update tags

Co-authored-by: coderabbitai[bot] <136622811+coderabbitai[bot]@users.noreply.github.com>

.coveragerc

@@ -2,7 +2,6 @@
 source = axolotl
 omit =
     */tests/*
-    setup.py
 
 [report]
 exclude_lines =

.github/CONTRIBUTING.md

@@ -29,13 +29,18 @@ PRs are **greatly welcome**!
 2. Set up the development environment by following the instructions in the [README.md](https://github.com/axolotl-ai-cloud/axolotl/tree/main/README.md) file.
 3. Explore the codebase, run tests, and verify that everything works as expected.
 
-Please run below to setup env
-```bash
-pip3 install -r requirements-dev.txt -r requirements-tests.txt
-pre-commit install
+Please run the below to setup:
 
-# test
-pytest tests/
+```bash
+git clone https://github.com/axolotl-ai-cloud/axolotl.git
+cd axolotl
+
+uv sync --dev && uv pip install flash-attn --no-build-isolation
+source .venv/bin/activate
+
+pre-commit install  # install pre-commit hooks
+
+pytest tests/  # optional; run test suite
 ```
 
 ## How to Contribute
@@ -68,12 +73,7 @@ You can skip certain CI checks by including specific keywords in your commit mes
 
 ### Code Style
 
-axolotl uses [Ruff](https://docs.astral.sh/ruff/) as its code style guide. Please ensure that your code follows these guidelines.
-
-Use the pre-commit linter to ensure that your code is formatted consistently.
-```bash
-pre-commit run --all-files
-```
+axolotl uses [{codestyle}]({URLofCodestyle}) as its code style guide. Please ensure that your code follows these guidelines.
 
 ### Commit Messages
 
@@ -83,6 +83,6 @@ Write clear and concise commit messages that briefly describe the changes made i
 
 - [GitHub Help](https://help.github.com/)
 - [GitHub Pull Request Documentation](https://docs.github.com/en/github/collaborating-with-issues-and-pull-requests)
-- [Ruff](https://docs.astral.sh/ruff/)
+- [{codestyle}]({URLofCodestyle})
 
 Thank you once again for your interest in contributing to axolotl. We look forward to collaborating with you and creating an even better project together!

.github/FUNDING.yml

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

.github/PULL_REQUEST_TEMPLATE.md

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

.github/workflows/base.yml

@@ -15,21 +15,44 @@ on:
       - '.github/workflows/base.yml'
   workflow_dispatch:
 
-permissions:
-  contents: read
-
 jobs:
   build-base:
     if: ${{ github.repository_owner == 'axolotl-ai-cloud' && (github.event_name != 'pull_request' || !github.event.pull_request.draft) }}
     timeout-minutes: 480
     # this job needs to be run on self-hosted GPU runners...
     runs-on: ubuntu-latest-m
-    env:
-      HAS_DOCKERHUB_CREDS: ${{ secrets.DOCKERHUB_USERNAME != '' && secrets.DOCKERHUB_TOKEN != '' }}
     strategy:
       fail-fast: false
       matrix:
         include:
+          - 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"
+            dockerfile: "Dockerfile-base"
+          - 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"
+            dockerfile: "Dockerfile-base"
+          - 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-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-base"
           - cuda: "128"
             cuda_version: 12.8.1
             cudnn_version: ""
@@ -37,71 +60,6 @@ jobs:
             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-base"
-            platforms: "linux/amd64"
-          - 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"
-            platforms: "linux/amd64,linux/arm64"
-          - cuda: "128"
-            cuda_version: 12.8.1
-            cudnn_version: ""
-            python_version: "3.11"
-            pytorch: 2.9.1
-            torch_cuda_arch_list: "7.0 7.5 8.0 8.6 8.7 8.9 9.0+PTX"
-            dockerfile: "Dockerfile-base"
-            platforms: "linux/amd64,linux/arm64"
-          - cuda: "128"
-            cuda_version: 12.8.1
-            cudnn_version: ""
-            python_version: "3.11"
-            pytorch: 2.10.0
-            torch_cuda_arch_list: "7.0 7.5 8.0 8.6 8.7 8.9 9.0+PTX"
-            dockerfile: "Dockerfile-base"
-            platforms: "linux/amd64,linux/arm64"
-          - cuda: "128"
-            cuda_version: 12.8.1
-            cudnn_version: ""
-            python_version: "3.12"
-            pytorch: 2.10.0
-            torch_cuda_arch_list: "7.0 7.5 8.0 8.6 8.7 8.9 9.0+PTX"
-            dockerfile: "Dockerfile-base"
-            platforms: "linux/amd64,linux/arm64"
-#          - cuda: "129"
-#            cuda_version: 12.9.1
-#            cudnn_version: ""
-#            python_version: "3.12"
-#            pytorch: 2.9.1
-#            torch_cuda_arch_list: "7.0 7.5 8.0 8.6 8.7 8.9 9.0+PTX"
-#            dockerfile: "Dockerfile-base"
-#            platforms: "linux/amd64,linux/arm64"
-          - cuda: "130"
-            cuda_version: 13.0.0
-            cudnn_version: ""
-            python_version: "3.11"
-            pytorch: 2.9.1
-            torch_cuda_arch_list: "9.0+PTX"
-            dockerfile: "Dockerfile-base"
-            platforms: "linux/amd64,linux/arm64"
-          - cuda: "130"
-            cuda_version: 13.0.0
-            cudnn_version: ""
-            python_version: "3.12"
-            pytorch: 2.9.1
-            torch_cuda_arch_list: "9.0+PTX"
-            dockerfile: "Dockerfile-base"
-            platforms: "linux/amd64,linux/arm64"
-          - cuda: "130"
-            cuda_version: 13.0.0
-            cudnn_version: ""
-            python_version: "3.12"
-            pytorch: 2.10.0
-            torch_cuda_arch_list: "9.0+PTX"
-            dockerfile: "Dockerfile-base"
-            platforms: "linux/amd64,linux/arm64"
 #          - cuda: "128"
 #            cuda_version: 12.8.1
 #            cudnn_version: ""
@@ -125,23 +83,24 @@ jobs:
         uses: docker/metadata-action@v5
         with:
           images: |
+            winglian/axolotl-base
             axolotlai/axolotl-base
       - name: Login to Docker Hub
-        uses: docker/login-action@v3
-        if: ${{ github.event_name != 'pull_request' && env.HAS_DOCKERHUB_CREDS == 'true' }}
+        uses: docker/login-action@v2
         with:
           username: ${{ secrets.DOCKERHUB_USERNAME }}
           password: ${{ secrets.DOCKERHUB_TOKEN }}
       - name: Set up Docker Buildx
         uses: docker/setup-buildx-action@v3
       - name: Build
-        uses: docker/build-push-action@v5
+        uses: docker/build-push-action@v4
         with:
           context: .
           file: ./docker/${{ matrix.dockerfile }}
-          platforms: ${{ matrix.platforms }}
           push: ${{ github.event_name != 'pull_request' }}
-          tags: ${{ steps.metadata.outputs.tags }}-base-py${{ matrix.python_version }}-cu${{ matrix.cuda }}-${{ matrix.pytorch }}${{ matrix.axolotl_extras != '' && '-' || '' }}${{ matrix.axolotl_extras }}
+          tags: |
+            ${{ steps.metadata.outputs.tags }}-base-py${{ matrix.python_version }}-cu${{ matrix.cuda }}-${{ matrix.pytorch }}${{ matrix.axolotl_extras != '' && '-' || '' }}${{ matrix.axolotl_extras }}
+            ${{ steps.metadata.outputs.tags }}-base-uv-py${{ 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 }}
@@ -154,12 +113,31 @@ jobs:
     if: ${{ github.repository_owner == 'axolotl-ai-cloud' && (github.event_name != 'pull_request' || !github.event.pull_request.draft) }}
     timeout-minutes: 480
     runs-on: ubuntu-latest-m
-    env:
-      HAS_DOCKERHUB_CREDS: ${{ secrets.DOCKERHUB_USERNAME != '' && secrets.DOCKERHUB_TOKEN != '' }}
     strategy:
       fail-fast: false
       matrix:
         include:
+          - cuda: "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"
+            dockerfile: "Dockerfile-uv-base"
+          - 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: ""
@@ -167,79 +145,6 @@ jobs:
             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"
-            platforms: "linux/amd64"
-          - cuda: "128"
-            cuda_version: 12.8.1
-            cudnn_version: ""
-            python_version: "3.11"
-            pytorch: 2.9.1
-            torch_cuda_arch_list: "7.0 7.5 8.0 8.6 8.7 8.9 9.0+PTX"
-            dockerfile: "Dockerfile-uv-base"
-            platforms: "linux/amd64,linux/arm64"
-          - cuda: "128"
-            cuda_version: 12.8.1
-            cudnn_version: ""
-            python_version: "3.12"
-            pytorch: 2.9.1
-            torch_cuda_arch_list: "7.0 7.5 8.0 8.6 8.7 8.9 9.0+PTX"
-            dockerfile: "Dockerfile-uv-base"
-            platforms: "linux/amd64,linux/arm64"
-          - cuda: "128"
-            cuda_version: 12.8.1
-            cudnn_version: ""
-            python_version: "3.11"
-            pytorch: 2.9.0
-            torch_cuda_arch_list: "7.0 7.5 8.0 8.6 8.7 8.9 9.0+PTX"
-            dockerfile: "Dockerfile-uv-base"
-            platforms: "linux/amd64,linux/arm64"
-          - cuda: "128"
-            cuda_version: 12.8.1
-            cudnn_version: ""
-            python_version: "3.11"
-            pytorch: 2.10.0
-            torch_cuda_arch_list: "7.0 7.5 8.0 8.6 8.7 8.9 9.0+PTX"
-            dockerfile: "Dockerfile-uv-base"
-            platforms: "linux/amd64,linux/arm64"
-          - cuda: "128"
-            cuda_version: 12.8.1
-            cudnn_version: ""
-            python_version: "3.12"
-            pytorch: 2.10.0
-            torch_cuda_arch_list: "7.0 7.5 8.0 8.6 8.7 8.9 9.0+PTX"
-            dockerfile: "Dockerfile-uv-base"
-            platforms: "linux/amd64,linux/arm64"
-#          - cuda: "129"
-#            cuda_version: 12.9.1
-#            cudnn_version: ""
-#            python_version: "3.12"
-#            pytorch: 2.9.1
-#            torch_cuda_arch_list: "7.0 7.5 8.0 8.6 8.7 8.9 9.0+PTX"
-#            dockerfile: "Dockerfile-uv-base"
-#            platforms: "linux/amd64,linux/arm64"
-          - cuda: "130"
-            cuda_version: 13.0.0
-            cudnn_version: ""
-            python_version: "3.11"
-            pytorch: 2.9.1
-            torch_cuda_arch_list: "9.0+PTX"
-            dockerfile: "Dockerfile-uv-base"
-            platforms: "linux/amd64,linux/arm64"
-          - cuda: "130"
-            cuda_version: 13.0.0
-            cudnn_version: ""
-            python_version: "3.12"
-            pytorch: 2.9.1
-            torch_cuda_arch_list: "9.0+PTX"
-            dockerfile: "Dockerfile-uv-base"
-            platforms: "linux/amd64,linux/arm64"
-          - cuda: "130"
-            cuda_version: 13.0.0
-            cudnn_version: ""
-            python_version: "3.12"
-            pytorch: 2.10.0
-            torch_cuda_arch_list: "9.0+PTX"
-            dockerfile: "Dockerfile-uv-base"
-            platforms: "linux/amd64,linux/arm64"
     steps:
       - name: Checkout
         uses: actions/checkout@v4
@@ -250,19 +155,17 @@ jobs:
           images: |
             axolotlai/axolotl-base-uv
       - name: Login to Docker Hub
-        uses: docker/login-action@v3
-        if: ${{ github.event_name != 'pull_request' && env.HAS_DOCKERHUB_CREDS == 'true' }}
+        uses: docker/login-action@v2
         with:
           username: ${{ secrets.DOCKERHUB_USERNAME }}
           password: ${{ secrets.DOCKERHUB_TOKEN }}
       - name: Set up Docker Buildx
         uses: docker/setup-buildx-action@v3
       - name: Build
-        uses: docker/build-push-action@v5
+        uses: docker/build-push-action@v4
         with:
           context: .
           file: ./docker/${{ matrix.dockerfile }}
-          platforms: ${{ matrix.platforms }}
           push: ${{ github.event_name != 'pull_request' }}
           tags: ${{ steps.metadata.outputs.tags }}-base-py${{ matrix.python_version }}-cu${{ matrix.cuda }}-${{ matrix.pytorch }}${{ matrix.axolotl_extras != '' && '-' || '' }}${{ matrix.axolotl_extras }}
           labels: ${{ steps.metadata.outputs.labels }}

.github/workflows/docs.yml

@@ -12,9 +12,6 @@ jobs:
     build-deploy:
         runs-on: ubuntu-latest
         steps:
-        - name: cleanup node
-          run: |
-            sudo rm -rf /usr/share/dotnet /usr/local/lib/android /opt/ghc /opt/hostedtoolcache/CodeQL
         - name: Check out repository
           uses: actions/checkout@v4
         - name: Set up Quarto
@@ -23,10 +20,14 @@ jobs:
           uses: actions/setup-python@v5
           with:
             python-version: '3.11'
+        - name: Install uv
+          uses: astral-sh/setup-uv@v4
+          with:
+            version: "latest"
         - name: Install dependencies
           run: |
-            python3 -m pip install jupyter quartodoc
-            python3 -m pip install -e .
+            uv pip install --system jupyter quartodoc
+            uv pip install --system -e .
         - name: Build autodoc
           run: quartodoc build
         - name: Publish to GitHub Pages (and render)

.github/workflows/lint.yml

@@ -6,16 +6,13 @@ on:
       types: [opened, synchronize, reopened, ready_for_review]
       paths:
        - '**.py'
-       - 'requirements.txt'
+       - 'pyproject.toml'
        - '.github/workflows/*.yml'
        - "*.[q]md"
        - "examples/**/*.y[a]?ml"
        - ".pre-commit-config.yaml"
   workflow_dispatch:
 
-permissions:
-  contents: read
-
 jobs:
   pre-commit:
     name: pre-commit
@@ -26,5 +23,4 @@ jobs:
       - uses: actions/setup-python@v5
         with:
           python-version: "3.11"
-          cache: 'pip' # caching pip dependencies
       - uses: pre-commit/action@v3.0.1

.github/workflows/main.yml

@@ -8,9 +8,6 @@ on:
       - "v*"
   workflow_dispatch:
 
-permissions:
-  contents: read
-
 jobs:
   build-axolotl:
     if: ${{ ! contains(github.event.commits[0].message, '[skip docker]') && github.repository_owner == 'axolotl-ai-cloud' }}
@@ -18,49 +15,27 @@ jobs:
       fail-fast: false
       matrix:
         include:
+          - cuda: 126
+            cuda_version: 12.6.3
+            python_version: "3.11"
+            pytorch: 2.6.0
+            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:
-            platforms: "linux/amd64"
-          - cuda: 128
-            cuda_version: 12.8.1
-            python_version: "3.11"
-            pytorch: 2.9.0
-            axolotl_extras:
-            platforms: "linux/amd64,linux/arm64"
-          - cuda: 128
-            cuda_version: 12.8.1
-            python_version: "3.11"
-            pytorch: 2.9.1
-            axolotl_extras:
-            platforms: "linux/amd64,linux/arm64"
-            is_latest: true
-          - cuda: 128
-            cuda_version: 12.8.1
-            python_version: "3.12"
-            pytorch: 2.10.0
-            axolotl_extras:
-            platforms: "linux/amd64,linux/arm64"
-#          - cuda: 129
-#            cuda_version: 12.9.1
-#            python_version: "3.12"
-#            pytorch: 2.9.1
-#            axolotl_extras:
-#            platforms: "linux/amd64,linux/arm64"
-          - cuda: 130
-            cuda_version: 13.0.0
-            python_version: "3.11"
-            pytorch: 2.9.1
-            axolotl_extras:
-            platforms: "linux/amd64,linux/arm64"
-          - cuda: 130
-            cuda_version: 13.0.0
-            python_version: "3.12"
-            pytorch: 2.10.0
-            axolotl_extras:
-            platforms: "linux/amd64,linux/arm64"
     runs-on: axolotl-gpu-runner
     steps:
       - name: Checkout
@@ -70,6 +45,7 @@ jobs:
         uses: docker/metadata-action@v5
         with:
           images: |
+            winglian/axolotl
             axolotlai/axolotl
           tags: |
             type=ref,event=branch
@@ -86,13 +62,14 @@ jobs:
         uses: docker/build-push-action@v5
         with:
           context: .
-          platforms: ${{ matrix.platforms }}
           build-args: |
             BASE_TAG=${{ github.ref_type == 'tag' && 'main' || github.ref_name }}-base-py${{ matrix.python_version }}-cu${{ matrix.cuda }}-${{ matrix.pytorch }}
             CUDA=${{ matrix.cuda }}
             PYTORCH_VERSION=${{ matrix.pytorch }}
             AXOLOTL_ARGS=${{ matrix.axolotl_args }}
             AXOLOTL_EXTRAS=${{ matrix.axolotl_extras}}
+            GIT_REF=${{ github.ref }}
+            GIT_SHA=${{ github.sha }}
           file: ./docker/Dockerfile
           push: ${{ github.event_name != 'pull_request' }}
           tags: |
@@ -101,134 +78,40 @@ jobs:
             ${{ (matrix.is_latest) && format('{0}-latest', steps.metadata.outputs.tags) || '' }}
           labels: ${{ steps.metadata.outputs.labels }}
 
-  build-axolotl-uv:
-    if: ${{ ! contains(github.event.commits[0].message, '[skip docker]') && github.repository_owner == 'axolotl-ai-cloud' }}
-    strategy:
-      fail-fast: false
-      matrix:
-        include:
-          - cuda: 128
-            cuda_version: 12.8.1
-            python_version: "3.11"
-            pytorch: 2.9.1
-            axolotl_extras:
-            platforms: "linux/amd64,linux/arm64"
-          - cuda: 128
-            cuda_version: 12.8.1
-            python_version: "3.12"
-            pytorch: 2.9.1
-            axolotl_extras:
-            platforms: "linux/amd64,linux/arm64"
-            is_latest: true
-          - cuda: 128
-            cuda_version: 12.8.1
-            python_version: "3.12"
-            pytorch: 2.10.0
-            axolotl_extras:
-            platforms: "linux/amd64,linux/arm64"
-          - cuda: 130
-            cuda_version: 13.0.0
-            python_version: "3.11"
-            pytorch: 2.9.1
-            axolotl_extras:
-            platforms: "linux/amd64,linux/arm64"
-          - cuda: 130
-            cuda_version: 13.0.0
-            python_version: "3.12"
-            pytorch: 2.10.0
-            axolotl_extras:
-            platforms: "linux/amd64,linux/arm64"
-    runs-on: axolotl-gpu-runner
-    steps:
-      - name: Checkout
-        uses: actions/checkout@v4
-      - name: Docker metadata
-        id: metadata
-        uses: docker/metadata-action@v5
-        with:
-          images: |
-            axolotlai/axolotl-uv
-          tags: |
-            type=ref,event=branch
-            type=pep440,pattern={{version}}
-      - name: Set up Docker Buildx
-        uses: docker/setup-buildx-action@v3
-      - name: Login to Docker Hub
-        uses: docker/login-action@v3
-        with:
-          username: ${{ secrets.DOCKERHUB_USERNAME }}
-          password: ${{ secrets.DOCKERHUB_TOKEN }}
-      # guidance for testing before pushing: https://docs.docker.com/build/ci/github-actions/test-before-push/
-      - name: Build and export to Docker
-        uses: docker/build-push-action@v5
-        with:
-          context: .
-          platforms: ${{ matrix.platforms }}
-          build-args: |
-            BASE_TAG=${{ github.ref_type == 'tag' && 'main' || github.ref_name }}-base-py${{ matrix.python_version }}-cu${{ matrix.cuda }}-${{ matrix.pytorch }}
-            CUDA=${{ matrix.cuda }}
-            PYTORCH_VERSION=${{ matrix.pytorch }}
-            AXOLOTL_ARGS=${{ matrix.axolotl_args }}
-            AXOLOTL_EXTRAS=${{ matrix.axolotl_extras}}
-          file: ./docker/Dockerfile-uv
-          push: ${{ github.event_name != 'pull_request' }}
-          tags: |
-            ${{ steps.metadata.outputs.tags }}-py${{ matrix.python_version }}-cu${{ matrix.cuda }}-${{ matrix.pytorch }}${{ matrix.axolotl_extras != '' && '-' || '' }}${{ matrix.axolotl_extras }}
-            ${{ steps.metadata.outputs.tags }}-py${{ matrix.python_version }}-cu${{ matrix.cuda }}-${{ matrix.pytorch }}
-            ${{ (matrix.is_latest) && format('{0}-latest', steps.metadata.outputs.tags) || '' }}
-          labels: ${{ steps.metadata.outputs.labels }}
-
   build-axolotl-cloud:
     needs: build-axolotl
     if: ${{ ! contains(github.event.commits[0].message, '[skip docker]') && github.repository_owner == 'axolotl-ai-cloud' }}
     # this job needs to be run on self-hosted GPU runners...
     strategy:
-      fail-fast: false
       matrix:
         include:
+          - cuda: 126
+            cuda_version: 12.6.3
+            python_version: "3.11"
+            pytorch: 2.6.0
+            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:
-            platforms: "linux/amd64"
-          - cuda: 128
-            cuda_version: 12.8.1
-            python_version: "3.11"
-            pytorch: 2.9.0
-            axolotl_extras:
-            platforms: "linux/amd64,linux/arm64"
-          - cuda: 128
-            cuda_version: 12.8.1
-            python_version: "3.11"
-            pytorch: 2.9.1
-            axolotl_extras:
-            is_latest: true
-            platforms: "linux/amd64,linux/arm64"
-          - cuda: 128
-            cuda_version: 12.8.1
-            python_version: "3.12"
-            pytorch: 2.10.0
-            axolotl_extras:
-            platforms: "linux/amd64,linux/arm64"
-#          - cuda: 129
-#            cuda_version: 12.9.1
-#            python_version: "3.12"
-#            pytorch: 2.9.1
-#            axolotl_extras:
-#            platforms: "linux/amd64,linux/arm64"
-          - cuda: 130
-            cuda_version: 13.0.0
-            python_version: "3.11"
-            pytorch: 2.9.1
-            axolotl_extras:
-            platforms: "linux/amd64,linux/arm64"
-          - cuda: 130
-            cuda_version: 13.0.0
-            python_version: "3.12"
-            pytorch: 2.10.0
-            axolotl_extras:
-            platforms: "linux/amd64,linux/arm64"
     runs-on: axolotl-gpu-runner
     steps:
       - name: Checkout
@@ -238,6 +121,7 @@ jobs:
         uses: docker/metadata-action@v5
         with:
           images: |
+            winglian/axolotl-cloud
             axolotlai/axolotl-cloud
           tags: |
             type=ref,event=branch
@@ -253,10 +137,11 @@ jobs:
         uses: docker/build-push-action@v5
         with:
           context: .
-          platforms: ${{ matrix.platforms }}
           build-args: |
             BASE_TAG=${{ github.ref_type == 'tag' && 'main' || github.ref_name }}-py${{ matrix.python_version }}-cu${{ matrix.cuda }}-${{ matrix.pytorch }}${{ matrix.axolotl_extras != '' && '-' || '' }}${{ matrix.axolotl_extras }}
             CUDA=${{ matrix.cuda }}
+            GIT_REF=${{ github.ref }}
+            GIT_SHA=${{ github.sha }}
           file: ./docker/Dockerfile-cloud
           push: ${{ github.event_name != 'pull_request' }}
           tags: |
@@ -264,98 +149,34 @@ jobs:
              ${{ (matrix.is_latest) && format('{0}-latest', steps.metadata.outputs.tags) || '' }}
           labels: ${{ steps.metadata.outputs.labels }}
 
-  build-axolotl-cloud-uv:
-    needs: build-axolotl-uv
-    if: ${{ ! contains(github.event.commits[0].message, '[skip docker]') && github.repository_owner == 'axolotl-ai-cloud' }}
-    # this job needs to be run on self-hosted GPU runners...
-    strategy:
-      fail-fast: false
-      matrix:
-        include:
-          - cuda: 128
-            cuda_version: 12.8.1
-            python_version: "3.11"
-            pytorch: 2.9.1
-            axolotl_extras:
-            platforms: "linux/amd64,linux/arm64"
-          - cuda: 128
-            cuda_version: 12.8.1
-            python_version: "3.12"
-            pytorch: 2.9.1
-            axolotl_extras:
-            is_latest: true
-            platforms: "linux/amd64,linux/arm64"
-          - cuda: 128
-            cuda_version: 12.8.1
-            python_version: "3.12"
-            pytorch: 2.10.0
-            axolotl_extras:
-            platforms: "linux/amd64,linux/arm64"
-          - cuda: 130
-            cuda_version: 13.0.0
-            python_version: "3.11"
-            pytorch: 2.9.1
-            axolotl_extras:
-            platforms: "linux/amd64,linux/arm64"
-          - cuda: 130
-            cuda_version: 13.0.0
-            python_version: "3.12"
-            pytorch: 2.10.0
-            axolotl_extras:
-            platforms: "linux/amd64,linux/arm64"
-    runs-on: axolotl-gpu-runner
-    steps:
-      - name: Checkout
-        uses: actions/checkout@v4
-      - name: Docker metadata
-        id: metadata
-        uses: docker/metadata-action@v5
-        with:
-          images: |
-            axolotlai/axolotl-cloud-uv
-          tags: |
-            type=ref,event=branch
-            type=pep440,pattern={{version}}
-      - name: Login to Docker Hub
-        uses: docker/login-action@v3
-        with:
-          username: ${{ secrets.DOCKERHUB_USERNAME }}
-          password: ${{ secrets.DOCKERHUB_TOKEN }}
-      - name: Set up Docker Buildx
-        uses: docker/setup-buildx-action@v3
-      - name: Build
-        uses: docker/build-push-action@v5
-        with:
-          context: .
-          platforms: ${{ matrix.platforms }}
-          build-args: |
-            BASE_TAG=${{ github.ref_type == 'tag' && 'main' || github.ref_name }}-py${{ matrix.python_version }}-cu${{ matrix.cuda }}-${{ matrix.pytorch }}${{ matrix.axolotl_extras != '' && '-' || '' }}${{ matrix.axolotl_extras }}
-            CUDA=${{ matrix.cuda }}
-          file: ./docker/Dockerfile-cloud-uv
-          push: ${{ github.event_name != 'pull_request' }}
-          tags: |
-             ${{ steps.metadata.outputs.tags }}-py${{ matrix.python_version }}-cu${{ matrix.cuda }}-${{ matrix.pytorch }}${{ matrix.axolotl_extras != '' && '-' || '' }}${{ matrix.axolotl_extras }}
-             ${{ (matrix.is_latest) && format('{0}-latest', steps.metadata.outputs.tags) || '' }}
-          labels: ${{ steps.metadata.outputs.labels }}
-
   build-axolotl-cloud-no-tmux:
     needs: build-axolotl
     if: ${{ ! contains(github.event.commits[0].message, '[skip docker]') && github.repository_owner == 'axolotl-ai-cloud' }}
     # this job needs to be run on self-hosted GPU runners...
     strategy:
-      fail-fast: false
       matrix:
         include:
+          - cuda: 126
+            cuda_version: 12.6.3
+            python_version: "3.11"
+            pytorch: 2.6.0
+            axolotl_extras:
+          - cuda: 126
+            cuda_version: 12.6.3
+            python_version: "3.11"
+            pytorch: 2.7.1
+            axolotl_extras:
+            is_latest:
+          - cuda: 126
+            cuda_version: 12.6.3
+            python_version: "3.11"
+            pytorch: 2.7.1
+            axolotl_extras: vllm
+            is_latest: true
           - cuda: 128
             cuda_version: 12.8.1
             python_version: "3.11"
-            pytorch: 2.9.1
-            axolotl_extras:
-            is_latest: true
-          - cuda: 130
-            cuda_version: 13.0.0
-            python_version: "3.11"
-            pytorch: 2.9.1
+            pytorch: 2.8.0
             axolotl_extras:
             is_latest:
     runs-on: axolotl-gpu-runner
@@ -367,6 +188,7 @@ jobs:
         uses: docker/metadata-action@v5
         with:
           images: |
+            winglian/axolotl-cloud-term
             axolotlai/axolotl-cloud-term
           tags: |
             type=ref,event=branch
@@ -382,10 +204,11 @@ jobs:
         uses: docker/build-push-action@v5
         with:
           context: .
-          platforms: linux/amd64,linux/arm64
           build-args: |
             BASE_TAG=${{ github.ref_type == 'tag' && 'main' || github.ref_name }}-py${{ matrix.python_version }}-cu${{ matrix.cuda }}-${{ matrix.pytorch }}${{ matrix.axolotl_extras != '' && '-' || '' }}${{ matrix.axolotl_extras }}
             CUDA=${{ matrix.cuda }}
+            GIT_REF=${{ github.ref }}
+            GIT_SHA=${{ github.sha }}
           file: ./docker/Dockerfile-cloud-no-tmux
           push: ${{ github.event_name != 'pull_request' }}
           tags: |

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

@@ -4,11 +4,8 @@ on:
   pull_request:
     paths:
       - 'tests/e2e/multigpu/**.py'
-      - 'requirements.txt'
-      - 'setup.py'
       - 'pyproject.toml'
       - '.github/workflows/multi-gpu-e2e.yml'
-      - 'scripts/cutcrossentropy_install.py'
       - 'src/axolotl/core/trainers/mixins/sequence_parallel.py'
       - 'src/axolotl/utils/distributed.py'
   workflow_dispatch:
@@ -20,12 +17,6 @@ concurrency:
   group: ${{ github.workflow }}-${{ github.ref }}
   cancel-in-progress: ${{ github.ref != 'refs/heads/main' }}
 
-permissions:
-  contents: read
-
-env:
-  MODAL_IMAGE_BUILDER_VERSION: "2025.06"
-
 jobs:
   test-axolotl-multigpu:
     if: ${{ ! contains(github.event.commits[0].message, '[skip e2e]') && github.repository_owner == 'axolotl-ai-cloud' && (github.event_name != 'pull_request' || !github.event.pull_request.draft) }}
@@ -33,33 +24,27 @@ jobs:
       fail-fast: false
       matrix:
         include:
+          - cuda: 126
+            cuda_version: 12.6.3
+            python_version: "3.11"
+            pytorch: 2.6.0
+            axolotl_extras:
+            num_gpus: 2
+            nightly_build: "true"
+          - cuda: 126
+            cuda_version: 12.6.3
+            python_version: "3.11"
+            pytorch: 2.7.1
+            axolotl_extras: vllm
+            num_gpus: 2
+            nightly_build: "true"
           - cuda: 128
             cuda_version: 12.8.1
             python_version: "3.11"
             pytorch: 2.8.0
             axolotl_extras: fbgemm-gpu
             num_gpus: 2
-#          - cuda: 129
-#            cuda_version: 12.9.1
-#            python_version: "3.12"
-#            pytorch: 2.9.1
-#            axolotl_extras: "fbgemm-gpu"
-#            num_gpus: 2
-#            dockerfile: "Dockerfile-uv.jinja"
-          - cuda: 130
-            cuda_version: 13.0.0
-            python_version: "3.11"
-            pytorch: 2.9.1
-            axolotl_extras:
-#            axolotl_extras: fbgemm-gpu
-            num_gpus: 2
-          - cuda: 128
-            cuda_version: 12.8.1
-            python_version: "3.11"
-            pytorch: 2.10.0
-            axolotl_extras: "fbgemm-gpu"
-            num_gpus: 2
-            dockerfile: "Dockerfile-uv.jinja"
+            nightly_build: "true"
     runs-on: [self-hosted, modal]
     timeout-minutes: 120
     steps:
@@ -69,21 +54,24 @@ jobs:
         uses: actions/setup-python@v5
         with:
           python-version: "3.11"
+      - name: Install uv
+        uses: astral-sh/setup-uv@v4
+        with:
+          version: "latest"
       - name: Install Modal
         run: |
           python -m pip install --upgrade pip
-          pip install modal==1.3.0.post1 jinja2
+          pip install modal==1.0.2 jinja2 protobuf
       - name: Update env vars
         run: |
-          echo "BASE_TAG=main-base-py${{ matrix.python_version }}-cu${{ matrix.cuda }}-${{ matrix.pytorch }}" >> $GITHUB_ENV
+          echo "BASE_TAG=${{ github.ref_name }}-base-py${{ matrix.python_version }}-cu${{ matrix.cuda }}-${{ matrix.pytorch }}" >> $GITHUB_ENV
           echo "PYTORCH_VERSION=${{ matrix.pytorch}}" >> $GITHUB_ENV
           echo "AXOLOTL_ARGS=${{ matrix.axolotl_args}}" >> $GITHUB_ENV
           echo "AXOLOTL_EXTRAS=${{ matrix.axolotl_extras}}" >> $GITHUB_ENV
           echo "CUDA=${{ matrix.cuda }}" >> $GITHUB_ENV
           echo "N_GPUS=${{ matrix.num_gpus }}" >> $GITHUB_ENV
-          echo "E2E_DOCKERFILE=${{ matrix.dockerfile || 'Dockerfile.jinja'}}" >> $GITHUB_ENV
+          echo "NIGHTLY_BUILD=${{ matrix.nightly_build }}" >> $GITHUB_ENV
+          echo "CODECOV_TOKEN=${{ secrets.CODECOV_TOKEN }}" >> $GITHUB_ENV
       - name: Run tests job on Modal
-        env:
-          CODECOV_TOKEN: ${{ secrets.CODECOV_TOKEN }}
         run: |
           modal run -m cicd.multigpu

.github/workflows/nightlies.yml

@@ -5,9 +5,6 @@ on:
   schedule:
     - cron: '0 0 * * *'  # Runs at 00:00 UTC every day
 
-permissions:
-  contents: read
-
 jobs:
   build-axolotl:
     if: ${{ ! contains(github.event.commits[0].message, '[skip docker]') && github.repository_owner == 'axolotl-ai-cloud' }}
@@ -15,15 +12,15 @@ jobs:
       fail-fast: false
       matrix:
         include:
-          - cuda: 128
-            cuda_version: 12.8.1
+          - cuda: 126
+            cuda_version: 12.6.3
             python_version: "3.11"
-            pytorch: 2.8.0
+            pytorch: 2.6.0
             axolotl_extras:
-          - cuda: 128
-            cuda_version: 12.8.1
+          - cuda: 126
+            cuda_version: 12.6.3
             python_version: "3.11"
-            pytorch: 2.9.1
+            pytorch: 2.7.1
             axolotl_extras:
     runs-on: axolotl-gpu-runner
     steps:
@@ -34,6 +31,7 @@ jobs:
         uses: docker/metadata-action@v5
         with:
           images: |
+            winglian/axolotl
             axolotlai/axolotl
           tags: |
             type=raw,value={{ branch }}-{{ date 'YYYYMMDD' }}
@@ -54,6 +52,8 @@ jobs:
             CUDA=${{ matrix.cuda }}
             PYTORCH_VERSION=${{ matrix.pytorch }}
             AXOLOTL_ARGS=${{ matrix.axolotl_args }}
+            GIT_REF=${{ github.ref }}
+            GIT_SHA=${{ github.sha }}
           file: ./docker/Dockerfile
           push: ${{ github.event_name != 'pull_request' }}
           tags: |
@@ -67,15 +67,15 @@ jobs:
     strategy:
       matrix:
         include:
-          - cuda: 128
-            cuda_version: 12.8.1
+          - cuda: 126
+            cuda_version: 12.6.3
             python_version: "3.11"
-            pytorch: 2.8.0
+            pytorch: 2.6.0
             axolotl_extras:
-          - cuda: 128
-            cuda_version: 12.8.1
+          - cuda: 126
+            cuda_version: 12.6.3
             python_version: "3.11"
-            pytorch: 2.9.1
+            pytorch: 2.7.1
             axolotl_extras:
     runs-on: axolotl-gpu-runner
     steps:
@@ -86,6 +86,7 @@ jobs:
         uses: docker/metadata-action@v5
         with:
           images: |
+            winglian/axolotl-cloud
             axolotlai/axolotl-cloud
           tags: |
             type=raw,value={{ branch }}-{{ date 'YYYYMMDD' }}
@@ -103,6 +104,8 @@ jobs:
           build-args: |
             BASE_TAG=${{ github.ref_name }}-py${{ matrix.python_version }}-cu${{ matrix.cuda }}-${{ matrix.pytorch }}${{ matrix.axolotl_extras != '' && '-' || '' }}${{ matrix.axolotl_extras }}
             CUDA=${{ matrix.cuda }}
+            GIT_REF=${{ github.ref }}
+            GIT_SHA=${{ github.sha }}
           file: ./docker/Dockerfile-cloud
           push: ${{ github.event_name != 'pull_request' }}
           tags: |

.github/workflows/precommit-autoupdate.yml

@@ -2,11 +2,9 @@ name: Pre-commit auto-update
 
 on:
   schedule:
-    - cron: '0 0 1 * *'  # Run monthly
+    - cron: '0 0 * * 0'  # Run weekly
   workflow_dispatch:  # Manual kickoff
 
-permissions: {}
-
 jobs:
   auto-update:
     runs-on: ubuntu-latest
@@ -20,10 +18,15 @@ jobs:
         with:
           python-version: '3.11'
 
+      - name: Install uv
+        uses: astral-sh/setup-uv@v4
+        with:
+          version: "latest"
+
       - name: Update pre-commit hooks
         id: update
         run: |
-          pip install pre-commit
+          uv pip install --system pre-commit
           pre-commit autoupdate
           if [[ -n $(git status --porcelain) ]]; then
             echo "changes=true" >> $GITHUB_OUTPUT

.github/workflows/preview-docs.yml

@@ -11,21 +11,22 @@ on:
       - '_quarto.yml'
       - docs/scripts/generate_config_docs.py
       - src/axolotl/utils/schemas/**.py
-      - .github/workflows/preview-docs.yml
 
 permissions:
-  contents: read
+  checks: write
+  contents: write
+  deployments: write
+  issues: write
+  discussions: write
+  pages: write
   pull-requests: write
+  statuses: write
 
 jobs:
   preview:
     runs-on: ubuntu-latest
     if: ${{ !github.event.pull_request.draft }}
     steps:
-      - name: cleanup node
-        run: |
-          sudo rm -rf /usr/share/dotnet /usr/local/lib/android /opt/ghc /opt/hostedtoolcache/CodeQL
-
       - name: Check out repository
         uses: actions/checkout@v4
         with:
@@ -39,10 +40,15 @@ jobs:
         with:
           python-version: '3.11'
 
+      - name: Install uv
+        uses: astral-sh/setup-uv@v4
+        with:
+          version: "latest"
+
       - name: Install dependencies
         run: |
-          python3 -m pip install jupyter quartodoc
-          python3 -m pip install -e .
+          uv pip install --system jupyter quartodoc
+          uv pip install --system -e .
 
       - name: Build autodoc
         run: quartodoc build

.github/workflows/pypi.yml

@@ -3,11 +3,9 @@ name: publish pypi
 on:
   push:
     tags:
-      - "v*"
+      - 'v*'
   workflow_dispatch:
 
-permissions: {}
-
 jobs:
   setup_release:
     name: Create Release
@@ -30,8 +28,7 @@ jobs:
       name: pypi
       url: https://pypi.org/p/axolotl
     permissions:
-      contents: read
-      id-token: write # IMPORTANT: this permission is mandatory for trusted publishing
+      id-token: write  # IMPORTANT: this permission is mandatory for trusted publishing
     steps:
       - name: Check out repository code
         uses: actions/checkout@v4
@@ -41,23 +38,24 @@ jobs:
         with:
           python-version: "3.11"
 
+      - name: Install uv
+        uses: astral-sh/setup-uv@v4
+        with:
+          version: "latest"
+
       - name: Install dependencies
         run: |
-          pip3 install wheel packaging==26.0
-          pip3 install --no-build-isolation -e .
-          pip3 install -r requirements-dev.txt -r requirements-tests.txt
+          uv pip install --system wheel packaging==23.2
+          uv pip install --system --no-build-isolation -e ".[dev]"
 
       - name: Extract tag name
         id: tag
-        run: echo "TAG_NAME=$(echo $GITHUB_REF | cut -d / -f 3)" >> "$GITHUB_OUTPUT"
+        run: echo "TAG_NAME=$(echo "$GITHUB_REF" | cut -d / -f 3)" >> "$GITHUB_OUTPUT"
 
-      - name: Update version in VERSION file
+      - name: Build package
         run: |
-          echo "${{ steps.tag.outputs.TAG_NAME }}" | sed 's/^v//' > VERSION
-
-      - name: Build a source dist
-        run: |
-          python setup.py sdist
+          uv pip install --system build
+          python -m build
 
       - name: Publish package distributions to PyPI
         uses: pypa/gh-action-pypi-publish@release/v1

.github/workflows/tests-nightly.yml

@@ -3,13 +3,6 @@ on:
   workflow_dispatch:
   schedule:
     - cron: '0 0 * * *'  # Runs at 00:00 UTC every day
-  pull_request:
-    types: [opened, synchronize, reopened, ready_for_review]
-    paths:
-      - '.github/workflows/tests-nightly.yml'
-
-permissions:
-  contents: read
 
 jobs:
   pre-commit:
@@ -20,31 +13,19 @@ jobs:
       - uses: actions/setup-python@v5
         with:
           python-version: "3.11"
-          cache: 'pip' # caching pip dependencies
       - uses: pre-commit/action@v3.0.1
         env:
           SKIP: no-commit-to-branch
 
-  prime-cdn-s3-cache:
-    name: Prefetch S3 once to prime the CDN cache
-    runs-on: ubuntu-latest
-    if: ${{ !github.event.pull_request.draft }}
-    timeout-minutes: 10
-    steps:
-      - name: Restore Cache from S3
-        id: hf-cache-restore-s3
-        run: |
-          curl -v -H "Range: bytes=0-1023" -L https://axolotl-ci.b-cdn.net/hf-cache.tar.zst > /dev/null
-
   pytest:
     name: PyTest
     runs-on: ubuntu-latest
-    needs: [prime-cdn-s3-cache]
     strategy:
       fail-fast: false
+      max-parallel: 2
       matrix:
-        python_version: ["3.12"]  # TODO include py3.14 once https://github.com/mistralai/mistral-common/pull/194 is merged
-        pytorch_version: ["2.8.0", "2.9.1", "2.10.0"]
+        python_version: ["3.11"]
+        pytorch_version: ["2.6.0", "2.7.0"]
     timeout-minutes: 20
 
     steps:
@@ -55,38 +36,36 @@ jobs:
         id: hf-cache-restore-s3
         run: |
           mkdir -p /home/runner/.cache/huggingface/hub
-          curl -L https://axolotl-ci.b-cdn.net/hf-cache.tar.zst | tar -xf - -C /home/runner/.cache/huggingface/hub/  --use-compress-program unzstd
+          curl -L https://d1dttdx32dkk5p.cloudfront.net/hf-cache.tar.zst | tar -xf - -C /home/runner/.cache/huggingface/hub/  --use-compress-program unzstd
 
       - name: Setup Python
         uses: actions/setup-python@v5
         with:
           python-version: ${{ matrix.python_version }}
-          cache: 'pip' # caching pip dependencies
 
-      - name: upgrade pip
-        run: |
-          pip3 install --upgrade pip
-          pip3 install --upgrade packaging==26.0 setuptools==78.1.1 wheel
+      - name: Install uv
+        uses: astral-sh/setup-uv@v4
+        with:
+          version: "latest"
 
       - name: Install PyTorch
         run: |
-          pip3 install torch==${{ matrix.pytorch_version }} torchvision
+          uv pip install --system torch==${{ matrix.pytorch_version }} torchvision
 
-      - name: Update requirements.txt
+      - name: Update pyproject.toml for nightly builds
         run: |
-          sed -i 's#^transformers.*#transformers @ git+https://github.com/huggingface/transformers.git@main#' requirements.txt
-          sed -i 's#^peft.*#peft @ git+https://github.com/huggingface/peft.git@main#' requirements.txt
-          sed -i 's#^accelerate.*#accelerate @ git+https://github.com/huggingface/accelerate.git@main#' requirements.txt
-          sed -i 's#^trl.*#trl @ git+https://github.com/huggingface/trl.git@main#' requirements.txt
-          sed -i 's#^datasets.*#datasets @ git+https://github.com/huggingface/datasets.git@main#' requirements.txt
+          sed -i 's#"transformers==.*"#"transformers @ git+https://github.com/huggingface/transformers.git@main"#' pyproject.toml
+          sed -i 's#"peft==.*"#"peft @ git+https://github.com/huggingface/peft.git@main"#' pyproject.toml
+          sed -i 's#"accelerate==.*"#"accelerate @ git+https://github.com/huggingface/accelerate.git@main"#' pyproject.toml
+          sed -i 's#"trl==.*"#"trl @ git+https://github.com/huggingface/trl.git@main"#' pyproject.toml
+          sed -i 's#"datasets==.*"#"datasets @ git+https://github.com/huggingface/datasets.git@main"#' pyproject.toml
 
       - name: Install dependencies
         run: |
-          pip3 show torch
-          pip3 install --no-build-isolation -U -e .
+          uv pip show --system torch
+          uv pip install --system --no-build-isolation -e ".[dev]"
           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: |
@@ -102,9 +81,6 @@ jobs:
           pytest -v --durations=10 tests/patched/
           pytest -v --durations=10 tests/cli/
 
-      - name: cleanup pip cache
-        run: |
-          find "$(pip cache dir)/http-v2" -type f -mtime +14 -exec rm {} \;
 
   docker-e2e-tests:
     if: github.repository_owner == 'axolotl-ai-cloud'
@@ -117,26 +93,19 @@ jobs:
       fail-fast: false
       matrix:
         include:
-          - cuda: 128
-            cuda_version: 12.8.1
+          - cuda: 126
+            cuda_version: 12.6.3
             python_version: "3.11"
-            pytorch: 2.9.1
+            pytorch: 2.6.0
             num_gpus: 1
             axolotl_extras:
             nightly_build: "true"
-          - cuda: 128
-            cuda_version: 12.8.1
+          - cuda: 126
+            cuda_version: 12.6.3
             python_version: "3.11"
-            pytorch: 2.10.0
+            pytorch: 2.7.1
             num_gpus: 1
             axolotl_extras:
-          - cuda: 130
-            cuda_version: 13.0.0
-            python_version: "3.12"
-            pytorch: 2.9.1
-            num_gpus: 1
-            axolotl_extras:
-            dockerfile: "Dockerfile-uv.jinja"
             nightly_build: "true"
     steps:
       - name: Checkout
@@ -145,25 +114,26 @@ jobs:
         uses: actions/setup-python@v5
         with:
           python-version: "3.11"
+      - name: Install uv
+        uses: astral-sh/setup-uv@v4
+        with:
+          version: "latest"
       - name: Install Modal
         run: |
-          python -m pip install --upgrade pip
-          pip install modal==1.3.0.post1 jinja2
+          uv pip install --system modal==1.0.2 jinja2
       - name: Update env vars
         run: |
-          echo "BASE_TAG=main-base-py${{ matrix.python_version }}-cu${{ matrix.cuda }}-${{ matrix.pytorch }}" >> $GITHUB_ENV
+          echo "BASE_TAG=main-base-uv-py${{ matrix.python_version }}-cu${{ matrix.cuda }}-${{ matrix.pytorch }}" >> $GITHUB_ENV
           echo "PYTORCH_VERSION=${{ matrix.pytorch}}" >> $GITHUB_ENV
           echo "AXOLOTL_ARGS=${{ matrix.axolotl_args}}" >> $GITHUB_ENV
           echo "AXOLOTL_EXTRAS=${{ matrix.axolotl_extras}}" >> $GITHUB_ENV
           echo "CUDA=${{ matrix.cuda }}" >> $GITHUB_ENV
           echo "N_GPUS=${{ matrix.num_gpus }}" >> $GITHUB_ENV
-          echo "E2E_DOCKERFILE=${{ matrix.dockerfile || 'Dockerfile.jinja'}}" >> $GITHUB_ENV
           echo "NIGHTLY_BUILD=${{ matrix.nightly_build }}" >> $GITHUB_ENV
+          echo "CODECOV_TOKEN=${{ secrets.CODECOV_TOKEN }}" >> $GITHUB_ENV
       - name: Run tests job on Modal
-        env:
-          CODECOV_TOKEN: ${{ secrets.CODECOV_TOKEN }}
         run: |
-          modal run cicd.e2e_tests
+          modal run -m cicd.e2e_tests
   docker-e2e-multigpu-tests:
     if: github.repository_owner == 'axolotl-ai-cloud'
     # this job needs to be run on self-hosted GPU runners...
@@ -175,10 +145,10 @@ jobs:
       fail-fast: false
       matrix:
         include:
-          - cuda: 128
-            cuda_version: 12.8.1
+          - cuda: 126
+            cuda_version: 12.6.3
             python_version: "3.11"
-            pytorch: 2.9.1
+            pytorch: 2.7.1
             num_gpus: 2
             axolotl_extras:
             nightly_build: "true"
@@ -189,21 +159,23 @@ jobs:
         uses: actions/setup-python@v5
         with:
           python-version: "3.11"
+      - name: Install uv
+        uses: astral-sh/setup-uv@v4
+        with:
+          version: "latest"
       - name: Install Modal
         run: |
-          python -m pip install --upgrade pip
-          pip install modal==1.3.0.post1 jinja2
+          uv pip install --system modal==1.0.2 jinja2
       - name: Update env vars
         run: |
-          echo "BASE_TAG=main-base-py${{ matrix.python_version }}-cu${{ matrix.cuda }}-${{ matrix.pytorch }}" >> $GITHUB_ENV
+          echo "BASE_TAG=main-base-uv-py${{ 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
-        env:
-          CODECOV_TOKEN: ${{ secrets.CODECOV_TOKEN }}
         run: |
           modal run cicd.multigpu

.github/workflows/tests.yml

@@ -7,18 +7,16 @@ on:
       - "main"
     paths:
       - '**.py'
-      - 'requirements.txt'
+      - 'pyproject.toml'
       - '.github/workflows/*.yml'
-      - 'requirements-tests.txt'
       - 'cicd/cicd.sh'
       - 'cicd/Dockerfile.jinja'
   pull_request:
       types: [opened, synchronize, reopened, ready_for_review]
       paths:
        - '**.py'
-       - 'requirements.txt'
+       - 'pyproject.toml'
        - '.github/workflows/*.yml'
-       - 'requirements-tests.txt'
        - 'cicd/cicd.sh'
        - 'cicd/Dockerfile.jinja'
   workflow_dispatch:
@@ -28,9 +26,6 @@ concurrency:
   group: ${{ github.workflow }}-${{ github.ref }}
   cancel-in-progress: ${{ github.ref != 'refs/heads/main' }}
 
-permissions:
-  contents: read
-
 env:
   TRANSFORMERS_IS_CI: "yes"
 
@@ -44,80 +39,55 @@ jobs:
       - uses: actions/setup-python@v5
         with:
           python-version: "3.11"
-          cache: 'pip' # caching pip dependencies
       - uses: pre-commit/action@v3.0.1
         env:
           SKIP: no-commit-to-branch
 
-  prime-cdn-s3-cache:
-    name: Prefetch S3 once to prime the CDN cache
-    runs-on: ubuntu-latest
-    if: ${{ !github.event.pull_request.draft }}
-    timeout-minutes: 10
-    steps:
-      - name: Restore Cache from S3
-        id: hf-cache-restore-s3
-        run: |
-          curl -v -H "Range: bytes=0-1023" -L https://axolotl-ci.b-cdn.net/hf-cache.tar.zst > /dev/null
-
   pytest:
     name: PyTest
     runs-on: ubuntu-latest
     if: ${{ !github.event.pull_request.draft }}
-    needs: [prime-cdn-s3-cache]
+#    needs: [preload-cache]
     strategy:
       fail-fast: false
       matrix:
-        python_version: ["3.12"]  # TODO include py3.14 once https://github.com/mistralai/mistral-common/pull/194 is merged
-        pytorch_version: ["2.8.0", "2.9.1", "2.10.0"]
-#        exclude:
-#          - python_version: "3.14"
-#            pytorch_version: "2.8.0"
-#          - python_version: "3.14"
-#            pytorch_version: "2.9.1"
+        python_version: ["3.11"]
+        pytorch_version: ["2.6.0", "2.7.1", "2.8.0"]
     timeout-minutes: 20
 
     steps:
-      - name: cleanup node
-        run: |
-          sudo rm -rf /usr/share/dotnet /usr/local/lib/android /opt/ghc /opt/hostedtoolcache/CodeQL
-
       - name: Check out repository code
         uses: actions/checkout@v4
 
       - name: Restore Cache from S3
         id: hf-cache-restore-s3
         run: |
-          mkdir -p ~/.cache/huggingface/hub
-          curl -L https://axolotl-ci.b-cdn.net/hf-cache.tar.zst | tar -xpf - -C ~/.cache/huggingface/hub/  --use-compress-program unzstd --strip-components=1
-          ls -ltr ~/.cache/huggingface/hub/
+          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: 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==26.0 setuptools==75.8.0 wheel
+      - name: Install uv
+        uses: astral-sh/setup-uv@v4
+        with:
+          version: "latest"
 
       - name: Install PyTorch
         run: |
-          pip3 install --no-cache-dir torch==${{ matrix.pytorch_version }} torchvision
+          uv pip install --system torch==${{ matrix.pytorch_version }} torchvision
 
       - name: Install dependencies
         run: |
-          pip3 show torch
-          pip3 install --no-cache-dir --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: cleanup pip cache
-        run: |
-          find "$(pip cache dir)/http-v2" -type f -mtime +14 -exec rm {} \;
+          uv pip show --system torch
+          uv pip install --system wheel
+          printf "torch==${{ matrix.pytorch_version }}\n" > torch-constraints.txt
+          uv pip install --system --no-cache-dir --no-build-isolation -e ".[dev]" --constraints torch-constraints.txt
+          set -o pipefail
+          python scripts/unsloth_install.py | bash
+          python scripts/cutcrossentropy_install.py | bash
 
       - name: Make sure PyTorch version wasn't clobbered
         run: |
@@ -129,24 +99,14 @@ jobs:
 
       - name: Pre-Download dataset fixture
         run: |
-          hf download --repo-type=dataset axolotl-ai-internal/axolotl-oss-dataset-fixtures
-
-      - name: Show HF cache
-        run: hf cache ls
+          huggingface-cli download --repo-type=dataset axolotl-ai-internal/axolotl-oss-dataset-fixtures
 
       - name: Run tests
         run: |
-          df -h
-          pytest -v --durations=10 -n4 --dist loadfile --ignore=tests/e2e/ --ignore=tests/patched/ --ignore=tests/cli/ --ignore=tests/monkeypatch/ tests/ --cov=axolotl --cov-report=xml
-          df -h
-          pytest -v --durations=10 tests/monkeypatch/ --cov=axolotl --cov-append --cov-report=xml
-          df -h
-          pytest -v --durations=10 tests/patched/ --cov=axolotl --cov-append --cov-report=xml
-          df -h
-          pytest -v --durations=10 tests/cli/ --cov=axolotl --cov-append --cov-report=xml
-
-      - name: Show HF cache
-        run: hf cache ls
+          python -m pytest -v --durations=10 -n 8 --dist loadfile --cov=axolotl --cov-report=xml --ignore=tests/e2e/ --ignore=tests/patched/ --ignore=tests/cli/ --ignore=tests/monkeypatch/ tests/
+          python -m pytest -v --durations=10 -n 8 --cov=axolotl --cov-append --cov-report=xml tests/monkeypatch/
+          python -m pytest -v --durations=10 -n 8 --cov=axolotl --cov-append --cov-report=xml tests/patched/
+          python -m pytest -v --durations=10 -n 8 --cov=axolotl --cov-append --cov-report=xml tests/cli/
 
       - name: Upload coverage to Codecov
         uses: codecov/codecov-action@v5
@@ -156,65 +116,52 @@ jobs:
           flags: unittests,pytorch-${{ matrix.pytorch_version }}
           fail_ci_if_error: false
 
+
   pytest-sdist:
     name: PyTest from Source Dist
     runs-on: ubuntu-latest
     if: ${{ !github.event.pull_request.draft }}
-    needs: [prime-cdn-s3-cache]
     strategy:
       fail-fast: false
       matrix:
-        python_version: ["3.12"]  # TODO include py3.14 once https://github.com/mistralai/mistral-common/pull/194 is merged
-        pytorch_version: ["2.8.0", "2.9.1", "2.10.0"]
-#        exclude:
-#          - python_version: "3.14"
-#            pytorch_version: "2.8.0"
-#          - python_version: "3.14"
-#            pytorch_version: "2.9.1"
-    timeout-minutes: 30
+        python_version: ["3.11"]
+        pytorch_version: ["2.6.0", "2.7.1", "2.8.0"]
+    timeout-minutes: 20
 
     steps:
-      - name: cleanup node
-        run: |
-          sudo rm -rf /usr/share/dotnet /usr/local/lib/android /opt/ghc /opt/hostedtoolcache/CodeQL
-
       - name: Check out repository code
         uses: actions/checkout@v4
 
       - name: Restore Cache from S3
         id: hf-cache-restore-s3
         run: |
-          mkdir -p ~/.cache/huggingface/hub
-          curl -L https://axolotl-ci.b-cdn.net/hf-cache.tar.zst | tar -xpf - -C ~/.cache/huggingface/hub/  --use-compress-program unzstd --strip-components=1
-          ls -ltr ~/.cache/huggingface/hub/
+          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: 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==26.0 setuptools==75.8.0 setuptools_scm build wheel psutil
+      - name: Install uv
+        uses: astral-sh/setup-uv@v4
+        with:
+          version: "latest"
 
       - name: Install PyTorch
         run: |
-          pip3 install --no-cache-dir torch==${{ matrix.pytorch_version }} torchvision
+          uv pip install --system torch==${{ matrix.pytorch_version }} torchvision
 
       - 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
+          uv pip show --system torch
+          uv pip install --system wheel build setuptools_scm
+          python -m build --sdist
+          printf "torch==${{ matrix.pytorch_version }}\n" > torch-constraints.txt
+          tarball_path=$(echo dist/axolotl*.tar.gz)
+          uv pip install --no-cache-dir --no-build-isolation --system "${tarball_path}[dev]" --constraints torch-constraints.txt
           python scripts/unsloth_install.py | sh
           python scripts/cutcrossentropy_install.py | sh
-          pip3 install -r requirements-dev.txt -r requirements-tests.txt
-
-      - name: cleanup pip cache
-        run: |
-          find "$(pip cache dir)/http-v2" -type f -mtime +14 -exec rm {} \;
 
       - name: Make sure PyTorch version wasn't clobbered
         run: |
@@ -225,19 +172,16 @@ jobs:
           axolotl --help
 
       - name: Show HF cache
-        run: hf cache ls
+        run: huggingface-cli scan-cache
 
       - name: Run tests
         run: |
-          pytest -v --durations=10 -n4 --dist loadfile --ignore=tests/e2e/ --ignore=tests/patched/ --ignore=tests/cli/ --ignore=tests/monkeypatch/ tests/ --cov=axolotl --cov-report=xml
-          pytest -v --durations=10 tests/monkeypatch/ --cov=axolotl --cov-append --cov-report=xml
-          pytest -v --durations=10 tests/cli/
-
-      - name: Show HF cache
-        run: hf cache ls
+          python -m pytest -v --durations=10 -n 8 --dist loadfile --cov=axolotl --cov-report=xml --ignore=tests/e2e/ --ignore=tests/patched/ --ignore=tests/cli/ --ignore=tests/monkeypatch/ tests/
+          python -m pytest -v --durations=10 -n 8 --cov=axolotl --cov-append --cov-report=xml tests/monkeypatch/
+          python -m pytest -v --durations=10 -n 8 tests/cli/
 
   gate-skip-e2e:
-    needs: [pre-commit]
+    needs: [pre-commit, pytest, pytest-sdist]
     runs-on: ubuntu-latest
     outputs:
       skip: ${{ steps.compute.outputs.skip }}
@@ -273,19 +217,25 @@ jobs:
     # this job needs to be run on self-hosted GPU runners...
     runs-on: [self-hosted, modal]
     timeout-minutes: 120
-    needs: [pre-commit, pytest]
+    needs: [pre-commit, pytest, pytest-sdist, gate-skip-e2e]
 
     strategy:
       fail-fast: false
       matrix:
         include:
-          - cuda: 130
-            cuda_version: 13.0.0
-            python_version: "3.12"
-            pytorch: 2.9.1
+          - cuda: 126
+            cuda_version: 12.6.3
+            python_version: "3.11"
+            pytorch: 2.7.1
             num_gpus: 1
             axolotl_extras:
-            dockerfile: "Dockerfile-uv.jinja"
+          - cuda: 126
+            cuda_version: 12.6.3
+            python_version: "3.11"
+            pytorch: 2.7.1
+            num_gpus: 1
+            axolotl_extras:
+            dockerfile: "Dockerfile.jinja"
     steps:
       - name: Checkout
         uses: actions/checkout@v4
@@ -293,23 +243,26 @@ jobs:
         uses: actions/setup-python@v5
         with:
           python-version: "3.11"
+      - name: Install uv
+        uses: astral-sh/setup-uv@v4
+        with:
+          version: "latest"
       - name: Install Modal
         run: |
           python -m pip install --upgrade pip
-          pip install modal==1.3.0.post1 jinja2
+          pip install modal==1.0.2 jinja2 protobuf
       - name: Update env vars
         run: |
-          echo "BASE_TAG=main-base-py${{ matrix.python_version }}-cu${{ matrix.cuda }}-${{ matrix.pytorch }}" >> $GITHUB_ENV
+          echo "BASE_TAG=${{ github.ref_name }}-base-uv-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 }}" >> $GITHUB_ENV
+          echo "CODECOV_TOKEN=${{ secrets.CODECOV_TOKEN }}" >> $GITHUB_ENV
           echo "E2E_DOCKERFILE=${{ matrix.dockerfile || 'Dockerfile.jinja'}}" >> $GITHUB_ENV
       - name: Run tests job on Modal
-        env:
-          CODECOV_TOKEN: ${{ secrets.CODECOV_TOKEN }}
         run: |
           modal run cicd.e2e_tests
 
@@ -329,6 +282,18 @@ jobs:
       fail-fast: false
       matrix:
         include:
+          - cuda: 126
+            cuda_version: 12.6.3
+            python_version: "3.11"
+            pytorch: 2.6.0
+            num_gpus: 1
+            axolotl_extras:
+          - cuda: 128
+            cuda_version: 12.8.1
+            python_version: "3.11"
+            pytorch: 2.7.1
+            num_gpus: 1
+            axolotl_extras:
           - cuda: 128
             cuda_version: 12.8.1
             python_version: "3.11"
@@ -336,24 +301,6 @@ jobs:
             num_gpus: 1
             gpu_type: "B200"
             axolotl_extras: fbgemm-gpu
-          - cuda: 128
-            cuda_version: 12.8.1
-            python_version: "3.11"
-            pytorch: 2.9.1
-            num_gpus: 1
-            axolotl_extras:
-          - cuda: 128
-            cuda_version: 12.8.1
-            python_version: "3.11"
-            pytorch: 2.10.0
-            num_gpus: 1
-            axolotl_extras:
-          - cuda: 130
-            cuda_version: 13.0.0
-            python_version: "3.11"
-            pytorch: 2.9.1
-            num_gpus: 1
-            axolotl_extras:
     steps:
       - name: Checkout
         uses: actions/checkout@v4
@@ -361,13 +308,17 @@ jobs:
         uses: actions/setup-python@v5
         with:
           python-version: "3.11"
+      - name: Install uv
+        uses: astral-sh/setup-uv@v4
+        with:
+          version: "latest"
       - name: Install Modal
         run: |
           python -m pip install --upgrade pip
-          pip install modal==1.3.0.post1 jinja2
+          pip install modal==1.0.2 jinja2 protobuf
       - name: Update env vars
         run: |
-          echo "BASE_TAG=main-base-py${{ matrix.python_version }}-cu${{ matrix.cuda }}-${{ matrix.pytorch }}" >> $GITHUB_ENV
+          echo "BASE_TAG=${{ github.ref_name }}-base-uv-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
@@ -375,10 +326,9 @@ jobs:
           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
-        env:
-          CODECOV_TOKEN: ${{ secrets.CODECOV_TOKEN }}
         run: |
           modal run cicd.e2e_tests
 
@@ -392,10 +342,10 @@ jobs:
       fail-fast: false
       matrix:
         include:
-          - cuda: 128
-            cuda_version: 12.8.1
+          - cuda: 126
+            cuda_version: 12.6.3
             python_version: "3.11"
-            pytorch: 2.9.1
+            pytorch: 2.7.1
             num_gpus: 1
             axolotl_extras:
     steps:
@@ -405,19 +355,24 @@ jobs:
         uses: actions/setup-python@v5
         with:
           python-version: "3.11"
+      - name: Install uv
+        uses: astral-sh/setup-uv@v4
+        with:
+          version: "latest"
       - name: Install Modal
         run: |
           python -m pip install --upgrade pip
-          pip install modal==1.3.0.post1 jinja2
+          pip install modal==1.0.2 jinja2 protobuf
       - name: Update env vars
         run: |
-          echo "BASE_TAG=main-base-py${{ matrix.python_version }}-cu${{ matrix.cuda }}-${{ matrix.pytorch }}" >> $GITHUB_ENV
+          echo "BASE_TAG=${{ github.ref_name }}-base-uv-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 }}" >> $GITHUB_ENV
+          echo "CODECOV_TOKEN=${{ secrets.CODECOV_TOKEN }}" >> $GITHUB_ENV
       - name: Run tests job on Modal
         run: |
           modal run cicd.cleanup

.gitignore

@@ -191,5 +191,5 @@ out/
 # vim
 *.swp
 
-# scm auto-versioning
+# setuptools-scm generated version file
 src/axolotl/_version.py

.pre-commit-config.yaml

@@ -11,13 +11,13 @@ repos:
     -   id: no-commit-to-branch
         args: ['--branch', 'main']
 -   repo: https://github.com/astral-sh/ruff-pre-commit
-    rev: v0.15.4
+    rev: v0.12.12
     hooks:
     -   id: ruff
         args: [--fix]
     -   id: ruff-format
 -   repo: https://github.com/pre-commit/mirrors-mypy
-    rev: v1.19.1
+    rev: v1.17.1
     hooks:
     - id: mypy
       additional_dependencies:
@@ -26,7 +26,7 @@ repos:
             'pydantic>=2.5.3',
         ]
 -   repo: https://github.com/PyCQA/bandit
-    rev: 1.9.4
+    rev: 1.8.6
     hooks:
     -   id: bandit
         args: [

.runpod/Dockerfile

@@ -1,16 +1,14 @@
 FROM axolotlai/axolotl-cloud:main-py3.11-cu124-2.6.0
 
 COPY .runpod/requirements.txt /requirements.txt
-RUN --mount=type=cache,target=/root/.cache/pip \
-    python3 -m pip install --upgrade pip && \
-    python3 -m pip install --upgrade -r /requirements.txt
+RUN curl -LsSf https://astral.sh/uv/install.sh | sh && \
+    /root/.local/bin/uv pip install --system -r /requirements.txt
 
 # Environment settings
 ARG BASE_VOLUME="/runpod-volume"
 ENV BASE_VOLUME=$BASE_VOLUME
 ENV HF_DATASETS_CACHE="${BASE_VOLUME}/huggingface-cache/datasets"
 ENV HUGGINGFACE_HUB_CACHE="${BASE_VOLUME}/huggingface-cache/hub"
-ENV HF_HUB_CACHE="${BASE_VOLUME}/huggingface-cache/hub"
 ENV TRANSFORMERS_CACHE="${BASE_VOLUME}/huggingface-cache/hub"
 
 COPY .runpod/src /src

.runpod/README.md

@@ -123,7 +123,7 @@ datasets:
 | --------------------------------- | -------------------------- | ----------------------------------- |
 | `dataset_prepared_path`           | `"data/last_run_prepared"` | Path for prepared dataset           |
 | `push_dataset_to_hub`             | `""`                       | Push dataset to HF hub              |
-| `dataset_num_proc`                | `4`                        | Number of preprocessing processes   |
+| `dataset_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 |

.runpod/src/config/config.yaml

@@ -39,6 +39,7 @@
 #     type: # linear | dynamic
 #     factor: # float
 
+
 # # Whether you are training a 4-bit GPTQ quantized model
 # gptq: true
 # gptq_groupsize: 128 # group size
@@ -106,7 +107,7 @@
 # push_dataset_to_hub: # repo path
 # # The maximum number of processes to use while preprocessing your input dataset. This defaults to `os.cpu_count()`
 # # if not set.
-# dataset_num_proc: # defaults to os.cpu_count() if not set
+# dataset_processes: # defaults to os.cpu_count() if not set
 # # push checkpoints to hub
 # hub_model_id: # repo path to push finetuned model
 # # how to push checkpoints to hub
@@ -223,6 +224,9 @@
 # eval_table_size: # Approximate number of predictions sent to wandb depending on batch size. Enabled above 0. Default is 0
 # eval_table_max_new_tokens: # Total number of tokens generated for predictions sent to wandb. Default is 128
 
+# # 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.
@@ -348,6 +352,8 @@
 # # Allow overwrite yml config using from cli
 # strict:
 
+
+
 base_model: ${BASE_MODEL}
 base_model_ignore_patterns: ${BASE_MODEL_IGNORE_PATTERNS}
 base_model_config: ${BASE_MODEL_CONFIG}
@@ -406,7 +412,7 @@ chat_template_jinja: ${CHAT_TEMPLATE_JINJA}
 default_system_message: ${DEFAULT_SYSTEM_MESSAGE}
 dataset_prepared_path: ${DATASET_PREPARED_PATH}
 push_dataset_to_hub: ${PUSH_DATASET_TO_HUB}
-dataset_num_proc: ${DATASET_NUM_PROC}
+dataset_processes: ${DATASET_PROCESSES}
 dataset_keep_in_memory: ${DATASET_KEEP_IN_MEMORY}
 hub_model_id: ${HUB_MODEL_ID}
 hub_strategy: ${HUB_STRATEGY}
@@ -506,6 +512,7 @@ profiler_steps: ${PROFILER_STEPS}
 loss_watchdog_threshold: ${LOSS_WATCHDOG_THRESHOLD}
 loss_watchdog_patience: ${LOSS_WATCHDOG_PATIENCE}
 
+save_safetensors: ${SAVE_SAFETENSORS}
 train_on_inputs: ${TRAIN_ON_INPUTS}
 group_by_length: ${GROUP_BY_LENGTH}
 gradient_checkpointing: ${GRADIENT_CHECKPOINTING}

MANIFEST.in

@@ -1,6 +1,5 @@
-include requirements.txt
+include pyproject.toml
 include README.md
 include LICENSE
-include src/setuptools_axolotl_dynamic_dependencies.py
 include src/axolotl/utils/chat_templates/templates/*.jinja
-recursive-include axolotl *.py
+recursive-include src/axolotl *.py

README.md

@@ -29,35 +29,21 @@
 
 ## 🎉 Latest Updates
 
-- 2026/03:
-  - New model support has been added in Axolotl for [Mistral Small 4](https://github.com/axolotl-ai-cloud/axolotl/tree/main/examples/mistral4), [Qwen3.5, Qwen3.5 MoE](https://github.com/axolotl-ai-cloud/axolotl/tree/main/examples/qwen3.5), [GLM-4.7-Flash](https://github.com/axolotl-ai-cloud/axolotl/tree/main/examples/glm47-flash), [GLM-4.6V](https://github.com/axolotl-ai-cloud/axolotl/tree/main/examples/glm46v), and [GLM-4.5-Air](https://github.com/axolotl-ai-cloud/axolotl/tree/main/examples/glm45).
-  - [MoE expert quantization](https://docs.axolotl.ai/docs/expert_quantization.html) support (via `quantize_moe_experts: true`) greatly reduces VRAM when training MoE models (FSDP2 compat).
-- 2026/02:
-  - [ScatterMoE LoRA](https://github.com/axolotl-ai-cloud/axolotl/pull/3410) support. LoRA fine-tuning directly on MoE expert weights using custom Triton kernels.
-  - Axolotl now has support for [SageAttention](https://github.com/axolotl-ai-cloud/axolotl/pull/2823) and [GDPO](https://github.com/axolotl-ai-cloud/axolotl/pull/3353) (Generalized DPO).
-- 2026/01:
-  - New integration for [EAFT](https://github.com/axolotl-ai-cloud/axolotl/pull/3366) (Entropy-Aware Focal Training), weights loss by entropy of the top-k logit distribution, and [Scalable Softmax](https://github.com/axolotl-ai-cloud/axolotl/pull/3338), improves long context in attention.
-- 2025/12:
-  - Axolotl now includes support for [Kimi-Linear](https://docs.axolotl.ai/docs/models/kimi-linear.html), [Plano-Orchestrator](https://docs.axolotl.ai/docs/models/plano.html), [MiMo](https://docs.axolotl.ai/docs/models/mimo.html), [InternVL 3.5](https://docs.axolotl.ai/docs/models/internvl3_5.html), [Olmo3](https://docs.axolotl.ai/docs/models/olmo3.html), [Trinity](https://docs.axolotl.ai/docs/models/trinity.html), and [Ministral3](https://docs.axolotl.ai/docs/models/ministral3.html).
-  - [Distributed Muon Optimizer](https://github.com/axolotl-ai-cloud/axolotl/pull/3264) support has been added for FSDP2 pretraining.
-- 2025/10: New model support has been added in Axolotl for: [Qwen3 Next](https://docs.axolotl.ai/docs/models/qwen3-next.html), [Qwen2.5-vl, Qwen3-vl](https://github.com/axolotl-ai-cloud/axolotl/tree/main/examples/qwen2_5-vl), [Qwen3, Qwen3MoE](https://docs.axolotl.ai/docs/models/qwen3.html), [Granite 4](https://docs.axolotl.ai/docs/models/granite4.html), [HunYuan](https://docs.axolotl.ai/docs/models/hunyuan.html), [Magistral 2509](https://docs.axolotl.ai/docs/models/magistral/vision.html), [Apertus](https://docs.axolotl.ai/docs/models/apertus.html), and [Seed-OSS](https://docs.axolotl.ai/docs/models/seed-oss.html).
+- 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>
 
-- 2025/09: Axolotl now has text diffusion training. Read more [here](https://github.com/axolotl-ai-cloud/axolotl/tree/main/src/axolotl/integrations/diffusion).
-- 2025/08: QAT has been updated to include NVFP4 support. See [PR](https://github.com/axolotl-ai-cloud/axolotl/pull/3107).
-- 2025/07:
-  - ND Parallelism support has been added into Axolotl. Compose Context Parallelism (CP), Tensor Parallelism (TP), and Fully Sharded Data Parallelism (FSDP) within a single node and across multiple nodes. Check out the [blog post](https://huggingface.co/blog/accelerate-nd-parallel) for more info.
-  - Axolotl adds more models: [GPT-OSS](https://docs.axolotl.ai/docs/models/gpt-oss.html), [Gemma 3n](https://docs.axolotl.ai/docs/models/gemma3n.html), [Liquid Foundation Model 2 (LFM2)](https://docs.axolotl.ai/docs/models/LiquidAI.html), and [Arcee Foundation Models (AFM)](https://docs.axolotl.ai/docs/models/arcee.html).
-  - FP8 finetuning with fp8 gather op is now possible in Axolotl via `torchao`. Get started [here](https://docs.axolotl.ai/docs/mixed_precision.html#sec-fp8)!
-  - [Voxtral](https://docs.axolotl.ai/docs/models/voxtral.html), [Magistral 1.1](https://docs.axolotl.ai/docs/models/magistral.html), and [Devstral](https://docs.axolotl.ai/docs/models/devstral.html) with mistral-common tokenizer support has been integrated in Axolotl!
-  - TiledMLP support for single-GPU to multi-GPU training with DDP, DeepSpeed and FSDP support has been added to support Arctic Long Sequence Training. (ALST). See [examples](https://github.com/axolotl-ai-cloud/axolotl/tree/main/examples/alst) for using ALST with Axolotl!
-- 2025/06: Magistral with mistral-common tokenizer support has been added to Axolotl. See [docs](https://docs.axolotl.ai/docs/models/magistral.html) to start training your own Magistral models with Axolotl!
-- 2025/05: Quantization Aware Training (QAT) support has been added to Axolotl. Explore the [docs](https://docs.axolotl.ai/docs/qat.html) to learn more!
-- 2025/04: Llama 4 support has been added in Axolotl. See [docs](https://docs.axolotl.ai/docs/models/llama-4.html) to start training your own Llama 4 models with Axolotl's linearized version!
-- 2025/03: Axolotl has implemented Sequence Parallelism (SP) support. Read the [blog](https://huggingface.co/blog/axolotl-ai-co/long-context-with-sequence-parallelism-in-axolotl) and [docs](https://docs.axolotl.ai/docs/sequence_parallelism.html) to learn how to scale your context length when fine-tuning.
+- 2025/06: 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!
@@ -72,22 +58,16 @@ Axolotl is a free and open-source tool designed to streamline post-training and
 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, GLM-4.6V, InternVL 3.5, Gemma 3n, and audio models like Voxtral with image, video, and audio support.
-- **Training Methods**: Full fine-tuning, LoRA, QLoRA, GPTQ, QAT, Preference Tuning (DPO, IPO, KTO, ORPO), RL (GRPO, GDPO), and Reward Modelling (RM) / Process Reward Modelling (PRM).
+- **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 2/3/4](https://docs.axolotl.ai/docs/attention.html#flash-attention), [Xformers](https://docs.axolotl.ai/docs/attention.html#xformers), [Flex Attention](https://docs.axolotl.ai/docs/attention.html#flex-attention), [SageAttention](https://docs.axolotl.ai/docs/attention.html#sageattention), [Liger Kernel](https://docs.axolotl.ai/docs/custom_integrations.html#liger-kernels), [Cut Cross Entropy](https://docs.axolotl.ai/docs/custom_integrations.html#cut-cross-entropy), [ScatterMoE](https://docs.axolotl.ai/docs/custom_integrations.html#kernels-integration), [Sequence Parallelism (SP)](https://docs.axolotl.ai/docs/sequence_parallelism.html), [LoRA optimizations](https://docs.axolotl.ai/docs/lora_optims.html), [Multi-GPU training (FSDP1, FSDP2, DeepSpeed)](https://docs.axolotl.ai/docs/multi-gpu.html), [Multi-node training (Torchrun, Ray)](https://docs.axolotl.ai/docs/multi-node.html), and many more!
+- **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.
 
-
-
 ## 🚀 Quick Start - LLM Fine-tuning in Minutes
 
-**Requirements**:
-
-- NVIDIA GPU (Ampere or newer for `bf16` and Flash Attention) or AMD GPU
-- Python 3.11
-- PyTorch ≥2.8.0
+**Requirements**: NVIDIA GPU (Ampere+) or AMD GPU, Python 3.11+
 
 ### Google Colab
 
@@ -95,15 +75,35 @@ Features:
 
 ### Installation
 
-#### Using pip
+#### Project setup (uv add)
 
 ```bash
-pip3 install -U packaging==26.0 setuptools==75.8.0 wheel ninja
-pip3 install --no-build-isolation axolotl[flash-attn,deepspeed]
+# Install uv
+curl -LsSf https://astral.sh/uv/install.sh | sh
+
+# Initialize or enter your project
+uv init my-project && cd my-project
+uv add axolotl
+uv pip install flash-attn --no-build-isolation
+source .venv/bin/activate
 
 # Download example axolotl configs, deepspeed configs
 axolotl fetch examples
-axolotl fetch deepspeed_configs  # OPTIONAL
+axolotl fetch deepspeed_configs  # optional
+```
+
+#### Quick try (uv pip)
+
+```bash
+# Install uv if needed
+curl -LsSf https://astral.sh/uv/install.sh | sh
+
+uv pip install axolotl
+uv pip install flash-attn --no-build-isolation
+
+# Download example axolotl configs, deepspeed configs
+axolotl fetch examples
+axolotl fetch deepspeed_configs  # optional
 ```
 
 #### Using Docker
@@ -168,13 +168,6 @@ 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.
 
-## 📈 Telemetry
-
-Axolotl has opt-out telemetry that helps us understand how the project is being used
-and prioritize improvements. We collect basic system information, model types, and
-error rates—never personal data or file paths. Telemetry is enabled by default. To
-disable it, set AXOLOTL_DO_NOT_TRACK=1. For more details, see our [telemetry documentation](https://docs.axolotl.ai/docs/telemetry.html).
-
 ## ❤️ Sponsors
 
 Interested in sponsoring? Contact us at [wing@axolotl.ai](mailto:wing@axolotl.ai)

VERSION

@@ -1 +0,0 @@
-0.16.0.dev0

_quarto.yml

@@ -1,8 +1,6 @@
 project:
   type: website
-  pre-render:
-   - docs/scripts/generate_config_docs.py
-   - docs/scripts/generate_examples_docs.py
+  pre-render: docs/scripts/generate_config_docs.py
 
 quartodoc:
   dir: docs/api
@@ -128,9 +126,11 @@ quartodoc:
         - monkeypatch.mistral_attn_hijack_flash
         - monkeypatch.multipack
         - monkeypatch.relora
+        - monkeypatch.llama_expand_mask
         - monkeypatch.lora_kernels
         - monkeypatch.utils
         - monkeypatch.btlm_attn_hijack_flash
+        - monkeypatch.llama_patch_multipack
         - monkeypatch.stablelm_attn_hijack_flash
         - monkeypatch.trainer_fsdp_optim
         - monkeypatch.transformers_fa_utils
@@ -240,48 +240,7 @@ website:
             - docs/getting-started.qmd
             - docs/installation.qmd
             - docs/inference.qmd
-            - section: "Model Guides"
-              contents:
-                - docs/models/kimi-linear.qmd
-                - docs/models/plano.qmd
-                - docs/models/mimo.qmd
-                - docs/models/internvl3_5.qmd
-                - docs/models/olmo3.qmd
-                - docs/models/trinity.qmd
-                - docs/models/arcee.qmd
-                - section: "Ministral3"
-                  contents:
-                    - docs/models/ministral3.qmd
-                    - docs/models/ministral3/think.qmd
-                    - docs/models/ministral3/vision.qmd
-                - section: "Magistral"
-                  contents:
-                    - docs/models/magistral.qmd
-                    - docs/models/magistral/think.qmd
-                    - docs/models/magistral/vision.qmd
-                - docs/models/ministral.qmd
-                - docs/models/mistral-small.qmd
-                - docs/models/voxtral.qmd
-                - docs/models/devstral.qmd
-                - docs/models/mistral.qmd
-                - docs/models/llama-4.qmd
-                - docs/models/llama-2.qmd
-                - docs/models/qwen3-next.qmd
-                - docs/models/qwen3.qmd
-                - docs/models/gemma3n.qmd
-                - docs/models/apertus.qmd
-                - docs/models/gpt-oss.qmd
-                - docs/models/seed-oss.qmd
-                - docs/models/phi.qmd
-                - docs/models/smolvlm2.qmd
-                - docs/models/granite4.qmd
-                - docs/models/LiquidAI.qmd
-                - docs/models/hunyuan.qmd
-                - docs/models/jamba.qmd
-                - docs/models/orpheus.qmd
-
             - docs/cli.qmd
-            - docs/telemetry.qmd
             - docs/config-reference.qmd
             - text: "API Reference"
               href: docs/api
@@ -318,7 +277,6 @@ website:
             - docs/multipack.qmd
             - docs/mixed_precision.qmd
             - docs/optimizers.qmd
-            - docs/attention.qmd
 
         - section: "Advanced Features"
           contents:
@@ -329,7 +287,6 @@ website:
             - docs/sequence_parallelism.qmd
             - docs/gradient_checkpointing.qmd
             - docs/nd_parallelism.qmd
-            - docs/expert_quantization.qmd
 
         - section: "Troubleshooting"
           contents:

benchmarks/bench_entropy.py

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

benchmarks/bench_scattermoe_lora.py

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

benchmarks/bench_selective_logsoftmax.py

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

cicd/Dockerfile-uv.jinja

@@ -1,54 +0,0 @@
-FROM axolotlai/axolotl-base-uv:{{ BASE_TAG }}
-
-ENV TORCH_CUDA_ARCH_LIST="7.0 7.5 8.0 8.6 9.0+PTX"
-ENV AXOLOTL_EXTRAS="{{ AXOLOTL_EXTRAS }}"
-ENV AXOLOTL_ARGS="{{ AXOLOTL_ARGS }}"
-ENV CUDA="{{ CUDA }}"
-ENV PYTORCH_VERSION="{{ PYTORCH_VERSION }}"
-ENV GITHUB_REF="{{ GITHUB_REF }}"
-ENV GITHUB_SHA="{{ GITHUB_SHA }}"
-ENV NIGHTLY_BUILD="{{ NIGHTLY_BUILD }}"
-ENV HF_HOME="{{ HF_HOME }}"
-
-RUN apt-get update && \
-    apt-get install -y --allow-change-held-packages vim curl nano zstd libnccl2 libnccl-dev ibverbs-providers ibverbs-utils infiniband-diags librdmacm-dev librdmacm1 rdmacm-utils slurm-wlm
-
-WORKDIR /workspace
-
-RUN git clone --depth=1 https://github.com/axolotl-ai-cloud/axolotl.git
-
-WORKDIR /workspace/axolotl
-
-RUN git fetch origin +$GITHUB_REF && \
-    git checkout FETCH_HEAD
-
-# If AXOLOTL_EXTRAS is set, append it in brackets
-RUN if [ "$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==26.0 setuptools==78.1.1
-RUN uv pip install torchvision
-RUN uv pip uninstall causal_conv1d
-RUN if [ "$AXOLOTL_EXTRAS" != "" ] ; then \
-        uv pip install --no-build-isolation -e .[deepspeed,flash-attn,ring-flash-attn,optimizers,ray,$AXOLOTL_EXTRAS] $AXOLOTL_ARGS; \
-    else \
-        uv pip install --no-build-isolation -e .[deepspeed,flash-attn,ring-flash-attn,optimizers,ray] $AXOLOTL_ARGS; \
-    fi
-
-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

cicd/Dockerfile.jinja

@@ -1,6 +1,10 @@
-FROM axolotlai/axolotl-base:{{ BASE_TAG }}
+FROM axolotlai/axolotl-base-uv:{{ BASE_TAG }}
 
-ENV TORCH_CUDA_ARCH_LIST="7.0 7.5 8.0 8.6 8.7 8.9 9.0+PTX"
+SHELL ["/bin/bash", "-euxo", "pipefail", "-c"]
+
+ARG VENV_PYTHON="/workspace/axolotl-venv/bin/python"
+
+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 }}"
@@ -9,10 +13,10 @@ 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"
+ENV VENV_PYTHON=$VENV_PYTHON
 
 RUN apt-get update && \
-    apt-get install -y --allow-change-held-packages vim curl nano zstd libnccl2 libnccl-dev ibverbs-providers ibverbs-utils infiniband-diags librdmacm-dev librdmacm1 rdmacm-utils slurm-wlm
+    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
 
@@ -25,26 +29,27 @@ RUN git fetch origin +$GITHUB_REF && \
 
 # 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; \
+        sed -i 's#"transformers[^"]*"#"transformers @ git+https://github.com/huggingface/transformers.git@main"#' pyproject.toml; \
+        sed -i 's#"peft[^"]*"#"peft @ git+https://github.com/huggingface/peft.git@main"#' pyproject.toml; \
+        sed -i 's#"accelerate[^"]*"#"accelerate @ git+https://github.com/huggingface/accelerate.git@main"#' pyproject.toml; \
+        sed -i 's#"trl[^"]*"#"trl @ git+https://github.com/huggingface/trl.git@main"#' pyproject.toml; \
+        sed -i 's#"datasets[^"]*"#"datasets @ git+https://github.com/huggingface/datasets.git@main"#' pyproject.toml; \
     fi
 
-RUN pip install packaging==26.0 setuptools==78.1.1 psutil
-RUN pip uninstall -y causal_conv1d
+RUN uv pip install --python "$VENV_PYTHON" packaging==23.2 setuptools==75.8.0 pip
 RUN if [ "$AXOLOTL_EXTRAS" != "" ] ; then \
-        pip install --no-build-isolation -e .[deepspeed,flash-attn,ring-flash-attn,optimizers,ray,$AXOLOTL_EXTRAS] $AXOLOTL_ARGS; \
+        uv pip install --python "$VENV_PYTHON" --no-build-isolation -e .[ring-flash-attn,optimizers,ray,${AXOLOTL_EXTRAS}] $AXOLOTL_ARGS; \
     else \
-        pip install --no-build-isolation -e .[deepspeed,flash-attn,ring-flash-attn,optimizers,ray] $AXOLOTL_ARGS; \
+        uv pip install --python "$VENV_PYTHON" --no-build-isolation -e .[ring-flash-attn,optimizers,ray] $AXOLOTL_ARGS; \
     fi
 
-RUN python scripts/unsloth_install.py | sh
-RUN python scripts/cutcrossentropy_install.py | sh
+RUN uv pip install --python "$VENV_PYTHON" --no-build-isolation flash-attn $AXOLOTL_ARGS
+
+RUN "$VENV_PYTHON" scripts/unsloth_install.py | sh
+RUN "$VENV_PYTHON" scripts/cutcrossentropy_install.py | sh
 
 # So we can test the Docker image
-RUN pip install -r requirements-dev.txt -r requirements-tests.txt
+RUN uv pip install --python "$VENV_PYTHON" -e ".[dev]"
 
 # fix so that git fetch/pull from remote works
 RUN git config remote.origin.fetch "+refs/heads/*:refs/remotes/origin/*" && \

cicd/cicd.sh

@@ -3,16 +3,8 @@ set -e
 
 python -c "import torch; assert '$PYTORCH_VERSION' in torch.__version__"
 
-set -o pipefail
-curl --silent --show-error --fail --retry 3 --retry-delay 5 -L https://axolotl-ci.b-cdn.net/hf-cache.tar.zst | tar -xpf - -C "${HF_HOME}/hub/" --use-compress-program unzstd --strip-components=1
-# hf download "NousResearch/Meta-Llama-3-8B"
-# hf download "NousResearch/Meta-Llama-3-8B-Instruct"
-# hf download "microsoft/Phi-4-reasoning"
-# hf download "microsoft/Phi-3.5-mini-instruct"
-# hf download "microsoft/Phi-3-medium-128k-instruct"
-
 # Run unit tests with initial coverage report
-pytest -v --durations=10 -n8 \
+uv run pytest -v --durations=10 -n8 \
   --ignore=tests/e2e/ \
   --ignore=tests/patched/ \
   --ignore=tests/cli \
@@ -20,36 +12,36 @@ pytest -v --durations=10 -n8 \
   --cov=axolotl
 
 # Run lora kernels tests with coverage append
-pytest -v --durations=10 \
+uv run pytest -v --durations=10 \
   /workspace/axolotl/tests/e2e/patched/lora_kernels \
   --cov=axolotl \
   --cov-append
 
 # Run patched tests excluding lora kernels with coverage append
-pytest --full-trace -vvv --durations=10 \
+uv run pytest --full-trace -vvv --durations=10 \
   --ignore=tests/e2e/patched/lora_kernels \
   /workspace/axolotl/tests/e2e/patched \
   --cov=axolotl \
   --cov-append
 
 # Run solo tests with coverage append
-pytest -v --durations=10 -n1 \
+uv run pytest -v --durations=10 -n1 \
   /workspace/axolotl/tests/e2e/solo/ \
   --cov=axolotl \
   --cov-append
 
 # Run integration tests with coverage append
-pytest -v --durations=10 \
+uv run pytest -v --durations=10 \
   /workspace/axolotl/tests/e2e/integrations/ \
   --cov=axolotl \
   --cov-append
 
-pytest -v --durations=10 /workspace/axolotl/tests/cli \
+uv run pytest -v --durations=10 /workspace/axolotl/tests/cli \
   --cov=axolotl \
   --cov-append
 
 # Run remaining e2e tests with coverage append and final report
-pytest -v --durations=10 \
+uv run pytest -v --durations=10 \
   --ignore=tests/e2e/solo/ \
   --ignore=tests/e2e/patched/ \
   --ignore=tests/e2e/multigpu/ \
@@ -60,4 +52,4 @@ pytest -v --durations=10 \
   --cov-append \
   --cov-report=xml:e2e-coverage.xml
 
-codecov upload-process -t $CODECOV_TOKEN -f e2e-coverage.xml -F e2e,pytorch-${PYTORCH_VERSION} || true
+uv run codecov upload-process -t $CODECOV_TOKEN -f e2e-coverage.xml -F e2e,pytorch-${PYTORCH_VERSION} || true

cicd/multigpu.py

@@ -17,22 +17,18 @@ 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_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"),
+    "BASE_TAG": os.environ.get("BASE_TAG", "main-base-uv-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)

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 --maxfail=3 \
+pytest -v --durations=10 -n2 \
   --ignore=/workspace/axolotl/tests/e2e/multigpu/solo/ \
   --ignore=/workspace/axolotl/tests/e2e/multigpu/patched/ \
   /workspace/axolotl/tests/e2e/multigpu/ \

cicd/single_gpu.py

@@ -23,7 +23,7 @@ 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"),
+    "BASE_TAG": os.environ.get("BASE_TAG", "main-base-uv-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", ""),
@@ -65,9 +65,8 @@ def run_cmd(cmd: str, run_folder: str):
     import subprocess  # nosec
 
     sp_env = os.environ.copy()
-    sp_env["AXOLOTL_DATASET_NUM_PROC"] = "8"
+    sp_env["AXOLOTL_DATASET_PROCESSES"] = "8"
 
     # Propagate errors from subprocess.
-    exit_code = subprocess.call(cmd.split(), cwd=run_folder, env=sp_env)  # nosec
-    if exit_code:
-        raise RuntimeError(f"Command '{cmd}' failed with exit code {exit_code}")
+    if exit_code := subprocess.call(cmd.split(), cwd=run_folder, env=sp_env):  # nosec
+        exit(exit_code)

codecov.yml

@@ -37,7 +37,6 @@ coverage:
         only_pulls: false
         flags: null
         paths: null
-        informational: true
 
 parsers:
   gcov:

devtools/dev_chat_template.yml

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

docker/Dockerfile

@@ -1,14 +1,19 @@
-ARG BASE_TAG=main-base
-FROM axolotlai/axolotl-base:$BASE_TAG
+ARG BASE_TAG=main-base-uv
+FROM axolotlai/axolotl-base-uv:$BASE_TAG
 
 ARG TORCH_CUDA_ARCH_LIST="7.0 7.5 8.0 8.6+PTX"
 ARG AXOLOTL_EXTRAS=""
 ARG AXOLOTL_ARGS=""
 ARG CUDA="118"
 ARG PYTORCH_VERSION="2.1.2"
-ARG TARGETARCH
+ARG GIT_REF="refs/heads/main"
+ARG GIT_SHA="HEAD"
+ARG VENV_PYTHON="/workspace/axolotl-venv/bin/python"
 
 ENV PYTORCH_VERSION=$PYTORCH_VERSION
+ENV GIT_REF=$GIT_REF
+ENV GIT_SHA=$GIT_SHA
+ENV VENV_PYTHON=$VENV_PYTHON
 
 RUN apt-get update && \
     apt-get install -y --allow-change-held-packages vim curl nano libnccl2 libnccl-dev rsync s3fs && \
@@ -21,21 +26,19 @@ RUN git clone --depth=1 https://github.com/axolotl-ai-cloud/axolotl.git
 
 WORKDIR /workspace/axolotl
 
-# If AXOLOTL_EXTRAS is set, append it in brackets; don't install deepspeed with arm64
-RUN pip uninstall -y causal_conv1d
-RUN if [ "$TARGETARCH" = "arm64" ]; then \
-        BASE_EXTRAS="flash-attn,ring-flash-attn,optimizers,ray"; \
+# Ensure we are on the expected commit and break Docker cache between revisions
+RUN git fetch origin "$GIT_REF" && git checkout "$GIT_SHA"
+
+# If AXOLOTL_EXTRAS is set, append it in brackets
+RUN if [ "$AXOLOTL_EXTRAS" != "" ] ; then \
+        uv pip install --python "$VENV_PYTHON" --no-build-isolation -e .[ring-flash-attn,optimizers,ray,$AXOLOTL_EXTRAS] $AXOLOTL_ARGS; \
     else \
-        BASE_EXTRAS="deepspeed,flash-attn,ring-flash-attn,optimizers,ray"; \
+        uv pip install --python "$VENV_PYTHON" --no-build-isolation -e .[ring-flash-attn,optimizers,ray] $AXOLOTL_ARGS; \
     fi && \
-    if [ "$AXOLOTL_EXTRAS" != "" ]; then \
-        pip install --no-build-isolation -e .[$BASE_EXTRAS,$AXOLOTL_EXTRAS] $AXOLOTL_ARGS; \
-    else \
-        pip install --no-build-isolation -e .[$BASE_EXTRAS] $AXOLOTL_ARGS; \
-    fi && \    python scripts/unsloth_install.py | sh && \
-    python scripts/cutcrossentropy_install.py | sh && \
-    pip install pytest && \
-    pip cache purge
+    uv pip install --python "$VENV_PYTHON" --no-build-isolation flash-attn $AXOLOTL_ARGS && \
+    "$VENV_PYTHON" scripts/unsloth_install.py | sh && \
+    "$VENV_PYTHON" scripts/cutcrossentropy_install.py | sh && \
+    uv pip install --python "$VENV_PYTHON" pytest
 
 # fix so that git fetch/pull from remote works with shallow clone
 RUN git config remote.origin.fetch "+refs/heads/*:refs/remotes/origin/*" && \

docker/Dockerfile-base

@@ -2,16 +2,14 @@ ARG CUDA_VERSION="11.8.0"
 ARG CUDNN_VERSION="8"
 ARG UBUNTU_VERSION="22.04"
 ARG MAX_JOBS=4
-ARG TARGETARCH
 
 FROM nvidia/cuda:$CUDA_VERSION-cudnn$CUDNN_VERSION-devel-ubuntu$UBUNTU_VERSION AS base-builder
 
 ENV PATH="/root/miniconda3/bin:${PATH}"
 
-ARG TARGETARCH
-ARG PYTHON_VERSION="3.11"
+ARG PYTHON_VERSION="3.10"
 ARG PYTORCH_VERSION="2.1.2"
-ARG CUDA="128"
+ARG CUDA="118"
 ARG TORCH_CUDA_ARCH_LIST="7.0 7.5 8.0 8.6 9.0+PTX"
 
 ENV PYTHON_VERSION=$PYTHON_VERSION
@@ -24,17 +22,11 @@ RUN apt-get update \
         librdmacm-dev librdmacm1 rdmacm-utils slurm-wlm \
     && rm -rf /var/cache/apt/archives \
     && rm -rf /var/lib/apt/lists/* \
-    && if [ "$TARGETARCH" = "amd64" ]; then \
-        MINICONDA_ARCH="x86_64"; \
-    elif [ "$TARGETARCH" = "arm64" ]; then \
-        MINICONDA_ARCH="aarch64"; \
-    else \
-        echo "Unsupported architecture: $TARGETARCH"; exit 1; \
-    fi \
-    && wget https://repo.anaconda.com/miniconda/Miniconda3-latest-Linux-${MINICONDA_ARCH}.sh \
+    && wget \
+    https://repo.anaconda.com/miniconda/Miniconda3-latest-Linux-x86_64.sh \
     && mkdir /root/.conda \
-    && bash Miniconda3-latest-Linux-${MINICONDA_ARCH}.sh -b \
-    && rm -f Miniconda3-latest-Linux-${MINICONDA_ARCH}.sh \
+    && 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}"
@@ -43,34 +35,18 @@ ENV PATH="/root/miniconda3/envs/py${PYTHON_VERSION}/bin:${PATH}"
 
 WORKDIR /workspace
 
-RUN python3 -m pip install --upgrade pip && pip3 install -U packaging==26.0 setuptools==75.8.0 wheel psutil && \
+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
 
-RUN if [ "$CUDA" != "130" ] ; then \
-        CAUSAL_CONV1D_FORCE_CXX11_ABI=TRUE CAUSAL_CONV1D_FORCE_BUILD=TRUE python3 -m pip install --no-cache-dir "causal_conv1d @ git+https://github.com/Dao-AILab/causal-conv1d.git@v1.5.4"; \
-        python3 -m pip install --no-cache-dir "mamba_ssm @ git+https://github.com/state-spaces/mamba.git@main"; \
-        python3 -m pip cache purge; \
-    fi
-
 RUN 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
 
-# Map Python version (e.g., 3.12 -> cp312)
-RUN PYTHON_CP="cp$(echo $PYTHON_VERSION | tr -d '.')" && \
-    # Map PyTorch version (e.g., 2.9.1 -> torch2.9, 2.10.0 -> torch2.10)
-    TORCH_TAG="torch$(echo $PYTORCH_VERSION | grep -oP '^\d+\.\d+')" && \
-    # Map architecture
-    case "$TARGETARCH" in \
-        amd64) ARCH_TAG="x86_64" ;; \
-        arm64) ARCH_TAG="aarch64" ;; \
-        *) echo "Unsupported architecture: $TARGETARCH"; exit 1 ;; \
-    esac && \
-    WHL_VERSION="v0.7.16" && \
-    WHL_FILE="flash_attn-2.8.3+cu${CUDA}${TORCH_TAG}-${PYTHON_CP}-${PYTHON_CP}-linux_${ARCH_TAG}.whl" && \
-    wget -nv "https://github.com/mjun0812/flash-attention-prebuild-wheels/releases/download/${WHL_VERSION}/${WHL_FILE}" && \
-    pip3 install --no-cache-dir "${WHL_FILE}" && \
-    rm "${WHL_FILE}"
+RUN if [ "$PYTORCH_VERSION" = "2.6.0" ] && [ "$CUDA" = "124" ] ; then \
+        FLASH_ATTENTION_FORCE_BUILD="TRUE" uv pip install --no-build-isolation flash-attn==2.8.0.post2; \
+    fi

docker/Dockerfile-base-nightly

@@ -30,7 +30,7 @@ ENV PATH="/root/miniconda3/envs/py${PYTHON_VERSION}/bin:${PATH}"
 
 WORKDIR /workspace
 
-RUN python3 -m pip install --upgrade pip && pip3 install -U packaging==26.0 setuptools==75.8.0 wheel && \
+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 --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" && \

docker/Dockerfile-cloud

@@ -12,8 +12,8 @@ EXPOSE 22
 COPY scripts/cloud-entrypoint.sh /root/cloud-entrypoint.sh
 COPY scripts/motd /etc/motd
 
-RUN pip install jupyterlab notebook ipywidgets && \
-    jupyter lab clean
+RUN uv pip install --python "$VENV_PYTHON" jupyterlab notebook ipywidgets && \
+    "$VENV_PYTHON" -m jupyter lab clean
 RUN apt update && \
     apt install --yes --no-install-recommends openssh-server tmux iproute2 nvtop && \
     rm -rf /var/cache/apt/archives && \

docker/Dockerfile-cloud-no-tmux

@@ -12,8 +12,8 @@ EXPOSE 22
 COPY scripts/cloud-entrypoint.sh /root/cloud-entrypoint.sh
 COPY scripts/motd /etc/motd
 
-RUN pip install jupyterlab notebook ipywidgets && \
-    jupyter lab clean
+RUN uv pip install --python "$VENV_PYTHON" jupyterlab notebook ipywidgets && \
+    "$VENV_PYTHON" -m 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 && \

docker/Dockerfile-cloud-uv

@@ -1,30 +0,0 @@
-ARG BASE_TAG=main
-FROM axolotlai/axolotl-uv:$BASE_TAG
-
-ENV HF_DATASETS_CACHE="/workspace/data/huggingface-cache/datasets"
-ENV HF_HUB_CACHE="/workspace/data/huggingface-cache/hub"
-ENV HF_HOME="/workspace/data/huggingface-cache/hub"
-ENV HF_HUB_ENABLE_HF_TRANSFER="1"
-
-EXPOSE 8888
-EXPOSE 22
-
-COPY scripts/cloud-entrypoint.sh /root/cloud-entrypoint.sh
-COPY scripts/motd /etc/motd
-
-RUN uv pip install jupyterlab notebook ipywidgets && \
-    jupyter lab clean
-RUN apt update && \
-    apt install --yes --no-install-recommends openssh-server tmux iproute2 nvtop && \
-    rm -rf /var/cache/apt/archives && \
-    rm -rf /var/lib/apt/lists/* && \
-    mkdir -p ~/.ssh && \
-    chmod 700 ~/.ssh && \
-    printf "\n[[ -z \"\$TMUX\"  ]] && { tmux attach-session -t ssh_tmux || tmux new-session -s ssh_tmux; exit; }\n" >> ~/.bashrc && \
-    printf "[ ! -z \"\$TERM\" -a -r /etc/motd ] && cat /etc/motd\n" >> ~/.bashrc && \
-    chmod +x /workspace/axolotl/scripts/cloud-entrypoint.sh && \
-    chmod +x /root/cloud-entrypoint.sh && \
-    echo 'set-option -g history-limit 5000' >> ~/.tmux.conf
-
-ENTRYPOINT ["/root/cloud-entrypoint.sh"]
-CMD ["sleep", "infinity"]

docker/Dockerfile-tests

@@ -24,13 +24,14 @@ RUN git fetch origin +$GITHUB_REF && \
 
 # If AXOLOTL_EXTRAS is set, append it in brackets
 RUN if [ "$AXOLOTL_EXTRAS" != "" ] ; then \
-        pip install --no-build-isolation -e .[deepspeed,flash-attn,mamba-ssm,$AXOLOTL_EXTRAS] $AXOLOTL_ARGS; \
+        uv pip install --no-build-isolation -e .[deepspeed,mamba-ssm,$AXOLOTL_EXTRAS] $AXOLOTL_ARGS; \
     else \
-        pip install --no-build-isolation -e .[deepspeed,flash-attn,mamba-ssm] $AXOLOTL_ARGS; \
-    fi
+        uv pip install --no-build-isolation -e .[deepspeed,mamba-ssm] $AXOLOTL_ARGS; \
+    fi && \
+    uv pip install --no-build-isolation flash-attn $AXOLOTL_ARGS
 
 # So we can test the Docker image
-RUN pip install pytest
+RUN uv pip install pytest
 
 # fix so that git fetch/pull from remote works
 RUN git config remote.origin.fetch "+refs/heads/*:refs/remotes/origin/*" && \

docker/Dockerfile-uv

@@ -1,48 +0,0 @@
-ARG BASE_TAG=main-base
-FROM axolotlai/axolotl-base-uv:$BASE_TAG
-
-ARG TORCH_CUDA_ARCH_LIST="7.0 7.5 8.0 8.6+PTX"
-ARG AXOLOTL_EXTRAS=""
-ARG AXOLOTL_ARGS=""
-ARG CUDA="118"
-ARG PYTORCH_VERSION="2.1.2"
-ARG TARGETARCH
-
-ENV PYTORCH_VERSION=$PYTORCH_VERSION
-
-RUN apt-get update && \
-    apt-get install -y --allow-change-held-packages vim curl nano libnccl2 libnccl-dev rsync s3fs && \
-    rm -rf /var/cache/apt/archives && \
-    rm -rf /var/lib/apt/lists/*
-
-WORKDIR /workspace
-
-RUN git clone --depth=1 https://github.com/axolotl-ai-cloud/axolotl.git
-
-WORKDIR /workspace/axolotl
-
-# If AXOLOTL_EXTRAS is set, append it in brackets; don't install deepspeed with arm64
-RUN uv pip uninstall causal_conv1d
-RUN if [ "$TARGETARCH" = "arm64" ]; then \
-        BASE_EXTRAS="flash-attn,ring-flash-attn,optimizers,ray"; \
-    else \
-        BASE_EXTRAS="deepspeed,flash-attn,ring-flash-attn,optimizers,ray"; \
-    fi && \
-    if [ "$AXOLOTL_EXTRAS" != "" ]; then \
-        uv pip install --no-build-isolation -e .[$BASE_EXTRAS,$AXOLOTL_EXTRAS] $AXOLOTL_ARGS; \
-    else \
-        uv pip install --no-build-isolation -e .[$BASE_EXTRAS] $AXOLOTL_ARGS; \
-    fi && \
-    python scripts/unsloth_install.py --uv | sh && \
-    python scripts/cutcrossentropy_install.py --uv | sh && \
-    uv pip install pytest && \
-    uv cache clean
-
-# fix so that git fetch/pull from remote works with shallow clone
-RUN git config remote.origin.fetch "+refs/heads/*:refs/remotes/origin/*" && \
-    git config --get remote.origin.fetch && \
-    git config --global credential.helper store
-
-COPY .axolotl-complete.bash /root/.axolotl-complete.bash
-RUN chmod +x /root/.axolotl-complete.bash && \
-    echo 'source /root/.axolotl-complete.bash' >> ~/.bashrc

docker/Dockerfile-uv-base

@@ -2,11 +2,9 @@ ARG CUDA_VERSION="12.6.3"
 ARG CUDNN_VERSION=""
 ARG UBUNTU_VERSION="22.04"
 ARG MAX_JOBS=4
-ARG TARGETARCH
 
 FROM nvidia/cuda:$CUDA_VERSION-cudnn$CUDNN_VERSION-devel-ubuntu$UBUNTU_VERSION AS base-builder
 
-ARG TARGETARCH
 ARG PYTHON_VERSION="3.11"
 ARG PYTORCH_VERSION="2.6.0"
 ARG CUDA="126"
@@ -15,6 +13,7 @@ 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}"
+ENV VENV_PYTHON=/workspace/axolotl-venv/bin/python
 
 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/* \
@@ -31,27 +30,8 @@ RUN uv venv --no-project --relocatable axolotl-venv
 
 ENV PATH="/workspace/axolotl-venv/bin:${PATH}"
 
-RUN uv pip install packaging setuptools wheel psutil \
-    && uv pip install torch==${PYTORCH_VERSION} torchvision \
-    && uv pip install awscli pydantic
-
-RUN if [ "$TARGETARCH" = "amd64" ]; then \
-        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"; \
-    fi
-
-# Map Python version (e.g., 3.12 -> cp312)
-RUN PYTHON_CP="cp$(echo $PYTHON_VERSION | tr -d '.')" && \
-    # Map PyTorch version (e.g., 2.9.1 -> torch2.9, 2.10.0 -> torch2.10)
-    TORCH_TAG="torch$(echo $PYTORCH_VERSION | grep -oP '^\d+\.\d+')" && \
-    # Map architecture
-    case "$TARGETARCH" in \
-        amd64) ARCH_TAG="x86_64" ;; \
-        arm64) ARCH_TAG="aarch64" ;; \
-        *) echo "Unsupported architecture: $TARGETARCH"; exit 1 ;; \
-    esac && \
-    WHL_VERSION="v0.7.16" && \
-    WHL_FILE="flash_attn-2.8.3+cu${CUDA}${TORCH_TAG}-${PYTHON_CP}-${PYTHON_CP}-linux_${ARCH_TAG}.whl" && \
-    wget -nv "https://github.com/mjun0812/flash-attention-prebuild-wheels/releases/download/${WHL_VERSION}/${WHL_FILE}" && \
-    uv pip install --no-cache-dir "${WHL_FILE}" && \
-    rm "${WHL_FILE}"
+RUN uv pip install --python "$VENV_PYTHON" packaging setuptools wheel psutil protobuf grpclib \
+    && uv pip install --python "$VENV_PYTHON" torch==${PYTORCH_VERSION} \
+    && uv pip install --python "$VENV_PYTHON" --no-build-isolation "causal_conv1d @ git+https://github.com/Dao-AILab/causal-conv1d.git@main" \
+    && uv pip install --python "$VENV_PYTHON" "mamba_ssm @ git+https://github.com/state-spaces/mamba.git@main" \
+    && uv pip install --python "$VENV_PYTHON" awscli pydantic

docs/.gitignore

@@ -3,5 +3,3 @@ _site/
 /api/*.qmd
 /api/*.html
 config-reference.qmd
-models/**/*.qmd
-models/**/*.html

docs/amd_hpc.qmd

@@ -86,7 +86,7 @@ export HF_DATASETS_OFFLINE=1
 Download a base model using the Hugging Face CLI:
 
 ```bash
-hf download meta-llama/Meta-Llama-3.1-8B --local-dir ~/hfdata/llama3.1-8B
+huggingface-cli download meta-llama/Meta-Llama-3.1-8B --local-dir ~/hfdata/llama3.1-8B
 ```
 
 ### 10. Create Axolotl Configuration

docs/attention.qmd

@@ -1,178 +0,0 @@
----
-title: Attention
-description: Supported attention modules in Axolotl
----
-
-## SDP Attention
-
-This is the default built-in attention in PyTorch.
-
-```yaml
-sdp_attention: true
-```
-
-For more details: [PyTorch docs](https://docs.pytorch.org/docs/stable/generated/torch.nn.functional.scaled_dot_product_attention.html)
-
-## Flash Attention
-
-Axolotl supports Flash Attention 2, 3, and 4. The best available version is used automatically
-based on your installed packages and GPU.
-
-```yaml
-flash_attention: true
-```
-
-For more details: [Flash Attention](https://github.com/Dao-AILab/flash-attention/)
-
-### Flash Attention 2
-
-Requirements: Ampere, Ada, or Hopper GPUs (Turing or lower not supported)
-
-```bash
-pip install flash-attn --no-build-isolation
-```
-
-::: {.callout-tip}
-
-If you get `undefined symbol` while training, ensure you installed PyTorch prior to Axolotl.
-Alternatively, try reinstall or downgrade a version.
-
-:::
-
-### Flash Attention 3
-
-Requirements: Hopper only and CUDA 12.8 (recommended)
-
-```bash
-git clone https://github.com/Dao-AILab/flash-attention.git
-cd flash-attention/hopper
-
-python setup.py install
-```
-
-### Flash Attention 4
-
-Requirements: Hopper or Blackwell GPUs
-
-```bash
-pip install flash-attn-4
-```
-
-Or from source:
-
-```bash
-git clone https://github.com/Dao-AILab/flash-attention.git
-cd flash-attention/flash_attn/cute
-
-pip install -e .
-
-# FA2's flash_attn package includes a cute/ stub that shadows FA4.
-# Remove it so Python can find the real FA4 module:
-rm -r $(python -c "import flash_attn; print(flash_attn.__path__[0])")/cute
-```
-
-::: {.callout-note}
-
-**Hopper (SM90) users**: The backward kernel is not yet included in the pip package. To use FA4
-for training on Hopper, install from source using the instructions above.
-
-:::
-
-::: {.callout-warning}
-
-FA4 only supports head dimensions up to 128 (`d ≤ 128`). The DeepSeek shape `(192, 128)` is
-also supported but only on Blackwell. Axolotl automatically detects incompatible head dimensions
-and falls back to FA2/3.
-
-:::
-
-For more details: [flash-attention/flash_attn/cute](https://github.com/Dao-AILab/flash-attention/tree/main/flash_attn/cute)
-
-### AMD
-
-Requirements: ROCm 6.0 and above.
-
-See [Flash Attention AMD docs](https://github.com/Dao-AILab/flash-attention/tree/main?tab=readme-ov-file#amd-rocm-support).
-
-## Flex Attention
-
-A flexible PyTorch API for attention used in combination with `torch.compile`.
-
-```yaml
-flex_attention: true
-
-# recommended
-torch_compile: true
-```
-
-::: {.callout-note}
-
-We recommend using latest stable version of PyTorch for best performance.
-
-:::
-
-For more details: [PyTorch docs](https://pytorch.org/blog/flexattention/)
-
-## SageAttention
-
-Attention kernels with QK Int8 and PV FP16 accumulator.
-
-```yaml
-sage_attention: true
-```
-
-Requirements: Ampere, Ada, or Hopper GPUs
-
-```bash
-pip install sageattention==2.2.0 --no-build-isolation
-```
-
-::: {.callout-warning}
-
-Only LoRA/QLoRA recommended at the moment. We found loss drop to 0 for full finetuning. See [GitHub Issue](https://github.com/thu-ml/SageAttention/issues/198).
-
-:::
-
-For more details: [Sage Attention](https://github.com/thu-ml/SageAttention)
-
-::: {.callout-note}
-
-We do not support SageAttention 3 at the moment. If you are interested on adding this or improving SageAttention implementation, please make an Issue.
-
-:::
-
-
-## xFormers
-
-```yaml
-xformers_attention: true
-```
-
-::: {.callout-tip}
-
-We recommend using with Turing GPUs or below (such as on Colab).
-
-:::
-
-For more details: [xFormers](https://github.com/facebookresearch/xformers)
-
-## Shifted Sparse Attention
-
-::: {.callout-warning}
-
-We plan to deprecate this! If you use this feature, we recommend switching to methods above.
-
-:::
-
-Requirements: LLaMA model architecture
-
-```yaml
-flash_attention: true
-s2_attention: true
-```
-
-::: {.callout-tip}
-
-No sample packing support!
-
-:::

docs/checkpoint_saving.qmd

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

docs/cli.qmd

@@ -210,8 +210,6 @@ axolotl lm-eval config.yml
 Configuration options:
 
 ```yaml
-lm_eval_model: # model to evaluate (local or hf path)
-
 # List of tasks to evaluate
 lm_eval_tasks:
   - arc_challenge
@@ -220,7 +218,7 @@ lm_eval_batch_size: # Batch size for evaluation
 output_dir: # Directory to save evaluation results
 ```
 
-See [LM Eval Harness integration docs](https://docs.axolotl.ai/docs/custom_integrations.html#language-model-evaluation-harness-lm-eval) for full configuration details.
+See [LM Eval Harness](https://github.com/EleutherAI/lm-evaluation-harness) for more details.
 
 ### delinearize-llama4
 

docs/dataset-formats/conversation.qmd

@@ -218,13 +218,6 @@ If you have tool arguments with same name but different dtypes (like `"time": st
 ```
 "arguments": "{\"...\": \"...\"}"
 ```
-
-The same is applicable for tool parameters.
-
-```
-"parameters": "{\"...\": \"...\"}"
-```
-
 :::
 
 Example config for Llama4:

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
@@ -72,8 +72,8 @@ datasets:
 Make sure you have an [editable install](https://setuptools.pypa.io/en/latest/userguide/development_mode.html) of Axolotl, which ensures that changes you make to the code are reflected at runtime.  Run the following commands from the root of this project:
 
 ```bash
-pip3 install packaging
-pip3 install --no-build-isolation -e '.[flash-attn,deepspeed]'
+uv sync --extra deepspeed
+uv pip install flash-attn --no-build-isolation
 ```
 
 #### Remote Hosts
@@ -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
@@ -213,8 +213,8 @@ docker run --privileged --gpus '"all"' --shm-size 10g --rm -it --name axolotl --
 You will now be in the container.  Next, perform an editable install of Axolotl:
 
 ```bash
-pip3 install packaging
-pip3 install --no-build-isolation -e '.[flash-attn,deepspeed]'
+uv sync --extra deepspeed
+uv pip install flash-attn --no-build-isolation
 ```
 
 ### Attach To Container

docs/docker.qmd

@@ -32,8 +32,11 @@ main-base-py{python_version}-cu{cuda_version}-{pytorch_version}
 
 Tags examples:
 
-- `main-base-py3.11-cu128-2.8.0`
-- `main-base-py3.11-cu128-2.9.1`
+- `main-base-py3.11-cu128-2.7.1`
+- `main-base-py3.11-cu126-2.7.1`
+- `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
 
@@ -71,12 +74,15 @@ There may be some extra tags appended to the image, like `-vllm` which installs
 
 Tags examples:
 
-- `main-py3.11-cu128-2.8.0`
-- `main-py3.11-cu128-2.9.1`
+- `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-latest`
 - `main-20250303-py3.11-cu124-2.6.0`
 - `main-20250303-py3.11-cu126-2.6.0`
-- `0.12.0`
+- `0.10.1`
 
 ## Cloud
 

docs/expert_quantization.qmd

@@ -1,67 +0,0 @@
----
-title: "MoE Expert Quantization"
-description: "Reduce VRAM usage when training MoE model adapters by quantizing expert weights on load"
----
-
-Transformers v5 changed MoE expert layers from `nn.Linear` to fused `nn.Parameter` (3D+ tensors).
-This means `bitsandbytes` can no longer quantize them during model loading, resulting in all expert
-weights being loaded in full bf16 precision and causing massive VRAM usage.
-
-`quantize_moe_experts` solves this by quantizing expert weights during model loading.
-It intercepts the weight loading process, quantizes each expert tensor on the fly, and
-immediately frees the original bf16 tensor from VRAM. This dramatically reduces peak memory.
-For example, GLM-4.7-Flash QLoRA drops from ~127GiB to ~23GiB reserved memory.
-
-## Usage
-
-Enable expert quantization in your Axolotl config:
-
-```yaml
-quantize_moe_experts: true
-```
-
-This works with both 4-bit (QLoRA) and 8-bit (LoRA) quantization.
-
-### Expert LoRA targeting
-
-You can optionally apply LoRA adapters directly to expert weights using `lora_target_parameters`:
-
-```yaml
-lora_target_parameters:
-  - mlp.experts.gate_up_proj
-  - mlp.experts.down_proj
-  # - mlp.gate.weight  # router
-```
-
-::: {.callout-note}
-`lora_dropout` must be `0` when using `lora_target_parameters`.
-:::
-
-## Requirements
-
-- Requires (`adapter: lora` and `load_in_8bit: true`) or (`adapter: qlora` and `load_in_4bit: true`)
-- CUDA GPUs only (not tested with ROCm or other backends)
-- FSDP2 compatible for distributed training
-
-## Limitations
-
-- `lora_target_linear` is not compatible with `quantize_moe_experts`. See [Expert LoRA targeting](#expert-lora-targeting) instead.
-- `cpu_ram_efficient_loading` hangs / takes long time with FSDP2 + QLoRA.
-- Total model parameter count may display incorrectly (trainable param count is correct).
-- FSDP LoRA (8-bit) may have a large initial VRAM spike at the first 1-2 steps, which then drops. QLoRA does not exhibit this.
-- FSDP2 may use more VRAM per GPU than single GPU training due to not all layers being properly sharded across ranks.
-- Model loading takes longer due to on-demand quantization, even on consecutive runs.
-- DeepSpeed has not been tested.
-
-## Implementation details
-
-The quantization is applied by patching transformers to intercept weight loading.
-When a 3D+ CUDA tensor with "expert" in its name is detected:
-
-- **4-bit mode:** Uses bitsandbytes NF4 parametrization (configurable via `bnb_4bit_quant_type`).
-- **8-bit mode:** Uses a custom row-wise int8 parametrization with bitsandbytes dequantization.
-
-The original bf16 tensor is freed immediately after quantization. Multiple sub-patches are applied to
-transformers, PEFT and accelerate FSDP2 to support these parametrized expert modules.
-
-For full implementation details, see [PR #3439](https://github.com/axolotl-ai-cloud/axolotl/pull/3439).

docs/faq.qmd

@@ -63,14 +63,6 @@ description: Frequently asked questions
 
 > 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' / ____`**

docs/gradient_checkpointing.qmd

@@ -1,5 +1,5 @@
 ---
-title: Gradient Checkpointing, Activation Offloading, and Layer Offloading
+title: Gradient Checkpointing and Activation Offloading
 ---
 
 Gradient checkpointing and activation offloading are techniques used to optimize the performance of deep learning
@@ -27,33 +27,3 @@ The `activation_offloading: legacy` naively offloads activations to CPU and with
 
 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.
-
-### Enabling Layer Offloading
-
-```yaml
-layer_offloading: true
-```
-
-Layer offloading reduces GPU memory usage by moving frozen (non-trainable) decoder layer parameters to CPU
-and streaming them back to GPU one layer at a time during the forward and backward passes. This is
-particularly useful for LoRA/QLoRA training where most of the model's parameters are frozen — only the
-trainable adapter weights stay on GPU permanently.
-
-During training, forward and backward hooks on each decoder layer handle the transfer automatically:
-
-- **Forward pass:** Before a layer executes, its frozen params are loaded to GPU. The next layer is
-  prefetched asynchronously on a separate CUDA stream for overlap.
-- **Backward pass:** Same pattern in reverse — the current layer's frozen params are loaded and the
-  previous layer is prefetched.
-
-After each layer finishes, its frozen params are offloaded back to CPU pinned memory.
-
-This approach trades some CPU-GPU transfer overhead for significant GPU memory savings — the freed memory
-is roughly equal to the size of all frozen parameters across all decoder layers, minus one layer's worth
-that is kept on GPU at any given time.
-
-**Requirements:**
-
-- CUDA GPU (CPU-only training is not supported for this feature)
-- Works with any HuggingFace model architecture that uses decoder layers (Llama, Mistral, Qwen, etc.)
-- Best combined with LoRA/QLoRA where most parameters are frozen

docs/installation.qmd

@@ -26,22 +26,43 @@ Follow the instructions at: [https://pytorch.org/get-started/locally/](https://p
 :::
 
 ::: {.callout-important}
-For Blackwell GPUs, please use Pytorch 2.9.1 and CUDA 12.8.
+For Blackwell GPUs, please use Pytorch 2.7.0 and CUDA 12.8.
 :::
 
-### PyPI Installation (Recommended) {#sec-pypi}
+### uv Installation (Recommended) {#sec-uv-quick}
 
 ```{.bash}
-pip3 install -U packaging setuptools wheel ninja
-pip3 install --no-build-isolation axolotl[flash-attn,deepspeed]
+# Install uv if not already installed
+curl -LsSf https://astral.sh/uv/install.sh | sh
+
+# Add Axolotl to a project (recommended)
+uv init my-project && cd my-project
+uv add axolotl
+uv pip install flash-attn --no-build-isolation
+source .venv/bin/activate
+```
+
+For a quick one-off install without creating a project:
+
+```{.bash}
+uv pip install axolotl
+uv pip install flash-attn --no-build-isolation
+```
+
+### pip Installation {#sec-pypi}
+
+```{.bash}
+pip install --no-build-isolation axolotl[deepspeed]
+pip install --no-build-isolation flash-attn
 ```
 
 We use `--no-build-isolation` in order to detect the installed PyTorch version (if
 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.
+co-dependencies. Flash Attention is resolved separately so it can be built against
+the environment configured by the previous step.
 
-### uv Installation {#sec-uv}
+### Advanced 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.
 
@@ -62,28 +83,38 @@ 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]
-
+uv pip install --no-build-isolation axolotl[deepspeed]
 # 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]
+# uv pip install --no-build-isolation axolotl[deepspeed,vllm]
+
+uv pip install flash-attn --no-build-isolation
 ```
 
 ### Edge/Development Build {#sec-edge-build}
 
 For the latest features between releases:
 
+#### Using uv (recommended)
 ```{.bash}
 git clone https://github.com/axolotl-ai-cloud/axolotl.git
 cd axolotl
-pip3 install -U packaging setuptools wheel ninja
-pip3 install --no-build-isolation -e '.[flash-attn,deepspeed]'
+curl -LsSf https://astral.sh/uv/install.sh | sh  # If not already installed
+uv sync
+uv pip install flash-attn --no-build-isolation
+```
+
+#### Using pip
+```{.bash}
+git clone https://github.com/axolotl-ai-cloud/axolotl.git
+cd axolotl
+pip install --no-build-isolation -e '.[deepspeed]'
+pip install --no-build-isolation flash-attn
 ```
 
 ### Docker {#sec-docker}
@@ -111,7 +142,7 @@ docker run --privileged --gpus '"all"' --shm-size 10g --rm -it \
 :::
 
 ::: {.callout-important}
-For Blackwell GPUs, please use `axolotlai/axolotl:main-py3.11-cu128-2.9.1` or the cloud variant `axolotlai/axolotl-cloud:main-py3.11-cu128-2.9.1`.
+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.
@@ -141,7 +172,7 @@ For providers supporting Docker:
 ### macOS {#sec-macos}
 
 ```{.bash}
-pip3 install --no-build-isolation -e '.'
+uv pip install --no-build-isolation -e '.'
 ```
 
 See @sec-troubleshooting for Mac-specific issues.
@@ -159,13 +190,18 @@ We recommend using WSL2 (Windows Subsystem for Linux) or Docker.
 1. Install Python ≥3.11
 2. Install PyTorch: https://pytorch.org/get-started/locally/
 3. Install Axolotl:
-   ```{.bash}
-   pip3 install -U packaging setuptools wheel ninja
-   pip3 install --no-build-isolation -e '.[flash-attn,deepspeed]'
-   ```
+```{.bash}
+# Option A: add Axolotl to the environment
+uv add axolotl
+uv pip install flash-attn --no-build-isolation
+
+# Option B: quick install
+uv pip install axolotl
+uv pip install flash-attn --no-build-isolation
+```
 4. (Optional) Login to Hugging Face:
    ```{.bash}
-   hf auth login
+   huggingface-cli login
    ```
 
 ## Troubleshooting {#sec-troubleshooting}

docs/lora_optims.qmd

@@ -89,10 +89,6 @@ lora_o_kernel: true
 Currently, LoRA kernels are not supported for RLHF training, only SFT.
 :::
 
-::: {.callout-warning}
-LoRA kernels do not support remote modeling code.
-:::
-
 ## Requirements
 
 - One or more NVIDIA or AMD GPUs (in order to use the Triton kernels)

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
-
-:::

docs/multi-gpu.qmd

@@ -4,7 +4,7 @@ format:
   html:
     toc: true
     toc-depth: 3
-    # number-sections: true
+    number-sections: true
     code-tools: true
 execute:
   enabled: false
@@ -14,18 +14,12 @@ This guide covers advanced training configurations for multi-GPU setups using Ax
 
 ## Overview {#sec-overview}
 
-When training on multiple GPUs, Axolotl supports 3 sharding/parallelism strategies. Additionally, you can layer specific optimization features on top of that strategy.
+Axolotl supports several methods for multi-GPU training:
 
-You generally cannot combine these strategies; they are mutually exclusive.
-
-1.  **DeepSpeed**: Powerful optimization library, supports ZeRO stages 1-3.
-2.  **FSDP (Fully Sharded Data Parallel)**: PyTorch's native sharding implementation (Recommended).
-3.  **DDP (Distributed Data Parallel)**: PyTorch's native parallelism implementation (Default if neither of the above are selected).
-
-These features can often be combined with the strategies above:
-
-*   **Sequence Parallelism**: Splits long sequences across GPUs (Compatible with DDP, DeepSpeed, and FSDP).
-*   **FSDP + QLoRA**: Combines 4-bit quantization with FSDP (Specific to FSDP).
+- DeepSpeed (recommended)
+- FSDP (Fully Sharded Data Parallel)
+- Sequence parallelism
+- FSDP + QLoRA
 
 ## DeepSpeed {#sec-deepspeed}
 
@@ -71,18 +65,12 @@ Start from Stage 1 -> Stage 2 -> Stage 3.
 
 ## Fully Sharded Data Parallel (FSDP) {#sec-fsdp}
 
-FSDP allows you to shard model parameters, gradients, and optimizer states across data parallel workers.
-
 ::: {.callout-note}
 
 FSDP2 is recommended for new users. FSDP1 is deprecated and will be removed in an upcoming release of Axolotl.
 
 :::
 
-### FSDP + QLoRA {#sec-fsdp-qlora}
-
-For combining FSDP with QLoRA, see our [dedicated guide](fsdp_qlora.qmd).
-
 ### 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
@@ -100,7 +88,6 @@ 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:
@@ -157,6 +144,10 @@ single sequence causes OOM errors during model training.
 
 See our [dedicated guide](sequence_parallelism.qmd) for more information.
 
+### FSDP + QLoRA {#sec-fsdp-qlora}
+
+For combining FSDP with QLoRA, see our [dedicated guide](fsdp_qlora.qmd).
+
 ## Performance Optimization {#sec-performance}
 
 ### Liger Kernel Integration {#sec-liger}

docs/multimodal.qmd

@@ -13,18 +13,14 @@ format:
 - [Pixtral](#sec-pixtral)
 - [Llava-1.5](#sec-llava-15)
 - [Mistral-Small-3.1](#sec-mistral-small-31)
-- [Mistral-Small-4](#sec-mistral-small-4)
 - [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)
-- [Qwen3.5](#sec-qwen3-5)
-- [GLM-4.6V](#sec-glm-4-6v)
 - [SmolVLM2](#sec-smolvlm2)
 - [LFM2-VL](#sec-lfm2-vl)
-- [Intern-VL](#sec-intern-vl)
 
 ## Usage
 
@@ -60,14 +56,10 @@ 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
@@ -103,23 +95,17 @@ 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'`
+Please make sure to install vision lib via `uv pip install 'mistral-common[opencv]==1.8.5'`
 :::
 
 ```yaml
 base_model: mistralai/Mistral-Small-3.1-24B-Instruct-2503
 ```
 
-### Mistral-Small-4 {#sec-mistral-small-4}
-
-```yaml
-base_model: mistralai/Mistral-Small-4-119B-2603
-```
-
 ### 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'`
+Please make sure to install vision lib via `uv pip install 'mistral-common[opencv]==1.8.5'`
 :::
 
 ```yaml
@@ -129,13 +115,11 @@ 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'`
+Please make sure to install audio lib via `uv pip install librosa==0.11.0 'mistral_common[audio]==1.8.3'`
 :::
 
 ```yaml
 base_model: mistralai/Voxtral-Mini-3B-2507
-
-processor_type: VoxtralProcessor
 ```
 
 ### Gemma-3 {#sec-gemma-3}
@@ -159,7 +143,7 @@ The model's initial loss and grad norm will be very high. We suspect this to be
 :::
 
 ::: {.callout-tip}
-Please make sure to install `timm` via `pip3 install timm==1.0.17`
+Please make sure to install `timm` via `uv pip install timm==1.0.17`
 :::
 
 ```yaml
@@ -184,38 +168,10 @@ 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
-```
-
-### Qwen3.5 {#sec-qwen3-5}
-
-```yaml
-base_model: Qwen/Qwen3.5-9B
-
-chat_template: qwen3_5
-```
-
-### GLM-4.6V {#sec-glm-4-6v}
-
-Both GLM-4.6V (106B MoE) and GLM-4.6V-Flash (9B) are supported.
-
-```yaml
-# GLM-4.6V (106B MoE version)
-base_model: zai-org/GLM-4.6V
-
-# OR GLM-4.6V-Flash (9B version)
-base_model: zai-org/GLM-4.6V-Flash
-```
-
 ### SmolVLM2 {#sec-smolvlm2}
 
 ::: {.callout-tip}
-Please make sure to install `num2words` via `pip3 install num2words==0.5.14`
+Please make sure to install `num2words` via `uv pip install num2words==0.5.14`
 :::
 
 ```yaml
@@ -225,23 +181,13 @@ base_model: HuggingFaceTB/SmolVLM2-500M-Video-Instruct
 ### LFM2-VL {#sec-lfm2-vl}
 
 ::: {.callout-warning}
-Please uninstall `causal-conv1d` via `pip3 uninstall -y causal-conv1d`
+Please uninstall `causal-conv1d` via `uv pip uninstall -y causal-conv1d`
 :::
 
 ```yaml
 base_model: LiquidAI/LFM2-VL-450M
 ```
 
-### Intern-VL {#sec-intern-vl}
-
-::: {.callout-tip}
-Please make sure to install `timm` via `pip3 install timm==1.0.19`
-:::
-
-```yaml
-base_model: OpenGVLab/InternVL3_5-8B
-```
-
 ## Dataset Format
 
 For multi-modal datasets, we adopt an extended `chat_template` format similar to OpenAI's Message format.
@@ -276,7 +222,7 @@ For audio loading, you can use the following keys within `content` alongside `"t
 
 ::: {.callout-tip}
 
-You may need to install `librosa` via `pip3 install librosa==0.11.0`.
+You may need to install `librosa` via `uv pip install librosa==0.11.0`.
 
 :::
 

docs/optimizations.qmd

@@ -54,13 +54,6 @@ These techniques save VRAM by changing how activations are handled.
 - 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)
 
-### Layer Offloading
-
-Offloads frozen (non-trainable) decoder layer parameters to CPU and streams them back to GPU one layer at a time during forward/backward passes using CUDA stream prefetching. Especially effective for LoRA/QLoRA where most parameters are frozen.
-
-- **Config:** `layer_offloading: true`
-- **Learn more:** [Layer Offloading Docs](gradient_checkpointing.qmd#enabling-layer-offloading)
-
 ### Cut Cross Entropy (CCE)
 
 Reduces VRAM usage by using an optimized cross-entropy loss calculation.
@@ -73,15 +66,6 @@ Provides efficient Triton kernels to improve training speed and reduce memory us
 
 - **Learn more:** [Custom Integrations - Liger Kernels](custom_integrations.qmd#liger-kernels)
 
-### Expert Kernels
-
-Optimized kernel implementations for Mixture of Experts (MoE) model training.
-
-- **ScatterMoE**: Triton-based MoE kernels with fused LoRA support.
-- **SonicMoE**: CUTLASS-based MoE kernels for NVIDIA Hopper and Blackwell GPUs.
-
-- **Learn more:** [Custom Integrations - Kernels Integration](custom_integrations.qmd#kernels-integration)
-
 ## Long Context Models
 
 Techniques to train models on sequences longer than their original context window.
@@ -147,10 +131,3 @@ Simulates quantization effects during training, helping the model adapt and pote
 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)
-
-### MoE Expert Quantization
-
-Quantizes MoE expert weights on load to reduce VRAM when training MoE models with adapters. Required for Transformers v5+ MoE models where experts use fused `nn.Parameter` tensors.
-
-- **Config:** `quantize_moe_experts: true`
-- **Learn more:** [MoE Expert Quantization](expert_quantization.qmd)

docs/rlhf.qmd

@@ -17,7 +17,6 @@ feedback. Various methods include, but not limited to:
 - [Kahneman-Tversky Optimization (KTO)](#kto)
 - [Odds Ratio Preference Optimization (ORPO)](#orpo)
 - [Group Relative Policy Optimization (GRPO)](#grpo)
-- [Group Reward-Decoupled Policy Optimization (GDPO)](#gdpo)
 
 
 ## RLHF using Axolotl
@@ -220,21 +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
@@ -598,116 +582,6 @@ To see other examples of custom reward functions, please see [TRL GRPO Docs](htt
 
 To see all configs, please see [TRLConfig](https://github.com/axolotl-ai-cloud/axolotl/blob/v0.9.2/src/axolotl/utils/schemas/trl.py).
 
-#### OpenEnv Rollout Functions
-
-GRPO supports custom rollout functions for OpenEnv-style environments, enabling interactive tasks like web browsing, code execution, or tool use. This allows you to implement custom generation logic that interacts with external environments.
-
-For example, to implement a simple math-solving environment with step-by-step verification:
-
-```python
-# math_env.py
-import re
-
-def math_solver_rollout(model, processing_class, prompts, generation_config=None):
-    """
-    Custom rollout function that generates step-by-step math solutions.
-
-    Args:
-        model: The language model
-        processing_class: The tokenizer/processing_class
-        prompts: List of prompt dicts (with 'messages' key for chat format)
-        generation_config: Optional generation configuration
-
-    Returns:
-        List of completion strings
-    """
-    completions = []
-
-    for prompt in prompts:
-        # Apply chat template to prompt
-        messages = prompt.get("messages", [])
-        formatted_prompt = processing_class.apply_chat_template(
-            messages, processing_class=False, add_generation_prompt=True
-        )
-
-        # Generate step-by-step solution
-        full_response = ""
-        for step in range(5):  # Max 5 reasoning steps
-            current_input = formatted_prompt + full_response + "\nNext step:"
-            inputs = processing_class(current_input, return_tensors="pt").to(model.device)
-
-            outputs = model.generate(
-                **inputs,
-                max_new_tokens=100,
-                generation_config=generation_config,
-            )
-            step_text = processing_class.decode(
-                outputs[0][inputs.input_ids.shape[1]:],
-                skip_special_tokens=True
-            )
-
-            # Check if solution is complete
-            if "FINAL ANSWER:" in step_text:
-                full_response += step_text
-                break
-            full_response += step_text + "\n"
-
-        completions.append(full_response)
-
-    return completions
-
-def math_reward(prompts, completions, answers, **kwargs):
-    """Reward function that checks mathematical correctness"""
-    rewards = []
-    for completion, correct_answer in zip(completions, answers):
-        # Extract predicted answer
-        match = re.search(r"FINAL ANSWER:\s*(.+)", completion)
-        predicted = match.group(1).strip() if match else ""
-
-        # Compare with correct answer
-        reward = 1.0 if predicted == str(correct_answer) else 0.0
-        rewards.append(reward)
-
-    return rewards
-
-def math_transform(cfg, *args, **kwargs):
-    """Transform dataset to GRPO format with answer field"""
-    def transform_fn(example, processing_class=None):
-        return {
-            "prompt": [{"role": "user", "content": example["question"]}],
-            "answer": str(example["answer"]),
-        }
-    return transform_fn, {"remove_columns": ["question"]}
-```
-
-```yaml
-rl: grpo
-
-trl:
-  beta: 0.001
-  max_completion_length: 512
-  num_generations: 4
-  rollout_func: "math_env.math_solver_rollout"  # Custom rollout function
-  reward_funcs: ["math_env.math_reward"]
-  reward_weights: [1.0]
-
-datasets:
-  - path: openai/gsm8k
-    name: main
-    type: math_env.math_transform
-```
-
-The `rollout_func` parameter accepts a fully qualified name (e.g., `module_name.function_name`) that points to a callable function in your local directory. The function receives:
-
-- `model`: The language model
-- `processing_class`: The tokenizer/processing class
-- `prompts`: List of prompt dictionaries
-- `generation_config` (optional): Generation configuration
-
-And should return a list of completion strings.
-
-For more OpenEnv examples, see [TRL OpenEnv Documentation](https://huggingface.co/docs/trl/main/en/openenv).
-
 #### 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.
@@ -721,309 +595,6 @@ trl:
 
 For more information, see [GRPO docs](https://huggingface.co/docs/trl/v0.17.0/en/grpo_trainer#loss-types).
 
-#### Async GRPO
-
-Async GRPO overlaps vLLM generation with training by producing rollouts in a background thread. While the model trains on the current batch, the next batch is already being generated. This can significantly reduce wall-clock time per step.
-
-```yaml
-trl:
-  use_data_producer: true     # Enable data producer protocol
-  use_vllm: true
-  async_prefetch: true         # Generate rollouts in background thread
-  prefetch_depth: 1            # Number of rollouts to prefetch
-  vllm_sync_interval: 2        # Sync weights to vLLM every N steps
-```
-
-::: {.callout-note}
-Because the background thread generates completions with slightly stale model weights, async GRPO uses importance sampling correction to account for the distribution shift. This is controlled by `vllm_importance_sampling_correction: true` (default when async is enabled).
-:::
-
-##### vLLM LoRA Sync
-
-By default, weight sync to vLLM merges the LoRA adapter into the base model and broadcasts all parameters via NCCL. LoRA sync is a faster alternative that saves only the adapter weights to the filesystem and has vLLM load them natively using Punica kernels.
-
-```yaml
-adapter: lora
-lora_r: 32
-lora_alpha: 64
-lora_target_linear: true
-
-trl:
-  vllm_lora_sync: true         # Enable native LoRA sync
-```
-
-When `vllm_lora_sync: true` is set, axolotl automatically selects the LoRA-aware vLLM serve module. Start vLLM as usual:
-
-```bash
-CUDA_VISIBLE_DEVICES=0 axolotl vllm-serve config.yaml
-```
-
-Then start training on a separate GPU:
-
-```bash
-CUDA_VISIBLE_DEVICES=1 axolotl train config.yaml
-```
-
-::: {.callout-tip}
-LoRA sync is especially beneficial with multi-GPU training (FSDP/DeepSpeed), where NCCL merge-sync can cause GPU contention with vLLM generation.
-:::
-
-##### Streaming Partial Batch
-
-Instead of scoring the entire batch at once, streaming mode scores one prompt group at a time. This enables finer-grained zero-advantage skipping and reduces peak memory usage during scoring.
-
-```yaml
-trl:
-  streaming_partial_batch: true
-```
-
-##### Importance Sampling Correction
-
-When using async prefetch, completions are generated from a slightly older version of the model. Importance sampling (IS) correction adjusts the policy gradient to account for this distribution shift.
-
-```yaml
-trl:
-  vllm_importance_sampling_correction: true   # Enable IS correction
-  importance_sampling_level: token             # 'token' or 'sequence'
-  off_policy_mask_threshold: 0.5              # Mask sequences with IS ratio below this
-```
-
-- `importance_sampling_level: token` applies per-token IS ratios (recommended with Liger kernel)
-- `importance_sampling_level: sequence` applies per-sequence IS ratios
-- `off_policy_mask_threshold` masks out sequences where the IS ratio indicates they are too far off-policy
-
-##### Replay Buffer
-
-The replay buffer caches rollout groups that had learning signal (non-zero reward variance) and uses them to replace zero-signal groups in later batches.
-
-```yaml
-trl:
-  replay_buffer_size: 100       # Max cached groups (0 = disabled)
-  replay_recompute_logps: true  # Recompute log-probs for replayed data (recommended)
-```
-
-::: {.callout-note}
-When `replay_recompute_logps: true` (default), old log-probabilities are recomputed using the current model weights. This fixes the IS mismatch that would otherwise occur when replaying stale data.
-:::
-
-##### Deferred Re-rolling
-
-Failed prompts (where the model produces zero reward for all generations) are buffered and re-injected into later batches when the model may be better equipped to solve them.
-
-```yaml
-trl:
-  reroll_start_fraction: 0.5    # Start re-rolling after 50% of training
-  reroll_max_groups: 1          # Max groups to replace per batch
-```
-
-##### Zero-Advantage Batch Skipping
-
-When all advantages in a micro-batch are zero (no learning signal), the forward/backward pass is skipped entirely. This is enabled by default and logged as `skipped_zero_adv_batches=1`.
-
-```yaml
-trl:
-  skip_zero_advantage_batches: true   # default
-```
-
-##### Parallel Reward Workers
-
-Reward functions that use `signal.alarm()` (e.g., `math_verify`) must run in the main thread. Parallel reward workers use subprocesses to work around this limitation while enabling concurrent reward computation.
-
-```yaml
-trl:
-  reward_num_workers: 4         # Number of subprocess workers (1 = no parallelism)
-```
-
-##### Full Async GRPO Example
-
-```yaml
-base_model: Qwen/Qwen2.5-1.5B-Instruct
-
-vllm:
-    host: 0.0.0.0
-    port: 8000
-    gpu_memory_utilization: 0.35
-    dtype: auto
-
-adapter: lora
-lora_r: 32
-lora_alpha: 64
-lora_target_linear: true
-
-rl: grpo
-trl:
-  use_data_producer: true
-  use_vllm: true
-  async_prefetch: true
-  prefetch_depth: 1
-  vllm_sync_interval: 2
-  vllm_lora_sync: true
-  streaming_partial_batch: true
-  vllm_importance_sampling_correction: true
-  off_policy_mask_threshold: 0.5
-  importance_sampling_level: token
-  num_generations: 8
-  max_completion_length: 512
-  reward_funcs:
-    - rewards.accuracy_reward
-  reroll_start_fraction: 0.5
-  replay_buffer_size: 100
-  reward_num_workers: 4
-  skip_zero_advantage_batches: true
-
-datasets:
-  - path: AI-MO/NuminaMath-TIR
-    type: rewards.prompt_transform
-    split: train
-
-gradient_accumulation_steps: 4
-micro_batch_size: 2
-max_steps: 500
-learning_rate: 1e-5
-bf16: true
-gradient_checkpointing: true
-```
-
-```bash
-# Terminal 1: Start vLLM on GPU 0
-CUDA_VISIBLE_DEVICES=0 axolotl vllm-serve config.yaml
-
-# Terminal 2: Train on GPU 1
-CUDA_VISIBLE_DEVICES=1 axolotl train config.yaml
-```
-
-##### Multi-GPU Async GRPO
-
-Async GRPO supports FSDP and DeepSpeed ZeRO-3 for multi-GPU training. vLLM runs on one GPU while training is distributed across the remaining GPUs.
-
-**FSDP:**
-
-```yaml
-fsdp:
-  - full_shard
-  - auto_wrap
-fsdp_config:
-  fsdp_transformer_layer_cls_to_wrap: Qwen2DecoderLayer
-gradient_checkpointing_kwargs:
-  use_reentrant: false
-```
-
-**DeepSpeed ZeRO-3:**
-
-```yaml
-deepspeed: deepspeed_configs/zero3_bf16.json
-gradient_checkpointing_kwargs:
-  use_reentrant: true   # Required for ZeRO-3
-```
-
-```bash
-# Terminal 1: Start vLLM on GPU 0
-CUDA_VISIBLE_DEVICES=0 axolotl vllm-serve config.yaml
-
-# Terminal 2: Train on GPUs 0,1
-CUDA_VISIBLE_DEVICES=0,1 accelerate launch --num_processes 2 -m axolotl.cli.train config.yaml
-```
-
-::: {.callout-important}
-With multi-GPU async prefetch, only rank 0 generates completions in the background thread. Results are broadcast to all ranks on the main thread. This avoids FSDP/DeepSpeed collective deadlocks from unsynchronized background threads.
-:::
-
-### GDPO
-
-GDPO (Group Reward-Decoupled Policy Optimization) extends GRPO for multi-reward training. It addresses the **reward advantage collapse** problem by normalizing each reward function independently before combining them.
-
-::: {.callout-tip}
-Use GDPO when training with multiple reward functions. For single reward, GRPO and GDPO produce equivalent results.
-:::
-
-Paper: [https://arxiv.org/pdf/2501.05242](https://arxiv.org/pdf/2501.05242)
-
-GDPO uses TRL's native `multi_objective_aggregation` parameter under the hood. When you set `rl: gdpo`, axolotl automatically configures TRL to use `normalize_then_sum` aggregation.
-
-```yaml
-base_model: Qwen/Qwen2.5-1.5B-Instruct
-
-vllm:
-    host: 0.0.0.0
-    port: 8000
-    tensor_parallel_size: 2
-    gpu_memory_utilization: 0.85
-
-rl: gdpo
-
-trl:
-    beta: 0.001
-    max_completion_length: 256
-    use_vllm: true
-    num_generations: 4
-    reward_funcs:
-        - rewards.format_reward
-        - rewards.correctness_reward
-    reward_weights: [1.0, 2.0]
-
-datasets:
-    - path: openai/gsm8k
-      name: main
-      type: rewards.oai_gsm8k_transform
-```
-
-You can also use GRPO with explicit aggregation control:
-
-```yaml
-rl: grpo
-trl:
-    multi_objective_aggregation: normalize_then_sum  # GDPO behavior
-    # or: sum_then_normalize  # Default GRPO behavior
-```
-
-#### GDPO vs GRPO
-
-| Aspect | GRPO | GDPO |
-|--------|------|------|
-| **Aggregation** | `sum_then_normalize` | `normalize_then_sum` |
-| **Multi-reward** | May collapse advantages | Preserves reward signals |
-| **Single reward** | Standard behavior | Equivalent to GRPO |
-
-#### Why GDPO?
-
-When using multiple rewards with GRPO, different reward combinations can produce identical advantages:
-
-```
-# Example: format + correctness rewards
-[format=0, correct=3] → sum=3
-[format=1, correct=2] → sum=3  ← GRPO sees these as equal!
-[format=2, correct=1] → sum=3
-[format=3, correct=0] → sum=3
-```
-
-GDPO normalizes each reward independently, preserving their relative differences.
-
-#### Reward Functions
-
-GDPO uses the same reward function format as GRPO:
-
-```python
-# rewards.py
-def format_reward(completions, **kwargs) -> list[float]:
-    return [1.0 if len(c) > 10 else 0.0 for c in completions]
-
-def correctness_reward(completions, answers, **kwargs) -> list[float]:
-    rewards = []
-    for completion, answer in zip(completions, answers):
-        # Your scoring logic here
-        rewards.append(score)
-    return rewards
-```
-
-#### Sequence Parallelism
-
-GDPO supports sequence parallelism for long-context training:
-
-```yaml
-rl: gdpo
-context_parallel_size: 2
-```
-
 ### SimPO
 
 SimPO uses [CPOTrainer](https://huggingface.co/docs/trl/main/en/cpo_trainer) but with alternative loss function.

docs/scripts/examples-allowlist.yml

@@ -1,90 +0,0 @@
-examples:
-  # December 2025
-  - name: kimi-linear
-    title: Kimi Linear
-  - name: plano
-    title: Plano Orchestrator
-  - name: mimo
-    title: MiMo
-  - name: internvl3_5
-    title: InternVL 3.5
-
-  # AllenAI
-  - name: olmo3
-    title: OLMo 3
-
-  # ArceeAI
-  - name: trinity
-    title: Trinity
-  - name: arcee
-    title: Arcee AFM
-
-  # MistralAI
-  - name: ministral3/think
-    title: Ministral 3 Thinking
-  - name: ministral3/vision
-    title: Ministral 3 Vision
-  - name: magistral/think
-    title: Magistral Thinking
-  - name: magistral/vision
-    title: Magistral Vision
-  - name: ministral
-    title: Ministral
-  - name: mistral-small
-    title: Mistral Small 3.1/3.2
-  - name: voxtral
-    title: Voxtral
-  - name: devstral
-    title: Devstral
-  - name: mistral
-    title: Mistral 7B
-
-  # Meta
-  - name: llama-4
-    title: Llama 4
-  - name: llama-2
-    title: Llama 2
-
-  # Alibaba
-  - name: qwen3-next
-    title: Qwen 3 Next
-  - name: qwen3
-    title: Qwen 3
-
-  # Google
-  - name: gemma3n
-    title: Gemma 3n
-
-  # Swiss AI
-  - name: apertus
-    title: Apertus
-
-  # GPT-OSS
-  - name: gpt-oss
-    title: GPT-OSS
-  - name: seed-oss
-    title: Seed-OSS
-
-  # Microsoft
-  - name: phi
-    title: Phi
-
-  # SmolVLM
-  - name: smolvlm2
-    title: SmolVLM 2
-
-  # IBM
-  - name: granite4
-    title: Granite 4
-
-  # LiquidAI
-  - name: LiquidAI
-    title: Liquid Foundation Models 2
-
-  # Other
-  - name: hunyuan
-    title: Hunyuan
-  - name: jamba
-    title: Jamba
-  - name: orpheus
-    title: Orpheus

docs/scripts/generate_examples_docs.py

@@ -1,424 +0,0 @@
-"""
-auto generate example docs from allowlist
-"""
-
-import re
-import shutil
-import sys
-from pathlib import Path
-
-import yaml
-
-# Paths
-THIS = Path(__file__).resolve()
-ROOT = THIS.parents[2]  # repo root (docs/scripts -> docs -> ROOT)
-EXAMPLES_DIR = ROOT / "examples"
-OUTPUT_DIR = ROOT / "docs" / "models"
-ALLOWLIST_YML = THIS.parent / "examples-allowlist.yml"
-
-
-def slugify(name: str) -> str:
-    """Convert a name to a slug (lowercase, hyphens for spaces)."""
-    s = re.sub(r"[^a-zA-Z0-9\s\-]+", "", name.strip())
-    s = re.sub(r"\s+", "-", s).strip("-").lower()
-    return s or "example"
-
-
-def read_allowlist():
-    with open(ALLOWLIST_YML, "r", encoding="utf-8") as f:
-        data = yaml.safe_load(f) or {}
-    items = data.get("examples", [])
-    if not isinstance(items, list):
-        raise ValueError("`examples` must be a list in examples-allowlist.yml")
-    return items
-
-
-def find_readme(folder: Path) -> Path | None:
-    for name in ("README.md", "Readme.md", "readme.md"):
-        p = folder / name
-        if p.exists():
-            return p
-    return None
-
-
-def remove_first_h1(md: str) -> tuple[str, str | None]:
-    """
-    Remove the first H1 from markdown and return (modified_md, h1_title).
-    The H1 is removed since we use the frontmatter title instead.
-    """
-    lines = md.splitlines()
-    result = []
-    h1_title = None
-    skipped_first = False
-
-    for line in lines:
-        if not skipped_first and line.startswith("# "):
-            h1_title = line[2:].strip()
-            skipped_first = True
-            continue
-        result.append(line)
-
-    return "\n".join(result), h1_title
-
-
-IMG_RE = re.compile(r"!\[[^\]]*\]\(([^)]+)\)")
-LINK_RE = re.compile(r"\[([^\]]+)\]\(([^)]+)\)")
-
-
-def rewrite_and_copy_assets(md: str, src_dir: Path, dest_assets_root: Path) -> str:
-    """
-    Copy local image assets referenced in markdown to
-    docs/examples/assets/... and rewrite the links.
-    """
-    dest_assets = dest_assets_root / "assets"
-
-    def repl(m):
-        url = m.group(1).strip()
-        if re.match(r"^(https?:)?//", url):
-            return m.group(0)  # leave remote URLs
-        src_path = (src_dir / url).resolve()
-        if not src_path.exists():
-            return m.group(0)  # leave as-is if not found
-        rel = src_path.relative_to(src_dir)
-        # Create a unique asset path based on source directory name
-        asset_name = src_dir.name.replace("/", "-")
-        dest_path = dest_assets / asset_name / rel
-        dest_path.parent.mkdir(parents=True, exist_ok=True)
-        shutil.copy2(src_path, dest_path)
-        new_rel = f"assets/{asset_name}/{rel.as_posix()}"
-        return m.group(0).replace(url, new_rel)
-
-    return IMG_RE.sub(repl, md)
-
-
-def rewrite_readme_links(
-    md: str,
-    src_dir: Path,
-    examples_dir: Path,
-    parent_index_only: set,
-    current_src_path: str,
-    allowlist_entries: set,
-    current_output_path: str,
-) -> str:
-    """
-    Rewrite links between README.md files to point to the correct .qmd files.
-    """
-
-    def repl(m):
-        text = m.group(1)
-        url = m.group(2).strip()
-
-        # Skip remote URLs and anchor links
-        if re.match(r"^(https?:)?//", url) or url.startswith("#"):
-            return m.group(0)
-
-        # Skip non-markdown files
-        if not url.lower().endswith(".md"):
-            return m.group(0)
-
-        # Resolve the target path
-        try:
-            target_path = (src_dir / url).resolve()
-
-            # Check if target is outside examples_dir
-            try:
-                rel_path = target_path.relative_to(examples_dir)
-            except ValueError:
-                # Target is outside examples_dir, leave as-is
-                return m.group(0)
-
-            parts = list(rel_path.parts)
-
-            # Determine the output path for the target
-            if len(parts) > 0 and parts[-1].lower() in ("readme.md", "readme"):
-                # This is a README link
-                if len(parts) == 1:
-                    # Link to root README -> index.qmd
-                    target_output = "index.qmd"
-                elif len(parts) == 2:
-                    if parts[0] == ".":
-                        # Current directory README
-                        target_output = "index.qmd"
-                    else:
-                        # subdir/README.md
-                        parent_dir = parts[0]
-                        if parent_dir in parent_index_only:
-                            target_output = f"{parent_dir}/index.qmd"
-                        else:
-                            target_output = f"{parent_dir}.qmd"
-                else:
-                    # Deeper nesting: parent/subdir/README.md
-                    # Build the full path like "parent/subdir"
-                    full_path = "/".join(parts[:-1])  # Remove README.md
-                    # Check if this exact path is in allowlist
-                    if full_path in allowlist_entries:
-                        # This is a sub-entry with its own entry -> use .qmd
-                        target_output = f"{full_path}.qmd"
-                    elif parts[0] == ".":
-                        # ./subdir/README.md -> check if subdir has own entry
-                        subdir = parts[1]
-                        if subdir in parent_index_only:
-                            target_output = f"{subdir}/index.qmd"
-                        else:
-                            target_output = f"{subdir}.qmd"
-                    else:
-                        # parent/subdir where parent doesn't have own entry
-                        target_output = f"{full_path}/index.qmd"
-            else:
-                # Regular .md file -> convert to .qmd, keep path structure
-                target_output = "/".join(parts)[:-2] + "qmd"
-
-            # Compute relative path from current output file to target
-            current_parts = current_output_path.split("/")
-            target_parts = target_output.split("/")
-
-            # Special case: if current is a subdir file and target is a single-component file at root
-            # Example: current="magistral/vision", target="magistral.qmd"
-            if len(current_parts) > 1 and len(target_parts) == 1:
-                # Current is in subdir, target is at root level
-                # Go up to root: ../ for each level
-                up_count = len(current_parts) - 1
-                rel_parts = [".."] * up_count + [target_parts[0]]
-                new_url = "/".join(rel_parts)
-            else:
-                # Find common prefix
-                i = 0
-                while (
-                    i < min(len(current_parts) - 1, len(target_parts))
-                    and current_parts[i] == target_parts[i]
-                ):
-                    i += 1
-
-                # Build relative path: go up (../) then down to target
-                up_count = len(current_parts) - 1 - i
-                rel_parts = [".."] * up_count + target_parts[i:]
-
-                if not rel_parts or rel_parts == [".."]:
-                    # Points to same directory or parent
-                    new_url = "/".join(rel_parts) if rel_parts else "."
-                else:
-                    new_url = "/".join(rel_parts)
-
-            return f"[{text}]({new_url})"
-        except (ValueError, IndexError):
-            return m.group(0)
-
-    return LINK_RE.sub(repl, md)
-
-
-def write_qmd(out_path: Path, title: str, body_md: str):
-    out_path.parent.mkdir(parents=True, exist_ok=True)
-    fm = f"---\ntitle: {title!r}\nexecute:\n  eval: false\nformat:\n  html:\n    toc: true\n---\n\n"
-    out_path.write_text(fm + body_md, encoding="utf-8")
-
-
-def update_quarto_yml(generated: list[tuple[str, str, str]]):
-    """
-    Update _quarto.yml with the generated example files in the correct order.
-    This keeps the sidebar in sync with the allowlist.
-
-    Model Guides is now nested under "Getting Started" section.
-    Creates nested sections for models with sub-entries (e.g., magistral, ministral3).
-    Parent pages are now flat files (e.g., ministral3.qmd) with sub-pages in subdirs.
-    """
-    quarto_yml = ROOT / "_quarto.yml"
-    if not quarto_yml.exists():
-        print(f"[WARN] {quarto_yml} not found, skipping update", file=sys.stderr)
-        return
-
-    content = quarto_yml.read_text(encoding="utf-8")
-
-    # First pass: find all parents that have sub-entries
-    parents_with_subs = set()
-    for path, _name, _title in generated:
-        if "/" in path:
-            parent = path.split("/")[0]
-            parents_with_subs.add(parent)
-
-    # Build the YAML contents while preserving allowlist order
-    lines = []
-    processed_sections = set()
-
-    for path, _name, title in generated:
-        # Check if this is a parent page that has sub-pages
-        if path in parents_with_subs:
-            # This is a parent page with sub-pages - create a nested section
-            if path not in processed_sections:
-                processed_sections.add(path)
-                section_title = (
-                    title or path.replace("-", " ").replace("_", " ").title()
-                )
-                lines.append(f'                - section: "{section_title}"')
-                lines.append("                  contents:")
-                # Add the parent page first
-                lines.append(f"                    - docs/models/{path}.qmd")
-                # Then add all sub-pages
-                for sub_path, _sub_name, _sub_title in generated:
-                    if "/" in sub_path and sub_path.split("/")[0] == path:
-                        lines.append(
-                            f"                    - docs/models/{sub_path}.qmd"
-                        )
-        elif "/" not in path:
-            # This is a flat item with no sub-pages
-            # Skip if it was already included as part of a parent section
-            if path not in processed_sections:
-                lines.append(f"                - docs/models/{path}.qmd")
-
-    yaml_content = "\n".join(lines) + "\n"
-
-    # Pattern to match only the Model Guides contents, stopping at the next item
-    # in Getting Started (lines starting with 12 spaces: same level as the section)
-    pattern = r'(            - section: "Model Guides"\n              contents:)([^\n]*|.*?)(?=\n            - |\n        - section:|\n\nformat:)'
-
-    def replacement(match):
-        prefix = match.group(1)
-        return prefix + "\n" + yaml_content
-
-    new_content = re.sub(pattern, replacement, content, flags=re.DOTALL)
-
-    if new_content != content:
-        quarto_yml.write_text(new_content, encoding="utf-8")
-        print(f"Updated {quarto_yml}")
-    else:
-        print(f"No changes needed for {quarto_yml}")
-
-
-def main():
-    allow = read_allowlist()
-    if not EXAMPLES_DIR.exists():
-        print(f"[WARN] {EXAMPLES_DIR} not found", file=sys.stderr)
-        return
-
-    (OUTPUT_DIR / "assets").mkdir(parents=True, exist_ok=True)
-
-    # First pass: identify which parents have their own entry vs only sub-entries
-    parent_entries = set()  # Parents that have their own entry
-    parent_with_subs = set()  # Parents that have sub-entries
-    allowlist_entries = set()  # All entries in allowlist
-
-    for item in allow:
-        if isinstance(item, str):
-            name = item
-        else:
-            name = item.get("name")
-
-        allowlist_entries.add(name)
-
-        if "/" in name:
-            parent = name.split("/")[0]
-            parent_with_subs.add(parent)
-        else:
-            parent_entries.add(name)
-
-    # Parents with subs that DON'T have their own entry -> use index.qmd
-    parent_index_only = parent_with_subs - parent_entries
-
-    generated = []
-    seen_dirs = set()  # Track which parent directories we've created index for
-
-    for item in allow:
-        if isinstance(item, str):
-            name = item
-            title = None
-        else:
-            name = item.get("name")
-            title = item.get("title")
-
-        if not name:
-            print(f"[WARN] Skipping item without name: {item}", file=sys.stderr)
-            continue
-
-        src_dir = EXAMPLES_DIR / name
-        if not src_dir.exists() or not src_dir.is_dir():
-            print(f"[WARN] Skipping {name} (not a directory)", file=sys.stderr)
-            continue
-
-        readme = find_readme(src_dir)
-        if not readme:
-            print(f"[WARN] Skipping {name} (no README.md)", file=sys.stderr)
-            continue
-
-        md = readme.read_text(encoding="utf-8")
-
-        # Determine output path first (needed for link rewriting)
-        parts = name.split("/")
-        if len(parts) == 1:
-            # Simple case: no subdirectory
-            out_path = OUTPUT_DIR / f"{parts[0]}.qmd"
-            sidebar_path = parts[0]
-        else:
-            # Has subdirectory: e.g., magistral/think
-            parent = parts[0]
-            child = "-".join(parts[1:])  # handle nested subdirs
-            out_path = OUTPUT_DIR / parent / f"{child}.qmd"
-            sidebar_path = f"{parent}/{child}"
-
-        # Remove the first H1 (we use frontmatter title instead)
-        md, _ = remove_first_h1(md)
-        # Rewrite links between README files
-        md = rewrite_readme_links(
-            md,
-            src_dir,
-            EXAMPLES_DIR,
-            parent_index_only,
-            name,
-            allowlist_entries,
-            sidebar_path,
-        )
-        md = rewrite_and_copy_assets(md, src_dir, OUTPUT_DIR)
-
-        # Handle parent page generation for sub-entries
-        if len(parts) > 1:
-            # Has subdirectory: e.g., magistral/think
-            parent = parts[0]
-
-            # Create parent.qmd if not already done and parent doesn't have own entry
-            if parent not in seen_dirs and parent in parent_index_only:
-                parent_readme = find_readme(EXAMPLES_DIR / parent)
-                if parent_readme:
-                    parent_md = parent_readme.read_text(encoding="utf-8")
-                    parent_md, _ = remove_first_h1(parent_md)
-                    parent_md = rewrite_readme_links(
-                        parent_md,
-                        EXAMPLES_DIR / parent,
-                        EXAMPLES_DIR,
-                        parent_index_only,
-                        parent,
-                        allowlist_entries,
-                        parent,
-                    )
-                    parent_md = rewrite_and_copy_assets(
-                        parent_md, EXAMPLES_DIR / parent, OUTPUT_DIR
-                    )
-                    parent_title = parent.replace("-", " ").replace("_", " ").title()
-                    write_qmd(OUTPUT_DIR / f"{parent}.qmd", parent_title, parent_md)
-                    generated.append((parent, parent, parent_title))
-                    seen_dirs.add(parent)
-
-        if not title:
-            title = name.replace("/", " ").replace("-", " ").title()
-
-        write_qmd(out_path, title, md)
-        generated.append((sidebar_path, name, title))
-
-    # Index page - preserve allowlist order
-    if generated:
-        listing = "\n".join(
-            [f"- [{title}]({path}.qmd)" for path, name, title in generated]
-        )
-        index_md = (
-            "# Model Guides\n\nBelow are the curated examples for training various model architectures:\n\n"
-            + listing
-            + "\n"
-        )
-        index_fm = (
-            "---\nexecute:\n  eval: false\nformat:\n  html:\n    toc: true\n---\n\n"
-        )
-        (OUTPUT_DIR / "index.qmd").write_text(index_fm + index_md, encoding="utf-8")
-
-        # Auto-update _quarto.yml to keep sidebar in sync
-        update_quarto_yml(generated)
-
-
-if __name__ == "__main__":
-    main()

docs/sequence_parallelism.qmd

@@ -49,9 +49,9 @@ When sequence parallelism is enabled:
 To use sequence parallelism, you need:
 
 - Multiple GPUs (at least 2)
-- The `ring-flash-attn` package. Install with:
-  - `pip install axolotl[ring-flash-attn]` (preferred)
-  - `pip install ring-flash-attn>=0.1.4`
+- The `ring-flash-attn` package. Install with either `uv sync --extra ring-flash-attn`
+  (from a cloned repository) or `uv pip install ring-flash-attn>=0.1.4`.
+- Flash Attention installed separately with `uv pip install flash-attn --no-build-isolation`.
 
 ## Limitations
 

docs/telemetry.qmd

@@ -1,61 +0,0 @@
----
-title: Telemetry
-description: A description of the telemetry implementation in Axolotl.
----
-
-# Telemetry in Axolotl
-
-Axolotl implements anonymous telemetry to help maintainers understand how the library
-is used and where users encounter issues. This data helps prioritize features, optimize
-performance, and fix bugs.
-
-## Data Collection
-
-We collect:
-
-- System info: OS, Python version, Axolotl version, PyTorch version, Transformers
-version, etc.
-- Hardware info: CPU count, memory, GPU count and models
-- Runtime metrics: Training progress, memory usage, timing information
-- Usage patterns: Models (from a whitelist) and configurations used
-- Error tracking: Stack traces and error messages (sanitized to remove personal
-information)
-
-Personally identifiable information (PII) is not collected.
-
-## Implementation
-
-Telemetry is implemented using PostHog and consists of:
-
-- `axolotl.telemetry.TelemetryManager`: A singleton class that initializes the
-telemetry system and provides methods for tracking events.
-- `axolotl.telemetry.errors.send_errors`: A decorator that captures exceptions and
-sends sanitized stack traces.
-- `axolotl.telemetry.runtime_metrics.RuntimeMetricsTracker`: A class that tracks
-runtime metrics during training.
-- `axolotl.telemetry.callbacks.TelemetryCallback`: A Trainer callback that sends
-runtime metrics telemetry.
-
-The telemetry system will block training startup for 10 seconds to ensure users are
-aware of data collection, unless telemetry is explicitly enabled or disabled.
-
-## Opt-Out Mechanism
-
-Telemetry is **enabled by default** on an opt-out basis. To disable it, set
-`AXOLOTL_DO_NOT_TRACK=1` or `DO_NOT_TRACK=1`.
-
-A warning message will be logged on start to clearly inform users about telemetry.
-We will remove this after some period.
-
-To hide the warning message about telemetry that is displayed on train, etc. startup,
-explicitly set: `AXOLOTL_DO_NOT_TRACK=0` (enable telemetry) or `AXOLOTL_DO_NOT_TRACK=1`
-(explicitly disable telemetry).
-
-## Privacy
-
-- All path-like config information is automatically redacted from telemetry data
-- Model information is only collected for whitelisted organizations
-    - See `axolotl/telemetry/whitelist.yaml` for the set of whitelisted organizations
-- Each run generates a unique anonymous ID
-    - This allows us to link different telemetry events in a single same training run
-- Telemetry is only sent from the main process to avoid duplicate events

examples/LiquidAI/README.md

@@ -6,17 +6,20 @@ LFM2 features a new hybrid Liquid architecture with multiplicative gates, short-
 
 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'
+    # Ensure you have a compatible version of PyTorch installed
+    # Option A: manage dependencies in your project
+    uv add 'axolotl>=0.12.0'
+    uv pip install flash-attn --no-build-isolation
+
+    # Option B: quick install
+    uv pip install 'axolotl>=0.12.0'
+    uv pip install flash-attn --no-build-isolation
     ```
 
 2.  Run one of the finetuning examples below.
@@ -33,19 +36,11 @@ Thanks to the team at LiquidAI for giving us early access to prepare for these r
     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
+  uv 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).
@@ -55,13 +50,14 @@ Thanks to the team at LiquidAI for giving us early access to prepare for these r
 
 ## Optimization Guides
 
-- [Optimizations Guide](https://docs.axolotl.ai/docs/optimizations.html)
+- [Multi-GPU Training](https://docs.axolotl.ai/docs/multi-gpu.html)
+- [LoRA Optimizations](https://docs.axolotl.ai/docs/lora_optims.html)
+- [Multi-Node Training](https://docs.axolotl.ai/docs/multi-node.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)

examples/LiquidAI/lfm2-350m-fft.yaml

@@ -1,7 +1,6 @@
 base_model: LiquidAI/LFM2-350M
 
-plugins:
-  - axolotl.integrations.cut_cross_entropy.CutCrossEntropyPlugin
+chunked_cross_entropy: true
 
 eot_tokens:
   - "<|im_end|>"

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

examples/LiquidAI/lfm2-vl-lora.yaml

@@ -3,9 +3,6 @@ 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

examples/apertus/README.md

@@ -15,8 +15,8 @@ This guide shows how to fine-tune it with Axolotl with multi-turn conversations
 git clone https://github.com/axolotl-ai-cloud/axolotl.git
 cd axolotl
 
-pip3 install packaging==26.0 setuptools==75.8.0 wheel ninja
-pip3 install --no-build-isolation -e '.[flash-attn]'
+uv sync
+uv pip install flash-attn --no-build-isolation
 
 # Install CCE https://docs.axolotl.ai/docs/custom_integrations.html#cut-cross-entropy
 python scripts/cutcrossentropy_install.py | sh
@@ -31,7 +31,7 @@ python scripts/cutcrossentropy_install.py | sh
 # 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
+uv pip install git+https://github.com/nickjbrowning/XIELU@59d6031 --no-build-isolation --no-deps
 ```
 
 For any installation errors, see [XIELU Installation Issues](#xielu-installation-issues)
@@ -67,7 +67,7 @@ If those didn't help, please try the below solutions:
 1. Pass env for CMAKE and try install again:
 
     ```bash
-    Python_EXECUTABLE=$(which python) pip3 install git+https://github.com/nickjbrowning/XIELU@59d6031 --no-build-isolation --no-deps
+    Python_EXECUTABLE=$(which python) uv pip install git+https://github.com/nickjbrowning/XIELU@59d6031 --no-build-isolation --no-deps
     ```
 
 2. Git clone the repo and manually hardcode python path:
@@ -92,7 +92,7 @@ If those didn't help, please try the below solutions:
     ```
 
     ```bash
-    pip3 install . --no-build-isolation --no-deps
+    uv pip install . --no-build-isolation --no-deps
     ```
 
 ## Optimization Guides

examples/arcee/README.md

@@ -17,8 +17,8 @@ Thanks to the team at Arcee.ai for using Axolotl in supervised fine-tuning the A
 git clone https://github.com/axolotl-ai-cloud/axolotl.git
 cd axolotl
 
-pip3 install packaging==26.0 setuptools==75.8.0 wheel ninja
-pip3 install --no-build-isolation -e '.[flash-attn]'
+uv sync
+uv pip install flash-attn --no-build-isolation
 
 # Install CCE https://docs.axolotl.ai/docs/custom_integrations.html#cut-cross-entropy
 python scripts/cutcrossentropy_install.py | sh

examples/colab-notebooks/colab-axolotl-example.ipynb

@@ -12,10 +12,10 @@
     "\n",
     "Axolotl is the most performant LLM post-training framework available, delivering faster training with efficient, consistent and stable performance. Train your workload and ship your product 30% faster; saving you both time and money.\n",
     "\n",
-    "- ⭐ us on [GitHub](https://github.com/axolotl-ai-cloud/axolotl)\n",
-    "- 📜 Read the [Docs](http://docs.axolotl.ai/)\n",
-    "- 💬 Chat with us on [Discord](https://discord.gg/mnpEYgRUmD)\n",
-    "- 📰 Get updates on [X/Twitter](https://x.com/axolotl_ai)\n"
+    "- \u2b50 us on [GitHub](https://github.com/axolotl-ai-cloud/axolotl)\n",
+    "- \ud83d\udcdc Read the [Docs](http://docs.axolotl.ai/)\n",
+    "- \ud83d\udcac Chat with us on [Discord](https://discord.gg/mnpEYgRUmD)\n",
+    "- \ud83d\udcf0 Get updates on [X/Twitter](https://x.com/axolotl_ai)\n"
    ]
   },
   {
@@ -39,8 +39,8 @@
    "source": [
     "%%capture\n",
     "# This step can take ~5-10 minutes to install dependencies\n",
-    "!pip install --no-build-isolation axolotl[flash-attn]>=0.9.1\n",
-    "!pip install \"cut-cross-entropy[transformers] @ git+https://github.com/axolotl-ai-cloud/ml-cross-entropy.git@63b15e6\""
+    "!uv pip install --no-build-isolation axolotl>=0.9.1\n!uv pip install flash-attn --no-build-isolation\n",
+    "!uv pip install \"cut-cross-entropy[transformers] @ git+https://github.com/axolotl-ai-cloud/ml-cross-entropy.git@147ea28\""
    ]
   },
   {
@@ -253,6 +253,7 @@
    "source": [
     "from axolotl.utils import set_pytorch_cuda_alloc_conf\n",
     "\n",
+    "# Set \"PYTORCH_CUDA_ALLOC_CONF\" env to save memory\n",
     "set_pytorch_cuda_alloc_conf()"
    ]
   },
@@ -1370,7 +1371,7 @@
        "version_minor": 0
       },
       "text/plain": [
-       "VBox(children=(HTML(value='<center> <img\\nsrc=https://huggingface.co/front/assets/huggingface_logo-noborder.sv…"
+       "VBox(children=(HTML(value='<center> <img\\nsrc=https://huggingface.co/front/assets/huggingface_logo-noborder.sv\u2026"
       ]
      },
      "metadata": {},
@@ -1728,9 +1729,9 @@
       "description": "",
       "description_tooltip": null,
       "layout": "IPY_MODEL_12815f401eba44658caa7b2e490137a8",
-      "placeholder": "​",
+      "placeholder": "\u200b",
       "style": "IPY_MODEL_30e02aa2d0d241979369e598287f2639",
-      "value": "Drop Samples with Zero Trainable Tokens (num_proc=2): 100%"
+      "value": "Drop\u2007Samples\u2007with\u2007Zero\u2007Trainable\u2007Tokens\u2007(num_proc=2):\u2007100%"
      }
     },
     "083f9cda8d754c168beee10d2f8955a2": {
@@ -1773,9 +1774,9 @@
       "description": "",
       "description_tooltip": null,
       "layout": "IPY_MODEL_b195f160ca20442fadd8b5aed0ee41af",
-      "placeholder": "​",
+      "placeholder": "\u200b",
       "style": "IPY_MODEL_ca65e32eb52f48c09a84b33cb18f22cd",
-      "value": " 11.4M/11.4M [00:00&lt;00:00, 21.8MB/s]"
+      "value": "\u200711.4M/11.4M\u2007[00:00&lt;00:00,\u200721.8MB/s]"
      }
     },
     "0a46ad75c198463d843fb35e813642cb": {
@@ -1916,7 +1917,7 @@
       "description": "",
       "description_tooltip": null,
       "layout": "IPY_MODEL_b1bea589efa14258a9982071b87938bf",
-      "placeholder": "​",
+      "placeholder": "\u200b",
       "style": "IPY_MODEL_590eef89881545aa8bbef9a8bbe7fb00",
       "value": "\n<b>Pro Tip:</b> If you don't already have one, you can create a dedicated\n'notebooks' token with 'write' access, that you can then easily reuse for all\nnotebooks. </center>"
      }
@@ -1937,9 +1938,9 @@
       "description": "",
       "description_tooltip": null,
       "layout": "IPY_MODEL_bfcdbba993b74972a9e3e575f86908ff",
-      "placeholder": "​",
+      "placeholder": "\u200b",
       "style": "IPY_MODEL_6ebb2ec171414e47a14765505f64bb3c",
-      "value": " 3.84G/3.84G [00:09&lt;00:00, 664MB/s]"
+      "value": "\u20073.84G/3.84G\u2007[00:09&lt;00:00,\u2007664MB/s]"
      }
     },
     "0e936d9dbf9c4fdd86bbfe9730dedc47": {
@@ -2295,9 +2296,9 @@
       "description": "",
       "description_tooltip": null,
       "layout": "IPY_MODEL_349eee9f56d64f0cba6fc24ff2c50c9b",
-      "placeholder": "​",
+      "placeholder": "\u200b",
       "style": "IPY_MODEL_7e5d3774060e4589aa65982da5ea4ef4",
-      "value": " 9985/9985 [00:04&lt;00:00, 2604.11 examples/s]"
+      "value": "\u20079985/9985\u2007[00:04&lt;00:00,\u20072604.11\u2007examples/s]"
      }
     },
     "16d1283741404b7bb319094c992fce01": {
@@ -2316,9 +2317,9 @@
       "description": "",
       "description_tooltip": null,
       "layout": "IPY_MODEL_a4e5789584564049b83df7c6c54a3e08",
-      "placeholder": "​",
+      "placeholder": "\u200b",
       "style": "IPY_MODEL_ff3a94b146a948b6907f5d80c7157f99",
-      "value": " 9985/0 [00:00&lt;00:00, 50763.46 examples/s]"
+      "value": "\u20079985/0\u2007[00:00&lt;00:00,\u200750763.46\u2007examples/s]"
      }
     },
     "1811cda0644e4190a9469d1774435d82": {
@@ -2389,9 +2390,9 @@
       "description": "",
       "description_tooltip": null,
       "layout": "IPY_MODEL_e366ae3fceec4566b9ed303d6c5f90af",
-      "placeholder": "​",
+      "placeholder": "\u200b",
       "style": "IPY_MODEL_5dd7d150dbe04f08b165ce7f2c27cd11",
-      "value": "model-00008-of-00008.safetensors: 100%"
+      "value": "model-00008-of-00008.safetensors:\u2007100%"
      }
     },
     "19127c7bb1554ccbac877059f9a82db0": {
@@ -2560,9 +2561,9 @@
       "description": "",
       "description_tooltip": null,
       "layout": "IPY_MODEL_0dea5caa27384f5689e3cab51f558727",
-      "placeholder": "​",
+      "placeholder": "\u200b",
       "style": "IPY_MODEL_a6f48410b9964fefba0c3009a77dc838",
-      "value": " 9.68k/9.68k [00:00&lt;00:00, 812kB/s]"
+      "value": "\u20079.68k/9.68k\u2007[00:00&lt;00:00,\u2007812kB/s]"
      }
     },
     "1f7d30f71bbd4547a9150d21da071055": {
@@ -2633,9 +2634,9 @@
       "description": "",
       "description_tooltip": null,
       "layout": "IPY_MODEL_f4a1795dc7514a718f478245f521f0ba",
-      "placeholder": "​",
+      "placeholder": "\u200b",
       "style": "IPY_MODEL_5e746eb25bbe416fb585fa24e79f5177",
-      "value": "model-00002-of-00008.safetensors: 100%"
+      "value": "model-00002-of-00008.safetensors:\u2007100%"
      }
     },
     "20352e5f58d24bb8b1f3940efd14fe4a": {
@@ -2706,9 +2707,9 @@
       "description": "",
       "description_tooltip": null,
       "layout": "IPY_MODEL_1c6f1f10667545aaab958016ba7e2c94",
-      "placeholder": "​",
+      "placeholder": "\u200b",
       "style": "IPY_MODEL_e6e969610738449887259063967f82b0",
-      "value": " 2.78M/2.78M [00:00&lt;00:00, 17.8MB/s]"
+      "value": "\u20072.78M/2.78M\u2007[00:00&lt;00:00,\u200717.8MB/s]"
      }
     },
     "258b7c635c1045329d4669e48c46ccd5": {
@@ -3055,9 +3056,9 @@
       "description": "",
       "description_tooltip": null,
       "layout": "IPY_MODEL_be724f04b03942b2a033a7e8898bb4fd",
-      "placeholder": "​",
+      "placeholder": "\u200b",
       "style": "IPY_MODEL_fcbab4d8dced41a18dfccce81e3a45a0",
-      "value": "model-00005-of-00008.safetensors: 100%"
+      "value": "model-00005-of-00008.safetensors:\u2007100%"
      }
     },
     "3036608c71904ce9ae4bb2a9fa8802d9": {
@@ -3076,9 +3077,9 @@
       "description": "",
       "description_tooltip": null,
       "layout": "IPY_MODEL_5ca6be24acb548cea130bd58e9954c7c",
-      "placeholder": "​",
+      "placeholder": "\u200b",
       "style": "IPY_MODEL_5cfb02ee044b4011a378efa8b54a370f",
-      "value": " 3.96G/3.96G [00:10&lt;00:00, 531MB/s]"
+      "value": "\u20073.96G/3.96G\u2007[00:10&lt;00:00,\u2007531MB/s]"
      }
     },
     "30a81da86f8043eca301e86a8651201a": {
@@ -3628,9 +3629,9 @@
       "description": "",
       "description_tooltip": null,
       "layout": "IPY_MODEL_8f5bd719974e41c3a8dd9a5b0d3d71e6",
-      "placeholder": "​",
+      "placeholder": "\u200b",
       "style": "IPY_MODEL_b87c84de30e84b3abf4871461fb9cbd3",
-      "value": "Loading checkpoint shards: 100%"
+      "value": "Loading\u2007checkpoint\u2007shards:\u2007100%"
      }
     },
     "41f3b32c2f6b4034ae7a3b9124e28bc7": {
@@ -3790,7 +3791,7 @@
       "description": "",
       "description_tooltip": null,
       "layout": "IPY_MODEL_39789237703c4a418134243055c9cbf5",
-      "placeholder": "​",
+      "placeholder": "\u200b",
       "style": "IPY_MODEL_a3a945817f684328b34651fe052393ec",
       "value": "Connecting..."
      }
@@ -4076,9 +4077,9 @@
       "description": "",
       "description_tooltip": null,
       "layout": "IPY_MODEL_4d468f96ec924681ad65eb671674b93e",
-      "placeholder": "​",
+      "placeholder": "\u200b",
       "style": "IPY_MODEL_ad7599de524549c48bf2d3124ad4b299",
-      "value": "Dropping Long Sequences (num_proc=2): 100%"
+      "value": "Dropping\u2007Long\u2007Sequences\u2007(num_proc=2):\u2007100%"
      }
     },
     "5ca240f31e6b44e3882c5eb37cd5a309": {
@@ -4470,9 +4471,9 @@
       "description": "",
       "description_tooltip": null,
       "layout": "IPY_MODEL_5e18768f7ad6434ba8b8b8a2e853e204",
-      "placeholder": "​",
+      "placeholder": "\u200b",
       "style": "IPY_MODEL_bb33aec33a6447078c31bfd728942994",
-      "value": " 728/728 [00:00&lt;00:00, 20.3kB/s]"
+      "value": "\u2007728/728\u2007[00:00&lt;00:00,\u200720.3kB/s]"
      }
     },
     "62e302ebdad64aada0ffe64ae1c873f3": {
@@ -4635,9 +4636,9 @@
       "description": "",
       "description_tooltip": null,
       "layout": "IPY_MODEL_81c3db71ac704280ad030072655f1537",
-      "placeholder": "​",
+      "placeholder": "\u200b",
       "style": "IPY_MODEL_042e091f75694c47aee761e760e76773",
-      "value": " 9985/9985 [00:02&lt;00:00, 3977.47 examples/s]"
+      "value": "\u20079985/9985\u2007[00:02&lt;00:00,\u20073977.47\u2007examples/s]"
      }
     },
     "67da6c4260574869aa24c3cbc1bc1654": {
@@ -4777,7 +4778,7 @@
       "description_tooltip": null,
       "disabled": false,
       "layout": "IPY_MODEL_2e257c8be2da40b4bb67a9e4ab6811f3",
-      "placeholder": "​",
+      "placeholder": "\u200b",
       "style": "IPY_MODEL_56e3768bef5a4b9db4168c5c17f509c2",
       "value": ""
      }
@@ -4822,9 +4823,9 @@
       "description": "",
       "description_tooltip": null,
       "layout": "IPY_MODEL_41f3b32c2f6b4034ae7a3b9124e28bc7",
-      "placeholder": "​",
+      "placeholder": "\u200b",
       "style": "IPY_MODEL_a10d0a76010f4e508c65a9b69ebc5156",
-      "value": "Tokenizing Prompts (num_proc=2): 100%"
+      "value": "Tokenizing\u2007Prompts\u2007(num_proc=2):\u2007100%"
      }
     },
     "704f2f5a9b1c49d5a75a0025a5dda11b": {
@@ -5070,9 +5071,9 @@
       "description": "",
       "description_tooltip": null,
       "layout": "IPY_MODEL_93a44a11aa4846fa8efc6c1413ef1627",
-      "placeholder": "​",
+      "placeholder": "\u200b",
       "style": "IPY_MODEL_a55060adc3564407ac81ad7297d34aaa",
-      "value": "train.jsonl: 100%"
+      "value": "train.jsonl:\u2007100%"
      }
     },
     "7be6f04c284e4326bb4ff3d301e7b3c6": {
@@ -5137,9 +5138,9 @@
       "description": "",
       "description_tooltip": null,
       "layout": "IPY_MODEL_7fd44cf9ca6e4726bfd7ac21846d6a14",
-      "placeholder": "​",
+      "placeholder": "\u200b",
       "style": "IPY_MODEL_366a343b62fa47d8985a3bd464d99f9e",
-      "value": "config.json: 100%"
+      "value": "config.json:\u2007100%"
      }
     },
     "7cd0b85ebd204b7aba908417811ce4e0": {
@@ -5338,9 +5339,9 @@
       "description": "",
       "description_tooltip": null,
       "layout": "IPY_MODEL_67da6c4260574869aa24c3cbc1bc1654",
-      "placeholder": "​",
+      "placeholder": "\u200b",
       "style": "IPY_MODEL_94b9088614464f60a203de39dbcae853",
-      "value": " 8/8 [01:47&lt;00:00, 11.64s/it]"
+      "value": "\u20078/8\u2007[01:47&lt;00:00,\u200711.64s/it]"
      }
     },
     "823f1c78f15043e38bbd4dca3932a86a": {
@@ -5487,7 +5488,7 @@
       "description": "",
       "description_tooltip": null,
       "layout": "IPY_MODEL_8640ac440fbc4644b9a3af7ba3ae7183",
-      "placeholder": "​",
+      "placeholder": "\u200b",
       "style": "IPY_MODEL_5cea7996f02040b187ece0bb2d6a8d1f",
       "value": "<center> <img\nsrc=https://huggingface.co/front/assets/huggingface_logo-noborder.svg\nalt='Hugging Face'> <br> Copy a token from <a\nhref=\"https://huggingface.co/settings/tokens\" target=\"_blank\">your Hugging Face\ntokens page</a> and paste it below. <br> Immediately click login after copying\nyour token or it might be stored in plain text in this notebook file. </center>"
      }
@@ -5508,9 +5509,9 @@
       "description": "",
       "description_tooltip": null,
       "layout": "IPY_MODEL_ef223e8504b64e3592589880326aaf41",
-      "placeholder": "​",
+      "placeholder": "\u200b",
       "style": "IPY_MODEL_598da69727bd4fb8b1caf465ac736d7a",
-      "value": " 1.67M/1.67M [00:00&lt;00:00, 19.0MB/s]"
+      "value": "\u20071.67M/1.67M\u2007[00:00&lt;00:00,\u200719.0MB/s]"
      }
     },
     "897b77a56c09479bb11d7f2a30997e55": {
@@ -5716,9 +5717,9 @@
       "description": "",
       "description_tooltip": null,
       "layout": "IPY_MODEL_37de928300e34184881039378bd75e7f",
-      "placeholder": "​",
+      "placeholder": "\u200b",
       "style": "IPY_MODEL_0e936d9dbf9c4fdd86bbfe9730dedc47",
-      "value": " 3.96G/3.96G [00:13&lt;00:00, 273MB/s]"
+      "value": "\u20073.96G/3.96G\u2007[00:13&lt;00:00,\u2007273MB/s]"
      }
     },
     "936d04b5fe1b4c63bf0b080e423d051b": {
@@ -6049,9 +6050,9 @@
       "description": "",
       "description_tooltip": null,
       "layout": "IPY_MODEL_d955dcaa0e944e719f3a06139dd54a03",
-      "placeholder": "​",
+      "placeholder": "\u200b",
       "style": "IPY_MODEL_d3de2662c7964f1ba96e58da382af720",
-      "value": "merges.txt: 100%"
+      "value": "merges.txt:\u2007100%"
      }
     },
     "9cd5211b5d8b457aa0002f1d17b80028": {
@@ -6070,9 +6071,9 @@
       "description": "",
       "description_tooltip": null,
       "layout": "IPY_MODEL_6932489232ec4ab18a160b1e7fbcdfe1",
-      "placeholder": "​",
+      "placeholder": "\u200b",
       "style": "IPY_MODEL_4540927d98f54466b434ba4c0edf045d",
-      "value": "model-00007-of-00008.safetensors: 100%"
+      "value": "model-00007-of-00008.safetensors:\u2007100%"
      }
     },
     "9d4897eefb5f48259ffb2d23e332f752": {
@@ -6302,9 +6303,9 @@
       "description": "",
       "description_tooltip": null,
       "layout": "IPY_MODEL_3aaecbf540f54a2db9ab0931e3b1fe57",
-      "placeholder": "​",
+      "placeholder": "\u200b",
       "style": "IPY_MODEL_9e333ed3b5014069ac1dd969255dd591",
-      "value": " 239/239 [00:00&lt;00:00, 30.9kB/s]"
+      "value": "\u2007239/239\u2007[00:00&lt;00:00,\u200730.9kB/s]"
      }
     },
     "a20927bf5f2c41f58c1e31ac858ab36c": {
@@ -6323,9 +6324,9 @@
       "description": "",
       "description_tooltip": null,
       "layout": "IPY_MODEL_1811cda0644e4190a9469d1774435d82",
-      "placeholder": "​",
+      "placeholder": "\u200b",
       "style": "IPY_MODEL_35c811d2ae8e43f3b5cecbdd3cfa857f",
-      "value": "tokenizer.json: 100%"
+      "value": "tokenizer.json:\u2007100%"
      }
     },
     "a3a945817f684328b34651fe052393ec": {
@@ -6359,9 +6360,9 @@
       "description": "",
       "description_tooltip": null,
       "layout": "IPY_MODEL_ed5ca967ad5342929e578ac6aa4dc4c0",
-      "placeholder": "​",
+      "placeholder": "\u200b",
       "style": "IPY_MODEL_af401d117d5047629d3a6e2361757b62",
-      "value": "model-00001-of-00008.safetensors: 100%"
+      "value": "model-00001-of-00008.safetensors:\u2007100%"
      }
     },
     "a4e5789584564049b83df7c6c54a3e08": {
@@ -6493,9 +6494,9 @@
       "description": "",
       "description_tooltip": null,
       "layout": "IPY_MODEL_fa1282ccc7544e4f818e2f03ccffe4a5",
-      "placeholder": "​",
+      "placeholder": "\u200b",
       "style": "IPY_MODEL_bbbf575d2a4b4c6ea8389be79b2a6039",
-      "value": "model.safetensors.index.json: 100%"
+      "value": "model.safetensors.index.json:\u2007100%"
      }
     },
     "ab93eabd7cea4b94b4b7a387f101e8a1": {
@@ -6581,9 +6582,9 @@
       "description": "",
       "description_tooltip": null,
       "layout": "IPY_MODEL_62e302ebdad64aada0ffe64ae1c873f3",
-      "placeholder": "​",
+      "placeholder": "\u200b",
       "style": "IPY_MODEL_bd1b0dfed6d34d16af33a4a58330f5ec",
-      "value": "Saving the dataset (1/1 shards): 100%"
+      "value": "Saving\u2007the\u2007dataset\u2007(1/1\u2007shards):\u2007100%"
      }
     },
     "ad7599de524549c48bf2d3124ad4b299": {
@@ -6966,9 +6967,9 @@
       "description": "",
       "description_tooltip": null,
       "layout": "IPY_MODEL_2b3a2659b12244bd8548320320016dbf",
-      "placeholder": "​",
+      "placeholder": "\u200b",
       "style": "IPY_MODEL_0cd7efffbb3c4c4b972e63749f61ab97",
-      "value": "Generating train split: "
+      "value": "Generating\u2007train\u2007split:\u2007"
      }
     },
     "b87c84de30e84b3abf4871461fb9cbd3": {
@@ -7084,9 +7085,9 @@
       "description": "",
       "description_tooltip": null,
       "layout": "IPY_MODEL_0f480e3a0b0a45d2a2d2dec3cad923f3",
-      "placeholder": "​",
+      "placeholder": "\u200b",
       "style": "IPY_MODEL_fcb30372e7404c5d8a1ad4df91e6c7b2",
-      "value": " 1.91G/1.91G [00:05&lt;00:00, 444MB/s]"
+      "value": "\u20071.91G/1.91G\u2007[00:05&lt;00:00,\u2007444MB/s]"
      }
     },
     "bd1b0dfed6d34d16af33a4a58330f5ec": {
@@ -7324,9 +7325,9 @@
       "description": "",
       "description_tooltip": null,
       "layout": "IPY_MODEL_158c8b85dbf34de6a94b4e35e2fc7d5a",
-      "placeholder": "​",
+      "placeholder": "\u200b",
       "style": "IPY_MODEL_0b4c9753a7cb4354b8e5f187e6e1ad7c",
-      "value": " 3.96G/3.96G [00:15&lt;00:00, 564MB/s]"
+      "value": "\u20073.96G/3.96G\u2007[00:15&lt;00:00,\u2007564MB/s]"
      }
     },
     "c0991cf63ee6458b96e9a75e7a88b61a": {
@@ -7345,9 +7346,9 @@
       "description": "",
       "description_tooltip": null,
       "layout": "IPY_MODEL_ed28e2e0410d4e0b855467e798e53d66",
-      "placeholder": "​",
+      "placeholder": "\u200b",
       "style": "IPY_MODEL_d93f134f802b4b69b575bdaf07dbd27c",
-      "value": "tokenizer_config.json: 100%"
+      "value": "tokenizer_config.json:\u2007100%"
      }
     },
     "c12ea43372ac4d57bb9605f1a429b397": {
@@ -7580,9 +7581,9 @@
       "description": "",
       "description_tooltip": null,
       "layout": "IPY_MODEL_8bc9d8ba866c442b9118d9630009939c",
-      "placeholder": "​",
+      "placeholder": "\u200b",
       "style": "IPY_MODEL_9f56a2d9979c4bd8928c644c22c3ecdf",
-      "value": "model-00003-of-00008.safetensors: 100%"
+      "value": "model-00003-of-00008.safetensors:\u2007100%"
      }
     },
     "c6164e05a1914ae48083db9ad7f4ef7c": {
@@ -7693,9 +7694,9 @@
       "description": "",
       "description_tooltip": null,
       "layout": "IPY_MODEL_e40d1c1ac9494b3bade9858324e7ffdf",
-      "placeholder": "​",
+      "placeholder": "\u200b",
       "style": "IPY_MODEL_d65b6b060d9845779299491ac5599c31",
-      "value": " 9985/9985 [01:04&lt;00:00, 189.08 examples/s]"
+      "value": "\u20079985/9985\u2007[01:04&lt;00:00,\u2007189.08\u2007examples/s]"
      }
     },
     "c7433acd3c4841e6958ae8f7e87b1808": {
@@ -7736,9 +7737,9 @@
       "description": "",
       "description_tooltip": null,
       "layout": "IPY_MODEL_0077aedc3d174560bce924ee89e9c006",
-      "placeholder": "​",
+      "placeholder": "\u200b",
       "style": "IPY_MODEL_00321cce58884f6f9b3855a21fcd9187",
-      "value": "Add position_id column (Sample Packing) (num_proc=2): 100%"
+      "value": "Add\u2007position_id\u2007column\u2007(Sample\u2007Packing)\u2007(num_proc=2):\u2007100%"
      }
     },
     "ca65e32eb52f48c09a84b33cb18f22cd": {
@@ -8161,9 +8162,9 @@
       "description": "",
       "description_tooltip": null,
       "layout": "IPY_MODEL_63580b6fb30642479fe3000915bf551a",
-      "placeholder": "​",
+      "placeholder": "\u200b",
       "style": "IPY_MODEL_8f726dbfb45d4528afa33e36a6313267",
-      "value": " 27.3M/27.3M [00:00&lt;00:00, 31.0MB/s]"
+      "value": "\u200727.3M/27.3M\u2007[00:00&lt;00:00,\u200731.0MB/s]"
      }
     },
     "d43c6df07ddb466587807d6dbe1ff614": {
@@ -8182,9 +8183,9 @@
       "description": "",
       "description_tooltip": null,
       "layout": "IPY_MODEL_8c4d4fc5a30f4e7cb3be53fe2adda33d",
-      "placeholder": "​",
+      "placeholder": "\u200b",
       "style": "IPY_MODEL_e90658f4bcb642baa78426012f863152",
-      "value": "model-00004-of-00008.safetensors: 100%"
+      "value": "model-00004-of-00008.safetensors:\u2007100%"
      }
     },
     "d65b6b060d9845779299491ac5599c31": {
@@ -8473,9 +8474,9 @@
       "description": "",
       "description_tooltip": null,
       "layout": "IPY_MODEL_34cf3df51fbc41cabfdbba153c007f0e",
-      "placeholder": "​",
+      "placeholder": "\u200b",
       "style": "IPY_MODEL_ac764024cf1c4e08ba7749afd2cd20ac",
-      "value": "vocab.json: 100%"
+      "value": "vocab.json:\u2007100%"
      }
     },
     "dfd2a2649b8341ef913207526708aff1": {
@@ -8668,9 +8669,9 @@
       "description": "",
       "description_tooltip": null,
       "layout": "IPY_MODEL_c6164e05a1914ae48083db9ad7f4ef7c",
-      "placeholder": "​",
+      "placeholder": "\u200b",
       "style": "IPY_MODEL_813621384dc748b0ad06775e22761c0b",
-      "value": " 9985/9985 [00:03&lt;00:00, 3622.89 examples/s]"
+      "value": "\u20079985/9985\u2007[00:03&lt;00:00,\u20073622.89\u2007examples/s]"
      }
     },
     "e400cbf14bcc446a9d33b210cd93550b": {
@@ -9064,9 +9065,9 @@
       "description": "",
       "description_tooltip": null,
       "layout": "IPY_MODEL_fba7aa824b38467ab3061b226114cdec",
-      "placeholder": "​",
+      "placeholder": "\u200b",
       "style": "IPY_MODEL_f3075dccbd2747b4a7913b66f44f2596",
-      "value": " 3.96G/3.96G [00:13&lt;00:00, 398MB/s]"
+      "value": "\u20073.96G/3.96G\u2007[00:13&lt;00:00,\u2007398MB/s]"
      }
     },
     "ec030fc3c346426f9abc3a89892258d3": {
@@ -9109,9 +9110,9 @@
       "description": "",
       "description_tooltip": null,
       "layout": "IPY_MODEL_936d04b5fe1b4c63bf0b080e423d051b",
-      "placeholder": "​",
+      "placeholder": "\u200b",
       "style": "IPY_MODEL_f1cef8e8dc2646fb9fd09f3b09081074",
-      "value": " 36.5k/36.5k [00:00&lt;00:00, 4.32MB/s]"
+      "value": "\u200736.5k/36.5k\u2007[00:00&lt;00:00,\u20074.32MB/s]"
      }
     },
     "ed28e2e0410d4e0b855467e798e53d66": {
@@ -9421,9 +9422,9 @@
       "description": "",
       "description_tooltip": null,
       "layout": "IPY_MODEL_735d4f225b24414294fc1b213c61223c",
-      "placeholder": "​",
+      "placeholder": "\u200b",
       "style": "IPY_MODEL_5e5e15b0569b474c9620083b3ec6af55",
-      "value": "generation_config.json: 100%"
+      "value": "generation_config.json:\u2007100%"
      }
     },
     "f4667818b9d34a09891cd727a429a610": {
@@ -9442,9 +9443,9 @@
       "description": "",
       "description_tooltip": null,
       "layout": "IPY_MODEL_4b27c267393640f28f6eae0875bd2ed9",
-      "placeholder": "​",
+      "placeholder": "\u200b",
       "style": "IPY_MODEL_9858cb74a09748a39e8149baac96702c",
-      "value": " 3.96G/3.96G [00:11&lt;00:00, 457MB/s]"
+      "value": "\u20073.96G/3.96G\u2007[00:11&lt;00:00,\u2007457MB/s]"
      }
     },
     "f4a1795dc7514a718f478245f521f0ba": {
@@ -9829,9 +9830,9 @@
       "description": "",
       "description_tooltip": null,
       "layout": "IPY_MODEL_d1f9b10c130542f094c8fd3d1e23b5e9",
-      "placeholder": "​",
+      "placeholder": "\u200b",
       "style": "IPY_MODEL_e575d87a7efe4ec7b1efde489839d4a6",
-      "value": "model-00006-of-00008.safetensors: 100%"
+      "value": "model-00006-of-00008.safetensors:\u2007100%"
      }
     },
     "fe18bba7f3fb4c31bf840541f36b3425": {
@@ -9872,9 +9873,9 @@
       "description": "",
       "description_tooltip": null,
       "layout": "IPY_MODEL_e5a82df528bb4e408797a3b6c2758f4a",
-      "placeholder": "​",
+      "placeholder": "\u200b",
       "style": "IPY_MODEL_f113ebd8c1c34806bea4dd7ed3035173",
-      "value": " 9985/9985 [00:00&lt;00:00, 44264.88 examples/s]"
+      "value": "\u20079985/9985\u2007[00:00&lt;00:00,\u200744264.88\u2007examples/s]"
      }
     },
     "fea1b70fb46745feb5111b3929175b5d": {
@@ -9930,9 +9931,9 @@
       "description": "",
       "description_tooltip": null,
       "layout": "IPY_MODEL_ab93eabd7cea4b94b4b7a387f101e8a1",
-      "placeholder": "​",
+      "placeholder": "\u200b",
       "style": "IPY_MODEL_704f2f5a9b1c49d5a75a0025a5dda11b",
-      "value": " 3.96G/3.96G [00:12&lt;00:00, 656MB/s]"
+      "value": "\u20073.96G/3.96G\u2007[00:12&lt;00:00,\u2007656MB/s]"
      }
     }
    }

examples/devstral/README.md

@@ -16,8 +16,13 @@ Thanks to the team at MistralAI for giving us early access to prepare for this r
 
 ```bash
 # Ensure you have Pytorch installed (Pytorch 2.6.0 min)
-pip3 install packaging==26.0 setuptools==75.8.0 wheel ninja
-pip3 install --no-build-isolation 'axolotl[flash-attn]>=0.12.0'
+# Option A: manage dependencies in your project
+uv add 'axolotl>=0.12.0'
+uv pip install flash-attn --no-build-isolation
+
+# Option B: quick install
+uv pip install 'axolotl>=0.12.0'
+uv pip install flash-attn --no-build-isolation
 ```
 
 2. Install [Cut Cross Entropy](https://docs.axolotl.ai/docs/custom_integrations.html#cut-cross-entropy) to reduce training VRAM usage

examples/devstral/devstral-small-qlora.yml

@@ -52,7 +52,6 @@ gradient_checkpointing: true
 resume_from_checkpoint:
 logging_steps: 1
 flash_attention: true
-scaling_softmax: true
 
 loss_watchdog_threshold: 5.0
 loss_watchdog_patience: 3

examples/eaft/eaft-example.yml

@@ -1,77 +0,0 @@
-base_model: google/gemma-3-1b-it
-
-model_type: Gemma3ForCausalLM
-cls_model_config: Gemma3TextConfig
-
-# gemma3 doesn't seem to play nice with ddp
-ddp_find_unused_parameters: true
-
-chat_template: gemma3
-eot_tokens:
-  - <end_of_turn>
-
-load_in_8bit: false
-load_in_4bit: false
-strict: false
-
-datasets:
-  - path: cgato/SlimOrcaDedupCleaned
-    type: chat_template
-    field_messages: conversations
-    message_property_mappings:
-      role: from
-      content: value
-
-dataset_prepared_path:
-val_set_size: 0
-output_dir: ./outputs/eaft-gemma-3-1b
-
-use_eaft: true
-eaft_alpha: 1.0
-eaft_k: 20
-
-sequence_len: 1024
-sample_packing: false
-
-adapter:
-lora_model_dir:
-
-wandb_project:
-wandb_entity:
-wandb_watch:
-wandb_name:
-wandb_log_model:
-
-gradient_accumulation_steps: 4
-micro_batch_size: 1
-eval_batch_size: 1
-max_steps: 1000
-evaluation_strategy: "no"
-optimizer: adamw_torch_fused
-lr_scheduler: cosine
-learning_rate: 5e-5
-
-train_on_inputs: false
-group_by_length: false
-bf16: auto
-fp16:
-tf32: true
-
-gradient_checkpointing: true
-gradient_checkpointing_kwargs:
-  use_reentrant: false
-
-early_stopping_patience:
-resume_from_checkpoint:
-local_rank:
-logging_steps: 1
-xformers_attention:
-flash_attention: true
-
-warmup_ratio: 0.1
-weight_decay: 0.0
-debug:
-deepspeed:
-fsdp:
-fsdp_config:
-special_tokens:

examples/gemma3/gemma-3-1b-qlora.yml

@@ -1,5 +1,7 @@
 base_model: google/gemma-3-1b-it
-
+# optionally might have model_type or tokenizer_type
+model_type: AutoModelForCausalLM
+tokenizer_type: AutoTokenizer
 # Automatically upload checkpoint and final model to HF
 # hub_model_id: username/custom_model_name
 
@@ -24,15 +26,10 @@ datasets:
 val_set_size: 0.0
 output_dir: ./outputs/out
 
-# Freeze vision tower
-unfrozen_parameters:
-  - ^model\.language_model\..*
-  - ^lm_head\..*
-
 adapter: qlora
 lora_r: 32
 lora_alpha: 16
-lora_dropout: 0
+lora_dropout: 0.05
 lora_target_linear: true
 
 sequence_len: 2048

examples/gemma3/gemma-3-270m-qlora.yml

@@ -1,5 +1,7 @@
 base_model: google/gemma-3-270m-it
-
+# optionally might have model_type or tokenizer_type
+model_type: AutoModelForCausalLM
+tokenizer_type: AutoTokenizer
 # Automatically upload checkpoint and final model to HF
 # hub_model_id: username/custom_model_name
 
@@ -24,15 +26,10 @@ datasets:
 val_set_size: 0.0
 output_dir: ./outputs/out
 
-# Freeze vision tower
-unfrozen_parameters:
-  - ^model\.language_model\..*
-  - ^lm_head\..*
-
 adapter: qlora
 lora_r: 32
 lora_alpha: 16
-lora_dropout: 0
+lora_dropout: 0.05
 lora_target_linear: true
 
 sequence_len: 2048

examples/gemma3/gemma-3-4b-qlora.yml

@@ -20,11 +20,6 @@ dataset_prepared_path: last_run_prepared
 val_set_size: 0.01
 output_dir: ./outputs/out
 
-# Freeze vision tower
-unfrozen_parameters:
-  - ^model\.language_model\..*
-  - ^lm_head\..*
-
 adapter: qlora
 lora_model_dir:
 
@@ -34,8 +29,8 @@ sample_packing: true
 
 lora_r: 32
 lora_alpha: 16
-lora_dropout: 0
-lora_target_linear: true
+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:

examples/gemma3/gemma-3-4b-vision-qlora.yml

@@ -31,7 +31,7 @@ pad_to_sequence_len: false
 
 lora_r: 32
 lora_alpha: 16
-lora_dropout: 0
+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:

examples/gemma3n/README.md

@@ -10,17 +10,22 @@ Gemma-3n is a family of multimodal models from Google found on [HuggingFace](htt
 
 ```bash
 # Ensure you have Pytorch installed (Pytorch 2.6.0 min)
-pip3 install packaging==26.0 setuptools==75.8.0 wheel ninja
-pip3 install --no-build-isolation 'axolotl[flash-attn]>=0.12.0'
+# Option A: manage dependencies in your project
+uv add 'axolotl>=0.12.0'
+uv pip install flash-attn --no-build-isolation
+
+# Option B: quick install
+uv pip install 'axolotl>=0.12.0'
+uv pip install flash-attn --no-build-isolation
 ```
 
 2. In addition to Axolotl's requirements, Gemma-3n requires:
 
 ```bash
-pip3 install timm==1.0.17
+uv pip install timm==1.0.17
 
 # for loading audio data
-pip3 install librosa==0.11.0
+uv pip install librosa==0.11.0
 ```
 
 3. Download sample dataset files

examples/glm45/README.md

@@ -1,72 +0,0 @@
-# Finetune Z.ai's GLM-4.5-Air with Axolotl
-
-[GLM-4.5-Air](https://huggingface.co/zai-org/GLM-4.5-Air) is a MoE model by Z.ai.
-
-This guide shows how to fine-tune it with Axolotl.
-
-## Getting started
-
-1. Install Axolotl following the [installation guide](https://docs.axolotl.ai/docs/installation.html).
-
-2. Install [Cut Cross Entropy](https://docs.axolotl.ai/docs/custom_integrations.html#cut-cross-entropy) to reduce training VRAM usage.
-
-3. Run the finetuning example:
-
-```bash
-# QLoRA (1x80GB @ ~63.4GiB/GPU)
-axolotl train examples/glm45/glm-45-air-qlora.yaml
-```
-
-### Dataset
-
-In addition to the standard OpenAI Messages format, GLM-4.5 supports an extra parameter for thinking in the assistant section.
-
-```json
-{
-    "role": "assistant",
-    "reasoning_content": "...",  // or have </think>...</think> in `content`
-    "content": "..."
-}
-```
-
-Make sure you set the below extra attributes if needed:
-
-```yaml
-datasets:
-  - path: ...
-    type: chat_template
-    message_property_mappings:
-      role: role
-      content: content
-
-    #   tool_calls: tool_calls  # uncomment if using tools
-    #   reasoning_content: reasoning_content  # uncomment if have reasoning
-
-# Uncomment if training on tool role (you would rarely if ever need this)
-# eot_tokens:
-#   - <|observation|>
-```
-
-### Tips
-
-- The role name for tools in this template is `tool`.
-- You will see this Axolotl WARNING — this is expected as the template does not use EOS:
-  ```
-  EOS token '<|endoftext|>' not found in chat_template. Please check if your template/EOS token is correct.
-  ```
-- You can run a full finetuning by removing `adapter: qlora`, `load_in_4bit: true`, and `quantize_moe_experts: true` from the config.
-- **LoRA kernels**: Incompatible with this model. Must be explicitly disabled (`lora_*_kernel: false`).
-- Read more on how to load your own dataset at [docs](https://docs.axolotl.ai/docs/dataset_loading.html).
-
-## Optimization Guides
-
-Please check the [Optimizations doc](https://docs.axolotl.ai/docs/optimizations.html).
-
-## Related Resources
-
-- [GLM-4.5-Air on HuggingFace](https://huggingface.co/zai-org/GLM-4.5-Air)
-- [GLM-4.5 Blog](https://z.ai/blog/glm-4.5)
-- [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)

examples/glm45/glm-45-air-qlora.yaml

@@ -1,64 +0,0 @@
-base_model: zai-org/GLM-4.5-Air
-
-# 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
-
-quantize_moe_experts: true # important
-
-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: 16
-lora_alpha: 8
-lora_dropout: 0
-lora_target_modules:
-  - q_proj
-  - v_proj
-  - k_proj
-  - o_proj
-
-# lora_target_parameters:
-#   - mlp.experts.gate_up_proj
-#   - mlp.experts.down_proj
-
-lora_mlp_kernel: false
-lora_qkv_kernel: false
-lora_o_kernel: false
-
-gradient_accumulation_steps: 2
-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

examples/glm46v/README.md

@@ -1,44 +0,0 @@
-# Finetune GLM-4.6V with Axolotl
-
-GLM-4.6V is a family of vision-language models from ZhipuAI found on [HuggingFace](https://huggingface.co/zai-org/GLM-4.6V). This guide shows how to fine-tune it with Axolotl for vision-language tasks.
-
-
-
-## Getting started
-
-1. Install Axolotl from source following the [installation guide](https://docs.axolotl.ai/docs/installation.html#sec-edge-build).
-
-2. Install [Cut Cross Entropy](https://docs.axolotl.ai/docs/custom_integrations.html#cut-cross-entropy) to reduce training VRAM usage.
-
-
-3. Run the fine-tuning:
-
-    glm-4-6v-flash(9B)
-    ```bash
-    axolotl train examples/glm46v/glm-4-6v-flash-qlora.yaml
-    ```
-
-Let us know how it goes. Happy finetuning! 🚀
-
-## Tips
-
-- Vision datasets should follow the format described in the [multimodal docs](https://docs.axolotl.ai/docs/multimodal.html#dataset-format)
-- 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 in the [dataset loading docs](https://docs.axolotl.ai/docs/dataset_loading.html).
-
-## Supported Models
-
-- **GLM-4.6V**: Full vision-language model (`zai-org/GLM-4.6V`)
-- **GLM-4.6V-Flash**: Faster variant (`zai-org/GLM-4.6V-Flash`)
-
-## Optimization Guides
-
-Please check the [Optimizations doc](https://docs.axolotl.ai/docs/optimizations.html).
-
-## Related Resources
-
-- [ZhipuAI GLM-4.6V](https://huggingface.co/zai-org/GLM-4.6V)
-- [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)

examples/glm46v/glm-4-6v-flash-ddp.yaml

@@ -1,53 +0,0 @@
-base_model: zai-org/GLM-4.6V-Flash
-trust_remote_code: true
-
-processor_type: AutoProcessor
-load_in_4bit: true
-
-# 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
-ddp_find_unused_parameters: true
-
-output_dir: ./outputs/glm-4-6v-flash-qlora
-datasets:
-  - path: HuggingFaceH4/llava-instruct-mix-vsft
-    type: chat_template
-    split: train[:1%]
-
-adapter: qlora
-lora_r: 16
-lora_alpha: 32
-lora_dropout: 0.05
-lora_target_modules:
-  - gate_proj
-  - down_proj
-  - up_proj
-  - q_proj
-  - v_proj
-  - k_proj
-  - o_proj
-
-sequence_len: 2048
-
-gradient_accumulation_steps: 4
-micro_batch_size: 1
-num_epochs: 1
-optimizer: adamw_8bit
-lr_scheduler: cosine
-learning_rate: 0.0002
-
-bf16: auto
-tf32: false
-
-gradient_checkpointing: true
-gradient_checkpointing_kwargs:
-  use_reentrant: false
-logging_steps: 1
-sdp_attention: true
-
-warmup_ratio: 0.1
-evals_per_epoch: 0
-saves_per_epoch: 1
-weight_decay: 0.0

examples/glm46v/glm-4-6v-flash-qlora.yaml

@@ -1,50 +0,0 @@
-base_model: zai-org/GLM-4.6V-Flash
-trust_remote_code: true
-
-processor_type: AutoProcessor
-load_in_4bit: true
-
-# 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
-
-output_dir: ./outputs/glm-4-6v-flash-qlora
-datasets:
-  - path: HuggingFaceH4/llava-instruct-mix-vsft
-    type: chat_template
-    split: train[:1%]
-
-adapter: qlora
-lora_r: 16
-lora_alpha: 32
-lora_dropout: 0.05
-lora_target_modules:
-  - gate_proj
-  - down_proj
-  - up_proj
-  - q_proj
-  - v_proj
-  - k_proj
-  - o_proj
-
-sequence_len: 2048
-
-gradient_accumulation_steps: 4
-micro_batch_size: 1
-num_epochs: 1
-optimizer: adamw_8bit
-lr_scheduler: cosine
-learning_rate: 0.0002
-
-bf16: auto
-tf32: false
-
-gradient_checkpointing: true
-logging_steps: 1
-sdp_attention: true
-
-warmup_ratio: 0.1
-evals_per_epoch: 0
-saves_per_epoch: 1
-weight_decay: 0.0

examples/glm47-flash/README.md

@@ -1,65 +0,0 @@
-# Finetune Z.ai's GLM-4.7-Flash with Axolotl
-
-[GLM-4.7-Flash](https://huggingface.co/zai-org/GLM-4.7-Flash) is a 30B-A3B MoE model by Z.ai.
-
-This guide shows how to fine-tune it with Axolotl.
-
-## Getting started
-
-1. Install Axolotl following the [installation guide](https://docs.axolotl.ai/docs/installation.html).
-
-2. Install [Cut Cross Entropy](https://docs.axolotl.ai/docs/custom_integrations.html#cut-cross-entropy) to reduce training VRAM usage.
-
-3. Run the finetuning example:
-
-```bash
-# QLoRA
-# - no target experts (1x48GB @ ~24GiB/GPU)
-# - target experts (1x48GB @ ~34GiB/GPU)
-axolotl train examples/glm47-flash/qlora.yaml
-
-# QLoRA FSDP2 no target experts (2x48GB @ ~29GiB/GPU)
-axolotl train examples/glm47-flash/qlora_fsdp.yaml
-```
-
-```bash
-# LoRA
-# - no target experts (1x48GB @ ~35GiB/GPU)
-# - target experts (1x48GB @ OOM. Projected ~45-50GiB/GPU)
-axolotl train examples/glm47-flash/lora.yaml
-
-# LoRA FSDP2 no target experts (2x48GB @ ~43GiB/GPU)
-axolotl train examples/glm47-flash/lora_fsdp.yaml
-```
-
-### MoE Expert Quantization & Expert LoRA
-
-This model quantize expert weights on load. To learn about expert quantization, expert LoRA targeting, and related limitations, see the [MoE Expert Quantization](https://docs.axolotl.ai/docs/expert_quantization.html) docs.
-
-## Limitations
-
-- **lora_target_linear**: Incompatible for this model.
-- **LoRA kernels**: Incompatible with this model due to non-standard attention projections (DSA). Must be explicitly disabled (`lora_*_kernel: false`).
-
-
-### TIPS
-
-- For inference, the official Z.ai team recommends these default settings (most tasks):
-  - `temperature: 1.0`
-  - `top_p: 0.95`
-  - `max_new_tokens: 131072`
-- You can run a full finetuning by removing `adapter: qlora`, `load_in_4bit: true`, and `quantize_moe_experts: true` from the config. This is heavy, so we have not tested this.
-- Read more on how to load your own dataset at [docs](https://docs.axolotl.ai/docs/dataset_loading.html).
-
-## Optimization Guides
-
-Please check the [Optimizations doc](https://docs.axolotl.ai/docs/optimizations.html).
-
-## Related Resources
-
-- [GLM-4.7-Flash on HuggingFace](https://huggingface.co/zai-org/GLM-4.7-Flash)
-- [GLM-4.7 Blog](https://z.ai/blog/glm-4.7)
-- [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)

examples/glm47-flash/lora.yaml

@@ -1,65 +0,0 @@
-base_model: zai-org/GLM-4.7-Flash
-
-plugins:
-  - axolotl.integrations.cut_cross_entropy.CutCrossEntropyPlugin
-
-load_in_8bit: true
-quantize_moe_experts: 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/glm4.7-flash-lora-8bit-out
-
-adapter: lora
-lora_model_dir:
-
-sequence_len: 2048
-sample_packing: true
-
-lora_r: 32
-lora_alpha: 16
-lora_dropout: 0
-lora_target_modules:
-  - q_proj
-  - v_proj
-  - k_proj
-  - o_proj
-
-# Uncomment to also target MoE expert weights:
-# lora_target_parameters:
-#   - mlp.experts.gate_up_proj
-#   - mlp.experts.down_proj
-
-# LoRA kernels incompatible with DSA attention
-lora_mlp_kernel: false
-lora_qkv_kernel: false
-lora_o_kernel: false
-
-wandb_project:
-wandb_entity:
-wandb_watch:
-wandb_name:
-wandb_log_model:
-
-gradient_accumulation_steps: 4
-micro_batch_size: 2
-num_epochs: 1
-optimizer: adamw_torch_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

examples/glm47-flash/lora_fsdp.yaml

@@ -1,75 +0,0 @@
-base_model: zai-org/GLM-4.7-Flash
-
-plugins:
-  - axolotl.integrations.cut_cross_entropy.CutCrossEntropyPlugin
-
-load_in_8bit: true
-quantize_moe_experts: 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/glm4.7-flash-lora-8bit-fsdp-out
-
-adapter: lora
-lora_model_dir:
-
-sequence_len: 2048
-sample_packing: true
-
-lora_r: 32
-lora_alpha: 16
-lora_dropout: 0
-lora_target_modules:
-  - q_proj
-  - v_proj
-  - k_proj
-  - o_proj
-
-# Uncomment to also target MoE expert weights:
-# lora_target_parameters:
-#   - mlp.experts.gate_up_proj
-#   - mlp.experts.down_proj
-
-# LoRA kernels incompatible with DSA attention
-lora_mlp_kernel: false
-lora_qkv_kernel: false
-lora_o_kernel: false
-
-wandb_project:
-wandb_entity:
-wandb_watch:
-wandb_name:
-wandb_log_model:
-
-gradient_accumulation_steps: 4
-micro_batch_size: 2
-num_epochs: 1
-optimizer: adamw_torch_8bit
-lr_scheduler: cosine
-learning_rate: 0.0002
-
-bf16: auto
-tf32: false
-
-resume_from_checkpoint:
-logging_steps: 1
-flash_attention: true
-
-warmup_ratio: 0.1
-evals_per_epoch: 1
-saves_per_epoch: 1
-
-fsdp_config:
-  fsdp_version: 2
-  offload_params: false
-  cpu_ram_efficient_loading: false
-  auto_wrap_policy: TRANSFORMER_BASED_WRAP
-  transformer_layer_cls_to_wrap: Glm4MoeLiteDecoderLayer
-  state_dict_type: FULL_STATE_DICT
-  sharding_strategy: FULL_SHARD
-  reshard_after_forward: true
-  activation_checkpointing: true

examples/glm47-flash/qlora.yaml

@@ -1,65 +0,0 @@
-base_model: zai-org/GLM-4.7-Flash
-
-plugins:
-  - axolotl.integrations.cut_cross_entropy.CutCrossEntropyPlugin
-
-load_in_4bit: true
-quantize_moe_experts: 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/glm4.7-flash-qlora-out
-
-adapter: qlora
-lora_model_dir:
-
-sequence_len: 2048
-sample_packing: true
-
-lora_r: 32
-lora_alpha: 16
-lora_dropout: 0
-lora_target_modules:
-  - q_proj
-  - v_proj
-  - k_proj
-  - o_proj
-
-# Uncomment to also target MoE expert weights:
-# lora_target_parameters:
-#   - mlp.experts.gate_up_proj
-#   - mlp.experts.down_proj
-
-# LoRA kernels incompatible with DSA attention
-lora_mlp_kernel: false
-lora_qkv_kernel: false
-lora_o_kernel: false
-
-wandb_project:
-wandb_entity:
-wandb_watch:
-wandb_name:
-wandb_log_model:
-
-gradient_accumulation_steps: 4
-micro_batch_size: 2
-num_epochs: 1
-optimizer: adamw_torch_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

examples/glm47-flash/qlora_fsdp.yaml

@@ -1,75 +0,0 @@
-base_model: zai-org/GLM-4.7-Flash
-
-plugins:
-  - axolotl.integrations.cut_cross_entropy.CutCrossEntropyPlugin
-
-load_in_4bit: true
-quantize_moe_experts: 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/glm4.7-flash-qlora-fsdp-out
-
-adapter: qlora
-lora_model_dir:
-
-sequence_len: 2048
-sample_packing: true
-
-lora_r: 32
-lora_alpha: 16
-lora_dropout: 0
-lora_target_modules:
-  - q_proj
-  - v_proj
-  - k_proj
-  - o_proj
-
-# Uncomment to also target MoE expert weights:
-# lora_target_parameters:
-#   - mlp.experts.gate_up_proj
-#   - mlp.experts.down_proj
-
-# LoRA kernels incompatible with DSA attention
-lora_mlp_kernel: false
-lora_qkv_kernel: false
-lora_o_kernel: false
-
-wandb_project:
-wandb_entity:
-wandb_watch:
-wandb_name:
-wandb_log_model:
-
-gradient_accumulation_steps: 4
-micro_batch_size: 2
-num_epochs: 1
-optimizer: adamw_torch_8bit
-lr_scheduler: cosine
-learning_rate: 0.0002
-
-bf16: auto
-tf32: false
-
-resume_from_checkpoint:
-logging_steps: 1
-flash_attention: true
-
-warmup_ratio: 0.1
-evals_per_epoch: 1
-saves_per_epoch: 1
-
-fsdp_config:
-  fsdp_version: 2
-  offload_params: false
-  cpu_ram_efficient_loading: false
-  auto_wrap_policy: TRANSFORMER_BASED_WRAP
-  transformer_layer_cls_to_wrap: Glm4MoeLiteDecoderLayer
-  state_dict_type: FULL_STATE_DICT
-  sharding_strategy: FULL_SHARD
-  reshard_after_forward: true
-  activation_checkpointing: true

examples/gpt-oss/README.md

@@ -2,8 +2,6 @@
 
 [GPT-OSS](https://huggingface.co/collections/openai/gpt-oss-68911959590a1634ba11c7a4) are a family of open-weight MoE models trained by OpenAI, released in August 2025. There are two variants: 20B and 120B.
 
-In October 2025, OpenAI released safeguard models built upon GPT-OSS called [GPT-OSS-Safeguard](https://huggingface.co/collections/openai/gpt-oss-safeguard). They use the same architecture, so the same examples below can be re-used.
-
 This guide shows how to fine-tune it with Axolotl with multi-turn conversations and proper masking.
 
 ## Getting started
@@ -14,8 +12,13 @@ This guide shows how to fine-tune it with Axolotl with multi-turn conversations
 
 ```bash
 # Ensure you have Pytorch installed (Pytorch 2.6.0 min)
-pip3 install packaging==26.0 setuptools==75.8.0 wheel ninja
-pip3 install --no-build-isolation 'axolotl[flash-attn]>=0.12.0'
+# Option A: manage dependencies in your project
+uv add 'axolotl>=0.12.0'
+uv pip install flash-attn --no-build-isolation
+
+# Option B: quick install
+uv pip install 'axolotl>=0.12.0'
+uv pip install flash-attn --no-build-isolation
 ```
 
 2. Choose one of the following configs below for training the 20B model. (for 120B, see [below](#training-120b))
@@ -66,16 +69,6 @@ axolotl merge-sharded-fsdp-weights examples/gpt-oss/gpt-oss-120b-fft-fsdp2-offlo
 mv ./outputs/gpt-oss-out/merged/* ./outputs/gpt-oss-out/
 ```
 
-### How to set reasoning_effort in template?
-
-The harmony template has a feature to set the `reasoning_effort` during prompt building. The default is `medium`. If you would like to adjust this, you can add the following to your config:
-
-```yaml
-chat_template_kwargs:
-  reasoning_effort: "high"  # low | medium | high
-```
-
-Currently, this applies globally. There is no method to apply per sample yet. If you are interested in adding this, please feel free to create an Issue to discuss.
 
 ### Inferencing your fine-tuned model
 
@@ -87,7 +80,7 @@ for more information about using a special vllm-openai docker image for inferenc
 Optionally, vLLM can be installed from nightly:
 
 ```bash
-pip install --no-build-isolation --pre -U vllm --extra-index-url https://wheels.vllm.ai/nightly
+uv pip install --no-build-isolation --pre -U vllm --extra-index-url https://wheels.vllm.ai/nightly
 ```
 and the vLLM server can be started with the following command (modify `--tensor-parallel-size 8` to match your environment):
 ```bash

examples/gpt-oss/gpt-oss-120b-fft-fsdp2-offload.yaml

@@ -32,10 +32,6 @@ wandb_watch:
 wandb_name:
 wandb_log_model:
 
-trackio_project_name:
-trackio_run_name:
-trackio_space_id:
-
 gradient_accumulation_steps: 2
 micro_batch_size: 1
 num_epochs: 1

examples/gpt-oss/gpt-oss-20b-fft-deepspeed-zero3.yaml

@@ -28,10 +28,6 @@ wandb_watch:
 wandb_name:
 wandb_log_model:
 
-trackio_project_name:
-trackio_run_name:
-trackio_space_id:
-
 gradient_accumulation_steps: 2
 micro_batch_size: 1
 num_epochs: 1

examples/gpt-oss/gpt-oss-20b-fft-fsdp2-offload.yaml

@@ -29,10 +29,6 @@ wandb_watch:
 wandb_name:
 wandb_log_model:
 
-trackio_project_name:
-trackio_run_name:
-trackio_space_id:
-
 gradient_accumulation_steps: 2
 micro_batch_size: 1
 num_epochs: 1

examples/gpt-oss/gpt-oss-20b-fft-fsdp2.yaml

@@ -28,10 +28,6 @@ wandb_watch:
 wandb_name:
 wandb_log_model:
 
-trackio_project_name:
-trackio_run_name:
-trackio_space_id:
-
 gradient_accumulation_steps: 2
 micro_batch_size: 1
 num_epochs: 1

examples/gpt-oss/gpt-oss-20b-sft-lora-singlegpu.yaml

@@ -41,10 +41,6 @@ wandb_watch:
 wandb_name:
 wandb_log_model:
 
-trackio_project_name:
-trackio_run_name:
-trackio_space_id:
-
 gradient_accumulation_steps: 8
 micro_batch_size: 1
 num_epochs: 1

examples/gpt-oss/gpt-oss-safeguard-20b-sft-lora-singlegpu.yaml

@@ -1,71 +0,0 @@
-base_model: openai/gpt-oss-safeguard-20b
-use_kernels: true
-model_quantization_config: Mxfp4Config
-model_quantization_config_kwargs:
-  dequantize: true
-
-plugins:
-  - axolotl.integrations.cut_cross_entropy.CutCrossEntropyPlugin
-
-experimental_skip_move_to_device: true  # prevent OOM by not putting model to GPU before sharding
-
-datasets:
-  - path: HuggingFaceH4/Multilingual-Thinking
-    type: chat_template
-    field_thinking: thinking
-    template_thinking_key: thinking
-
-dataset_prepared_path: last_run_prepared
-val_set_size: 0
-output_dir: ./outputs/gpt-oss-safeguard-out/
-
-sequence_len: 4096
-sample_packing: true
-
-adapter: lora
-lora_r: 8
-lora_alpha: 16
-lora_dropout: 0.0  # dropout not supported when using LoRA over expert parameters
-lora_target_linear: true
-
-# TODO: not supported for now, see peft#2710
-#lora_target_parameters:  # target the experts in the last two layers
-#  - "22._checkpoint_wrapped_module.mlp.experts.gate_up_proj"
-#  - "22._checkpoint_wrapped_module.mlp.experts.down_proj"
-#  - "23._checkpoint_wrapped_module.mlp.experts.gate_up_proj"
-#  - "23._checkpoint_wrapped_module.mlp.experts.down_proj"
-
-wandb_project:
-wandb_entity:
-wandb_watch:
-wandb_name:
-wandb_log_model:
-
-trackio_project_name:
-trackio_run_name:
-trackio_space_id:
-
-gradient_accumulation_steps: 8
-micro_batch_size: 1
-num_epochs: 1
-
-optimizer: adamw_torch_8bit
-lr_scheduler: constant_with_warmup
-learning_rate: 2e-4
-
-bf16: true
-tf32: true
-
-flash_attention: true
-attn_implementation: kernels-community/vllm-flash-attn3  # this is not needed if using flash_attn >= 2.8.3
-
-gradient_checkpointing: true
-activation_offloading: true
-
-logging_steps: 1
-saves_per_epoch: 1
-warmup_ratio: 0.1
-
-special_tokens:
-eot_tokens:
-  - "<|end|>"

examples/granite4/README.md

@@ -1,65 +0,0 @@
-# Finetune IBM's Granite 4.0 with Axolotl
-
-[Granite 4.0](https://huggingface.co/collections/ibm-granite/granite-40-language-models) are a family of open source models trained by IBM Research.
-
-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 Granite4 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 2.7.1 min)
-git clone https://github.com/axolotl-ai-cloud/axolotl.git
-cd axolotl
-
-pip3 install packaging==26.0 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/granite4/granite-4.0-tiny-fft.yaml
-```
-
-This config uses about 40.8GiB VRAM.
-
-Let us know how it goes. Happy finetuning! 🚀
-
-### TIPS
-
-- 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).
-
-### Limitation
-
-Adapter finetuning does not work at the moment. It would error with
-
-```bash
-RuntimeError: mat1 and mat2 shapes cannot be multiplied (4096x3072 and 1x1179648)
-```
-
-In addition, if adapter training works, `lora_target_linear: true` will not work due to:
-```bash
-ValueError: Target module GraniteMoeHybridParallelExperts() is not supported.
-```
-
-## 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
-
-- [Granite Docs](https://www.ibm.com/granite/docs/models/granite)
-- [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)

examples/granite4/granite-4.0-tiny-fft.yaml

@@ -1,45 +0,0 @@
-base_model: ibm-granite/granite-4.0-tiny-preview
-
-# Automatically upload checkpoint and final model to HF
-# hub_model_id: username/custom_model_name
-
-plugins:
-  - axolotl.integrations.cut_cross_entropy.CutCrossEntropyPlugin
-
-datasets:
-  - path: fozziethebeat/alpaca_messages_2k_test
-    type: chat_template
-
-dataset_prepared_path: last_run_prepared
-val_set_size: 0.1
-output_dir: ./outputs/model-out
-
-sequence_len: 2048
-sample_packing: true
-
-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