fsdp2 w evals fixed upstream

.bandit

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

.coderabbit.yaml

@@ -12,6 +12,5 @@ reviews:
   auto_review:
     enabled: true
     drafts: false
-    auto_incremental_review: false
 chat:
   auto_reply: true

.flake8

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

.github/CONTRIBUTING.md

@@ -57,13 +57,6 @@ We welcome ideas for improvements and new features. To suggest an enhancement, o
 5. Push your branch to your fork on GitHub.
 6. Open a new pull request against the `main` branch of the axolotl repository. Include a clear and concise description of your changes, referencing any related issues.
 
-#### Skipping CI Checks
-
-You can skip certain CI checks by including specific keywords in your commit messages:
-
-- `[skip ci]` or `skip ci` - Skips all CI checks for that commit
-- `[skip-e2e]` or `skip-e2e` - Skips only end-to-end tests while running other CI checks. You may also include this in the title of your PR to disable end-to-end tests for the entire PR.
-
 ## Style Guidelines
 
 ### Code Style

.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/workflows/base.yml

@@ -25,6 +25,41 @@ jobs:
       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: 2.7.0
+            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: ""
@@ -32,27 +67,6 @@ 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"
-          - cuda: "128"
-            cuda_version: 12.8.1
-            cudnn_version: ""
-            python_version: "3.11"
-            pytorch: 2.9.0
-            torch_cuda_arch_list: "7.0 7.5 8.0 8.6 8.7 8.9 9.0+PTX"
-            dockerfile: "Dockerfile-base"
-          - cuda: "128"
-            cuda_version: 12.8.1
-            cudnn_version: ""
-            python_version: "3.11"
-            pytorch: 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"
-          - 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"
 #          - cuda: "128"
 #            cuda_version: 12.8.1
 #            cudnn_version: ""
@@ -76,6 +90,7 @@ jobs:
         uses: docker/metadata-action@v5
         with:
           images: |
+            winglian/axolotl-base
             axolotlai/axolotl-base
       - name: Login to Docker Hub
         uses: docker/login-action@v2
@@ -107,6 +122,27 @@ jobs:
       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: ""
@@ -114,27 +150,6 @@ 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"
-          - 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"
-          - 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"
-          - 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"
     steps:
       - name: Checkout
         uses: actions/checkout@v4

.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

.github/workflows/main.yml

@@ -15,27 +15,27 @@ 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: 126
+            cuda_version: 12.6.3
+            python_version: "3.11"
+            pytorch: 2.7.0
+            axolotl_extras:
+          - cuda: 126
+            cuda_version: 12.6.3
+            python_version: "3.11"
+            pytorch: 2.7.1
+            axolotl_extras: vllm
             is_latest: true
           - cuda: 128
             cuda_version: 12.8.1
             python_version: "3.11"
-            pytorch: 2.9.0
+            pytorch: 2.7.1
             axolotl_extras:
-          - cuda: 128
-            cuda_version: 12.8.1
-            python_version: "3.11"
-            pytorch: 2.9.1
-            axolotl_extras:
-#          - cuda: 130
-#            cuda_version: 13.0.0
-#            python_version: "3.11"
-#            pytorch: 2.9.1
-#            axolotl_extras:
     runs-on: axolotl-gpu-runner
     steps:
       - name: Checkout
@@ -45,6 +45,7 @@ jobs:
         uses: docker/metadata-action@v5
         with:
           images: |
+            winglian/axolotl
             axolotlai/axolotl
           tags: |
             type=ref,event=branch
@@ -82,27 +83,33 @@ 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: 126
+            cuda_version: 12.6.3
+            python_version: "3.11"
+            pytorch: 2.7.0
+            axolotl_extras:
+          - cuda: 126
+            cuda_version: 12.6.3
+            python_version: "3.11"
+            pytorch: 2.7.1
+            axolotl_extras:
+            is_latest:
+          - cuda: 126
+            cuda_version: 12.6.3
+            python_version: "3.11"
+            pytorch: 2.7.1
+            axolotl_extras: vllm
             is_latest: true
           - cuda: 128
             cuda_version: 12.8.1
             python_version: "3.11"
-            pytorch: 2.9.0
+            pytorch: 2.7.1
             axolotl_extras:
-          - cuda: 128
-            cuda_version: 12.8.1
-            python_version: "3.11"
-            pytorch: 2.9.1
-            axolotl_extras:
-#          - cuda: 130
-#            cuda_version: 13.0.0
-#            python_version: "3.11"
-#            pytorch: 2.9.1
-#            axolotl_extras:
     runs-on: axolotl-gpu-runner
     steps:
       - name: Checkout
@@ -112,6 +119,7 @@ jobs:
         uses: docker/metadata-action@v5
         with:
           images: |
+            winglian/axolotl-cloud
             axolotlai/axolotl-cloud
           tags: |
             type=ref,event=branch
@@ -144,18 +152,23 @@ 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: 126
+            cuda_version: 12.6.3
+            python_version: "3.11"
+            pytorch: 2.7.1
             axolotl_extras:
             is_latest:
-          - cuda: 128
-            cuda_version: 12.8.1
+          - cuda: 126
+            cuda_version: 12.6.3
             python_version: "3.11"
-            pytorch: 2.9.1
-            axolotl_extras:
-            is_latest:
+            pytorch: 2.7.1
+            axolotl_extras: vllm
+            is_latest: true
     runs-on: axolotl-gpu-runner
     steps:
       - name: Checkout
@@ -165,6 +178,7 @@ jobs:
         uses: docker/metadata-action@v5
         with:
           images: |
+            winglian/axolotl-cloud-term
             axolotlai/axolotl-cloud-term
           tags: |
             type=ref,event=branch

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

@@ -19,9 +19,6 @@ concurrency:
   group: ${{ github.workflow }}-${{ github.ref }}
   cancel-in-progress: ${{ github.ref != 'refs/heads/main' }}
 
-env:
-  MODAL_IMAGE_BUILDER_VERSION: "2025.06"
-
 jobs:
   test-axolotl-multigpu:
     if: ${{ ! contains(github.event.commits[0].message, '[skip e2e]') && github.repository_owner == 'axolotl-ai-cloud' && (github.event_name != 'pull_request' || !github.event.pull_request.draft) }}
@@ -29,18 +26,25 @@ jobs:
       fail-fast: false
       matrix:
         include:
-          - cuda: 128
-            cuda_version: 12.8.1
+          - cuda: 126
+            cuda_version: 12.6.3
             python_version: "3.11"
-            pytorch: 2.8.0
-            axolotl_extras: fbgemm-gpu
+            pytorch: 2.6.0
+            axolotl_extras:
             num_gpus: 2
             nightly_build: "true"
-          - cuda: 128
-            cuda_version: 12.8.1
+          - cuda: 126
+            cuda_version: 12.6.3
             python_version: "3.11"
-            pytorch: 2.9.1
-            axolotl_extras: fbgemm-gpu
+            pytorch: 2.7.0
+            axolotl_extras:
+            num_gpus: 2
+            nightly_build: "true"
+          - cuda: 126
+            cuda_version: 12.6.3
+            python_version: "3.11"
+            pytorch: 2.7.1
+            axolotl_extras: vllm
             num_gpus: 2
             nightly_build: "true"
     runs-on: [self-hosted, modal]
@@ -55,7 +59,7 @@ jobs:
       - name: Install Modal
         run: |
           python -m pip install --upgrade pip
-          pip install modal==1.3.0.post1 jinja2
+          pip install modal==1.0.2 jinja2
       - name: Update env vars
         run: |
           echo "BASE_TAG=main-base-py${{ matrix.python_version }}-cu${{ matrix.cuda }}-${{ matrix.pytorch }}" >> $GITHUB_ENV
@@ -68,4 +72,4 @@ jobs:
           echo "CODECOV_TOKEN=${{ secrets.CODECOV_TOKEN }}" >> $GITHUB_ENV
       - name: Run tests job on Modal
         run: |
-          modal run -m cicd.multigpu
+          modal run cicd.multigpu

.github/workflows/nightlies.yml

@@ -12,15 +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:
@@ -31,6 +31,7 @@ jobs:
         uses: docker/metadata-action@v5
         with:
           images: |
+            winglian/axolotl
             axolotlai/axolotl
           tags: |
             type=raw,value={{ branch }}-{{ date 'YYYYMMDD' }}
@@ -64,15 +65,15 @@ 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:
@@ -83,6 +84,7 @@ jobs:
         uses: docker/metadata-action@v5
         with:
           images: |
+            winglian/axolotl-cloud
             axolotlai/axolotl-cloud
           tags: |
             type=raw,value={{ branch }}-{{ date 'YYYYMMDD' }}

.github/workflows/precommit-autoupdate.yml

@@ -2,7 +2,7 @@ name: Pre-commit auto-update
 
 on:
   schedule:
-    - cron: '0 0 1 * *'  # Run monthly
+    - cron: '0 0 * * 0'  # Run weekly
   workflow_dispatch:  # Manual kickoff
 
 jobs:

.github/workflows/preview-docs.yml

@@ -11,7 +11,6 @@ on:
       - '_quarto.yml'
       - docs/scripts/generate_config_docs.py
       - src/axolotl/utils/schemas/**.py
-      - .github/workflows/preview-docs.yml
 
 permissions:
   checks: write
@@ -28,10 +27,6 @@ jobs:
     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:

.github/workflows/tests-nightly.yml

@@ -26,7 +26,7 @@ jobs:
       max-parallel: 2
       matrix:
         python_version: ["3.11"]
-        pytorch_version: ["2.8.0", "2.9.0", "2.9.1"]
+        pytorch_version: ["2.6.0", "2.7.0"]
     timeout-minutes: 20
 
     steps:
@@ -99,17 +99,17 @@ 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
             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: 2.9.1
+            pytorch: 2.7.1
             num_gpus: 1
             axolotl_extras:
             nightly_build: "true"
@@ -123,7 +123,7 @@ jobs:
       - 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
       - name: Update env vars
         run: |
           echo "BASE_TAG=main-base-py${{ matrix.python_version }}-cu${{ matrix.cuda }}-${{ matrix.pytorch }}" >> $GITHUB_ENV
@@ -148,10 +148,10 @@ jobs:
       fail-fast: false
       matrix:
         include:
-          - cuda: 128
-            cuda_version: 12.8.1
+          - cuda: 126
+            cuda_version: 12.6.3
             python_version: "3.11"
-            pytorch: 2.9.1
+            pytorch: 2.7.1
             num_gpus: 2
             axolotl_extras:
             nightly_build: "true"
@@ -165,7 +165,7 @@ jobs:
       - name: Install Modal
         run: |
           python -m pip install --upgrade pip
-          pip install modal==1.3.0.post1 jinja2
+          pip install modal==1.0.2 jinja2
       - name: Update env vars
         run: |
           echo "BASE_TAG=main-base-py${{ matrix.python_version }}-cu${{ matrix.cuda }}-${{ matrix.pytorch }}" >> $GITHUB_ENV

.github/workflows/tests.yml

@@ -55,23 +55,18 @@ jobs:
       fail-fast: false
       matrix:
         python_version: ["3.11"]
-        pytorch_version: ["2.8.0", "2.9.0", "2.9.1"]
+        pytorch_version: ["2.6.0", "2.7.0", "2.7.1"]
     timeout-minutes: 20
 
     steps:
-      - name: cleanup node
-        run: |
-          sudo rm -rf /usr/share/dotnet /usr/local/lib/android /opt/ghc /opt/hostedtoolcache/CodeQL
-
       - name: Check out repository code
         uses: actions/checkout@v4
 
       - name: Restore Cache from S3
         id: hf-cache-restore-s3
         run: |
-          mkdir -p ~/.cache/huggingface/hub
-          curl -L https://d1dttdx32dkk5p.cloudfront.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
@@ -86,20 +81,16 @@ jobs:
 
       - name: Install PyTorch
         run: |
-          pip3 install --no-cache-dir torch==${{ matrix.pytorch_version }} torchvision
+          pip3 install torch==${{ matrix.pytorch_version }} torchvision
 
       - name: Install dependencies
         run: |
           pip3 show torch
-          pip3 install --no-cache-dir --no-build-isolation -U -e .
+          pip3 install --no-build-isolation -U -e .
           python scripts/unsloth_install.py | sh
           python scripts/cutcrossentropy_install.py | sh
           pip3 install -r requirements-dev.txt -r requirements-tests.txt
 
-      - name: cleanup pip cache
-        run: |
-          find "$(pip cache dir)/http-v2" -type f -mtime +14 -exec rm {} \;
-
       - name: Make sure PyTorch version wasn't clobbered
         run: |
           python -c "import torch; assert '${{ matrix.pytorch_version }}' in torch.__version__"
@@ -112,23 +103,13 @@ jobs:
         run: |
           huggingface-cli download --repo-type=dataset axolotl-ai-internal/axolotl-oss-dataset-fixtures
 
-      - name: Show HF cache
-        run: hf cache scan
-
       - 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 -n8 --dist loadfile --ignore=tests/e2e/ --ignore=tests/patched/ --ignore=tests/cli/ --ignore=tests/monkeypatch/ tests/ --cov=axolotl --cov-report=xml
           pytest -v --durations=10 tests/monkeypatch/ --cov=axolotl --cov-append --cov-report=xml
-          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 scan
-
       - name: Upload coverage to Codecov
         uses: codecov/codecov-action@v5
         with:
@@ -137,6 +118,10 @@ jobs:
           flags: unittests,pytorch-${{ matrix.pytorch_version }}
           fail_ci_if_error: false
 
+      - name: cleanup pip cache
+        run: |
+          find "$(pip cache dir)/http-v2" -type f -mtime +14 -exec rm {} \;
+
   pytest-sdist:
     name: PyTest from Source Dist
     runs-on: ubuntu-latest
@@ -145,23 +130,18 @@ jobs:
       fail-fast: false
       matrix:
         python_version: ["3.11"]
-        pytorch_version: ["2.8.0", "2.9.0", "2.9.1"]
+        pytorch_version: ["2.6.0", "2.7.0", "2.7.1"]
     timeout-minutes: 20
 
     steps:
-      - name: cleanup node
-        run: |
-          sudo rm -rf /usr/share/dotnet /usr/local/lib/android /opt/ghc /opt/hostedtoolcache/CodeQL
-
       - name: Check out repository code
         uses: actions/checkout@v4
 
       - name: Restore Cache from S3
         id: hf-cache-restore-s3
         run: |
-          mkdir -p ~/.cache/huggingface/hub
-          curl -L https://d1dttdx32dkk5p.cloudfront.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
@@ -172,25 +152,21 @@ jobs:
       - name: upgrade pip
         run: |
           pip3 install --upgrade pip
-          pip3 install --upgrade packaging==23.2 setuptools==75.8.0 setuptools_scm build wheel psutil
+          pip3 install --upgrade packaging==23.2 setuptools==75.8.0 setuptools_scm build wheel
 
       - name: Install PyTorch
         run: |
-          pip3 install --no-cache-dir torch==${{ matrix.pytorch_version }} torchvision
+          pip3 install torch==${{ 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
+          pip3 install --no-build-isolation dist/axolotl*.tar.gz
           python scripts/unsloth_install.py | sh
           python scripts/cutcrossentropy_install.py | sh
           pip3 install -r requirements-dev.txt -r requirements-tests.txt
 
-      - name: cleanup pip cache
-        run: |
-          find "$(pip cache dir)/http-v2" -type f -mtime +14 -exec rm {} \;
-
       - name: Make sure PyTorch version wasn't clobbered
         run: |
           python -c "import torch; assert '${{ matrix.pytorch_version }}' in torch.__version__"
@@ -200,64 +176,40 @@ jobs:
           axolotl --help
 
       - name: Show HF cache
-        run: hf cache scan
+        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 -n8 --dist loadfile --ignore=tests/e2e/ --ignore=tests/patched/ --ignore=tests/cli/ --ignore=tests/monkeypatch/ tests/ --cov=axolotl --cov-report=xml
           pytest -v --durations=10 tests/monkeypatch/ --cov=axolotl --cov-append --cov-report=xml
           pytest -v --durations=10 tests/cli/
 
-      - name: Show HF cache
-        run: hf cache scan
-
-  gate-skip-e2e:
-    needs: [pre-commit, pytest, pytest-sdist]
-    runs-on: ubuntu-latest
-    outputs:
-      skip: ${{ steps.compute.outputs.skip }}
-    steps:
-      - uses: actions/github-script@v7
-        id: compute
-        with:
-          script: |
-            const token = /\[skip-e2e\]/i;
-            let msg = '';
-            if (context.eventName === 'push') {
-              msg = context.payload.head_commit?.message || '';
-            } else if (context.eventName === 'pull_request') {
-              const { owner, repo } = context.repo;
-              const prNumber = context.payload.pull_request.number;
-              const commits = await github.paginate(
-                github.rest.pulls.listCommits,
-                { owner, repo, pull_number: prNumber, per_page: 100 }
-              );
-              msg = commits.at(-1)?.commit?.message || '';
-            }
-            const title = context.payload.pull_request?.title || '';
-            const body  = context.payload.pull_request?.body  || '';
-            const skip = token.test(msg) || token.test(title) || token.test(body);
-            core.setOutput('skip', String(skip));
+      - name: cleanup pip cache
+        run: |
+          find "$(pip cache dir)/http-v2" -type f -mtime +14 -exec rm {} \;
 
   docker-e2e-tests-1st:
     # Run this job first as a gate for running the remainder of the test matrix
-    if: >
-      github.repository_owner == 'axolotl-ai-cloud' &&
-      (github.event_name != 'pull_request' || !github.event.pull_request.draft) &&
-      needs.gate-skip-e2e.outputs.skip != 'true'
+    if: ${{ ! contains(github.event.commits[0].message, '[skip e2e]') && github.repository_owner == 'axolotl-ai-cloud' && !github.event.pull_request.draft }}
     # this job needs to be run on self-hosted GPU runners...
     runs-on: [self-hosted, modal]
     timeout-minutes: 120
-    needs: [pre-commit, pytest, pytest-sdist, gate-skip-e2e]
+    needs: [pre-commit, pytest, pytest-sdist]
 
     strategy:
       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: 2.7.1
+            num_gpus: 1
+            axolotl_extras:
+          - cuda: 126
+            cuda_version: 12.6.3
+            python_version: "3.11"
+            pytorch: 2.6.0
             num_gpus: 1
             axolotl_extras:
             dockerfile: "Dockerfile-uv.jinja"
@@ -271,7 +223,7 @@ jobs:
       - 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
       - name: Update env vars
         run: |
           echo "BASE_TAG=main-base-py${{ matrix.python_version }}-cu${{ matrix.cuda }}-${{ matrix.pytorch }}" >> $GITHUB_ENV
@@ -288,32 +240,28 @@ jobs:
           modal run cicd.e2e_tests
 
   docker-e2e-tests:
-    if: >
-      github.repository_owner == 'axolotl-ai-cloud' &&
-      (github.event_name != 'pull_request' || !github.event.pull_request.draft) &&
-      needs.gate-skip-e2e.outputs.skip != 'true'
+    if: ${{ github.repository_owner == 'axolotl-ai-cloud' && !github.event.pull_request.draft }}
     # this job needs to be run on self-hosted GPU runners...
     runs-on: [self-hosted, modal]
     timeout-minutes: 120
     # Only run the remainder of the matrix if the first e2e check passed;
     # this is to save on wasted compute costs for known failures that get caught in the first run
-    needs: [pre-commit, pytest, gate-skip-e2e, docker-e2e-tests-1st]
+    needs: [pre-commit, pytest, docker-e2e-tests-1st]
 
     strategy:
       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
             num_gpus: 1
-            gpu_type: "B200"
-            axolotl_extras: fbgemm-gpu
+            axolotl_extras:
           - cuda: 128
             cuda_version: 12.8.1
             python_version: "3.11"
-            pytorch: 2.9.1
+            pytorch: 2.7.1
             num_gpus: 1
             axolotl_extras:
     steps:
@@ -326,7 +274,7 @@ jobs:
       - 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
       - name: Update env vars
         run: |
           echo "BASE_TAG=main-base-py${{ matrix.python_version }}-cu${{ matrix.cuda }}-${{ matrix.pytorch }}" >> $GITHUB_ENV
@@ -336,7 +284,6 @@ jobs:
           echo "CUDA=${{ matrix.cuda }}" >> $GITHUB_ENV
           echo "MODAL_IMAGE_BUILDER_VERSION=2024.10" >> $GITHUB_ENV
           echo "N_GPUS=${{ matrix.num_gpus }}" >> $GITHUB_ENV
-          echo "GPU_TYPE=${{ matrix.gpu_type || 'L40S'}}" >> $GITHUB_ENV
           echo "CODECOV_TOKEN=${{ secrets.CODECOV_TOKEN }}" >> $GITHUB_ENV
           echo "E2E_DOCKERFILE=${{ matrix.dockerfile || 'Dockerfile.jinja'}}" >> $GITHUB_ENV
       - name: Run tests job on Modal
@@ -353,10 +300,10 @@ jobs:
       fail-fast: false
       matrix:
         include:
-          - cuda: 128
-            cuda_version: 12.8.1
+          - cuda: 124
+            cuda_version: 12.4.1
             python_version: "3.11"
-            pytorch: 2.9.1
+            pytorch: 2.6.0
             num_gpus: 1
             axolotl_extras:
     steps:
@@ -369,7 +316,7 @@ jobs:
       - name: Install Modal
         run: |
           python -m pip install --upgrade pip
-          pip install modal==1.3.0.post1 jinja2
+          pip install modal==1.0.2 jinja2
       - name: Update env vars
         run: |
           echo "BASE_TAG=main-base-py${{ matrix.python_version }}-cu${{ matrix.cuda }}-${{ matrix.pytorch }}" >> $GITHUB_ENV

.gitignore

@@ -190,6 +190,3 @@ out/
 
 # vim
 *.swp
-
-# scm auto-versioning
-src/axolotl/_version.py

.isort.cfg

@@ -0,0 +1,4 @@
+[settings]
+profile=black
+known_third_party=wandb,comet_ml
+known_local_folder=src,tests

.pre-commit-config.yaml

@@ -10,14 +10,24 @@ repos:
     -   id: trailing-whitespace
     -   id: no-commit-to-branch
         args: ['--branch', 'main']
--   repo: https://github.com/astral-sh/ruff-pre-commit
-    rev: v0.14.10
+-   repo: https://github.com/psf/black
+    rev: 25.1.0
     hooks:
-    -   id: ruff
-        args: [--fix]
-    -   id: ruff-format
+    -   id: black
+-   repo: https://github.com/pycqa/isort
+    rev: 6.0.1
+    hooks:
+      - id: isort
+-   repo: https://github.com/PyCQA/flake8
+    rev: 7.3.0
+    hooks:
+    - id: flake8
+-   repo: https://github.com/pylint-dev/pylint
+    rev: v3.3.8
+    hooks:
+    - id: pylint
 -   repo: https://github.com/pre-commit/mirrors-mypy
-    rev: v1.19.1
+    rev: v1.17.1
     hooks:
     - id: mypy
       additional_dependencies:
@@ -26,7 +36,7 @@ repos:
             'pydantic>=2.5.3',
         ]
 -   repo: https://github.com/PyCQA/bandit
-    rev: 1.9.2
+    rev: 1.8.6
     hooks:
     -   id: bandit
         args: [

.pylintrc

@@ -0,0 +1,15 @@
+[MASTER]
+init-hook="from pylint.config import find_default_config_files; import sys; sys.path.append(next(find_default_config_files()).parent.as_posix())"
+
+[TYPECHECK]
+
+# List of members which are set dynamically and missed by Pylint inference
+# system, and so shouldn't trigger E1101 when accessed.
+generated-members=numpy.*, torch.*
+
+
+[pylint.messages_control]
+disable=missing-function-docstring, line-too-long, import-error,
+    too-many-arguments, too-many-locals, too-many-statements, too-many-branches, too-few-public-methods,
+    too-many-instance-attributes, fixme, import-outside-toplevel, logging-fstring-interpolation,
+    too-many-positional-arguments, possibly-used-before-assignment

.runpod/Dockerfile

@@ -10,7 +10,6 @@ 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

CITATION.cff

@@ -1,6 +1,6 @@
 cff-version: 1.2.0
 type: software
-title: "Axolotl: Open Source LLM Post-Training"
+title: "Axolotl: Post-Training for AI Models"
 message: "If you use this software, please cite it as below."
 authors:
   - name: "Axolotl maintainers and contributors"

README.md

@@ -5,9 +5,6 @@
         <img alt="Axolotl" src="https://raw.githubusercontent.com/axolotl-ai-cloud/axolotl/887513285d98132142bf5db2a74eb5e0928787f1/image/axolotl_logo_digital_black.svg" width="400" height="104" style="max-width: 100%;">
     </picture>
 </p>
-  <p align="center">
-      <strong>A Free and Open Source LLM Fine-tuning Framework</strong><br>
-  </p>
 
 <p align="center">
     <img src="https://img.shields.io/github/license/axolotl-ai-cloud/axolotl.svg?color=blue" alt="GitHub License">
@@ -20,7 +17,6 @@
     <br/>
     <a href="https://discord.com/invite/HhrNrHJPRb"><img src="https://img.shields.io/badge/discord-7289da.svg?style=flat-square&logo=discord" alt="discord" style="height: 20px;"></a>
     <a href="https://twitter.com/axolotl_ai"><img src="https://img.shields.io/twitter/follow/axolotl_ai?style=social" alt="twitter" style="height: 20px;"></a>
-    <a href="https://colab.research.google.com/github/axolotl-ai-cloud/axolotl/blob/main/examples/colab-notebooks/colab-axolotl-example.ipynb"><img src="https://colab.research.google.com/assets/colab-badge.svg" alt="google-colab" style="height: 20px;"></a>
     <br/>
     <img src="https://github.com/axolotl-ai-cloud/axolotl/actions/workflows/tests-nightly.yml/badge.svg" alt="tests-nightly">
     <img src="https://github.com/axolotl-ai-cloud/axolotl/actions/workflows/multi-gpu-e2e.yml/badge.svg" alt="multigpu-semi-weekly tests">
@@ -29,25 +25,21 @@
 
 ## 🎉 Latest Updates
 
-- 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).
-- 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).
-- 2025/09: 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).
+  - 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://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!
+  - [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>
 
-- 2025/03: 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 [docs](https://docs.axolotl.ai/docs/models/magistral.html) to start training your own Magistral models with Axolotl!
-- 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!
+- 2025/06: 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!
@@ -57,31 +49,26 @@
 
 ## ✨ Overview
 
-Axolotl is a free and open-source tool designed to streamline post-training and fine-tuning for the latest large language models (LLMs).
+Axolotl is a tool designed to streamline post-training for various AI models.
 
 Features:
 
-- **Multiple Model Support**: Train various models like GPT-OSS, LLaMA, Mistral, Mixtral, Pythia, and many more models available on the Hugging Face Hub.
-- **Multimodal Training**: Fine-tune vision-language models (VLMs) including LLaMA-Vision, Qwen2-VL, Pixtral, LLaVA, SmolVLM2, and audio models like Voxtral with image, video, and audio support.
-- **Training Methods**: Full fine-tuning, LoRA, QLoRA, GPTQ, QAT, Preference Tuning (DPO, IPO, KTO, ORPO), RL (GRPO), and Reward Modelling (RM) / Process Reward Modelling (PRM).
-- **Easy Configuration**: Re-use a single YAML configuration file across the full fine-tuning pipeline: dataset preprocessing, training, evaluation, quantization, and inference.
+- **Multiple Model Support**: Train various models like LLaMA, Mistral, Mixtral, Pythia, and more. We are compatible with HuggingFace transformers causal language models.
+- **Training Methods**: Full fine-tuning, LoRA, QLoRA, GPTQ, QAT, Preference Tuning (DPO, IPO, KTO, ORPO), RL (GRPO), Multimodal, and Reward Modelling (RM) / Process Reward Modelling (PRM).
+- **Easy Configuration**: Re-use a single YAML file between dataset preprocess, training, evaluation, quantization, and inference.
 - **Performance Optimizations**: [Multipacking](https://docs.axolotl.ai/docs/multipack.html), [Flash Attention](https://github.com/Dao-AILab/flash-attention), [Xformers](https://github.com/facebookresearch/xformers), [Flex Attention](https://pytorch.org/blog/flexattention/), [Liger Kernel](https://github.com/linkedin/Liger-Kernel), [Cut Cross Entropy](https://github.com/apple/ml-cross-entropy/tree/main), [Sequence Parallelism (SP)](https://docs.axolotl.ai/docs/sequence_parallelism.html), [LoRA optimizations](https://docs.axolotl.ai/docs/lora_optims.html), [Multi-GPU training (FSDP1, FSDP2, DeepSpeed)](https://docs.axolotl.ai/docs/multi-gpu.html), [Multi-node training (Torchrun, Ray)](https://docs.axolotl.ai/docs/multi-node.html), and many more!
 - **Flexible Dataset Handling**: Load from local, HuggingFace, and cloud (S3, Azure, GCP, OCI) datasets.
 - **Cloud Ready**: We ship [Docker images](https://hub.docker.com/u/axolotlai) and also [PyPI packages](https://pypi.org/project/axolotl/) for use on cloud platforms and local hardware.
 
 
 
-## 🚀 Quick Start - LLM Fine-tuning in Minutes
+## 🚀 Quick Start
 
 **Requirements**:
 
 - NVIDIA GPU (Ampere or newer for `bf16` and Flash Attention) or AMD GPU
 - Python 3.11
-- PyTorch ≥2.8.0
-
-### Google Colab
-
-[![Open In Colab](https://colab.research.google.com/assets/colab-badge.svg)](https://colab.research.google.com/github/axolotl-ai-cloud/axolotl/blob/main/examples/colab-notebooks/colab-axolotl-example.ipynb#scrollTo=msOCO4NRmRLa)
+- PyTorch ≥2.6.0
 
 ### Installation
 
@@ -158,13 +145,6 @@ 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)
@@ -175,7 +155,7 @@ If you use Axolotl in your research or projects, please cite it as follows:
 
 ```bibtex
 @software{axolotl,
-  title = {Axolotl: Open Source LLM Post-Training},
+  title = {Axolotl: Post-Training for AI Models},
   author = {{Axolotl maintainers and contributors}},
   url = {https://github.com/axolotl-ai-cloud/axolotl},
   license = {Apache-2.0},

TODO.md

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

_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
@@ -155,7 +153,7 @@ quartodoc:
         - utils.distributed
         - utils.dict
         - utils.optimizers.adopt
-        - utils.data.streaming
+        - utils.data.pretraining
         - utils.data.sft
         - utils.quantization
     - title: Schemas
@@ -242,48 +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
-                - docs/models/mistral.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/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
@@ -310,13 +267,11 @@ website:
             - docs/dataset_loading.qmd
             - docs/qat.qmd
             - docs/quantize.qmd
-            - docs/optimizations.qmd
 
         - section: "Core Concepts"
           contents:
             - docs/batch_vs_grad.qmd
             - docs/dataset_preprocessing.qmd
-            - docs/streaming.qmd
             - docs/multipack.qmd
             - docs/mixed_precision.qmd
             - docs/optimizers.qmd

cicd/Dockerfile-uv.jinja

@@ -32,7 +32,6 @@ RUN if [ "$NIGHTLY_BUILD" = "true" ] ; then \
     fi
 
 RUN uv pip install packaging==23.2 setuptools==75.8.0
-RUN uv pip install torchvision
 RUN if [ "$AXOLOTL_EXTRAS" != "" ] ; then \
         uv pip install --no-build-isolation -e .[deepspeed,flash-attn,ring-flash-attn,optimizers,ray,$AXOLOTL_EXTRAS] $AXOLOTL_ARGS; \
     else \

cicd/Dockerfile.jinja

@@ -1,6 +1,6 @@
 FROM axolotlai/axolotl-base:{{ BASE_TAG }}
 
-ENV TORCH_CUDA_ARCH_LIST="7.0 7.5 8.0 8.6 8.7 8.9 9.0+PTX"
+ENV TORCH_CUDA_ARCH_LIST="7.0 7.5 8.0 8.6+PTX"
 ENV AXOLOTL_EXTRAS="{{ AXOLOTL_EXTRAS }}"
 ENV AXOLOTL_ARGS="{{ AXOLOTL_ARGS }}"
 ENV CUDA="{{ CUDA }}"
@@ -9,7 +9,7 @@ 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 AXOLOTL_DATASET_PROCESSES="8"
 
 RUN apt-get update && \
     apt-get install -y --allow-change-held-packages vim curl nano libnccl2 libnccl-dev ibverbs-providers ibverbs-utils infiniband-diags librdmacm-dev librdmacm1 rdmacm-utils slurm-wlm
@@ -32,7 +32,7 @@ RUN if [ "$NIGHTLY_BUILD" = "true" ] ; then \
         sed -i 's#^datasets.*#datasets @ git+https://github.com/huggingface/datasets.git@main#' requirements.txt; \
     fi
 
-RUN pip install packaging==23.2 setuptools==75.8.0 psutil
+RUN pip install packaging==23.2 setuptools==75.8.0
 RUN if [ "$AXOLOTL_EXTRAS" != "" ] ; then \
         pip install --no-build-isolation -e .[deepspeed,flash-attn,ring-flash-attn,optimizers,ray,$AXOLOTL_EXTRAS] $AXOLOTL_ARGS; \
     else \

cicd/multigpu.py

@@ -2,6 +2,8 @@
 modal application to run axolotl gpu tests in Modal
 """
 
+# pylint: disable=duplicate-code
+
 import os
 import pathlib
 import tempfile
@@ -61,7 +63,7 @@ def run_cmd(cmd: str, run_folder: str):
 
     # Propagate errors from subprocess.
     if exit_code := subprocess.call(cmd.split(), cwd=run_folder):  # nosec
-        exit(exit_code)
+        exit(exit_code)  # pylint: disable=consider-using-sys-exit
 
 
 @app.function(

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=4 \
+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

@@ -1,5 +1,7 @@
 """Modal app to run axolotl GPU tests"""
 
+# pylint: disable=duplicate-code
+
 import os
 import pathlib
 import tempfile
@@ -57,21 +59,15 @@ VOLUME_CONFIG = {
 }
 
 N_GPUS = int(os.environ.get("N_GPUS", 1))
-GPU_TYPE = os.environ.get("GPU_TYPE", "L40S")
-GPU_CONFIG = f"{GPU_TYPE}:{N_GPUS}"
+GPU_CONFIG = f"L40S:{N_GPUS}"
 
 
 def run_cmd(cmd: str, run_folder: str):
     import subprocess  # nosec
 
     sp_env = os.environ.copy()
-    sp_env["AXOLOTL_DATASET_NUM_PROC"] = "8"
+    sp_env["AXOLOTL_DATASET_PROCESSES"] = "8"
 
     # Propagate errors from subprocess.
-    try:
-        exit_code = subprocess.call(cmd.split(), cwd=run_folder, env=sp_env)  # nosec
-        if exit_code:
-            print(f"Command '{cmd}' failed with exit code {exit_code}")
-            return exit_code
-    except Exception as e:  # pylint: disable=broad-except
-        print(f"Command '{cmd}' failed with exception {e}")
+    if exit_code := subprocess.call(cmd.split(), cwd=run_folder, env=sp_env):  # nosec
+        exit(exit_code)  # pylint: disable=consider-using-sys-exit

codecov.yml

@@ -12,7 +12,7 @@ coverage:
       default:
         # basic
         target: auto
-        threshold: 1%
+        threshold: 0%
         base: auto
         # advanced
         branches: null
@@ -27,7 +27,7 @@ coverage:
       default:
         # basic
         target: auto
-        threshold: 1%
+        threshold: 0%
         base: auto
         # advanced
         branches: null

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-base

@@ -35,24 +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==23.2 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 && \
+    python3 -m pip install --no-cache-dir "causal_conv1d @ git+https://github.com/Dao-AILab/causal-conv1d.git@main" && \
+    python3 -m pip install --no-cache-dir "mamba_ssm @ git+https://github.com/state-spaces/mamba.git@main" && \
     python3 -m pip cache purge
 
-RUN 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
 
-RUN if [ "$PYTORCH_VERSION" =~ ^2\.9\.[0-9]+$ ] && [ "$CUDA" = "128" ] ; then \
-        wget https://github.com/mjun0812/flash-attention-prebuild-wheels/releases/download/v0.4.17/flash_attn-2.8.3+cu128torch2.9-cp311-cp311-linux_x86_64.whl; \
-        pip3 install --no-cache-dir flash_attn-2.8.3+cu128torch2.9-cp311-cp311-linux_x86_64.whl; \
-        rm flash_attn-2.8.3+cu128torch2.9-cp311-cp311-linux_x86_64.whl; \
+RUN if [ "$PYTORCH_VERSION" = "2.6.0" ] && [ "$CUDA" = "124" ] ; then \
+        FLASH_ATTENTION_FORCE_BUILD="TRUE" pip3 install --no-build-isolation flash-attn==2.8.0.post2; \
     fi

docker/Dockerfile-uv-base

@@ -30,13 +30,7 @@ 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==${PYTORCH_VERSION} torchvision \
+    && uv pip install torch==${PYTORCH_VERSION} \
     && uv pip install --no-build-isolation "causal_conv1d @ git+https://github.com/Dao-AILab/causal-conv1d.git@main" \
     && uv pip install "mamba_ssm @ git+https://github.com/state-spaces/mamba.git@main" \
     && uv pip install awscli pydantic
-
-RUN if [ "$PYTORCH_VERSION" = "2.9.0" ] && [ "$CUDA" = "128" ] ; then \
-        wget https://github.com/mjun0812/flash-attention-prebuild-wheels/releases/download/v0.4.17/flash_attn-2.8.3+cu128torch2.9-cp311-cp311-linux_x86_64.whl; \
-        uv pip install --no-cache-dir flash_attn-2.8.3+cu128torch2.9-cp311-cp311-linux_x86_64.whl; \
-        rm flash_attn-2.8.3+cu128torch2.9-cp311-cp311-linux_x86_64.whl; \
-    fi

docs/.gitignore

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

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/dataset-formats/conversation.qmd

@@ -212,21 +212,6 @@ Instead of passing `tools` via the system prompt, an alternative method would be
 Tools need to follow [JSON schema](https://json-schema.org/learn/getting-started-step-by-step).
 :::
 
-::: {.callout-warning}
-If you have tool arguments with same name but different dtypes (like `"time": string` and `"time": number`), please save `arguments: ` as JSON string to prevent `datasets` from having casting issues.
-
-```
-"arguments": "{\"...\": \"...\"}"
-```
-
-The same is applicable for tool parameters.
-
-```
-"parameters": "{\"...\": \"...\"}"
-```
-
-:::
-
 Example config for Llama4:
 ```yaml
 chat_template: llama4

docs/dataset-formats/index.qmd

@@ -61,7 +61,7 @@ While we recommend `.jsonl`, you can also use the other formats (`csv`, `parquet
 
 ### Pre-training without streaming
 
-In the case that the dataset is small and can be loaded entirely into memory, another approach to running pre-training is to use the `completion` format. This would mean that the entire dataset is pre-tokenized instead of on-demand in streaming.
+On the rare case that the dataset is small and can be loaded entirely into memory, another approach to running pre-training is to use the `completion` format. This would mean that the entire dataset is pre-tokenized instead of on-demand in streaming.
 
 One benefit of this is that the tokenization can be performed separately on a CPU-only machine, and then transferred to a GPU machine for training to save costs.
 

docs/debugging.qmd

@@ -29,7 +29,7 @@ While debugging it's helpful to simplify your test scenario as much as possible.
 1. **Make sure you are using the latest version of axolotl**:  This project changes often and bugs get fixed fast.  Check your git branch and make sure you have pulled the latest changes from `main`.
 1. **Eliminate concurrency**: Restrict the number of processes to 1 for both training and data preprocessing:
     - Set `CUDA_VISIBLE_DEVICES` to a single GPU, ex: `export CUDA_VISIBLE_DEVICES=0`.
-    - Set `dataset_num_proc: 1` in your axolotl config or run the training command with `--dataset_num_proc=1`.
+    - Set `dataset_processes: 1` in your axolotl config or run the training command with `--dataset_processes=1`.
 2. **Use a small dataset**: Construct or use a small dataset from HF Hub. When using a small dataset, you will often have to make sure `sample_packing: False` and `eval_sample_packing: False` to avoid errors.  If you are in a pinch and don't have time to construct a small dataset but want to use from the HF Hub, you can shard the data (this will still tokenize the entire dataset, but will only use a fraction of the data for training.  For example, to shard the dataset into 20 pieces, add the following to your axolotl config):
 
     ```yaml
@@ -101,7 +101,7 @@ For example, to mimic the command `cd devtools && CUDA_VISIBLE_DEVICES=0 acceler
                 "-m", "axolotl.cli.train", "dev_chat_template.yml",
                 // The flags below simplify debugging by overriding the axolotl config
                 // with the debugging tips above.  Modify as needed.
-                "--dataset_num_proc=1",      // limits data preprocessing to one process
+                "--dataset_processes=1",      // limits data preprocessing to one process
                 "--max_steps=1",              // limits training to just one step
                 "--batch_size=1",             // minimizes batch size
                 "--micro_batch_size=1",       // minimizes batch size

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/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' / ____`**
@@ -148,7 +140,3 @@ description: Frequently asked questions
 **Q: `ValueError("Backward pass should have cleared tracker of all tensors")`
 
 > A: This may happen due to edge cases in using the modern OffloadActivations context manager for CUDA streams. If you encounter this error, you may have success using the naive implementation with `offload_activations: legacy` in your YAML.
-
-**Q: `Error parsing tool_calls arguments as JSON.`
-
-> A: There is an error parsing string arguments to a dict. Please check your dataset and the error message for more details.

docs/fsdp_qlora.qmd

@@ -1,5 +1,5 @@
 ---
-title: "FSDP + QLoRA"
+title: "FDSP + QLoRA"
 description: Use FSDP with QLoRA to fine-tune large LLMs on consumer GPUs.
 format:
   html:
@@ -23,12 +23,6 @@ To enable `QLoRA` with `FSDP`, you need to perform the following steps:
 2. Enable FSDP in your axolotl config, as [described here](multi-gpu.qmd#sec-fsdp).
 3. Use one of the supported model types: `llama`, `mistral` or `mixtral`.
 
-## Enabling Swap for FSDP2
-
-If available memory is insufficient even after FSDP's CPU offloading, you can enable swap memory usage by setting `cpu_offload_pin_memory: false` alongside `offload_params: true` in FSDP config.
-
-This disables memory pinning, allowing FSDP to use disk swap space as fallback. Disabling memory pinning itself incurs performance overhead, and actually having to use swap adds more, but it may enable training larger models that would otherwise cause OOM errors on resource constrained systems.
-
 ## Example Config
 
 [examples/llama-2/qlora-fsdp.yml](../examples/llama-2/qlora-fsdp.yml) contains an example of how to enable QLoRA + FSDP in axolotl.

docs/installation.qmd

@@ -26,7 +26,7 @@ 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}
@@ -111,7 +111,7 @@ 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.
@@ -134,7 +134,7 @@ For providers supporting Docker:
 
 ### Google Colab {#sec-colab}
 
-[![](https://colab.research.google.com/assets/colab-badge.svg)](https://colab.research.google.com/github/axolotl-ai-cloud/axolotl/blob/main/examples/colab-notebooks/colab-axolotl-example.ipynb#scrollTo=msOCO4NRmRLa)
+Use our [example notebook](../examples/colab-notebooks/colab-axolotl-example.ipynb).
 
 ## Platform-Specific Instructions {#sec-platform-specific}
 

docs/lora_optims.qmd

@@ -5,11 +5,10 @@ description: "Custom autograd functions and Triton kernels in Axolotl for optimi
 
 Inspired by [Unsloth](https://github.com/unslothai/unsloth), we've implemented two
 optimizations for LoRA and QLoRA fine-tuning, supporting both single GPU and multi-GPU
-(including the DDP, DeepSpeed, and FSDP2 settings) training. These include (1) SwiGLU
-and GEGLU activation function Triton kernels, and (2) LoRA MLP and attention custom
-autograd functions. Our goal was to leverage operator fusion and tensor re-use in order
-to improve speed and reduce memory usage during the forward and backward passes of
-these calculations.
+(in the DDP and DeepSpeed settings) training. These include (1) SwiGLU and GEGLU activation function
+Triton kernels, and (2) LoRA MLP and attention custom autograd functions. Our goal was
+to leverage operator fusion and tensor re-use in order to improve speed and reduce
+memory usage during the forward and backward passes of these calculations.
 
 We currently support several common model architectures, including (but not limited to):
 
@@ -132,5 +131,6 @@ computation path.
 ## Future Work
 
 - Support for additional model architectures
+- Support for the FSDP setting
 - Support for dropout and bias
 - Additional operator fusions

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}
 
@@ -69,9 +63,16 @@ Start from Stage 1 -> Stage 2 -> Stage 3.
 
 :::
 
-## Fully Sharded Data Parallel (FSDP) {#sec-fsdp}
+::: {.callout-tip}
 
-FSDP allows you to shard model parameters, gradients, and optimizer states across data parallel workers.
+Using ZeRO Stage 3 with Single-GPU training
+
+ZeRO Stage 3 can be used for training on a single GPU by manually setting the environment variables:
+`WORLD_SIZE=1 LOCAL_RANK=0 MASTER_ADDR=0.0.0.0 MASTER_PORT=29500`
+
+:::
+
+## Fully Sharded Data Parallel (FSDP) {#sec-fsdp}
 
 ::: {.callout-note}
 
@@ -79,10 +80,6 @@ FSDP2 is recommended for new users. FSDP1 is deprecated and will be removed in a
 
 :::
 
-### 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 +97,6 @@ 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 +153,10 @@ 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

@@ -13,15 +13,10 @@ format:
 - [Pixtral](#sec-pixtral)
 - [Llava-1.5](#sec-llava-15)
 - [Mistral-Small-3.1](#sec-mistral-small-31)
-- [Magistral-Small-2509](#sec-magistral-small-2509)
-- [Voxtral](#sec-voxtral)
 - [Gemma-3](#sec-gemma-3)
 - [Gemma-3n](#sec-gemma-3n)
 - [Qwen2-VL](#sec-qwen2-vl)
 - [Qwen2.5-VL](#sec-qwen25-vl)
-- [SmolVLM2](#sec-smolvlm2)
-- [LFM2-VL](#sec-lfm2-vl)
-- [Intern-VL](#sec-intern-vl)
 
 ## Usage
 
@@ -36,13 +31,14 @@ skip_prepare_dataset: true
 remove_unused_columns: false  # leave columns in place as they are needed to handle image embeddings during training
 sample_packing: false  # not yet supported with multimodal
 
-chat_template:  # see in next section if specified
+chat_template:  # see in next section
 
 # example dataset
 datasets:
   - path: HuggingFaceH4/llava-instruct-mix-vsft
     type: chat_template
     split: train[:1%]
+    field_messages: messages
 
 # (optional) if doing lora, only finetune the Language model,
 # leave the vision model and vision tower frozen
@@ -57,14 +53,10 @@ 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
@@ -99,34 +91,10 @@ chat_template: llava
 
 ### Mistral-Small-3.1 {#sec-mistral-small-31}
 
-::: {.callout-tip}
-Please make sure to install vision lib via `pip install 'mistral-common[opencv]==1.8.5'`
-:::
-
 ```yaml
 base_model: mistralai/Mistral-Small-3.1-24B-Instruct-2503
-```
 
-### Magistral-Small-2509 {#sec-magistral-small-2509}
-
-::: {.callout-tip}
-Please make sure to install vision lib via `pip install 'mistral-common[opencv]==1.8.5'`
-:::
-
-```yaml
-base_model: mistralai/Magistral-Small-2509
-```
-
-### Voxtral {#sec-voxtral}
-
-::: {.callout-tip}
-Please make sure to install audio lib via `pip3 install librosa==0.11.0 'mistral_common[audio]==1.8.3'`
-:::
-
-```yaml
-base_model: mistralai/Voxtral-Mini-3B-2507
-
-processor_type: VoxtralProcessor
+chat_template: mistral_v7_tekken
 ```
 
 ### Gemma-3 {#sec-gemma-3}
@@ -175,44 +143,6 @@ base_model: Qwen/Qwen2.5-VL-7B-Instruct
 chat_template: qwen2_vl  # same as qwen2-vl
 ```
 
-### Qwen3-VL {#sec-qwen3-vl}
-
-```yaml
-base_model: Qwen/Qwen3-VL-4B-Instruct
-
-chat_template: qwen2_vl  # same as qwen2-vl
-```
-
-### SmolVLM2 {#sec-smolvlm2}
-
-::: {.callout-tip}
-Please make sure to install `num2words` via `pip3 install num2words==0.5.14`
-:::
-
-```yaml
-base_model: HuggingFaceTB/SmolVLM2-500M-Video-Instruct
-```
-
-### LFM2-VL {#sec-lfm2-vl}
-
-::: {.callout-warning}
-Please uninstall `causal-conv1d` via `pip3 uninstall -y causal-conv1d`
-:::
-
-```yaml
-base_model: LiquidAI/LFM2-VL-450M
-```
-
-### 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.
@@ -251,20 +181,6 @@ You may need to install `librosa` via `pip3 install librosa==0.11.0`.
 
 :::
 
-### Video
-
-::: {.callout-warning}
-
-This is not well tested at the moment. We welcome contributors!
-
-:::
-
-For video loading, you can use the following keys within `content` alongside `"type": "video"`:
-
-- `"path": "/path/to/video.mp4"`
-- `"url": "https://example.com/video.mp4"`
-- `"video": np.ndarray | list[PIL.Image.Image] | torch.Tensor` (or list of the aforementioned)
-
 ### Example
 
 Here is an example of a multi-modal dataset:

docs/optimizations.qmd

@@ -1,133 +0,0 @@
----
-title: Optimizations Guide
-description: A guide to the performance and memory optimizations available in Axolotl.
----
-
-Axolotl includes numerous optimizations to speed up training, reduce memory usage, and handle large models.
-
-This guide provides a high-level overview and directs you to the detailed documentation for each feature.
-
-## Speed Optimizations
-
-These optimizations focus on increasing training throughput and reducing total training time.
-
-### Sample Packing
-
-Improves GPU utilization by combining multiple short sequences into a single packed sequence for training. This requires enabling one of the [attention](#attention-implementations) implementations below.
-
-- **Config:** `sample_packing: true`
-- **Learn more:** [Sample Packing](multipack.qmd)
-
-### Attention Implementations
-
-Using an optimized attention implementation is critical for training speed.
-
-- **[Flash Attention 2](https://github.com/Dao-AILab/flash-attention)**: `flash_attention: true`. **(Recommended)** The industry standard for fast attention on modern GPUs. Requires Ampere or higher. For AMD, check [AMD Support](https://github.com/Dao-AILab/flash-attention?tab=readme-ov-file#amd-rocm-support).
-- **[Flex Attention](https://pytorch.org/blog/flexattention/)**: `flex_attention: true`.
-- **[SDP Attention](https://docs.pytorch.org/docs/stable/generated/torch.nn.functional.scaled_dot_product_attention.html)**: `sdp_attention: true`. PyTorch's native implementation.
-- **[Xformers](https://github.com/facebookresearch/xformers)**: `xformers_attention: true`. Works with FP16.
-
-*Note: You should only enable one attention backend.*
-
-### LoRA Optimizations
-
-Leverages optimized kernels to accelerate LoRA training and reduce memory usage.
-
-- **Learn more:** [LoRA Optimizations Documentation](lora_optims.qmd)
-
-## Memory Optimizations
-
-These techniques help you fit larger models or use bigger batch sizes on your existing hardware.
-
-### Parameter Efficient Finetuning (LoRA & QLoRA)
-
-Drastically reduces memory by training a small set of "adapter" parameters instead of the full model. This is the most common and effective memory-saving technique.
-
-- Examples: Find configs with `lora` or `qlora` in the [examples directory](https://github.com/axolotl-ai-cloud/axolotl/tree/main/examples/llama-3).
-- Config Reference: See `adapter`, `load_in_4bit`, and `load_in_8bit` in the [Configuration Reference](config-reference.qmd).
-
-### Gradient Checkpointing & Activation Offloading
-
-These techniques save VRAM by changing how activations are handled.
-
-- Gradient Checkpointing: re-computes activations during the backward pass, trading compute time for VRAM.
-- Activation Offloading: moves activations to CPU RAM or disk, trading I/O overhead for VRAM.
-- Learn more: [Gradient Checkpointing and Offloading Docs](gradient_checkpointing.qmd)
-
-### Cut Cross Entropy (CCE)
-
-Reduces VRAM usage by using an optimized cross-entropy loss calculation.
-
-- **Learn more:** [Custom Integrations - CCE](custom_integrations.qmd#cut-cross-entropy)
-
-### Liger Kernels
-
-Provides efficient Triton kernels to improve training speed and reduce memory usage.
-
-- **Learn more:** [Custom Integrations - Liger Kernels](custom_integrations.qmd#liger-kernels)
-
-## Long Context Models
-
-Techniques to train models on sequences longer than their original context window.
-
-### RoPE Scaling
-
-Extends a model's context window by interpolating its Rotary Position Embeddings.
-
-- **Config:** Pass the `rope_scaling` config under the `overrides_of_model_config: `. To learn how to set RoPE, check the respective model config.
-
-### Sequence Parallelism
-
-Splits long sequences across multiple GPUs, enabling training with sequence lengths that would not fit on a single device.
-
-- **Learn more:** [Sequence Parallelism Documentation](sequence_parallelism.qmd)
-
-### Artic Long Sequence Training (ALST)
-
-ALST is a recipe that combines several techniques to train long-context models efficiently. It typically involves:
-
-- TiledMLP to reduce memory usage in MLP layers.
-- Tiled Loss functions (like [CCE](#cut-cross-entropy-(cce) or [Liger](#liger-kernels)).
-- Activation Offloading to CPU.
-
-- Example: [ALST Example Configuration](https://github.com/axolotl-ai-cloud/axolotl/tree/main/examples/alst)
-
-## Large Models (Distributed Training)
-
-To train models that don't fit on a single GPU, you'll need to use a distributed training strategy like FSDP or DeepSpeed. These frameworks shard the model weights, gradients, and optimizer states across multiple GPUs and nodes.
-
-- **Learn more:** [Multi-GPU Guide](multi-gpu.qmd)
-- **Learn more:** [Multi-Node Guide](multi-node.qmd)
-
-### N-D Parallelism (Beta)
-
-For advanced scaling, Axolotl allows you to compose different parallelism techniques (e.g., Data, Tensor, Sequence Parallelism). This is a powerful approach to train an extremely large model by overcoming multiple bottlenecks at once.
-
-- **Learn more:** [N-D Parallelism Guide](nd_parallelism.qmd)
-
-
-## Quantization
-
-Techniques to reduce the precision of model weights for memory savings.
-
-### 4-bit Training (QLoRA)
-
-The recommended approach for quantization-based training. It loads the base model in 4-bit using `bitsandbytes` and then trains QLoRA adapters. See [Adapter Finetuning](#adapter-finetuning-lora-qlora) for details.
-
-### FP8 Training
-
-Enables training with 8-bit floating point precision on supported hardware (e.g., NVIDIA Hopper series GPUs) for significant speed and memory gains.
-
-- **Example:** [Llama 3 FP8 FSDP Example](https://github.com/axolotl-ai-cloud/axolotl/blob/main/examples/llama-3/3b-fp8-fsdp2.yaml)
-
-### Quantization Aware Training (QAT)
-
-Simulates quantization effects during training, helping the model adapt and potentially improving the final accuracy of the quantized model.
-
-- **Learn more:** [QAT Documentation](qat.qmd)
-
-### GPTQ
-
-Allows you to finetune LoRA adapters on top of a model that has already been quantized using the GPTQ method.
-
-- **Example:** [GPTQ LoRA Example](https://github.com/axolotl-ai-cloud/axolotl/blob/main/examples/llama-2/gptq-lora.yml)

docs/qat.qmd

@@ -23,18 +23,10 @@ To enable QAT in axolotl, add the following to your configuration file:
 
 ```yaml
 qat:
-  activation_dtype: # Optional[str] = "int8". Fake quantization layout to use for activation quantization. Valid options are "int4", "int8", "float8"
-  weight_dtype: # Optional[str] = "int8". Fake quantization layout to use for weight quantization. Valid options are "int4", "fp8", and "nvfp4".
+  activation_dtype: # Optional[str] = "int8". Fake quantization layout to use for activation quantization. Valid options are "int4" and "int8"
+  weight_dtype: # Optional[str] = "int8". Fake quantization layout to use for weight quantization. Valid options are "int4" and "int8"
   group_size: # Optional[int] = 32. The number of elements in each group for per-group fake quantization
   fake_quant_after_n_steps: # Optional[int] = None. The number of steps to apply fake quantization after
 ```
 
-We support the following quantization schemas:
-
-- `Int4WeightOnly` (requires the `fbgemm-gpu` extra when installing Axolotl)
-- `Int8DynamicActivationInt4Weight`
-- `Float8DynamicActivationFloat8Weight`
-- `Float8DynamicActivationInt4Weight`
-- `NVFP4`
-
 Once you have finished training, you must quantize your model by using the same quantization configuration which you used to train the model with. You can use the [`quantize`](./quantize.qmd) command to do this.

docs/quantize.qmd

@@ -22,8 +22,8 @@ Quantization is configured using the `quantization` key in your configuration fi
 ```yaml
 base_model: # The path to the model to quantize.
 quantization:
-  activation_dtype: # Optional[str] = "int8". Fake quantization layout to use for activation quantization. Valid options are "int4", "int8", "float8"
-  weight_dtype: # Optional[str] = "int8". Fake quantization layout to use for weight quantization. Valid options are "int4", "fp8", and "nvfp4".
+  weight_dtype: # Optional[str] = "int8". Fake quantization layout to use for weight quantization. Valid options are uintX for X in [1, 2, 3, 4, 5, 6, 7], or int4, or int8
+  activation_dtype: # Optional[str] = "int8". Fake quantization layout to use for activation quantization. Valid options are "int4" and "int8"
   group_size: # Optional[int] = 32. The number of elements in each group for per-group fake quantization
   quantize_embedding: # Optional[bool] = False. Whether to quantize the embedding layer.
 
@@ -39,8 +39,9 @@ you used to train the model:
 # qat.yml
 qat:
   activation_dtype: int8
-  weight_dtype: int4
+  weight_dtype: int8
   group_size: 256
+  quantize_embedding: true
 
 output_dir: # The path to the output directory used during training where the final checkpoint has been saved.
 ```
@@ -50,11 +51,3 @@ axolotl quantize qat.yml
 ```
 
 This ensures that an identical quantization configuration is used to quantize the model as was used to train it.
-
-
-::: {.callout-note}
-
-If you have configured pushing to hub with `hub_model_id`, your model hub name will have the quantization schema appended to it,
-e.g. `axolotl-ai-cloud/qat-nvfp4-llama3B` will become `axolotl-ai-cloud/qat-nvfp4-llama3B-nvfp4w`
-
-:::

docs/reward_modelling.qmd

@@ -11,7 +11,6 @@ We support the reward modelling techniques supported by `trl`.
 ### (Outcome) Reward Models
 
 Outcome reward models are trained using data which contains preference annotations for an entire interaction between the user and model (e.g. rather than per-turn or per-step).
-For improved training stability, you can use the `center_rewards_coefficient` parameter to encourage mean-zero reward outputs ([see TRL docs](https://huggingface.co/docs/trl/v0.10.1/en/reward_trainer#centering-rewards)).
 
 ```yaml
 base_model: google/gemma-2-2b

docs/rlhf.qmd

@@ -219,21 +219,6 @@ DPO supports the following types with the following dataset format:
 }
 ```
 
-#### chat_template.argilla_chat
-
-```json
-{
-    "chosen": [
-        {"role": "user", "content": "..."},
-        {"role": "assistant", "content": "..."}
-    ],
-    "rejected": [
-        {"role": "user", "content": "..."},
-        {"role": "assistant", "content": "..."}
-    ]
-}
-```
-
 #### chat_template.default
 
 ```yaml
@@ -597,116 +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.

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_config_docs.py

@@ -47,6 +47,7 @@ class QuartoGenerator:
         """Check if a type is a Pydantic BaseModel."""
         return inspect.isclass(type_obj) and issubclass(type_obj, BaseModel)
 
+    # pylint: disable=too-many-return-statements
     def _extract_nested_type(self, field_type) -> Any:
         """Extract the actual type from complex type annotations."""
         # Handle Annotated types (Python 3.9+)
@@ -123,6 +124,7 @@ class QuartoGenerator:
 
         return field_type
 
+    # pylint: disable=too-many-return-statements
     def _extract_all_pydantic_models_from_type(
         self, field_type
     ) -> list[type[BaseModel]]:
@@ -316,6 +318,7 @@ class QuartoGenerator:
 
         return all_groups
 
+    # pylint: disable=too-many-return-statements
     def _extract_field_groups_from_source(
         self, model_class: type[BaseModel]
     ) -> list[dict]:
@@ -500,7 +503,7 @@ class QuartoGenerator:
                     nested_schema = nested_model.model_json_schema()
                     nested_properties = nested_schema.get("properties", {})
                     nested_required = nested_schema.get("required", [])
-                except Exception:
+                except Exception:  # pylint: disable=broad-exception-caught
                     # Fallback: use model fields directly
                     nested_properties = {}
                     nested_required = []
@@ -604,7 +607,7 @@ class QuartoGenerator:
             schema = model_class.model_json_schema()
             properties = schema.get("properties", {})
             required = schema.get("required", [])
-        except Exception as e:
+        except Exception as e:  # pylint: disable=broad-exception-caught
             print(
                 f"Warning: Could not generate JSON schema ({e}). Using model fields instead."
             )

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/streaming.qmd

@@ -1,120 +0,0 @@
----
-title: Streaming Datasets
-description: How to use streaming mode for large-scale datasets and memory-efficient training
-order: 10
----
-
-Streaming enables memory-efficient training with large datasets by loading data
-incrementally rather than loading the entire dataset into memory at once.
-
-Use streaming when:
-
-- Your dataset is too large to fit in memory (e.g. when you're doing pretraining with massive text corpora)
-- You want to start training immediately without preprocessing the entire dataset
-
-Streaming works with both remote and locally stored datasets!
-
-::: {.callout-note}
-Streaming currently only supports a single dataset. Multi-dataset support will be added soon.
-:::
-
-
-## Configuration
-
-### Basic Streaming
-
-Enable streaming mode by setting the `streaming` flag:
-
-```yaml
-streaming: true
-```
-
-### Pretraining with Streaming
-
-For pretraining tasks, streaming is automatically enabled when using `pretraining_dataset`:
-
-```yaml
-pretraining_dataset:
-  - path: HuggingFaceFW/fineweb-edu
-    type: pretrain
-    text_column: text
-    split: train
-
-# Optionally, enable sample packing
-streaming_multipack_buffer_size: 10000
-sample_packing: true
-```
-
-### SFT with Streaming
-
-For supervised fine-tuning with streaming:
-
-```yaml
-streaming: true
-datasets:
-  - path: tatsu-lab/alpaca
-    type: alpaca
-    split: train
-
-# Optionally, enable sample packing
-streaming_multipack_buffer_size: 10000
-sample_packing: true
-```
-
-## Configuration Options
-
-### `streaming_multipack_buffer_size`
-
-Controls the buffer size for multipack streaming (default: 10,000). This determines how
-many samples are buffered before packing. Larger buffers can improve packing efficiency
-but use more memory.
-
-### `shuffle_merged_datasets`
-
-When enabled, shuffles the streaming dataset using the buffer. This requires additional
-memory for the shuffle buffer.
-
-## Sample Packing with Streaming
-
-Sample packing is supported for streaming datasets. When enabled, multiple samples are
-packed into a single sequence to maximize GPU utilization:
-
-```yaml
-sample_packing: true
-streaming_multipack_buffer_size: 10000
-
-# For SFT: attention is automatically isolated between packed samples
-# For pretraining: control with pretrain_multipack_attn
-pretrain_multipack_attn: true  # prevent cross-attention between packed samples
-```
-
-For more information, see our [documentation](multipack.qmd) on multipacking.
-
-## Important Considerations
-
-### Memory Usage
-
-While streaming reduces memory usage compared to loading entire datasets, you still need
-to consider:
-
-- You can control the memory usage by adjusting `streaming_multipack_buffer_size`
-- Sample packing requires buffering multiple samples
-- Shuffling requires additional memory for the shuffle buffer
-
-### Performance
-
-- Streaming may have slightly higher latency compared to preprocessed datasets, as samples are processed on-the-fly
-- Network speed and disk read speed are important when streaming from remote sources or a local dataset, respectively
-- Consider using `axolotl preprocess` for smaller or more frequently used datasets
-
-### Evaluation Datasets
-
-Evaluation datasets are not streamed to ensure consistent evaluation metrics. They're
-loaded normally even when training uses streaming.
-
-## Examples
-
-See the `examples/streaming/` directory for complete configuration examples:
-
-- `pretrain.yaml`: Pretraining with streaming dataset
-- `sft.yaml`: Supervised fine-tuning with streaming

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

@@ -1,67 +0,0 @@
-# Finetune Liquid Foundation Models 2 (LFM2) with Axolotl
-
-[Liquid Foundation Models 2 (LFM2)](https://huggingface.co/collections/LiquidAI/lfm2-686d721927015b2ad73eaa38) are a family of small, open-weight models from [Liquid AI](https://www.liquid.ai/) focused on quality, speed, and memory efficiency. Liquid AI released text-only [LFM2](https://huggingface.co/collections/LiquidAI/lfm2-686d721927015b2ad73eaa38) and text+vision [LFM2-VL](https://huggingface.co/collections/LiquidAI/lfm2-vl-68963bbc84a610f7638d5ffa) models.
-
-LFM2 features a new hybrid Liquid architecture with multiplicative gates, short-range convolutions, and grouped query attention, enabling fast training and inference.
-
-This guide shows how to fine-tune both the LFM2 and LFM2-VL models with Axolotl.
-
-Thanks to the team at LiquidAI for giving us early access to prepare for these releases.
-
-## Getting Started
-
-1.  Install Axolotl following the [installation guide](https://docs.axolotl.ai/docs/installation.html).
-
-    Here is an example of how to install from pip:
-    ```bash
-    # Ensure you have a compatible version of Pytorch installed
-    pip3 install packaging setuptools wheel ninja
-    pip3 install --no-build-isolation 'axolotl[flash-attn]>=0.12.0'
-    ```
-
-2.  Run one of the finetuning examples below.
-
-    **LFM2**
-    ```bash
-    # FFT SFT (1x48GB @ 25GiB)
-    axolotl train examples/LiquidAI/lfm2-350m-fft.yaml
-    ```
-
-    **LFM2-VL**
-    ```bash
-    # LoRA SFT (1x48GB @ 2.7GiB)
-    axolotl train examples/LiquidAI/lfm2-vl-lora.yaml
-    ```
-
-    **LFM2-MoE**
-    ```bash
-    pip install git+https://github.com/huggingface/transformers.git@0c9a72e4576fe4c84077f066e585129c97bfd4e6
-
-    # LoRA SFT (1x48GB @ 16.2GiB)
-    axolotl train examples/LiquidAI/lfm2-8b-a1b-lora.yaml
-    ```
-
-### TIPS
-
-- **Installation Error**: If you encounter `ImportError: ... undefined symbol ...` or `ModuleNotFoundError: No module named 'causal_conv1d_cuda'`, the `causal-conv1d` package may have been installed incorrectly. Try uninstalling it:
-  ```bash
-  pip uninstall -y causal-conv1d
-  ```
-
-- **Dataset Loading**: Read more on how to load your own dataset in our [documentation](https://docs.axolotl.ai/docs/dataset_loading.html).
-- **Dataset Formats**:
-  - For LFM2 models, the dataset format follows the OpenAI Messages format as seen [here](https://docs.axolotl.ai/docs/dataset-formats/conversation.html#chat_template).
-  - For LFM2-VL models, Axolotl follows the multi-content Messages format. See our [Multimodal docs](https://docs.axolotl.ai/docs/multimodal.html#dataset-format) for details.
-
-## Optimization Guides
-
-- [Optimizations Guide](https://docs.axolotl.ai/docs/optimizations.html)
-
-## Related Resources
-
-- [LFM2 Blog](https://www.liquid.ai/blog/liquid-foundation-models-v2-our-second-series-of-generative-ai-models)
-- [LFM2-VL Blog](https://www.liquid.ai/blog/lfm2-vl-efficient-vision-language-models)
-- [LFM2-MoE Blog](https://www.liquid.ai/blog/lfm2-8b-a1b-an-efficient-on-device-mixture-of-experts)
-- [Axolotl Docs](https://docs.axolotl.ai)
-- [Axolotl GitHub](https://github.com/axolotl-ai-cloud/axolotl)
-- [Axolotl Discord](https://discord.gg/7m9sfhzaf3)

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

@@ -1,61 +0,0 @@
-base_model: LiquidAI/LFM2-VL-450M
-trust_remote_code: true
-model_type: AutoModelForImageTextToText
-processor_type: AutoProcessor
-
-plugins:
-  - axolotl.integrations.cut_cross_entropy.CutCrossEntropyPlugin
-
-# these 3 lines are needed for now to handle vision chat templates w images
-skip_prepare_dataset: true
-remove_unused_columns: false
-sample_packing: false
-
-datasets:
-  - path: HuggingFaceH4/llava-instruct-mix-vsft
-    type: chat_template
-    split: train[:1%]
-
-dataset_prepared_path: last_run_prepared
-val_set_size: 0.0
-output_dir: ./outputs/out
-
-adapter: lora
-lora_model_dir:
-
-sequence_len: 8192
-pad_to_sequence_len: false
-
-lora_r: 32
-lora_alpha: 16
-lora_dropout: 0.05
-lora_target_modules: 'model.language_model.layers.[\d]+.(mlp|cross_attn|self_attn).(up|down|gate|q|k|v|o)_proj'
-
-wandb_project:
-wandb_entity:
-wandb_watch:
-wandb_name:
-wandb_log_model:
-
-gradient_accumulation_steps: 4
-micro_batch_size: 1
-num_epochs: 1
-optimizer: adamw_bnb_8bit
-lr_scheduler: cosine
-learning_rate: 0.0002
-
-bf16: true
-fp16:
-tf32: true
-
-gradient_checkpointing: true
-logging_steps: 1
-flash_attention: true
-eager_attention:
-
-warmup_ratio: 0.1
-evals_per_epoch: 1
-saves_per_epoch: 1
-weight_decay: 0.0
-
-# save_first_step: true  # uncomment this to validate checkpoint saving works with your config

examples/alst/README.md

@@ -7,24 +7,3 @@ techniques. It is a combination of:
 - Activation Offloading: Offload activations to CPU RAM to reduce memory usage
 
 For more information, you can check out the ALST paper [here](https://www.arxiv.org/abs/2506.13996).
-
-## Usage
-
-```yaml
-tiled_mlp: true
-
-# See Sequence Parallelism docs
-# https://docs.axolotl.ai/docs/sequence_parallelism.html
-context_parallel_size: int
-
-plugins:
-# See Cut Cross Entropy docs
-# https://docs.axolotl.ai/docs/custom_integrations.html#cut-cross-entropy
-  - axolotl.integrations.cut_cross_entropy.CutCrossEntropyPlugin
-
-# or Liger Kernel docs
-# https://docs.axolotl.ai/docs/custom_integrations.html#liger-kernels
-  - axolotl.integrations.liger.LigerPlugin
-# ...
-
-```

examples/apertus/README.md

@@ -1,110 +0,0 @@
-# Finetune Swiss-AI's Apertus with Axolotl
-
-[Apertus](https://huggingface.co/collections/swiss-ai/apertus-llm-68b699e65415c231ace3b059) is a family of opensource models trained by Swiss-ai.
-
-This guide shows how to fine-tune it with Axolotl with multi-turn conversations and proper masking.
-
-## Getting started
-
-1. Install Axolotl following the [installation guide](https://docs.axolotl.ai/docs/installation.html). You need to install from main as Apertus is only on nightly or use our latest [Docker images](https://docs.axolotl.ai/docs/docker.html).
-
-    Here is an example of how to install from main for pip:
-
-```bash
-# Ensure you have Pytorch installed (Pytorch 2.6.0 min)
-git clone https://github.com/axolotl-ai-cloud/axolotl.git
-cd axolotl
-
-pip3 install packaging==23.2 setuptools==75.8.0 wheel ninja
-pip3 install --no-build-isolation -e '.[flash-attn]'
-
-# Install CCE https://docs.axolotl.ai/docs/custom_integrations.html#cut-cross-entropy
-python scripts/cutcrossentropy_install.py | sh
-```
-
-2. (Optional, highly recommended) Install XIELU CUDA
-
-```bash
-## Recommended for reduced VRAM and faster speeds
-
-# Point to CUDA toolkit directory
-# For those using our Docker image, use the below path.
-export CUDA_HOME=/usr/local/cuda
-
-pip3 install git+https://github.com/nickjbrowning/XIELU@59d6031 --no-build-isolation --no-deps
-```
-
-For any installation errors, see [XIELU Installation Issues](#xielu-installation-issues)
-
-3. Run the finetuning example:
-
-```bash
-axolotl train examples/apertus/apertus-8b-qlora.yaml
-```
-
-This config uses about 8.7 GiB VRAM.
-
-Let us know how it goes. Happy finetuning! 🚀
-
-### Tips
-
-- For inference, the official Apertus team recommends `top_p=0.9` and `temperature=0.8`.
-- You can instead use full paremter fine-tuning by removing the `adapter: qlora` and `load_in_4bit: true` from the config.
-- Read more on how to load your own dataset at [docs](https://docs.axolotl.ai/docs/dataset_loading.html).
-- The dataset format follows the OpenAI Messages format as seen [here](https://docs.axolotl.ai/docs/dataset-formats/conversation.html#chat_template).
-
-### XIELU Installation Issues
-
-#### `ModuleNotFoundError: No module named 'torch'`
-
-Please check these one by one:
-- Running in correct environment
-- Env has PyTorch installed
-- CUDA toolkit is at `CUDA_HOME`
-
-If those didn't help, please try the below solutions:
-
-1. Pass env for CMAKE and try install again:
-
-    ```bash
-    Python_EXECUTABLE=$(which python) pip3 install git+https://github.com/nickjbrowning/XIELU@59d6031 --no-build-isolation --no-deps
-    ```
-
-2. Git clone the repo and manually hardcode python path:
-
-    ```bash
-    git clone https://github.com/nickjbrowning/XIELU
-    cd xielu
-    git checkout 59d6031
-
-    cd xielu
-    nano CMakeLists.txt  # or vi depending on your preference
-    ```
-
-    ```diff
-    execute_process(
-    -    COMMAND ${Python_EXECUTABLE} -c "import torch.utils; print(torch.utils.cmake_prefix_path)"
-    +    COMMAND /root/miniconda3/envs/py3.11/bin/python -c "import torch.utils; print(torch.utils.cmake_prefix_path)"
-        RESULT_VARIABLE TORCH_CMAKE_PATH_RESULT
-        OUTPUT_VARIABLE TORCH_CMAKE_PATH_OUTPUT
-        ERROR_VARIABLE TORCH_CMAKE_PATH_ERROR
-    )
-    ```
-
-    ```bash
-    pip3 install . --no-build-isolation --no-deps
-    ```
-
-## Optimization Guides
-
-- [Multi-GPU Training](https://docs.axolotl.ai/docs/multi-gpu.html)
-- [Multi-Node Training](https://docs.axolotl.ai/docs/multi-node.html)
-- [LoRA Optimizations](https://docs.axolotl.ai/docs/lora_optims.html)
-
-## Related Resources
-
-- [Apertus Tech Report](https://github.com/swiss-ai/apertus-tech-report/blob/main/Apertus_Tech_Report.pdf)
-- [Axolotl Docs](https://docs.axolotl.ai)
-- [Axolotl Website](https://axolotl.ai)
-- [Axolotl GitHub](https://github.com/axolotl-ai-cloud/axolotl)
-- [Axolotl Discord](https://discord.gg/7m9sfhzaf3)

examples/apertus/apertus-8b-qlora.yaml

@@ -1,64 +0,0 @@
-base_model: swiss-ai/Apertus-8B-Instruct-2509
-
-# Automatically upload checkpoint and final model to HF
-# hub_model_id: username/custom_model_name
-
-plugins:
-  - axolotl.integrations.cut_cross_entropy.CutCrossEntropyPlugin
-
-load_in_8bit: false
-load_in_4bit: true
-
-datasets:
-  - path: fozziethebeat/alpaca_messages_2k_test
-    type: chat_template
-
-dataset_prepared_path: last_run_prepared
-val_set_size: 0.1
-output_dir: ./outputs/lora-out
-
-adapter: qlora
-lora_model_dir:
-
-sequence_len: 2048
-sample_packing: true
-
-lora_r: 32
-lora_alpha: 16
-lora_dropout: 0.05
-lora_target_linear: true
-lora_target_modules:
-  - gate_proj
-  - down_proj
-  - up_proj
-  - q_proj
-  - v_proj
-  - k_proj
-  - o_proj
-
-wandb_project:
-wandb_entity:
-wandb_watch:
-wandb_name:
-wandb_log_model:
-
-gradient_accumulation_steps: 4
-micro_batch_size: 2
-num_epochs: 1
-optimizer: adamw_bnb_8bit
-lr_scheduler: cosine
-learning_rate: 0.0002
-
-bf16: auto
-tf32: false
-
-gradient_checkpointing: true
-resume_from_checkpoint:
-logging_steps: 1
-flash_attention: true
-
-warmup_ratio: 0.1
-evals_per_epoch: 1
-saves_per_epoch: 1
-
-# save_first_step: true  # uncomment this to validate checkpoint saving works with your config

examples/arcee/README.md

@@ -19,9 +19,6 @@ cd axolotl
 
 pip3 install packaging==23.2 setuptools==75.8.0 wheel ninja
 pip3 install --no-build-isolation -e '.[flash-attn]'
-
-# Install CCE https://docs.axolotl.ai/docs/custom_integrations.html#cut-cross-entropy
-python scripts/cutcrossentropy_install.py | sh
 ```
 
 2. Run the finetuning example:

examples/archived/deepcoder/deepcoder-14B-preview-lora.yml

@@ -9,6 +9,10 @@ strict: false
 datasets:
   - path: fozziethebeat/alpaca_messages_2k_test
     type: chat_template
+    field_messages: messages
+    message_property_mappings:
+      role: role
+      content: content
 
 dataset_prepared_path:
 val_set_size: 0.05

examples/cloud/baseten.yaml

@@ -1,10 +0,0 @@
-provider: baseten
-project_name:
-
-secrets:
-  - HF_TOKEN
-  - WANDB_API_KEY
-
-gpu: h100
-gpu_count: 8
-node_count: 1

examples/deepcogito/cogito-v1-preview-llama-3B-lora.yml

@@ -9,6 +9,10 @@ strict: false
 datasets:
   - path: fozziethebeat/alpaca_messages_2k_test
     type: chat_template
+    field_messages: messages
+    message_property_mappings:
+      role: role
+      content: content
 
 dataset_prepared_path:
 val_set_size: 0.05

examples/deepcogito/cogito-v1-preview-qwen-14B-lora.yml

@@ -9,6 +9,10 @@ strict: false
 datasets:
   - path: fozziethebeat/alpaca_messages_2k_test
     type: chat_template
+    field_messages: messages
+    message_property_mappings:
+      role: role
+      content: content
 
 dataset_prepared_path:
 val_set_size: 0.05

examples/devstral/README.md

@@ -20,13 +20,7 @@ pip3 install packaging==23.2 setuptools==75.8.0 wheel ninja
 pip3 install --no-build-isolation 'axolotl[flash-attn]>=0.12.0'
 ```
 
-2. Install [Cut Cross Entropy](https://docs.axolotl.ai/docs/custom_integrations.html#cut-cross-entropy) to reduce training VRAM usage
-
-```bash
-python scripts/cutcrossentropy_install.py | sh
-```
-
-3. Run the finetuning example:
+2. Run the finetuning example:
 
 ```bash
 axolotl train examples/devstral/devstral-small-qlora.yml

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

@@ -1,7 +1,7 @@
 base_model: google/gemma-3-1b-it
-
-model_type: Gemma3ForCausalLM
-
+# 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
 

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

@@ -1,68 +0,0 @@
-base_model: google/gemma-3-270m-it
-
-model_type: Gemma3ForCausalLM
-
-# Automatically upload checkpoint and final model to HF
-# hub_model_id: username/custom_model_name
-
-# gemma3 doesn't seem to play nice with ddp
-ddp_find_unused_parameters: true
-
-load_in_8bit: false
-load_in_4bit: true
-
-# huggingface repo
-chat_template: gemma3
-eot_tokens:
-  - <end_of_turn>
-datasets:
-  - path: cgato/SlimOrcaDedupCleaned
-    type: chat_template
-    field_messages: conversations
-    message_property_mappings:
-      role: from
-      content: value
-
-val_set_size: 0.0
-output_dir: ./outputs/out
-
-adapter: qlora
-lora_r: 32
-lora_alpha: 16
-lora_dropout: 0.05
-lora_target_linear: true
-
-sequence_len: 2048
-sample_packing: true
-eval_sample_packing: false
-
-
-wandb_project:
-wandb_entity:
-wandb_watch:
-wandb_name:
-wandb_log_model:
-
-
-gradient_accumulation_steps: 4
-micro_batch_size: 1
-num_epochs: 1
-optimizer: adamw_bnb_8bit
-lr_scheduler: cosine
-learning_rate: 0.0002
-
-bf16: auto
-tf32: true
-
-gradient_checkpointing: true
-gradient_checkpointing_kwargs:
-  use_reentrant: false
-resume_from_checkpoint:
-logging_steps: 1
-flash_attention: true
-
-warmup_ratio: 0.1
-evals_per_epoch:
-saves_per_epoch: 1
-weight_decay: 0.0
-special_tokens:

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

@@ -1,8 +1,5 @@
 base_model: google/gemma-3-4b-it
 
-# Need to set else transformers tries to load vision too
-model_type: Gemma3ForCausalLM
-
 load_in_4bit: true
 
 # gemma3 doesn't seem to play nice with ddp

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

@@ -18,7 +18,7 @@ datasets:
   - path: HuggingFaceH4/llava-instruct-mix-vsft
     type: chat_template
     split: train[:1%]
-
+    field_messages: messages
 dataset_prepared_path: last_run_prepared
 val_set_size: 0.01
 output_dir: ./outputs/out

examples/gemma3n/README.md

@@ -23,15 +23,7 @@ pip3 install timm==1.0.17
 pip3 install librosa==0.11.0
 ```
 
-3. Download sample dataset files
-
-```bash
-# for text + vision + audio only
-wget https://huggingface.co/datasets/Nanobit/text-vision-audio-2k-test/resolve/main/African_elephant.jpg
-wget https://huggingface.co/datasets/Nanobit/text-vision-audio-2k-test/resolve/main/En-us-African_elephant.oga
-```
-
-4. Run the finetuning example:
+3. Run the finetuning example:
 
 ```bash
 # text only

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
@@ -35,74 +33,13 @@ Note: Memory usage taken from `device_mem_reserved(gib)` from logs.
 
 ### Training 120B
 
-On 8xH100s, make sure you have ~3TB of free disk space. With each checkpoint clocking in at ~720GB, along with the base
-model, and final model output, you may need at least 3TB of free disk space to keep at least 2 checkpoints.
+On 8xH100s
 
 ```bash
 # FFT SFT with offloading (8x80GB @ ~49GiB/GPU)
 axolotl train examples/gpt-oss/gpt-oss-120b-fft-fsdp2-offload.yaml
 ```
 
-To simplify fine-tuning across 2 nodes × 8x H100 (80GB) GPUs, we've partnered with [Baseten](https://baseten.co) to showcase multi-node
-training of the 120B model using Baseten Truss. You can read more about this recipe on
-[Baseten's blog](https://www.baseten.co/blog/how-to-fine-tune-gpt-oss-120b-with-baseten-and-axolotl/). The recipe can
-be found on their
-[GitHub](https://github.com/basetenlabs/ml-cookbook/tree/main/examples/oss-gpt-120b-axolotl/training).
-
-ERRATA: Transformers saves the model Architecture prefixed with `FSDP` which needs to be manually renamed in `config.json`.
-See https://github.com/huggingface/transformers/pull/40207 for the status of this issue.
-
-```bash
-sed -i 's/FSDPGptOssForCausalLM/GptOssForCausalLM/g' ./outputs/gpt-oss-out/config.json
-```
-
-When using SHARDED_STATE_DICT with FSDP, the final checkpoint should automatically merge the sharded weights to your
-configured `output_dir`. However, if that step fails due to a disk space error, you can take an additional step to
-merge the sharded weights.  This step will automatically determine the last checkpoint directory and merge the sharded
-weights to `{output_dir}/merged`.
-
-```bash
-axolotl merge-sharded-fsdp-weights examples/gpt-oss/gpt-oss-120b-fft-fsdp2-offload.yaml
-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
-
-#### vLLM
-
-GPT-OSS support in vLLM does not exist in a stable release yet. See https://x.com/MaziyarPanahi/status/1955741905515323425
-for more information about using a special vllm-openai docker image for inferencing with vLLM.
-
-Optionally, vLLM can be installed from nightly:
-
-```bash
-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
-vllm serve ./outputs/gpt-oss-out/ --served-model-name axolotl/gpt-oss-20b --host 0.0.0.0 --port 8888  --tensor-parallel-size 8
-```
-
-#### SGLang
-
-SGLang has 0-day support in main, see https://github.com/sgl-project/sglang/issues/8833 for infomation on installing
-SGLang from source. Once you've installed SGLang, run the following command to launch a SGLang server:
-
-```bash
-python3 -m sglang.launch_server --model ./outputs/gpt-oss-out/ --served-model-name axolotl/gpt-oss-120b --host 0.0.0.0 --port 8888 --tp 8
-```
-
 ### Tool use
 
 GPT-OSS has a comprehensive tool understanding. Axolotl supports tool calling datasets for Supervised Fine-tuning.
@@ -118,16 +55,6 @@ See [Nanobit/text-tools-2k-test](https://huggingface.co/datasets/Nanobit/text-to
 
 Refer to [our docs](https://docs.axolotl.ai/docs/dataset-formats/conversation.html#using-tool-use) for more info.
 
-### Thinking and chat_template masking conflict
-
-OpenAI’s Harmony template hides `thinking` in all non-final turns, which conflicts with Axolotl’s `chat_template` masking.
-
-If your dataset has `thinking` content mid-turn, there are two paths we recommend:
-
-- Train only on the last turn. This can be accomplished via chat_template's [train on last doc](https://docs.axolotl.ai/docs/dataset-formats/conversation.html#training-on-last-message).
-
-- Adjust your dataset to only have `thinking` content in the last turn.
-
 ### TIPS
 
 - Read more on how to load your own dataset at [docs](https://docs.axolotl.ai/docs/dataset_loading.html).

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

@@ -20,7 +20,6 @@ datasets:
 dataset_prepared_path: last_run_prepared
 val_set_size: 0
 output_dir: ./outputs/gpt-oss-out/
-save_total_limit: 2  # the 120B model can use up to 720GB of disk space per checkpoint, so let's only keep the last 2
 
 sequence_len: 4096
 sample_packing: true
@@ -32,10 +31,6 @@ 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
@@ -48,7 +43,7 @@ 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
+attn_implementation: kernels-community/vllm-flash-attn3
 
 gradient_checkpointing: true
 activation_offloading: true

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
@@ -44,7 +40,7 @@ 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
+attn_implementation: kernels-community/vllm-flash-attn3
 
 gradient_checkpointing: true
 activation_offloading: true

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

@@ -15,7 +15,7 @@ datasets:
     field_thinking: thinking
     template_thinking_key: thinking
 
-dataset_prepared_path: ./outputs/last_run_prepared
+dataset_prepared_path: last_run_prepared
 val_set_size: 0
 output_dir: ./outputs/gpt-oss-out/
 
@@ -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
@@ -45,7 +41,7 @@ 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
+attn_implementation: kernels-community/vllm-flash-attn3
 
 gradient_checkpointing: true
 activation_offloading: true

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

@@ -15,7 +15,7 @@ datasets:
     field_thinking: thinking
     template_thinking_key: thinking
 
-dataset_prepared_path: ./outputs/last_run_prepared
+dataset_prepared_path: last_run_prepared
 val_set_size: 0
 output_dir: ./outputs/gpt-oss-out/
 
@@ -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
@@ -44,7 +40,7 @@ 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
+attn_implementation: kernels-community/vllm-flash-attn3
 
 gradient_checkpointing: true
 activation_offloading: true

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
@@ -57,7 +53,7 @@ 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
+attn_implementation: kernels-community/vllm-flash-attn3
 
 gradient_checkpointing: true
 activation_offloading: true

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==23.2 setuptools==75.8.0 wheel ninja
-pip3 install --no-build-isolation -e '.[flash-attn]'
-
-# Install CCE https://docs.axolotl.ai/docs/custom_integrations.html#cut-cross-entropy
-python scripts/cutcrossentropy_install.py | sh
-```
-
-2. Run the finetuning example:
-
-```bash
-axolotl train examples/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

examples/hunyuan/README.md

@@ -1,85 +0,0 @@
-# Finetune HunYuan with Axolotl
-
-Tencent released a family of opensource models called HunYuan with varying parameter scales of 0.5B, 1.8B, 4B, and 7B scale for both Pre-trained and Instruct variants. The models can be found at [HuggingFace](https://huggingface.co/collections/tencent/hunyuan-dense-model-6890632cda26b19119c9c5e7). 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 HunYuan is only on nightly or use our latest [Docker images](https://docs.axolotl.ai/docs/docker.html).
-
-    Here is an example of how to install from main for pip:
-
-```bash
-# Ensure you have Pytorch installed (Pytorch 2.6.0 min)
-git clone https://github.com/axolotl-ai-cloud/axolotl.git
-cd axolotl
-
-pip3 install packaging==23.2 setuptools==75.8.0 wheel ninja
-pip3 install --no-build-isolation -e '.[flash-attn]'
-
-# Install CCE https://docs.axolotl.ai/docs/custom_integrations.html#cut-cross-entropy
-python scripts/cutcrossentropy_install.py | sh
-```
-
-2. Run the finetuning example:
-
-```bash
-axolotl train examples/hunyuan/hunyuan-v1-dense-qlora.yaml
-```
-
-This config uses about 4.7 GB VRAM.
-
-Let us know how it goes. Happy finetuning! 🚀
-
-### Dataset
-
-HunYuan Instruct models can choose to enter a slow think or fast think pattern. For best performance on fine-tuning their Instruct models, your dataset should be adjusted to match their pattern.
-
-```python
-# fast think pattern
-messages = [
-    {"role": "system", "content": "You are a helpful assistant."},
-    {"role": "user", "content": "/no_think What color is the sun?" },
-    {"role": "assistant", "content": "<think>\n\n</think>\n<answer>\nThe sun is yellow.\n</answer>"}
-]
-
-# slow think pattern
-messages = [
-    {"role": "system", "content": "You are a helpful assistant."},
-    {"role": "user", "content": "/no_think What color is the sun?" },
-    {"role": "assistant", "content": "<think>\nThe user is asking about the color of the sun. I need to ...\n</think>\n<answer>\nThe sun is yellow.\n</answer>"}
-]
-```
-
-### TIPS
-
-- For inference, the official Tencent team recommends
-
-```json
-
-{
-  "do_sample": true,
-  "top_k": 20,
-  "top_p": 0.8,
-  "repetition_penalty": 1.05,
-  "temperature": 0.7
-}
-
-```
-
-- You can run a full finetuning by removing the `adapter: qlora` and `load_in_4bit: true` from the config.
-- Read more on how to load your own dataset at [docs](https://docs.axolotl.ai/docs/dataset_loading.html).
-- The dataset format follows the OpenAI Messages format as seen [here](https://docs.axolotl.ai/docs/dataset-formats/conversation.html#chat_template).
-
-## Optimization Guides
-
-- [Multi-GPU Training](https://docs.axolotl.ai/docs/multi-gpu.html)
-- [Multi-Node Training](https://docs.axolotl.ai/docs/multi-node.html)
-- [LoRA Optimizations](https://docs.axolotl.ai/docs/lora_optims.html)
-
-## Related Resources
-
-- [Tencent HunYuan Blog](https://hunyuan.tencent.com/)
-- [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/hunyuan/hunyuan-v1-dense-qlora.yaml

@@ -1,64 +0,0 @@
-base_model: tencent/Hunyuan-0.5B-Instruct
-
-# Automatically upload checkpoint and final model to HF
-# hub_model_id: username/custom_model_name
-
-plugins:
-  - axolotl.integrations.cut_cross_entropy.CutCrossEntropyPlugin
-
-load_in_8bit: false
-load_in_4bit: true
-
-datasets:
-  - path: fozziethebeat/alpaca_messages_2k_test
-    type: chat_template
-
-dataset_prepared_path: last_run_prepared
-val_set_size: 0.1
-output_dir: ./outputs/lora-out
-
-adapter: qlora
-lora_model_dir:
-
-sequence_len: 2048
-sample_packing: true
-
-lora_r: 32
-lora_alpha: 16
-lora_dropout: 0.05
-lora_target_linear: true
-lora_target_modules:
-  - gate_proj
-  - down_proj
-  - up_proj
-  - q_proj
-  - v_proj
-  - k_proj
-  - o_proj
-
-wandb_project:
-wandb_entity:
-wandb_watch:
-wandb_name:
-wandb_log_model:
-
-gradient_accumulation_steps: 4
-micro_batch_size: 2
-num_epochs: 1
-optimizer: adamw_bnb_8bit
-lr_scheduler: cosine
-learning_rate: 0.0002
-
-bf16: auto
-tf32: false
-
-gradient_checkpointing: true
-resume_from_checkpoint:
-logging_steps: 1
-flash_attention: true
-
-warmup_ratio: 0.1
-evals_per_epoch: 1
-saves_per_epoch: 1
-
-# save_first_step: true  # uncomment this to validate checkpoint saving works with your config

examples/internvl3_5/README.md

@@ -1,43 +0,0 @@
-# Finetune OpenGV's InternVL with Axolotl
-
-[InternVL 3.5](https://huggingface.co/OpenGVLab/InternVL3_5-8B-HF) is a family of powerful vision-language models supporting dynamic resolution and multi-image understanding by OpenGV. It features a ViT-style vision encoder and strong language model backbone for tasks like visual question answering, OCR, and scene text understanding.
-
-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 `timm` for vision model support:
-
-    ```bash
-    pip install timm==1.0.19
-    ```
-
-3. Install [Cut Cross Entropy](https://docs.axolotl.ai/docs/custom_integrations.html#cut-cross-entropy) to reduce training VRAM usage.
-
-4. Run the finetuning example:
-
-    ```bash
-    axolotl train examples/internvl3_5/internvl3_5-8b-qlora.yml
-    ```
-
-This config uses about 8.21 GiB VRAM. Let us know how it goes. Happy finetuning! 🚀
-
-### Tips
-
-- You can run a full finetuning by removing the `adapter: qlora` and `load_in_4bit: true` from the config.
-- Read more on how to load your own dataset at [docs](https://docs.axolotl.ai/docs/dataset_loading.html).
-- The dataset format follows the multi-modal format as seen [here](https://docs.axolotl.ai/docs/multimodal.html#dataset-format).
-
-## Optimization Guides
-
-Please check the [Optimizations doc](https://docs.axolotl.ai/docs/optimizations.html).
-
-## Related Resources
-
-- [InternVL Paper](https://huggingface.co/papers/2508.18265)
-- [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/internvl3_5/internvl3_5-8b-qlora.yml

@@ -1,61 +0,0 @@
-base_model: OpenGVLab/InternVL3_5-8B-HF
-processor_type: AutoProcessor
-
-plugins:
-  - axolotl.integrations.cut_cross_entropy.CutCrossEntropyPlugin
-
-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
-
-datasets:
-  - path: HuggingFaceH4/llava-instruct-mix-vsft
-    type: chat_template
-    split: train[:1%]
-    field_messages: messages
-
-dataset_prepared_path: last_run_prepared
-val_set_size: 0.01
-output_dir: ./outputs/out
-
-adapter: qlora
-lora_model_dir:
-
-sequence_len: 2048
-
-lora_r: 32
-lora_alpha: 16
-lora_dropout: 0.05
-lora_target_modules: 'model.language_model.layers.[\d]+.(mlp|cross_attn|self_attn).(up|down|gate|q|k|v|o)_proj'
-
-wandb_project:
-wandb_entity:
-wandb_watch:
-wandb_name:
-wandb_log_model:
-
-gradient_accumulation_steps: 4
-micro_batch_size: 2
-num_epochs: 1
-optimizer: adamw_bnb_8bit
-lr_scheduler: cosine
-learning_rate: 0.0002
-
-bf16: true
-fp16:
-tf32: true
-
-gradient_checkpointing: true
-logging_steps: 1
-flash_attention: true
-eager_attention:
-
-warmup_ratio: 0.1
-evals_per_epoch: 1
-saves_per_epoch: 1
-weight_decay: 0.0
-
-# save_first_step: true  # uncomment this to validate checkpoint saving works with your config

examples/kimi-linear/README.md

@@ -1,47 +0,0 @@
-# Finetune MoonshotAI's Kimi Linear with Axolotl
-
-[Kimi Linear](https://huggingface.co/collections/moonshotai/kimi-linear-a3b) is a MoE model (48B total, 3B active) by MoonshotAI using a hybrid linear attention architecture to achieve a 1M token context length. It uses Kimi Delta Attention (KDA), a refined version of Gated DeltaNet that reduces KV cache size by up to 75% and boosts decoding throughput by up to 6x for long contexts.
-
-This guide shows how to fine-tune it with Axolotl with multi-turn conversations and proper masking.
-
-**Note:** Axolotl uses experimental training code for Kimi Linear as their original modeling code is inference-only.
-
-## Getting started
-
-1. Install Axolotl following the [installation guide](https://docs.axolotl.ai/docs/installation.html).
-
-2. Install CCE via [docs](https://docs.axolotl.ai/docs/custom_integrations.html#cut-cross-entropy)
-
-3. Run the finetuning example:
-
-    ```bash
-    axolotl train examples/kimi-linear/kimi-48b-lora.yaml
-    ```
-
-This config uses about 98.7GiB VRAM.
-
-Let us know how it goes. Happy finetuning!
-
-### TIPS
-
-- Kimi Linear requires `trust_remote_code: true`.
-- You can run a full finetuning by removing the `adapter: lora` and `load_in_8bit: true`.
-- Read more on how to load your own dataset at [docs](https://docs.axolotl.ai/docs/dataset_loading.html)
-- The dataset format follows the OpenAI Messages format as seen [here](https://docs.axolotl.ai/docs/dataset-formats/conversation.html#chat_template)
-
-## Optimization Guides
-
-See 👉 [docs](https://docs.axolotl.ai/docs/optimizations.html).
-
-## Limitations
-
-This is not yet compatible with MoE kernels from transformers v5.
-
-## Related Resources
-
-- [Kimi Linear Paper](https://huggingface.co/papers/2510.26692)
-- [Kimi Linear GitHub](https://github.com/MoonshotAI/Kimi-Linear)
-- [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/kimi-linear/kimi-48b-lora.yaml

@@ -1,81 +0,0 @@
-base_model: moonshotai/Kimi-Linear-48B-A3B-Instruct
-
-# Automatically upload checkpoint and final model to HF
-# hub_model_id: username/custom_model_name
-
-trust_remote_code: true
-
-plugins:
-  - axolotl.integrations.cut_cross_entropy.CutCrossEntropyPlugin
-
-load_in_8bit: true
-load_in_4bit: false
-strict: false
-
-datasets:
-  - path: fozziethebeat/alpaca_messages_2k_test
-    type: chat_template
-    split: train
-
-dataset_prepared_path: last_run_prepared
-val_set_size: 0.2
-output_dir: ./outputs/lora-out
-
-adapter: lora
-lora_model_dir:
-
-sequence_len: 2048
-sample_packing: true
-pad_to_sequence_len: true
-
-lora_r: 16
-lora_alpha: 32
-lora_dropout: 0.05
-lora_fan_in_fan_out:
-lora_target_modules:
-  - gate_proj
-  - down_proj
-  - up_proj
-  - q_proj
-  - v_proj
-  - k_proj
-  - o_proj
-
-wandb_project:
-wandb_entity:
-wandb_watch:
-wandb_name:
-wandb_log_model:
-
-gradient_accumulation_steps: 2
-micro_batch_size: 2
-num_epochs: 1
-optimizer: adamw_8bit
-lr_scheduler: cosine
-learning_rate: 0.0002
-
-train_on_inputs: false
-group_by_length: false
-bf16: auto
-fp16:
-tf32: false
-
-gradient_checkpointing: true
-early_stopping_patience:
-resume_from_checkpoint:
-local_rank:
-logging_steps: 1
-flash_attention: true
-
-loss_watchdog_threshold: 5.0
-loss_watchdog_patience: 3
-
-warmup_ratio: 0.1
-evals_per_epoch: 2
-saves_per_epoch: 1
-debug:
-deepspeed:
-weight_decay: 0.0
-fsdp:
-fsdp_config:
-special_tokens:

examples/lfm2/README.md

@@ -0,0 +1,7 @@
+# Liquid Foundation Models 2
+
+LFM2 support in transformers exists in the main branch, but is not yet included in the transformers release.
+
+```bash
+pip install --upgrade --no-deps --force-reinstall git+https://github.com/huggingface/transformers.git
+```

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

@@ -1,8 +1,8 @@
 base_model: LiquidAI/LFM2-350M
 
-plugins:
-  - axolotl.integrations.cut_cross_entropy.CutCrossEntropyPlugin
+chunked_cross_entropy: true
 
+chat_template: tokenizer_default
 eot_tokens:
   - "<|im_end|>"
 datasets:

examples/llama-2/qlora-fsdp.yml

@@ -66,7 +66,6 @@ fsdp_config:
   fsdp_auto_wrap_policy: TRANSFORMER_BASED_WRAP
   fsdp_transformer_layer_cls_to_wrap: LlamaDecoderLayer
   fsdp_state_dict_type: FULL_STATE_DICT
-  # fsdp_cpu_offload_pin_memory: false  # uncomment to enable swap memory usage when RAM is insufficient
 special_tokens:
 
 # save_first_step: true  # uncomment this to validate checkpoint saving works with your config

examples/llama-3/3b-fp8-fsdp2.yaml

@@ -29,6 +29,7 @@ flex_attention: true
 flex_attn_compile_kwargs:
   dynamic: false
   mode: max-autotune-no-cudagraphs
+
 torch_compile: true
 
 wandb_project:

examples/llama-3/3b-qat-fsdp2.yaml

@@ -15,18 +15,20 @@ liger_glu_activation: true
 liger_layer_norm: true
 liger_fused_linear_cross_entropy: true
 
-
 datasets:
   - path: yahma/alpaca-cleaned
     type: alpaca
-    split: train[:95%]
 
 output_dir: ./outputs/qat_out/
-dataset_prepared_path: ./outputs/qat_out/dataset_prepared
 
-sample_packing: false
-sequence_len: 8192
-flash_attention: true
+sample_packing: true
+
+sequence_len: 512
+
+flex_attention: true
+flex_attn_compile_kwargs:
+  dynamic: false
+  mode: max-autotune-no-cudagraphs
 
 qat:
   activation_dtype: int8
@@ -65,7 +67,7 @@ fsdp:
 fsdp_config:
   fsdp_version: 2
   fsdp_offload_params: false
-  fsdp_cpu_ram_efficient_loading: false
+  fsdp_cpu_ram_efficient_loading: true
   fsdp_auto_wrap_policy: TRANSFORMER_BASED_WRAP
   fsdp_transformer_layer_cls_to_wrap: LlamaDecoderLayer
   fsdp_state_dict_type: FULL_STATE_DICT
@@ -74,6 +76,6 @@ fsdp_config:
   fsdp_activation_checkpointing: true
 
 special_tokens:
-  pad_token: <|finetune_right_pad_id|>
+  pad_token: <|end_of_text|>
 
 # save_first_step: true  # uncomment this to validate checkpoint saving works with your config

examples/llama-3/3b-qat-nvfp4.yaml

@@ -1,64 +0,0 @@
-base_model: meta-llama/Llama-3.2-3B
-# Automatically upload checkpoint and final model to HF
-# hub_model_id: username/custom_model_name
-
-load_in_8bit: false
-load_in_4bit: false
-strict: false
-
-plugins:
-  - axolotl.integrations.liger.LigerPlugin
-
-liger_rope: true
-liger_rms_norm: true
-liger_glu_activation: true
-liger_layer_norm: true
-liger_fused_linear_cross_entropy: true
-
-datasets:
-  - path: yahma/alpaca-cleaned
-    type: alpaca
-    split: train[:95%]
-
-output_dir: ./outputs/qat_out/
-dataset_prepared_path: ./outputs/dataset_prepared
-
-sequence_len: 8192
-flash_attention: true
-
-qat:
-  activation_dtype: nvfp4
-  weight_dtype: nvfp4
-  group_size: 16 # only group_size of 16 is supported with nvfp4
-
-wandb_project:
-wandb_entity:
-wandb_watch:
-wandb_name:
-wandb_log_model:
-
-gradient_checkpointing: true
-gradient_accumulation_steps: 1
-micro_batch_size: 64
-num_epochs: 1
-optimizer: adamw_torch_fused
-
-cosine_constant_lr_ratio: 0
-cosine_min_lr_ratio: 1.0
-learning_rate: 2e-5
-save_only_model: true
-bf16: true
-
-resume_from_checkpoint:
-logging_steps: 1
-
-evals_per_epoch: 1
-saves_per_epoch: 1
-
-warmup_ratio: 0.1
-weight_decay: 0.0
-
-special_tokens:
-  pad_token: <|finetune_right_pad_id|>
-
-# save_first_step: true  # uncomment this to validate checkpoint saving works with your config

examples/llama-3/diffusion/pretrain-1b.yaml

@@ -1,56 +0,0 @@
-base_model: meta-llama/Llama-3.2-1B
-# Automatically upload checkpoint and final model to HF
-# hub_model_id: username/custom_model_name
-
-pretraining_dataset:
-  - path: wikitext
-    name: wikitext-103-raw-v1
-    type: completion
-    field: text
-
-plugins:
-  - axolotl.integrations.diffusion.DiffusionPlugin
-
-diffusion:
-  noise_schedule: cosine
-  min_mask_ratio: 0.15
-  max_mask_ratio: 0.85
-  num_diffusion_steps: 128
-  eps: 5e-4
-  importance_weighting: true
-  mask_token_id: 128002
-  generate_samples: true
-  generation_interval: 250
-
-output_dir: ./outputs/model-out
-
-sequence_len: 512
-sample_packing: true
-
-gradient_accumulation_steps: 8
-micro_batch_size: 4
-max_steps: 10000
-warmup_ratio: 0.1
-
-optimizer: adamw_8bit
-lr_scheduler: cosine
-learning_rate: 3e-4
-sdp_attention: true
-
-bf16: auto
-tf32: true
-
-logging_steps: 1
-save_strategy: steps
-save_steps: 1000
-
-special_tokens:
-  pad_token: "<|end_of_text|>"
-
-wandb_project:
-wandb_entity:
-wandb_watch:
-wandb_name:
-wandb_log_model:
-
-# save_first_step: true  # uncomment this to validate checkpoint saving works with your config

examples/llama-3/diffusion/sft-1b.yaml

@@ -1,59 +0,0 @@
-base_model: meta-llama/Llama-3.2-1B
-# Automatically upload checkpoint and final model to HF
-# hub_model_id: username/custom_model_name
-
-datasets:
-  - path: teknium/GPT4-LLM-Cleaned
-    type: alpaca
-val_set_size: 0.05
-
-plugins:
-  - axolotl.integrations.diffusion.DiffusionPlugin
-
-diffusion:
-  noise_schedule: cosine
-  min_mask_ratio: 0.1
-  max_mask_ratio: 0.9
-  num_diffusion_steps: 128
-  eps: 1e-3
-  importance_weighting: true
-  mask_token_id: 128002
-  generate_samples: true
-  generation_interval: 250
-
-output_dir: ./outputs/model-out
-
-sequence_len: 512
-sample_packing: true
-eval_sample_packing: true
-
-gradient_accumulation_steps: 4
-micro_batch_size: 4
-num_epochs: 1
-warmup_steps: 0.1
-
-optimizer: adamw_8bit
-lr_scheduler: cosine
-learning_rate: 1e-5
-
-bf16: auto
-tf32: true
-
-gradient_checkpointing: true
-resume_from_checkpoint:
-sdp_attention: true
-
-logging_steps: 1
-save_strategy: best
-eval_strategy: epoch
-
-special_tokens:
-  pad_token: "<|end_of_text|>"
-
-wandb_project:
-wandb_entity:
-wandb_watch:
-wandb_name:
-wandb_log_model:
-
-# save_first_step: true  # uncomment this to validate checkpoint saving works with your config

examples/llama-3/instruct-lora-8b.yml

@@ -12,6 +12,15 @@ chat_template: llama3
 datasets:
   - path: fozziethebeat/alpaca_messages_2k_test
     type: chat_template
+    field_messages: messages
+    message_property_mappings:
+      role: role
+      content: content
+    roles:
+      user:
+        - user
+      assistant:
+        - assistant
 
 dataset_prepared_path:
 val_set_size: 0.05

examples/llama-3/opentelemetry-qlora.yml

@@ -1,50 +0,0 @@
-base_model: NousResearch/Llama-3.2-1B
-model_type: AutoModelForCausalLM
-tokenizer_type: AutoTokenizer
-
-load_in_4bit: true
-
-datasets:
-  - path: mhenrichsen/alpaca_2k_test
-    type: alpaca
-
-output_dir: ./outputs/opentelemetry-example
-
-adapter: qlora
-sequence_len: 512
-sample_packing: false
-
-lora_r: 32
-lora_alpha: 16
-lora_dropout: 0.05
-lora_target_linear: true
-
-# OpenTelemetry Configuration
-use_otel_metrics: true
-otel_metrics_host: "localhost"
-otel_metrics_port: 8000
-
-# Disable WandB
-use_wandb: false
-
-gradient_accumulation_steps: 4
-micro_batch_size: 2
-num_epochs: 1
-optimizer: paged_adamw_32bit
-lr_scheduler: cosine
-learning_rate: 0.0002
-
-bf16: auto
-tf32: false
-
-gradient_checkpointing: true
-logging_steps: 1
-flash_attention: false
-
-warmup_ratio: 0.1
-evals_per_epoch: 2
-saves_per_epoch: 1
-weight_decay: 0.0
-
-special_tokens:
-  pad_token: "<|end_of_text|>"

examples/llama-4/do-no-use-fa2/scout-vision-qlora-fsdp.yaml

@@ -46,6 +46,7 @@ datasets:
   - path: HuggingFaceH4/llava-instruct-mix-vsft
     type: chat_template
     split: train[:1%]
+    field_messages: messages
 
 dataset_prepared_path: last_run_prepared
 val_set_size: 0.0

examples/llama-4/scout-vision-qlora-fsdp2-flex.yaml

@@ -45,6 +45,7 @@ datasets:
   - path: HuggingFaceH4/llava-instruct-mix-vsft
     type: chat_template
     split: train[:1%]
+    field_messages: messages
 
 dataset_prepared_path: last_run_prepared
 val_set_size: 0.0