exclusive with `shards`)
+
+    name: # Optional[str] name of dataset configuration to load
+    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 chat template. Used only if `chat_template: jinja` or empty.
+    chat_template_jinja:
+
+    # Key containing the messages (default: "messages")
+    field_messages: messages
+
+    # Key containing the system message (default: "system")
+    # If the system message is not present in the dataset sample, it will be loaded from the field_system property.
+    field_system: system
+
+    # Mapping of properties from the input dataset to the chat template.
+    # (default: message_property_mappings={'role':'role', 'content':'content'})
+    # If a property exists in the template but not in this mapping, the system will attempt
+    # to load it directly from the message using the property name as the key.
+    # Example: In the mapping below, 'from' is loaded from input dataset and used as 'role',
+    # while 'value' is loaded and used as 'content' in the chat template.
+    message_property_mappings:
+      role: from
+      content: value
+      # ...
+
+    # Optional[Dict[str, List]]. Roles mapping in the messages.
+    # The format is {target_role: [source_roles]}. All source roles will be mapped to the target role.
+    # The default is:
+    roles:
+      user: ["human", "user"]
+      assistant: ["gpt", "assistant"]
+      system: ["system"]
+      tool: ["tool"]
+
+    # Optional[bool]. Whether to drop the system turn from the dataset. Only works with chat_template.
+    # This does not drop the default system message from chat_template if it exists. If you wish to,
+    # we recommend using a custom jinja template with the default system message removed or
+    # adding a system turn with empty content.
+    drop_system_message:
+
+    # Optional[bool]. (for Qwen3 template only) Whether to split the assistant content based on a reasoning trace inside delimited tags
+    # See example at `docs/dataset-formats/conversation.qmd`
+    split_thinking:
+
+    # IMPORTANT: The following fields determine which parts of the conversation to train on.
+    # Priority order: message_field_training > message_field_training_detail > train_on_inputs or role in roles_to_train
+    # See examples at `docs/dataset-formats/conversation.qmd`
+    # Note: If the below 5 fields are empty, defaults to training only on the last message.
+
+    # Optional[List[str]]. Roles to train on. The tokens from these roles will be considered for the loss.
+    roles_to_train: ["assistant"]  # default
+    # Optional[str]. Which EOS tokens to train on in the conversation. Possible values are:
+    # - all: train on all EOS tokens
+    # - turn (default): train on the EOS token at the end of each trainable turn
+    # - last: train on the last EOS token in the conversation
+    # TIP: Please make sure that your `tokenizer.eos_token` is same as EOS/EOT token in template. Otherwise, set `eos_token` under `special_tokens`.
+    train_on_eos: turn
+    # Optional[str]. Which EOT (End-of-Turn) tokens to train on in the conversation. Possible values are:
+    # - all: train on all EOT tokens
+    # - turn: train on the EOT token at the end of each trainable turn
+    # - last: train on the last EOT token in the conversation
+    # If not specified, defaults to the value of train_on_eos for backward compatibility.
+    train_on_eot:
+    # 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).
+    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', 'simpo', 'orpo', 'grpo'
+rl:
+rl_beta:  # Optional[float]. The beta parameter for the RL training.
+
+# dpo
+dpo_use_weighting:  # Optional[bool]. Whether to perform weighting.
+rpo_alpha: # Optional[float]. Weighting of NLL term in loss from RPO paper.
+
+# orpo
+orpo_alpha: 0.1  # Parameter controlling the relative ratio loss weight in the ORPO loss. Passed to `beta` in `ORPOConfig` due to trl mapping.
+
+# kto
+kto_desirable_weight: # Optional[float]. Factor for desirable loss term in KTO loss.
+kto_undesirable_weight: # Optional[float]. Factor for undesirable loss term in KTO loss.
+
+# simpo
+cpo_alpha: 1.0  # Weight of the BC regularizer
+simpo_gamma: 0.5  # Target reward margin for the SimPO loss
+
+# grpo
+trl:
+  use_vllm: # Optional[bool]. Whether to use VLLM for RL training.
+  vllm_server_host: # Optional[str]. Host of the vLLM server to connect to.
+  vllm_server_port: # Optional[int]. Port of the vLLM server to connect to.
+  vllm_server_timeout: # Optional[int]. Total timeout (in seconds) to wait for the vLLM server to respond.
+  vllm_guided_decoding_regex: # Optional[str]. Regex for vLLM guided decoding.
+
+  beta: # Optional[float]. Beta parameter for the RL training. Same as `rl_beta`. Use
+  max_completion_length: # Optional[int]. Maximum length of the completion for RL training.
+
+  reward_funcs: # Optional[list[str]]. List of reward functions to load. Paths must be importable from current dir.
+  reward_weights: # Optional[list[float]]. List of reward weights for the reward functions.
+
+  num_generations: # Optional[int]. Number of generations to sample.
+  log_completions: # Optional[bool]. Whether to log completions.
+
+  sync_ref_model: # Optional[bool]. Whether to sync the reference model.
+  ref_model_mixup_alpha: # Optional[float]. Mixup alpha for the reference model.
+  ref_model_sync_steps: # Optional[int]. Sync steps for the reference model.
+
+
+# reward modelling: `True` or `False`
+reward_model:
+
+# process reward modelling: `True` or `False`
+process_reward_model:
+
+# 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
+# Optional[List[str]]. Custom EOT (End-of-Turn) tokens to mask/unmask during training.
+# These tokens mark the boundaries between conversation turns.
+# For example: ["/INST", "</s>", "[/SYSTEM_PROMPT]"]
+# If not specified, defaults to just the model's eos_token.
+# This is useful for templates that use multiple delimiter tokens.
+eot_tokens:
+  # - "</s>"
+  # - "[/INST]"
+  # - "[/SYSTEM_PROMPT]"
+# 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: # Optional[str] repo_org/repo_name
+# 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
+sample_pack_sequentially: # Optional[bool]. Whether to pack samples sequentially.
+
+# whether to concatenate samples during pretraining
+pretraining_sample_concatenation:
+
+curriculum_sampling: # Optional[bool]. Whether to use sequential sampling for curriculum learning
+
+# Use batch flattening for speedups when not using sample_packing
+batch_flattening:
+
+# 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
+
+# List[int] | int. # The layer indices to transform, otherwise, apply to all layers
+# https://huggingface.co/docs/peft/v0.15.0/en/package_reference/lora#peft.LoraConfig.layers_to_transform
+peft_layers_to_transform:
+
+# Optional[bool]. Whether to use DoRA.
+# https://huggingface.co/docs/peft/v0.15.0/en/developer_guides/lora#weight-decomposed-low-rank-adaptation-dora
+peft_use_dora:
+
+# Optional[bool]. Whether to use RSLoRA.
+# https://huggingface.co/docs/peft/v0.15.0/en/developer_guides/lora#rank-stabilized-lora
+peft_use_rslora:
+
+# Optional[list[tuple[int, int]]]. List of layer indices to replicate.
+# https://huggingface.co/docs/peft/v0.15.0/en/developer_guides/lora#memory-efficient-layer-replication-with-lora
+peft_layer_replication:
+
+# bool | Literal["gaussian", "eva", "olora", "pissa", "pissa_niter_[number of iters]", "corda", "loftq"]
+# How to initialize LoRA weights. Default to True which is MS original implementation.
+# https://huggingface.co/docs/peft/v0.15.0/en/developer_guides/lora#initialization
+peft_init_lora_weights:
+
+# 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
+
+# Apply custom LoRA autograd functions and activation function Triton kernels for
+# speed and memory savings
+# See: https://docs.axolotl.ai/docs/lora_optims.html
+lora_mlp_kernel: true
+lora_qkv_kernel: true
+lora_o_kernel: true
+
+# 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.
+
+# Tensorboard
+use_tensorboard: # Optional[bool]
+
+# Where to save the full-finetuned model to
+output_dir: ./completed-model
+
+# Whether to use torch.compile and which backend to use
+# setting to `auto` will enable torch compile when torch>=2.5.1
+torch_compile:  # Optional[Union[Literal["auto"], 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, integer for every N steps. float for fraction of total steps
+evals_per_epoch: # number of times per epoch to run evals, mutually exclusive with eval_steps
+eval_strategy: # Set to `"no"` to skip evaluation, `"epoch"` at end of each epoch, leave empty to infer from `eval_steps`.
+save_strategy: # Set to `"no"` to skip checkpoint saves, `"epoch"` at end of each epoch, `"best"` when better result is achieved, leave empty to infer from `save_steps`.
+save_steps: # Leave empty to save at each epoch, integer for every N steps. float for fraction of total steps
+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:
+
+# bool of whether to include tokens trainer per second in the training metrics. This iterates over the entire dataset once, so it takes some time.
+include_tokens_per_second: # Optional[bool]
+
+# whether to find batch size that fits in memory. Passed to underlying transformers Trainer
+auto_find_batch_size: # Optional[bool]
+
+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
+do_causal_lm_eval: # Whether to run causal language model evaluation for metrics in `eval_causal_lm_metrics`.
+eval_causal_lm_metrics: # HF evaluate metrics used during evaluation. Default is ["sacrebleu", "comet", "ter", "chrf", "perplexity"]
+
+profiler_steps: # enable the pytorch profiler to capture the first N steps of training to the output_dir.
+                # see https://pytorch.org/blog/understanding-gpu-memory-1/ for more information
+                # snapshots can be visualized @ https://pytorch.org/memory_viz
+
+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. Available options are: true, false, "offload".
+# 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' | 'rex' | 'log_sweep' | 'linear' | 'cosine_with_restarts' | 'polynomial' | 'constant' | 'constant_with_warmup' | 'inverse_sqrt' | 'reduce_lr_on_plateau' | 'cosine_with_min_lr' | 'warmup_stable_decay' | 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/cbf924b76c03828101a34069a96d209314114fd5/src/transformers/training_args.py#L144-L189
+#
+# 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_torch
+# - adamw_torch_fused
+# - adamw_torch_xla
+# - adamw_torch_npu_fused
+# - adamw_apex_fused
+# - adopt_adamw  (an EXPERIMENTAL optimizer, only for torch version >= 2.5.1)
+# - adafactor
+# - adamw_anyprecision
+# - adamw_torch_4bit
+# - ademamix
+# - sgd
+# - adagrad
+# - adamw_bnb_8bit
+# - adamw_8bit   # alias for adamw_bnb_8bit
+# - ademamix_8bit
+# - lion_8bit
+# - lion_32bit
+# - paged_adamw_32bit
+# - paged_adamw_8bit
+# - paged_ademamix_32bit
+# - paged_ademamix_8bit
+# - paged_lion_32bit
+# - paged_lion_8bit
+# - rmsprop
+# - rmsprop_bnb
+# - rmsprop_bnb_8bit
+# - rmsprop_bnb_32bit
+# - galore_adamw
+# - galore_adamw_8bit
+# - galore_adafactor
+# - galore_adamw_layerwise
+# - galore_adamw_8bit_layerwise
+# - galore_adafactor_layerwise
+# - lomo
+# - adalomo
+# - grokadamw
+# - schedule_free_adamw
+# - schedule_free_sgd
+# - apollo_adamw
+# - apollo_adamw_layerwise
+#
+# Additional custom optimizers include:
+# - optimi_adamw
+# - ao_adamw_8bit
+# - ao_adamw_fp8
+# - came_pytorch
+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:
+
+# Optional[bool]. Whether to bettertransformers
+flash_optimum:
+
+# Note: Only one of the following attention patches can be used at a time.
+# For example, if you set `xformers_attention` to `true`, do not set `flash_attention` to `true`.
+
+# Optional[bool]. Whether to use xformers attention patch https://github.com/facebookresearch/xformers:
+xformers_attention:
+# Optional[bool]. Whether to use flash attention patch https://github.com/Dao-AILab/flash-attention:
+flash_attention:
+flash_attn_cross_entropy:  # Optional[bool]. Whether to use flash-attention cross entropy implementation - advanced use only
+flash_attn_rms_norm:  # Optional[bool]. Whether to use flash-attention rms norm implementation - advanced use only
+flash_attn_fuse_qkv: # Optional[bool]. Whether to fuse QKV into a single operation
+flash_attn_fuse_mlp: # Optional[bool]. Whether to fuse part of the MLP into a single operation
+# Optional[bool]. Whether to use scaled-dot-product attention
+# https://pytorch.org/docs/stable/generated/torch.nn.functional.scaled_dot_product_attention.html
+sdp_attention:
+# Optional[bool]. Shifted-sparse attention (only llama) - https://arxiv.org/pdf/2309.12307.pdf
+s2_attention:
+
+# Optional[bool]. Whether to use low_cpu_mem_usage
+low_cpu_mem_usage:
+# Optional[str]. Resume from a specific checkpoint dir
+resume_from_checkpoint:
+# Optional[bool]. 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
+
+## Multimodal section
+# int | tuple[int, int] | None . Size to resize images to, width x height.
+# Will read from model/processor config if not set.
+image_size:
+# str. Algorithm to use for image resizing. "bilinear", "bicubic", "lanczos". Default is "bilinear".
+image_resize_algorithm: 'bilinear'
+## End of multimodal section
+
+# 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]"
+
+# Optional[list[str]]. Add extra tokens to the tokenizer.
+tokens:
+  # - "<|startoftext|>"
+  # - "<|endoftext|>"
+
+# Mapping token_id to new_token_string to override reserved added_tokens in the tokenizer.
+# Only works for tokens that are not part of the base vocab (aka are added_tokens).
+# Can be checked if they exist in tokenizer.json added_tokens.
+added_tokens_overrides:  # Dict[int, str]
+#  128041: "<|im_start|>"
+#  128042: "<|im_end|>"
+
+# 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:
+
+# Sequence parallelism
+# Set to a divisor of the number of GPUs available to split sequences into chunks of equal size.
+# Use in long context training to prevent OOM when sequences cannot fit into a single GPU's VRAM.
+# E.g., if 4 GPUs are available, set this value to 2 to split each sequence into two equal-sized
+# subsequences, or set to 4 to split into four equal-sized subsequences.
+# See https://docs.axolotl.ai/docs/sequence_parallelism.html for more details.
+sequence_parallel_degree:
+# Optional; strides across the key dimension. Larger values use more memory but should make training faster.
+# Must evenly divide the number of KV heads in your model.
+heads_k_stride: 1
+# One of "varlen_llama3", "batch_ring", "batch_zigzag", "batch_stripe". Defaults to "varlen_llama3"
+# in the sample packing case, and "batch_ring" in the non-sample packing case.
+ring_attn_func:
+
+# 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:
+```
--- a/docs/custom_integrations.qmd
+++ b/docs/custom_integrations.qmd
@@ -7,7 +7,6 @@ toc-depth: 3
 ```{python}
 #| echo: false

-import os
 import re

 def process_readme(integration_name):
@@ -54,24 +53,6 @@ sections = [
    ("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))
 ```
--- a/docs/dataset-formats/conversation.qmd
+++ b/docs/dataset-formats/conversation.qmd
@@ -9,10 +9,10 @@ order: 3
 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 [configs](../config.qmd) for full configs and supported templates.

 ### Migrating from sharegpt

@@ -52,9 +52,7 @@ 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. (Legacy) Using the default chat template in the tokenizer_config.json on OpenAI messages format, training on only last message.

 ```yaml
 datasets:
@@ -68,9 +66,7 @@ datasets:
 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
@@ -80,13 +76,7 @@ datasets:
    roles_to_train: ["assistant"]  # default value
 ```

-::: {.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
@@ -95,9 +85,7 @@ datasets:
    type: chat_template
 ```

-#### 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
@@ -112,9 +100,7 @@ datasets:
 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.
+5. 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:
@@ -130,16 +116,16 @@ datasets:
 ```

 ::: {.callout-tip}
-See [config documentation](../config-reference.qmd) for detailed explanations of "turn", "last", and "all" options for training on tokens.
+See [config documentation](../config.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.
+You can add those tokens as new tokens under `tokens: ` or (recommended) override unused added_tokens via `added_tokens_overrides: `. See [config](../config.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`.
+6. 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:
@@ -159,84 +145,7 @@ If EOS token only appears at the end of a prompt, `train_on_eos: last` is equiva
 :::


-#### 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": "{\"...\": \"...\"}"
-```
-:::
-
-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
+7. (Advanced) Using fine-grained control over tokens and turns to train in a conversation

 For a data sample that looks like:

@@ -287,9 +196,7 @@ datasets:
 It is not necessary to set both `message_field_training` and `message_field_training_detail` at once.
 :::

-#### 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.
+8. (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:
--- a/docs/dataset-formats/index.qmd
+++ b/docs/dataset-formats/index.qmd
@@ -36,6 +36,10 @@ It is typically recommended to save your dataset as `.jsonl` due to its flexibil

 Axolotl supports loading from a Hugging Face hub repo or from local files.

+::: {.callout-important}
+For pre-training only, Axolotl would split texts if it exceeds the context length into multiple smaller prompts.
+:::
+
 ### Pre-training from Hugging Face hub datasets

 As an example, to train using a Hugging Face dataset `hf_org/name`, you can pass the following config:
@@ -61,7 +65,7 @@ While we recommend `.jsonl`, you can also use the other formats (`csv`, `parquet

 ### Pre-training without streaming

-In the case that the dataset is small and can be loaded entirely into memory, another approach to running pre-training is to use the `completion` format. This would mean that the entire dataset is pre-tokenized instead of on-demand in streaming.
+On the rare case that the dataset is small and can be loaded entirely into memory, another approach to running pre-training is to use the `completion` format. This would mean that the entire dataset is pre-tokenized instead of on-demand in streaming.

 One benefit of this is that the tokenization can be performed separately on a CPU-only machine, and then transferred to a GPU machine for training to save costs.

@@ -73,21 +77,18 @@ datasets:
    type: completion
 ```

-From local files:
+From local files (either example works):

 ```yaml
 datasets:
  - path: A.jsonl
    type: completion

-  - path: B.jsonl
+  - path: json
+    data_files: ["A.jsonl", "B.jsonl", "C.jsonl"]
    type: completion
 ```

-::: {.callout-important}
-For `completion` only, Axolotl would split texts if it exceeds the context length into multiple smaller prompts. If you are interested in having this for `pretraining_dataset` too, please let us know or help make a PR!
-:::
-
 ### Pre-training dataset configuration tips

 #### Setting max_steps
--- a/docs/dataset-formats/inst_tune.qmd
+++ b/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).
--- a/docs/dataset_loading.qmd
+++ b/docs/dataset_loading.qmd
@@ -36,7 +36,7 @@ This matches the API of [`datasets.load_dataset`](https://github.com/huggingface

 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).
+For full details on the config, see [config.qmd](config.qmd).

 ::: {.callout-note}

@@ -54,7 +54,7 @@ datasets:

 #### Files

-To load a JSON file, you would do something like this:
+Usually, to load a JSON file, you would do something like this:

 ```python
 from datasets import load_dataset
@@ -66,11 +66,19 @@ Which translates to the following config:

 ```yaml
 datasets:
-  - path: data.json
-    ds_type: json
+  - path: json
+    data_files: /path/to/your/file.jsonl
 ```

-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.
+However, to make things easier, we have added a few shortcuts for loading local dataset files.
+
+You can just point the `path` to the file or directory along with the `ds_type` to load the dataset. The below example shows for a JSON file:
+
+```yaml
+datasets:
+  - path: /path/to/your/file.jsonl
+    ds_type: json
+```

 This works for CSV, JSON, Parquet, and Arrow files.

--- a/docs/debugging.qmd
+++ b/docs/debugging.qmd
@@ -29,7 +29,7 @@ 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
@@ -101,7 +101,7 @@ For example, to mimic the command `cd devtools && CUDA_VISIBLE_DEVICES=0 acceler
                "-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
--- a/docs/docker.qmd
+++ b/docs/docker.qmd
@@ -8,10 +8,6 @@ format:

 This section describes the different Docker images that are released by AxolotlAI at [Docker Hub](https://hub.docker.com/u/axolotlai).

-::: {.callout-important}
-For Blackwell GPUs, please use the tags with PyTorch 2.7.1 and CUDA 12.8.
-:::
-
 ## Base

 The base image is the most minimal image that can install Axolotl. It is based on the `nvidia/cuda` image. It includes python, torch, git, git-lfs, awscli, pydantic, and more.
@@ -32,11 +28,11 @@ main-base-py{python_version}-cu{cuda_version}-{pytorch_version}

 Tags examples:

- `main-base-py3.11-cu128-2.7.1`
- `main-base-py3.11-cu126-2.7.1`
+- `main-base-py3.11-cu128-2.7.0`
 - `main-base-py3.11-cu126-2.7.0`
- `main-base-py3.11-cu126-2.6.0`
 - `main-base-py3.11-cu124-2.6.0`
+- `main-base-py3.11-cu124-2.5.1`
+- `main-base-py3.11-cu124-2.4.1`

 ## Main

@@ -74,15 +70,15 @@ There may be some extra tags appended to the image, like `-vllm` which installs

 Tags examples:

- `main-py3.11-cu128-2.7.1`
- `main-py3.11-cu126-2.7.1`
 - `main-py3.11-cu126-2.7.0`
- `main-py3.11-cu126-2.6.0`
 - `main-py3.11-cu124-2.6.0`
+- `main-py3.11-cu124-2.5.1`
+- `main-py3.11-cu124-2.4.1`
 - `main-latest`
 - `main-20250303-py3.11-cu124-2.6.0`
- `main-20250303-py3.11-cu126-2.6.0`
- `0.10.1`
+- `main-20250303-py3.11-cu124-2.5.1`
+- `main-20250303-py3.11-cu124-2.4.1`
+- `0.7.1`

 ## Cloud

--- a/docs/faq.qmd
+++ b/docs/faq.qmd
@@ -9,11 +9,11 @@ description: Frequently asked questions

 > A: Usually an issue with the GPUs communicating with each other. See the [NCCL doc](nccl.qmd)

-**Q: exitcode: -9**
+**Q: Exitcode -9**

 > A: This usually happens when you run out of system RAM.

-**Q: exitcode: -7 while using deepspeed**
+**Q: Exitcode -7 while using deepspeed**

 > A: Try upgrading deepspeed w: `pip install -U deepspeed`

@@ -51,26 +51,6 @@ description: Frequently asked questions
 >   pad_token: "..."
 > ```

-**Q: `IterableDataset error` or `KeyError: 'input_ids'` when using `preprocess` CLI**
-
-> A: This is because you may be using `preprocess` CLI with `pretraining_dataset:` or `skip_prepare_dataset: true` respectively. Please use `axolotl train` CLI directly instead as these datasets are prepared on demand.
-
-**Q: vLLM is not working with Axolotl**
-
-> A: We currently recommend torch 2.6.0 for use with `vllm`. Please ensure you use the right version. For Docker, please use the `main-py3.11-cu124-2.6.0` tag.
-
-**Q: FA2 2.8.0 `undefined symbol` runtime error on CUDA 12.4**
-
-> A: There seems to be a wheel issue with FA2 2.8.0 on CUDA 12.4. Try CUDA 12.6 instead or downgrade to FA2 2.7.4. Please refer to the upstream issue: https://github.com/Dao-AILab/flash-attention/issues/1717.
-
-**Q: Can we mix text and text+image datasets for VLM training?**
-
-> A: Yes, you can for newer VLM arch. The ones that would not work are LLaVA / Pixtral arch. If you notice one not working, please let us know!
-
-**Q: Why is `memory/max_*` different from `nvidia-smi`?**
-
-> A: We use `torch` APIs to retrieve this information. You can see https://docs.pytorch.org/docs/stable/notes/cuda.html#cuda-memory-management for more information.
-
 ### Chat templates

 **Q: `jinja2.exceptions.UndefinedError: 'dict object' has no attribute 'content' / 'role' / ____`**
@@ -130,25 +110,3 @@ description: Frequently asked questions
 > A: If `eot_tokens: ` is not provided, the default behavior is the same as before. EOS tokens used to delimit turns are masked/unmasked depending on whether the turn is trainable.

 > Internally, `eot_tokens: tokenizer.eos_token` and `train_on_eot: train_on_eos` (which defaults to `turn`). This transition helps clarify the naming and behavior of EOT/EOS tokens.
-
-**Q: `Data processing error: CAS service error`**
-
-> A: Try disabling XET with `export HF_HUB_DISABLE_XET=1`
-
-**Q: `torch._inductor.exc.LoweringException: NoValidChoicesError: No choices to select, please consider adding ATEN into max_autotune_gemm_backends config (defined in torch/_inductor/config.py) to allow at least one choice. `**
-
-> A: Depending on the version of torch, you may need to include this in your YAML:
-
-> ```yaml
-> flex_attn_compile_kwargs:
->   dynamic: false
->   mode: max-autotune-no-cudagraphs
-> ```
-
-**Q: `ValueError("Backward pass should have cleared tracker of all tensors")`
-
-> A: This may happen due to edge cases in using the modern OffloadActivations context manager for CUDA streams. If you encounter this error, you may have success using the naive implementation with `offload_activations: legacy` in your YAML.
-
-**Q: `Error parsing tool_calls arguments as JSON.`
-
-> A: There is an error parsing string arguments to a dict. Please check your dataset and the error message for more details.
--- a/docs/fsdp_qlora.qmd
+++ b/docs/fsdp_qlora.qmd
@@ -1,5 +1,5 @@
 ---
-title: "FSDP + QLoRA"
+title: "FDSP + QLoRA"
 description: Use FSDP with QLoRA to fine-tune large LLMs on consumer GPUs.
 format:
  html:
@@ -20,15 +20,9 @@ To enable `QLoRA` with `FSDP`, you need to perform the following steps:
 > See the [example config](#example-config) file in addition to reading these instructions.

 1. Set `adapter: qlora` in your axolotl config file.
-2. Enable FSDP in your axolotl config, as [described here](multi-gpu.qmd#sec-fsdp).
+2. Enable FSDP in your axolotl config, as [described here](https://github.com/axolotl-ai-cloud/axolotl?tab=readme-ov-file#fsdp).
 3. Use one of the supported model types: `llama`, `mistral` or `mixtral`.

-## Enabling Swap for FSDP2
-
-If available memory is insufficient even after FSDP's CPU offloading, you can enable swap memory usage by setting `cpu_offload_pin_memory: false` alongside `offload_params: true` in FSDP config.
-
-This disables memory pinning, allowing FSDP to use disk swap space as fallback. Disabling memory pinning itself incurs performance overhead, and actually having to use swap adds more, but it may enable training larger models that would otherwise cause OOM errors on resource constrained systems.
-
 ## Example Config

 [examples/llama-2/qlora-fsdp.yml](../examples/llama-2/qlora-fsdp.yml) contains an example of how to enable QLoRA + FSDP in axolotl.
--- a/docs/getting-started.qmd
+++ b/docs/getting-started.qmd
@@ -55,7 +55,7 @@ output_dir: ./outputs/lora-out
 - To perform QLoRA finetuning, replace with `load_in_4bit: true` and `adapter: qlora`.
 :::

-See our [config options](config-reference.qmd) for more details.
+See our [Config options](config.qmd) for more details.

 ### Training {#sec-training}

@@ -104,7 +104,7 @@ the `alpaca` dataset format, which has the following format:
 Please see our [Dataset Formats](dataset-formats) for more dataset formats and how to
 format them.

-2. Prepare your JSONL data in the specified format (in this case, the expected `alpaca`
+2. Prepare your JSONL data in the specified format (in this case, the expected `alpaca
 format):

 ```json
@@ -120,12 +120,6 @@ axolotl train my_training.yml

 ## Common Tasks {#sec-common-tasks}

-::: {.callout-tip}
-
-The same yaml file is used for training, inference, and merging.
-
-:::
-
 ### Testing Your Model {#sec-testing}

 After training, test your model:
@@ -134,16 +128,6 @@ After training, test your model:
 axolotl inference my_training.yml --lora-model-dir="./outputs/lora-out"
 ```

-More details can be found in [Inference](inference.qmd).
-
-### Using a UI {#sec-ui}
-
-Launch a Gradio interface:
-
-```bash
-axolotl inference my_training.yml --lora-model-dir="./outputs/lora-out" --gradio
-```
-
 ### Preprocessing Data {#sec-preprocessing}

 For large datasets, preprocess first:
@@ -152,22 +136,14 @@ For large datasets, preprocess first:
 axolotl preprocess my_training.yml
 ```

-Please make sure to set `dataset_prepared_path: ` in your config to set the path to save the prepared dataset.
+### Using a UI {#sec-ui}

-More details can be found in [Dataset Preprocessing](dataset_preprocessing.qmd).
-
-### Merging LoRA weights {#sec-merging-lora}
-
-To merge the LoRA weights back into the base model, run:
+Launch a Gradio interface:

 ```bash
-axolotl merge-lora my_training.yml --lora-model-dir="./outputs/lora-out"
+axolotl inference my_training.yml --lora-model-dir="./outputs/lora-out" --gradio
 ```

-The merged model will be saved in the `{output_dir}/merged` directory.
-
-More details can be found in [Merging LoRA weights](inference.qmd#sec-merging).
-
 ## Next Steps {#sec-next-steps}

 Now that you have the basics, you might want to:
@@ -179,8 +155,7 @@ Now that you have the basics, you might want to:

 Check our other guides for details on these topics:

- [Configuration Guide](config-reference.qmd) - Full configuration options
- [Dataset Loading](dataset_loading.qmd) - Loading datasets from various sources
+- [Configuration Guide](config.qmd) - Full configuration options
 - [Dataset Formats](dataset-formats) - Working with different data formats
 - [Multi-GPU Training](multi-gpu.qmd)
 - [Multi-Node Training](multi-node.qmd)
--- a/docs/gradient_checkpointing.qmd
+++ b/docs/gradient_checkpointing.qmd
@@ -1,29 +0,0 @@
---
-title: Gradient Checkpointing and Activation Offloading
---
-
-Gradient checkpointing and activation offloading are techniques used to optimize the performance of deep learning
-models by reducing the memory footprint and improving computational efficiency.
-
-### Enabling Gradient Checkpointing
-
-```yaml
-gradient_checkpointing: true
-```
-
-### Enabling Activation Offloading
-
-```yaml
-gradient_checkpointing: true  # required for activation offloading
-activation_offloading: true
-```
-
-Activation offloading variants:
-
-The default `activation_offloading: true` offloads activations to CPU and uses CUDA streams
-to overlap the communications and computations when offloading.
-
-The `activation_offloading: legacy` naively offloads activations to CPU and without additional optimizations.
-
-For resource constrained environments with limited CPU memory, `activation_offloading: disk` offloads
-activations to disk instead of CPU RAM so that much larger context lengths can be trained with minimal memory.
--- a/docs/installation.qmd
+++ b/docs/installation.qmd
@@ -14,8 +14,8 @@ This guide covers all the ways you can install and set up Axolotl for your envir
 ## Requirements {#sec-requirements}

 - NVIDIA GPU (Ampere architecture or newer for `bf16` and Flash Attention) or AMD GPU
- Python ≥3.11
- PyTorch ≥2.6.0
+- Python ≥3.10
+- PyTorch ≥2.4.1

 ## Installation Methods {#sec-installation-methods}

@@ -25,10 +25,6 @@ Please make sure to have Pytorch installed before installing Axolotl in your loc
 Follow the instructions at: [https://pytorch.org/get-started/locally/](https://pytorch.org/get-started/locally/)
 :::

-::: {.callout-important}
-For Blackwell GPUs, please use Pytorch 2.7.0 and CUDA 12.8.
-:::
-
 ### PyPI Installation (Recommended) {#sec-pypi}

 ```{.bash}
@@ -41,40 +37,6 @@ installed) in order not to clobber it, and so that we set the correct version of
 dependencies that are specific to the PyTorch version or other installed
 co-dependencies.

-### uv Installation {#sec-uv}
-
-uv is a fast, reliable Python package installer and resolver built in Rust. It offers significant performance improvements over pip and provides better dependency resolution, making it an excellent choice for complex environments.
-
-Install uv if not already installed
-```{.bash}
-curl -LsSf https://astral.sh/uv/install.sh | sh
-source $HOME/.local/bin/env
-```
-
-Choose your CUDA version to use with PyTorch; e.g. `cu124`, `cu126`, `cu128`,
-then create the venv and activate
-```{.bash}
-export UV_TORCH_BACKEND=cu126
-uv venv --no-project --relocatable
-source .venv/bin/activate
-```
-
-Install PyTorch
- PyTorch 2.6.0 recommended
-```{.bash}
-uv pip install packaging setuptools wheel
-uv pip install torch==2.6.0
-uv pip install awscli pydantic
-```
-
-Install axolotl from PyPi
-```{.bash}
-uv pip install --no-build-isolation axolotl[deepspeed,flash-attn]
-
-# optionally install with vLLM if you're using torch==2.6.0 and want to train w/ GRPO
-uv pip install --no-build-isolation axolotl[deepspeed,flash-attn,vllm]
-```
-
 ### Edge/Development Build {#sec-edge-build}

 For the latest features between releases:
@@ -110,10 +72,6 @@ docker run --privileged --gpus '"all"' --shm-size 10g --rm -it \
 ```
 :::

-::: {.callout-important}
-For Blackwell GPUs, please use `axolotlai/axolotl:main-py3.11-cu128-2.7.0` or the cloud variant `axolotlai/axolotl-cloud:main-py3.11-cu128-2.7.0`.
-:::
-
 Please refer to the [Docker documentation](docker.qmd) for more information on the different Docker images that are available.

 ## Cloud Environments {#sec-cloud}
@@ -124,17 +82,14 @@ For providers supporting Docker:

 - Use `axolotlai/axolotl-cloud:main-latest`
 - Available on:
-    - [RunPod](https://runpod.io/gsc?template=v2ickqhz9s&ref=6i7fkpdz)
-    - [Vast.ai](https://cloud.vast.ai?ref_id=62897&template_id=bdd4a49fa8bce926defc99471864cace&utm_source=axolotl&utm_medium=partner&utm_campaign=template_launch_july2025&utm_content=docs_link)
-    - [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)
+  - [Latitude.sh](https://latitude.sh/blueprint/989e0e79-3bf6-41ea-a46b-1f246e309d5c)
+  - [JarvisLabs.ai](https://jarvislabs.ai/templates/axolotl)
+  - [RunPod](https://runpod.io/gsc?template=v2ickqhz9s&ref=6i7fkpdz)
+  - [Novita](https://novita.ai/gpus-console?templateId=311)

 ### Google Colab {#sec-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)
+Use our [example notebook](../examples/colab-notebooks/colab-axolotl-example.ipynb).

 ## Platform-Specific Instructions {#sec-platform-specific}

@@ -156,7 +111,7 @@ We recommend using WSL2 (Windows Subsystem for Linux) or Docker.

 ### Conda/Pip venv {#sec-conda}

-1. Install Python ≥3.11
+1. Install Python ≥3.10
 2. Install PyTorch: https://pytorch.org/get-started/locally/
 3. Install Axolotl:
   ```{.bash}
--- a/docs/lora_optims.qmd
+++ b/docs/lora_optims.qmd
@@ -5,11 +5,10 @@ description: "Custom autograd functions and Triton kernels in Axolotl for optimi

 Inspired by [Unsloth](https://github.com/unslothai/unsloth), we've implemented two
 optimizations for LoRA and QLoRA fine-tuning, supporting both single GPU and multi-GPU
-(including the DDP, DeepSpeed, and FSDP2 settings) training. These include (1) SwiGLU
-and GEGLU activation function Triton kernels, and (2) LoRA MLP and attention custom
-autograd functions. Our goal was to leverage operator fusion and tensor re-use in order
-to improve speed and reduce memory usage during the forward and backward passes of
-these calculations.
+(in the DDP and DeepSpeed settings) training. These include (1) SwiGLU and GEGLU activation function
+Triton kernels, and (2) LoRA MLP and attention custom autograd functions. Our goal was
+to leverage operator fusion and tensor re-use in order to improve speed and reduce
+memory usage during the forward and backward passes of these calculations.

 We currently support several common model architectures, including (but not limited to):

@@ -85,10 +84,6 @@ lora_qkv_kernel: true
 lora_o_kernel: true
 ```

-::: {.callout-note}
-Currently, LoRA kernels are not supported for RLHF training, only SFT.
-:::
-
 ## Requirements

 - One or more NVIDIA or AMD GPUs (in order to use the Triton kernels)
@@ -132,5 +127,6 @@ computation path.
 ## Future Work

 - Support for additional model architectures
+- Support for the FSDP setting
 - Support for dropout and bias
 - Additional operator fusions
--- a/docs/lr_groups.qmd
+++ b/docs/lr_groups.qmd
@@ -27,9 +27,3 @@ learning_rate: 2e-5
 In this example, we have a default learning rate of 2e-5 across the entire model, but we have a separate learning rate
 of 1e-6 for all the self attention `o_proj` modules across all layers, and a learning are of 1e-5 to the 3rd layer's
 self attention `q_proj` module.
-
-::: {.callout-note}
-
-We currently only support varying `lr` for now. If you're interested in adding support for others (`weight_decay`), we welcome PRs. See https://github.com/axolotl-ai-cloud/axolotl/blob/613bcf90e58f3ab81d3827e7fc572319908db9fb/src/axolotl/core/trainers/mixins/optimizer.py#L17
-
-:::
--- a/docs/mixed_precision.qmd
+++ b/docs/mixed_precision.qmd
@@ -1,149 +0,0 @@
---
-title: "Mixed Precision Training"
-format:
-  html:
-    toc: true
-    toc-depth: 3
-    number-sections: true
-    code-tools: true
-execute:
-  enabled: false
---
-
-Mixed precision training uses lower precision data types to reduce memory usage and increase training speed while maintaining model quality. Axolotl supports several mixed precision formats:
-
- **FP16** - Half precision 16-bit (Pascal generation+)
- **BF16** - Brain Float 16-bit (Ampere generation+)
- **FP8** - 8-bit floating point (Hopper generation+)
-
-## FP16 Mixed Precision {#sec-fp16}
-
-### Overview {#sec-fp16-overview}
-
-FP16 is the traditional half-precision format, supported on older GPUs but can be less numerically stable than BF16.
-
-### Configuration {#sec-fp16-config}
-
-```{.yaml}
-fp16: true
-```
-
-### FP16 Considerations {#sec-fp16-considerations}
-
- May require gradient scaling to prevent underflow
- Less numerically stable than BF16
- Can cause training instability with some model architectures
- Consider using BF16 if your hardware supports it
-
-## BF16 Mixed Precision {#sec-bf16}
-
-### Overview {#sec-bf16-overview}
-
-BF16 (Brain Float 16) offers better numerical stability than FP16 and is the recommended mixed precision format for modern GPUs. It provides the same dynamic range as FP32 while using half the memory.
-
-### Configuration {#sec-bf16-config}
-
-```{.yaml}
-# Automatic BF16 detection (recommended)
-bf16: auto
-
-# Or explicitly enable
-bf16: true
-
-# For evaluation with BF16
-bf16: full  # Equivalent to bf16_full_eval in the HF trainer
-```
-
-## FP8 Mixed Precision {#sec-fp8}
-
-::: {.callout-note}
-FP8 support is experimental and requires compatible hardware (H100, H200) and recent PyTorch versions with TorchAO.
-:::
-
-### What is FP8? {#sec-fp8-overview}
-
-FP8 (8-bit floating point) can provide significant time savings compared to FP16/BF16 while maintaining training stability. Axolotl's implementation uses PyTorch's TorchAO library with "tensorwise" scaling strategy.
-
-### Requirements {#sec-fp8-software}
-
- Hopper+ GPUs (H100/H200)
- PyTorch 2.7+ (+ compatible TorchAO version)
- CUDA 12.4+
-
-### Configuration {#sec-fp8-config}
-
-Add to your YAML config:
-
-```{.yaml}
-# Enable FP8 mixed precision
-fp8: true
-
-# Optional: Enable FP8 for FSDP all-gather operations
-fp8_enable_fsdp_float8_all_gather: true
-
-# Enable torch.compile (almost always necessary for FP8 speedups)
-torch_compile: true
-```
-
-::: {.callout-important}
-**torch.compile is critical for FP8 performance**
-
-FP8 training requires `torch_compile: true` to see meaningful speedups. Without compilation, FP8 may actually be slower and use more memory than FP16/BF16.
-:::
-
-### Advanced FP8 Configs {#sec-fp8-advanced}
-
-For [FSDP](multi-gpu.qmd#sec-fsdp) (Fully Sharded Data Parallel) training:
-
-```{.yaml}
-fp8: true
-fp8_enable_fsdp_float8_all_gather: true
-
-torch_compile: true
-
-# FSDP configuration
-fsdp_version: 2
-fsdp_config:
-  offload_params: false
-  cpu_ram_efficient_loading: true
-  auto_wrap_policy: TRANSFORMER_BASED_WRAP
-  transformer_layer_cls_to_wrap: LlamaDecoderLayer
-  state_dict_type: FULL_STATE_DICT
-  reshard_after_forward: true
-```
-
-## Best Practices {#sec-best-practices}
-
-### Choosing Precision Format {#sec-choosing-format}
-
- **Start with automatic detection**: `bf16: auto`
- **For Hopper+ (H100/H200)**: Try FP8 + torch.compile for maximum speed
- **For Ampere (A100/RTX 30/40)**: Use BF16
- **For older Pascal/Turing GPUs**: Use FP16 with caution
- **For very old or unsupported GPUs**: Use FP32
-
-### Validation and Testing {#sec-validation}
-
-Always validate your mixed precision setup:
-
- **Start with a small dataset** to verify stability
- **Monitor loss curves** for irregularities
- **Compare with FP32 baseline** when possible
- **Test evaluation metrics** match expectations
-
-### FP8 Particulars {#sec-fp8-details}
-
- Use cases
-  - Single GPU training
-  - Multi GPU training with FSDP2 or Deepspeed
- Speedups
-  - Please refer to the [TorchAO FP8 training benchmarks](https://github.com/pytorch/ao/tree/main/torchao/float8#rowwise-scaling) for expected matmul speedups for different (M, K, N) settings
-  - Concrete number for LLaMA 3 8B training can be found [here](https://github.com/pytorch/ao/tree/main/torchao/float8#training-benchmarks)
- Known issues:
-  - FP8 + DDP + `torch.compile` (causes [error](https://gist.github.com/djsaunde/0c1664c32e44a64d31b5e01b4aafe5c4))
-  - FP8 + FSDP2 + `torch.compile` + FSDP2 activation checkpointing tends to be _slower_ than the BF16 equivalent training
-  - Flash Attention 2 does not play nicely with `torch.compile`
-
-See `examples/llama-3/3b-fp8-fsdp2.yaml` for an optimized example config. Enabling FP8 mixed precision + FP8 all-gather training results in ~10% faster iterations per second vs. BF16 for a relatively small (3B param) model
-
-For more information on multi-GPU training, see our [Multi-GPU guide](multi-gpu.qmd).
--- a/docs/multi-gpu.qmd
+++ b/docs/multi-gpu.qmd
@@ -23,6 +23,8 @@ Axolotl supports several methods for multi-GPU training:

 ## DeepSpeed {#sec-deepspeed}

+DeepSpeed is the recommended approach for multi-GPU training due to its stability and performance. It provides various optimization levels through ZeRO stages.
+
 ### Configuration {#sec-deepspeed-config}

 Add to your YAML config:
@@ -30,6 +32,7 @@ Add to your YAML config:
 ```{.yaml}
 deepspeed: deepspeed_configs/zero1.json
 ```
+
 ### Usage {#sec-deepspeed-usage}

 ```{.bash}
@@ -63,67 +66,9 @@ Start from Stage 1 -> Stage 2 -> Stage 3.

 :::

-## Fully Sharded Data Parallel (FSDP) {#sec-fsdp}
+## FSDP {#sec-fsdp}

-::: {.callout-note}
-
-FSDP2 is recommended for new users. FSDP1 is deprecated and will be removed in an upcoming release of Axolotl.
-
-:::
-
-### Migrating from FSDP1 to FSDP2 {#sec-migrate-fsdp1-fsdp2}
-
-To migrate your config from FSDP1 to FSDP2, you must use the `fsdp_version` top-level config field to specify the FSDP version, and
-also follow the config field mapping below to update field names.
-
-#### Config mapping
-
-FSDP1 | FSDP2
-------- | --------
-fsdp_sharding_strategy | reshard_after_forward
-fsdp_backward_prefetch_policy | **REMOVED**
-fsdp_backward_prefetch | **REMOVED**
-fsdp_forward_prefetch | **REMOVED**
-fsdp_sync_module_states | **REMOVED**
-fsdp_cpu_ram_efficient_loading | cpu_ram_efficient_loading
-fsdp_state_dict_type | state_dict_type
-fsdp_use_orig_params | **REMOVED**
-fsdp_activation_checkpointing | activation_checkpointing
-
-For more details, please see the migration guide in the [torchtitan repo](https://github.com/pytorch/torchtitan/blob/main/docs/fsdp.md). In Axolotl,
-if you were using the following FSDP1 config:
-
-```{.yaml}
-fsdp_version: 1
-fsdp_config:
-  fsdp_offload_params: false
-  fsdp_cpu_ram_efficient_loading: true
-  fsdp_auto_wrap_policy: TRANSFORMER_BASED_WRAP
-  fsdp_transformer_layer_cls_to_wrap: Qwen3DecoderLayer
-  fsdp_state_dict_type: FULL_STATE_DICT
-  fsdp_sharding_strategy: FULL_SHARD
-```
-
-You can migrate to the following FSDP2 config:
-
-```{.yaml}
-fsdp_version: 2
-fsdp_config:
-  offload_params: false
-  cpu_ram_efficient_loading: true
-  auto_wrap_policy: TRANSFORMER_BASED_WRAP
-  transformer_layer_cls_to_wrap: Qwen3DecoderLayer
-  state_dict_type: FULL_STATE_DICT
-  reshard_after_forward: true
-```
-
-### FSDP1 (deprecated) {#sec-fsdp-config}
-
-::: {.callout-note}
-
-Using `fsdp` to configure FSDP is deprecated and will be removed in an upcoming release of Axolotl. Please use `fsdp_config` as above instead.
-
-:::
+### Basic FSDP Configuration {#sec-fsdp-config}

 ```{.yaml}
 fsdp:
@@ -135,7 +80,6 @@ fsdp_config:
  fsdp_transformer_layer_cls_to_wrap: LlamaDecoderLayer
 ```

-
 ## Sequence parallelism {#sec-sequence-parallelism}

 We support sequence parallelism (SP) via the
@@ -143,7 +87,20 @@ We support sequence parallelism (SP) via the
 allows one to split up sequences across GPUs, which is useful in the event that a
 single sequence causes OOM errors during model training.

-See our [dedicated guide](sequence_parallelism.qmd) for more information.
+First, install `ring-flash-attn`, recommended via `pip install axolotl[ring-flash-attn]`,
+or from source with `pip install .[ring-flash-attn]`.
+
+Your Axolotl YAML config should contain the following lines:
+
+```{.yaml}
+sequence_parallel_degree: 4  # Split each sequence into 4 parts, one per GPU
+flash_attention: true  # Required with sequence parallelism
+
+# Optional; strides across the key dimension. Larger values use more memory but will make training faster.
+heads_k_stride: 1
+```
+
+See our [dedicated guide](sequence_parallelism.qmd) for more details.

 ### FSDP + QLoRA {#sec-fsdp-qlora}

--- a/docs/multi-node.qmd
+++ b/docs/multi-node.qmd
@@ -40,13 +40,13 @@ use_cpu: false

 Configure your model to use FSDP in the Axolotl yaml. For example:
 ```yaml
-fsdp_version: 2
+fsdp:
+  - full_shard
+  - auto_wrap
 fsdp_config:
-  offload_params: true
-  state_dict_type: FULL_STATE_DICT
-  auto_wrap_policy: TRANSFORMER_BASED_WRAP
-  transformer_layer_cls_to_wrap: LlamaDecoderLayer
-  reshard_after_forward: true
+  fsdp_offload_params: true
+  fsdp_state_dict_type: FULL_STATE_DICT
+  fsdp_transformer_layer_cls_to_wrap: LlamaDecoderLayer
 ```

 All you have to do now is launch using accelerate as you would usually do on each machine and voila, the processes will start once you have launched accelerate on every machine.
@@ -69,19 +69,11 @@ export NCCL_BUFFSIZE=2097152

 Run the following on each node:

-### Option 1: New Axolotl CLI with launcher args (Recommended)
-
-```bash
-axolotl train config.yaml --launcher torchrun -- --nnodes $num_nodes --nproc_per_node $gpu_per_node --rdzv_id $rdzv_id --rdzv_backend c10d --rdzv_endpoint "$head_node_ip:$head_node_port"
-```
-
-### Option 2: Direct torchrun (Legacy)
-
 ```bash
 torchrun --nnodes $num_nodes --nproc_per_node $gpu_per_node --rdzv_id $rdzv_id --rdzv_backend c10d --rdzv_endpoint "$head_node_ip:$head_node_port" -m axolotl.cli.train config.yaml
 ```

-Please make sure to substitute the placeholder variables:
+Please make sure to substitute the placeholder variables.

 - `num_nodes`: Number of nodes (containing GPUs)
 - `gpu_per_node`: Number of gpus per node
@@ -89,6 +81,8 @@ Please make sure to substitute the placeholder variables:
 - `head_node_port`: Port of the head node (make sure other machines can connect to this. Default 29400)
 - `rdzv_id`: A unique job ID that is used by the job across nodes.

-The new CLI approach (Option 1) is recommended as it provides consistent argument handling and works seamlessly with other Axolotl CLI features.
+::: {.callout-note}
+You need to call `axolotl.cli.train` instead of `axolotl train` as the latter calls accelerate under the hood
+:::

 More info on the available configs can be found on the Pytorch docs [here](https://pytorch.org/docs/stable/elastic/run.html)
--- a/docs/multimodal.qmd
+++ b/docs/multimodal.qmd
@@ -13,14 +13,9 @@ format:
 - [Pixtral](#sec-pixtral)
 - [Llava-1.5](#sec-llava-15)
 - [Mistral-Small-3.1](#sec-mistral-small-31)
- [Magistral-Small-2509](#sec-magistral-small-2509)
- [Voxtral](#sec-voxtral)
 - [Gemma-3](#sec-gemma-3)
- [Gemma-3n](#sec-gemma-3n)
 - [Qwen2-VL](#sec-qwen2-vl)
 - [Qwen2.5-VL](#sec-qwen25-vl)
- [SmolVLM2](#sec-smolvlm2)
- [LFM2-VL](#sec-lfm2-vl)

 ## Usage

@@ -35,19 +30,20 @@ skip_prepare_dataset: true
 remove_unused_columns: false  # leave columns in place as they are needed to handle image embeddings during training
 sample_packing: false  # not yet supported with multimodal

-chat_template:  # see in next section if specified
+chat_template:  # see in next section

 # example dataset
 datasets:
  - path: HuggingFaceH4/llava-instruct-mix-vsft
    type: chat_template
    split: train[:1%]
+    field_messages: messages

 # (optional) if doing lora, only finetune the Language model,
 # leave the vision model and vision tower frozen
 # load_in_8bit: true
 adapter: lora
-lora_target_modules: 'model.language_model.layers.[\d]+.(mlp|cross_attn|self_attn).(up|down|gate|q|k|v|o)_proj'
+lora_target_modules: 'language_model.model.layers.[\d]+.(mlp|cross_attn|self_attn).(up|down|gate|q|k|v|o)_proj'

 # (optional) if you want to resize images to a set size
 image_size: 512
@@ -56,14 +52,10 @@ image_resize_algorithm: bilinear

 Please see [examples](https://github.com/axolotl-ai/axolotl/tree/main/examples) folder for full configs.

-::: {.callout-tip}
+::: {.callout-warning}
 Some of our chat_templates have been extended to support broader dataset types. This should not break any existing configs.
 :::

-::: {.callout-note}
-As of now, we do not truncate nor drop samples based on `sequence_len` as each arch has different ways to process non-text tokens. We are looking for help on this.
-:::
-
 ### Mllama {#sec-mllama}

 ```yaml
@@ -98,32 +90,10 @@ chat_template: llava

 ### Mistral-Small-3.1 {#sec-mistral-small-31}

-::: {.callout-tip}
-Please make sure to install vision lib via `pip install 'mistral-common[opencv]==1.8.5'`
-:::
-
 ```yaml
 base_model: mistralai/Mistral-Small-3.1-24B-Instruct-2503
-```

-### Magistral-Small-2509 {#sec-magistral-small-2509}
-
-::: {.callout-tip}
-Please make sure to install vision lib via `pip install 'mistral-common[opencv]==1.8.5'`
-:::
-
-```yaml
-base_model: mistralai/Magistral-Small-2509
-```
-
-### Voxtral {#sec-voxtral}
-
-::: {.callout-tip}
-Please make sure to install audio lib via `pip3 install librosa==0.11.0 'mistral_common[audio]==1.8.3'`
-:::
-
-```yaml
-base_model: mistralai/Voxtral-Mini-3B-2507
+chat_template: mistral_v7_tekken
 ```

 ### Gemma-3 {#sec-gemma-3}
@@ -140,22 +110,6 @@ base_model: google/gemma-3-4b-it
 chat_template: gemma3
 ```

-### Gemma-3n {#sec-gemma-3n}
-
-::: {.callout-warning}
-The model's initial loss and grad norm will be very high. We suspect this to be due to the Conv in the vision layers.
-:::
-
-::: {.callout-tip}
-Please make sure to install `timm` via `pip3 install timm==1.0.17`
-:::
-
-```yaml
-base_model: google/gemma-3n-E2B-it
-
-chat_template: gemma3n
-```
-
 ### Qwen2-VL {#sec-qwen2-vl}

 ```yaml
@@ -172,43 +126,13 @@ base_model: Qwen/Qwen2.5-VL-7B-Instruct
 chat_template: qwen2_vl  # same as qwen2-vl
 ```

-### Qwen3-VL {#sec-qwen3-vl}
-
-```yaml
-base_model: Qwen/Qwen3-VL-4B-Instruct
-
-chat_template: qwen2_vl  # same as qwen2-vl
-```
-
-### SmolVLM2 {#sec-smolvlm2}
-
-::: {.callout-tip}
-Please make sure to install `num2words` via `pip3 install num2words==0.5.14`
-:::
-
-```yaml
-base_model: HuggingFaceTB/SmolVLM2-500M-Video-Instruct
-```
-
-### LFM2-VL {#sec-lfm2-vl}
-
-::: {.callout-warning}
-Please uninstall `causal-conv1d` via `pip3 uninstall -y causal-conv1d`
-:::
-
-```yaml
-base_model: LiquidAI/LFM2-VL-450M
-```
-
 ## Dataset Format

 For multi-modal datasets, we adopt an extended `chat_template` format similar to OpenAI's Message format.

 - A message is a list of `role` and `content`.
 - `role` can be `system`, `user`, `assistant`, etc.
- `content` is a list of `type` and (`text`, `image`, `path`, `url`, `base64`, or `audio`).
-
-### Image
+- `content` is a list of `type` and (`text` or `image` or `path` or `url` or `base64`).

 ::: {.callout-note}
 For backwards compatibility:
@@ -217,43 +141,15 @@ For backwards compatibility:
 - If `content` is a string, it will be converted to a list with `type` as `text`.
 :::

+::: {.callout-tip}
 For image loading, you can use the following keys within `content` alongside `"type": "image"`:

 - `"path": "/path/to/image.jpg"`
 - `"url": "https://example.com/image.jpg"`
 - `"base64": "..."`
 - `"image": PIL.Image`
-
-### Audio
-
-For audio loading, you can use the following keys within `content` alongside `"type": "audio"`:
-
- `"path": "/path/to/audio.mp3"`
- `"url": "https://example.com/audio.mp3"`
- `"audio": np.ndarray`
-
-::: {.callout-tip}
-
-You may need to install `librosa` via `pip3 install librosa==0.11.0`.
-
 :::

-### Video
-
-::: {.callout-warning}
-
-This is not well tested at the moment. We welcome contributors!
-
-:::
-
-For video loading, you can use the following keys within `content` alongside `"type": "video"`:
-
- `"path": "/path/to/video.mp4"`
- `"url": "https://example.com/video.mp4"`
- `"video": np.ndarray | list[PIL.Image.Image] | torch.Tensor` (or list of the aforementioned)
-
-### Example
-
 Here is an example of a multi-modal dataset:
 ```json
 [
@@ -282,9 +178,3 @@ Here is an example of a multi-modal dataset:
  }
 ]
 ```
-
-## FAQ
-
-1. `PIL.UnidentifiedImageError: cannot identify image file ...`
-
-`PIL` could not retrieve the file at `url` using `requests`. Please check for typo. One alternative reason is that the request is blocked by the server.
--- a/docs/nd_parallelism.qmd
+++ b/docs/nd_parallelism.qmd
@@ -1,108 +0,0 @@
---
-title: "N-D Parallelism (Beta)"
---
-
-Axolotl enables training models at scale by composing different parallelism techniques. This is essential when:
-
- A model's weights are too large to fit on a single GPU's memory.
- A model's activations, especially with very long contexts, are too large for a single GPU.
- You want to accelerate training by using multiple GPUs or nodes.
-
-or combinations of the above!
-
-## Core Concepts
-
-Parallelism strategies can be combined. The key is understanding how each one divides the workload. PyTorch's `DeviceMesh` is the modern way to manage these combinations, creating a logical grid of your GPUs and assigning different parallel strategies to different dimensions of the grid.
-
-### Data Parallelism {#sec-dp}
-
-Data Parallelism focuses on splitting the global data batch across GPUs.
-
- Distributed Data Parallel (DDP): The classic approach. The full model is replicated on every GPU. Each GPU processes a different slice of the data batch. Gradients are then averaged across all GPUs after the backward pass to keep the models synchronized. This can substantially improve data throughput compared to single-device training, but requires that each GPU is able to hold the entire model, its gradients, and optimizer states.
-
- [Fully Sharded Data Parallel (FSDP)](multi-gpu.qmd#fully-sharded-data-parallel-(fsdp)): A highly memory-efficient form of data parallelism (inspired by DeepSpeed's ZeRO). Instead of replicating the model, FSDP shards the model's *parameters, gradients, and optimizer states* across the GPUs in the data-parallel group. During computation, each GPU receives the specific parameters it needs via an `all_gather` operation just before they are used, and they can be discarded immediately after (`reshard-after-forward`).
-    - FSDP maps to ZeRO stages:
-        - ZeRO-2 (`reshard_after_forward=False`): Shards gradients and optimizer states. Model weights are replicated on each GPU.
-        - ZeRO-3 (`reshard_after_forward=True`): Shards gradients, optimizer states, AND model parameters. This provides the most memory savings at the cost of more communication (re-gathering parameters for both forward and backward passes).
-
-### [Experimental] Tensor Parallelism (TP) {#sec-tp}
-
-Also known as "horizontal model parallelism," as described in the [Megatron-LM paper](https://arxiv.org/pdf/1909.08053.pdf). Instead of splitting the batch, TP splits the model's layers themselves across GPUs.
-
- How it works: For a linear layer `Y = XA`, the weight matrix `A` is split column-wise (`A = [A_1, A_2]`). The computation becomes `Y_1 = XA_1` and `Y_2 = XA_2`, which can happen in parallel on different GPUs. The final output `Y` is simply the concatenation of `Y_1` and `Y_2`. Check [this comment](https://github.com/huggingface/transformers/issues/10321#issuecomment-783543530) for more detailed info.
- Requirement: TP involves frequent, small communications within a forward/backward pass. It requires a very fast interconnect between GPUs (e.g., NVLink) and is typically not recommended across different nodes.
-
-### Context Parallelism (CP) {#sec-cp}
-
-Context Parallelism, also called [Sequence Parallelism](sequence_parallelism.qmd), addresses the memory bottleneck from long sequences. The input sequence itself is split along the sequence length dimension and distributed across GPUs.
-
- How it works: If you have a sequence of 8192 tokens and a `context_parallel_size` of 4, each GPU will only handle a chunk of 2048 tokens.
- The Challenge: Attention is not local; every token needs to "attend to" every other token. Splitting the sequence breaks this.
- The Solution (`ring-flash-attention`): An efficient communication protocol is used. To compute attention for its local sequence chunk, each GPU passes its Key-Value (KV) cache to its neighbor in a "ring." After `N-1` steps, every GPU has seen the KV-cache from all other GPUs, allowing it to compute the correct attention values for its chunk. This is implemented using the highly optimized `flash-attention` kernel at each step.
-
-### Hybrid Sharding Data Parallel (HSDP) {#sec-hsdp}
-
-HSDP is a 2D strategy that intelligently combines FSDP and DDP, typically for multi-node training.
-
- Intra-Node (within a machine): Use FSDP. This is efficient because GPUs on the same node have fast interconnects (NVLink), making the `all_gather` operations for sharded parameters fast.
- Inter-Node (across machines): Use DDP. The gradient synchronization between nodes is less frequent than FSDP's parameter gathering, making it a better fit for the slower node-to-node network (e.g., Ethernet/Infiniband).
- Example: With 2 nodes of 8 GPUs each (16 total), you could have `dp_shard_size=8` (FSDP within each node) and `dp_replicate_size=2` (DDP across the two nodes).
-
-## Usage
-
-```yaml
-# FSDP config. See https://docs.axolotl.ai/docs/multi-gpu.html#sec-fsdp
-fsdp_version: 2
-fsdp_config:
-  # ...
-
-# The number of GPUs to shard the model parameters across (FSDP dimension).
-dp_shard_size: 4
-
-# The number of times to replicate the sharded model (DDP dimension).
-dp_replicate_size: 2
-
-# Number of GPUs for Tensor Parallelism.
-tensor_parallel_size: 1  # (default is 1, no TP)
-
-# Number of GPUs for Context/Sequence Parallelism.
-context_parallel_size: 1 # (default is 1, no CP)
-```
-
-Note: We recommend FSDP. DeepSpeed is only compatible with `tensor_parallel_size`.
-
-## Examples
-
-::: {.callout-tip}
-See our example configs [here](https://github.com/axolotl-ai-cloud/axolotl/tree/main/examples/distributed-parallel).
-:::
-
-1.  HSDP on 2 nodes with 4 GPUs each (8 GPUs total):
-    - You want FSDP within each node and DDP across nodes.
-    - Set `dp_shard_size: 4` and `dp_replicate_size: 2`.
-
-2.  FSDP + TP on a single 8-GPU node:
-    - You want to split the model across 4 GPUs using FSDP, and further split each layer across 2 GPUs with TP.
-    - Set `dp_shard_size: 4` and `tensor_parallel_size: 2`.
-
-3.  FSDP + CP on a single 8-GPU node for long context:
-    - You want to shard the model across all 8 GPUs and also split the sequence length across all 8 GPUs.
-    - Set `dp_shard_size: 8` and `context_parallel_size: 8`. Note: this means the data parallel group and context parallel group are the same. A more common setup might be to shard across a smaller group.
-
-## Support Matrix
-
-This matrix describes how different parallelism methods can be combined in Axolotl.
-
-| Combination | `dp_replicate_size` | `dp_shard_size` | `tp_size` | `cp_size` | Status & Notes |
-| --- | :---: | :---: |:---:|:---:|---|
-| **FSDP** (ZeRO-3) | 1 | >1 | 1 | 1 | ✅ Fully supported. Shards model across all GPUs. |
-| **HSDP** | >1 | >1 | 1 | 1 | ✅ Fully supported. FSDP intra-node, DDP inter-node. |
-| **FSDP + TP** | 1 | >1 | >1 | 1 | ✅ **2D Parallelism**. Shards the model across a `dp_shard` group, and TP-splits layers within the `tp` group. |
-| **HSDP + TP** | >1 | >1 | >1 | 1 | ✅ **3D Parallelism**. A powerful but complex combination. |
-| **FSDP + CP** | 1 | >1 | 1 | >1 | ✅ **2D Parallelism**. Combines FSDP with context parallelism. |
-| **FSDP + TP + CP**| 1 | >1 | >1| >1| ✅ **3D Parallelism**. Another advanced combination. |
-| DDP + TP/CP | >1 | 1 | >1 | >1 | ❌ **Not Supported**. The `ParallelismConfig` explicitly prevents this, as composing pure DDP with TP or CP is currently not supported. You should use FSDP + TP/CP instead (`dp_shard_size > 1`). |
-| Just TP / CP | 1 | 1 | >1 | >1 | ✅ Supported. Useful for inference or when the model fits on one GPU but context is too long. |
-
- `tp_size` refers to `tensor_parallel_size`
- `cp_size` refers to `context_parallel_size`
--- a/docs/optimizations.qmd
+++ b/docs/optimizations.qmd
@@ -1,133 +0,0 @@
---
-title: Optimizations Guide
-description: A guide to the performance and memory optimizations available in Axolotl.
---
-
-Axolotl includes numerous optimizations to speed up training, reduce memory usage, and handle large models.
-
-This guide provides a high-level overview and directs you to the detailed documentation for each feature.
-
-## Speed Optimizations
-
-These optimizations focus on increasing training throughput and reducing total training time.
-
-### Sample Packing
-
-Improves GPU utilization by combining multiple short sequences into a single packed sequence for training. This requires enabling one of the [attention](#attention-implementations) implementations below.
-
- **Config:** `sample_packing: true`
- **Learn more:** [Sample Packing](multipack.qmd)
-
-### Attention Implementations
-
-Using an optimized attention implementation is critical for training speed.
-
- **[Flash Attention 2](https://github.com/Dao-AILab/flash-attention)**: `flash_attention: true`. **(Recommended)** The industry standard for fast attention on modern GPUs. Requires Ampere or higher. For AMD, check [AMD Support](https://github.com/Dao-AILab/flash-attention?tab=readme-ov-file#amd-rocm-support).
- **[Flex Attention](https://pytorch.org/blog/flexattention/)**: `flex_attention: true`.
- **[SDP Attention](https://docs.pytorch.org/docs/stable/generated/torch.nn.functional.scaled_dot_product_attention.html)**: `sdp_attention: true`. PyTorch's native implementation.
- **[Xformers](https://github.com/facebookresearch/xformers)**: `xformers_attention: true`. Works with FP16.
-
-*Note: You should only enable one attention backend.*
-
-### LoRA Optimizations
-
-Leverages optimized kernels to accelerate LoRA training and reduce memory usage.
-
- **Learn more:** [LoRA Optimizations Documentation](lora_optims.qmd)
-
-## Memory Optimizations
-
-These techniques help you fit larger models or use bigger batch sizes on your existing hardware.
-
-### Parameter Efficient Finetuning (LoRA & QLoRA)
-
-Drastically reduces memory by training a small set of "adapter" parameters instead of the full model. This is the most common and effective memory-saving technique.
-
- Examples: Find configs with `lora` or `qlora` in the [examples directory](https://github.com/axolotl-ai-cloud/axolotl/tree/main/examples/llama-3).
- Config Reference: See `adapter`, `load_in_4bit`, and `load_in_8bit` in the [Configuration Reference](config-reference.qmd).
-
-### Gradient Checkpointing & Activation Offloading
-
-These techniques save VRAM by changing how activations are handled.
-
- Gradient Checkpointing: re-computes activations during the backward pass, trading compute time for VRAM.
- Activation Offloading: moves activations to CPU RAM or disk, trading I/O overhead for VRAM.
- Learn more: [Gradient Checkpointing and Offloading Docs](gradient_checkpointing.qmd)
-
-### Cut Cross Entropy (CCE)
-
-Reduces VRAM usage by using an optimized cross-entropy loss calculation.
-
- **Learn more:** [Custom Integrations - CCE](custom_integrations.qmd#cut-cross-entropy)
-
-### Liger Kernels
-
-Provides efficient Triton kernels to improve training speed and reduce memory usage.
-
- **Learn more:** [Custom Integrations - Liger Kernels](custom_integrations.qmd#liger-kernels)
-
-## Long Context Models
-
-Techniques to train models on sequences longer than their original context window.
-
-### RoPE Scaling
-
-Extends a model's context window by interpolating its Rotary Position Embeddings.
-
- **Config:** Pass the `rope_scaling` config under the `overrides_of_model_config: `. To learn how to set RoPE, check the respective model config.
-
-### Sequence Parallelism
-
-Splits long sequences across multiple GPUs, enabling training with sequence lengths that would not fit on a single device.
-
- **Learn more:** [Sequence Parallelism Documentation](sequence_parallelism.qmd)
-
-### Artic Long Sequence Training (ALST)
-
-ALST is a recipe that combines several techniques to train long-context models efficiently. It typically involves:
-
- TiledMLP to reduce memory usage in MLP layers.
- Tiled Loss functions (like [CCE](#cut-cross-entropy-(cce) or [Liger](#liger-kernels)).
- Activation Offloading to CPU.
-
- Example: [ALST Example Configuration](https://github.com/axolotl-ai-cloud/axolotl/tree/main/examples/alst)
-
-## Large Models (Distributed Training)
-
-To train models that don't fit on a single GPU, you'll need to use a distributed training strategy like FSDP or DeepSpeed. These frameworks shard the model weights, gradients, and optimizer states across multiple GPUs and nodes.
-
- **Learn more:** [Multi-GPU Guide](multi-gpu.qmd)
- **Learn more:** [Multi-Node Guide](multi-node.qmd)
-
-### N-D Parallelism (Beta)
-
-For advanced scaling, Axolotl allows you to compose different parallelism techniques (e.g., Data, Tensor, Sequence Parallelism). This is a powerful approach to train an extremely large model by overcoming multiple bottlenecks at once.
-
- **Learn more:** [N-D Parallelism Guide](nd_parallelism.qmd)
-
-
-## Quantization
-
-Techniques to reduce the precision of model weights for memory savings.
-
-### 4-bit Training (QLoRA)
-
-The recommended approach for quantization-based training. It loads the base model in 4-bit using `bitsandbytes` and then trains QLoRA adapters. See [Adapter Finetuning](#adapter-finetuning-lora-qlora) for details.
-
-### FP8 Training
-
-Enables training with 8-bit floating point precision on supported hardware (e.g., NVIDIA Hopper series GPUs) for significant speed and memory gains.
-
- **Example:** [Llama 3 FP8 FSDP Example](https://github.com/axolotl-ai-cloud/axolotl/blob/main/examples/llama-3/3b-fp8-fsdp2.yaml)
-
-### Quantization Aware Training (QAT)
-
-Simulates quantization effects during training, helping the model adapt and potentially improving the final accuracy of the quantized model.
-
- **Learn more:** [QAT Documentation](qat.qmd)
-
-### GPTQ
-
-Allows you to finetune LoRA adapters on top of a model that has already been quantized using the GPTQ method.
-
- **Example:** [GPTQ LoRA Example](https://github.com/axolotl-ai-cloud/axolotl/blob/main/examples/llama-2/gptq-lora.yml)
--- a/docs/optimizers.qmd
+++ b/docs/optimizers.qmd
@@ -1,129 +0,0 @@
---
-title: Optimizers
-description: Configuring optimizers
---
-
-## Overview
-
-Axolotl supports all optimizers supported by [transformers OptimizerNames](https://github.com/huggingface/transformers/blob/51f94ea06d19a6308c61bbb4dc97c40aabd12bad/src/transformers/training_args.py#L142-L187)
-
-Here is a list of optimizers supported by transformers as of `v4.54.0`:
-
- `adamw_torch`
- `adamw_torch_fused`
- `adamw_torch_xla`
- `adamw_torch_npu_fused`
- `adamw_apex_fused`
- `adafactor`
- `adamw_anyprecision`
- `adamw_torch_4bit`
- `adamw_torch_8bit`
- `ademamix`
- `sgd`
- `adagrad`
- `adamw_bnb_8bit`
- `adamw_8bit`  # alias for adamw_bnb_8bit
- `ademamix_8bit`
- `lion_8bit`
- `lion_32bit`
- `paged_adamw_32bit`
- `paged_adamw_8bit`
- `paged_ademamix_32bit`
- `paged_ademamix_8bit`
- `paged_lion_32bit`
- `paged_lion_8bit`
- `rmsprop`
- `rmsprop_bnb`
- `rmsprop_bnb_8bit`
- `rmsprop_bnb_32bit`
- `galore_adamw`
- `galore_adamw_8bit`
- `galore_adafactor`
- `galore_adamw_layerwise`
- `galore_adamw_8bit_layerwise`
- `galore_adafactor_layerwise`
- `lomo`
- `adalomo`
- `grokadamw`
- `schedule_free_radam`
- `schedule_free_adamw`
- `schedule_free_sgd`
- `apollo_adamw`
- `apollo_adamw_layerwise`
- `stable_adamw`
-
-
-## Custom Optimizers
-
-Enable custom optimizers by passing a string to the `optimizer` argument. Each optimizer will receive beta and epsilon args, however, some may accept additional args which are detailed below.
-
-### optimi_adamw
-
-```yaml
-optimizer: optimi_adamw
-```
-
-### ao_adamw_4bit
-
-Deprecated: Please use `adamw_torch_4bit`.
-
-### ao_adamw_8bit
-
-Deprecated: Please use `adamw_torch_8bit`.
-
-### ao_adamw_fp8
-
-
-```yaml
-optimizer: ao_adamw_fp8
-```
-
-### adopt_adamw
-
-GitHub: [https://github.com/iShohei220/adopt](https://github.com/iShohei220/adopt)
-Paper: [https://arxiv.org/abs/2411.02853](https://arxiv.org/abs/2411.02853)
-
-```yaml
-optimizer: adopt_adamw
-```
-
-### came_pytorch
-
-GitHub: [https://github.com/yangluo7/CAME/tree/master](https://github.com/yangluo7/CAME/tree/master)
-Paper: [https://arxiv.org/abs/2307.02047](https://arxiv.org/abs/2307.02047)
-
-```yaml
-optimizer: came_pytorch
-
-# optional args (defaults below)
-adam_beta1: 0.9
-adam_beta2: 0.999
-adam_beta3: 0.9999
-adam_epsilon: 1e-30
-adam_epsilon2: 1e-16
-```
-
-### muon
-
-Blog: [https://kellerjordan.github.io/posts/muon/](https://kellerjordan.github.io/posts/muon/)
-Paper: [https://arxiv.org/abs/2502.16982v1](https://arxiv.org/abs/2502.16982v1)
-
-```yaml
-optimizer: muon
-```
-
-### dion
-
-Microsoft's Dion (DIstributed OrthoNormalization) optimizer is a scalable and communication-efficient
-orthonormalizing optimizer that uses low-rank approximations to reduce gradient communication.
-
-GitHub: [https://github.com/microsoft/dion](https://github.com/microsoft/dion)
-Paper: [https://arxiv.org/pdf/2504.05295](https://arxiv.org/pdf/2504.05295)
-Note: Implementation written for PyTorch 2.7+ for DTensor
-
-```yaml
-optimizer: dion
-dion_lr: 0.01
-dion_momentum: 0.95
-lr: 0.00001  # learning rate for embeddings and parameters that fallback to AdamW
-```
--- a/docs/qat.qmd
+++ b/docs/qat.qmd
@@ -1,40 +0,0 @@
---
-title: "Quantization Aware Training (QAT)"
-back-to-top-navigation: true
-toc: true
-toc-expand: 2
-toc-depth: 4
---
-
-## Overview
-
-[Quantization Aware Training](https://pytorch.org/blog/introduction-to-quantization-on-pytorch/#quantization-aware-training) (QAT) is a technique for improving the accuracy of models which are quantized
-by applying "fake" quantizations to the model's weights (and optionally, activations) during training. This fake
-quantization allows for the model to adjust for noise introduced by the quantization, so when the model is eventually
-quantized, the accuracy loss is minimized. We use the quantization techniques implemented in [torchao](https://github.com/pytorch/ao) to provide
-support for QAT and post-training quantization (PTQ) in axolotl.
-
-We recommend reviewing the excellent QAT tutorial in the [torchtune library](https://pytorch.org/torchtune/main/tutorials/qat_finetune.html#quantizing-the-qat-model),
-and the QAT documentation in the [torchao library](https://github.com/pytorch/ao/tree/main/torchao/quantization/qat), for more details.
-
-## Configuring QAT in Axolotl
-
-To enable QAT in axolotl, add the following to your configuration file:
-
-```yaml
-qat:
-  activation_dtype: # Optional[str] = "int8". Fake quantization layout to use for activation quantization. Valid options are "int4", "int8", "float8"
-  weight_dtype: # Optional[str] = "int8". Fake quantization layout to use for weight quantization. Valid options are "int4", "fp8", and "nvfp4".
-  group_size: # Optional[int] = 32. The number of elements in each group for per-group fake quantization
-  fake_quant_after_n_steps: # Optional[int] = None. The number of steps to apply fake quantization after
-```
-
-We support the following quantization schemas:
-
- `Int4WeightOnly` (requires the `fbgemm-gpu` extra when installing Axolotl)
- `Int8DynamicActivationInt4Weight`
- `Float8DynamicActivationFloat8Weight`
- `Float8DynamicActivationInt4Weight`
- `NVFP4`
-
-Once you have finished training, you must quantize your model by using the same quantization configuration which you used to train the model with. You can use the [`quantize`](./quantize.qmd) command to do this.
--- a/docs/quantize.qmd
+++ b/docs/quantize.qmd
@@ -1,60 +0,0 @@
---
-title: "Quantization with torchao"
-back-to-top-navigation: true
-toc: true
-toc-expand: 2
-toc-depth: 4
---
-
-Quantization is a technique to lower the memory footprint of your model, potentially at the cost of accuracy or model performance. We support quantizing your model using the [torchao](https://github.com/pytorch/ao) library. Quantization is supported for both post-training quantization (PTQ) and quantization-aware training (QAT).
-
-
-::: {.callout-note}
-
-We do not currently support quantization techniques such as GGUF/GPTQ,EXL2 at the moment.
-
-:::
-
-## Configuring Quantization in Axolotl
-
-Quantization is configured using the `quantization` key in your configuration file.
-
-```yaml
-base_model: # The path to the model to quantize.
-quantization:
-  activation_dtype: # Optional[str] = "int8". Fake quantization layout to use for activation quantization. Valid options are "int4", "int8", "float8"
-  weight_dtype: # Optional[str] = "int8". Fake quantization layout to use for weight quantization. Valid options are "int4", "fp8", and "nvfp4".
-  group_size: # Optional[int] = 32. The number of elements in each group for per-group fake quantization
-  quantize_embedding: # Optional[bool] = False. Whether to quantize the embedding layer.
-
-output_dir:  # The path to the output directory.
-```
-
-Once quantization is complete, your quantized model will be saved in the `{output_dir}/quantized` directory.
-
-You may also use the `quantize` command to quantize a model which has been trained with [QAT](./qat.qmd) - you can do this by using the existing QAT configuration file which
-you used to train the model:
-
-```yaml
-# qat.yml
-qat:
-  activation_dtype: int8
-  weight_dtype: int4
-  group_size: 256
-
-output_dir: # The path to the output directory used during training where the final checkpoint has been saved.
-```
-
-```bash
-axolotl quantize qat.yml
-```
-
-This ensures that an identical quantization configuration is used to quantize the model as was used to train it.
-
-
-::: {.callout-note}
-
-If you have configured pushing to hub with `hub_model_id`, your model hub name will have the quantization schema appended to it,
-e.g. `axolotl-ai-cloud/qat-nvfp4-llama3B` will become `axolotl-ai-cloud/qat-nvfp4-llama3B-nvfp4w`
-
-:::
--- a/docs/reward_modelling.qmd
+++ b/docs/reward_modelling.qmd
@@ -11,7 +11,6 @@ We support the reward modelling techniques supported by `trl`.
 ### (Outcome) Reward Models

 Outcome reward models are trained using data which contains preference annotations for an entire interaction between the user and model (e.g. rather than per-turn or per-step).
-For improved training stability, you can use the `center_rewards_coefficient` parameter to encourage mean-zero reward outputs ([see TRL docs](https://huggingface.co/docs/trl/v0.10.1/en/reward_trainer#centering-rewards)).

 ```yaml
 base_model: google/gemma-2-2b
--- a/docs/rlhf.qmd
+++ b/docs/rlhf.qmd
@@ -16,7 +16,7 @@ feedback. Various methods include, but not limited to:
 - [Identity Preference Optimization (IPO)](#ipo)
 - [Kahneman-Tversky Optimization (KTO)](#kto)
 - [Odds Ratio Preference Optimization (ORPO)](#orpo)
- [Group Relative Policy Optimization (GRPO)](#grpo)
+- Proximal Policy Optimization (PPO) (not yet supported in axolotl)


 ## RLHF using Axolotl
@@ -219,21 +219,6 @@ DPO supports the following types with the following dataset format:
 }
 ```

-#### chat_template.argilla_chat
-
-```json
-{
-    "chosen": [
-        {"role": "user", "content": "..."},
-        {"role": "assistant", "content": "..."}
-    ],
-    "rejected": [
-        {"role": "user", "content": "..."},
-        {"role": "assistant", "content": "..."}
-    ]
-}
-```
-
 #### chat_template.default

 ```yaml
@@ -289,14 +274,15 @@ rl: dpo
 datasets:
  - path: ...
    split: train
-    type:
-      field_prompt: "prompt"
-      field_system: "system"
-      field_chosen: "chosen"
-      field_rejected: "rejected"
-      prompt_format: "{prompt}"
-      chosen_format: "{chosen}"
-      rejected_format: "{rejected}"
+    type: user_defined.default
+
+    field_prompt: "prompt"
+    field_system: "system"
+    field_chosen: "chosen"
+    field_rejected: "rejected"
+    prompt_format: "{prompt}"
+    chosen_format: "{chosen}"
+    rejected_format: "{rejected}"
 ```

 The input format is a simple JSON input with customizable fields based on the above config.
@@ -489,13 +475,14 @@ rl: kto
 datasets:
  - path: ...
    split: train
-    type:
-      field_prompt: "prompt"
-      field_system: "system"
-      field_completion: "completion"
-      field_label: "label"
-      prompt_format: "{prompt}"
-      completion_format: "{completion}"
+    type: user_defined.default
+
+    field_prompt: "prompt"
+    field_system: "system"
+    field_completion: "completion"
+    field_label: "label"
+    prompt_format: "{prompt}"
+    completion_format: "{completion}"
 ```

 The input format is a simple JSON input with customizable fields based on the above config.
@@ -512,7 +499,7 @@ The input format is a simple JSON input with customizable fields based on the ab
 ### GRPO

 ::: {.callout-tip}
-Check out our [GRPO cookbook](https://github.com/axolotl-ai-cloud/grpo_code).
+Check out our [GRPO cookbook](https://github.com/axolotl-ai-cloud/axolotl-cookbook/tree/main/grpo#training-an-r1-style-large-language-model-using-grpo).
 :::

 In the latest GRPO implementation, `vLLM` is used to significantly speedup trajectory generation during training. In this example, we're using 4 GPUs - 2 for training, and 2 for vLLM:
@@ -595,20 +582,7 @@ datasets:

 To see other examples of custom reward functions, please see [TRL GRPO Docs](https://github.com/huggingface/trl/blob/main/docs/source/grpo_trainer.md#using-a-custom-reward-function).

-To see all configs, please see [TRLConfig](https://github.com/axolotl-ai-cloud/axolotl/blob/v0.9.2/src/axolotl/utils/schemas/trl.py).
-
-#### GRPO with DAPO/Dr. GRPO loss
-
-The DAPO paper and subsequently Dr. GRPO paper proposed an alternative loss function for GRPO to remediate the penalty in longer responses.
-
-```yaml
-trl:
-  loss_type: dr_grpo
-  # Normalizes loss based on max completion length (default: 256)
-  max_completion_length:
-```
-
-For more information, see [GRPO docs](https://huggingface.co/docs/trl/v0.17.0/en/grpo_trainer#loss-types).
+To see description of the configs, please see [TRLConfig](https://github.com/axolotl-ai-cloud/axolotl/blob/main/src/axolotl/utils/config/models/input/v0_4_1/trl.py).

 ### SimPO

--- a/docs/scripts/generate_config_docs.py
+++ b/docs/scripts/generate_config_docs.py
@@ -1,749 +0,0 @@
-# type: ignore
-
-"""
-Quarto documentation generation from Pydantic models. Uses Pydantic model source code
-to automatically group fields, including inherited fields from parent classes.
-"""
-
-import ast
-import inspect
-import textwrap
-import types
-import typing
-from typing import Any, FrozenSet, Type, Union
-
-from pydantic import BaseModel
-
-from axolotl.utils.schemas.config import AxolotlInputConfig
-
-
-class QuartoGenerator:
-    """Generate Quarto documentation from Pydantic models."""
-
-    def __init__(self):
-        self._class_fields_cache = {}
-        self._inheritance_map_cache = {}
-        self._nested_models_cache = {}
-
-    def _get_direct_fields(self, cls: Type[BaseModel]) -> FrozenSet[str]:
-        """Get fields defined directly in a single class (not inherited)."""
-        if cls in self._class_fields_cache:
-            return self._class_fields_cache[cls]
-
-        fields = set()
-
-        # Get annotated fields
-        if hasattr(cls, "__annotations__"):
-            fields.update(cls.__annotations__.keys())
-
-        # Filter out private/special methods
-        fields = {f for f in fields if not f.startswith("_")}
-
-        result = frozenset(fields)
-        self._class_fields_cache[cls] = result
-        return result
-
-    def _is_pydantic_model(self, type_obj) -> bool:
-        """Check if a type is a Pydantic BaseModel."""
-        return inspect.isclass(type_obj) and issubclass(type_obj, BaseModel)
-
-    def _extract_nested_type(self, field_type) -> Any:
-        """Extract the actual type from complex type annotations."""
-        # Handle Annotated types (Python 3.9+)
-        if hasattr(typing, "get_origin") and hasattr(typing, "get_args"):
-            origin = typing.get_origin(field_type)
-            args = typing.get_args(field_type)
-
-            if origin is not None:
-                # Handle Annotated[SomeType, ...] - extract the first argument
-                if hasattr(typing, "Annotated") and origin is typing.Annotated:
-                    if args:
-                        return self._extract_nested_type(
-                            args[0]
-                        )  # Recursively process the actual type
-
-                # Handle list[SomeType], List[SomeType], etc.
-                elif origin in (list, typing.List):
-                    if args:
-                        return self._extract_nested_type(
-                            args[0]
-                        )  # Extract element type
-
-                # Handle Union types (including | syntax)
-                elif origin is typing.Union:
-                    # Get non-None types from the Union
-                    non_none_types = [arg for arg in args if arg is not type(None)]
-                    if len(non_none_types) >= 1:
-                        # Prioritize Pydantic models over primitive types
-                        pydantic_models = [
-                            arg
-                            for arg in non_none_types
-                            if self._is_pydantic_model(arg)
-                        ]
-                        if pydantic_models:
-                            # Return the first Pydantic model found
-                            return self._extract_nested_type(pydantic_models[0])
-
-                        # No Pydantic models, return the first non-None type
-                        return self._extract_nested_type(non_none_types[0])
-
-        # Handle new Python 3.10+ union syntax (PeftConfig | None)
-        if hasattr(field_type, "__class__") and field_type.__class__ is types.UnionType:
-            # Get non-None types from the Union
-            non_none_types = [
-                arg for arg in field_type.__args__ if arg is not type(None)
-            ]
-            if len(non_none_types) >= 1:
-                # Prioritize Pydantic models over primitive types
-                pydantic_models = [
-                    arg for arg in non_none_types if self._is_pydantic_model(arg)
-                ]
-                if pydantic_models:
-                    return self._extract_nested_type(pydantic_models[0])
-                return self._extract_nested_type(non_none_types[0])
-
-        # Handle old typing.Union syntax (fallback)
-        if hasattr(field_type, "__origin__"):
-            if field_type.__origin__ is Union:
-                # Get non-None types from the Union
-                non_none_types = [
-                    arg for arg in field_type.__args__ if arg is not type(None)
-                ]
-                if len(non_none_types) >= 1:
-                    # Prioritize Pydantic models over primitive types
-                    pydantic_models = [
-                        arg for arg in non_none_types if self._is_pydantic_model(arg)
-                    ]
-                    if pydantic_models:
-                        return self._extract_nested_type(pydantic_models[0])
-                    return self._extract_nested_type(non_none_types[0])
-            # Handle other generic types like dict[str, Any], etc.
-            elif hasattr(field_type, "__args__"):
-                return field_type
-
-        return field_type
-
-    def _extract_all_pydantic_models_from_type(
-        self, field_type
-    ) -> list[type[BaseModel]]:
-        """Extract all Pydantic models from a type annotation, including from Unions."""
-        models = []
-
-        if field_type is None:
-            return models
-
-        # Handle Annotated types
-        if hasattr(typing, "get_origin") and hasattr(typing, "get_args"):
-            origin = typing.get_origin(field_type)
-            args = typing.get_args(field_type)
-
-            if origin is not None:
-                # Handle Annotated[SomeType, ...] - extract from the first argument
-                if hasattr(typing, "Annotated") and origin is typing.Annotated:
-                    if args:
-                        models.extend(
-                            self._extract_all_pydantic_models_from_type(args[0])
-                        )
-                    return models
-
-                # Handle list[SomeType], List[SomeType], etc.
-                if origin in (list, typing.List):
-                    if args:
-                        models.extend(
-                            self._extract_all_pydantic_models_from_type(args[0])
-                        )
-                    return models
-
-                # Handle Union types
-                if origin is typing.Union:
-                    for arg in args:
-                        if arg is not type(None):  # Skip None type
-                            models.extend(
-                                self._extract_all_pydantic_models_from_type(arg)
-                            )
-                    return models
-
-        # Handle new Python 3.10+ union syntax
-        if hasattr(field_type, "__class__") and field_type.__class__ is types.UnionType:
-            for arg in field_type.__args__:
-                if arg is not type(None):  # Skip None type
-                    models.extend(self._extract_all_pydantic_models_from_type(arg))
-            return models
-
-        # Handle old typing.Union syntax (fallback)
-        if hasattr(field_type, "__origin__") and field_type.__origin__ is Union:
-            for arg in field_type.__args__:
-                if arg is not type(None):  # Skip None type
-                    models.extend(self._extract_all_pydantic_models_from_type(arg))
-            return models
-
-        # Check if this type itself is a Pydantic model
-        if self._is_pydantic_model(field_type):
-            models.append(field_type)
-
-        return models
-
-    def _get_nested_models(
-        self, model_class: type[BaseModel], visited=None
-    ) -> dict[str, type[BaseModel]]:
-        """Get all nested Pydantic models from a model class."""
-        if visited is None:
-            visited = set()
-
-        # Avoid infinite recursion
-        if model_class in visited:
-            return {}
-
-        if model_class in self._nested_models_cache:
-            return self._nested_models_cache[model_class]
-
-        visited.add(model_class)
-        nested_models = {}
-
-        # Check all fields in the model
-        for field_info in model_class.model_fields.values():
-            field_type = self._extract_nested_type(field_info.annotation)
-
-            if self._is_pydantic_model(field_type):
-                nested_models[field_type.__name__] = field_type
-                # Recursively get nested models from this nested model
-                deeper_nested = self._get_nested_models(field_type, visited.copy())
-                nested_models.update(deeper_nested)
-
-        self._nested_models_cache[model_class] = nested_models
-        return nested_models
-
-    def _build_inheritance_map(self, child_class: Type[BaseModel]):
-        """Build inheritance map for a class and all its parents."""
-        if child_class in self._inheritance_map_cache:
-            return self._inheritance_map_cache[child_class]
-
-        inheritance_map = {}
-
-        # Get MRO and filter out BaseModel and object
-        mro_classes = [
-            cls
-            for cls in child_class.__mro__
-            if cls not in (BaseModel, object) and hasattr(cls, "__annotations__")
-        ]
-
-        # Process each class in the MRO
-        for cls in mro_classes:
-            inheritance_map[cls] = self._get_direct_fields(cls)
-
-        self._inheritance_map_cache[child_class] = inheritance_map
-        return inheritance_map
-
-    def _wrap_comment(self, text: str, width: int = 88) -> list[str]:
-        """Wrap a comment to specified width, accounting for '# ' prefix."""
-        if not text.strip():
-            return ["#"]
-
-        # Account for "# " prefix (2 characters)
-        content_width = width - 2
-        wrapped_lines = textwrap.wrap(text, width=content_width)
-        return [f"# {line}" for line in wrapped_lines]
-
-    def _extract_type_from_source(
-        self, model_class: type[BaseModel], field_name: str
-    ) -> str:
-        """Extract the actual type annotation text from source code, checking inheritance chain."""
-        # Use inheritance map to check classes efficiently
-        inheritance_map = self._build_inheritance_map(model_class)
-
-        # Check classes in MRO order
-        for cls in model_class.__mro__:
-            if cls in inheritance_map and field_name in inheritance_map[cls]:
-                type_annotation = self._get_type_from_class_source(cls, field_name)
-                if type_annotation != "unknown":
-                    return type_annotation
-
-        return "unknown"
-
-    def _get_type_from_class_source(self, class_obj: type, field_name: str) -> str:
-        """Extract type annotation from a specific class's source code."""
-        try:
-            source = inspect.getsource(class_obj)
-            tree = ast.parse(source)
-        except (OSError, TypeError):
-            return "unknown"
-
-        # Find the class definition
-        for node in tree.body:
-            if isinstance(node, ast.ClassDef) and node.name == class_obj.__name__:
-                # Find the field assignment
-                for body_node in node.body:
-                    if isinstance(body_node, ast.AnnAssign) and isinstance(
-                        body_node.target, ast.Name
-                    ):
-                        if body_node.target.id == field_name and body_node.annotation:
-                            return ast.unparse(body_node.annotation)
-                break
-
-        return "unknown"
-
-    def _extract_field_groups_from_all_classes(
-        self, model_class: type[BaseModel]
-    ) -> list[dict]:
-        """Extract field groups from all classes in the inheritance hierarchy."""
-        all_groups = []
-        inheritance_map = self._build_inheritance_map(model_class)
-
-        # Get all Pydantic base classes in MRO order (most specific first)
-        # This puts AxolotlInputConfig fields first, then parent class fields
-        pydantic_classes = [
-            cls
-            for cls in model_class.__mro__
-            if cls in inheritance_map and inheritance_map[cls]
-        ]
-
-        # Extract groups from each class
-        for cls in pydantic_classes:
-            class_groups = self._extract_field_groups_from_source(cls)
-            for group in class_groups:
-                all_groups.append(group)
-
-        # If no groups found, create a default grouping by class
-        if not all_groups:
-            for cls in pydantic_classes:
-                fields_in_class = inheritance_map[cls]
-                if fields_in_class:
-                    all_groups.append(
-                        {
-                            "fields": list(fields_in_class),
-                        }
-                    )
-
-        return all_groups
-
-    def _extract_field_groups_from_source(
-        self, model_class: type[BaseModel]
-    ) -> list[dict]:
-        """Extract field groups from source code based on blank lines and comments."""
-        try:
-            source = inspect.getsource(model_class)
-            tree = ast.parse(source)
-        except (OSError, TypeError):
-            # Fallback if we can't get source code
-            fields_in_class = self._get_direct_fields(model_class)
-            if fields_in_class:
-                return [
-                    {
-                        "fields": list(fields_in_class),
-                    }
-                ]
-            return []
-
-        groups = []
-        current_group_fields = []
-        current_group_comment = None
-
-        # Find the class definition
-        class_node = None
-        for node in ast.walk(tree):
-            if isinstance(node, ast.ClassDef) and node.name == model_class.__name__:
-                class_node = node
-                break
-
-        if not class_node:
-            fields_in_class = self._get_direct_fields(model_class)
-            if fields_in_class:
-                return [
-                    {
-                        "fields": list(fields_in_class),
-                    }
-                ]
-            return []
-
-        # Parse the source lines to detect groupings
-        source_lines = source.split("\n")
-
-        # Get fields that are actually defined in this specific class
-        fields_in_class = self._get_direct_fields(model_class)
-
-        # Find assignments that correspond to model fields for THIS class only
-        field_assignments = []
-        for node in class_node.body:
-            if isinstance(node, ast.AnnAssign) and isinstance(node.target, ast.Name):
-                field_name = node.target.id
-                if field_name in fields_in_class:
-                    field_assignments.append(
-                        {
-                            "name": field_name,
-                            "lineno": node.lineno,
-                            "end_lineno": getattr(node, "end_lineno", node.lineno),
-                        }
-                    )
-
-        if not field_assignments:
-            if fields_in_class:
-                return [
-                    {
-                        "fields": list(fields_in_class),
-                    }
-                ]
-            return []
-
-        # Sort by line number
-        field_assignments.sort(key=lambda x: x["lineno"])
-
-        # Group fields based on blank lines and comments
-        for i, field_info in enumerate(field_assignments):
-            field_name = field_info["name"]
-            current_line = field_info["lineno"]
-
-            # Check if this starts a new group (blank line before or significant gap)
-            is_new_group = False
-
-            if i == 0:
-                is_new_group = True
-            else:
-                prev_end_line = field_assignments[i - 1]["end_lineno"]
-
-                # Check for blank lines or comments between fields
-                lines_between = source_lines[prev_end_line : current_line - 1]
-                has_blank_line = any(line.strip() == "" for line in lines_between)
-                has_comment = any(
-                    line.strip().startswith("#") for line in lines_between
-                )
-
-                # Start new group if there's a blank line or comment, or significant gap
-                if has_blank_line or has_comment or (current_line - prev_end_line > 3):
-                    is_new_group = True
-
-            if is_new_group and current_group_fields:
-                # Save the previous group
-                groups.append(
-                    {
-                        "fields": current_group_fields.copy(),
-                        "description": current_group_comment,
-                    }
-                )
-                current_group_fields = []
-                current_group_comment = None
-
-            current_group_fields.append(field_name)
-
-        # Add the final group
-        if current_group_fields:
-            groups.append(
-                {
-                    "fields": current_group_fields,
-                    "description": current_group_comment,
-                }
-            )
-
-        return groups
-
-    def _generate_field_documentation(
-        self,
-        model_class: type[BaseModel],
-        field_name: str,
-        field_info: dict,
-        field_type_str: str,
-        is_required: bool,
-        indent_level: int = 0,
-        visited_models: set = None,
-    ) -> list[str]:
-        """Generate documentation for a single field, expanding nested models inline."""
-        if visited_models is None:
-            visited_models = set()
-
-        lines = []
-        indent = "  " * indent_level
-
-        # Get the actual field type for nested model detection
-        if field_name in model_class.model_fields:
-            pydantic_field_info = model_class.model_fields[field_name]
-            actual_field_type = pydantic_field_info.annotation
-        else:
-            actual_field_type = None
-
-        # Add description comment if available
-        description = field_info.get("description", "")
-        if description:
-            wrapped_lines = self._wrap_comment(description, width=88 - len(indent))
-            for line in wrapped_lines:
-                lines.append(f"{indent}{line}")
-
-        # Extract nested Pydantic models from the type annotation
-        nested_models = self._extract_all_pydantic_models_from_type(actual_field_type)
-
-        # Filter out already visited models to prevent infinite recursion
-        expandable_models = [
-            model for model in nested_models if model not in visited_models
-        ]
-
-        if expandable_models:
-            # This field contains Pydantic models that can be expanded
-
-            # Show the field with its full type annotation
-            field_line = f"{indent}{field_name}: {field_type_str}"
-            if field_info.get("default") is not None:
-                field_line += f" = {field_info['default']}"
-            if is_required:
-                field_line += " (required)"
-            lines.append(field_line)
-
-            # Add to visited to prevent infinite recursion
-            new_visited = visited_models.copy()
-            new_visited.update(expandable_models)
-
-            # Expand each nested Pydantic model
-            for i, nested_model in enumerate(expandable_models):
-                if i > 0:
-                    lines.append("\n")
-                lines.append(f"{indent}  # For {nested_model.__name__}:")
-
-                # Get nested model schema
-                try:
-                    nested_schema = nested_model.model_json_schema()
-                    nested_properties = nested_schema.get("properties", {})
-                    nested_required = nested_schema.get("required", [])
-                except Exception:
-                    # Fallback: use model fields directly
-                    nested_properties = {}
-                    nested_required = []
-                    for (
-                        nested_field_name,
-                        nested_field_info,
-                    ) in nested_model.model_fields.items():
-                        nested_description = ""
-                        if (
-                            hasattr(nested_field_info, "json_schema_extra")
-                            and nested_field_info.json_schema_extra
-                        ):
-                            nested_description = (
-                                nested_field_info.json_schema_extra.get(
-                                    "description", ""
-                                )
-                            )
-                        elif (
-                            hasattr(nested_field_info, "description")
-                            and nested_field_info.description
-                        ):
-                            nested_description = nested_field_info.description
-
-                        nested_default_val = None
-                        if (
-                            hasattr(nested_field_info, "default")
-                            and nested_field_info.default is not None
-                        ):
-                            if str(nested_field_info.default) != "PydanticUndefined":
-                                nested_default_val = nested_field_info.default
-
-                        nested_properties[nested_field_name] = {
-                            "type": "unknown",
-                            "description": nested_description,
-                            "default": nested_default_val,
-                        }
-
-                        if nested_field_info.is_required():
-                            nested_required.append(nested_field_name)
-
-                # Get field groups for the nested model
-                nested_field_groups = self._extract_field_groups_from_all_classes(
-                    nested_model
-                )
-
-                # Generate nested fields with increased indentation
-                for i, group in enumerate(nested_field_groups):
-                    if not group["fields"]:
-                        continue
-
-                    # Add blank line between groups (except before first group)
-                    if i > 0:
-                        lines.append("")
-
-                    # Process nested fields
-                    for nested_field_name in group["fields"]:
-                        if nested_field_name not in nested_properties:
-                            continue
-
-                        nested_field_info = nested_properties[nested_field_name]
-                        nested_field_type = self._extract_type_from_source(
-                            nested_model, nested_field_name
-                        )
-                        nested_is_required = nested_field_name in nested_required
-
-                        # Recursively generate documentation for nested field
-                        nested_lines = self._generate_field_documentation(
-                            nested_model,
-                            nested_field_name,
-                            nested_field_info,
-                            nested_field_type,
-                            nested_is_required,
-                            indent_level + 1,
-                            new_visited,
-                        )
-                        lines.extend(nested_lines)
-        else:
-            # Regular field (no expandable nested models)
-            field_line = f"{indent}{field_name}: {field_type_str}"
-            if field_info.get("default") is not None:
-                field_line += f" = {field_info['default']}"
-            if is_required:
-                field_line += " (required)"
-            lines.append(field_line)
-
-        return lines
-
-    def generate_qmd(
-        self,
-        model_class: type[BaseModel],
-        title: str | None = None,
-        expand_nested: bool = True,
-    ) -> str:
-        """Auto-generate config reference documentation including inherited fields."""
-
-        if title is None:
-            title = f"{model_class.__name__} Reference"
-
-        # Try to get JSON schema, with fallback for serialization issues
-        try:
-            schema = model_class.model_json_schema()
-            properties = schema.get("properties", {})
-            required = schema.get("required", [])
-        except Exception as e:
-            print(
-                f"Warning: Could not generate JSON schema ({e}). Using model fields instead."
-            )
-            # Fallback: use model fields directly
-            properties = {}
-            required = []
-            for field_name, field_info in model_class.model_fields.items():
-                # Extract description from json_schema_extra or field info
-                description = ""
-                if (
-                    hasattr(field_info, "json_schema_extra")
-                    and field_info.json_schema_extra
-                ):
-                    description = field_info.json_schema_extra.get("description", "")
-                elif hasattr(field_info, "description") and field_info.description:
-                    description = field_info.description
-
-                # Get default value
-                default_val = None
-                if hasattr(field_info, "default") and field_info.default is not None:
-                    # Handle special Pydantic default markers
-                    if str(field_info.default) != "PydanticUndefined":
-                        default_val = field_info.default
-
-                properties[field_name] = {
-                    "type": "unknown",
-                    "description": description,
-                    "default": default_val,
-                }
-
-                if field_info.is_required():
-                    required.append(field_name)
-
-        # Extract field groups from all classes in inheritance hierarchy
-        field_groups = self._extract_field_groups_from_all_classes(model_class)
-
-        # Start building QMD content
-        qmd_lines = [
-            "---",
-            f"title: {title}",
-            "description: A complete list of all configuration options.",
-            "---",
-            "",
-        ]
-
-        # Generate one big code block with all fields (inline nested expansion)
-        qmd_lines.append("```yaml")
-
-        for i, group in enumerate(field_groups):
-            if not group["fields"]:
-                continue
-
-            # Add blank line between groups (except before first group)
-            if i > 0:
-                qmd_lines.append("")
-
-            # Process fields in the order they appear in source
-            for field_name in group["fields"]:
-                if field_name not in properties:
-                    continue
-
-                field_info = properties[field_name]
-                field_type = self._extract_type_from_source(model_class, field_name)
-                is_required = field_name in required
-
-                if expand_nested:
-                    # Check if this field has nested models
-                    if field_name in model_class.model_fields:
-                        pydantic_field_info = model_class.model_fields[field_name]
-                        nested_models = self._extract_all_pydantic_models_from_type(
-                            pydantic_field_info.annotation
-                        )
-                        has_nested = bool(nested_models)
-                    else:
-                        has_nested = False
-
-                    # Add blank line before nested config
-                    if has_nested:
-                        qmd_lines.append("")
-
-                    # Use the new inline generation method
-                    field_lines = self._generate_field_documentation(
-                        model_class,
-                        field_name,
-                        field_info,
-                        field_type,
-                        is_required,
-                        indent_level=0,
-                        visited_models=set(),
-                    )
-                    qmd_lines.extend(field_lines)
-
-                    # Add blank line after nested config
-                    if has_nested:
-                        qmd_lines.append("")
-                else:
-                    # Original simple approach
-                    description = field_info.get("description", "")
-                    default = field_info.get("default")
-
-                    # Add wrapped comment for description
-                    if description:
-                        wrapped_lines = self._wrap_comment(description)
-                        qmd_lines.extend(wrapped_lines)
-
-                    line = f"{field_name}: {field_type}"
-                    if default is not None:
-                        line += f" = {default}"
-                    if is_required:
-                        line += " (required)"
-                    qmd_lines.append(line)
-
-        qmd_lines.append("```")
-
-        # Join all lines and clean up any double newlines
-        content = "\n".join(qmd_lines)
-
-        # Replace multiple consecutive newlines with just two newlines (one blank line)
-        import re
-
-        content = re.sub(r"\n{3,}", "\n\n", content)
-
-        # Ensure single newline at the very end
-        content = content.rstrip("\n") + "\n"
-
-        return content
-
-
-def main():
-    generator = QuartoGenerator()
-
-    print("Generating config reference content...")
-    qmd_content = generator.generate_qmd(AxolotlInputConfig, "Config Reference", True)
-
-    print("Writing to file...")
-    with open("docs/config-reference.qmd", "w", encoding="utf-8") as f:
-        f.write(qmd_content)
-    print("Done!")
-
-
-if __name__ == "__main__":
-    main()
--- a/docs/sequence_parallelism.qmd
+++ b/docs/sequence_parallelism.qmd
@@ -3,6 +3,8 @@ title: Sequence Parallelism
 description: Train with long sequences split across multiple GPUs.
 ---

+# Sequence Parallelism
+
 Sequence parallelism is a technique that splits sequences across multiple GPUs,
 allowing you to train with very long sequences that wouldn't fit on a single GPU. Each
 GPU processes a different portion of the sequence, and the results are aggregated
@@ -22,15 +24,15 @@ To enable sequence parallelism, add the following to your configuration file:

 ```yaml
 # Set to a divisor (> 1) of the number of GPUs available
-context_parallel_size: 4  # Split sequences across 4 GPUs
+sequence_parallel_degree: 4  # Split sequences across 4 GPUs
 # Optional; strides across the key dimension. Larger values use more memory but should make training faster.
 heads_k_stride: 1
-# Optional; one of "varlen_llama3" or "batch_ring". Defaults to
+# Optional; one of "varlen_llama3", "batch_ring", "batch_zigzag", "batch_stripe". Defaults to
 # "varlen_llama3" when `sample_packing: true`, and "batch_ring" otherwise.
 ring_attn_func:
 ```

-The `context_parallel_size` should be a divisor of the total number of GPUs. For example:
+The `sequence_parallel_degree` should be a divisor of the total number of GPUs. For example:

 - With 8 GPUs, valid values would be 2, 4, or 8
 - With 4 GPUs, valid values would be 2 or 4
@@ -41,7 +43,7 @@ When sequence parallelism is enabled:

 1. Each sequence is divided into equal chunks across the GPUs in a sequence parallel group
 2. The data collator handles the chunking of input_ids, attention_mask, labels, and position_ids
-3. Position IDs are adjusted to maintain proper relative positions
+3. Position IDs are adjusted to maintain proper relative positions, especially for packed sequences
 4. The trainer uses special ring communication patterns for attention operations

 ## Requirements
@@ -66,12 +68,10 @@ sequence_len: 8192

 ...

-context_parallel_size: 4  # Split each sequence into 4 parts, one per GPU
+sequence_parallel_degree: 4  # Split each sequence into 4 parts, one per GPU
+flash_attention: true  # Required with sequence parallelism
 # Optional; strides across the key dimension. Larger values use more memory but should make training faster.
 heads_k_stride: 1
-# Optional; one of "varlen_llama3" or "batch_ring". Defaults to
-# "varlen_llama3" when `sample_packing: true`, and "batch_ring" otherwise.
-ring_attn_func:

 ...
 ```
@@ -89,12 +89,12 @@ Sequence parallelism is compatible with Axolotl's sample packing functionality.

 ## Effect on Batch Size

-When using sequence parallelism, your effective global batch size is **divided** by the `context_parallel_size`. This happens because:
+When using sequence parallelism, your effective global batch size is **divided** by the `sequence_parallel_degree`. This happens because:

- Each group of `context_parallel_size` GPUs works on the same batch (just different parts of each sequence)
+- Each group of `sequence_parallel_degree` GPUs works on the same batch (just different parts of each sequence)
 - The number of batches processed per step decreases

 For example:
 - With 8 GPUs and no sequence parallelism: 8 different batches processed per step
- With 8 GPUs and `context_parallel_size=4`: Only 2 different batches processed per step (each split across 4 GPUs)
+- With 8 GPUs and `sequence_parallel_degree=4`: Only 2 different batches processed per step (each split across 4 GPUs)
 - If your per-GPU `micro_batch_size` is 2, the global batch size decreases from 16 to 4
--- a/docs/streaming.qmd
+++ b/docs/streaming.qmd
@@ -1,120 +0,0 @@
---
-title: Streaming Datasets
-description: How to use streaming mode for large-scale datasets and memory-efficient training
-order: 10
---
-
-Streaming enables memory-efficient training with large datasets by loading data
-incrementally rather than loading the entire dataset into memory at once.
-
-Use streaming when:
-
- Your dataset is too large to fit in memory (e.g. when you're doing pretraining with massive text corpora)
- You want to start training immediately without preprocessing the entire dataset
-
-Streaming works with both remote and locally stored datasets!
-
-::: {.callout-note}
-Streaming currently only supports a single dataset. Multi-dataset support will be added soon.
-:::
-
-
-## Configuration
-
-### Basic Streaming
-
-Enable streaming mode by setting the `streaming` flag:
-
-```yaml
-streaming: true
-```
-
-### Pretraining with Streaming
-
-For pretraining tasks, streaming is automatically enabled when using `pretraining_dataset`:
-
-```yaml
-pretraining_dataset:
-  - path: HuggingFaceFW/fineweb-edu
-    type: pretrain
-    text_column: text
-    split: train
-
-# Optionally, enable sample packing
-streaming_multipack_buffer_size: 10000
-sample_packing: true
-```
-
-### SFT with Streaming
-
-For supervised fine-tuning with streaming:
-
-```yaml
-streaming: true
-datasets:
-  - path: tatsu-lab/alpaca
-    type: alpaca
-    split: train
-
-# Optionally, enable sample packing
-streaming_multipack_buffer_size: 10000
-sample_packing: true
-```
-
-## Configuration Options
-
-### `streaming_multipack_buffer_size`
-
-Controls the buffer size for multipack streaming (default: 10,000). This determines how
-many samples are buffered before packing. Larger buffers can improve packing efficiency
-but use more memory.
-
-### `shuffle_merged_datasets`
-
-When enabled, shuffles the streaming dataset using the buffer. This requires additional
-memory for the shuffle buffer.
-
-## Sample Packing with Streaming
-
-Sample packing is supported for streaming datasets. When enabled, multiple samples are
-packed into a single sequence to maximize GPU utilization:
-
-```yaml
-sample_packing: true
-streaming_multipack_buffer_size: 10000
-
-# For SFT: attention is automatically isolated between packed samples
-# For pretraining: control with pretrain_multipack_attn
-pretrain_multipack_attn: true  # prevent cross-attention between packed samples
-```
-
-For more information, see our [documentation](multipack.qmd) on multipacking.
-
-## Important Considerations
-
-### Memory Usage
-
-While streaming reduces memory usage compared to loading entire datasets, you still need
-to consider:
-
- You can control the memory usage by adjusting `streaming_multipack_buffer_size`
- Sample packing requires buffering multiple samples
- Shuffling requires additional memory for the shuffle buffer
-
-### Performance
-
- Streaming may have slightly higher latency compared to preprocessed datasets, as samples are processed on-the-fly
- Network speed and disk read speed are important when streaming from remote sources or a local dataset, respectively
- Consider using `axolotl preprocess` for smaller or more frequently used datasets
-
-### Evaluation Datasets
-
-Evaluation datasets are not streamed to ensure consistent evaluation metrics. They're
-loaded normally even when training uses streaming.
-
-## Examples
-
-See the `examples/streaming/` directory for complete configuration examples:
-
- `pretrain.yaml`: Pretraining with streaming dataset
- `sft.yaml`: Supervised fine-tuning with streaming
--- a/examples/LiquidAI/README.md
+++ b/examples/LiquidAI/README.md
@@ -1,67 +0,0 @@
-# Finetune Liquid Foundation Models 2 (LFM2) with Axolotl
-
-[Liquid Foundation Models 2 (LFM2)](https://huggingface.co/collections/LiquidAI/lfm2-686d721927015b2ad73eaa38) are a family of small, open-weight models from [Liquid AI](https://www.liquid.ai/) focused on quality, speed, and memory efficiency. Liquid AI released text-only [LFM2](https://huggingface.co/collections/LiquidAI/lfm2-686d721927015b2ad73eaa38) and text+vision [LFM2-VL](https://huggingface.co/collections/LiquidAI/lfm2-vl-68963bbc84a610f7638d5ffa) models.
-
-LFM2 features a new hybrid Liquid architecture with multiplicative gates, short-range convolutions, and grouped query attention, enabling fast training and inference.
-
-This guide shows how to fine-tune both the LFM2 and LFM2-VL models with Axolotl.
-
-Thanks to the team at LiquidAI for giving us early access to prepare for these releases.
-
-## Getting Started
-
-1.  Install Axolotl following the [installation guide](https://docs.axolotl.ai/docs/installation.html).
-
-    Here is an example of how to install from pip:
-    ```bash
-    # Ensure you have a compatible version of Pytorch installed
-    pip3 install packaging setuptools wheel ninja
-    pip3 install --no-build-isolation 'axolotl[flash-attn]>=0.12.0'
-    ```
-
-2.  Run one of the finetuning examples below.
-
-    **LFM2**
-    ```bash
-    # FFT SFT (1x48GB @ 25GiB)
-    axolotl train examples/LiquidAI/lfm2-350m-fft.yaml
-    ```
-
-    **LFM2-VL**
-    ```bash
-    # LoRA SFT (1x48GB @ 2.7GiB)
-    axolotl train examples/LiquidAI/lfm2-vl-lora.yaml
-    ```
-
-    **LFM2-MoE**
-    ```bash
-    pip install git+https://github.com/huggingface/transformers.git@0c9a72e4576fe4c84077f066e585129c97bfd4e6
-
-    # LoRA SFT (1x48GB @ 16.2GiB)
-    axolotl train examples/LiquidAI/lfm2-8b-a1b-lora.yaml
-    ```
-
-### TIPS
-
- **Installation Error**: If you encounter `ImportError: ... undefined symbol ...` or `ModuleNotFoundError: No module named 'causal_conv1d_cuda'`, the `causal-conv1d` package may have been installed incorrectly. Try uninstalling it:
-  ```bash
-  pip uninstall -y causal-conv1d
-  ```
-
- **Dataset Loading**: Read more on how to load your own dataset in our [documentation](https://docs.axolotl.ai/docs/dataset_loading.html).
- **Dataset Formats**:
-  - For LFM2 models, the dataset format follows the OpenAI Messages format as seen [here](https://docs.axolotl.ai/docs/dataset-formats/conversation.html#chat_template).
-  - For LFM2-VL models, Axolotl follows the multi-content Messages format. See our [Multimodal docs](https://docs.axolotl.ai/docs/multimodal.html#dataset-format) for details.
-
-## Optimization Guides
-
- [Optimizations Guide](https://docs.axolotl.ai/docs/optimizations.html)
-
-## Related Resources
-
- [LFM2 Blog](https://www.liquid.ai/blog/liquid-foundation-models-v2-our-second-series-of-generative-ai-models)
- [LFM2-VL Blog](https://www.liquid.ai/blog/lfm2-vl-efficient-vision-language-models)
- [LFM2-MoE Blog](https://www.liquid.ai/blog/lfm2-8b-a1b-an-efficient-on-device-mixture-of-experts)
- [Axolotl Docs](https://docs.axolotl.ai)
- [Axolotl GitHub](https://github.com/axolotl-ai-cloud/axolotl)
- [Axolotl Discord](https://discord.gg/7m9sfhzaf3)
--- a/examples/LiquidAI/lfm2-350m-fft.yaml
+++ b/examples/LiquidAI/lfm2-350m-fft.yaml
@@ -1,50 +0,0 @@
-base_model: LiquidAI/LFM2-350M
-
-plugins:
-  - axolotl.integrations.cut_cross_entropy.CutCrossEntropyPlugin
-
-eot_tokens:
-  - "<|im_end|>"
-datasets:
-  - path: mlabonne/FineTome-100k
-    type: chat_template
-    split: train[:20%]
-    field_messages: conversations
-    message_field_role: from
-    message_field_content: value
-dataset_prepared_path: last_run_prepared
-val_set_size: 0.05
-output_dir: ./outputs/out
-
-sequence_len: 4096
-sample_packing: true
-
-
-wandb_project:
-wandb_entity:
-wandb_watch:
-wandb_name:
-wandb_log_model:
-
-gradient_accumulation_steps: 2
-micro_batch_size: 4
-num_epochs: 1
-optimizer: adamw_torch_fused
-lr_scheduler: cosine
-learning_rate: 5e-5
-
-bf16: true
-tf32: true
-
-gradient_checkpointing: false
-resume_from_checkpoint:
-logging_steps: 1
-flash_attention: true
-
-warmup_ratio: 0.1
-evals_per_epoch: 2
-saves_per_epoch: 1
-
-weight_decay: 0.0
-
-# save_first_step: true  # uncomment this to validate checkpoint saving works with your config
--- a/examples/LiquidAI/lfm2-8b-a1b-lora.yaml
+++ b/examples/LiquidAI/lfm2-8b-a1b-lora.yaml
@@ -1,59 +0,0 @@
-base_model: LiquidAI/LFM2-8B-A1B
-
-plugins:
-  - axolotl.integrations.cut_cross_entropy.CutCrossEntropyPlugin
-
-load_in_8bit: true
-
-eot_tokens:
-  - "<|im_end|>"
-datasets:
-  - path: mlabonne/FineTome-100k
-    type: chat_template
-    split: train[:20%]
-    field_messages: conversations
-    message_field_role: from
-    message_field_content: value
-dataset_prepared_path: last_run_prepared
-val_set_size: 0.05
-output_dir: ./outputs/out
-
-sequence_len: 4096
-sample_packing: true
-
-adapter: lora
-lora_model_dir:
-
-lora_r: 32
-lora_alpha: 16
-lora_dropout: 0.05
-lora_target_modules: 'model.layers.[\d]+.(mlp|cross_attn|self_attn).(up|down|gate|q|k|v|o)_proj'
-
-wandb_project:
-wandb_entity:
-wandb_watch:
-wandb_name:
-wandb_log_model:
-
-gradient_accumulation_steps: 2
-micro_batch_size: 4
-num_epochs: 1
-optimizer: adamw_torch_fused
-lr_scheduler: cosine
-learning_rate: 5e-5
-
-bf16: true
-tf32: true
-
-gradient_checkpointing: true
-resume_from_checkpoint:
-logging_steps: 1
-flash_attention: true
-
-warmup_ratio: 0.1
-evals_per_epoch: 2
-saves_per_epoch: 1
-
-weight_decay: 0.0
-
-# save_first_step: true  # uncomment this to validate checkpoint saving works with your config
--- a/examples/LiquidAI/lfm2-vl-lora.yaml
+++ b/examples/LiquidAI/lfm2-vl-lora.yaml
@@ -1,61 +0,0 @@
-base_model: LiquidAI/LFM2-VL-450M
-trust_remote_code: true
-model_type: AutoModelForImageTextToText
-processor_type: AutoProcessor
-
-plugins:
-  - axolotl.integrations.cut_cross_entropy.CutCrossEntropyPlugin
-
-# these 3 lines are needed for now to handle vision chat templates w images
-skip_prepare_dataset: true
-remove_unused_columns: false
-sample_packing: false
-
-datasets:
-  - path: HuggingFaceH4/llava-instruct-mix-vsft
-    type: chat_template
-    split: train[:1%]
-
-dataset_prepared_path: last_run_prepared
-val_set_size: 0.0
-output_dir: ./outputs/out
-
-adapter: lora
-lora_model_dir:
-
-sequence_len: 8192
-pad_to_sequence_len: false
-
-lora_r: 32
-lora_alpha: 16
-lora_dropout: 0.05
-lora_target_modules: 'model.language_model.layers.[\d]+.(mlp|cross_attn|self_attn).(up|down|gate|q|k|v|o)_proj'
-
-wandb_project:
-wandb_entity:
-wandb_watch:
-wandb_name:
-wandb_log_model:
-
-gradient_accumulation_steps: 4
-micro_batch_size: 1
-num_epochs: 1
-optimizer: adamw_bnb_8bit
-lr_scheduler: cosine
-learning_rate: 0.0002
-
-bf16: true
-fp16:
-tf32: true
-
-gradient_checkpointing: true
-logging_steps: 1
-flash_attention: true
-eager_attention:
-
-warmup_ratio: 0.1
-evals_per_epoch: 1
-saves_per_epoch: 1
-weight_decay: 0.0
-
-# save_first_step: true  # uncomment this to validate checkpoint saving works with your config
--- a/examples/alst/README.md
+++ b/examples/alst/README.md
@@ -1,30 +0,0 @@
-# Arctic Long Sequence Training (ALST)
-
-Artic Long Sequence Training (ALST) is a technique for training long context models using a variety of optimization
-techniques. It is a combination of:
- TiledMLP: Leverage tiling over the sequence dimension on MLP layers to reduce memory usage
- Tiled Loss: Using optimized loss functions like Liger-Kernel or Cut Cross Entropy to reduce memory usage
- Activation Offloading: Offload activations to CPU RAM to reduce memory usage
-
-For more information, you can check out the ALST paper [here](https://www.arxiv.org/abs/2506.13996).
-
-## Usage
-
-```yaml
-tiled_mlp: true
-
-# See Sequence Parallelism docs
-# https://docs.axolotl.ai/docs/sequence_parallelism.html
-context_parallel_size: int
-
-plugins:
-# See Cut Cross Entropy docs
-# https://docs.axolotl.ai/docs/custom_integrations.html#cut-cross-entropy
-  - axolotl.integrations.cut_cross_entropy.CutCrossEntropyPlugin
-
-# or Liger Kernel docs
-# https://docs.axolotl.ai/docs/custom_integrations.html#liger-kernels
-  - axolotl.integrations.liger.LigerPlugin
-# ...
-
-```
--- a/examples/alst/llama3-8b-deepspeed-alst.yaml
+++ b/examples/alst/llama3-8b-deepspeed-alst.yaml
@@ -1,53 +0,0 @@
-base_model: meta-llama/Llama-3.1-8B
-# Automatically upload checkpoint and final model to HF
-# hub_model_id: username/custom_model_name
-
-datasets:
-  - path: togethercomputer/Long-Data-Collections
-    type: completion
-    field: text
-    data_files:
-      - pretrain/rp_sub.jsonl.zst
-  - path: princeton-nlp/TextbookChapters
-    type: completion
-    field: chapter
-dataset_prepared_path: last_run_prepared
-val_set_size: 0.0
-output_dir: ./outputs/out
-
-sequence_len: 500_000
-min_sample_len: 200_000
-sample_packing: true
-
-tiled_mlp: true
-context_parallel_size: 8
-plugins:
-  - axolotl.integrations.cut_cross_entropy.CutCrossEntropyPlugin
-
-gradient_accumulation_steps: 1
-micro_batch_size: 1
-num_epochs: 1
-optimizer: adamw_torch_8bit
-lr_scheduler: cosine
-learning_rate: 2e-5
-
-bf16: auto
-tf32: true
-
-gradient_checkpointing: true
-activation_offloading: legacy
-
-resume_from_checkpoint:
-logging_steps: 1
-flash_attention: true
-
-warmup_steps: 100
-saves_per_epoch: 1
-evals_per_epoch: 2
-weight_decay: 0.0
-special_tokens:
-  pad_token: <|end_of_text|>
-
-deepspeed: deepspeed_configs/zero3_bf16_cpuoffload_all.json
-
-# save_first_step: true  # uncomment this to validate checkpoint saving works with your config
--- a/examples/alst/llama3-8b-fsdp2-alst.yaml
+++ b/examples/alst/llama3-8b-fsdp2-alst.yaml
@@ -1,59 +0,0 @@
-base_model: meta-llama/Llama-3.1-8B
-# Automatically upload checkpoint and final model to HF
-# hub_model_id: username/custom_model_name
-
-datasets:
-  - path: togethercomputer/Long-Data-Collections
-    type: completion
-    field: text
-    data_files:
-      - pretrain/rp_sub.jsonl.zst
-  - path: princeton-nlp/TextbookChapters
-    type: completion
-    field: chapter
-dataset_prepared_path: last_run_prepared
-val_set_size: 0.0
-output_dir: ./outputs/out
-
-sequence_len: 500_000
-min_sample_len: 200_000
-sample_packing: true
-
-tiled_mlp: true
-context_parallel_size: 8
-plugins:
-  - axolotl.integrations.cut_cross_entropy.CutCrossEntropyPlugin
-
-gradient_accumulation_steps: 1
-micro_batch_size: 1
-num_epochs: 1
-optimizer: adamw_torch_8bit
-lr_scheduler: cosine
-learning_rate: 2e-5
-
-bf16: auto
-tf32: true
-
-gradient_checkpointing: true
-activation_offloading: legacy
-
-resume_from_checkpoint:
-logging_steps: 1
-flash_attention: true
-
-warmup_steps: 100
-saves_per_epoch: 1
-evals_per_epoch: 2
-weight_decay: 0.0
-special_tokens:
-  pad_token: <|end_of_text|>
-
-fsdp_version: 2
-fsdp_config:
-  offload_params: false  # offloading is currently not compatible with SP + torchao optimizer
-  state_dict_type: SHARDED_STATE_DICT
-  auto_wrap_policy: TRANSFORMER_BASED_WRAP
-  transformer_layer_cls_to_wrap: LlamaDecoderLayer
-  reshard_after_forward: true
-
-# save_first_step: true  # uncomment this to validate checkpoint saving works with your config
--- a/examples/apertus/README.md
+++ b/examples/apertus/README.md
@@ -1,110 +0,0 @@
-# Finetune Swiss-AI's Apertus with Axolotl
-
-[Apertus](https://huggingface.co/collections/swiss-ai/apertus-llm-68b699e65415c231ace3b059) is a family of opensource models trained by Swiss-ai.
-
-This guide shows how to fine-tune it with Axolotl with multi-turn conversations and proper masking.
-
-## Getting started
-
-1. Install Axolotl following the [installation guide](https://docs.axolotl.ai/docs/installation.html). You need to install from main as Apertus is only on nightly or use our latest [Docker images](https://docs.axolotl.ai/docs/docker.html).
-
-    Here is an example of how to install from main for pip:
-
-```bash
-# Ensure you have Pytorch installed (Pytorch 2.6.0 min)
-git clone https://github.com/axolotl-ai-cloud/axolotl.git
-cd axolotl
-
-pip3 install packaging==23.2 setuptools==75.8.0 wheel ninja
-pip3 install --no-build-isolation -e '.[flash-attn]'
-
-# Install CCE https://docs.axolotl.ai/docs/custom_integrations.html#cut-cross-entropy
-python scripts/cutcrossentropy_install.py | sh
-```
-
-2. (Optional, highly recommended) Install XIELU CUDA
-
-```bash
-## Recommended for reduced VRAM and faster speeds
-
-# Point to CUDA toolkit directory
-# For those using our Docker image, use the below path.
-export CUDA_HOME=/usr/local/cuda
-
-pip3 install git+https://github.com/nickjbrowning/XIELU@59d6031 --no-build-isolation --no-deps
-```
-
-For any installation errors, see [XIELU Installation Issues](#xielu-installation-issues)
-
-3. Run the finetuning example:
-
-```bash
-axolotl train examples/apertus/apertus-8b-qlora.yaml
-```
-
-This config uses about 8.7 GiB VRAM.
-
-Let us know how it goes. Happy finetuning! 🚀
-
-### Tips
-
- For inference, the official Apertus team recommends `top_p=0.9` and `temperature=0.8`.
- You can instead use full paremter fine-tuning by removing the `adapter: qlora` and `load_in_4bit: true` from the config.
- Read more on how to load your own dataset at [docs](https://docs.axolotl.ai/docs/dataset_loading.html).
- The dataset format follows the OpenAI Messages format as seen [here](https://docs.axolotl.ai/docs/dataset-formats/conversation.html#chat_template).
-
-### XIELU Installation Issues
-
-#### `ModuleNotFoundError: No module named 'torch'`
-
-Please check these one by one:
- Running in correct environment
- Env has PyTorch installed
- CUDA toolkit is at `CUDA_HOME`
-
-If those didn't help, please try the below solutions:
-
-1. Pass env for CMAKE and try install again:
-
-    ```bash
-    Python_EXECUTABLE=$(which python) pip3 install git+https://github.com/nickjbrowning/XIELU@59d6031 --no-build-isolation --no-deps
-    ```
-
-2. Git clone the repo and manually hardcode python path:
-
-    ```bash
-    git clone https://github.com/nickjbrowning/XIELU
-    cd xielu
-    git checkout 59d6031
-
-    cd xielu
-    nano CMakeLists.txt  # or vi depending on your preference
-    ```
-
-    ```diff
-    execute_process(
-    -    COMMAND ${Python_EXECUTABLE} -c "import torch.utils; print(torch.utils.cmake_prefix_path)"
-    +    COMMAND /root/miniconda3/envs/py3.11/bin/python -c "import torch.utils; print(torch.utils.cmake_prefix_path)"
-        RESULT_VARIABLE TORCH_CMAKE_PATH_RESULT
-        OUTPUT_VARIABLE TORCH_CMAKE_PATH_OUTPUT
-        ERROR_VARIABLE TORCH_CMAKE_PATH_ERROR
-    )
-    ```
-
-    ```bash
-    pip3 install . --no-build-isolation --no-deps
-    ```
-
-## Optimization Guides
-
- [Multi-GPU Training](https://docs.axolotl.ai/docs/multi-gpu.html)
- [Multi-Node Training](https://docs.axolotl.ai/docs/multi-node.html)
- [LoRA Optimizations](https://docs.axolotl.ai/docs/lora_optims.html)
-
-## Related Resources
-
- [Apertus Tech Report](https://github.com/swiss-ai/apertus-tech-report/blob/main/Apertus_Tech_Report.pdf)
- [Axolotl Docs](https://docs.axolotl.ai)
- [Axolotl Website](https://axolotl.ai)
- [Axolotl GitHub](https://github.com/axolotl-ai-cloud/axolotl)
- [Axolotl Discord](https://discord.gg/7m9sfhzaf3)
--- a/examples/apertus/apertus-8b-qlora.yaml
+++ b/examples/apertus/apertus-8b-qlora.yaml
@@ -1,64 +0,0 @@
-base_model: swiss-ai/Apertus-8B-Instruct-2509
-
-# Automatically upload checkpoint and final model to HF
-# hub_model_id: username/custom_model_name
-
-plugins:
-  - axolotl.integrations.cut_cross_entropy.CutCrossEntropyPlugin
-
-load_in_8bit: false
-load_in_4bit: true
-
-datasets:
-  - path: fozziethebeat/alpaca_messages_2k_test
-    type: chat_template
-
-dataset_prepared_path: last_run_prepared
-val_set_size: 0.1
-output_dir: ./outputs/lora-out
-
-adapter: qlora
-lora_model_dir:
-
-sequence_len: 2048
-sample_packing: true
-
-lora_r: 32
-lora_alpha: 16
-lora_dropout: 0.05
-lora_target_linear: true
-lora_target_modules:
-  - gate_proj
-  - down_proj
-  - up_proj
-  - q_proj
-  - v_proj
-  - k_proj
-  - o_proj
-
-wandb_project:
-wandb_entity:
-wandb_watch:
-wandb_name:
-wandb_log_model:
-
-gradient_accumulation_steps: 4
-micro_batch_size: 2
-num_epochs: 1
-optimizer: adamw_bnb_8bit
-lr_scheduler: cosine
-learning_rate: 0.0002
-
-bf16: auto
-tf32: false
-
-gradient_checkpointing: true
-resume_from_checkpoint:
-logging_steps: 1
-flash_attention: true
-
-warmup_ratio: 0.1
-evals_per_epoch: 1
-saves_per_epoch: 1
-
-# save_first_step: true  # uncomment this to validate checkpoint saving works with your config
--- a/examples/arcee/README.md
+++ b/examples/arcee/README.md
@@ -1,56 +0,0 @@
-# Finetune ArceeAI's AFM with Axolotl
-
-[Arcee Foundation Models (AFM)](https://huggingface.co/collections/arcee-ai/afm-45b-68823397c351603014963473) are a family of 4.5B parameter open weight models trained by Arcee.ai.
-
-This guide shows how to fine-tune it with Axolotl with multi-turn conversations and proper masking.
-
-Thanks to the team at Arcee.ai for using Axolotl in supervised fine-tuning the AFM model.
-
-## Getting started
-
-1. Install Axolotl following the [installation guide](https://docs.axolotl.ai/docs/installation.html). You need to install from main as AFM is only on nightly or use our latest [Docker images](https://docs.axolotl.ai/docs/docker.html).
-
-    Here is an example of how to install from main for pip:
-
-```bash
-# Ensure you have Pytorch installed (Pytorch 2.6.0 min)
-git clone https://github.com/axolotl-ai-cloud/axolotl.git
-cd axolotl
-
-pip3 install packaging==23.2 setuptools==75.8.0 wheel ninja
-pip3 install --no-build-isolation -e '.[flash-attn]'
-
-# Install CCE https://docs.axolotl.ai/docs/custom_integrations.html#cut-cross-entropy
-python scripts/cutcrossentropy_install.py | sh
-```
-
-2. Run the finetuning example:
-
-```bash
-axolotl train examples/arcee/afm-4.5b-qlora.yaml
-```
-
-This config uses about 7.8GiB VRAM.
-
-Let us know how it goes. Happy finetuning! 🚀
-
-### TIPS
-
- For inference, the official Arcee.ai team recommends `top_p: 0.95`, `temperature: 0.5`, `top_k: 50`, and `repeat_penalty: 1.1`.
- You can run a full finetuning by removing the `adapter: qlora` and `load_in_4bit: true` from the config.
- Read more on how to load your own dataset at [docs](https://docs.axolotl.ai/docs/dataset_loading.html).
- The dataset format follows the OpenAI Messages format as seen [here](https://docs.axolotl.ai/docs/dataset-formats/conversation.html#chat_template).
-
-## Optimization Guides
-
- [Multi-GPU Training](https://docs.axolotl.ai/docs/multi-gpu.html)
- [Multi-Node Training](https://docs.axolotl.ai/docs/multi-node.html)
- [LoRA Optimizations](https://docs.axolotl.ai/docs/lora_optims.html)
-
-## Related Resources
-
- [AFM Blog](https://docs.arcee.ai/arcee-foundation-models/introduction-to-arcee-foundation-models)
- [Axolotl Docs](https://docs.axolotl.ai)
- [Axolotl Website](https://axolotl.ai)
- [Axolotl GitHub](https://github.com/axolotl-ai-cloud/axolotl)
- [Axolotl Discord](https://discord.gg/7m9sfhzaf3)
--- a/examples/arcee/afm-4.5b-qlora.yaml
+++ b/examples/arcee/afm-4.5b-qlora.yaml
@@ -1,64 +0,0 @@
-base_model: arcee-ai/AFM-4.5B
-
-# Automatically upload checkpoint and final model to HF
-# hub_model_id: username/custom_model_name
-
-plugins:
-  - axolotl.integrations.cut_cross_entropy.CutCrossEntropyPlugin
-
-load_in_8bit: false
-load_in_4bit: true
-
-datasets:
-  - path: fozziethebeat/alpaca_messages_2k_test
-    type: chat_template
-
-dataset_prepared_path: last_run_prepared
-val_set_size: 0.1
-output_dir: ./outputs/lora-out
-
-adapter: qlora
-lora_model_dir:
-
-sequence_len: 2048
-sample_packing: true
-
-lora_r: 32
-lora_alpha: 16
-lora_dropout: 0.05
-lora_target_linear: true
-lora_target_modules:
-  - gate_proj
-  - down_proj
-  - up_proj
-  - q_proj
-  - v_proj
-  - k_proj
-  - o_proj
-
-wandb_project:
-wandb_entity:
-wandb_watch:
-wandb_name:
-wandb_log_model:
-
-gradient_accumulation_steps: 4
-micro_batch_size: 2
-num_epochs: 1
-optimizer: adamw_bnb_8bit
-lr_scheduler: cosine
-learning_rate: 0.0002
-
-bf16: auto
-tf32: false
-
-gradient_checkpointing: true
-resume_from_checkpoint:
-logging_steps: 1
-flash_attention: true
-
-warmup_ratio: 0.1
-evals_per_epoch: 1
-saves_per_epoch: 1
-
-# save_first_step: true  # uncomment this to validate checkpoint saving works with your config
--- a/examples/archived/README.md
+++ b/examples/archived/README.md
@@ -1,5 +0,0 @@
-# Archived Examples
-
-This directory contains examples that are no longer maintained and may no longer be functional.
-
-We keep them around for archival purposes in case they are useful to others.
--- a/examples/archived/cerebras/btlm-ft.yml
+++ b/examples/archived/cerebras/btlm-ft.yml
@@ -66,7 +66,7 @@ flash_optimum:
 gptq_groupsize:
 gptq_model_v1:

-warmup_ratio: 0.1
+warmup_steps: 32
 evals_per_epoch: 4
 saves_per_epoch: 1
 save_total_limit:
--- a/examples/archived/cerebras/qlora.yml
+++ b/examples/archived/cerebras/qlora.yml
@@ -43,7 +43,7 @@ xformers_attention: true
 flash_attention:
 gptq_groupsize:
 gptq_model_v1:
-warmup_ratio: 0.1
+warmup_steps: 10
 evals_per_epoch: 4
 saves_per_epoch: 1
 weight_decay: 0.1
--- a/examples/cloud/baseten.yaml
+++ b/examples/cloud/baseten.yaml
@@ -1,10 +0,0 @@
-provider: baseten
-project_name:
-
-secrets:
-  - HF_TOKEN
-  - WANDB_API_KEY
-
-gpu: h100
-gpu_count: 8
-node_count: 1
--- a/examples/archived/code-llama/13b/lora.yml
+++ b/examples/archived/code-llama/13b/lora.yml
@@ -17,7 +17,7 @@ output_dir: ./outputs/lora-out

 sequence_len: 4096
 sample_packing: true
-
+pad_to_sequence_len: true

 adapter: lora
 lora_model_dir:
@@ -47,7 +47,7 @@ resume_from_checkpoint:
 logging_steps: 1
 flash_attention: true

-warmup_ratio: 0.1
+warmup_steps: 10
 evals_per_epoch: 4
 saves_per_epoch: 1
 weight_decay: 0.0
--- a/examples/archived/code-llama/13b/qlora.yml
+++ b/examples/archived/code-llama/13b/qlora.yml
@@ -20,7 +20,7 @@ lora_model_dir:

 sequence_len: 4096
 sample_packing: true
-
+pad_to_sequence_len: true

 lora_r: 32
 lora_alpha: 16
@@ -48,7 +48,7 @@ resume_from_checkpoint:
 logging_steps: 1
 flash_attention: true

-warmup_ratio: 0.1
+warmup_steps: 10
 evals_per_epoch: 4
 saves_per_epoch: 1
 weight_decay: 0.0
--- a/examples/archived/code-llama/34b/lora.yml
+++ b/examples/archived/code-llama/34b/lora.yml
@@ -17,7 +17,7 @@ output_dir: ./outputs/lora-out

 sequence_len: 4096
 sample_packing: true
-
+pad_to_sequence_len: true

 adapter: lora
 lora_model_dir:
@@ -47,7 +47,7 @@ resume_from_checkpoint:
 logging_steps: 1
 flash_attention: true

-warmup_ratio: 0.1
+warmup_steps: 10
 evals_per_epoch: 4
 saves_per_epoch: 1
 weight_decay: 0.0
--- a/examples/archived/code-llama/34b/qlora.yml
+++ b/examples/archived/code-llama/34b/qlora.yml
@@ -20,7 +20,7 @@ lora_model_dir:

 sequence_len: 4096
 sample_packing: true
-
+pad_to_sequence_len: true

 lora_r: 32
 lora_alpha: 16
@@ -48,7 +48,7 @@ resume_from_checkpoint:
 logging_steps: 1
 flash_attention: true

-warmup_ratio: 0.1
+warmup_steps: 10
 evals_per_epoch: 4
 saves_per_epoch: 1
 weight_decay: 0.0
--- a/Show More
+++ b/Show More