add missing file

- Kernel: fused_rms_norm_rope crashed when cos.shape[-1] < x.shape[-1].
  Triton forward/backward take an n_rot runtime arg that restricts
  rotate_half to [0, n_rot) and treats trailing cols as RMSNorm-only
  pass-through (cos=1, sin=0 defaults). Wrapper also expands cos/sin
  that broadcast over batch.

- Forward: _make_fused_forward used a stale shared_kv_states kwarg the
  current decoder layer no longer passes. Now mirrors stock attention,
  reading/writing past_key_values.shared_layers.

.github/CONTRIBUTING.md

@@ -68,7 +68,12 @@ You can skip certain CI checks by including specific keywords in your commit mes
 
 ### Code Style
 
-axolotl uses [{codestyle}]({URLofCodestyle}) as its code style guide. Please ensure that your code follows these guidelines.
+axolotl uses [Ruff](https://docs.astral.sh/ruff/) as its code style guide. Please ensure that your code follows these guidelines.
+
+Use the pre-commit linter to ensure that your code is formatted consistently.
+```bash
+pre-commit run --all-files
+```
 
 ### Commit Messages
 
@@ -78,6 +83,6 @@ Write clear and concise commit messages that briefly describe the changes made i
 
 - [GitHub Help](https://help.github.com/)
 - [GitHub Pull Request Documentation](https://docs.github.com/en/github/collaborating-with-issues-and-pull-requests)
-- [{codestyle}]({URLofCodestyle})
+- [Ruff](https://docs.astral.sh/ruff/)
 
 Thank you once again for your interest in contributing to axolotl. We look forward to collaborating with you and creating an even better project together!

.github/FUNDING.yml

@@ -1,13 +1,13 @@
 # These are supported funding model platforms
 
-github: [winglian, OpenAccess-AI-Collective] # Replace with up to 4 GitHub Sponsors-enabled usernames e.g., [user1, user2]
+github: # 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: axolotl_ai # Replace with a single Ko-fi username
+ko_fi: # 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: ['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']
+custom: # Replace with up to 4 custom sponsorship URLs e.g., ['link1', 'link2']

.github/PULL_REQUEST_TEMPLATE.md

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

.github/workflows/base.yml

@@ -15,58 +15,85 @@ on:
       - '.github/workflows/base.yml'
   workflow_dispatch:
 
+permissions:
+  contents: read
+
 jobs:
   build-base:
     if: ${{ github.repository_owner == 'axolotl-ai-cloud' && (github.event_name != 'pull_request' || !github.event.pull_request.draft) }}
     timeout-minutes: 480
     # this job needs to be run on self-hosted GPU runners...
     runs-on: ubuntu-latest-m
+    env:
+      HAS_DOCKERHUB_CREDS: ${{ secrets.DOCKERHUB_USERNAME != '' && secrets.DOCKERHUB_TOKEN != '' }}
     strategy:
       fail-fast: false
       matrix:
         include:
-          - cuda: "124"
-            cuda_version: 12.4.1
-            cudnn_version: ""
-            python_version: "3.11"
-            pytorch: 2.6.0
-            torch_cuda_arch_list: "7.0 7.5 8.0 8.6 8.7 8.9 9.0+PTX"
-            dockerfile: "Dockerfile-base"
-          - cuda: "126"
-            cuda_version: 12.6.3
-            cudnn_version: ""
-            python_version: "3.11"
-            pytorch: 2.6.0
-            torch_cuda_arch_list: "7.0 7.5 8.0 8.6 8.7 8.9 9.0+PTX"
-            dockerfile: "Dockerfile-base"
-          - cuda: "126"
-            cuda_version: 12.6.3
-            cudnn_version: ""
-            python_version: "3.11"
-            pytorch: 2.7.0
-            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
+            pytorch: 2.9.0
             torch_cuda_arch_list: "7.0 7.5 8.0 8.6 8.7 8.9 9.0+PTX"
             dockerfile: "Dockerfile-base"
+            platforms: "linux/amd64,linux/arm64"
           - cuda: "128"
             cuda_version: 12.8.1
             cudnn_version: ""
             python_version: "3.11"
-            pytorch: 2.8.0
+            pytorch: 2.9.1
             torch_cuda_arch_list: "7.0 7.5 8.0 8.6 8.7 8.9 9.0+PTX"
             dockerfile: "Dockerfile-base"
+            platforms: "linux/amd64,linux/arm64"
+          - cuda: "128"
+            cuda_version: 12.8.1
+            cudnn_version: ""
+            python_version: "3.11"
+            pytorch: 2.10.0
+            torch_cuda_arch_list: "7.0 7.5 8.0 8.6 8.7 8.9 9.0+PTX"
+            dockerfile: "Dockerfile-base"
+            platforms: "linux/amd64,linux/arm64"
+          - cuda: "128"
+            cuda_version: 12.8.1
+            cudnn_version: ""
+            python_version: "3.12"
+            pytorch: 2.10.0
+            torch_cuda_arch_list: "7.0 7.5 8.0 8.6 8.7 8.9 9.0+PTX"
+            dockerfile: "Dockerfile-base"
+            platforms: "linux/amd64,linux/arm64"
+#          - cuda: "129"
+#            cuda_version: 12.9.1
+#            cudnn_version: ""
+#            python_version: "3.12"
+#            pytorch: 2.9.1
+#            torch_cuda_arch_list: "7.0 7.5 8.0 8.6 8.7 8.9 9.0+PTX"
+#            dockerfile: "Dockerfile-base"
+#            platforms: "linux/amd64,linux/arm64"
+          - cuda: "130"
+            cuda_version: 13.0.0
+            cudnn_version: ""
+            python_version: "3.11"
+            pytorch: 2.9.1
+            torch_cuda_arch_list: "9.0+PTX"
+            dockerfile: "Dockerfile-base"
+            platforms: "linux/amd64,linux/arm64"
+          - cuda: "130"
+            cuda_version: 13.0.0
+            cudnn_version: ""
+            python_version: "3.12"
+            pytorch: 2.9.1
+            torch_cuda_arch_list: "9.0+PTX"
+            dockerfile: "Dockerfile-base"
+            platforms: "linux/amd64,linux/arm64"
+          - cuda: "130"
+            cuda_version: 13.0.0
+            cudnn_version: ""
+            python_version: "3.12"
+            pytorch: 2.10.0
+            torch_cuda_arch_list: "9.0+PTX"
+            dockerfile: "Dockerfile-base"
+            platforms: "linux/amd64,linux/arm64"
 #          - cuda: "128"
 #            cuda_version: 12.8.1
 #            cudnn_version: ""
@@ -90,20 +117,21 @@ jobs:
         uses: docker/metadata-action@v5
         with:
           images: |
-            winglian/axolotl-base
             axolotlai/axolotl-base
       - name: Login to Docker Hub
-        uses: docker/login-action@v2
+        uses: docker/login-action@v3
+        if: ${{ github.event_name != 'pull_request' && env.HAS_DOCKERHUB_CREDS == 'true' }}
         with:
           username: ${{ secrets.DOCKERHUB_USERNAME }}
           password: ${{ secrets.DOCKERHUB_TOKEN }}
       - name: Set up Docker Buildx
         uses: docker/setup-buildx-action@v3
       - name: Build
-        uses: docker/build-push-action@v4
+        uses: docker/build-push-action@v5
         with:
           context: .
           file: ./docker/${{ matrix.dockerfile }}
+          platforms: ${{ matrix.platforms }}
           push: ${{ github.event_name != 'pull_request' }}
           tags: ${{ steps.metadata.outputs.tags }}-base-py${{ matrix.python_version }}-cu${{ matrix.cuda }}-${{ matrix.pytorch }}${{ matrix.axolotl_extras != '' && '-' || '' }}${{ matrix.axolotl_extras }}
           labels: ${{ steps.metadata.outputs.labels }}
@@ -118,38 +146,84 @@ jobs:
     if: ${{ github.repository_owner == 'axolotl-ai-cloud' && (github.event_name != 'pull_request' || !github.event.pull_request.draft) }}
     timeout-minutes: 480
     runs-on: ubuntu-latest-m
+    env:
+      HAS_DOCKERHUB_CREDS: ${{ secrets.DOCKERHUB_USERNAME != '' && secrets.DOCKERHUB_TOKEN != '' }}
     strategy:
       fail-fast: false
       matrix:
         include:
-          - cuda: "126"
-            cuda_version: 12.6.3
-            cudnn_version: ""
-            python_version: "3.11"
-            pytorch: 2.6.0
-            torch_cuda_arch_list: "7.0 7.5 8.0 8.6 8.7 8.9 9.0+PTX"
-            dockerfile: "Dockerfile-uv-base"
-          - cuda: "126"
-            cuda_version: 12.6.3
-            cudnn_version: ""
-            python_version: "3.11"
-            pytorch: 2.7.1
-            torch_cuda_arch_list: "7.0 7.5 8.0 8.6 8.7 8.9 9.0+PTX"
-            dockerfile: "Dockerfile-uv-base"
           - cuda: "128"
             cuda_version: 12.8.1
             cudnn_version: ""
             python_version: "3.11"
-            pytorch: 2.7.1
+            pytorch: 2.9.1
             torch_cuda_arch_list: "7.0 7.5 8.0 8.6 8.7 8.9 9.0+PTX"
             dockerfile: "Dockerfile-uv-base"
+            platforms: "linux/amd64,linux/arm64"
+          - cuda: "128"
+            cuda_version: 12.8.1
+            cudnn_version: ""
+            python_version: "3.12"
+            pytorch: 2.9.1
+            torch_cuda_arch_list: "7.0 7.5 8.0 8.6 8.7 8.9 9.0+PTX"
+            dockerfile: "Dockerfile-uv-base"
+            platforms: "linux/amd64,linux/arm64"
           - cuda: "128"
             cuda_version: 12.8.1
             cudnn_version: ""
             python_version: "3.11"
-            pytorch: 2.8.0
+            pytorch: 2.9.0
             torch_cuda_arch_list: "7.0 7.5 8.0 8.6 8.7 8.9 9.0+PTX"
             dockerfile: "Dockerfile-uv-base"
+            platforms: "linux/amd64,linux/arm64"
+          - cuda: "128"
+            cuda_version: 12.8.1
+            cudnn_version: ""
+            python_version: "3.11"
+            pytorch: 2.10.0
+            torch_cuda_arch_list: "7.0 7.5 8.0 8.6 8.7 8.9 9.0+PTX"
+            dockerfile: "Dockerfile-uv-base"
+            platforms: "linux/amd64,linux/arm64"
+          - cuda: "128"
+            cuda_version: 12.8.1
+            cudnn_version: ""
+            python_version: "3.12"
+            pytorch: 2.10.0
+            torch_cuda_arch_list: "7.0 7.5 8.0 8.6 8.7 8.9 9.0+PTX"
+            dockerfile: "Dockerfile-uv-base"
+            platforms: "linux/amd64,linux/arm64"
+#          - cuda: "129"
+#            cuda_version: 12.9.1
+#            cudnn_version: ""
+#            python_version: "3.12"
+#            pytorch: 2.9.1
+#            torch_cuda_arch_list: "7.0 7.5 8.0 8.6 8.7 8.9 9.0+PTX"
+#            dockerfile: "Dockerfile-uv-base"
+#            platforms: "linux/amd64,linux/arm64"
+          - cuda: "130"
+            cuda_version: 13.0.0
+            cudnn_version: ""
+            python_version: "3.11"
+            pytorch: 2.9.1
+            torch_cuda_arch_list: "9.0+PTX"
+            dockerfile: "Dockerfile-uv-base"
+            platforms: "linux/amd64,linux/arm64"
+          - cuda: "130"
+            cuda_version: 13.0.0
+            cudnn_version: ""
+            python_version: "3.12"
+            pytorch: 2.9.1
+            torch_cuda_arch_list: "9.0+PTX"
+            dockerfile: "Dockerfile-uv-base"
+            platforms: "linux/amd64,linux/arm64"
+          - cuda: "130"
+            cuda_version: 13.0.0
+            cudnn_version: ""
+            python_version: "3.12"
+            pytorch: 2.10.0
+            torch_cuda_arch_list: "9.0+PTX"
+            dockerfile: "Dockerfile-uv-base"
+            platforms: "linux/amd64,linux/arm64"
     steps:
       - name: Checkout
         uses: actions/checkout@v4
@@ -160,17 +234,19 @@ jobs:
           images: |
             axolotlai/axolotl-base-uv
       - name: Login to Docker Hub
-        uses: docker/login-action@v2
+        uses: docker/login-action@v3
+        if: ${{ github.event_name != 'pull_request' && env.HAS_DOCKERHUB_CREDS == 'true' }}
         with:
           username: ${{ secrets.DOCKERHUB_USERNAME }}
           password: ${{ secrets.DOCKERHUB_TOKEN }}
       - name: Set up Docker Buildx
         uses: docker/setup-buildx-action@v3
       - name: Build
-        uses: docker/build-push-action@v4
+        uses: docker/build-push-action@v5
         with:
           context: .
           file: ./docker/${{ matrix.dockerfile }}
+          platforms: ${{ matrix.platforms }}
           push: ${{ github.event_name != 'pull_request' }}
           tags: ${{ steps.metadata.outputs.tags }}-base-py${{ matrix.python_version }}-cu${{ matrix.cuda }}-${{ matrix.pytorch }}${{ matrix.axolotl_extras != '' && '-' || '' }}${{ matrix.axolotl_extras }}
           labels: ${{ steps.metadata.outputs.labels }}

.github/workflows/docs.yml

@@ -12,6 +12,9 @@ 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/lint.yml

@@ -13,6 +13,9 @@ on:
        - ".pre-commit-config.yaml"
   workflow_dispatch:
 
+permissions:
+  contents: read
+
 jobs:
   pre-commit:
     name: pre-commit

.github/workflows/main.yml

@@ -8,6 +8,9 @@ on:
       - "v*"
   workflow_dispatch:
 
+permissions:
+  contents: read
+
 jobs:
   build-axolotl:
     if: ${{ ! contains(github.event.commits[0].message, '[skip docker]') && github.repository_owner == 'axolotl-ai-cloud' }}
@@ -15,32 +18,43 @@ jobs:
       fail-fast: false
       matrix:
         include:
-          - cuda: 126
-            cuda_version: 12.6.3
+          - cuda: 128
+            cuda_version: 12.8.1
             python_version: "3.11"
-            pytorch: 2.6.0
+            pytorch: 2.9.0
             axolotl_extras:
-          - cuda: 126
-            cuda_version: 12.6.3
+            platforms: "linux/amd64,linux/arm64"
+          - cuda: 128
+            cuda_version: 12.8.1
             python_version: "3.11"
-            pytorch: 2.7.0
+            pytorch: 2.9.1
             axolotl_extras:
-          - cuda: 126
-            cuda_version: 12.6.3
-            python_version: "3.11"
-            pytorch: 2.7.1
-            axolotl_extras: vllm
+            platforms: "linux/amd64,linux/arm64"
             is_latest: true
           - cuda: 128
             cuda_version: 12.8.1
-            python_version: "3.11"
-            pytorch: 2.7.1
+            python_version: "3.12"
+            pytorch: 2.10.0
             axolotl_extras:
-          - cuda: 128
-            cuda_version: 12.8.1
+            platforms: "linux/amd64,linux/arm64"
+#          - cuda: 129
+#            cuda_version: 12.9.1
+#            python_version: "3.12"
+#            pytorch: 2.9.1
+#            axolotl_extras:
+#            platforms: "linux/amd64,linux/arm64"
+          - cuda: 130
+            cuda_version: 13.0.0
             python_version: "3.11"
-            pytorch: 2.8.0
+            pytorch: 2.9.1
             axolotl_extras:
+            platforms: "linux/amd64,linux/arm64"
+          - cuda: 130
+            cuda_version: 13.0.0
+            python_version: "3.12"
+            pytorch: 2.10.0
+            axolotl_extras:
+            platforms: "linux/amd64,linux/arm64"
     runs-on: axolotl-gpu-runner
     steps:
       - name: Checkout
@@ -50,7 +64,6 @@ jobs:
         uses: docker/metadata-action@v5
         with:
           images: |
-            winglian/axolotl
             axolotlai/axolotl
           tags: |
             type=ref,event=branch
@@ -67,6 +80,7 @@ jobs:
         uses: docker/build-push-action@v5
         with:
           context: .
+          platforms: ${{ matrix.platforms }}
           build-args: |
             BASE_TAG=${{ github.ref_type == 'tag' && 'main' || github.ref_name }}-base-py${{ matrix.python_version }}-cu${{ matrix.cuda }}-${{ matrix.pytorch }}
             CUDA=${{ matrix.cuda }}
@@ -81,45 +95,128 @@ jobs:
             ${{ (matrix.is_latest) && format('{0}-latest', steps.metadata.outputs.tags) || '' }}
           labels: ${{ steps.metadata.outputs.labels }}
 
-  build-axolotl-cloud:
-    needs: build-axolotl
+  build-axolotl-uv:
     if: ${{ ! contains(github.event.commits[0].message, '[skip docker]') && github.repository_owner == 'axolotl-ai-cloud' }}
-    # this job needs to be run on self-hosted GPU runners...
     strategy:
+      fail-fast: false
       matrix:
         include:
-          - cuda: 126
-            cuda_version: 12.6.3
+          - cuda: 128
+            cuda_version: 12.8.1
             python_version: "3.11"
-            pytorch: 2.6.0
+            pytorch: 2.9.1
             axolotl_extras:
-          - cuda: 126
-            cuda_version: 12.6.3
-            python_version: "3.11"
-            pytorch: 2.7.0
+            platforms: "linux/amd64,linux/arm64"
+          - cuda: 128
+            cuda_version: 12.8.1
+            python_version: "3.12"
+            pytorch: 2.9.1
             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
+            platforms: "linux/amd64,linux/arm64"
             is_latest: true
           - cuda: 128
             cuda_version: 12.8.1
-            python_version: "3.11"
-            pytorch: 2.7.1
+            python_version: "3.12"
+            pytorch: 2.10.0
             axolotl_extras:
-          - cuda: 128
-            cuda_version: 12.8.1
+            platforms: "linux/amd64,linux/arm64"
+          - cuda: 130
+            cuda_version: 13.0.0
             python_version: "3.11"
-            pytorch: 2.8.0
+            pytorch: 2.9.1
             axolotl_extras:
+            platforms: "linux/amd64,linux/arm64"
+          - cuda: 130
+            cuda_version: 13.0.0
+            python_version: "3.12"
+            pytorch: 2.10.0
+            axolotl_extras:
+            platforms: "linux/amd64,linux/arm64"
+    runs-on: axolotl-gpu-runner
+    steps:
+      - name: Checkout
+        uses: actions/checkout@v4
+      - name: Docker metadata
+        id: metadata
+        uses: docker/metadata-action@v5
+        with:
+          images: |
+            axolotlai/axolotl-uv
+          tags: |
+            type=ref,event=branch
+            type=pep440,pattern={{version}}
+      - name: Set up Docker Buildx
+        uses: docker/setup-buildx-action@v3
+      - name: Login to Docker Hub
+        uses: docker/login-action@v3
+        with:
+          username: ${{ secrets.DOCKERHUB_USERNAME }}
+          password: ${{ secrets.DOCKERHUB_TOKEN }}
+      # guidance for testing before pushing: https://docs.docker.com/build/ci/github-actions/test-before-push/
+      - name: Build and export to Docker
+        uses: docker/build-push-action@v5
+        with:
+          context: .
+          platforms: ${{ matrix.platforms }}
+          build-args: |
+            BASE_TAG=${{ github.ref_type == 'tag' && 'main' || github.ref_name }}-base-py${{ matrix.python_version }}-cu${{ matrix.cuda }}-${{ matrix.pytorch }}
+            CUDA=${{ matrix.cuda }}
+            PYTORCH_VERSION=${{ matrix.pytorch }}
+            AXOLOTL_ARGS=${{ matrix.axolotl_args }}
+            AXOLOTL_EXTRAS=${{ matrix.axolotl_extras}}
+          file: ./docker/Dockerfile-uv
+          push: ${{ github.event_name != 'pull_request' }}
+          tags: |
+            ${{ steps.metadata.outputs.tags }}-py${{ matrix.python_version }}-cu${{ matrix.cuda }}-${{ matrix.pytorch }}${{ matrix.axolotl_extras != '' && '-' || '' }}${{ matrix.axolotl_extras }}
+            ${{ steps.metadata.outputs.tags }}-py${{ matrix.python_version }}-cu${{ matrix.cuda }}-${{ matrix.pytorch }}
+            ${{ (matrix.is_latest) && format('{0}-latest', steps.metadata.outputs.tags) || '' }}
+          labels: ${{ steps.metadata.outputs.labels }}
+
+  build-axolotl-cloud:
+    needs: build-axolotl
+    if: ${{ ! contains(github.event.commits[0].message, '[skip docker]') && github.repository_owner == 'axolotl-ai-cloud' }}
+    # this job needs to be run on self-hosted GPU runners...
+    strategy:
+      fail-fast: false
+      matrix:
+        include:
+          - cuda: 128
+            cuda_version: 12.8.1
+            python_version: "3.11"
+            pytorch: 2.9.0
+            axolotl_extras:
+            platforms: "linux/amd64,linux/arm64"
+          - cuda: 128
+            cuda_version: 12.8.1
+            python_version: "3.11"
+            pytorch: 2.9.1
+            axolotl_extras:
+            is_latest: true
+            platforms: "linux/amd64,linux/arm64"
+          - cuda: 128
+            cuda_version: 12.8.1
+            python_version: "3.12"
+            pytorch: 2.10.0
+            axolotl_extras:
+            platforms: "linux/amd64,linux/arm64"
+#          - cuda: 129
+#            cuda_version: 12.9.1
+#            python_version: "3.12"
+#            pytorch: 2.9.1
+#            axolotl_extras:
+#            platforms: "linux/amd64,linux/arm64"
+          - cuda: 130
+            cuda_version: 13.0.0
+            python_version: "3.11"
+            pytorch: 2.9.1
+            axolotl_extras:
+            platforms: "linux/amd64,linux/arm64"
+          - cuda: 130
+            cuda_version: 13.0.0
+            python_version: "3.12"
+            pytorch: 2.10.0
+            axolotl_extras:
+            platforms: "linux/amd64,linux/arm64"
     runs-on: axolotl-gpu-runner
     steps:
       - name: Checkout
@@ -129,7 +226,6 @@ jobs:
         uses: docker/metadata-action@v5
         with:
           images: |
-            winglian/axolotl-cloud
             axolotlai/axolotl-cloud
           tags: |
             type=ref,event=branch
@@ -145,6 +241,7 @@ jobs:
         uses: docker/build-push-action@v5
         with:
           context: .
+          platforms: ${{ matrix.platforms }}
           build-args: |
             BASE_TAG=${{ github.ref_type == 'tag' && 'main' || github.ref_name }}-py${{ matrix.python_version }}-cu${{ matrix.cuda }}-${{ matrix.pytorch }}${{ matrix.axolotl_extras != '' && '-' || '' }}${{ matrix.axolotl_extras }}
             CUDA=${{ matrix.cuda }}
@@ -155,34 +252,98 @@ jobs:
              ${{ (matrix.is_latest) && format('{0}-latest', steps.metadata.outputs.tags) || '' }}
           labels: ${{ steps.metadata.outputs.labels }}
 
+  build-axolotl-cloud-uv:
+    needs: build-axolotl-uv
+    if: ${{ ! contains(github.event.commits[0].message, '[skip docker]') && github.repository_owner == 'axolotl-ai-cloud' }}
+    # this job needs to be run on self-hosted GPU runners...
+    strategy:
+      fail-fast: false
+      matrix:
+        include:
+          - cuda: 128
+            cuda_version: 12.8.1
+            python_version: "3.11"
+            pytorch: 2.9.1
+            axolotl_extras:
+            platforms: "linux/amd64,linux/arm64"
+          - cuda: 128
+            cuda_version: 12.8.1
+            python_version: "3.12"
+            pytorch: 2.9.1
+            axolotl_extras:
+            is_latest: true
+            platforms: "linux/amd64,linux/arm64"
+          - cuda: 128
+            cuda_version: 12.8.1
+            python_version: "3.12"
+            pytorch: 2.10.0
+            axolotl_extras:
+            platforms: "linux/amd64,linux/arm64"
+          - cuda: 130
+            cuda_version: 13.0.0
+            python_version: "3.11"
+            pytorch: 2.9.1
+            axolotl_extras:
+            platforms: "linux/amd64,linux/arm64"
+          - cuda: 130
+            cuda_version: 13.0.0
+            python_version: "3.12"
+            pytorch: 2.10.0
+            axolotl_extras:
+            platforms: "linux/amd64,linux/arm64"
+    runs-on: axolotl-gpu-runner
+    steps:
+      - name: Checkout
+        uses: actions/checkout@v4
+      - name: Docker metadata
+        id: metadata
+        uses: docker/metadata-action@v5
+        with:
+          images: |
+            axolotlai/axolotl-cloud-uv
+          tags: |
+            type=ref,event=branch
+            type=pep440,pattern={{version}}
+      - name: Login to Docker Hub
+        uses: docker/login-action@v3
+        with:
+          username: ${{ secrets.DOCKERHUB_USERNAME }}
+          password: ${{ secrets.DOCKERHUB_TOKEN }}
+      - name: Set up Docker Buildx
+        uses: docker/setup-buildx-action@v3
+      - name: Build
+        uses: docker/build-push-action@v5
+        with:
+          context: .
+          platforms: ${{ matrix.platforms }}
+          build-args: |
+            BASE_TAG=${{ github.ref_type == 'tag' && 'main' || github.ref_name }}-py${{ matrix.python_version }}-cu${{ matrix.cuda }}-${{ matrix.pytorch }}${{ matrix.axolotl_extras != '' && '-' || '' }}${{ matrix.axolotl_extras }}
+            CUDA=${{ matrix.cuda }}
+          file: ./docker/Dockerfile-cloud-uv
+          push: ${{ github.event_name != 'pull_request' }}
+          tags: |
+             ${{ steps.metadata.outputs.tags }}-py${{ matrix.python_version }}-cu${{ matrix.cuda }}-${{ matrix.pytorch }}${{ matrix.axolotl_extras != '' && '-' || '' }}${{ matrix.axolotl_extras }}
+             ${{ (matrix.is_latest) && format('{0}-latest', steps.metadata.outputs.tags) || '' }}
+          labels: ${{ steps.metadata.outputs.labels }}
+
   build-axolotl-cloud-no-tmux:
     needs: build-axolotl
     if: ${{ ! contains(github.event.commits[0].message, '[skip docker]') && github.repository_owner == 'axolotl-ai-cloud' }}
     # this job needs to be run on self-hosted GPU runners...
     strategy:
+      fail-fast: false
       matrix:
         include:
-          - cuda: 126
-            cuda_version: 12.6.3
-            python_version: "3.11"
-            pytorch: 2.6.0
-            axolotl_extras:
-          - cuda: 126
-            cuda_version: 12.6.3
-            python_version: "3.11"
-            pytorch: 2.7.1
-            axolotl_extras:
-            is_latest:
-          - cuda: 126
-            cuda_version: 12.6.3
-            python_version: "3.11"
-            pytorch: 2.7.1
-            axolotl_extras: vllm
-            is_latest: true
           - cuda: 128
             cuda_version: 12.8.1
             python_version: "3.11"
-            pytorch: 2.8.0
+            pytorch: 2.9.1
+            axolotl_extras:
+            is_latest: true
+          - cuda: 130
+            cuda_version: 13.0.0
+            python_version: "3.11"
+            pytorch: 2.9.1
             axolotl_extras:
             is_latest:
     runs-on: axolotl-gpu-runner
@@ -194,7 +355,6 @@ jobs:
         uses: docker/metadata-action@v5
         with:
           images: |
-            winglian/axolotl-cloud-term
             axolotlai/axolotl-cloud-term
           tags: |
             type=ref,event=branch
@@ -210,6 +370,7 @@ jobs:
         uses: docker/build-push-action@v5
         with:
           context: .
+          platforms: linux/amd64,linux/arm64
           build-args: |
             BASE_TAG=${{ github.ref_type == 'tag' && 'main' || github.ref_name }}-py${{ matrix.python_version }}-cu${{ matrix.cuda }}-${{ matrix.pytorch }}${{ matrix.axolotl_extras != '' && '-' || '' }}${{ matrix.axolotl_extras }}
             CUDA=${{ matrix.cuda }}

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

@@ -8,6 +8,7 @@ on:
       - 'setup.py'
       - 'pyproject.toml'
       - '.github/workflows/multi-gpu-e2e.yml'
+      - 'scripts/cutcrossentropy_install.py'
       - 'src/axolotl/core/trainers/mixins/sequence_parallel.py'
       - 'src/axolotl/utils/distributed.py'
   workflow_dispatch:
@@ -19,6 +20,12 @@ concurrency:
   group: ${{ github.workflow }}-${{ github.ref }}
   cancel-in-progress: ${{ github.ref != 'refs/heads/main' }}
 
+permissions:
+  contents: read
+
+env:
+  MODAL_IMAGE_BUILDER_VERSION: "2025.06"
+
 jobs:
   test-axolotl-multigpu:
     if: ${{ ! contains(github.event.commits[0].message, '[skip e2e]') && github.repository_owner == 'axolotl-ai-cloud' && (github.event_name != 'pull_request' || !github.event.pull_request.draft) }}
@@ -26,27 +33,27 @@ jobs:
       fail-fast: false
       matrix:
         include:
-          - cuda: 126
-            cuda_version: 12.6.3
+#          - cuda: 129
+#            cuda_version: 12.9.1
+#            python_version: "3.12"
+#            pytorch: 2.9.1
+#            axolotl_extras: "fbgemm-gpu"
+#            num_gpus: 2
+#            dockerfile: "Dockerfile-uv.jinja"
+          - cuda: 130
+            cuda_version: 13.0.0
             python_version: "3.11"
-            pytorch: 2.6.0
+            pytorch: 2.9.1
             axolotl_extras:
+#            axolotl_extras: fbgemm-gpu
             num_gpus: 2
-            nightly_build: "true"
-          - cuda: 126
-            cuda_version: 12.6.3
-            python_version: "3.11"
-            pytorch: 2.7.1
-            axolotl_extras: vllm
-            num_gpus: 2
-            nightly_build: "true"
           - cuda: 128
             cuda_version: 12.8.1
             python_version: "3.11"
-            pytorch: 2.8.0
-            axolotl_extras: fbgemm-gpu
+            pytorch: 2.10.0
+            axolotl_extras: "fbgemm-gpu"
             num_gpus: 2
-            nightly_build: "true"
+            dockerfile: "Dockerfile-uv.jinja"
     runs-on: [self-hosted, modal]
     timeout-minutes: 120
     steps:
@@ -59,7 +66,7 @@ jobs:
       - name: Install Modal
         run: |
           python -m pip install --upgrade pip
-          pip install modal==1.0.2 jinja2
+          pip install modal==1.3.0.post1 jinja2
       - name: Update env vars
         run: |
           echo "BASE_TAG=main-base-py${{ matrix.python_version }}-cu${{ matrix.cuda }}-${{ matrix.pytorch }}" >> $GITHUB_ENV
@@ -68,8 +75,9 @@ jobs:
           echo "AXOLOTL_EXTRAS=${{ matrix.axolotl_extras}}" >> $GITHUB_ENV
           echo "CUDA=${{ matrix.cuda }}" >> $GITHUB_ENV
           echo "N_GPUS=${{ matrix.num_gpus }}" >> $GITHUB_ENV
-          echo "NIGHTLY_BUILD=${{ matrix.nightly_build }}" >> $GITHUB_ENV
-          echo "CODECOV_TOKEN=${{ secrets.CODECOV_TOKEN }}" >> $GITHUB_ENV
+          echo "E2E_DOCKERFILE=${{ matrix.dockerfile || 'Dockerfile.jinja'}}" >> $GITHUB_ENV
       - name: Run tests job on Modal
+        env:
+          CODECOV_TOKEN: ${{ secrets.CODECOV_TOKEN }}
         run: |
-          modal run cicd.multigpu
+          modal run -m cicd.multigpu

.github/workflows/nightlies.yml

@@ -5,6 +5,9 @@ on:
   schedule:
     - cron: '0 0 * * *'  # Runs at 00:00 UTC every day
 
+permissions:
+  contents: read
+
 jobs:
   build-axolotl:
     if: ${{ ! contains(github.event.commits[0].message, '[skip docker]') && github.repository_owner == 'axolotl-ai-cloud' }}
@@ -12,15 +15,10 @@ jobs:
       fail-fast: false
       matrix:
         include:
-          - cuda: 126
-            cuda_version: 12.6.3
+          - cuda: 128
+            cuda_version: 12.8.1
             python_version: "3.11"
-            pytorch: 2.6.0
-            axolotl_extras:
-          - cuda: 126
-            cuda_version: 12.6.3
-            python_version: "3.11"
-            pytorch: 2.7.1
+            pytorch: 2.9.1
             axolotl_extras:
     runs-on: axolotl-gpu-runner
     steps:
@@ -31,7 +29,6 @@ jobs:
         uses: docker/metadata-action@v5
         with:
           images: |
-            winglian/axolotl
             axolotlai/axolotl
           tags: |
             type=raw,value={{ branch }}-{{ date 'YYYYMMDD' }}
@@ -65,15 +62,10 @@ jobs:
     strategy:
       matrix:
         include:
-          - cuda: 126
-            cuda_version: 12.6.3
+          - cuda: 128
+            cuda_version: 12.8.1
             python_version: "3.11"
-            pytorch: 2.6.0
-            axolotl_extras:
-          - cuda: 126
-            cuda_version: 12.6.3
-            python_version: "3.11"
-            pytorch: 2.7.1
+            pytorch: 2.9.1
             axolotl_extras:
     runs-on: axolotl-gpu-runner
     steps:
@@ -84,7 +76,6 @@ 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

@@ -2,9 +2,11 @@ name: Pre-commit auto-update
 
 on:
   schedule:
-    - cron: '0 0 * * 0'  # Run weekly
+    - cron: '0 0 1 * *'  # Run monthly
   workflow_dispatch:  # Manual kickoff
 
+permissions: {}
+
 jobs:
   auto-update:
     runs-on: ubuntu-latest

.github/workflows/preview-docs.yml

@@ -11,22 +11,21 @@ on:
       - '_quarto.yml'
       - docs/scripts/generate_config_docs.py
       - src/axolotl/utils/schemas/**.py
+      - .github/workflows/preview-docs.yml
 
 permissions:
-  checks: write
-  contents: write
-  deployments: write
-  issues: write
-  discussions: write
-  pages: write
+  contents: read
   pull-requests: write
-  statuses: write
 
 jobs:
   preview:
     runs-on: ubuntu-latest
     if: ${{ !github.event.pull_request.draft }}
     steps:
+      - name: cleanup node
+        run: |
+          sudo rm -rf /usr/share/dotnet /usr/local/lib/android /opt/ghc /opt/hostedtoolcache/CodeQL
+
       - name: Check out repository
         uses: actions/checkout@v4
         with:

.github/workflows/pypi.yml

@@ -3,9 +3,11 @@ name: publish pypi
 on:
   push:
     tags:
-      - 'v*'
+      - "v*"
   workflow_dispatch:
 
+permissions: {}
+
 jobs:
   setup_release:
     name: Create Release
@@ -28,7 +30,8 @@ jobs:
       name: pypi
       url: https://pypi.org/p/axolotl
     permissions:
-      id-token: write  # IMPORTANT: this permission is mandatory for trusted publishing
+      contents: read
+      id-token: write # IMPORTANT: this permission is mandatory for trusted publishing
     steps:
       - name: Check out repository code
         uses: actions/checkout@v4
@@ -40,17 +43,17 @@ jobs:
 
       - name: Install dependencies
         run: |
-          pip3 install wheel packaging==23.2
+          pip3 install wheel packaging==26.0
           pip3 install --no-build-isolation -e .
           pip3 install -r requirements-dev.txt -r requirements-tests.txt
 
       - name: Extract tag name
         id: tag
-        run: echo ::set-output name=TAG_NAME::$(echo $GITHUB_REF | cut -d / -f 3)
+        run: echo "TAG_NAME=$(echo $GITHUB_REF | cut -d / -f 3)" >> "$GITHUB_OUTPUT"
 
-      - name: Update version in setup.py
+      - name: Update version in VERSION file
         run: |
-          sed -i -E 's/version="([0-9.]+)",/version="${{ steps.tag.outputs.TAG_NAME }}",/g' setup.py
+          echo "${{ steps.tag.outputs.TAG_NAME }}" | sed 's/^v//' > VERSION
 
       - name: Build a source dist
         run: |

.github/workflows/tests-nightly.yml

@@ -3,6 +3,13 @@ on:
   workflow_dispatch:
   schedule:
     - cron: '0 0 * * *'  # Runs at 00:00 UTC every day
+  pull_request:
+    types: [opened, synchronize, reopened, ready_for_review]
+    paths:
+      - '.github/workflows/tests-nightly.yml'
+
+permissions:
+  contents: read
 
 jobs:
   pre-commit:
@@ -18,15 +25,26 @@ jobs:
         env:
           SKIP: no-commit-to-branch
 
+  prime-cdn-s3-cache:
+    name: Prefetch S3 once to prime the CDN cache
+    runs-on: ubuntu-latest
+    if: ${{ !github.event.pull_request.draft }}
+    timeout-minutes: 10
+    steps:
+      - name: Restore Cache from S3
+        id: hf-cache-restore-s3
+        run: |
+          curl -v -H "Range: bytes=0-1023" -L https://axolotl-ci.b-cdn.net/hf-cache.tar.zst > /dev/null
+
   pytest:
     name: PyTest
     runs-on: ubuntu-latest
+    needs: [prime-cdn-s3-cache]
     strategy:
       fail-fast: false
-      max-parallel: 2
       matrix:
-        python_version: ["3.11"]
-        pytorch_version: ["2.6.0", "2.7.0"]
+        python_version: ["3.12"]  # TODO include py3.14 once https://github.com/mistralai/mistral-common/pull/194 is merged
+        pytorch_version: ["2.9.1", "2.10.0"]
     timeout-minutes: 20
 
     steps:
@@ -37,7 +55,7 @@ jobs:
         id: hf-cache-restore-s3
         run: |
           mkdir -p /home/runner/.cache/huggingface/hub
-          curl -L https://d1dttdx32dkk5p.cloudfront.net/hf-cache.tar.zst | tar -xf - -C /home/runner/.cache/huggingface/hub/  --use-compress-program unzstd
+          curl -L https://axolotl-ci.b-cdn.net/hf-cache.tar.zst | tar -xf - -C /home/runner/.cache/huggingface/hub/  --use-compress-program unzstd
 
       - name: Setup Python
         uses: actions/setup-python@v5
@@ -48,7 +66,7 @@ jobs:
       - name: upgrade pip
         run: |
           pip3 install --upgrade pip
-          pip3 install --upgrade packaging==23.2 setuptools==75.8.0 wheel
+          pip3 install --upgrade packaging==26.0 setuptools==78.1.1 wheel
 
       - name: Install PyTorch
         run: |
@@ -99,19 +117,26 @@ jobs:
       fail-fast: false
       matrix:
         include:
-          - cuda: 126
-            cuda_version: 12.6.3
+          - cuda: 128
+            cuda_version: 12.8.1
             python_version: "3.11"
-            pytorch: 2.6.0
+            pytorch: 2.9.1
             num_gpus: 1
             axolotl_extras:
             nightly_build: "true"
-          - cuda: 126
-            cuda_version: 12.6.3
+          - cuda: 128
+            cuda_version: 12.8.1
             python_version: "3.11"
-            pytorch: 2.7.1
+            pytorch: 2.10.0
             num_gpus: 1
             axolotl_extras:
+          - cuda: 130
+            cuda_version: 13.0.0
+            python_version: "3.12"
+            pytorch: 2.9.1
+            num_gpus: 1
+            axolotl_extras:
+            dockerfile: "Dockerfile-uv.jinja"
             nightly_build: "true"
     steps:
       - name: Checkout
@@ -123,7 +148,7 @@ jobs:
       - name: Install Modal
         run: |
           python -m pip install --upgrade pip
-          pip install modal==1.0.2 jinja2
+          pip install modal==1.3.0.post1 jinja2
       - name: Update env vars
         run: |
           echo "BASE_TAG=main-base-py${{ matrix.python_version }}-cu${{ matrix.cuda }}-${{ matrix.pytorch }}" >> $GITHUB_ENV
@@ -132,9 +157,11 @@ jobs:
           echo "AXOLOTL_EXTRAS=${{ matrix.axolotl_extras}}" >> $GITHUB_ENV
           echo "CUDA=${{ matrix.cuda }}" >> $GITHUB_ENV
           echo "N_GPUS=${{ matrix.num_gpus }}" >> $GITHUB_ENV
+          echo "E2E_DOCKERFILE=${{ matrix.dockerfile || 'Dockerfile.jinja'}}" >> $GITHUB_ENV
           echo "NIGHTLY_BUILD=${{ matrix.nightly_build }}" >> $GITHUB_ENV
-          echo "CODECOV_TOKEN=${{ secrets.CODECOV_TOKEN }}" >> $GITHUB_ENV
       - name: Run tests job on Modal
+        env:
+          CODECOV_TOKEN: ${{ secrets.CODECOV_TOKEN }}
         run: |
           modal run cicd.e2e_tests
   docker-e2e-multigpu-tests:
@@ -148,10 +175,10 @@ jobs:
       fail-fast: false
       matrix:
         include:
-          - cuda: 126
-            cuda_version: 12.6.3
+          - cuda: 128
+            cuda_version: 12.8.1
             python_version: "3.11"
-            pytorch: 2.7.1
+            pytorch: 2.9.1
             num_gpus: 2
             axolotl_extras:
             nightly_build: "true"
@@ -165,7 +192,7 @@ jobs:
       - name: Install Modal
         run: |
           python -m pip install --upgrade pip
-          pip install modal==1.0.2 jinja2
+          pip install modal==1.3.0.post1 jinja2
       - name: Update env vars
         run: |
           echo "BASE_TAG=main-base-py${{ matrix.python_version }}-cu${{ matrix.cuda }}-${{ matrix.pytorch }}" >> $GITHUB_ENV
@@ -175,7 +202,8 @@ jobs:
           echo "CUDA=${{ matrix.cuda }}" >> $GITHUB_ENV
           echo "N_GPUS=${{ matrix.num_gpus }}" >> $GITHUB_ENV
           echo "NIGHTLY_BUILD=${{ matrix.nightly_build }}" >> $GITHUB_ENV
-          echo "CODECOV_TOKEN=${{ secrets.CODECOV_TOKEN }}" >> $GITHUB_ENV
       - name: Run tests job on Modal
+        env:
+          CODECOV_TOKEN: ${{ secrets.CODECOV_TOKEN }}
         run: |
           modal run cicd.multigpu

.github/workflows/tests.yml

@@ -28,6 +28,9 @@ concurrency:
   group: ${{ github.workflow }}-${{ github.ref }}
   cancel-in-progress: ${{ github.ref != 'refs/heads/main' }}
 
+permissions:
+  contents: read
+
 env:
   TRANSFORMERS_IS_CI: "yes"
 
@@ -46,27 +49,46 @@ jobs:
         env:
           SKIP: no-commit-to-branch
 
+  prime-cdn-s3-cache:
+    name: Prefetch S3 once to prime the CDN cache
+    runs-on: ubuntu-latest
+    if: ${{ !github.event.pull_request.draft }}
+    timeout-minutes: 10
+    steps:
+      - name: Restore Cache from S3
+        id: hf-cache-restore-s3
+        run: |
+          curl -v -H "Range: bytes=0-1023" -L https://axolotl-ci.b-cdn.net/hf-cache.tar.zst > /dev/null
+
   pytest:
     name: PyTest
     runs-on: ubuntu-latest
     if: ${{ !github.event.pull_request.draft }}
-#    needs: [preload-cache]
+    needs: [prime-cdn-s3-cache]
     strategy:
       fail-fast: false
       matrix:
-        python_version: ["3.11"]
-        pytorch_version: ["2.6.0", "2.7.1", "2.8.0"]
+        python_version: ["3.12", "3.14"]
+        pytorch_version: ["2.9.1", "2.10.0"]
+        exclude:
+          - python_version: "3.14"
+            pytorch_version: "2.9.1"
     timeout-minutes: 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 /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
+          mkdir -p ~/.cache/huggingface/hub
+          curl -L https://axolotl-ci.b-cdn.net/hf-cache.tar.zst | tar -xpf - -C ~/.cache/huggingface/hub/  --use-compress-program unzstd --strip-components=1
+          ls -ltr ~/.cache/huggingface/hub/
 
       - name: Setup Python
         uses: actions/setup-python@v5
@@ -77,20 +99,24 @@ jobs:
       - name: upgrade pip
         run: |
           pip3 install --upgrade pip
-          pip3 install --upgrade packaging==23.2 setuptools==75.8.0 wheel
+          pip3 install --upgrade packaging==26.0 setuptools==75.8.0 wheel
 
       - name: Install PyTorch
         run: |
-          pip3 install torch==${{ matrix.pytorch_version }} torchvision
+          pip3 install --no-cache-dir torch==${{ matrix.pytorch_version }} torchvision
 
       - name: Install dependencies
         run: |
           pip3 show torch
-          pip3 install --no-build-isolation -U -e .
+          pip3 install --no-cache-dir --no-build-isolation -U -e .
           python scripts/unsloth_install.py | sh
           python scripts/cutcrossentropy_install.py | sh
           pip3 install -r requirements-dev.txt -r requirements-tests.txt
 
+      - name: cleanup pip cache
+        run: |
+          find "$(pip cache dir)/http-v2" -type f -mtime +14 -exec rm {} \;
+
       - name: Make sure PyTorch version wasn't clobbered
         run: |
           python -c "import torch; assert '${{ matrix.pytorch_version }}' in torch.__version__"
@@ -101,15 +127,25 @@ jobs:
 
       - name: Pre-Download dataset fixture
         run: |
-          huggingface-cli download --repo-type=dataset axolotl-ai-internal/axolotl-oss-dataset-fixtures
+          hf download --repo-type=dataset axolotl-ai-internal/axolotl-oss-dataset-fixtures
+
+      - name: Show HF cache
+        run: hf cache ls
 
       - name: Run tests
         run: |
-          pytest -v --durations=10 -n8 --dist loadfile --ignore=tests/e2e/ --ignore=tests/patched/ --ignore=tests/cli/ --ignore=tests/monkeypatch/ tests/ --cov=axolotl --cov-report=xml
+          df -h
+          pytest -v --durations=10 -n4 --dist loadfile --ignore=tests/e2e/ --ignore=tests/patched/ --ignore=tests/cli/ --ignore=tests/monkeypatch/ tests/ --cov=axolotl --cov-report=xml
+          df -h
           pytest -v --durations=10 tests/monkeypatch/ --cov=axolotl --cov-append --cov-report=xml
+          df -h
           pytest -v --durations=10 tests/patched/ --cov=axolotl --cov-append --cov-report=xml
+          df -h
           pytest -v --durations=10 tests/cli/ --cov=axolotl --cov-append --cov-report=xml
 
+      - name: Show HF cache
+        run: hf cache ls
+
       - name: Upload coverage to Codecov
         uses: codecov/codecov-action@v5
         with:
@@ -118,30 +154,35 @@ 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
     if: ${{ !github.event.pull_request.draft }}
+    needs: [prime-cdn-s3-cache]
     strategy:
       fail-fast: false
       matrix:
-        python_version: ["3.11"]
-        pytorch_version: ["2.6.0", "2.7.1", "2.8.0"]
-    timeout-minutes: 20
+        python_version: ["3.12", "3.14"]
+        pytorch_version: ["2.9.1", "2.10.0"]
+        exclude:
+          - python_version: "3.14"
+            pytorch_version: "2.9.1"
+    timeout-minutes: 30
 
     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 /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
+          mkdir -p ~/.cache/huggingface/hub
+          curl -L https://axolotl-ci.b-cdn.net/hf-cache.tar.zst | tar -xpf - -C ~/.cache/huggingface/hub/  --use-compress-program unzstd --strip-components=1
+          ls -ltr ~/.cache/huggingface/hub/
 
       - name: Setup Python
         uses: actions/setup-python@v5
@@ -152,21 +193,25 @@ jobs:
       - name: upgrade pip
         run: |
           pip3 install --upgrade pip
-          pip3 install --upgrade packaging==23.2 setuptools==75.8.0 setuptools_scm build wheel
+          pip3 install --upgrade packaging==26.0 setuptools==75.8.0 setuptools_scm build wheel psutil
 
       - name: Install PyTorch
         run: |
-          pip3 install torch==${{ matrix.pytorch_version }} torchvision
+          pip3 install --no-cache-dir torch==${{ matrix.pytorch_version }} torchvision
 
       - name: Install dependencies
         run: |
           pip3 show torch
           python -m build --no-isolation --sdist
-          pip3 install --no-build-isolation dist/axolotl*.tar.gz
+          pip3 install --no-cache-dir --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__"
@@ -175,21 +220,30 @@ jobs:
         run: |
           axolotl --help
 
+      - name: Verify agent docs are discoverable
+        run: |
+          # Agent docs live in docs/agents/ (source of truth) and are resolved
+          # at runtime from the repo checkout or via `axolotl fetch docs`
+          axolotl agent-docs --list
+          axolotl agent-docs | grep -q "Fine-tuning framework"
+          axolotl agent-docs grpo | grep -q "GRPO"
+          axolotl agent-docs sft | grep -q "SFT"
+          python -c "from axolotl.cli.agent_docs import get_doc, list_topics; assert len(list_topics()) >= 5; assert 'GRPO' in get_doc('grpo')"
+
       - name: Show HF cache
-        run: huggingface-cli scan-cache
+        run: hf cache ls
 
       - name: Run tests
         run: |
-          pytest -v --durations=10 -n8 --dist loadfile --ignore=tests/e2e/ --ignore=tests/patched/ --ignore=tests/cli/ --ignore=tests/monkeypatch/ tests/ --cov=axolotl --cov-report=xml
+          pytest -v --durations=10 -n4 --dist loadfile --ignore=tests/e2e/ --ignore=tests/patched/ --ignore=tests/cli/ --ignore=tests/monkeypatch/ tests/ --cov=axolotl --cov-report=xml
           pytest -v --durations=10 tests/monkeypatch/ --cov=axolotl --cov-append --cov-report=xml
           pytest -v --durations=10 tests/cli/
 
-      - name: cleanup pip cache
-        run: |
-          find "$(pip cache dir)/http-v2" -type f -mtime +14 -exec rm {} \;
+      - name: Show HF cache
+        run: hf cache ls
 
   gate-skip-e2e:
-    needs: [pre-commit, pytest, pytest-sdist]
+    needs: [pre-commit]
     runs-on: ubuntu-latest
     outputs:
       skip: ${{ steps.compute.outputs.skip }}
@@ -225,22 +279,16 @@ jobs:
     # 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]
 
     strategy:
       fail-fast: false
       matrix:
         include:
-          - cuda: 126
-            cuda_version: 12.6.3
-            python_version: "3.11"
-            pytorch: 2.7.1
-            num_gpus: 1
-            axolotl_extras:
-          - cuda: 126
-            cuda_version: 12.6.3
-            python_version: "3.11"
-            pytorch: 2.7.1
+          - cuda: 130
+            cuda_version: 13.0.0
+            python_version: "3.12"
+            pytorch: 2.9.1
             num_gpus: 1
             axolotl_extras:
             dockerfile: "Dockerfile-uv.jinja"
@@ -254,7 +302,7 @@ jobs:
       - name: Install Modal
         run: |
           python -m pip install --upgrade pip
-          pip install modal==1.0.2 jinja2
+          pip install modal==1.3.0.post1 jinja2
       - name: Update env vars
         run: |
           echo "BASE_TAG=main-base-py${{ matrix.python_version }}-cu${{ matrix.cuda }}-${{ matrix.pytorch }}" >> $GITHUB_ENV
@@ -264,9 +312,10 @@ jobs:
           echo "CUDA=${{ matrix.cuda }}" >> $GITHUB_ENV
           echo "MODAL_IMAGE_BUILDER_VERSION=2024.10" >> $GITHUB_ENV
           echo "N_GPUS=${{ matrix.num_gpus }}" >> $GITHUB_ENV
-          echo "CODECOV_TOKEN=${{ secrets.CODECOV_TOKEN }}" >> $GITHUB_ENV
           echo "E2E_DOCKERFILE=${{ matrix.dockerfile || 'Dockerfile.jinja'}}" >> $GITHUB_ENV
       - name: Run tests job on Modal
+        env:
+          CODECOV_TOKEN: ${{ secrets.CODECOV_TOKEN }}
         run: |
           modal run cicd.e2e_tests
 
@@ -286,25 +335,24 @@ jobs:
       fail-fast: false
       matrix:
         include:
-          - cuda: 126
-            cuda_version: 12.6.3
+          - cuda: 128
+            cuda_version: 12.8.1
             python_version: "3.11"
-            pytorch: 2.6.0
+            pytorch: 2.9.1
             num_gpus: 1
             axolotl_extras:
           - cuda: 128
             cuda_version: 12.8.1
             python_version: "3.11"
-            pytorch: 2.7.1
+            pytorch: 2.10.0
             num_gpus: 1
             axolotl_extras:
-          - cuda: 128
-            cuda_version: 12.8.1
+          - cuda: 130
+            cuda_version: 13.0.0
             python_version: "3.11"
-            pytorch: 2.8.0
+            pytorch: 2.9.1
             num_gpus: 1
-            gpu_type: "B200"
-            axolotl_extras: fbgemm-gpu
+            axolotl_extras:
     steps:
       - name: Checkout
         uses: actions/checkout@v4
@@ -315,7 +363,7 @@ jobs:
       - name: Install Modal
         run: |
           python -m pip install --upgrade pip
-          pip install modal==1.0.2 jinja2
+          pip install modal==1.3.0.post1 jinja2
       - name: Update env vars
         run: |
           echo "BASE_TAG=main-base-py${{ matrix.python_version }}-cu${{ matrix.cuda }}-${{ matrix.pytorch }}" >> $GITHUB_ENV
@@ -326,9 +374,10 @@ jobs:
           echo "MODAL_IMAGE_BUILDER_VERSION=2024.10" >> $GITHUB_ENV
           echo "N_GPUS=${{ matrix.num_gpus }}" >> $GITHUB_ENV
           echo "GPU_TYPE=${{ matrix.gpu_type || 'L40S'}}" >> $GITHUB_ENV
-          echo "CODECOV_TOKEN=${{ secrets.CODECOV_TOKEN }}" >> $GITHUB_ENV
           echo "E2E_DOCKERFILE=${{ matrix.dockerfile || 'Dockerfile.jinja'}}" >> $GITHUB_ENV
       - name: Run tests job on Modal
+        env:
+          CODECOV_TOKEN: ${{ secrets.CODECOV_TOKEN }}
         run: |
           modal run cicd.e2e_tests
 
@@ -342,10 +391,10 @@ jobs:
       fail-fast: false
       matrix:
         include:
-          - cuda: 126
-            cuda_version: 12.6.3
+          - cuda: 128
+            cuda_version: 12.8.1
             python_version: "3.11"
-            pytorch: 2.7.1
+            pytorch: 2.9.1
             num_gpus: 1
             axolotl_extras:
     steps:
@@ -358,7 +407,7 @@ jobs:
       - name: Install Modal
         run: |
           python -m pip install --upgrade pip
-          pip install modal==1.0.2 jinja2
+          pip install modal==1.3.0.post1 jinja2
       - name: Update env vars
         run: |
           echo "BASE_TAG=main-base-py${{ matrix.python_version }}-cu${{ matrix.cuda }}-${{ matrix.pytorch }}" >> $GITHUB_ENV
@@ -368,7 +417,6 @@ jobs:
           echo "CUDA=${{ matrix.cuda }}" >> $GITHUB_ENV
           echo "MODAL_IMAGE_BUILDER_VERSION=2024.10" >> $GITHUB_ENV
           echo "N_GPUS=${{ matrix.num_gpus }}" >> $GITHUB_ENV
-          echo "CODECOV_TOKEN=${{ secrets.CODECOV_TOKEN }}" >> $GITHUB_ENV
       - name: Run tests job on Modal
         run: |
           modal run cicd.cleanup

.pre-commit-config.yaml

@@ -11,13 +11,13 @@ repos:
     -   id: no-commit-to-branch
         args: ['--branch', 'main']
 -   repo: https://github.com/astral-sh/ruff-pre-commit
-    rev: v0.12.12
+    rev: v0.15.8
     hooks:
     -   id: ruff
         args: [--fix]
     -   id: ruff-format
 -   repo: https://github.com/pre-commit/mirrors-mypy
-    rev: v1.17.1
+    rev: v1.19.1
     hooks:
     - id: mypy
       additional_dependencies:
@@ -26,7 +26,7 @@ repos:
             'pydantic>=2.5.3',
         ]
 -   repo: https://github.com/PyCQA/bandit
-    rev: 1.8.6
+    rev: 1.9.4
     hooks:
     -   id: bandit
         args: [

.runpod/Dockerfile

@@ -10,6 +10,7 @@ ARG BASE_VOLUME="/runpod-volume"
 ENV BASE_VOLUME=$BASE_VOLUME
 ENV HF_DATASETS_CACHE="${BASE_VOLUME}/huggingface-cache/datasets"
 ENV HUGGINGFACE_HUB_CACHE="${BASE_VOLUME}/huggingface-cache/hub"
+ENV HF_HUB_CACHE="${BASE_VOLUME}/huggingface-cache/hub"
 ENV TRANSFORMERS_CACHE="${BASE_VOLUME}/huggingface-cache/hub"
 
 COPY .runpod/src /src

.runpod/README.md

@@ -123,7 +123,7 @@ datasets:
 | --------------------------------- | -------------------------- | ----------------------------------- |
 | `dataset_prepared_path`           | `"data/last_run_prepared"` | Path for prepared dataset           |
 | `push_dataset_to_hub`             | `""`                       | Push dataset to HF hub              |
-| `dataset_processes`               | `4`                        | Number of preprocessing processes   |
+| `dataset_num_proc`                | `4`                        | Number of preprocessing processes   |
 | `dataset_keep_in_memory`          | `false`                    | Keep dataset in memory              |
 | `shuffle_merged_datasets`         | `true`                     | Shuffle merged datasets             |
 | `shuffle_before_merging_datasets` | `false`                    | Shuffle each dataset before merging |

.runpod/src/config/config.yaml

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

AGENTS.md

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

MANIFEST.in

@@ -3,4 +3,6 @@ include README.md
 include LICENSE
 include src/setuptools_axolotl_dynamic_dependencies.py
 include src/axolotl/utils/chat_templates/templates/*.jinja
+include AGENTS.md
+recursive-include docs/agents *.md
 recursive-include axolotl *.py

README.md

@@ -29,21 +29,35 @@
 
 ## 🎉 Latest Updates
 
-- 2025/07:
-  - ND Parallelism support has been added into Axolotl. Compose Context Parallelism (CP), Tensor Parallelism (TP), and Fully Sharded Data Parallelism (FSDP) within a single node and across multiple nodes. Check out the [blog post](https://huggingface.co/blog/accelerate-nd-parallel) for more info.
-  - Axolotl adds more models: [GPT-OSS](https://github.com/axolotl-ai-cloud/axolotl/tree/main/examples/gpt-oss), [Gemma 3n](https://github.com/axolotl-ai-cloud/axolotl/tree/main/examples/gemma3n), [Liquid Foundation Model 2 (LFM2)](https://github.com/axolotl-ai-cloud/axolotl/tree/main/examples/lfm2), and [Arcee Foundation Models (AFM)](https://github.com/axolotl-ai-cloud/axolotl/tree/main/examples/afm).
-  - FP8 finetuning with fp8 gather op is now possible in Axolotl via `torchao`. Get started [here](https://docs.axolotl.ai/docs/mixed_precision.html#sec-fp8)!
-  - [Voxtral](https://github.com/axolotl-ai-cloud/axolotl/tree/main/examples/voxtral), [Magistral 1.1](https://github.com/axolotl-ai-cloud/axolotl/tree/main/examples/magistral), and [Devstral](https://github.com/axolotl-ai-cloud/axolotl/tree/main/examples/devstral) with mistral-common tokenizer support has been integrated in Axolotl!
-  - TiledMLP support for single-GPU to multi-GPU training with DDP, DeepSpeed and FSDP support has been added to support Arctic Long Sequence Training. (ALST). See [examples](https://github.com/axolotl-ai-cloud/axolotl/tree/main/examples/alst) for using ALST with Axolotl!
-- 2025/05: Quantization Aware Training (QAT) support has been added to Axolotl. Explore the [docs](https://docs.axolotl.ai/docs/qat.html) to learn more!
-- 2025/03: Axolotl has implemented Sequence Parallelism (SP) support. Read the [blog](https://huggingface.co/blog/axolotl-ai-co/long-context-with-sequence-parallelism-in-axolotl) and [docs](https://docs.axolotl.ai/docs/sequence_parallelism.html) to learn how to scale your context length when fine-tuning.
+- 2026/03:
+  - New model support has been added in Axolotl for [Mistral Small 4](https://github.com/axolotl-ai-cloud/axolotl/tree/main/examples/mistral4), [Qwen3.5, Qwen3.5 MoE](https://github.com/axolotl-ai-cloud/axolotl/tree/main/examples/qwen3.5), [GLM-4.7-Flash](https://github.com/axolotl-ai-cloud/axolotl/tree/main/examples/glm47-flash), [GLM-4.6V](https://github.com/axolotl-ai-cloud/axolotl/tree/main/examples/glm46v), and [GLM-4.5-Air](https://github.com/axolotl-ai-cloud/axolotl/tree/main/examples/glm45).
+  - [MoE expert quantization](https://docs.axolotl.ai/docs/expert_quantization.html) support (via `quantize_moe_experts: true`) greatly reduces VRAM when training MoE models (FSDP2 compat).
+- 2026/02:
+  - [ScatterMoE LoRA](https://github.com/axolotl-ai-cloud/axolotl/pull/3410) support. LoRA fine-tuning directly on MoE expert weights using custom Triton kernels.
+  - Axolotl now has support for [SageAttention](https://github.com/axolotl-ai-cloud/axolotl/pull/2823) and [GDPO](https://github.com/axolotl-ai-cloud/axolotl/pull/3353) (Generalized DPO).
+- 2026/01:
+  - New integration for [EAFT](https://github.com/axolotl-ai-cloud/axolotl/pull/3366) (Entropy-Aware Focal Training), weights loss by entropy of the top-k logit distribution, and [Scalable Softmax](https://github.com/axolotl-ai-cloud/axolotl/pull/3338), improves long context in attention.
+- 2025/12:
+  - Axolotl now includes support for [Kimi-Linear](https://docs.axolotl.ai/docs/models/kimi-linear.html), [Plano-Orchestrator](https://docs.axolotl.ai/docs/models/plano.html), [MiMo](https://docs.axolotl.ai/docs/models/mimo.html), [InternVL 3.5](https://docs.axolotl.ai/docs/models/internvl3_5.html), [Olmo3](https://docs.axolotl.ai/docs/models/olmo3.html), [Trinity](https://docs.axolotl.ai/docs/models/trinity.html), and [Ministral3](https://docs.axolotl.ai/docs/models/ministral3.html).
+  - [Distributed Muon Optimizer](https://github.com/axolotl-ai-cloud/axolotl/pull/3264) support has been added for FSDP2 pretraining.
+- 2025/10: New model support has been added in Axolotl for: [Qwen3 Next](https://docs.axolotl.ai/docs/models/qwen3-next.html), [Qwen2.5-vl, Qwen3-vl](https://github.com/axolotl-ai-cloud/axolotl/tree/main/examples/qwen2_5-vl), [Qwen3, Qwen3MoE](https://docs.axolotl.ai/docs/models/qwen3.html), [Granite 4](https://docs.axolotl.ai/docs/models/granite4.html), [HunYuan](https://docs.axolotl.ai/docs/models/hunyuan.html), [Magistral 2509](https://docs.axolotl.ai/docs/models/magistral/vision.html), [Apertus](https://docs.axolotl.ai/docs/models/apertus.html), and [Seed-OSS](https://docs.axolotl.ai/docs/models/seed-oss.html).
 
 <details>
 
 <summary>Expand older updates</summary>
 
-- 2025/06: 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!
+- 2025/09: Axolotl now has text diffusion training. Read more [here](https://github.com/axolotl-ai-cloud/axolotl/tree/main/src/axolotl/integrations/diffusion).
+- 2025/08: QAT has been updated to include NVFP4 support. See [PR](https://github.com/axolotl-ai-cloud/axolotl/pull/3107).
+- 2025/07:
+  - ND Parallelism support has been added into Axolotl. Compose Context Parallelism (CP), Tensor Parallelism (TP), and Fully Sharded Data Parallelism (FSDP) within a single node and across multiple nodes. Check out the [blog post](https://huggingface.co/blog/accelerate-nd-parallel) for more info.
+  - Axolotl adds more models: [GPT-OSS](https://docs.axolotl.ai/docs/models/gpt-oss.html), [Gemma 3n](https://docs.axolotl.ai/docs/models/gemma3n.html), [Liquid Foundation Model 2 (LFM2)](https://docs.axolotl.ai/docs/models/LiquidAI.html), and [Arcee Foundation Models (AFM)](https://docs.axolotl.ai/docs/models/arcee.html).
+  - FP8 finetuning with fp8 gather op is now possible in Axolotl via `torchao`. Get started [here](https://docs.axolotl.ai/docs/mixed_precision.html#sec-fp8)!
+  - [Voxtral](https://docs.axolotl.ai/docs/models/voxtral.html), [Magistral 1.1](https://docs.axolotl.ai/docs/models/magistral.html), and [Devstral](https://docs.axolotl.ai/docs/models/devstral.html) with mistral-common tokenizer support has been integrated in Axolotl!
+  - TiledMLP support for single-GPU to multi-GPU training with DDP, DeepSpeed and FSDP support has been added to support Arctic Long Sequence Training. (ALST). See [examples](https://github.com/axolotl-ai-cloud/axolotl/tree/main/examples/alst) for using ALST with Axolotl!
+- 2025/06: Magistral with mistral-common tokenizer support has been added to Axolotl. See [docs](https://docs.axolotl.ai/docs/models/magistral.html) to start training your own Magistral models with Axolotl!
+- 2025/05: Quantization Aware Training (QAT) support has been added to Axolotl. Explore the [docs](https://docs.axolotl.ai/docs/qat.html) to learn more!
+- 2025/04: Llama 4 support has been added in Axolotl. See [docs](https://docs.axolotl.ai/docs/models/llama-4.html) to start training your own Llama 4 models with Axolotl's linearized version!
+- 2025/03: Axolotl has implemented Sequence Parallelism (SP) support. Read the [blog](https://huggingface.co/blog/axolotl-ai-co/long-context-with-sequence-parallelism-in-axolotl) and [docs](https://docs.axolotl.ai/docs/sequence_parallelism.html) to learn how to scale your context length when fine-tuning.
 - 2025/03: (Beta) Fine-tuning Multimodal models is now supported in Axolotl. Check out the [docs](https://docs.axolotl.ai/docs/multimodal.html) to fine-tune your own!
 - 2025/02: Axolotl has added LoRA optimizations to reduce memory usage and improve training speed for LoRA and QLoRA in single GPU and multi-GPU training (DDP and DeepSpeed). Jump into the [docs](https://docs.axolotl.ai/docs/lora_optims.html) to give it a try.
 - 2025/02: Axolotl has added GRPO support. Dive into our [blog](https://huggingface.co/blog/axolotl-ai-co/training-llms-w-interpreter-feedback-wasm) and [GRPO example](https://github.com/axolotl-ai-cloud/grpo_code) and have some fun!
@@ -58,10 +72,10 @@ Axolotl is a free and open-source tool designed to streamline post-training and
 Features:
 
 - **Multiple Model Support**: Train various models like GPT-OSS, LLaMA, Mistral, Mixtral, Pythia, and many more models available on the Hugging Face Hub.
-- **Multimodal Training**: Fine-tune vision-language models (VLMs) including LLaMA-Vision, Qwen2-VL, Pixtral, LLaVA, SmolVLM2, 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).
+- **Multimodal Training**: Fine-tune vision-language models (VLMs) including LLaMA-Vision, Qwen2-VL, Pixtral, LLaVA, SmolVLM2, GLM-4.6V, InternVL 3.5, Gemma 3n, and audio models like Voxtral with image, video, and audio support.
+- **Training Methods**: Full fine-tuning, LoRA, QLoRA, GPTQ, QAT, Preference Tuning (DPO, IPO, KTO, ORPO), RL (GRPO, GDPO), and Reward Modelling (RM) / Process Reward Modelling (PRM).
 - **Easy Configuration**: Re-use a single YAML configuration file across the full fine-tuning pipeline: dataset preprocessing, training, evaluation, quantization, and inference.
-- **Performance Optimizations**: [Multipacking](https://docs.axolotl.ai/docs/multipack.html), [Flash Attention](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!
+- **Performance Optimizations**: [Multipacking](https://docs.axolotl.ai/docs/multipack.html), [Flash Attention 2/3/4](https://docs.axolotl.ai/docs/attention.html#flash-attention), [Xformers](https://docs.axolotl.ai/docs/attention.html#xformers), [Flex Attention](https://docs.axolotl.ai/docs/attention.html#flex-attention), [SageAttention](https://docs.axolotl.ai/docs/attention.html#sageattention), [Liger Kernel](https://docs.axolotl.ai/docs/custom_integrations.html#liger-kernels), [Cut Cross Entropy](https://docs.axolotl.ai/docs/custom_integrations.html#cut-cross-entropy), [ScatterMoE](https://docs.axolotl.ai/docs/custom_integrations.html#kernels-integration), [Sequence Parallelism (SP)](https://docs.axolotl.ai/docs/sequence_parallelism.html), [LoRA optimizations](https://docs.axolotl.ai/docs/lora_optims.html), [Multi-GPU training (FSDP1, FSDP2, DeepSpeed)](https://docs.axolotl.ai/docs/multi-gpu.html), [Multi-node training (Torchrun, Ray)](https://docs.axolotl.ai/docs/multi-node.html), and many more!
 - **Flexible Dataset Handling**: Load from local, HuggingFace, and cloud (S3, Azure, GCP, OCI) datasets.
 - **Cloud Ready**: We ship [Docker images](https://hub.docker.com/u/axolotlai) and also [PyPI packages](https://pypi.org/project/axolotl/) for use on cloud platforms and local hardware.
 
@@ -72,8 +86,8 @@ Features:
 **Requirements**:
 
 - NVIDIA GPU (Ampere or newer for `bf16` and Flash Attention) or AMD GPU
-- Python 3.11
-- PyTorch ≥2.6.0
+- Python >=3.11 (3.12 recommended)
+- PyTorch ≥2.9.1
 
 ### Google Colab
 
@@ -81,10 +95,38 @@ Features:
 
 ### Installation
 
+#### Using uv (recommended)
+
+```bash
+# install uv if you don't already have it installed
+curl -LsSf https://astral.sh/uv/install.sh | sh
+source $HOME/.local/bin/env
+
+# CUDA 12.8.1 tends to have better package compatibility
+export UV_TORCH_BACKEND=cu128
+
+# create a new virtual environment
+uv venv --python 3.12
+source .venv/bin/activate
+
+uv pip install torch==2.10.0 torchvision
+uv pip install --no-build-isolation axolotl[deepspeed]
+
+# recommended - install cut-cross-entropy
+uv pip install "cut-cross-entropy[transformers] @ git+https://github.com/axolotl-ai-cloud/ml-cross-entropy.git@main"
+
+# (optional) - prefetch flash-attn2 and causal-conv1d kernels
+uv run --python 3.12 python -c "from kernels import get_kernel; get_kernel('kernels-community/flash-attn2'); get_kernel('kernels-community/causal-conv1d')"
+
+# Download example axolotl configs, deepspeed configs
+axolotl fetch examples
+axolotl fetch deepspeed_configs  # OPTIONAL
+```
+
 #### Using pip
 
 ```bash
-pip3 install -U packaging==23.2 setuptools==75.8.0 wheel ninja
+pip3 install -U packaging==26.0 setuptools==75.8.0 wheel ninja
 pip3 install --no-build-isolation axolotl[flash-attn,deepspeed]
 
 # Download example axolotl configs, deepspeed configs
@@ -143,6 +185,29 @@ That's it! Check out our [Getting Started Guide](https://docs.axolotl.ai/docs/ge
 - [API Reference](https://docs.axolotl.ai/docs/api/) - Auto-generated code documentation
 - [FAQ](https://docs.axolotl.ai/docs/faq.html) - Frequently asked questions
 
+## AI Agent Support
+
+Axolotl ships with built-in documentation optimized for AI coding agents (Claude Code, Cursor, Copilot, etc.). These docs are bundled with the pip package — no repo clone needed.
+
+```bash
+# Show overview and available training methods
+axolotl agent-docs
+
+# Topic-specific references
+axolotl agent-docs sft                 # supervised fine-tuning
+axolotl agent-docs grpo                # GRPO online RL
+axolotl agent-docs preference_tuning   # DPO, KTO, ORPO, SimPO
+axolotl agent-docs reward_modelling    # outcome and process reward models
+axolotl agent-docs pretraining         # continual pretraining
+axolotl agent-docs --list              # list all topics
+
+# Dump config schema for programmatic use
+axolotl config-schema
+axolotl config-schema --field adapter
+```
+
+If you're working with the source repo, agent docs are also available at `docs/agents/` and the project overview is in `AGENTS.md`.
+
 ## 🤝 Getting Help
 
 - Join our [Discord community](https://discord.gg/HhrNrHJPRb) for support
@@ -154,6 +219,13 @@ That's it! Check out our [Getting Started Guide](https://docs.axolotl.ai/docs/ge
 
 Contributions are welcome! Please see our [Contributing Guide](https://github.com/axolotl-ai-cloud/axolotl/blob/main/.github/CONTRIBUTING.md) for details.
 
+## 📈 Telemetry
+
+Axolotl has opt-out telemetry that helps us understand how the project is being used
+and prioritize improvements. We collect basic system information, model types, and
+error rates—never personal data or file paths. Telemetry is enabled by default. To
+disable it, set AXOLOTL_DO_NOT_TRACK=1. For more details, see our [telemetry documentation](https://docs.axolotl.ai/docs/telemetry.html).
+
 ## ❤️ Sponsors
 
 Interested in sponsoring? Contact us at [wing@axolotl.ai](mailto:wing@axolotl.ai)

VERSION

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

_quarto.yml

@@ -1,6 +1,8 @@
 project:
   type: website
-  pre-render: docs/scripts/generate_config_docs.py
+  pre-render:
+   - docs/scripts/generate_config_docs.py
+   - docs/scripts/generate_examples_docs.py
 
 quartodoc:
   dir: docs/api
@@ -126,11 +128,9 @@ quartodoc:
         - monkeypatch.mistral_attn_hijack_flash
         - monkeypatch.multipack
         - monkeypatch.relora
-        - monkeypatch.llama_expand_mask
         - monkeypatch.lora_kernels
         - monkeypatch.utils
         - monkeypatch.btlm_attn_hijack_flash
-        - monkeypatch.llama_patch_multipack
         - monkeypatch.stablelm_attn_hijack_flash
         - monkeypatch.trainer_fsdp_optim
         - monkeypatch.transformers_fa_utils
@@ -238,9 +238,51 @@ website:
         - section: "Getting Started"
           contents:
             - docs/getting-started.qmd
+            - docs/choosing_method.qmd
             - docs/installation.qmd
             - docs/inference.qmd
+            - section: "Model Guides"
+              contents:
+                - docs/models/kimi-linear.qmd
+                - docs/models/plano.qmd
+                - docs/models/mimo.qmd
+                - docs/models/internvl3_5.qmd
+                - docs/models/olmo3.qmd
+                - docs/models/trinity.qmd
+                - docs/models/arcee.qmd
+                - section: "Ministral3"
+                  contents:
+                    - docs/models/ministral3.qmd
+                    - docs/models/ministral3/think.qmd
+                    - docs/models/ministral3/vision.qmd
+                - section: "Magistral"
+                  contents:
+                    - docs/models/magistral.qmd
+                    - docs/models/magistral/think.qmd
+                    - docs/models/magistral/vision.qmd
+                - docs/models/ministral.qmd
+                - docs/models/mistral-small.qmd
+                - docs/models/voxtral.qmd
+                - docs/models/devstral.qmd
+                - docs/models/mistral.qmd
+                - docs/models/llama-4.qmd
+                - docs/models/llama-2.qmd
+                - docs/models/qwen3-next.qmd
+                - docs/models/qwen3.qmd
+                - docs/models/gemma3n.qmd
+                - docs/models/apertus.qmd
+                - docs/models/gpt-oss.qmd
+                - docs/models/seed-oss.qmd
+                - docs/models/phi.qmd
+                - docs/models/smolvlm2.qmd
+                - docs/models/granite4.qmd
+                - docs/models/LiquidAI.qmd
+                - docs/models/hunyuan.qmd
+                - docs/models/jamba.qmd
+                - docs/models/orpheus.qmd
+
             - docs/cli.qmd
+            - docs/telemetry.qmd
             - docs/config-reference.qmd
             - text: "API Reference"
               href: docs/api
@@ -261,12 +303,16 @@ website:
           contents:
             - docs/multimodal.qmd
             - docs/rlhf.qmd
+            - docs/grpo.qmd
+            - docs/ebft.qmd
+            - docs/vllm_serving.qmd
             - docs/reward_modelling.qmd
             - docs/lr_groups.qmd
             - docs/lora_optims.qmd
             - docs/dataset_loading.qmd
             - docs/qat.qmd
             - docs/quantize.qmd
+            - docs/optimizations.qmd
 
         - section: "Core Concepts"
           contents:
@@ -276,6 +322,7 @@ website:
             - docs/multipack.qmd
             - docs/mixed_precision.qmd
             - docs/optimizers.qmd
+            - docs/attention.qmd
 
         - section: "Advanced Features"
           contents:
@@ -285,12 +332,13 @@ website:
             - docs/custom_integrations.qmd
             - docs/sequence_parallelism.qmd
             - docs/gradient_checkpointing.qmd
-            - docs/moe_backends.md
             - docs/nd_parallelism.qmd
+            - docs/expert_quantization.qmd
 
         - section: "Troubleshooting"
           contents:
             - docs/faq.qmd
+            - docs/training_stability.qmd
             - docs/debugging.qmd
             - docs/nccl.qmd
 

benchmarks/bench_entropy.py

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

benchmarks/bench_scattermoe_lora.py

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

benchmarks/bench_selective_logsoftmax.py

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

cicd/Dockerfile-uv.jinja

@@ -11,7 +11,7 @@ ENV NIGHTLY_BUILD="{{ NIGHTLY_BUILD }}"
 ENV HF_HOME="{{ HF_HOME }}"
 
 RUN apt-get update && \
-    apt-get install -y --allow-change-held-packages vim curl nano libnccl2 libnccl-dev ibverbs-providers ibverbs-utils infiniband-diags librdmacm-dev librdmacm1 rdmacm-utils slurm-wlm
+    apt-get install -y --allow-change-held-packages vim curl nano zstd libnccl2 libnccl-dev ibverbs-providers ibverbs-utils infiniband-diags librdmacm-dev librdmacm1 rdmacm-utils slurm-wlm
 
 WORKDIR /workspace
 
@@ -31,7 +31,9 @@ RUN if [ "$NIGHTLY_BUILD" = "true" ] ; then \
         sed -i 's#^datasets.*#datasets @ git+https://github.com/huggingface/datasets.git@main#' requirements.txt; \
     fi
 
-RUN uv pip install packaging==23.2 setuptools==75.8.0
+RUN uv pip install packaging==26.0 setuptools==78.1.1
+RUN uv pip install torchvision
+RUN uv pip uninstall causal_conv1d
 RUN if [ "$AXOLOTL_EXTRAS" != "" ] ; then \
         uv pip install --no-build-isolation -e .[deepspeed,flash-attn,ring-flash-attn,optimizers,ray,$AXOLOTL_EXTRAS] $AXOLOTL_ARGS; \
     else \

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+PTX"
+ENV TORCH_CUDA_ARCH_LIST="7.0 7.5 8.0 8.6 8.7 8.9 9.0+PTX"
 ENV AXOLOTL_EXTRAS="{{ AXOLOTL_EXTRAS }}"
 ENV AXOLOTL_ARGS="{{ AXOLOTL_ARGS }}"
 ENV CUDA="{{ CUDA }}"
@@ -9,10 +9,10 @@ ENV GITHUB_REF="{{ GITHUB_REF }}"
 ENV GITHUB_SHA="{{ GITHUB_SHA }}"
 ENV NIGHTLY_BUILD="{{ NIGHTLY_BUILD }}"
 ENV HF_HOME="{{ HF_HOME }}"
-ENV AXOLOTL_DATASET_PROCESSES="8"
+ENV AXOLOTL_DATASET_NUM_PROC="8"
 
 RUN apt-get update && \
-    apt-get install -y --allow-change-held-packages vim curl nano libnccl2 libnccl-dev ibverbs-providers ibverbs-utils infiniband-diags librdmacm-dev librdmacm1 rdmacm-utils slurm-wlm
+    apt-get install -y --allow-change-held-packages vim curl nano zstd libnccl2 libnccl-dev ibverbs-providers ibverbs-utils infiniband-diags librdmacm-dev librdmacm1 rdmacm-utils slurm-wlm
 
 WORKDIR /workspace
 
@@ -32,7 +32,8 @@ 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
+RUN pip install packaging==26.0 setuptools==78.1.1 psutil
+RUN pip uninstall -y causal_conv1d
 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/cicd.sh

@@ -3,6 +3,24 @@ set -e
 
 python -c "import torch; assert '$PYTORCH_VERSION' in torch.__version__"
 
+set -o pipefail
+for i in 1 2 3; do
+  if curl --silent --show-error --fail -L \
+    https://axolotl-ci.b-cdn.net/hf-cache.tar.zst \
+    | tar -xpf - -C "${HF_HOME}/hub/" --use-compress-program unzstd --strip-components=1; then
+    echo "HF cache extracted successfully"
+    break
+  fi
+  echo "Attempt $i failed, cleaning up and retrying in 15s..."
+  rm -rf "${HF_HOME}/hub/"*
+  sleep 15
+done
+# hf download "NousResearch/Meta-Llama-3-8B"
+# hf download "NousResearch/Meta-Llama-3-8B-Instruct"
+# hf download "microsoft/Phi-4-reasoning"
+# hf download "microsoft/Phi-3.5-mini-instruct"
+# hf download "microsoft/Phi-3-medium-128k-instruct"
+
 # Run unit tests with initial coverage report
 pytest -v --durations=10 -n8 \
   --ignore=tests/e2e/ \

cicd/multigpu.py

@@ -17,7 +17,8 @@ template_loader = jinja2.FileSystemLoader(searchpath=cicd_path)
 template_env = jinja2.Environment(
     loader=template_loader, autoescape=select_autoescape()
 )
-df_template = template_env.get_template("Dockerfile.jinja")
+dockerfile = os.environ.get("E2E_DOCKERFILE", "Dockerfile.jinja")
+df_template = template_env.get_template(dockerfile)
 
 df_args = {
     "AXOLOTL_EXTRAS": os.environ.get("AXOLOTL_EXTRAS", ""),
@@ -27,8 +28,11 @@ df_args = {
     "CUDA": os.environ.get("CUDA", "126"),
     "GITHUB_REF": os.environ.get("GITHUB_REF", "refs/heads/main"),
     "GITHUB_SHA": os.environ.get("GITHUB_SHA", ""),
+    "NIGHTLY_BUILD": os.environ.get("NIGHTLY_BUILD", ""),
     "CODECOV_TOKEN": os.environ.get("CODECOV_TOKEN", ""),
     "HF_HOME": "/workspace/data/huggingface-cache/hub",
+    "PYTHONUNBUFFERED": os.environ.get("PYTHONUNBUFFERED", "1"),
+    "DEEPSPEED_LOG_LEVEL": os.environ.get("DEEPSPEED_LOG_LEVEL", "WARNING"),
 }
 
 dockerfile_contents = df_template.render(**df_args)

cicd/multigpu.sh

@@ -2,7 +2,7 @@
 set -e
 
 # Only run two tests at a time to avoid OOM on GPU (with coverage collection)
-pytest -v --durations=10 -n2 \
+pytest -v --durations=10 -n2 --maxfail=3 \
   --ignore=/workspace/axolotl/tests/e2e/multigpu/solo/ \
   --ignore=/workspace/axolotl/tests/e2e/multigpu/patched/ \
   /workspace/axolotl/tests/e2e/multigpu/ \

cicd/single_gpu.py

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

codecov.yml

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

devtools/dev_chat_template.yml

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

docker/Dockerfile

@@ -6,6 +6,7 @@ ARG AXOLOTL_EXTRAS=""
 ARG AXOLOTL_ARGS=""
 ARG CUDA="118"
 ARG PYTORCH_VERSION="2.1.2"
+ARG TARGETARCH
 
 ENV PYTORCH_VERSION=$PYTORCH_VERSION
 
@@ -20,13 +21,18 @@ RUN git clone --depth=1 https://github.com/axolotl-ai-cloud/axolotl.git
 
 WORKDIR /workspace/axolotl
 
-# If AXOLOTL_EXTRAS is set, append it in brackets
-RUN if [ "$AXOLOTL_EXTRAS" != "" ] ; then \
-        pip install --no-build-isolation -e .[deepspeed,flash-attn,ring-flash-attn,optimizers,ray,$AXOLOTL_EXTRAS] $AXOLOTL_ARGS; \
+# If AXOLOTL_EXTRAS is set, append it in brackets; don't install deepspeed with arm64
+RUN pip uninstall -y causal_conv1d
+RUN if [ "$TARGETARCH" = "arm64" ]; then \
+        BASE_EXTRAS="flash-attn,ring-flash-attn,optimizers,ray"; \
     else \
-        pip install --no-build-isolation -e .[deepspeed,flash-attn,ring-flash-attn,optimizers,ray] $AXOLOTL_ARGS; \
+        BASE_EXTRAS="deepspeed,flash-attn,ring-flash-attn,optimizers,ray"; \
     fi && \
-    python scripts/unsloth_install.py | sh && \
+    if [ "$AXOLOTL_EXTRAS" != "" ]; then \
+        pip install --no-build-isolation -e .[$BASE_EXTRAS,$AXOLOTL_EXTRAS] $AXOLOTL_ARGS; \
+    else \
+        pip install --no-build-isolation -e .[$BASE_EXTRAS] $AXOLOTL_ARGS; \
+    fi && \    python scripts/unsloth_install.py | sh && \
     python scripts/cutcrossentropy_install.py | sh && \
     pip install pytest && \
     pip cache purge

docker/Dockerfile-base

@@ -2,14 +2,16 @@ ARG CUDA_VERSION="11.8.0"
 ARG CUDNN_VERSION="8"
 ARG UBUNTU_VERSION="22.04"
 ARG MAX_JOBS=4
+ARG TARGETARCH
 
 FROM nvidia/cuda:$CUDA_VERSION-cudnn$CUDNN_VERSION-devel-ubuntu$UBUNTU_VERSION AS base-builder
 
 ENV PATH="/root/miniconda3/bin:${PATH}"
 
-ARG PYTHON_VERSION="3.10"
+ARG TARGETARCH
+ARG PYTHON_VERSION="3.11"
 ARG PYTORCH_VERSION="2.1.2"
-ARG CUDA="118"
+ARG CUDA="128"
 ARG TORCH_CUDA_ARCH_LIST="7.0 7.5 8.0 8.6 9.0+PTX"
 
 ENV PYTHON_VERSION=$PYTHON_VERSION
@@ -22,11 +24,17 @@ RUN apt-get update \
         librdmacm-dev librdmacm1 rdmacm-utils slurm-wlm \
     && rm -rf /var/cache/apt/archives \
     && rm -rf /var/lib/apt/lists/* \
-    && wget \
-    https://repo.anaconda.com/miniconda/Miniconda3-latest-Linux-x86_64.sh \
+    && if [ "$TARGETARCH" = "amd64" ]; then \
+        MINICONDA_ARCH="x86_64"; \
+    elif [ "$TARGETARCH" = "arm64" ]; then \
+        MINICONDA_ARCH="aarch64"; \
+    else \
+        echo "Unsupported architecture: $TARGETARCH"; exit 1; \
+    fi \
+    && wget https://repo.anaconda.com/miniconda/Miniconda3-latest-Linux-${MINICONDA_ARCH}.sh \
     && mkdir /root/.conda \
-    && bash Miniconda3-latest-Linux-x86_64.sh -b \
-    && rm -f Miniconda3-latest-Linux-x86_64.sh \
+    && bash Miniconda3-latest-Linux-${MINICONDA_ARCH}.sh -b \
+    && rm -f Miniconda3-latest-Linux-${MINICONDA_ARCH}.sh \
     && conda tos accept --override-channels --channel https://repo.anaconda.com/pkgs/main \
     && conda tos accept --override-channels --channel https://repo.anaconda.com/pkgs/r \
     && conda create -n "py${PYTHON_VERSION}" python="${PYTHON_VERSION}"
@@ -35,18 +43,34 @@ ENV PATH="/root/miniconda3/envs/py${PYTHON_VERSION}/bin:${PATH}"
 
 WORKDIR /workspace
 
-RUN python3 -m pip install --upgrade pip && pip3 install -U packaging==23.2 setuptools==75.8.0 wheel && \
+RUN python3 -m pip install --upgrade pip && pip3 install -U packaging==26.0 setuptools==75.8.0 wheel psutil && \
     python3 -m pip install --no-cache-dir -U torch==${PYTORCH_VERSION}+cu${CUDA} torchvision --extra-index-url https://download.pytorch.org/whl/cu$CUDA && \
-    CAUSAL_CONV1D_FORCE_CXX11_ABI=TRUE CAUSAL_CONV1D_FORCE_BUILD=TRUE python3 -m pip install --no-cache-dir causal_conv1d==1.5.2 && \
-    python3 -m pip install --no-cache-dir "mamba_ssm @ git+https://github.com/state-spaces/mamba.git@main" && \
     python3 -m pip cache purge
 
+RUN if [ "$CUDA" != "130" ] ; then \
+        CAUSAL_CONV1D_FORCE_CXX11_ABI=TRUE CAUSAL_CONV1D_FORCE_BUILD=TRUE python3 -m pip install --no-cache-dir "causal_conv1d @ git+https://github.com/Dao-AILab/causal-conv1d.git@v1.5.4"; \
+        python3 -m pip install --no-cache-dir "mamba_ssm @ git+https://github.com/state-spaces/mamba.git@main"; \
+        python3 -m pip cache purge; \
+    fi
+
 RUN git lfs install --skip-repo && \
     pip3 install awscli && \
     # The base image ships with `pydantic==1.8.2` which is not working
     pip3 install -U --no-cache-dir pydantic==1.10.10 && \
     pip3 cache purge
 
-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
+# Map Python version (e.g., 3.12 -> cp312)
+RUN PYTHON_CP="cp$(echo $PYTHON_VERSION | tr -d '.')" && \
+    # Map PyTorch version (e.g., 2.9.1 -> torch2.9, 2.10.0 -> torch2.10)
+    TORCH_TAG="torch$(echo $PYTORCH_VERSION | grep -oP '^\d+\.\d+')" && \
+    # Map architecture
+    case "$TARGETARCH" in \
+        amd64) ARCH_TAG="x86_64" ;; \
+        arm64) ARCH_TAG="aarch64" ;; \
+        *) echo "Unsupported architecture: $TARGETARCH"; exit 1 ;; \
+    esac && \
+    WHL_VERSION="v0.7.16" && \
+    WHL_FILE="flash_attn-2.8.3+cu${CUDA}${TORCH_TAG}-${PYTHON_CP}-${PYTHON_CP}-linux_${ARCH_TAG}.whl" && \
+    wget -nv "https://github.com/mjun0812/flash-attention-prebuild-wheels/releases/download/${WHL_VERSION}/${WHL_FILE}" && \
+    pip3 install --no-cache-dir "${WHL_FILE}" && \
+    rm "${WHL_FILE}"

docker/Dockerfile-base-nightly

@@ -30,7 +30,7 @@ ENV PATH="/root/miniconda3/envs/py${PYTHON_VERSION}/bin:${PATH}"
 
 WORKDIR /workspace
 
-RUN python3 -m pip install --upgrade pip && pip3 install -U packaging==23.2 setuptools==75.8.0 wheel && \
+RUN python3 -m pip install --upgrade pip && pip3 install -U packaging==26.0 setuptools==75.8.0 wheel && \
     python3 -m pip install --no-cache-dir -U torch --extra-index-url https://download.pytorch.org/whl/nightly/cu$CUDA && \
     python3 -m pip install --no-cache-dir "causal_conv1d @ git+https://github.com/Dao-AILab/causal-conv1d.git@main" && \
     python3 -m pip install --no-cache-dir "mamba_ssm @ git+https://github.com/state-spaces/mamba.git@main" && \

docker/Dockerfile-cloud-uv

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

docker/Dockerfile-uv

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

docker/Dockerfile-uv-base

@@ -2,9 +2,11 @@ ARG CUDA_VERSION="12.6.3"
 ARG CUDNN_VERSION=""
 ARG UBUNTU_VERSION="22.04"
 ARG MAX_JOBS=4
+ARG TARGETARCH
 
 FROM nvidia/cuda:$CUDA_VERSION-cudnn$CUDNN_VERSION-devel-ubuntu$UBUNTU_VERSION AS base-builder
 
+ARG TARGETARCH
 ARG PYTHON_VERSION="3.11"
 ARG PYTORCH_VERSION="2.6.0"
 ARG CUDA="126"
@@ -30,7 +32,26 @@ 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==${PYTORCH_VERSION} \
-    && uv pip install --no-build-isolation "causal_conv1d @ git+https://github.com/Dao-AILab/causal-conv1d.git@main" \
-    && uv pip install "mamba_ssm @ git+https://github.com/state-spaces/mamba.git@main" \
+    && uv pip install torch==${PYTORCH_VERSION} torchvision \
     && uv pip install awscli pydantic
+
+RUN if [ "$TARGETARCH" = "amd64" ]; then \
+        MAMBA_SKIP_CUDA_BUILD=TRUE CAUSAL_CONV1D_SKIP_CUDA_BUILD=TRUE uv pip install --no-build-isolation mamba_ssm causal_conv1d; \
+    fi
+
+# Map Python version (e.g., 3.12 -> cp312)
+RUN PYTHON_CP="cp$(echo $PYTHON_VERSION | tr -d '.')" && \
+    # Map PyTorch version (e.g., 2.9.1 -> torch2.9, 2.10.0 -> torch2.10)
+    TORCH_TAG="torch$(echo $PYTORCH_VERSION | grep -oP '^\d+\.\d+')" && \
+    LINUX_TAG="manylinux_" && \
+    # Map architecture
+    case "$TARGETARCH" in \
+        amd64) ARCH_TAG="2_24_x86_64.manylinux_2_28_x86_64" ;; \
+        arm64) ARCH_TAG="2_34_aarch64" ;; \
+        *) echo "Unsupported architecture: $TARGETARCH"; exit 1 ;; \
+    esac && \
+    WHL_VERSION="v0.7.16" && \
+    WHL_FILE="flash_attn-2.8.3+cu${CUDA}${TORCH_TAG}-${PYTHON_CP}-${PYTHON_CP}-${LINUX_TAG}${ARCH_TAG}.whl" && \
+    wget -nv "https://github.com/mjun0812/flash-attention-prebuild-wheels/releases/download/${WHL_VERSION}/${WHL_FILE}" && \
+    uv pip install --no-cache-dir "${WHL_FILE}" && \
+    rm "${WHL_FILE}"

docs/.gitignore

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

docs/agents/grpo.md

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

docs/agents/model_architectures.md

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

docs/agents/new_model_support.md

@@ -0,0 +1,181 @@
+# New Model Support — Agent Reference
+
+Guide for debugging and adding support for new model architectures in axolotl. Based on lessons learned from Gemma4, Gemma3, Qwen2-VL, and other multimodal/MoE models.
+
+## Quick Validation Checklist
+
+When testing a new model, run through these checks in order:
+
+1. **Does the model load?** `axolotl preprocess config.yaml` — catches config schema errors
+2. **Does LoRA apply?** Check for "Unsupported layer type" warnings from PEFT
+3. **Is the initial loss sane?** First-step loss for a pretrained model should be 0.5–2.0 for SFT
+4. **Does sample packing work?** Compare loss with `sample_packing: true` vs `false` — should be similar
+5. **Is CCE active?** Check for "Applying Cut Cross Entropy" log and verify peak VRAM is lower
+
+## Loss Debugging
+
+### Expected initial loss
+A pretrained model doing SFT should start with loss roughly in the 0.5–2.0 range. If loss starts above 3.0, something is wrong. If it's near `log(vocab_size)` (≈ 12 for 262K vocab), the model is predicting at random — attention masking or model weights are broken.
+
+### Direct comparison technique
+The fastest way to isolate a loss issue — bypass the trainer entirely:
+
+```python
+# Load model via axolotl's pipeline (applies all patches)
+from axolotl.cli.config import load_cfg
+from axolotl.utils.config import normalize_config, prepare_plugins
+from axolotl.loaders.tokenizer import load_tokenizer
+from axolotl.loaders.model import ModelLoader
+
+cfg = load_cfg("your_config.yaml")
+normalize_config(cfg)
+prepare_plugins(cfg)
+tokenizer = load_tokenizer(cfg)
+model, _ = ModelLoader(cfg, tokenizer).load()
+
+# Forward pass on preprocessed data
+model.train()
+out = model(input_ids, labels=labels)
+print(f"Direct loss: {out.loss.item()}")  # Compare to trainer's reported loss
+```
+
+If direct loss is correct (~1.0) but trainer reports 3–4x higher, check `model_accepts_loss_kwargs` (see below).
+
+### `model_accepts_loss_kwargs` inflation
+HF Trainer checks if the model's `forward()` has `**kwargs` and sets `model_accepts_loss_kwargs=True`. This changes loss normalization: the trainer does NOT divide loss by `gradient_accumulation_steps` before logging. The gradient is correct — only the logged loss is inflated.
+
+**Symptom**: Logged loss ≈ actual_loss × gradient_accumulation_steps.
+
+**Which models are affected**: Any model with `**kwargs` in forward (common in multimodal models for extra inputs like `mm_token_type_ids`, `pixel_values`, etc.).
+
+**Fix location**: `src/axolotl/core/trainers/base.py` `__init__()` — after `super().__init__()`, check if the unwrapped model actually has `num_items_in_batch` in its forward signature. If not, set `self.model_accepts_loss_kwargs = False`.
+
+## Multimodal Models (ForConditionalGeneration)
+
+Many recent models use `ForConditionalGeneration` as the top-level class, not `ForCausalLM`:
+- Gemma3 → `Gemma3ForConditionalGeneration`
+- Gemma4 → `Gemma4ForConditionalGeneration`
+- Qwen2-VL → `Qwen2VLForConditionalGeneration`
+- LLaVA → `LlavaForConditionalGeneration`
+
+### Why this matters
+
+| Component | Targets `ForCausalLM` | Needs `ForConditionalGeneration` |
+|-----------|----------------------|--------------------------------|
+| CCE patches | ✅ (default) | ❌ silently inactive if not patched |
+| PEFT LoRA | ✅ | May fail on custom layer types |
+| HF Trainer label handling | ✅ | May need extra inputs |
+
+### Required extra inputs
+Multimodal models require special inputs during training even for text-only data:
+
+| Model | Required Input | Value for Text-Only |
+|-------|---------------|-------------------|
+| Gemma4 | `mm_token_type_ids` | `torch.zeros_like(input_ids)` |
+| Gemma3 | `token_type_ids` | `torch.zeros_like(input_ids)` |
+
+Auto-inject in `compute_loss()` when not provided by the data collator. See `core/trainers/base.py`.
+
+### Custom layer types and PEFT
+Vision towers often use custom module wrappers that PEFT doesn't support:
+
+| Model | Custom Layer | Wraps | Fix |
+|-------|-------------|-------|-----|
+| Gemma4 | `Gemma4ClippableLinear` | `nn.Linear` | Redirect to `.linear` child |
+
+Fix location: `src/axolotl/loaders/adapter.py` `_patch_peft_clippable_linear()`.
+
+## Sample Packing
+
+### How packed sequence detection works (transformers ≥ 5.x)
+`transformers.masking_utils._preprocess_mask_arguments()` detects packed sequences from `position_ids` resets. But **only when `attention_mask is None`**:
+
+```python
+# From masking_utils.py:
+if position_ids is not None and attention_mask is None and past_key_values is None:
+    packed_sequence_mask = find_packed_sequence_indices(position_ids)
+```
+
+If the collator provides an all-ones `attention_mask`, packing detection is **skipped** and the model builds a single causal mask spanning all packed sequences → cross-sequence attention leakage → very high loss.
+
+### Fix for models using `create_causal_mask_mapping`
+For Gemma3, Gemma4, and similar models that use the new transformers masking system, remove `attention_mask` from inputs when sample packing is active:
+
+```python
+# In compute_loss():
+if (
+    self.args.sample_packing
+    and model_type in ("gemma4", "gemma3")
+    and "attention_mask" in inputs
+    and "position_ids" in inputs
+):
+    del inputs["attention_mask"]
+```
+
+Fix location: `src/axolotl/core/trainers/base.py` `compute_loss()`.
+
+### Models that DON'T need this fix
+Older models that use `_prepare_4d_causal_attention_mask` (Llama, Mistral, Qwen2, etc.) handle sample packing via axolotl's multipack attention monkeypatch instead. Only models using the new `create_causal_mask_mapping` / `create_causal_mask` masking system need the `attention_mask` removal.
+
+## Attention Backend Selection
+
+| Backend | Config | head_dim limit | torch_compile | Notes |
+|---------|--------|---------------|---------------|-------|
+| FA2 | `flash_attention: true` | 256 | ✅ | Fastest when supported |
+| FA4 | auto with `flash_attention: true` | 256 (SM90+) | ✅ | Auto-detected on H100+ |
+| SDPA | `sdp_attention: true` | None | ✅ | Universal fallback |
+| flex | `flex_attention: true` | None | ⚠️ Triton OOM for large head_dim | Good for variable head dims |
+| eager | neither set | None | ✅ | Slowest, always works |
+
+**Check model support**: Look at `_supports_flash_attn_2`, `_supports_flex_attn`, `_supports_sdpa` attributes on the model class.
+
+**head_dim gotcha**: The 256 limit is specific to flash-attn CUDA kernels, NOT PyTorch-level. SDPA and flex_attention both handle arbitrary head_dim. Models with `global_head_dim > 256` (Gemma4: 512) must use SDPA or flex.
+
+**flex + compile gotcha**: `torch_compile` with flex_attention can hit Triton shared memory OOM for large head_dim. Falls back to eager per-function (not a crash, but slower). Unsloth disables flex for Gemma4 for this reason.
+
+## Cut Cross Entropy (CCE)
+
+### How CCE patches work
+CCE replaces the model's `forward()` with a fused version that computes loss from hidden states + lm_head weight without materializing the full logits tensor. This saves ~`batch × seq_len × vocab_size × dtype_bytes` of VRAM.
+
+### Adding CCE for a new model
+1. Check if the model type is in `cut_cross_entropy.transformers.patch.PATCH_FNS`
+2. If not, axolotl's generic fallback (`integrations/cut_cross_entropy/__init__.py` `patch_llama_like()`) patches `{Prefix}ForCausalLM.forward` with `cce_forward`
+3. For multimodal models (`ForConditionalGeneration`), a model-specific patch is needed in `ml-cross-entropy` repo
+4. The multimodal `cce_forward` must accept all extra kwargs (pixel_values, mm_token_type_ids, etc.) and pop any that would conflict before calling `self.model()`
+
+### Common CCE pitfall
+If CCE appears active (log says "Applying Cut Cross Entropy") but peak VRAM doesn't decrease, check which class was patched. If the model loads as `ForConditionalGeneration` but CCE patched `ForCausalLM`, the patch is silently inactive.
+
+## MoE Models
+
+### Dense MLP vs MoE experts
+Some MoE models (e.g., Gemma4) have BOTH dense MLP layers and MoE expert layers at every decoder layer:
+- `gate_proj/up_proj/down_proj` → targets the **dense MLP** (`Gemma4TextMLP`)
+- `experts.gate_up_proj/experts.down_proj` → targets the **MoE experts** (`Gemma4TextExperts`)
+
+LoRA on the dense MLP works normally. Expert LoRA via `lora_target_parameters` requires PEFT support for the specific expert module type (may warn "Unsupported layer type").
+
+### ScatterMoE kernels
+`use_scattermoe: true` with `experts_implementation: scattermoe` registers fused expert kernels via transformers' `ExpertsInterface`. Significant speedup for MoE models. Requires the kernels plugin:
+```yaml
+plugins:
+  - axolotl.integrations.kernels.KernelsPlugin
+use_kernels: true
+use_scattermoe: true
+experts_implementation: scattermoe
+```
+
+## Where to Add Model-Specific Fixes
+
+| What | Where | Example |
+|------|-------|---------|
+| Missing forward inputs | `core/trainers/base.py` `compute_loss()` | mm_token_type_ids injection |
+| Attention mask fixes | `core/trainers/base.py` `compute_loss()` | Sample packing mask removal |
+| Loss logging fixes | `core/trainers/base.py` `__init__()` | model_accepts_loss_kwargs override |
+| PEFT/LoRA patches | `loaders/adapter.py` | ClippableLinear redirect |
+| Attention patches | `monkeypatch/attention/` | FA4 tuple fix |
+| Model-specific patches | `loaders/patch_manager.py` `_apply_model_specific_patches()` | Llama4, Kimi, NemotronH |
+| CCE patches | `ml-cross-entropy` repo `transformers/` | Per-model cce_forward |
+| Example configs | `examples/<model>/` | Validated YAML |
+| Config validation | `utils/schemas/validation.py` | Compatibility checks |

docs/agents/preference_tuning.md

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

docs/agents/pretraining.md

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

docs/agents/reward_modelling.md

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

docs/agents/sft.md

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

docs/amd_hpc.qmd

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

docs/attention.qmd

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

docs/checkpoint_saving.qmd

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

docs/choosing_method.qmd

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

docs/cli.qmd

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

docs/dataset-formats/conversation.qmd

@@ -108,6 +108,14 @@ datasets:
     type: chat_template
 ```
 
+::: {.callout-tip}
+`chat_template_jinja` also accepts a file path to a `.jinja2` file instead of an inline string:
+
+```yaml
+chat_template_jinja: ./path/to/my_template.jinja2
+```
+:::
+
 ::: {.callout-important}
 Please make sure that your `tokenizer.eos_token` is same as EOS (End-of-Sequence) token in template. Otherwise, set `eos_token` under `special_tokens: `.
 :::
@@ -212,6 +220,21 @@ 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
@@ -279,6 +302,113 @@ datasets:
 It is not necessary to set both `message_field_training` and `message_field_training_detail` at once.
 :::
 
+#### Content parts with per-part training control
+
+Instead of using character offsets with `train_detail`, you can split a message's content into a list of parts, each with its own training flag. This is useful when you want to mask specific sections of a response (e.g., mask reasoning but train on the answer).
+
+```{.json filename="data.jsonl"}
+{
+  "messages": [
+    {"role": "user", "content": [{"type": "text", "text": "What is 2+2?"}]},
+    {
+      "role": "assistant",
+      "content": [
+        {"type": "text", "text": "Let me think step by step...", "train": false},
+        {"type": "text", "text": " The answer is 4.", "train": true}
+      ]
+    }
+  ]
+}
+```
+
+The configuration is the same as standard `chat_template` — no extra fields needed:
+
+```yaml
+datasets:
+  - path: ...
+    type: chat_template
+    roles_to_train: ["assistant"]
+```
+
+Each content part supports:
+
+- `type`: `"text"` (required)
+- `text`: the text value (also accepts `content` or `value` as the key)
+- `train`: `true`/`false` (optional) — whether to train on this part
+- `weight`: `0`/`1` (optional) — alternative to `train`
+
+If a part has no `train` or `weight` flag, it inherits the turn-level training decision (from `roles_to_train`, `message_field_training`, or `train_on_inputs`).
+
+::: {.callout-warning title="Whitespace at part boundaries"}
+BPE tokenizers (used by Llama, Qwen, Mistral, GPT, etc.) prepend spaces to word tokens. For example, `" answer"` is a single token — the space is part of it. This means **where you place whitespace between content parts matters**:
+
+**Split BEFORE spaces** (space goes with the next part):
+
+```json
+[
+  {"type": "text", "text": "Let me think...", "train": false},
+  {"type": "text", "text": " The answer is 4.", "train": true}
+]
+```
+
+**DON'T put trailing spaces** on a part (the space merges with the next word into one token that straddles the boundary, and straddling tokens are masked):
+
+```json
+[
+  {"type": "text", "text": "Let me think... ", "train": false},
+  {"type": "text", "text": "The answer is 4.", "train": true}
+]
+```
+
+In the bad example, `" The"` becomes a single token that spans both parts. Because it straddles the boundary, it is conservatively **masked** (not trained) — even though the second part has `train: true`.
+
+**Newlines** typically merge with preceding punctuation (e.g., `":\n"` is one token). Keep newlines with the preceding part:
+
+```json
+[
+  {"type": "text", "text": "Thinking:\n", "train": false},
+  {"type": "text", "text": "The answer is 4.", "train": true}
+]
+```
+
+Axolotl will log a warning if it detects trailing whitespace at a boundary between parts with different training flags.
+:::
+
+::: {.callout-note}
+When all content parts in a message are strings, they are concatenated before being passed to the chat template. This means content parts work with **any** Jinja template — the template sees a plain string, and the per-part training flags are applied during tokenization.
+:::
+
+##### Per-part training on reasoning_content
+
+For templates that support a separate `reasoning_content` field (e.g., `qwen3`), the same content-parts format works on `reasoning_content`. This is useful for masking incorrect reasoning steps while training on self-corrections:
+
+```{.json filename="data.jsonl"}
+{
+  "messages": [
+    {"role": "user", "content": [{"type": "text", "text": "What is 2+2?"}]},
+    {
+      "role": "assistant",
+      "reasoning_content": [
+        {"type": "text", "text": "Hmm maybe 2+2=5.", "train": false},
+        {"type": "text", "text": " Wait no, 2+2=4.", "train": true}
+      ],
+      "content": [
+        {"type": "text", "text": "The answer is 4.", "train": true}
+      ]
+    }
+  ]
+}
+```
+
+The `reasoning_content` and `content` fields are handled independently — each has its own token boundaries and per-part masking. No additional configuration is needed beyond what the template already requires.
+
+::: {.callout-tip}
+When `reasoning_content` is provided as a separate field, `split_thinking` is not needed — the reasoning is already separated from the content in the data.
+:::
+
+The same whitespace rules apply to `reasoning_content` parts as to `content` parts — split before spaces, keep newlines with the preceding part.
+
+
 #### Reasoning split
 
 (For Qwen3 template only) Enable reasoning split, where the reasoning is split from the content and passed as a separate field into the template.

docs/dataset-formats/index.qmd

@@ -22,89 +22,46 @@ For `pretraining_dataset:` specifically, please refer to the [Pre-training secti
 
 ## Pre-training
 
-When aiming to train on large corpora of text datasets, pre-training is your go-to choice. Due to the size of these datasets, downloading the entire-datasets before beginning training would be prohibitively time-consuming. Axolotl supports [streaming](https://huggingface.co/docs/datasets/en/stream) to only load batches into memory at a time.
-
-A sample format for a pre-training dataset is as follows:
+Pre-training trains on raw text corpora with no input masking. The dataset format is simple:
 
 ```json
 {"text": "first row"}
 {"text": "second row"}
-...
 ```
 
-It is typically recommended to save your dataset as `.jsonl` due to its flexibility and simplicity.
+Axolotl supports two approaches:
 
-Axolotl supports loading from a Hugging Face hub repo or from local files.
+### Streaming (large datasets)
 
-### Pre-training from Hugging Face hub datasets
-
-As an example, to train using a Hugging Face dataset `hf_org/name`, you can pass the following config:
-
-```yaml
-pretraining_dataset: hf_org/name
-```
-
-### Pre-training from local dataset files
-
-Given a few corpus files: `A.jsonl`, `B.jsonl`, and `C.jsonl`, your config will look like the below:
+For large corpora that don't fit in memory, use `pretraining_dataset` with [streaming](../streaming.qmd). Data is tokenized on-demand during training.
 
 ```yaml
 pretraining_dataset:
-  - path: json
-    data_files:
-      - A.jsonl
-      - B.jsonl
-      - C.jsonl
-```
-
-While we recommend `.jsonl`, you can also use the other formats (`csv`, `parquet`, `arrow`, `SQL`, `Webdataset`) that are supported by [`Dataset.load_dataset`](https://huggingface.co/docs/datasets/loading#local-and-remote-files)
-
-### Pre-training without 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.
-
-From Hugging Face:
-
-```yaml
-datasets:
-  - path: hf_org/name
-    type: completion
-```
-
-From local files:
-
-```yaml
-datasets:
-  - path: A.jsonl
-    type: completion
-
-  - path: B.jsonl
-    type: completion
+  - path: HuggingFaceFW/fineweb-edu
+    type: pretrain
+    text_column: text
+    split: train
 ```
 
 ::: {.callout-important}
-For `completion` only, Axolotl would split texts if it exceeds the context length into multiple smaller prompts. If you are interested in having this for `pretraining_dataset` too, please let us know or help make a PR!
+Streaming requires `max_steps` in your config — Axolotl cannot infer the dataset size. One step = `sequence_len * micro_batch_size * gradient_accumulation_steps * num_gpus` tokens.
 :::
 
-### Pre-training dataset configuration tips
+See [Streaming Datasets](../streaming.qmd) for full configuration details.
 
-#### Setting max_steps
+### Non-streaming (smaller datasets)
 
-When using streaming for large datasets, Axolotl does not know in advance how large the dataset is and does not know when to stop.
+For datasets that fit in memory, use `type: completion` under `datasets:`. The entire dataset is pre-tokenized before training, which can be done on a CPU-only machine.
 
-Therefore, it is necessary to set `max_steps: int` in your config for pre-training to run, so that Axolotl knows when to stop training.
+```yaml
+datasets:
+  - path: my_corpus
+    type: completion
+```
 
-One step is equal to `sequence_len * micro_batch_size * gradient_accumulation_steps * total_num_gpus` tokens.
-
-#### Group_by_length
-
-It is recommended to leave this off if downloading from Hugging Face hub as it would download the entire dataset which can be very large.
-
-### Reference
-
-Please see docs [here](pretraining.qmd).
+::: {.callout-note}
+With `completion`, texts exceeding `sequence_len` are split into multiple samples automatically.
+:::
 
 ## Supervised fine-tuning (SFT)
 

docs/dataset-formats/pretraining.qmd

@@ -4,29 +4,9 @@ description: Data format for a pre-training completion task.
 order: 1
 ---
 
-For pretraining, there is no prompt template or roles.  The only required field is `text`:
-
-```{.json filename="data.jsonl"}
-{"text": "first row"}
-{"text": "second row"}
-...
-```
-
-:::{.callout-note}
-
-### Streaming is recommended for large datasets
-
-Axolotl usually loads the entire dataset into memory. This will be challenging for large datasets. Use the following config to enable streaming:
-
-```{.yaml filename="config.yaml"}
-pretraining_dataset:
-  - name:
-    path:
-    split:
-    text_column: # column in dataset with the data, usually `text`
-    type: pretrain
-    trust_remote_code:
-    skip: # number of rows of data to skip over from the beginning
-```
+::: {.callout-note}
+Pre-training documentation has been consolidated:
 
+- **Streaming pretraining** (large datasets): See [Streaming Datasets](../streaming.qmd#pretraining-with-streaming)
+- **Non-streaming pretraining** (`type: completion`): See [Dataset Formats](index.qmd#pre-training)
 :::

docs/debugging.qmd

@@ -6,6 +6,10 @@ description: How to debug Axolotl
 
 This document provides some tips and tricks for debugging Axolotl.  It also provides an example configuration for debugging with VSCode.  A good debugging setup is essential to understanding how Axolotl code works behind the scenes.
 
+::: {.callout-tip}
+For training-specific debugging (loss spikes, NaN gradients, OOM errors, RL training stability), see [Training Stability & Debugging](training_stability.qmd).
+:::
+
 ## Table of Contents
 
 - [General Tips](#general-tips)
@@ -29,7 +33,7 @@ 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_processes: 1` in your axolotl config or run the training command with `--dataset_processes=1`.
+    - Set `dataset_num_proc: 1` in your axolotl config or run the training command with `--dataset_num_proc=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
@@ -85,7 +89,7 @@ If you developing on a remote host, you can easily use VSCode to debug remotely.
 
 The easiest way to get started is to modify the [.vscode/launch.json](../.vscode/launch.json) file in this project.  This is just an example configuration, so you may need to modify or copy it to suit your needs.
 
-For example, to mimic the command `cd devtools && CUDA_VISIBLE_DEVICES=0 accelerate launch -m axolotl.cli.train dev_chat_template.yml`, you would use the below configuration[^1].  Note that we add additional flags that override the axolotl config and incorporate the tips above (see the comments). We also set the working directory to `devtools` and set the `env` variable `HF_HOME` to a temporary folder that is later partially deleted.  This is because we want to delete the HF dataset cache before each run in order to ensure that the data preprocessing code is run from scratch.
+For example, to mimic the command `cd devtools && CUDA_VISIBLE_DEVICES=0 axolotl train dev_chat_template.yml`, you would use the below configuration[^1].  Note that we add additional flags that override the axolotl config and incorporate the tips above (see the comments). We also set the working directory to `devtools` and set the `env` variable `HF_HOME` to a temporary folder that is later partially deleted.  This is because we want to delete the HF dataset cache before each run in order to ensure that the data preprocessing code is run from scratch.
 
 ```json
 // .vscode/launch.json
@@ -101,7 +105,7 @@ 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_processes=1",      // limits data preprocessing to one process
+                "--dataset_num_proc=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
@@ -242,6 +246,6 @@ style="border-radius: 10px; display: block; margin: auto;" width="560" height="3
 </div>
 <br>
 
-[^1]: The config actually mimics the command `CUDA_VISIBLE_DEVICES=0 python -m accelerate.commands.launch -m axolotl.cli.train devtools/chat_template.yml`, but this is the same thing.
+[^1]: The VSCode config uses `accelerate.commands.launch` as the Python module entry point, which is what `axolotl train` invokes under the hood.
 
 [^2]: Many of the below flags are recommended best practices by Nvidia when using nvidia-container-toolkit.  You can read more about these flags [here](https://docs.nvidia.com/deeplearning/frameworks/user-guide/index.html).

docs/docker.qmd

@@ -32,11 +32,8 @@ main-base-py{python_version}-cu{cuda_version}-{pytorch_version}
 
 Tags examples:
 
-- `main-base-py3.11-cu128-2.7.1`
-- `main-base-py3.11-cu126-2.7.1`
-- `main-base-py3.11-cu126-2.7.0`
-- `main-base-py3.11-cu126-2.6.0`
-- `main-base-py3.11-cu124-2.6.0`
+- `main-base-py3.11-cu128-2.8.0`
+- `main-base-py3.11-cu128-2.9.1`
 
 ## Main
 
@@ -74,15 +71,12 @@ There may be some extra tags appended to the image, like `-vllm` which installs
 
 Tags examples:
 
-- `main-py3.11-cu128-2.7.1`
-- `main-py3.11-cu126-2.7.1`
-- `main-py3.11-cu126-2.7.0`
-- `main-py3.11-cu126-2.6.0`
-- `main-py3.11-cu124-2.6.0`
+- `main-py3.11-cu128-2.8.0`
+- `main-py3.11-cu128-2.9.1`
 - `main-latest`
 - `main-20250303-py3.11-cu124-2.6.0`
 - `main-20250303-py3.11-cu126-2.6.0`
-- `0.10.1`
+- `0.12.0`
 
 ## Cloud
 

docs/ebft.qmd

@@ -0,0 +1,556 @@
+---
+title: "EBFT Training"
+description: "Energy-Based Fine-Tuning uses feature-matching rewards from internal representations to train language models without external reward functions."
+order: 9
+back-to-top-navigation: true
+toc: true
+toc-expand: 2
+toc-depth: 4
+---
+
+## Overview
+
+Energy-Based Fine-Tuning (EBFT) is a training method that optimizes language models by matching the **internal feature representations** of generated text to those of ground-truth completions. Instead of relying on external reward models or hand-crafted reward functions, EBFT extracts hidden states from intermediate layers of a frozen copy of the model and uses cosine similarity between generated and reference features as the reward signal.
+
+Paper: ["Matching Features, Not Tokens: Energy-Based Fine-Tuning of Language Models"](https://arxiv.org/abs/2603.12248) (Jelassi et al., 2026)
+
+### How EBFT Differs from Other RL Methods
+
+| Method | Reward Signal | Requires | Best For |
+|--------|--------------|----------|----------|
+| **GRPO** | External reward function(s) | Custom reward code or reward model | Tasks with verifiable answers (math, code) |
+| **DPO** | Preference pairs (chosen vs rejected) | Paired preference data | Alignment with human preferences |
+| **EBFT** | Feature similarity to ground truth | Ground-truth completions | Any task with reference outputs |
+
+EBFT's key advantage is that it needs only ground-truth completions -- no reward engineering, no preference annotation, and no reward model training. The model's own internal representations serve as the reward signal. This makes it particularly effective for:
+
+- Code generation (match features of known-good solutions)
+- Instruction following with reference outputs
+- Continual pretraining on unstructured text (strided mode)
+- Multi-turn dialogue with reference conversations
+
+### Reward Formulation
+
+The EBFT reward for each generated completion is:
+
+```
+reward = alignment_coef * cosine_similarity(gen_features, gt_features)
+       - diversity_coef * mean_pairwise_similarity(gen_features)
+```
+
+- **Alignment**: How closely the generated output's internal representations match the ground truth. Higher is better.
+- **Diversity**: Penalizes generated samples that are too similar to each other (prevents mode collapse). Lower is better.
+- **CFM loss** (Cross-Feature Matching): Tracks `||mean(gen_features) - gt_features||^2` as a diagnostic. This is the quantity that EBFT ultimately minimizes.
+
+## Modes
+
+EBFT supports three operational modes, each suited to different use cases.
+
+### Structured Mode (Sync)
+
+Uses vLLM on a separate GPU for generation, with sequential generate-score-train steps. This is the simplest mode and recommended for getting started.
+
+```
+GPU 0: vLLM Server (generates completions, receives weight syncs)
+GPU 1: Trainer (feature extraction, reward computation, GRPO training)
+```
+
+**When to use**: Standard instruction-following or QA datasets where you have prompt/completion pairs. Requires 2 GPUs.
+
+### Structured Mode (Async)
+
+Same architecture as sync, but overlaps generation of the next batch with training on the current batch. Faster throughput at the cost of slightly stale weights during generation.
+
+**When to use**: Same data as sync mode, but when you want faster training and can tolerate weight staleness (controlled by `vllm_sync_interval`).
+
+### Strided Mode
+
+Runs entirely on a single GPU with no vLLM dependency. Places anchor points throughout a document and generates short rollouts at each anchor using block-parallel attention patterns.
+
+```
+Single GPU: Base model + LoRA adapter
+  - Strided block-parallel generation (flex_attention)
+  - Feature extraction via disable_adapter()
+  - No vLLM needed
+```
+
+**When to use**: Unstructured text data (raw code, prose, documents) where there is no natural prompt/completion split. Also works with structured data that includes prompt boundaries. Requires only 1 GPU.
+
+## Quick Start
+
+### Structured Mode
+
+This minimal example fine-tunes Qwen2-0.5B on code data using EBFT with vLLM generation.
+
+**Step 1**: Create a config file `ebft_quickstart.yaml`:
+
+```yaml
+base_model: Qwen/Qwen2-0.5B-Instruct
+
+rl: ebft
+
+ebft:
+  feature_layers: [0.25, 0.5, 0.75]
+  embed_method: last_token
+  alignment_coef: 1.0
+  diversity_coef: 1.0
+
+trl:
+  num_generations: 4
+  max_completion_length: 256
+  temperature: 0.7
+  use_vllm: true
+  vllm_server_host: 0.0.0.0
+  vllm_server_port: 8000
+  vllm_lora_sync: true
+  vllm_sync_interval: 3
+  use_data_producer: true
+  async_prefetch: false
+  scale_rewards: true
+  loss_type: grpo
+
+vllm:
+  gpu_memory_utilization: 0.5
+  max_model_len: 1024
+
+datasets:
+  - path: nvidia/OpenCodeInstruct
+    type: ebft_opencode.transform
+    split: train[:500]
+
+# Standard training settings (see getting-started.qmd for details)
+adapter: lora
+lora_r: 16
+lora_alpha: 32
+lora_target_linear: true
+sequence_len: 1024
+micro_batch_size: 2
+gradient_accumulation_steps: 4
+max_steps: 20
+learning_rate: 5.0e-6
+bf16: auto
+flash_attention: true
+gradient_checkpointing: true
+output_dir: ./outputs/ebft-quickstart
+```
+
+**Step 2**: Start vLLM on GPU 0:
+
+```bash
+CUDA_VISIBLE_DEVICES=0 axolotl vllm-serve ebft_quickstart.yaml
+```
+
+**Step 3**: Wait approximately 30 seconds for vLLM to initialize, then start training on GPU 1:
+
+```bash
+CUDA_VISIBLE_DEVICES=1 axolotl train ebft_quickstart.yaml
+```
+
+::: {.callout-important}
+The `micro_batch_size` must be divisible by `num_generations`. For example, with `num_generations: 4`, valid values are 4, 8, 12, etc.
+:::
+
+### Dataset Format
+
+Structured mode datasets must produce two fields after the transform:
+
+- `prompt`: Either a string or a list of chat messages (`[{"role": "user", "content": "..."}]`)
+- `ground_truth`: A string containing the reference completion
+
+Example raw dataset row:
+
+```json
+{
+  "input": "Write a function to compute fibonacci numbers.",
+  "output": "def fibonacci(n):\n    if n <= 1:\n        return n\n    return fibonacci(n-1) + fibonacci(n-2)"
+}
+```
+
+The `ebft_opencode.transform` converts this to the required `{prompt, ground_truth}` format automatically.
+
+## Feature Extraction
+
+EBFT extracts hidden states from intermediate transformer layers and pools them into per-sequence embeddings. These embeddings are compared between generated and ground-truth completions to compute rewards.
+
+### Feature Layers
+
+The `feature_layers` parameter specifies which layers to extract, as fractions of total model depth:
+
+```yaml
+ebft:
+  feature_layers: [0.25, 0.5, 0.75]  # Quarter, middle, three-quarter depth
+```
+
+For a 32-layer model, this extracts layers 8, 16, and 24. The hidden states from all selected layers are concatenated along the feature dimension, producing embeddings of size `num_layers * hidden_dim`.
+
+::: {.callout-tip}
+Using multiple layers captures both low-level syntactic features (early layers) and high-level semantic features (later layers). The default `[0.25, 0.5, 0.75]` works well across model sizes.
+:::
+
+### Embed Methods
+
+The `embed_method` controls how per-token hidden states are pooled into a single vector per sequence:
+
+| Method | Description | Output Shape | Notes |
+|--------|-------------|-------------|-------|
+| `last_token` | Hidden state at the last non-padding token | `(B, D)` | Default. Good for autoregressive models where the last token summarizes the sequence. |
+| `mean_pooling` | Mean of all non-padding token states | `(B, D)` | Considers the entire sequence equally. |
+| `completion_mean` | Mean over completion tokens only (excludes prompt) | `(B, D)` | Focuses reward signal on generated content. Requires prompt length information. |
+| `concat` | Concatenation of states at 25%, 50%, 75% positions | `(B, 3*D)` | Captures positional structure. Higher dimensional. |
+
+```yaml
+ebft:
+  embed_method: completion_mean  # Focus on completion features
+```
+
+### SVD Whitening
+
+Whitening decorrelates the feature dimensions so that no single direction dominates the feature-matching loss. This is computed via SVD on the generated embeddings, with the same transform applied to the ground-truth embeddings.
+
+```yaml
+ebft:
+  use_whitening: true
+```
+
+When whitening is enabled, the reward computation applies a whitening matrix `W = U @ diag(1/S) @ U^T` derived from the SVD of generated embeddings. This ensures all feature dimensions contribute equally to the alignment reward.
+
+::: {.callout-note}
+Singular values scale with `sqrt(batch_size)`, so reward magnitudes are batch-size dependent. This is acceptable because the number of samples per prompt (`n_samples_per_prompt` or `num_generations`) is fixed during training.
+:::
+
+### Alignment and Diversity Coefficients
+
+The two reward components are weighted by coefficients:
+
+```yaml
+ebft:
+  alignment_coef: 1.0   # Weight for cosine similarity with ground truth
+  diversity_coef: 1.0   # Weight for pairwise similarity penalty
+```
+
+Both values are scaled by 2 internally (per paper equation 7). The final reward per sample is:
+
+```
+reward_j = 2 * alignment_coef * cos(gen_j, gt)
+         - 2 * diversity_coef * (1/(n-1)) * sum_{j' != j} dot(gen_j, gen_j')
+```
+
+Setting `diversity_coef: 0.0` disables the diversity penalty entirely, which may be appropriate when `num_generations` is small (e.g., 2).
+
+## Strided Mode
+
+Strided mode is designed for training on unstructured text data where there is no natural prompt/completion boundary. Instead of generating full completions with vLLM, it places **anchor points** at regular intervals throughout each document and generates short rollouts at each anchor using block-parallel attention.
+
+### How Block-Parallel Generation Works
+
+Given a document of length `S` tokens:
+
+1. **Anchor placement**: Starting at position `anchor_offset`, place anchors every `stride` tokens. Each anchor defines a block.
+2. **Context window**: Each block sees `context_length` tokens of preceding context from the original document.
+3. **Generation**: At each anchor, generate `generate_max_len` tokens autoregressively, conditioned only on the context window.
+4. **Parallelism**: All blocks are processed in a single forward pass using a specialized attention mask that prevents information leakage between blocks.
+
+```
+Document:   [tok0, tok1, ..., tok_S]
+                    |         |         |
+                 anchor_0   anchor_1  anchor_2
+                    |         |         |
+             [ctx][gen]  [ctx][gen]  [ctx][gen]
+```
+
+The attention mask ensures:
+
+- Prompt tokens use standard causal attention
+- Each generated block attends to its own context window and its own preceding generated tokens
+- Blocks do not attend to each other's generated tokens
+
+When `flex_attention` is available (PyTorch >= 2.5), the mask is compiled into efficient fused kernels. Otherwise, a dense 4D attention mask is used as a fallback.
+
+### Strided Mode Configuration
+
+```yaml
+base_model: meta-llama/Llama-3.2-1B
+rl: ebft
+
+ebft:
+  mode: strided
+  stride: 8                    # Tokens between anchor points
+  context_length: 8            # Context window per block
+  generate_max_len: 8          # Tokens to generate per block
+  n_samples_per_prompt: 4      # Independent rollouts per document
+  temperature: 0.6
+  feature_layers: [0.25, 0.5, 0.75]
+  embed_method: last_token
+  use_whitening: true
+  alignment_coef: 1.0
+  diversity_coef: 1.0
+  rl_coef: 1.0                # RL policy gradient loss weight
+  ce_coef: 0.03               # Cross-entropy loss on GT tokens
+  advantage_estimator: rloo    # rloo, group_norm, or reinforce
+  min_completion_prefix: 8     # Skip anchors in prompt region
+
+datasets:
+  - path: nvidia/OpenCodeInstruct
+    type: ebft_strided_structured.transform
+    split: train[:1%]
+
+sequence_len: 2048
+micro_batch_size: 1
+gradient_accumulation_steps: 2
+
+adapter: lora
+lora_r: 16
+lora_alpha: 32
+lora_target_linear: true
+
+bf16: auto
+flex_attention: true
+gradient_checkpointing: true
+gradient_checkpointing_kwargs:
+  use_reentrant: true          # Required with flex_attention
+```
+
+Run with a single command (no vLLM needed):
+
+```bash
+CUDA_VISIBLE_DEVICES=0 axolotl train config.yaml
+```
+
+### Advantage Estimators
+
+Strided mode supports three advantage estimation methods:
+
+| Estimator | Formula | Requirements |
+|-----------|---------|-------------|
+| `rloo` | Leave-one-out baseline: `reward_j - mean(rewards_{-j})` | `n_samples_per_prompt >= 2` |
+| `group_norm` | Group normalization: `(reward_j - mean) / std` | `n_samples_per_prompt >= 2` |
+| `reinforce` | Raw reward as advantage (no baseline) | Works with `n_samples_per_prompt = 1` |
+
+::: {.callout-warning}
+When `n_samples_per_prompt: 1`, the trainer automatically falls back to `reinforce` and disables the diversity penalty (which requires multiple samples).
+:::
+
+### Strided Mode Constraints
+
+- **`flex_attention: true`** is strongly recommended. Without it, dense 4D masks consume significantly more memory.
+- **`torch_compile: true`** must NOT be set. `flex_attention` compiles its own kernels internally; adding `torch_compile` causes conflicts and OOM.
+- **Gradient checkpointing** must use `use_reentrant: true`. Non-reentrant checkpointing causes `CheckpointError` with `flex_attention` block masks.
+- **`activation_offloading`** is incompatible with `flex_attention`.
+
+### Cross-Entropy Loss
+
+Strided mode supports an optional cross-entropy loss term on ground-truth tokens. This acts as a regularizer to prevent the model from drifting too far from the original distribution:
+
+```yaml
+ebft:
+  ce_coef: 0.03    # Small CE coefficient
+  rl_coef: 1.0     # RL loss coefficient
+```
+
+The total loss is `rl_coef * rl_loss + ce_coef * ce_loss`. For structured mode, `ce_coef` is typically `0.0` since vLLM generation provides sufficient learning signal.
+
+## Dataset Formats
+
+EBFT provides several built-in dataset transforms in `src/axolotl/prompt_strategies/ebft/`.
+
+### Built-In Transforms
+
+| Transform | Input Format | Output Fields | Use Case |
+|-----------|-------------|---------------|----------|
+| `ebft_opencode.transform` | `{input, output}` | `{prompt, ground_truth}` | OpenCodeInstruct, structured QA |
+| `ebft_strided_structured.transform` | `{input, output}` | `{input_ids, labels, prompt_length}` | Strided mode with structured data |
+| `ebft_strided_chat.transform` | `{messages: [...]}` | `{input_ids, labels, prompt_length}` | Strided mode with chat data |
+| `ebft_chat_multiturn.transform` | `{messages: [...]}` | `{prompt, ground_truth, remaining_turns}` | Multi-turn: first-turn target |
+| `ebft_chat_multiturn.transform_last_turn` | `{messages: [...]}` | `{prompt, ground_truth}` | Multi-turn: last-turn target |
+| `ebft_chat_multiturn.transform_all_turns` | `{messages: [...]}` | `{prompt[], ground_truth[]}` | Multi-turn: one example per turn |
+| `ebft_reasoning.transform` | `{messages: [...]}` (with `<think>`) | `{prompt, ground_truth}` | Reasoning/thinking datasets |
+
+### Structured Mode Datasets
+
+For structured (sync/async) mode, the transform must produce `prompt` and `ground_truth` fields:
+
+```yaml
+datasets:
+  - path: nvidia/OpenCodeInstruct
+    type: ebft_opencode.transform
+    split: train[:500]
+```
+
+### Multi-Turn Datasets
+
+Multi-turn transforms extract conversation data for sequential rollout. The `transform` variant targets the first assistant turn, while `transform_last_turn` targets the final turn:
+
+```yaml
+datasets:
+  - path: your/multiturn-dataset
+    type: ebft_chat_multiturn.transform
+```
+
+When `remaining_turns` is present in the dataset output, the trainer performs sequential rollouts: it generates the first assistant turn with vLLM, then continues generating subsequent turns by building up the conversation history.
+
+### Strided Mode Datasets
+
+Strided transforms tokenize the full document and produce `input_ids`, `labels`, and `prompt_length`:
+
+```yaml
+datasets:
+  - path: nvidia/OpenCodeInstruct
+    type: ebft_strided_structured.transform
+    split: train[:1%]
+```
+
+### Custom Transforms
+
+To use your own dataset format, write a transform function:
+
+```python
+def transform(cfg, **kwargs):
+    def transform_fn(example, tokenizer=None):
+        return {
+            "prompt": [{"role": "user", "content": example["question"]}],
+            "ground_truth": example["answer"],
+        }
+    return transform_fn, {"remove_columns": "__all__"}
+```
+
+The `"__all__"` sentinel removes all original dataset columns after the mapping step. Reference this transform in your config:
+
+```yaml
+datasets:
+  - path: your/dataset
+    type: your_module.transform
+```
+
+## Configuration Reference
+
+### Common Parameters (All Modes)
+
+These parameters are set under the `ebft:` key in the YAML config.
+
+| Parameter | Type | Default | Description |
+|-----------|------|---------|-------------|
+| `mode` | `"structured"` or `"strided"` | `"structured"` | EBFT operating mode |
+| `feature_layers` | `list[float]` | `[0.25, 0.5, 0.75]` | Fractional layer depths for feature extraction |
+| `embed_method` | `string` | `"last_token"` | Pooling method: `last_token`, `mean_pooling`, `completion_mean`, or `concat` |
+| `use_whitening` | `bool` | `false` | Apply SVD whitening to feature embeddings before reward computation |
+| `alignment_coef` | `float` | `1.0` | Weight for alignment reward (cosine similarity with ground truth) |
+| `diversity_coef` | `float` | `1.0` | Weight for diversity penalty (pairwise dot product between samples) |
+| `ce_coef` | `float` | `0.0` | Cross-entropy loss coefficient on ground-truth tokens |
+| `adaptive_max_tokens` | `bool` | `true` | Dynamically set vLLM `max_tokens` based on ground-truth length (structured mode) |
+| `gt_length_multiplier` | `float` | `1.5` | Multiplier for ground-truth token count when computing adaptive max tokens (min 0.1) |
+
+### Strided Mode Parameters
+
+These additional parameters apply only when `mode: strided`.
+
+| Parameter | Type | Default | Description |
+|-----------|------|---------|-------------|
+| `stride` | `int` | `8` | Number of tokens between anchor points (must be >= 1) |
+| `context_length` | `int` | `8` | Context window size for each generated block (must be >= 1) |
+| `generate_max_len` | `int` | `8` | Number of tokens to generate per block (must be >= 1) |
+| `n_samples_per_prompt` | `int` | `4` | Number of independent rollouts per document (must be >= 1) |
+| `temperature` | `float` | `0.6` | Sampling temperature for strided generation |
+| `top_p` | `float` | `1.0` | Top-p nucleus sampling threshold |
+| `rl_coef` | `float` | `1.0` | RL policy gradient loss coefficient |
+| `advantage_estimator` | `string` | `"rloo"` | Advantage estimation method: `rloo`, `group_norm`, or `reinforce` |
+| `min_completion_prefix` | `int` | `0` | Minimum tokens into the completion span before placing anchors |
+
+### Structured Mode TRL Parameters
+
+These are set under the `trl:` key and control the GRPO training loop.
+
+| Parameter | Type | Default | Description |
+|-----------|------|---------|-------------|
+| `num_generations` | `int` | -- | Number of completions generated per prompt |
+| `max_completion_length` | `int` | -- | Maximum tokens per generated completion |
+| `temperature` | `float` | `0.7` | Sampling temperature for vLLM generation |
+| `use_vllm` | `bool` | -- | Enable vLLM generation backend |
+| `vllm_lora_sync` | `bool` | `false` | Sync LoRA adapters via filesystem (recommended) |
+| `vllm_sync_interval` | `int` | `1` | Steps between weight syncs to vLLM |
+| `use_data_producer` | `bool` | -- | Required for sync mode with LoRA sync |
+| `async_prefetch` | `bool` | `false` | Enable async generation (overlaps with training) |
+| `streaming_partial_batch` | `bool` | `false` | Score groups incrementally (async mode) |
+| `skip_zero_advantage_batches` | `bool` | `false` | Skip micro-batches where all advantages are zero |
+| `scale_rewards` | `bool` | -- | Normalize rewards within each prompt group |
+| `loss_type` | `string` | `"grpo"` | Loss type for policy optimization |
+| `epsilon` | `float` | `0.2` | Clipping parameter for importance sampling |
+
+### Stop Tokens
+
+vLLM needs explicit stop token IDs for generation. Common configurations:
+
+```yaml
+trl:
+  generation_kwargs:
+    stop_token_ids: [151645, 151643]   # Qwen: <|im_end|>, <|endoftext|>
+```
+
+### Multi-Turn Chat Settings
+
+For multi-turn conversations with Qwen3.5, disable thinking mode to prevent `<think>` tags in completions:
+
+```yaml
+trl:
+  chat_template_kwargs:
+    enable_thinking: false
+```
+
+## Monitoring
+
+### Key Metrics
+
+EBFT logs several custom metrics to wandb and the training console. Here is what to watch for:
+
+| Metric | Healthy Range | Interpretation |
+|--------|--------------|----------------|
+| `ebft/alignment` | 0.3 -- 0.9, trending upward | Cosine similarity between generated and ground-truth features. Higher means the model is learning to produce representations that match the reference. |
+| `ebft/diversity` | 0.01 -- 0.1 | Mean pairwise similarity between different generations for the same prompt. Values above 1.0 indicate mode collapse. |
+| `ebft/cfm_loss` | Below 10, trending downward | Cross-Feature Matching loss. This is the core quantity being minimized. Consistently above 100 indicates instability. |
+| `ebft/reward` | Trending upward (may start negative) | Combined reward signal. If stuck at -1.0, the diversity penalty is dominating alignment. |
+| `grad_norm` | 0.1 -- 3.0 | Gradient magnitude. Values of 0.0 indicate zero-advantage skip (normal). Values above 10 suggest instability. |
+| `entropy` | 0.05 -- 0.5 | Policy entropy. Values below 0.01 suggest mode collapse. |
+| `IS ratio min` | Above 0.1 | Importance sampling ratio minimum. Near-zero values mean the policy is too far off-policy; increase `vllm_sync_interval`. |
+
+### Console Log Example
+
+During training, you will see periodic EBFT reward logs:
+
+```
+ebft reward | align +0.412 ^ | divers +0.023 v | cfm 4.231 v | reward +0.389 ^
+```
+
+The arrows indicate the desired direction: alignment and reward should trend upward, while diversity and CFM loss should trend downward.
+
+### Troubleshooting
+
+| Symptom | Likely Cause | Fix |
+|---------|-------------|-----|
+| `alignment` stays below 0.1 | Feature layers not capturing useful information | Try different `feature_layers` or `embed_method` |
+| `diversity` exceeds 1.0 | Mode collapse -- generations are too similar | Increase `diversity_coef` or `temperature` |
+| `reward` stuck at -1.0 | Diversity penalty dominates alignment | Reduce `diversity_coef` or increase `alignment_coef` |
+| `grad_norm` consistently 0.0 | All micro-batches have zero advantage | Increase `num_generations` or check data quality |
+| `CheckpointError` in strided mode | Incompatible gradient checkpointing settings | Set `use_reentrant: true` in `gradient_checkpointing_kwargs` |
+| OOM during training | Logits tensor too large | Reduce `sequence_len` or `micro_batch_size`; strided mode uses chunked lm_head to mitigate this |
+| vLLM 500 errors | `truncate_prompt_tokens` not supported | Ensure you are using `axolotl vllm-serve` (not `trl vllm-serve`) |
+
+### Feature Network Memory
+
+In PEFT (LoRA) mode, the feature network shares base weights with the actor model by using the `disable_adapter()` context manager. This saves an entire model copy in VRAM (approximately 1--16 GB depending on model size). For non-PEFT training, a separate frozen deepcopy is created.
+
+::: {.callout-note}
+The `disable_adapter()` approach relies on an invariant: `merge_adapter()` is never called on the base weights. All weight sync paths (LoRA sync, HTTP, NCCL) compute merged weights as new tensors or save the adapter to the filesystem, leaving base weights unmodified.
+:::
+
+## Examples
+
+Complete example configurations are available in `examples/ebft/`:
+
+| Config | Model | Mode | Description |
+|--------|-------|------|-------------|
+| `llama-1b-ebft-strided-structured.yaml` | Llama 3.2 1B | Strided | Single-GPU strided training on code data |
+| `qwen3-4b-ebft-structured.yaml` | Qwen3 4B | Structured (sync) | Two-GPU structured training |
+| `qwen3-4b-ebft-structured-async.yaml` | Qwen3 4B | Structured (async) | Two-GPU async training with prefetch |
+| `qwen3-8b-ebft-structured.yaml` | Qwen3 8B | Structured (sync) | Two-GPU structured training for larger model |
+| `qwen35-4b-ebft-structured.yaml` | Qwen3.5 4B | Structured (sync) | Two-GPU with Qwen3.5 |
+| `qwen35-4b-ebft-structured-async.yaml` | Qwen3.5 4B | Structured (async) | Two-GPU async with Qwen3.5 |
+| `qwen35-9b-ebft-structured.yaml` | Qwen3.5 9B | Structured (sync) | Two-GPU structured for 9B model |

docs/expert_quantization.qmd

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

docs/faq.qmd

@@ -63,6 +63,14 @@ 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' / ____`**
@@ -140,3 +148,7 @@ 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: "FDSP + QLoRA"
+title: "FSDP + QLoRA"
 description: Use FSDP with QLoRA to fine-tune large LLMs on consumer GPUs.
 format:
   html:
@@ -23,6 +23,12 @@ 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/getting-started.qmd

@@ -170,17 +170,26 @@ More details can be found in [Merging LoRA weights](inference.qmd#sec-merging).
 
 ## Next Steps {#sec-next-steps}
 
-Now that you have the basics, you might want to:
+Now that you have the basics, explore these guides based on what you want to do:
 
-- Try different model architectures
-- Experiment with hyperparameters
-- Use more advanced training methods
-- Scale up to larger models
+**Choose your path:**
 
-Check our other guides for details on these topics:
+- [Choosing a Fine-Tuning Method](choosing_method.qmd) — SFT vs LoRA vs QLoRA vs GRPO vs DPO, with hardware recommendations
 
-- [Configuration Guide](config-reference.qmd) - Full configuration options
-- [Dataset Loading](dataset_loading.qmd) - Loading datasets from various sources
-- [Dataset Formats](dataset-formats) - Working with different data formats
-- [Multi-GPU Training](multi-gpu.qmd)
-- [Multi-Node Training](multi-node.qmd)
+**Core guides:**
+
+- [Dataset Loading](dataset_loading.qmd) — Loading datasets from various sources
+- [Dataset Formats](dataset-formats) — Working with different data formats
+- [Optimizations](optimizations.qmd) — Flash attention, gradient checkpointing, sample packing
+- [Training Stability & Debugging](training_stability.qmd) — Monitoring metrics, fixing NaN, OOM debugging
+
+**Advanced training methods:**
+
+- [RLHF / Preference Learning](rlhf.qmd) — DPO, KTO, GRPO, EBFT
+- [GRPO Training](grpo.qmd) — RL with custom rewards and vLLM generation
+- [vLLM Serving](vllm_serving.qmd) — Setting up vLLM for GRPO
+
+**Scaling up:**
+
+- [Multi-GPU Training](multi-gpu.qmd) — DeepSpeed, FSDP, DDP
+- [Multi-Node Training](multi-node.qmd) — Distributed training across machines

docs/gradient_checkpointing.qmd

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

docs/grpo.qmd

@@ -0,0 +1,611 @@
+---
+title: "GRPO Training"
+description: "Group Relative Policy Optimization — a reinforcement learning method for training language models with verifiable reward functions."
+order: 8
+---
+
+## Overview
+
+Group Relative Policy Optimization (GRPO) is a reinforcement learning method that improves language models by generating multiple completions per prompt, scoring them with reward functions, and using the relative ranking within each group to compute advantage estimates. Unlike DPO, which requires pre-collected preference pairs, GRPO generates its own training data online and can work with any programmatic reward signal (math correctness, format compliance, code execution results, etc.).
+
+Use GRPO when you have a task with a verifiable reward signal and want the model to discover solution strategies on its own. Use DPO when you already have human preference data. Use SFT when you have gold-standard completions to imitate directly.
+
+Axolotl's GRPO implementation builds on TRL and adds async generation, streaming scoring, importance sampling correction, replay buffers, and multi-GPU scaling via FSDP and DeepSpeed.
+
+
+## Architecture
+
+GRPO training uses a two-process architecture: a vLLM server for fast generation and a trainer process for scoring and gradient updates.
+
+```
+Terminal 1 (GPU 0)                    Terminal 2 (GPU 1)
+┌──────────────────────┐              ┌──────────────────────────────────┐
+│  vLLM Server         │              │  Trainer                         │
+│                      │   HTTP       │                                  │
+│  Serves base model   │◄────────────►│  Background thread:              │
+│  + LoRA adapter      │  /generate   │    Send prompts to vLLM          │
+│                      │  /set_lora   │    Pad & collate completions     │
+│  Punica kernels for  │              │                                  │
+│  LoRA inference      │              │  Main thread:                    │
+│                      │              │    Score completions (rewards)   │
+└──────────────────────┘              │    Compute policy log-probs      │
+                                      │    Calculate advantages          │
+                                      │    PPO-clip gradient update      │
+                                      │    Sync LoRA weights to vLLM     │
+                                      └──────────────────────────────────┘
+```
+
+**Data flow for each training step:**
+
+1. The background thread sends prompts to vLLM, which generates `num_generations` completions per prompt.
+2. The main thread scores completions using your reward functions.
+3. Advantages are computed within each prompt group (group-relative normalization).
+4. Policy log-probabilities are computed by running a forward pass on the training model.
+5. The PPO-clip loss is computed and gradients are applied.
+6. Periodically, LoRA adapter weights are synced back to vLLM so future generations reflect the updated policy.
+
+With async prefetch enabled, step 1 for the *next* batch runs concurrently with steps 2-6 for the *current* batch.
+
+
+## Quick Start
+
+A GRPO training run requires three components: a YAML config, a reward module (Python file), and a running vLLM server.
+
+### 1. Write a reward module
+
+Create a file called `rewards.py` in your working directory:
+
+```python
+# rewards.py
+import re
+
+
+def accuracy_reward(completions, answer, **kwargs) -> list[float]:
+    """Check if the completion contains the correct numerical answer."""
+    rewards = []
+    for completion, correct in zip(completions, answer):
+        text = completion[0]["content"]
+        # Extract the last number from the completion
+        numbers = re.findall(r"-?\d+(?:\.\d+)?", text)
+        predicted = numbers[-1] if numbers else ""
+        rewards.append(1.0 if predicted == str(correct) else 0.0)
+    return rewards
+
+
+def format_reward(completions, **kwargs) -> list[float]:
+    """Reward completions that use a structured thinking format."""
+    rewards = []
+    for completion in completions:
+        text = completion[0]["content"]
+        has_think = "<think>" in text and "</think>" in text
+        has_answer = "<answer>" in text and "</answer>" in text
+        rewards.append(1.0 if has_think and has_answer else 0.0)
+    return rewards
+
+
+def prompt_transform(cfg, *args, **kwargs):
+    """Convert GSM8K dataset rows into chat prompts."""
+    def transform_fn(example, tokenizer=None):
+        label = example["answer"].split("####")[-1].strip().replace(",", "")
+        return {
+            "prompt": [
+                {"role": "system", "content": "Solve the math problem. Show your reasoning in <think> tags and your final numerical answer in <answer> tags."},
+                {"role": "user", "content": example["question"]},
+            ],
+            "answer": label,
+        }
+    return transform_fn, {"remove_columns": ["question"]}
+```
+
+### 2. Write the config
+
+Create `config.yaml`:
+
+```yaml
+base_model: Qwen/Qwen2.5-1.5B-Instruct
+
+rl: grpo
+chat_template: tokenizer_default
+
+vllm:
+  host: 0.0.0.0
+  port: 8000
+  gpu_memory_utilization: 0.85
+  dtype: auto
+  max_model_len: 2048
+
+adapter: lora
+lora_r: 32
+lora_alpha: 64
+lora_target_linear: true
+
+trl:
+  use_vllm: true
+  use_data_producer: true
+  vllm_server_host: 0.0.0.0
+  vllm_server_port: 8000
+  vllm_server_timeout: 300
+  vllm_lora_sync: true
+  num_generations: 8
+  max_completion_length: 512
+  temperature: 0.7
+  reward_funcs:
+    - rewards.accuracy_reward
+    - rewards.format_reward
+  reward_weights:
+    - 1.0
+    - 0.5
+
+datasets:
+  - path: openai/gsm8k
+    name: main
+    type: rewards.prompt_transform
+    split: train
+
+skip_prepare_dataset: true
+val_set_size: 0.0
+sequence_len: 512
+micro_batch_size: 2
+gradient_accumulation_steps: 4
+max_steps: 200
+learning_rate: 5.0e-6
+optimizer: adamw_torch_fused
+lr_scheduler: cosine
+warmup_steps: 10
+
+bf16: true
+flash_attention: true
+gradient_checkpointing: true
+
+special_tokens:
+  pad_token: "<|endoftext|>"
+
+output_dir: ./grpo-output
+logging_steps: 1
+```
+
+### 3. Start vLLM and train
+
+```bash
+# Terminal 1: Start vLLM server on GPU 0
+CUDA_VISIBLE_DEVICES=0 axolotl vllm-serve config.yaml
+
+# Wait 30-90 seconds for model loading and CUDA graph capture
+
+# Terminal 2: Train on GPU 1
+CUDA_VISIBLE_DEVICES=1 axolotl train config.yaml
+```
+
+:::{.callout-tip}
+Use `tmux` or separate terminal sessions to manage the two processes. The vLLM server must remain running for the entire training duration.
+:::
+
+
+## Custom Reward Functions
+
+### Function signature
+
+TRL calls reward functions with this signature:
+
+```python
+def my_reward(completions, **kwargs) -> list[float]:
+```
+
+- `completions` is a list of single-element lists, where each element is a dict `{"role": "assistant", "content": "..."}`. So `completions[i][0]["content"]` gives you the text of the i-th completion.
+- `**kwargs` contains all dataset columns that were *not* removed by the dataset transform. This is how you pass ground truth answers, metadata, or any other information to your reward function.
+- Return a `list[float]` with the same length as `completions`. You may return `None` for individual elements to exclude them from aggregation.
+
+### Example: accuracy reward with answer extraction
+
+```python
+def accuracy_reward(completions, answer, **kwargs) -> list[float]:
+    rewards = []
+    for completion, correct_answer in zip(completions, answer):
+        text = completion[0]["content"]
+        # Extract answer from <answer>...</answer> tags
+        match = re.search(r"<answer>(.*?)</answer>", text, re.DOTALL)
+        predicted = match.group(1).strip() if match else ""
+        rewards.append(1.0 if predicted == str(correct_answer) else 0.0)
+    return rewards
+```
+
+### Example: length penalty
+
+```python
+def length_penalty(completions, **kwargs) -> list[float]:
+    """Penalize very short or very long completions."""
+    rewards = []
+    for completion in completions:
+        length = len(completion[0]["content"])
+        if length < 50:
+            rewards.append(-0.5)
+        elif length > 2000:
+            rewards.append(-0.2)
+        else:
+            rewards.append(0.0)
+    return rewards
+```
+
+### Multiple rewards and weighting
+
+You can combine multiple reward functions with different weights:
+
+```yaml
+trl:
+  reward_funcs:
+    - rewards.accuracy_reward
+    - rewards.format_reward
+    - rewards.length_penalty
+  reward_weights:
+    - 1.0    # accuracy is most important
+    - 0.5    # format compliance
+    - 0.1    # mild length preference
+```
+
+Rewards are combined by the `multi_objective_aggregation` strategy:
+
+- `sum_then_normalize` (default): weights and sums all rewards first, then normalizes across the group.
+- `normalize_then_sum` (GDPO): normalizes each reward independently, then sums. This prevents one reward from dominating and is recommended when using multiple reward functions with different scales.
+
+```yaml
+trl:
+  multi_objective_aggregation: normalize_then_sum
+```
+
+### Dataset transforms
+
+The dataset transform converts raw HuggingFace dataset rows into chat-format prompts:
+
+```python
+def prompt_transform(cfg, *args, **kwargs):
+    def map_fn(example, tokenizer=None):
+        return {
+            "prompt": [
+                {"role": "system", "content": "You are a helpful assistant."},
+                {"role": "user", "content": example["question"]},
+            ],
+            # Keep 'answer' column for the reward function
+            "answer": example["answer"],
+        }
+    # Remove columns consumed by the transform; keep columns needed by rewards
+    return map_fn, {"remove_columns": ["question"]}
+```
+
+The transform returns a tuple of `(map_function, kwargs_dict)`. The `remove_columns` in the kwargs dict removes columns that are no longer needed. Columns that your reward functions reference via `**kwargs` (like `answer`) must *not* be removed.
+
+:::{.callout-warning}
+The reward module must be importable from the directory where you run `axolotl train`. If your reward file is `rewards.py`, the import path is `rewards.accuracy_reward`. If it is inside a package `my_rewards/scoring.py`, use `my_rewards.scoring.accuracy_reward`.
+:::
+
+### Reward models (neural network rewards)
+
+Instead of a Python function, you can pass a HuggingFace model path as a reward function. TRL will load it as a reward model and use its scalar output as the reward:
+
+```yaml
+trl:
+  reward_funcs:
+    - OpenAssistant/reward-model-deberta-v3-large-v2
+    - rewards.format_reward
+  reward_weights:
+    - 1.0
+    - 0.3
+```
+
+### Using math_verify
+
+The `math_verify` library provides robust mathematical answer verification but uses `signal.alarm()` internally, which only works in the main thread. If you use `math_verify` in a reward function, set `reward_num_workers` to use subprocess workers:
+
+```yaml
+trl:
+  reward_num_workers: 4
+```
+
+Each worker runs in its own subprocess with its own main thread, so `signal.alarm()` works correctly.
+
+
+## vLLM Setup
+
+GRPO requires a running vLLM server for generation. For a complete guide on server modes, LoRA sync, weight synchronization, and restart procedures, see [vLLM Serving](vllm_serving.qmd).
+
+The minimal setup:
+
+```yaml
+vllm:
+  host: 0.0.0.0
+  port: 8000
+  gpu_memory_utilization: 0.85
+
+trl:
+  use_vllm: true
+  vllm_lora_sync: true         # Recommended with LoRA — faster sync, no NCCL contention
+  vllm_sync_interval: 5        # Sync weights every 5 steps
+```
+
+```bash
+CUDA_VISIBLE_DEVICES=0 axolotl vllm-serve config.yaml   # GPU 0: vLLM
+CUDA_VISIBLE_DEVICES=1 axolotl train config.yaml         # GPU 1: training
+```
+
+:::{.callout-warning}
+vLLM must be restarted between experiments — stale weight syncs corrupt server state. See [Restart Requirements](vllm_serving.qmd#sec-restart).
+:::
+
+
+## Async Training Features
+
+Async GRPO overlaps generation and training to reduce wall-clock time. While the model trains on the current batch, the next batch is already being generated by vLLM.
+
+### Enabling async prefetch
+
+```yaml
+trl:
+  use_data_producer: true
+  async_prefetch: true
+  prefetch_depth: 1
+  vllm_sync_interval: 2
+```
+
+- `use_data_producer: true` enables the data producer protocol (required for all async features).
+- `async_prefetch: true` runs generation in a background thread.
+- `prefetch_depth` controls how many batches to prefetch ahead (1 is usually sufficient).
+- `vllm_sync_interval` controls how often LoRA weights are synced to vLLM (every N optimizer steps). Lower values mean fresher generations but more sync overhead.
+
+:::{.callout-tip}
+Because the background thread generates with slightly stale model weights, async mode benefits from importance sampling correction (see next section). Enable `vllm_importance_sampling_correction: true` when using `async_prefetch: true`.
+:::
+
+### Streaming partial batch
+
+Instead of scoring the entire batch at once, streaming mode scores one prompt group at a time. This reduces peak memory during scoring and enables finer-grained zero-advantage skipping.
+
+```yaml
+trl:
+  streaming_partial_batch: true
+  streaming_min_groups: 1
+```
+
+`streaming_min_groups` controls the minimum number of prompt groups scored per chunk. Setting it to 1 gives maximum granularity.
+
+### Zero-advantage batch skipping
+
+When all advantages in a micro-batch are zero (every completion in the group got the same reward), there is no learning signal. This feature skips the forward/backward pass entirely for such micro-batches.
+
+```yaml
+trl:
+  skip_zero_advantage_batches: true   # default
+```
+
+This is enabled by default and logged as `skipped_zero_adv_batches` in training metrics. It is a safety net, not a major optimization -- it only saves significant time when the model cannot solve any prompts in the batch.
+
+### Replay buffer
+
+The replay buffer caches rollout groups that had learning signal (non-zero reward variance) and replaces zero-signal groups in later batches. This improves data utilization when many prompts yield no reward variance.
+
+```yaml
+trl:
+  replay_buffer_size: 100
+  replay_recompute_logps: true
+```
+
+:::{.callout-warning}
+When `replay_recompute_logps: false`, replayed data uses stale log-probabilities which creates an IS mismatch. Keep the default `true` unless you have a specific reason to disable it.
+:::
+
+### Deferred re-rolling
+
+Prompts where the model gets zero reward for all generations are buffered and re-injected into later batches, when the model may have improved enough to produce useful completions.
+
+```yaml
+trl:
+  reroll_start_fraction: 0.5   # Start re-rolling after 50% of training
+  reroll_max_groups: 1          # Max groups to replace per batch
+```
+
+Set `reroll_start_fraction: 1.0` to disable. This is most useful for tasks where the model starts weak but steadily improves.
+
+### Parallel reward workers
+
+Reward functions that use `signal.alarm()` (like `math_verify`) only work in the main thread. Parallel reward workers run each function in its own subprocess:
+
+```yaml
+trl:
+  reward_num_workers: 4
+```
+
+Work is sharded across workers by prompt group. For simple reward functions, a single worker is usually sufficient -- the overhead of IPC can exceed the computation time.
+
+
+## Importance Sampling and Off-Policy Correction
+
+When using async prefetch, completions are generated from a slightly older policy. IS correction adjusts the gradient to account for this mismatch.
+
+```yaml
+trl:
+  vllm_importance_sampling_correction: true
+  importance_sampling_level: token     # 'token' recommended (especially with Liger kernel)
+  off_policy_mask_threshold: 0.5       # KL threshold — masks sequences that are too off-policy
+```
+
+Use `token` level IS. Sequence-level has numerical issues with Liger's chunked computation. The `off_policy_mask_threshold` (OPSM) is a safety net that drops sequences where KL divergence exceeds the threshold — 0.5 is a reasonable starting point.
+
+For detailed coverage of IS modes (`token_mask`, `token_truncate`, etc.), capping, and bias-corrected KL, see [vLLM Serving — IS Correction](vllm_serving.qmd#sec-weight-sync).
+
+
+## Scaling
+
+### FP8 training
+
+FP8 quantization halves model VRAM usage with minimal impact on training quality. It does not significantly speed up computation for small models but allows larger models to fit in memory.
+
+```yaml
+fp8: true
+torch_compile: true
+```
+
+:::{.callout-warning}
+FP8 requires patching for zero-padding edge cases. The `act_quant_kernel` can produce NaN when input is all zeros (padding positions). If you see NaN in grad norms, check whether your padding token embedding is non-zero.
+:::
+
+### FSDP (Fully Sharded Data Parallel)
+
+FSDP distributes model parameters across multiple GPUs for training while vLLM runs on a separate GPU:
+
+```yaml
+fsdp:
+  - full_shard
+  - auto_wrap
+fsdp_config:
+  fsdp_transformer_layer_cls_to_wrap: Qwen2DecoderLayer
+gradient_checkpointing_kwargs:
+  use_reentrant: false
+```
+
+Launch with:
+
+```bash
+# GPU 0: vLLM
+CUDA_VISIBLE_DEVICES=0 axolotl vllm-serve config.yaml
+
+# GPUs 0,1: Training (FSDP will use both visible GPUs)
+CUDA_VISIBLE_DEVICES=0,1 axolotl train config.yaml
+```
+
+:::{.callout-warning}
+`async_prefetch: true` can deadlock with FSDP because background threads perform unsynchronized FSDP collectives across ranks. With multi-GPU FSDP, only rank 0 generates in the background thread and results are broadcast to all ranks. If you still see hangs, set `async_prefetch: false`.
+:::
+
+### DeepSpeed ZeRO-3
+
+```yaml
+deepspeed: deepspeed_configs/zero3_bf16.json
+gradient_checkpointing_kwargs:
+  use_reentrant: true   # Required -- non-reentrant causes CheckpointError with ZeRO-3
+```
+
+:::{.callout-note}
+DeepSpeed ZeRO-3 requires `use_reentrant: true` for gradient checkpointing. This is the opposite of the FSDP recommendation. Non-reentrant checkpointing causes tensor metadata mismatches during recomputation with ZeRO-3's parameter partitioning.
+:::
+
+### Multi-GPU considerations
+
+| Concern | Recommendation |
+|---------|---------------|
+| vLLM GPU allocation | Dedicate one or more GPUs to vLLM; do not share with trainer GPUs |
+| Weight sync contention | Use `vllm_lora_sync: true` to avoid NCCL contention between training and vLLM |
+| FSDP + async | Use `async_prefetch: false` or rely on rank-0-only background generation |
+| DeepSpeed + gradient checkpoint | Must use `use_reentrant: true` |
+| OOM during scoring | Reduce `micro_batch_size` or `num_generations`. The logits tensor scales with `batch_size * vocab_size` |
+
+
+## Monitoring and Debugging
+
+For detailed metric ranges, failure diagnosis, and OOM debugging, see [Training Stability & Debugging](training_stability.qmd).
+
+Quick health checks during GRPO training:
+
+- `rewards/*/mean` should be > 0.15 within 20 steps — if it stays at 0, test your reward function standalone
+- `reward_std` should be > 0 on most steps — all-zero means no learning signal
+- `entropy` in 0.05-0.5 — below 0.01 suggests mode collapse
+- `grad_norm` in 0.001-1.0 — > 10 is unstable, 0.0 is expected when zero-advantage skip fires
+
+:::{.callout-tip}
+Pipe training output to a log file: `axolotl train config.yaml 2>&1 | tee /tmp/training.log`
+:::
+
+
+## Configuration Reference
+
+All GRPO-specific options live under the `trl:` key in your config. Standard training options (`learning_rate`, `micro_batch_size`, etc.) are set at the top level as usual.
+
+### Core GRPO
+
+| Option | Type | Default | Description |
+|--------|------|---------|-------------|
+| `use_vllm` | bool | `false` | Enable vLLM for generation |
+| `vllm_mode` | `"server"` or `"colocate"` | `null` | vLLM deployment mode |
+| `vllm_server_host` | str | `"0.0.0.0"` | vLLM server hostname |
+| `vllm_server_port` | int | `8000` | vLLM server port |
+| `vllm_server_timeout` | int | `null` | Timeout (seconds) for vLLM responses |
+| `num_generations` | int | `null` | Completions generated per prompt |
+| `generation_batch_size` | int | `null` | Number of unique prompts per generation step |
+| `max_completion_length` | int | `null` | Maximum tokens per completion |
+| `beta` | float | `null` | KL penalty coefficient |
+| `num_iterations` | int | `null` | Iterations per batch (mu in the GRPO paper) |
+| `epsilon` | float | `null` | PPO clipping lower bound |
+| `epsilon_high` | float | `null` | PPO clipping upper bound |
+| `loss_type` | str | `null` | Loss formulation: `grpo`, `bnpo`, or `dr_grpo` |
+| `scale_rewards` | bool | `true` | Normalize rewards by standard deviation |
+| `mask_truncated_completions` | bool | `false` | Exclude truncated completions from loss |
+
+### Reward functions
+
+| Option | Type | Default | Description |
+|--------|------|---------|-------------|
+| `reward_funcs` | list[str] | `null` | Import paths to reward functions or HF model IDs |
+| `reward_weights` | list[float] | `null` | Relative weights for each reward function |
+| `multi_objective_aggregation` | str | `null` | `"sum_then_normalize"` (GRPO) or `"normalize_then_sum"` (GDPO) |
+| `rollout_func` | str | `null` | Import path to custom rollout function for OpenEnv-style tasks |
+
+### Generation parameters
+
+| Option | Type | Default | Description |
+|--------|------|---------|-------------|
+| `temperature` | float | `null` | Sampling temperature |
+| `top_p` | float | `null` | Nucleus sampling probability |
+| `top_k` | int | `null` | Top-k sampling |
+| `min_p` | float | `null` | Minimum probability threshold |
+| `repetition_penalty` | float | `null` | Penalty for repeated tokens |
+| `generation_kwargs` | dict | `null` | Additional vLLM SamplingParams (e.g., `stop_token_ids`) |
+| `chat_template_kwargs` | dict | `null` | Chat template kwargs (e.g., `{enable_thinking: false}`) |
+| `vllm_guided_decoding_regex` | str | `null` | Regex constraint for guided decoding |
+
+### Async pipeline
+
+| Option | Type | Default | Description |
+|--------|------|---------|-------------|
+| `use_data_producer` | bool | `false` | Enable data producer protocol (required for async features) |
+| `async_prefetch` | bool | `false` | Generate next batch in background thread |
+| `prefetch_depth` | int | `null` | Number of batches to prefetch ahead |
+| `vllm_sync_interval` | int | `null` | Sync LoRA weights to vLLM every N steps |
+| `vllm_lora_sync` | bool | `false` | Use filesystem LoRA sync instead of NCCL merge |
+| `streaming_partial_batch` | bool | `null` | Score prompt groups incrementally |
+| `streaming_min_groups` | int | `null` | Minimum groups per streaming chunk |
+| `skip_zero_advantage_batches` | bool | `true` | Skip micro-batches with zero learning signal |
+| `reward_num_workers` | int | `1` | Subprocess workers for reward computation |
+| `vllm_enable_sleep_mode` | bool | `null` | Offload vLLM weights when idle (colocate mode) |
+
+### Importance sampling
+
+| Option | Type | Default | Description |
+|--------|------|---------|-------------|
+| `vllm_importance_sampling_correction` | bool | `null` | Enable IS correction for async distribution shift |
+| `importance_sampling_level` | `"token"` or `"sequence"` | `null` | Granularity of IS ratios. Use `token` with Liger |
+| `vllm_importance_sampling_mode` | str | `null` | `token_mask`, `token_truncate`, `sequence_mask`, or `sequence_truncate` |
+| `vllm_importance_sampling_cap` | float | `null` | Cap C for IS ratio clipping/masking |
+| `off_policy_mask_threshold` | float | `null` | KL threshold for off-policy sequence masking (OPSM) |
+| `use_bias_correction_kl` | bool | `null` | Apply IS correction to KL divergence term |
+
+### Replay and re-roll
+
+| Option | Type | Default | Description |
+|--------|------|---------|-------------|
+| `replay_buffer_size` | int | `0` | Max cached high-signal groups. 0 = disabled |
+| `replay_recompute_logps` | bool | `true` | Recompute log-probs for replayed data with current model |
+| `reroll_start_fraction` | float | `1.0` | Start re-rolling failed prompts after this fraction of training. 1.0 = disabled |
+| `reroll_max_groups` | int | `1` | Max prompt groups to replace with re-rolls per batch |
+
+### Reference model
+
+| Option | Type | Default | Description |
+|--------|------|---------|-------------|
+| `sync_ref_model` | bool | `false` | Periodically sync reference model with training model |
+| `ref_model_mixup_alpha` | float | `0.9` | EMA coefficient for reference model sync |
+| `ref_model_sync_steps` | int | `64` | Sync reference model every N steps |
+
+### Logging
+
+| Option | Type | Default | Description |
+|--------|------|---------|-------------|
+| `log_completions` | bool | `false` | Log sample completions to W&B |
+| `num_completions_to_print` | int | `null` | Number of completions to print per step |
+| `use_liger_loss` | bool | `null` | Use Liger fused kernel for GRPO loss (reduces VRAM) |

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 2.7.0 and CUDA 12.8.
+For Blackwell GPUs, please use Pytorch 2.9.1 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-cu128-2.7.0` or the cloud variant `axolotlai/axolotl-cloud:main-py3.11-cu128-2.7.0`.
+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`.
 :::
 
 Please refer to the [Docker documentation](docker.qmd) for more information on the different Docker images that are available.
@@ -165,7 +165,7 @@ We recommend using WSL2 (Windows Subsystem for Linux) or Docker.
    ```
 4. (Optional) Login to Hugging Face:
    ```{.bash}
-   huggingface-cli login
+   hf auth login
    ```
 
 ## Troubleshooting {#sec-troubleshooting}

docs/lora_optims.qmd

@@ -5,10 +5,11 @@ 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
-(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.
+(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.
 
 We currently support several common model architectures, including (but not limited to):
 
@@ -88,6 +89,10 @@ lora_o_kernel: true
 Currently, LoRA kernels are not supported for RLHF training, only SFT.
 :::
 
+::: {.callout-warning}
+LoRA kernels do not support remote modeling code.
+:::
+
 ## Requirements
 
 - One or more NVIDIA or AMD GPUs (in order to use the Triton kernels)
@@ -131,6 +136,5 @@ 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

@@ -27,3 +27,9 @@ 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/moe_backends.md

@@ -1,18 +0,0 @@
-MoE Backends in Axolotl
-
-Axolotl supports selecting a Mixture-of-Experts (MoE) compute backend via the training config (YAML):
-
-- Set `moe_backend: auto|torch_grouped|naive`
-
-Behavior
-- auto (default): prefers PyTorch 2.8+ grouped GEMM; otherwise naive.
-- torch_grouped: targets PyTorch 2.8+ grouped GEMM (H100/SM90+ recommended).
-- naive: keeps the reference per-expert loop.
-
-Notes
-- Current implementation wires the backend selector and routes Mixtral MoE through it. Torch grouped uses cuBLASLt grouped GEMM when available; otherwise, the code falls back to the naive per-expert loop.
-- No changes to training scripts are required; selection happens inside the model forward.
-
-Example
-moe_backend: torch_grouped
-accelerate launch -m axolotl.cli.train path/to/config.yaml

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
@@ -14,12 +14,18 @@ This guide covers advanced training configurations for multi-GPU setups using Ax
 
 ## Overview {#sec-overview}
 
-Axolotl supports several methods for multi-GPU training:
+When training on multiple GPUs, Axolotl supports 3 sharding/parallelism strategies. Additionally, you can layer specific optimization features on top of that strategy.
 
-- DeepSpeed (recommended)
-- FSDP (Fully Sharded Data Parallel)
-- Sequence parallelism
-- FSDP + QLoRA
+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 {#sec-deepspeed}
 
@@ -65,12 +71,18 @@ Start from Stage 1 -> Stage 2 -> Stage 3.
 
 ## Fully Sharded Data Parallel (FSDP) {#sec-fsdp}
 
+FSDP allows you to shard model parameters, gradients, and optimizer states across data parallel workers.
+
 ::: {.callout-note}
 
 FSDP2 is recommended for new users. FSDP1 is deprecated and will be removed in an upcoming release of Axolotl.
 
 :::
 
+### FSDP + QLoRA {#sec-fsdp-qlora}
+
+For combining FSDP with QLoRA, see our [dedicated guide](fsdp_qlora.qmd).
+
 ### Migrating from FSDP1 to FSDP2 {#sec-migrate-fsdp1-fsdp2}
 
 To migrate your config from FSDP1 to FSDP2, you must use the `fsdp_version` top-level config field to specify the FSDP version, and
@@ -88,6 +100,7 @@ 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:
@@ -144,10 +157,6 @@ 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

@@ -8,18 +8,24 @@ format:
 
 ## Supported Models
 
+- [Gemma-4](#sec-gemma-4) *(NEW)*
 - [Mllama](#sec-mllama)
 - [Llama4](#sec-llama4)
 - [Pixtral](#sec-pixtral)
 - [Llava-1.5](#sec-llava-15)
 - [Mistral-Small-3.1](#sec-mistral-small-31)
+- [Mistral-Small-4](#sec-mistral-small-4)
+- [Magistral-Small-2509](#sec-magistral-small-2509)
 - [Voxtral](#sec-voxtral)
 - [Gemma-3](#sec-gemma-3)
 - [Gemma-3n](#sec-gemma-3n)
 - [Qwen2-VL](#sec-qwen2-vl)
 - [Qwen2.5-VL](#sec-qwen25-vl)
+- [Qwen3.5](#sec-qwen3-5)
+- [GLM-4.6V](#sec-glm-4-6v)
 - [SmolVLM2](#sec-smolvlm2)
 - [LFM2-VL](#sec-lfm2-vl)
+- [Intern-VL](#sec-intern-vl)
 
 ## Usage
 
@@ -41,7 +47,6 @@ 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
@@ -56,10 +61,14 @@ image_resize_algorithm: bilinear
 
 Please see [examples](https://github.com/axolotl-ai/axolotl/tree/main/examples) folder for full configs.
 
-::: {.callout-warning}
+::: {.callout-tip}
 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
@@ -94,10 +103,28 @@ 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
+```
 
-chat_template: mistral_v7_tekken
+### Mistral-Small-4 {#sec-mistral-small-4}
+
+```yaml
+base_model: mistralai/Mistral-Small-4-119B-2603
+```
+
+### Magistral-Small-2509 {#sec-magistral-small-2509}
+
+::: {.callout-tip}
+Please make sure to install vision lib via `pip install 'mistral-common[opencv]==1.8.5'`
+:::
+
+```yaml
+base_model: mistralai/Magistral-Small-2509
 ```
 
 ### Voxtral {#sec-voxtral}
@@ -108,8 +135,44 @@ Please make sure to install audio lib via `pip3 install librosa==0.11.0 'mistral
 
 ```yaml
 base_model: mistralai/Voxtral-Mini-3B-2507
+
+processor_type: VoxtralProcessor
 ```
 
+### Gemma-4 {#sec-gemma-4}
+
+All Gemma 4 variants (E2B, E4B, 26B-A4B, 31B) load as multimodal models even for text-only training.
+
+```yaml
+base_model: google/gemma-4-E2B-it  # or E4B-it, 26B-A4B, 31B
+
+chat_template: gemma4
+freeze_mm_modules: true  # freeze vision/audio encoders for text-only or vision LoRA
+
+# For the 26B-A4B MoE model, enable ScatterMoE and expert LoRA:
+plugins:
+  - axolotl.integrations.cut_cross_entropy.CutCrossEntropyPlugin
+  - axolotl.integrations.kernels.KernelsPlugin
+use_kernels: true
+use_scattermoe: true
+experts_implementation: scattermoe
+
+lora_target_modules: 'model.language_model.layers.[\d]+.(_checkpoint_wrapped_module.)?(mlp|self_attn).(up|down|gate|q|k|v|o)_proj'
+
+# MoE expert LoRA (3D tensors, not nn.Linear) — only for 26B-A4B:
+lora_target_parameters:
+  - experts.gate_up_proj
+  - experts.down_proj
+```
+
+::: {.callout-warning}
+Gemma 4 VLM training starts with high loss (~8-15). This is expected — see the [training stability guide](training_stability.qmd) for details.
+:::
+
+::: {.callout-tip}
+For DDP training, axolotl auto-detects Gemma4 and sets `use_reentrant=False` and `ddp_find_unused_parameters=True`. However, when `activation_offloading: true`, `ddp_find_unused_parameters` is skipped (checkpoint wrappers conflict with it); use `freeze_mm_modules: true` instead to handle unused vision/audio params. For FSDP2, use `fsdp_transformer_layer_cls_to_wrap: Gemma4TextDecoderLayer`.
+:::
+
 ### Gemma-3 {#sec-gemma-3}
 
 ::: {.callout-tip}
@@ -156,6 +219,34 @@ base_model: Qwen/Qwen2.5-VL-7B-Instruct
 chat_template: qwen2_vl  # same as qwen2-vl
 ```
 
+### Qwen3-VL {#sec-qwen3-vl}
+
+```yaml
+base_model: Qwen/Qwen3-VL-4B-Instruct
+
+chat_template: qwen2_vl  # same as qwen2-vl
+```
+
+### Qwen3.5 {#sec-qwen3-5}
+
+```yaml
+base_model: Qwen/Qwen3.5-9B
+
+chat_template: qwen3_5
+```
+
+### GLM-4.6V {#sec-glm-4-6v}
+
+Both GLM-4.6V (106B MoE) and GLM-4.6V-Flash (9B) are supported.
+
+```yaml
+# GLM-4.6V (106B MoE version)
+base_model: zai-org/GLM-4.6V
+
+# OR GLM-4.6V-Flash (9B version)
+base_model: zai-org/GLM-4.6V-Flash
+```
+
 ### SmolVLM2 {#sec-smolvlm2}
 
 ::: {.callout-tip}
@@ -176,6 +267,16 @@ Please uninstall `causal-conv1d` via `pip3 uninstall -y causal-conv1d`
 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.

docs/optimizations.qmd

@@ -0,0 +1,156 @@
+---
+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)
+
+### Layer Offloading
+
+Offloads frozen (non-trainable) decoder layer parameters to CPU and streams them back to GPU one layer at a time during forward/backward passes using CUDA stream prefetching. Especially effective for LoRA/QLoRA where most parameters are frozen.
+
+- **Config:** `layer_offloading: true`
+- **Learn more:** [Layer Offloading Docs](gradient_checkpointing.qmd#enabling-layer-offloading)
+
+### Cut Cross Entropy (CCE)
+
+Reduces VRAM usage by using an optimized cross-entropy loss calculation.
+
+- **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)
+
+### Expert Kernels
+
+Optimized kernel implementations for Mixture of Experts (MoE) model training.
+
+- **ScatterMoE**: Triton-based MoE kernels with fused LoRA support.
+- **SonicMoE**: CUTLASS-based MoE kernels for NVIDIA Hopper and Blackwell GPUs.
+
+- **Learn more:** [Custom Integrations - Kernels Integration](custom_integrations.qmd#kernels-integration)
+
+## Long Context Models
+
+Techniques to train models on sequences longer than their original context window.
+
+### 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)
+
+### MoE Expert Quantization
+
+Quantizes MoE expert weights on load to reduce VRAM when training MoE models with adapters. Required for Transformers v5+ MoE models where experts use fused `nn.Parameter` tensors.
+
+- **Config:** `quantize_moe_experts: true`
+- **Learn more:** [MoE Expert Quantization](expert_quantization.qmd)

docs/qat.qmd

@@ -30,6 +30,7 @@ qat:
 ```
 
 We support the following quantization schemas:
+
 - `Int4WeightOnly` (requires the `fbgemm-gpu` extra when installing Axolotl)
 - `Int8DynamicActivationInt4Weight`
 - `Float8DynamicActivationFloat8Weight`

docs/rlhf.qmd

@@ -16,7 +16,12 @@ feedback. Various methods include, but not limited to:
 - [Identity Preference Optimization (IPO)](#ipo)
 - [Kahneman-Tversky Optimization (KTO)](#kto)
 - [Odds Ratio Preference Optimization (ORPO)](#orpo)
-- [Group Relative Policy Optimization (GRPO)](#grpo)
+- [Group Relative Policy Optimization (GRPO)](#grpo) — see also the [GRPO deep dive](grpo.qmd) for async features, custom rewards, and scaling
+- [Group Reward-Decoupled Policy Optimization (GDPO)](#gdpo)
+- [Energy-Based Fine-Tuning (EBFT)](#ebft) — see also the [EBFT guide](ebft.qmd) for detailed mode comparisons and configuration
+- [NeMo Gym Integration](#nemo-gym-integration)
+
+For help choosing between these methods, see [Choosing a Fine-Tuning Method](choosing_method.qmd).
 
 
 ## RLHF using Axolotl
@@ -219,6 +224,21 @@ 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
@@ -497,7 +517,7 @@ The input format is a simple JSON input with customizable fields based on the ab
 ### GRPO
 
 ::: {.callout-tip}
-Check out our [GRPO cookbook](https://github.com/axolotl-ai-cloud/grpo_code).
+Check out our [GRPO cookbook](https://github.com/axolotl-ai-cloud/grpo_code). For a comprehensive guide covering async training, custom rewards, importance sampling, and scaling, see the [GRPO deep dive](grpo.qmd).
 :::
 
 In the latest GRPO implementation, `vLLM` is used to significantly speedup trajectory generation during training. In this example, we're using 4 GPUs - 2 for training, and 2 for vLLM:
@@ -582,6 +602,116 @@ 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.
@@ -595,6 +725,309 @@ trl:
 
 For more information, see [GRPO docs](https://huggingface.co/docs/trl/v0.17.0/en/grpo_trainer#loss-types).
 
+#### Async GRPO
+
+Async GRPO overlaps vLLM generation with training by producing rollouts in a background thread. While the model trains on the current batch, the next batch is already being generated. This can significantly reduce wall-clock time per step.
+
+```yaml
+trl:
+  use_data_producer: true     # Enable data producer protocol
+  use_vllm: true
+  async_prefetch: true         # Generate rollouts in background thread
+  prefetch_depth: 1            # Number of rollouts to prefetch
+  vllm_sync_interval: 2        # Sync weights to vLLM every N steps
+```
+
+::: {.callout-note}
+Because the background thread generates completions with slightly stale model weights, async GRPO uses importance sampling correction to account for the distribution shift. This is controlled by `vllm_importance_sampling_correction: true` (default when async is enabled).
+:::
+
+##### vLLM LoRA Sync
+
+By default, weight sync to vLLM merges the LoRA adapter into the base model and broadcasts all parameters via NCCL. LoRA sync is a faster alternative that saves only the adapter weights to the filesystem and has vLLM load them natively using Punica kernels.
+
+```yaml
+adapter: lora
+lora_r: 32
+lora_alpha: 64
+lora_target_linear: true
+
+trl:
+  vllm_lora_sync: true         # Enable native LoRA sync
+```
+
+When `vllm_lora_sync: true` is set, axolotl automatically selects the LoRA-aware vLLM serve module. Start vLLM as usual:
+
+```bash
+CUDA_VISIBLE_DEVICES=0 axolotl vllm-serve config.yaml
+```
+
+Then start training on a separate GPU:
+
+```bash
+CUDA_VISIBLE_DEVICES=1 axolotl train config.yaml
+```
+
+::: {.callout-tip}
+LoRA sync is especially beneficial with multi-GPU training (FSDP/DeepSpeed), where NCCL merge-sync can cause GPU contention with vLLM generation.
+:::
+
+##### Streaming Partial Batch
+
+Instead of scoring the entire batch at once, streaming mode scores one prompt group at a time. This enables finer-grained zero-advantage skipping and reduces peak memory usage during scoring.
+
+```yaml
+trl:
+  streaming_partial_batch: true
+```
+
+##### Importance Sampling Correction
+
+When using async prefetch, completions are generated from a slightly older version of the model. Importance sampling (IS) correction adjusts the policy gradient to account for this distribution shift.
+
+```yaml
+trl:
+  vllm_importance_sampling_correction: true   # Enable IS correction
+  importance_sampling_level: token             # 'token' or 'sequence'
+  off_policy_mask_threshold: 0.5              # Mask sequences with IS ratio below this
+```
+
+- `importance_sampling_level: token` applies per-token IS ratios (recommended with Liger kernel)
+- `importance_sampling_level: sequence` applies per-sequence IS ratios
+- `off_policy_mask_threshold` masks out sequences where the IS ratio indicates they are too far off-policy
+
+##### Replay Buffer
+
+The replay buffer caches rollout groups that had learning signal (non-zero reward variance) and uses them to replace zero-signal groups in later batches.
+
+```yaml
+trl:
+  replay_buffer_size: 100       # Max cached groups (0 = disabled)
+  replay_recompute_logps: true  # Recompute log-probs for replayed data (recommended)
+```
+
+::: {.callout-note}
+When `replay_recompute_logps: true` (default), old log-probabilities are recomputed using the current model weights. This fixes the IS mismatch that would otherwise occur when replaying stale data.
+:::
+
+##### Deferred Re-rolling
+
+Failed prompts (where the model produces zero reward for all generations) are buffered and re-injected into later batches when the model may be better equipped to solve them.
+
+```yaml
+trl:
+  reroll_start_fraction: 0.5    # Start re-rolling after 50% of training
+  reroll_max_groups: 1          # Max groups to replace per batch
+```
+
+##### Zero-Advantage Batch Skipping
+
+When all advantages in a micro-batch are zero (no learning signal), the forward/backward pass is skipped entirely. This is enabled by default and logged as `skipped_zero_adv_batches=1`.
+
+```yaml
+trl:
+  skip_zero_advantage_batches: true   # default
+```
+
+##### Parallel Reward Workers
+
+Reward functions that use `signal.alarm()` (e.g., `math_verify`) must run in the main thread. Parallel reward workers use subprocesses to work around this limitation while enabling concurrent reward computation.
+
+```yaml
+trl:
+  reward_num_workers: 4         # Number of subprocess workers (1 = no parallelism)
+```
+
+##### Full Async GRPO Example
+
+```yaml
+base_model: Qwen/Qwen2.5-1.5B-Instruct
+
+vllm:
+    host: 0.0.0.0
+    port: 8000
+    gpu_memory_utilization: 0.35
+    dtype: auto
+
+adapter: lora
+lora_r: 32
+lora_alpha: 64
+lora_target_linear: true
+
+rl: grpo
+trl:
+  use_data_producer: true
+  use_vllm: true
+  async_prefetch: true
+  prefetch_depth: 1
+  vllm_sync_interval: 2
+  vllm_lora_sync: true
+  streaming_partial_batch: true
+  vllm_importance_sampling_correction: true
+  off_policy_mask_threshold: 0.5
+  importance_sampling_level: token
+  num_generations: 8
+  max_completion_length: 512
+  reward_funcs:
+    - rewards.accuracy_reward
+  reroll_start_fraction: 0.5
+  replay_buffer_size: 100
+  reward_num_workers: 4
+  skip_zero_advantage_batches: true
+
+datasets:
+  - path: AI-MO/NuminaMath-TIR
+    type: rewards.prompt_transform
+    split: train
+
+gradient_accumulation_steps: 4
+micro_batch_size: 2
+max_steps: 500
+learning_rate: 1e-5
+bf16: true
+gradient_checkpointing: true
+```
+
+```bash
+# Terminal 1: Start vLLM on GPU 0
+CUDA_VISIBLE_DEVICES=0 axolotl vllm-serve config.yaml
+
+# Terminal 2: Train on GPU 1
+CUDA_VISIBLE_DEVICES=1 axolotl train config.yaml
+```
+
+##### Multi-GPU Async GRPO
+
+Async GRPO supports FSDP and DeepSpeed ZeRO-3 for multi-GPU training. vLLM runs on one GPU while training is distributed across the remaining GPUs.
+
+**FSDP:**
+
+```yaml
+fsdp:
+  - full_shard
+  - auto_wrap
+fsdp_config:
+  fsdp_transformer_layer_cls_to_wrap: Qwen2DecoderLayer
+gradient_checkpointing_kwargs:
+  use_reentrant: false
+```
+
+**DeepSpeed ZeRO-3:**
+
+```yaml
+deepspeed: deepspeed_configs/zero3_bf16.json
+gradient_checkpointing_kwargs:
+  use_reentrant: true   # Required for ZeRO-3
+```
+
+```bash
+# Terminal 1: Start vLLM on GPU 0
+CUDA_VISIBLE_DEVICES=0 axolotl vllm-serve config.yaml
+
+# Terminal 2: Train on GPUs 0,1
+CUDA_VISIBLE_DEVICES=0,1 axolotl train config.yaml
+```
+
+::: {.callout-important}
+With multi-GPU async prefetch, only rank 0 generates completions in the background thread. Results are broadcast to all ranks on the main thread. This avoids FSDP/DeepSpeed collective deadlocks from unsynchronized background threads.
+:::
+
+### GDPO
+
+GDPO (Group Reward-Decoupled Policy Optimization) extends GRPO for multi-reward training. It addresses the **reward advantage collapse** problem by normalizing each reward function independently before combining them.
+
+::: {.callout-tip}
+Use GDPO when training with multiple reward functions. For single reward, GRPO and GDPO produce equivalent results.
+:::
+
+Paper: [https://arxiv.org/pdf/2501.05242](https://arxiv.org/pdf/2501.05242)
+
+GDPO uses TRL's native `multi_objective_aggregation` parameter under the hood. When you set `rl: gdpo`, axolotl automatically configures TRL to use `normalize_then_sum` aggregation.
+
+```yaml
+base_model: Qwen/Qwen2.5-1.5B-Instruct
+
+vllm:
+    host: 0.0.0.0
+    port: 8000
+    tensor_parallel_size: 2
+    gpu_memory_utilization: 0.85
+
+rl: gdpo
+
+trl:
+    beta: 0.001
+    max_completion_length: 256
+    use_vllm: true
+    num_generations: 4
+    reward_funcs:
+        - rewards.format_reward
+        - rewards.correctness_reward
+    reward_weights: [1.0, 2.0]
+
+datasets:
+    - path: openai/gsm8k
+      name: main
+      type: rewards.oai_gsm8k_transform
+```
+
+You can also use GRPO with explicit aggregation control:
+
+```yaml
+rl: grpo
+trl:
+    multi_objective_aggregation: normalize_then_sum  # GDPO behavior
+    # or: sum_then_normalize  # Default GRPO behavior
+```
+
+#### GDPO vs GRPO
+
+| Aspect | GRPO | GDPO |
+|--------|------|------|
+| **Aggregation** | `sum_then_normalize` | `normalize_then_sum` |
+| **Multi-reward** | May collapse advantages | Preserves reward signals |
+| **Single reward** | Standard behavior | Equivalent to GRPO |
+
+#### Why GDPO?
+
+When using multiple rewards with GRPO, different reward combinations can produce identical advantages:
+
+```
+# Example: format + correctness rewards
+[format=0, correct=3] → sum=3
+[format=1, correct=2] → sum=3  ← GRPO sees these as equal!
+[format=2, correct=1] → sum=3
+[format=3, correct=0] → sum=3
+```
+
+GDPO normalizes each reward independently, preserving their relative differences.
+
+#### Reward Functions
+
+GDPO uses the same reward function format as GRPO:
+
+```python
+# rewards.py
+def format_reward(completions, **kwargs) -> list[float]:
+    return [1.0 if len(c) > 10 else 0.0 for c in completions]
+
+def correctness_reward(completions, answers, **kwargs) -> list[float]:
+    rewards = []
+    for completion, answer in zip(completions, answers):
+        # Your scoring logic here
+        rewards.append(score)
+    return rewards
+```
+
+#### Sequence Parallelism
+
+GDPO supports sequence parallelism for long-context training:
+
+```yaml
+rl: gdpo
+context_parallel_size: 2
+```
+
 ### SimPO
 
 SimPO uses [CPOTrainer](https://huggingface.co/docs/trl/main/en/cpo_trainer) but with alternative loss function.
@@ -608,6 +1041,306 @@ simpo_gamma: 0.5  # default in CPOTrainer
 
 This method uses the same dataset format as [DPO](#dpo).
 
+### EBFT {#ebft}
+
+::: {.callout-tip}
+For a detailed guide on EBFT modes, feature extraction, and configuration, see the [EBFT guide](ebft.qmd).
+:::
+
+EBFT (Energy-Based Fine-Tuning) fine-tunes language models by optimizing a **feature-matching loss** rather than relying on external reward functions. A frozen copy of the model extracts embeddings from both generated and ground-truth completions, and the generator is updated via REINFORCE to match the ground-truth feature moments.
+
+Paper: ["Matching Features, Not Tokens: Energy-Based Fine-Tuning of Language Models"](https://arxiv.org/abs/2603.12248) (Jelassi et al., 2026)
+
+**Key advantages:**
+
+- No reward model or verifier required — works on any (prompt, completion) data
+- Applicable to non-verifiable tasks (code, translation, creative writing)
+- Operates on model rollouts (not teacher forcing), reducing distribution shift
+
+EBFT supports two modes:
+
+- **Structured mode**: For QA/instruction data with prompt + completion pairs. Uses vLLM for generation (like GRPO).
+- **Strided mode**: For unstructured text without prompt/completion splits. Uses strided block-parallel generation with flex_attention — no vLLM needed.
+
+#### Structured Mode
+
+```yaml
+base_model: Qwen/Qwen3-4B
+
+rl: ebft
+
+ebft:
+  feature_layers: [0.25, 0.5, 0.75]    # Extract features at 25%, 50%, 75% depth
+  embed_method: last_token
+  use_whitening: false
+  alignment_coef: 1.0                    # Cosine similarity reward weight
+  diversity_coef: 1.0                    # Pairwise dot product penalty
+  ce_coef: 0.0                          # Cross-entropy on GT tokens (0 = off)
+
+trl:
+  num_generations: 4
+  max_completion_length: 256
+  temperature: 0.7
+  use_vllm: true
+  vllm_server_host: 0.0.0.0
+  vllm_server_port: 8000
+  vllm_lora_sync: true                   # LoRA adapter sync (recommended)
+  vllm_sync_interval: 3
+  use_data_producer: true
+  async_prefetch: true                   # Set false for sync mode
+  scale_rewards: true
+  loss_type: grpo
+  epsilon: 0.2
+
+vllm:
+  gpu_memory_utilization: 0.5
+  max_model_len: 2048
+
+datasets:
+  - path: nvidia/OpenCodeInstruct
+    type: ebft_opencode.transform
+    split: train[:500]
+
+adapter: lora
+lora_r: 16
+lora_alpha: 32
+lora_target_linear: true
+```
+
+```bash
+# Terminal 1: Start vLLM
+CUDA_VISIBLE_DEVICES=0 axolotl vllm-serve config.yaml
+
+# Terminal 2: Train
+CUDA_VISIBLE_DEVICES=1 axolotl train config.yaml
+```
+
+#### Strided Mode
+
+For unstructured text (raw code, prose). No vLLM needed — runs on a single GPU.
+
+```yaml
+base_model: meta-llama/Llama-3.2-1B
+
+rl: ebft
+
+ebft:
+  mode: strided
+  stride: 8
+  context_length: 8
+  generate_max_len: 8
+  n_samples_per_prompt: 4
+  temperature: 0.6
+  feature_layers: [0.25, 0.5, 0.75]
+  embed_method: last_token
+  use_whitening: true
+  alignment_coef: 1.0
+  diversity_coef: 1.0
+  rl_coef: 1.0
+  ce_coef: 0.03
+  advantage_estimator: rloo
+
+datasets:
+  - path: nvidia/OpenCodeInstruct
+    type: ebft_strided_structured.transform
+    split: train[:1%]
+
+flash_attention: false
+flex_attention: true     # Strided mode uses flex_attention
+gradient_checkpointing: true
+gradient_checkpointing_kwargs:
+  use_reentrant: true    # Required for flex_attention
+```
+
+```bash
+CUDA_VISIBLE_DEVICES=0 axolotl train config.yaml
+```
+
+::: {.callout-tip}
+See `examples/ebft/` for complete example configs covering Llama 1B/3B/8B and Qwen3 4B/8B models in both modes.
+:::
+
+#### EBFT Configuration Reference
+
+| Parameter | Default | Description |
+|-----------|---------|-------------|
+| `ebft.feature_layers` | `[0.25, 0.5, 0.75]` | Layer depths for feature extraction (fractional) |
+| `ebft.embed_method` | `last_token` | Feature pooling: `last_token`, `mean_pooling`, `concat` |
+| `ebft.use_whitening` | `false` | SVD whitening of feature dimensions |
+| `ebft.alignment_coef` | `1.0` | Cosine similarity reward weight |
+| `ebft.diversity_coef` | `1.0` | Pairwise dot product penalty weight |
+| `ebft.ce_coef` | `0.0` | Cross-entropy loss on ground-truth tokens |
+| `ebft.mode` | `structured` | `structured` (vLLM) or `strided` (no vLLM) |
+| `ebft.stride` | — | Tokens between anchor points (strided mode) |
+| `ebft.context_length` | — | Context window per block (strided mode) |
+| `ebft.generate_max_len` | — | Tokens to generate per block (strided mode) |
+| `ebft.n_samples_per_prompt` | — | Rollouts per document (strided mode) |
+| `ebft.advantage_estimator` | `grpo` | `grpo` or `rloo` (strided mode) |
+
+### NeMo Gym Integration
+
+[NeMo Gym](https://github.com/NVIDIA-NeMo/Gym) provides 50+ verified RL environments (math, coding, tool-use, reasoning) with deterministic reward signals. The axolotl integration supports both **single-turn** (call `/verify` after generation) and **multi-turn** (agent-based tool execution via `/run`).
+
+#### Single-Turn (Simplest)
+
+For environments that only need answer verification (math, coding challenges). No agent server needed — the reward function calls `/verify` directly on the resource server.
+
+```yaml
+base_model: Qwen/Qwen2.5-0.5B-Instruct
+
+rl: grpo
+chat_template: tokenizer_default
+
+trl:
+  use_vllm: false                          # Colocate mode (single GPU)
+  num_generations: 4
+  max_completion_length: 128
+  temperature: 0.9
+  reward_funcs:
+    - axolotl.integrations.nemo_gym.rewards.reward_nemo_gym_verify
+
+plugins:
+  - axolotl.integrations.nemo_gym.NemoGymPlugin
+
+nemo_gym_enabled: true
+nemo_gym_dir: ~/Gym
+nemo_gym_auto_start: false
+nemo_gym_head_port: 11000
+nemo_gym_datasets:
+  - path: resources_servers/reasoning_gym/data/train_basic_arithmetic.jsonl
+    server_name: reasoning_gym
+
+datasets:
+  - path: ~/Gym/resources_servers/reasoning_gym/data/train_basic_arithmetic.jsonl
+    type: chat_template
+    field_messages: responses_create_params.input
+    message_field_content: content
+    message_field_role: role
+```
+
+```bash
+# Terminal 1: Start NeMo Gym resource server
+cd ~/Gym && .venv/bin/ng_run \
+    "+config_paths=[resources_servers/reasoning_gym/configs/resources_only.yaml]" \
+    "+skip_venv_if_present=true"
+
+# Terminal 2: Train
+CUDA_VISIBLE_DEVICES=0 axolotl train config.yaml
+```
+
+::: {.callout-note}
+`nemo_gym_datasets.path` is relative to `nemo_gym_dir`. Don't use absolute paths or they will be double-joined.
+:::
+
+#### Multi-Turn with Async GRPO (Recommended)
+
+For environments with tool-use (weather, search, databases). An agent server orchestrates multi-turn interactions: generate → parse tool calls → execute tools → feed results back → repeat until done.
+
+```yaml
+base_model: Qwen/Qwen3-0.6B
+
+rl: grpo
+chat_template: tokenizer_default
+
+adapter: lora
+lora_r: 16
+lora_alpha: 32
+lora_target_modules: [q_proj, k_proj, v_proj, o_proj, gate_proj, up_proj, down_proj]
+
+trl:
+  use_vllm: true
+  vllm_mode: server
+  vllm_server_host: localhost
+  vllm_server_port: 8000
+  vllm_lora_sync: true
+  vllm_sync_interval: 5
+  use_data_producer: true
+  async_prefetch: true                     # 3x speedup
+  num_generations: 4
+  max_completion_length: 512
+  temperature: 0.8
+  reward_funcs:
+    - axolotl.integrations.nemo_gym.rewards.reward_env
+
+plugins:
+  - axolotl.integrations.nemo_gym.NemoGymPlugin
+
+nemo_gym_enabled: true
+nemo_gym_auto_start: false
+nemo_gym_head_port: 11000
+nemo_gym_multi_turn: true
+nemo_gym_verify_timeout: 120
+nemo_gym_datasets:
+  - path: resources_servers/example_single_tool_call/data/weather_tool_calling.jsonl
+    server_name: example_single_tool_call
+
+datasets:
+  - path: ~/Gym/resources_servers/example_single_tool_call/data/weather_tool_calling.jsonl
+    type: chat_template
+    field_messages: responses_create_params.input
+    message_field_content: content
+    message_field_role: role
+
+vllm:
+  gpu_memory_utilization: 0.85
+  max_model_len: 2048
+```
+
+Multi-turn requires three services running:
+
+```bash
+# Terminal 1: vLLM with LoRA + tool calling
+VLLM_ALLOW_RUNTIME_LORA_UPDATING=1 CUDA_VISIBLE_DEVICES=0 \
+    python -m vllm.entrypoints.openai.api_server \
+    --model Qwen/Qwen3-0.6B --max-model-len 2048 \
+    --gpu-memory-utilization 0.85 \
+    --enable-lora --max-lora-rank 64 \
+    --enable-auto-tool-choice --tool-call-parser hermes
+
+# Terminal 2: NeMo Gym servers (resource + model proxy + agent)
+cd ~/Gym && .venv/bin/ng_run \
+    "+config_paths=[configs/axolotl_tool_calling.yaml]" \
+    "+skip_venv_if_present=true"
+
+# Terminal 3: Training
+CUDA_VISIBLE_DEVICES=1 axolotl train config.yaml
+```
+
+::: {.callout-important}
+Multi-turn requires a NeMo Gym agent config YAML that defines three components: a resource server (tools + `/verify`), a model server proxy (forwards to your vLLM), and an agent server (orchestrates `/run`). See the [NeMo Gym README](https://github.com/NVIDIA-NeMo/Gym) for agent config format.
+:::
+
+#### NeMo Gym Prerequisites
+
+```bash
+# Clone and set up NeMo Gym
+git clone https://github.com/NVIDIA-NeMo/Gym.git ~/Gym
+cd ~/Gym
+uv venv --python 3.12 && source .venv/bin/activate && uv sync
+
+# Fix pycosat build (GCC 13+)
+CFLAGS="" uv pip install pycosat --python .venv/bin/python --no-build-isolation
+```
+
+#### NeMo Gym Configuration Reference
+
+| Parameter | Type | Default | Description |
+|-----------|------|---------|-------------|
+| `nemo_gym_enabled` | bool | — | Enable the NeMo Gym integration |
+| `nemo_gym_dir` | str | `~/Gym` | Path to NeMo Gym repo |
+| `nemo_gym_auto_start` | bool | `true` | Auto-start resource servers |
+| `nemo_gym_head_port` | int | `11000` | Head server port |
+| `nemo_gym_multi_turn` | bool | `false` | Enable multi-turn via agent `/run` |
+| `nemo_gym_verify_timeout` | int | `30` | Per-request timeout (seconds) |
+| `nemo_gym_datasets` | list | required | Dataset configs with `path` and `server_name` |
+
+#### Reward Functions
+
+| Function | Mode | Description |
+|----------|------|-------------|
+| `axolotl.integrations.nemo_gym.rewards.reward_nemo_gym_verify` | Single-turn | Calls `/verify`, returns binary reward |
+| `axolotl.integrations.nemo_gym.rewards.reward_env` | Multi-turn | Passthrough reward from agent `/run` |
+
 ### Using local dataset files
 
 ```yaml

docs/scripts/examples-allowlist.yml

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

docs/scripts/generate_examples_docs.py

@@ -0,0 +1,424 @@
+"""
+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/telemetry.qmd

@@ -0,0 +1,61 @@
+---
+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

docs/training_stability.qmd

@@ -0,0 +1,399 @@
+---
+title: "Training Stability & Debugging"
+order: 15
+description: "Guide to monitoring, debugging, and stabilizing training runs in axolotl"
+---
+
+This guide covers practical techniques for monitoring training health, diagnosing instability, and resolving common failures in both supervised fine-tuning (SFT) and reinforcement learning (GRPO/EBFT) workflows.
+
+## Monitoring Training
+
+### Key Metrics for SFT
+
+Every SFT run should be monitored through at least these four metrics:
+
+| Metric | What It Tells You | Healthy Range |
+|--------|-------------------|---------------|
+| `train/loss` | How well the model fits training data | Decreasing; typically 0.5--2.0 for chat fine-tuning |
+| `eval/loss` | Generalization performance | Tracks train loss with small gap; divergence signals overfitting |
+| `grad_norm` | Gradient magnitude | 0.1--10.0; spikes above 100 indicate instability |
+| `learning_rate` | Current LR from scheduler | Should follow expected schedule (warmup then decay) |
+
+::: {.callout-tip}
+## Set Up Logging Early
+Enable W&B or TensorBoard from the start. Debugging a failed run without metrics is guesswork.
+
+```yaml
+wandb_project: my-project
+wandb_run_id:   # optional, for resuming
+logging_steps: 1
+```
+:::
+
+### Key Metrics for RL (GRPO)
+
+GRPO training logs a richer set of metrics. These are the critical ones:
+
+| Metric | Healthy Range | Red Flag |
+|--------|---------------|----------|
+| `rewards/<name>/mean` | > 0.15 within 20 steps | Stays at 0 -- reward function is broken or task is too hard |
+| `reward_std` | > 0 on most steps | Always 0 -- no learning signal (all completions get the same reward) |
+| `frac_reward_zero_std` | < 0.8 | 1.0 on every step -- zero-advantage skip fires constantly, no gradient updates |
+| `grad_norm` | 0.001--1.0 | 0.0 is acceptable occasionally (zero-adv skip); > 10.0 is unstable |
+| `entropy` | 0.05--0.5 | < 0.01 suggests mode collapse; > 1.0 suggests the model is not converging |
+| `kl` | 0.0--0.5 | > 2.0 suggests policy has diverged too far from reference |
+| `sampling/sampling_logp_difference/mean` | < 0.1 | > 1.0 means policy has diverged far from vLLM server weights |
+| `sampling/importance_sampling_ratio/min` | > 0.1 | Near 0 indicates stale off-policy data; increase `vllm_sync_interval` |
+| `clip_ratio/region_mean` | < 0.1 | > 0.3 means PPO clipping is too aggressive |
+| `completions/mean_length` | Task-dependent | Monotonically increasing to max length suggests reward hacking |
+| `completions/clipped_ratio` | < 0.3 | > 0.8 means most completions hit `max_completion_length` -- increase it |
+
+::: {.callout-note}
+## EBFT-Specific Metrics
+For EBFT training, also monitor `ebft/alignment` (should trend upward, healthy 0.3--0.9), `ebft/diversity` (healthy 0.01--0.1; > 1.0 indicates mode collapse), and `ebft/cfm_loss` (should trend downward, < 10).
+:::
+
+## SFT Stability
+
+### Loss Plateau
+
+**Symptom**: Loss stops decreasing early in training, well above expected values.
+
+**Causes and fixes**:
+
+- **Learning rate too low**: Increase by 2--5x. Typical ranges: full fine-tune 1e-5 to 5e-5, LoRA 1e-4 to 3e-4.
+- **Insufficient warmup**: Set `warmup_steps` to 5--10% of total steps. Too-aggressive learning at the start can push the model into a flat region.
+- **Data quality**: Check that labels are correctly masked. Use `axolotl preprocess` and inspect tokenized samples to confirm only the target tokens are trainable.
+- **Weight decay too high**: Default 0.01 is usually fine. Values above 0.1 can suppress learning in LoRA.
+
+### Loss Spikes
+
+**Symptom**: Loss suddenly jumps by 2--10x then (possibly) recovers.
+
+**Causes and fixes**:
+
+- **Bad data samples**: A single malformed or extremely long example can cause a spike. Enable `sample_packing: false` temporarily and check if spikes correlate with specific batches.
+- **Learning rate too high**: Reduce by 2--5x, or increase warmup.
+- **Gradient accumulation mismatch**: Effective batch size = `micro_batch_size * gradient_accumulation_steps * num_gpus`. Very large effective batch sizes amplify gradient noise.
+- **Mixed precision issues**: With `bf16: true`, some operations can lose precision. If spikes are severe, try `fp32` for diagnosis.
+
+### Overfitting
+
+**Symptom**: Train loss keeps decreasing but eval loss starts increasing.
+
+**Fixes**:
+
+- Increase `val_set_size` (e.g., 0.05) and monitor `eval/loss`.
+- Reduce `num_epochs` or `max_steps`.
+- Increase `weight_decay` (try 0.01--0.1).
+- Use a smaller LoRA rank (`lora_r`). Typical values: 8--32.
+- Increase dropout: `lora_dropout: 0.05`.
+
+## RL/GRPO Stability
+
+### Reward Never Increases
+
+If `rewards/*/mean` stays at 0 for more than 20 steps:
+
+1. **Test reward function standalone**: Run it outside training with known inputs to verify it returns nonzero values.
+   ```bash
+   cd experiments && python -c "import my_rewards; print(my_rewards.accuracy_reward(...))"
+   ```
+2. **Check dataset columns**: The reward function receives `**kwargs` containing dataset columns. Verify the columns it needs (e.g., `answer`) are not removed by the dataset transform.
+3. **Check completion content**: Enable `log_completions: true` in the `trl:` config and inspect logged completions in W&B. If completions are empty or incoherent, the model may be too weak for the task.
+4. **Verify vLLM is serving the right model**: Hit the vLLM health endpoint and confirm the model name matches your config.
+
+### Entropy Collapse (Mode Collapse)
+
+**Symptom**: `entropy` drops below 0.01; all completions become nearly identical.
+
+**Fixes**:
+
+- Increase `temperature` in generation kwargs (try 0.8--1.0).
+- Reduce learning rate.
+- Add a KL penalty term (`beta` parameter in GRPO config).
+- Check that `num_generations` is sufficient (16+ gives better advantage estimates).
+
+### IS Ratio Divergence
+
+**Symptom**: `sampling/importance_sampling_ratio/min` drops near 0, or `sampling/sampling_logp_difference/mean` exceeds 1.0.
+
+This means the policy has diverged significantly from the weights used by vLLM for generation. The importance sampling correction becomes unreliable.
+
+**Fixes**:
+
+- Decrease `vllm_sync_interval` (sync weights more often).
+- Enable `off_policy_mask_threshold` (e.g., 0.5) to mask stale off-policy samples.
+- Use `importance_sampling_level: token` for finer-grained correction.
+
+### Gradient Norm Instability
+
+**Symptom**: `grad_norm` oscillates wildly or exceeds 10.0 regularly.
+
+**Fixes**:
+
+- Enable gradient clipping: `max_grad_norm: 1.0` (default in most configs).
+- Reduce learning rate.
+- Increase `gradient_accumulation_steps` to smooth out noisy batches.
+- Check for NaN issues (see next section).
+
+## NaN and Inf Handling
+
+### Common Causes
+
+| Cause | Where It Manifests | Detection |
+|-------|-------------------|-----------|
+| FP8 zero-scale division | Forward pass logits | `grad_norm: nan`, loss becomes NaN immediately |
+| Gradient explosion | Backward pass | `grad_norm` spikes to inf, then loss goes NaN |
+| Bad data (empty sequences) | Logprob computation | NaN in specific batches only |
+| Numerical overflow in log-softmax | Loss computation | Large negative logprobs cause exp() overflow |
+
+### FP8-Specific NaN Issues
+
+FP8 quantization (`fp8: true`) can produce NaN when the activation quantization kernel divides by `max(abs(x)) / 448`. If the input tensor is all zeros (e.g., padding positions), the scale becomes 0, causing division by zero.
+
+**Fixes applied in axolotl**:
+
+- The `act_quant_kernel` has a zero-guard: `s = tl.where(s == 0, 1.0, s)`.
+- A safety net `nan_to_num(logits, nan=0.0)` is applied in `_get_per_token_logps_and_entropies`.
+- Embedding padding is zero-padded for FP8 compatibility.
+
+::: {.callout-important}
+## After Modifying Triton Kernels
+If you patch any Triton JIT kernel (e.g., the FP8 quantization kernels in transformers), you must clear the Triton cache for changes to take effect:
+
+```bash
+rm -rf ~/.triton/cache
+```
+:::
+
+### General NaN Debugging Steps
+
+1. **Enable anomaly detection** (slow, but pinpoints the source):
+   ```python
+   torch.autograd.set_detect_anomaly(True)
+   ```
+2. **Check grad_norm**: If it goes to NaN, the backward pass is the problem. If loss is NaN but grad_norm was fine on the previous step, the forward pass is the problem.
+3. **Reduce to single GPU, single batch**: Eliminate distributed training variables.
+4. **Inspect data**: Print the batch that triggers NaN. Look for empty sequences, extreme token IDs, or unexpected padding patterns.
+
+## OOM Debugging
+
+Out-of-memory errors are the most common training failure. Use this systematic approach, from least to most disruptive:
+
+### Step 1: Reduce Batch Size
+
+The single highest-impact change. VRAM scales roughly linearly with batch size.
+
+```yaml
+micro_batch_size: 1              # Start here
+gradient_accumulation_steps: 16  # Increase to maintain effective batch size
+```
+
+For GRPO specifically, the logits tensor for policy logprob computation can be very large. `batch_size * num_generations * seq_len * vocab_size` in bf16. For example, with `num_generations: 16` and `micro_batch_size: 8`, the logits tensor alone is:
+
+```
+8 * 16 * 2048 * 151936 * 2 bytes = ~75 GB  (way too large)
+```
+
+Reduce `micro_batch_size` to 2--4 for GRPO.
+
+### Step 2: Enable Gradient Checkpointing
+
+Trades compute for memory by recomputing activations during the backward pass instead of storing them.
+
+```yaml
+gradient_checkpointing: true
+gradient_checkpointing_kwargs:
+  use_reentrant: false     # Recommended default
+```
+
+::: {.callout-warning}
+## Reentrant Checkpointing Exceptions
+Some configurations require `use_reentrant: true`:
+
+- DeepSpeed ZeRO-3 (non-reentrant causes `CheckpointError`)
+- EBFT strided mode with flex_attention
+:::
+
+### Step 3: Use Quantization
+
+Load the base model in reduced precision:
+
+```yaml
+# 4-bit QLoRA
+adapter: qlora
+load_in_4bit: true
+
+# 8-bit
+load_in_8bit: true
+
+# FP8 (saves ~50% model VRAM, same compute speed as bf16)
+fp8: true
+```
+
+### Step 4: Reduce Sequence Length
+
+```yaml
+sequence_len: 1024     # Down from 2048 or 4096
+```
+
+For GRPO, also reduce `max_completion_length`. Memory scales quadratically with sequence length when using standard attention.
+
+### Step 5: Use Flash Attention
+
+Reduces attention memory from O(n^2) to O(n):
+
+```yaml
+flash_attention: true
+```
+
+### Step 6: Offload with DeepSpeed
+
+For extreme cases, offload optimizer states or parameters to CPU:
+
+```yaml
+deepspeed: deepspeed_configs/zero3_bf16.json
+```
+
+### Diagnosing the Specific Culprit
+
+Use the `profiler_steps` config option to capture GPU memory snapshots:
+
+```yaml
+profiler_steps: [1, 2]
+```
+
+This generates PyTorch profiler traces you can inspect to see exactly which tensor allocation caused the OOM.
+
+## Common Errors
+
+| Error Message | Likely Cause | Fix |
+|---------------|-------------|-----|
+| `exitcode: -9` | System RAM exhaustion | Reduce dataset size, `dataset_num_proc`, or number of data workers |
+| `exitcode: -7` (DeepSpeed) | DeepSpeed version issue | `pip install -U deepspeed` |
+| `CUDA out of memory` | GPU VRAM exhaustion | Follow OOM debugging steps above |
+| `RuntimeError: NCCL communicator was aborted` | GPU communication failure | See [NCCL docs](nccl.qmd); check `NCCL_DEBUG=INFO` output |
+| `ValueError: Asking to pad but the tokenizer does not have a padding token` | Missing pad token | Add `special_tokens: { pad_token: "<\|endoftext\|>" }` to config |
+| `'DummyOptim' object has no attribute 'step'` | DeepSpeed on single GPU | Remove `deepspeed:` section from config |
+| `unable to load strategy X` then `None is not callable` | Reward module not importable | Run `cd experiments && python -c "import my_rewards"` to check |
+| `generation_batch_size not divisible by num_generations` | micro_batch_size too small | Set `micro_batch_size >= num_generations` and make it divisible |
+| `'weight' must be 2-D` | FSDP1 flattened parameters | Use `fsdp_version: 2` or skip `unwrap_model` when FSDP is enabled |
+| `CheckpointError` (tensor count mismatch) | Non-reentrant checkpointing + ZeRO-3 or flex_attention | Set `use_reentrant: true` in `gradient_checkpointing_kwargs` |
+| `BFloat16` TypeError during weight sync | NumPy does not support bf16 | Fixed in axolotl's `weight_serde.py` (auto bf16 to fp16 conversion) |
+| `Content end boundary is before start boundary` | Chat template parsing issue | Check `eos_token` matches template; file a GitHub issue if persistent |
+| `CAS service error` during data processing | HuggingFace XET issue | Set `export HF_HUB_DISABLE_XET=1` |
+| Training hangs (multi-GPU) | FSDP + async prefetch deadlock | Set `async_prefetch: false` with FSDP |
+
+## Profiling
+
+### PyTorch Profiler
+
+Axolotl supports PyTorch profiler integration via the config:
+
+```yaml
+profiler_steps: [1, 2, 3]
+```
+
+This captures profiler traces for the specified steps. View them in TensorBoard:
+
+```bash
+tensorboard --logdir output_dir/runs
+```
+
+Or open the `.json` trace file in `chrome://tracing`.
+
+### CUDA Memory Snapshots
+
+For detailed memory analysis, use PyTorch's memory snapshot API. Add this to your training script or use it interactively:
+
+```python
+import torch
+
+# Enable memory history tracking
+torch.cuda.memory._record_memory_history()
+
+# ... run your training step ...
+
+# Save snapshot
+torch.cuda.memory._dump_snapshot("memory_snapshot.pickle")
+```
+
+Visualize with PyTorch's memory visualizer:
+
+```bash
+python -m torch.cuda.memory._viz memory_snapshot.pickle
+```
+
+### Quick GPU Memory Check
+
+During training, monitor GPU utilization in a separate terminal:
+
+```bash
+watch -n 1 nvidia-smi
+```
+
+For programmatic access within axolotl, the logged metrics `memory/max_alloc` and `memory/max_reserved` come from `torch.cuda.max_memory_allocated()` and `torch.cuda.max_memory_reserved()`. Note these report PyTorch's view of memory, which may differ from `nvidia-smi` (see [FAQ](faq.qmd)).
+
+## W&B and Logging
+
+### Enabling Logging
+
+```yaml
+wandb_project: my-project
+wandb_entity: my-team          # optional
+wandb_run_id: run-123          # optional, for resuming
+wandb_name: experiment-name    # optional
+logging_steps: 1               # log every step (recommended for RL)
+```
+
+### Debug Logging
+
+For detailed axolotl-internal debug output:
+
+```bash
+AXOLOTL_LOG_LEVEL=DEBUG axolotl train config.yaml 2>&1 | tee /tmp/training.log
+```
+
+::: {.callout-tip}
+## Always Log to a File
+Pipe training output to a log file so you can inspect it after the run:
+
+```bash
+axolotl train config.yaml 2>&1 | tee /tmp/my_run.log
+```
+:::
+
+### What Axolotl Logs
+
+**SFT metrics** (logged every `logging_steps`):
+
+- `train/loss`, `eval/loss` -- training and validation loss
+- `train/grad_norm` -- gradient L2 norm (before clipping)
+- `train/learning_rate` -- current learning rate
+- `memory/max_alloc`, `memory/max_reserved` -- peak GPU memory
+
+**GRPO/RL metrics** (logged every step):
+
+- `rewards/<name>/mean`, `rewards/<name>/std` -- per-reward-function statistics
+- `reward`, `reward_std` -- aggregated reward across all reward functions
+- `frac_reward_zero_std` -- fraction of prompt groups where all completions got the same reward
+- `completions/mean_length`, `completions/min_length`, `completions/max_length` -- completion token lengths
+- `completions/clipped_ratio` -- fraction of completions that hit the max length
+- `completions/mean_terminated_length`, `completions/min_terminated_length`, `completions/max_terminated_length` -- lengths of naturally terminated completions
+- `kl` -- KL divergence between policy and reference
+- `entropy` -- policy entropy (measure of output diversity)
+- `clip_ratio/region_mean`, `clip_ratio/low_mean`, `clip_ratio/high_mean` -- PPO clipping statistics
+- `sampling/sampling_logp_difference/mean`, `sampling/sampling_logp_difference/max` -- log-probability difference between policy and sampling distribution
+- `sampling/importance_sampling_ratio/min`, `sampling/importance_sampling_ratio/mean`, `sampling/importance_sampling_ratio/max` -- IS ratio statistics for off-policy correction
+- `num_tokens` -- total tokens processed
+
+### Reading W&B Charts
+
+For a healthy GRPO run, expect to see:
+
+1. **`reward/mean`**: Gradual upward trend. May start near 0 and reach 0.3--0.8 depending on task difficulty. Not monotonic -- fluctuations are normal.
+2. **`entropy`**: Gradual decrease from initial values (often 0.3--0.6) as the model becomes more confident. Should not collapse to near-zero.
+3. **`grad_norm`**: Mostly in the 0.001--1.0 range. Occasional 0.0 values are fine (zero-advantage skip). Persistent values above 10.0 need investigation.
+4. **`kl`**: Starts near 0 and grows slowly. If it shoots up rapidly, the policy is diverging from the reference.
+5. **`completions/mean_length`**: Should reflect the task's natural answer length. If it steadily increases to `max_completion_length`, the model may be reward-hacking by generating longer outputs.

docs/vllm_serving.qmd

@@ -0,0 +1,318 @@
+---
+title: "vLLM Serving for GRPO Training"
+description: "How to configure and run vLLM as a generation backend for GRPO reinforcement learning in Axolotl."
+format:
+  html:
+    toc: true
+    toc-depth: 3
+    number-sections: true
+execute:
+  enabled: false
+---
+
+## Overview {#sec-overview}
+
+GRPO (Group Relative Policy Optimization) trains a language model by generating completions, scoring them with reward functions, and updating the policy to favor higher-reward outputs. The generation step is the bottleneck: producing thousands of tokens per training step with the policy model is slow using standard HuggingFace generation.
+
+Axolotl uses [vLLM](https://github.com/vllm-project/vllm) as a high-throughput generation backend. vLLM runs as a separate process (either on a dedicated GPU or colocated on the training GPU) and serves completions via an HTTP API. The trainer sends prompts to vLLM, receives completions, scores them, and performs gradient updates.
+
+```
+┌──────────────────────┐       HTTP        ┌──────────────────────┐
+│   Trainer (GPU 1)    │ ───────────────── │   vLLM Server (GPU 0)│
+│                      │  prompts/compls   │                      │
+│  - Policy model      │ ◄──────────────── │  - Same base model   │
+│  - Reward scoring    │                   │  - Fast generation   │
+│  - Gradient updates  │  weight sync      │  - LoRA adapter      │
+│  - LoRA adapter      │ ─────────────────►│    (periodically     │
+│                      │  (every N steps)  │     updated)         │
+└──────────────────────┘                   └──────────────────────┘
+```
+
+::: {.callout-important}
+vLLM must serve the **same base model** specified in your training config. If the models do not match, weight synchronization will silently produce incorrect results.
+:::
+
+## Server Mode {#sec-server-mode}
+
+Server mode runs vLLM as an external process on dedicated GPU(s). This is the recommended configuration for most setups.
+
+### Starting the Server
+
+Use the `axolotl vllm-serve` command with your training config:
+
+```bash
+# Terminal 1: Start vLLM on GPU 0
+CUDA_VISIBLE_DEVICES=0 axolotl vllm-serve grpo_config.yaml
+```
+
+```bash
+# Terminal 2: Start training on GPU 1
+CUDA_VISIBLE_DEVICES=1 axolotl train grpo_config.yaml
+```
+
+The server reads vLLM settings from the `vllm:` section of your config and starts an HTTP server (default: `http://0.0.0.0:8000`).
+
+::: {.callout-tip}
+Use `tmux` or `screen` to manage the vLLM server process. Typical startup time is 30-90 seconds depending on model size and whether CUDA graphs are captured.
+:::
+
+### Minimal Server Config
+
+```yaml
+base_model: Qwen/Qwen2.5-1.5B-Instruct
+
+vllm:
+  host: 0.0.0.0
+  port: 8000
+  gpu_memory_utilization: 0.85
+  dtype: auto
+  max_model_len: 4096
+
+rl: grpo
+trl:
+  use_vllm: true
+  vllm_server_host: 0.0.0.0
+  vllm_server_port: 8000
+  vllm_server_timeout: 300
+```
+
+### Multi-GPU vLLM
+
+For larger models, use tensor parallelism across multiple GPUs:
+
+```yaml
+vllm:
+  tensor_parallel_size: 2
+  gpu_memory_utilization: 0.85
+```
+
+```bash
+# vLLM on GPUs 2,3; training on GPUs 0,1
+CUDA_VISIBLE_DEVICES=2,3 axolotl vllm-serve grpo_config.yaml
+CUDA_VISIBLE_DEVICES=0,1 axolotl train grpo_config.yaml --num-processes 2
+```
+
+::: {.callout-note}
+Due to how TRL maps vLLM device indices, the vLLM instance should use the **last** N GPUs (highest device indices), while training uses the first N.
+:::
+
+## Colocate Mode {#sec-colocate-mode}
+
+Colocate mode runs vLLM on the same GPU as the trainer. This is useful when you only have a single GPU.
+
+```yaml
+trl:
+  use_vllm: true
+  vllm_mode: colocate
+  vllm_enable_sleep_mode: true
+```
+
+With `vllm_enable_sleep_mode: true`, vLLM offloads its VRAM allocation when not actively generating, freeing memory for training. When the trainer needs new completions, vLLM wakes up and reclaims VRAM.
+
+::: {.callout-warning}
+Colocate mode is significantly slower than server mode because generation and training cannot overlap. The GPU alternates between the two workloads. This mode is practical only for smaller models (up to ~3B on a 24 GB GPU).
+:::
+
+**When to use colocate mode:**
+
+- You have exactly one GPU
+- The model fits in memory with both vLLM and training active (with sleep mode), or is small enough to time-share
+- You accept the performance tradeoff for simpler setup (no separate vLLM process to manage)
+
+**When to use server mode:**
+
+- You have two or more GPUs
+- You want maximum throughput (generation overlaps with training via async prefetch)
+- You are running larger models (7B+)
+
+## LoRA Sync {#sec-lora-sync}
+
+LoRA sync is the recommended weight synchronization method when training with LoRA adapters. Instead of merging adapter weights into the base model and broadcasting the full merged weights over NCCL, it saves only the LoRA adapter files to the filesystem and tells vLLM to load them natively.
+
+### How It Works
+
+1. The trainer calls `model.save_pretrained()` to write the LoRA adapter weights to a temporary directory
+2. The trainer sends an HTTP POST to `/set_lora_adapter/` on the vLLM server
+3. vLLM loads the adapter using its native LoRA support (Punica kernels)
+4. Generation uses the updated adapter on the next request
+
+### Benefits
+
+- **Smaller sync payload**: Transfers ~40 MB of LoRA weights instead of ~1.4 GB+ of merged model weights (for a typical 0.5-3B model)
+- **No NCCL communicator**: Eliminates the need for a cross-GPU NCCL communication channel, removing GPU contention between vLLM generation and weight sync
+- **Faster sync**: ~200 ms per sync vs. 350 ms to 5+ seconds for NCCL merge sync
+- **Simpler multi-GPU**: No need to set up NCCL groups between trainer and vLLM processes
+
+### Configuration
+
+```yaml
+adapter: lora
+lora_r: 32
+lora_alpha: 64
+lora_target_linear: true
+
+trl:
+  vllm_lora_sync: true    # Enables LoRA sync mode
+  vllm_sync_interval: 5   # Sync every 5 training steps
+```
+
+Setting `vllm_lora_sync: true` automatically selects the LoRA-aware vLLM serve script (`axolotl.scripts.vllm_serve_lora`). You do not need to set `vllm.serve_module` manually.
+
+::: {.callout-important}
+LoRA sync requires that you are training with a LoRA adapter (`adapter: lora` or `adapter: qlora`). It is not applicable to full fine-tuning.
+:::
+
+## Weight Synchronization {#sec-weight-sync}
+
+During GRPO training, the policy model on the trainer is continuously updated via gradient steps. The vLLM server, however, still holds the old weights. Periodically, the trainer must push updated weights to vLLM so that future generations reflect the improved policy.
+
+### Sync Interval
+
+The `vllm_sync_interval` parameter controls how often weights are synced:
+
+```yaml
+trl:
+  vllm_sync_interval: 5   # Sync every 5 optimizer steps
+```
+
+**Tradeoffs:**
+
+- **Lower interval** (e.g., 1-3): Fresher generations, better on-policy data, but more sync overhead per step
+- **Higher interval** (e.g., 5-10): Less overhead, but generations become increasingly off-policy between syncs
+- **Recommended**: 3-5 for most setups. Axolotl includes importance sampling correction (`vllm_importance_sampling_correction: true`) to handle mild distribution mismatch from stale vLLM weights.
+
+### Sync Methods
+
+| Method | Config | Payload | Mechanism | Typical Time |
+|--------|--------|---------|-----------|-------------|
+| **LoRA sync** | `vllm_lora_sync: true` | LoRA adapter only (~40 MB) | Filesystem + HTTP | ~200 ms |
+| **NCCL merge sync** | Default (no lora_sync) | Full merged weights (~1.4 GB+) | HTTP trigger + NCCL broadcast | 350 ms - 5 s |
+
+::: {.callout-tip}
+If you are training with LoRA (which is recommended for GRPO), always enable `vllm_lora_sync: true`. The performance difference is substantial, especially as training progresses and NCCL contention increases.
+:::
+
+### Importance Sampling Correction
+
+When vLLM weights are stale (between syncs), the generated data is slightly off-policy. Axolotl can correct for this:
+
+```yaml
+trl:
+  vllm_importance_sampling_correction: true
+  importance_sampling_level: token          # 'token' or 'sequence'
+  off_policy_mask_threshold: 0.5            # KL threshold for masking stale sequences
+```
+
+- **Token-level IS** is recommended when using Liger kernel (sequence-level has numerical issues with chunked computation)
+- **Off-policy sequence masking (OPSM)** drops sequences that have diverged too far from the current policy, providing a safety net against stale data
+
+## Restart Requirements {#sec-restart}
+
+::: {.callout-warning}
+**vLLM must be restarted between training runs.** Weight syncs from a previous run leave the server in a corrupted state. If you start a new training run against a stale vLLM server, the model may fail to learn.
+:::
+
+### When to Restart
+
+- Before every new training experiment
+- After a training run crashes or is interrupted
+- If you change the base model in your config
+
+### How to Restart
+
+Killing vLLM reliably requires terminating both the main process and its background EngineCore subprocess:
+
+```bash
+# Kill all vLLM-related processes
+pkill -9 -f "vllm|EngineCore"
+
+# Verify GPU memory is freed
+nvidia-smi
+
+# Restart the server
+CUDA_VISIBLE_DEVICES=0 axolotl vllm-serve grpo_config.yaml
+```
+
+::: {.callout-tip}
+A single `kill` often does not fully stop vLLM. Always use `kill -9` and verify with `nvidia-smi` that GPU memory has been released before restarting.
+:::
+
+### Health Check
+
+The vLLM server exposes a health endpoint. Wait for it to return 200 before starting training:
+
+```bash
+# For the LoRA serve script (trailing slash required)
+curl http://localhost:8000/health/
+
+# For the default TRL serve script
+curl http://localhost:8000/health
+```
+
+## Configuration Reference {#sec-config-reference}
+
+### vLLM Server Options (`vllm:` section)
+
+These control the vLLM server process started by `axolotl vllm-serve`.
+
+| Option | Type | Default | Description |
+|--------|------|---------|-------------|
+| `host` | str | `0.0.0.0` | Host address for the vLLM server |
+| `port` | int | `8000` | Port for the vLLM server |
+| `device` | str | `auto` | Device to use for vLLM |
+| `tensor_parallel_size` | int | `None` | Number of GPUs for tensor parallelism |
+| `data_parallel_size` | int | `None` | Number of data parallel replicas |
+| `gpu_memory_utilization` | float | `0.9` | Fraction of GPU memory for vLLM (0.0-1.0) |
+| `dtype` | str | `auto` | Data type (`auto`, `float16`, `bfloat16`) |
+| `max_model_len` | int | `None` | Maximum model context length. Set explicitly if the default is too large for your GPU |
+| `enable_prefix_caching` | bool | `None` | Enable prefix caching for repeated prompt prefixes |
+| `enable_reasoning` | bool | `None` | Enable reasoning mode for models with thinking tokens |
+| `reasoning_parser` | str | `None` | Parser for reasoning output |
+| `enforce_eager` | bool | `None` | Disable CUDA graph capture (required for some architectures like Qwen3.5 hybrid attention) |
+| `serve_module` | str | `None` | Python module for vLLM serve script. Auto-set when `vllm_lora_sync: true` |
+| `worker_extension_cls` | str | `None` | vLLM worker extension class for weight sync |
+
+### Trainer vLLM Options (`trl:` section)
+
+These control how the trainer interacts with vLLM.
+
+| Option | Type | Default | Description |
+|--------|------|---------|-------------|
+| `use_vllm` | bool | `false` | Enable vLLM for generation |
+| `vllm_mode` | str | `None` | `server` (external process) or `colocate` (same GPU) |
+| `vllm_server_host` | str | `0.0.0.0` | Host of the vLLM server to connect to |
+| `vllm_server_port` | int | `8000` | Port of the vLLM server to connect to |
+| `vllm_server_timeout` | int | `None` | Timeout in seconds for vLLM requests |
+| `vllm_lora_sync` | bool | `false` | Sync LoRA adapters via filesystem instead of NCCL merge |
+| `vllm_sync_interval` | int | `None` | Sync weights every N optimizer steps |
+| `vllm_enable_sleep_mode` | bool | `None` | Offload vLLM VRAM when idle (colocate mode) |
+| `vllm_guided_decoding_regex` | str | `None` | Regex constraint for guided decoding |
+
+For async pipeline and off-policy correction options, see the [GRPO Configuration Reference](grpo.qmd#configuration-reference).
+
+## Complete Example {#sec-complete-example}
+
+For a full working GRPO config including vLLM, LoRA sync, async generation, rewards, and dataset setup, see the [GRPO Quick Start](grpo.qmd#quick-start). That config includes all the vLLM settings covered in this guide.
+
+```bash
+# Terminal 1: Start vLLM
+CUDA_VISIBLE_DEVICES=0 axolotl vllm-serve grpo_config.yaml
+
+# Wait for health check to pass
+curl http://localhost:8000/health/
+
+# Terminal 2: Start training
+CUDA_VISIBLE_DEVICES=1 axolotl train grpo_config.yaml
+```
+
+## Troubleshooting {#sec-troubleshooting}
+
+| Problem | Likely Cause | Solution |
+|---------|-------------|----------|
+| Training hangs waiting for vLLM | Server not started or wrong port | Check `curl http://localhost:8000/health/` and verify `vllm_server_host`/`vllm_server_port` match |
+| OOM on vLLM GPU | `gpu_memory_utilization` too high or `max_model_len` too large | Reduce `gpu_memory_utilization` to 0.7 or set `max_model_len` explicitly |
+| OOM on training GPU | Batch too large for policy logprobs | Reduce `micro_batch_size` or `num_generations` |
+| Accuracy stays at zero | Stale vLLM from previous run | Restart vLLM: `pkill -9 -f "vllm\|EngineCore"`, verify with `nvidia-smi`, restart |
+| `ResponseValidationError` from vLLM | Missing logprobs in response | Ensure you are using the correct serve module (auto-selected with `vllm_lora_sync: true`) |
+| Weight sync takes 5+ seconds | NCCL contention with vLLM generation | Switch to `vllm_lora_sync: true` to eliminate NCCL |
+| `async_prefetch` deadlocks with FSDP | Background threads run unsynchronized FSDP collectives | Set `async_prefetch: false` when using FSDP or DeepSpeed multi-GPU |

examples/LiquidAI/README.md

@@ -6,6 +6,8 @@ LFM2 features a new hybrid Liquid architecture with multiplicative gates, short-
 
 This guide shows how to fine-tune both the LFM2 and LFM2-VL models with Axolotl.
 
+Thanks to the team at LiquidAI for giving us early access to prepare for these releases.
+
 ## Getting Started
 
 1.  Install Axolotl following the [installation guide](https://docs.axolotl.ai/docs/installation.html).
@@ -31,6 +33,14 @@ This guide shows how to fine-tune both the LFM2 and LFM2-VL models with Axolotl.
     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:
@@ -45,14 +55,13 @@ This guide shows how to fine-tune both the LFM2 and LFM2-VL models with Axolotl.
 
 ## Optimization Guides
 
-- [Multi-GPU Training](https://docs.axolotl.ai/docs/multi-gpu.html)
-- [LoRA Optimizations](https://docs.axolotl.ai/docs/lora_optims.html)
-- [Multi-Node Training](https://docs.axolotl.ai/docs/multi-node.html)
+- [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/lfm2-350m-fft.yaml

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

examples/LiquidAI/lfm2-8b-a1b-lora.yaml

@@ -0,0 +1,59 @@
+base_model: LiquidAI/LFM2-8B-A1B
+
+plugins:
+  - axolotl.integrations.cut_cross_entropy.CutCrossEntropyPlugin
+
+load_in_8bit: true
+
+eot_tokens:
+  - "<|im_end|>"
+datasets:
+  - path: mlabonne/FineTome-100k
+    type: chat_template
+    split: train[:20%]
+    field_messages: conversations
+    message_field_role: from
+    message_field_content: value
+dataset_prepared_path: last_run_prepared
+val_set_size: 0.05
+output_dir: ./outputs/out
+
+sequence_len: 4096
+sample_packing: true
+
+adapter: lora
+lora_model_dir:
+
+lora_r: 32
+lora_alpha: 16
+lora_dropout: 0.05
+lora_target_modules: 'model.layers.[\d]+.(mlp|cross_attn|self_attn).(up|down|gate|q|k|v|o)_proj'
+
+wandb_project:
+wandb_entity:
+wandb_watch:
+wandb_name:
+wandb_log_model:
+
+gradient_accumulation_steps: 2
+micro_batch_size: 4
+num_epochs: 1
+optimizer: adamw_torch_fused
+lr_scheduler: cosine
+learning_rate: 5e-5
+
+bf16: true
+tf32: true
+
+gradient_checkpointing: true
+resume_from_checkpoint:
+logging_steps: 1
+flash_attention: true
+
+warmup_ratio: 0.1
+evals_per_epoch: 2
+saves_per_epoch: 1
+
+weight_decay: 0.0
+
+# save_first_step: true  # uncomment this to validate checkpoint saving works with your config

examples/LiquidAI/lfm2-vl-lora.yaml

@@ -3,6 +3,9 @@ trust_remote_code: true
 model_type: AutoModelForImageTextToText
 processor_type: AutoProcessor
 
+plugins:
+  - axolotl.integrations.cut_cross_entropy.CutCrossEntropyPlugin
+
 # these 3 lines are needed for now to handle vision chat templates w images
 skip_prepare_dataset: true
 remove_unused_columns: false

examples/alst/README.md

@@ -7,3 +7,24 @@ 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

@@ -0,0 +1,110 @@
+# 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==26.0 setuptools==75.8.0 wheel ninja
+pip3 install --no-build-isolation -e '.[flash-attn]'
+
+# Install CCE https://docs.axolotl.ai/docs/custom_integrations.html#cut-cross-entropy
+python scripts/cutcrossentropy_install.py | sh
+```
+
+2. (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

@@ -0,0 +1,64 @@
+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

@@ -17,8 +17,11 @@ Thanks to the team at Arcee.ai for using Axolotl in supervised fine-tuning the A
 git clone https://github.com/axolotl-ai-cloud/axolotl.git
 cd axolotl
 
-pip3 install packaging==23.2 setuptools==75.8.0 wheel ninja
+pip3 install packaging==26.0 setuptools==75.8.0 wheel ninja
 pip3 install --no-build-isolation -e '.[flash-attn]'
+
+# Install CCE https://docs.axolotl.ai/docs/custom_integrations.html#cut-cross-entropy
+python scripts/cutcrossentropy_install.py | sh
 ```
 
 2. Run the finetuning example:

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

@@ -9,10 +9,6 @@ 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/colab-notebooks/colab-axolotl-example.ipynb

@@ -40,7 +40,7 @@
     "%%capture\n",
     "# This step can take ~5-10 minutes to install dependencies\n",
     "!pip install --no-build-isolation axolotl[flash-attn]>=0.9.1\n",
-    "!pip install \"cut-cross-entropy[transformers] @ git+https://github.com/axolotl-ai-cloud/ml-cross-entropy.git@c6a32c5\""
+    "!pip install \"cut-cross-entropy[transformers] @ git+https://github.com/axolotl-ai-cloud/ml-cross-entropy.git@fec1a88\""
    ]
   },
   {
@@ -253,7 +253,6 @@
    "source": [
     "from axolotl.utils import set_pytorch_cuda_alloc_conf\n",
     "\n",
-    "# Set \"PYTORCH_CUDA_ALLOC_CONF\" env to save memory\n",
     "set_pytorch_cuda_alloc_conf()"
    ]
   },

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

@@ -9,10 +9,6 @@ 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

@@ -9,10 +9,6 @@ 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

@@ -16,7 +16,7 @@ Thanks to the team at MistralAI for giving us early access to prepare for this r
 
 ```bash
 # Ensure you have Pytorch installed (Pytorch 2.6.0 min)
-pip3 install packaging==23.2 setuptools==75.8.0 wheel ninja
+pip3 install packaging==26.0 setuptools==75.8.0 wheel ninja
 pip3 install --no-build-isolation 'axolotl[flash-attn]>=0.12.0'
 ```
 

examples/devstral/devstral-small-qlora.yml

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

examples/eaft/eaft-example.yml

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

examples/ebft/README.md

@@ -0,0 +1,211 @@
+# Energy-Based Fine-Tuning (EBFT)
+
+EBFT is an integration of ["Matching Features, Not Tokens: Energy-Based Fine-Tuning of Language Models"](https://arxiv.org/abs/2603.12248) (Jelassi et al., 2026) into axolotl.
+
+## Overview
+
+EBFT fine-tunes language models by optimizing a **feature-matching loss** rather than relying on external reward functions or verifiers. A frozen copy of the model (the "feature network") extracts embeddings from both generated and ground-truth completions, and the generator is updated via REINFORCE to match the ground-truth feature moments.
+
+**Key advantages over SFT:**
+- Operates on model rollouts (not teacher forcing), reducing distribution shift
+- Provides dense sequence-level supervision without a task-specific verifier
+- Improves both downstream accuracy and validation cross-entropy simultaneously
+
+**Key advantages over RLVR:**
+- No reward model or verifier required — works on any (prompt, completion) data
+- Applicable to non-verifiable tasks (e.g., raw code, translation, creative writing)
+- Maintains distributional calibration (low feature-matching loss)
+
+## Two Modes
+
+EBFT supports two modes depending on your data format:
+
+### Structured Mode (`mode: structured`, default)
+For **QA/instruction data** with prompt + completion pairs (e.g., OpenCodeInstruct, ALMA translation).
+- Extends GRPOTrainer — uses vLLM for fast rollout generation
+- RLOO advantages and clipped policy gradient from GRPO
+- Feature-matching rewards replace external reward functions
+
+### Strided Mode (`mode: strided`)
+For **unstructured text** without prompt/completion splits (e.g., raw code, prose, SwallowCode).
+- Uses **strided block-parallel generation** — multiple short rollouts at different anchor points within a document
+- No vLLM needed — generation uses custom strided attention masks
+- Uses **torch flex_attention** with compiled block masks for efficient fused attention kernels (~2x faster than eager attention)
+- Compatible with gradient checkpointing via automatic dtype normalization
+- This is the core EBFT algorithm from the paper (Section F)
+
+### Common to both modes:
+- **Frozen feature network** — deep copy of the model at initialization (frozen, eval mode)
+- **Feature extraction** — hidden states at configurable layer depths (default: 25%, 50%, 75%), L2-normalized per layer before concatenation
+- **Feature-matching rewards** — cosine similarity (alignment) minus pairwise dot-product (diversity), scaled by 2 per paper equation (7)
+- **SVD whitening** — decorrelates feature dimensions; the paper shows removing it causes the largest degradation
+- **CFM loss tracking** — conditional feature-matching loss (paper eq 2) logged as `ebft/cfm_loss`
+- **FSDP2 compatible** — feature network stays outside FSDP wrapping (frozen, inference-only)
+
+## Quick Start
+
+### Structured Mode (QA data + vLLM)
+
+```bash
+# 1. Start vLLM server (LoRA serve module auto-selected when vllm_lora_sync: true)
+CUDA_VISIBLE_DEVICES=0 axolotl vllm-serve examples/ebft/qwen3-4b-ebft-structured-async.yaml
+
+# 2. Train on a separate GPU
+CUDA_VISIBLE_DEVICES=1 axolotl train examples/ebft/qwen3-4b-ebft-structured-async.yaml
+```
+
+### Strided Mode (unstructured text)
+
+```bash
+# No vLLM needed — strided generation is built-in
+axolotl train examples/ebft/llama-3b-ebft-strided-fft.yaml
+```
+
+## Configuration
+
+### Common EBFT Settings
+
+```yaml
+rl: ebft
+
+ebft:
+  # Feature network: which layers to extract hidden states from
+  # Values are fractions of total depth (0.0 = embedding, 1.0 = final layer)
+  feature_layers: [0.25, 0.5, 0.75]
+
+  # How to pool per-token hidden states into sequence embeddings
+  # Options: "last_token" (recommended), "mean_pooling", "concat"
+  embed_method: last_token
+
+  # SVD whitening — strongly recommended (paper shows largest degradation without it)
+  use_whitening: true
+
+  # Reward = alignment_coef * alignment - diversity_coef * diversity
+  # Per paper Variant (i) (eq 49): alignment uses cosine similarity (normalized),
+  # diversity uses raw dot product — both are bounded after whitening.
+  alignment_coef: 1.0
+  diversity_coef: 1.0
+
+  # Cross-entropy loss on ground-truth tokens (mixed objective, paper Section 2.1)
+  # 0.0 = pure feature matching; 0.03 = recommended balance; 0.1 = CE-dominated
+  ce_coef: 0.0
+```
+
+### Strided Mode Settings
+
+```yaml
+ebft:
+  mode: strided
+  stride: 8              # tokens between anchor points (paper default: 8)
+  context_length: 8      # context window per block (paper default: 8)
+  generate_max_len: 8    # tokens generated per block (paper default: 8)
+  n_samples_per_prompt: 4  # independent rollouts per document (>= 2 for RLOO)
+  temperature: 0.6
+  rl_coef: 1.0           # RL loss weight
+  advantage_estimator: rloo  # rloo (recommended), group_norm, or reinforce
+```
+
+### Structured Mode Settings (via TRL)
+
+```yaml
+trl:
+  num_generations: 4           # samples per prompt
+  max_completion_length: 256   # max tokens to generate
+  temperature: 1.0
+  use_vllm: true
+  scale_rewards: true
+  loss_type: grpo
+  epsilon: 0.2
+```
+
+### Dataset Format
+
+**Structured mode** — QA data with prompt + ground-truth completion:
+```yaml
+datasets:
+  - path: nvidia/OpenCodeInstruct
+    type: ebft_opencode.transform
+```
+Transform returns: `{"prompt": ..., "ground_truth": ...}`
+
+**Strided mode** — raw text tokenized to fixed length:
+```yaml
+datasets:
+  - path: sjelassi/swallow_code_20m
+    type: ebft_pretrain.transform
+```
+Transform returns: `{"input_ids": ..., "attention_mask": ..., "labels": ...}`
+
+## How It Works
+
+### Structured Mode
+1. **Generate**: For each prompt, generate `num_generations` completions via vLLM
+2. **Extract features**: Forward both generated and ground-truth sequences through the frozen feature network
+3. **Compute rewards**: `2 * alignment - 2 * diversity` (paper eq 7)
+4. **RLOO advantages**: subtract leave-one-out group mean
+5. **Policy gradient**: clipped PPO-style loss
+
+### Strided Mode
+1. **Anchor selection**: Pick `num_blocks = (seq_len - gen_len - ctx_len) / stride + 1` anchor points across the document
+2. **Block-parallel generation**: At each anchor, generate `gen_len` tokens using a custom strided attention mask via `flex_attention` compiled block masks
+3. **Feature extraction**: Forward the full sequence (prompt + generated) through the frozen feature network **with the strided attention mask** — this is critical for correct feature representations
+4. **Per-block rewards**:
+   - **Alignment** = `2 * cosine_similarity(gen_block_emb, gt_block_emb)` — normalized, bounded in [-2, 2]
+   - **Diversity** = `2 * mean_pairwise_dot_product(gen_block_embs)` — raw dot product on whitened vectors
+   - **Reward** = `alignment_coef * alignment - diversity_coef * diversity`
+5. **RLOO advantages**: leave-one-out baseline across `n_samples_per_prompt` rollouts per block
+6. **Policy gradient**: REINFORCE loss on generated tokens, weighted by per-block advantages
+
+### Tracked Metrics
+
+| Metric | Description |
+|--------|-------------|
+| `ebft/alignment` | Mean cosine similarity between generated and GT features (higher = better) |
+| `ebft/diversity` | Mean pairwise similarity between samples (lower = more diverse) |
+| `ebft/mean_reward` | alignment - diversity (should trend upward) |
+| `ebft/cfm_loss` | Conditional feature-matching loss ‖E[φ(ŷ)] - φ(y)‖² (paper eq 2, lower = better) |
+| `ebft/rl_loss` | REINFORCE policy gradient loss |
+| `ebft/ce_loss` | Cross-entropy loss on ground-truth tokens (when `ce_coef > 0`) |
+| `ebft/advantages_std` | RLOO advantage standard deviation (should be non-zero) |
+
+## Tips and Recommendations
+
+### Reward coefficients
+- **`use_whitening: true`**: Strongly recommended. The paper's ablation (Figure 7) shows removing whitening causes the largest performance degradation. Safe to use with `diversity_coef > 0`.
+- **`diversity_coef`**: Default 1.0. Per the paper's Variant (i) (eq 49), alignment uses cosine similarity while diversity uses raw dot product. After whitening, both are bounded and on compatible scales.
+- **`n_samples_per_prompt`**: Must be >= 2 for diversity and RLOO. 4 is the paper's default.
+- **`ce_coef`**: The paper ablates `γ ∈ {0, 0.03, 0.1}`. `0.03` balances CE and RL signals; `0.1` causes CE to dominate the gradient. `0.0` gives pure feature matching.
+
+### Feature extraction
+- **`feature_layers: [0.25, 0.5, 0.75]`**: Extracts and concatenates hidden states from 25%, 50%, 75% depth. Each layer is L2-normalized independently before concatenation. The paper shows this works better than mean pooling or single-layer extraction.
+- **`embed_method: last_token`**: Uses the last token's hidden state per block. The paper shows this outperforms mean pooling (Figure 7).
+
+### Performance
+- **`torch_compile: true`**: Recommended for strided mode. Provides additional speedup via graph compilation.
+- **flex_attention**: Strided mode automatically uses `flex_attention` with compiled block masks when available (~2x faster than eager attention). Works with gradient checkpointing via automatic dtype normalization. Falls back to eager attention with dense 4D masks if flex_attention is unavailable.
+
+### Memory
+- EBFT requires a frozen copy of the model (the feature network), roughly doubling model memory.
+- **LoRA** is recommended to reduce trainable parameter memory. The feature network is always a frozen copy of the base model (without LoRA adapters).
+- With 2 GPUs visible, the trainer automatically places the feature network on the second GPU.
+- **FSDP2** is supported — the feature network stays outside FSDP wrapping since it's frozen and inference-only. With `cpu_ram_efficient_loading`, the feature network is loaded separately from pretrained weights.
+
+## Example Configs
+
+| Config | Mode | Model | Description |
+|--------|------|-------|-------------|
+| `llama-1b-ebft-opencode.yaml` | Structured | Llama-3.2-1B | QA coding with vLLM |
+| `llama-1b-ebft-opencode-novllm.yaml` | Structured | Llama-3.2-1B | QA coding without vLLM |
+| `llama-3b-ebft-strided-fft.yaml` | Strided | Llama-3.2-3B | Unstructured code with LoRA |
+| `llama-1b-ebft-strided.yaml` | Strided | Llama-3.2-1B | Quick validation |
+
+## Citation
+
+```bibtex
+@article{jelassi2026matching,
+  title={Matching Features, Not Tokens: Energy-Based Fine-Tuning of Language Models},
+  author={Jelassi, Samy and Kwun, Mujin and Zhao, Rosie and Li, Yuanzhi and Fusi, Nicolo and Du, Yilun and Kakade, Sham M. and Domingo-Enrich, Carles},
+  journal={arXiv preprint arXiv:2603.12248},
+  year={2026}
+}
+```

examples/ebft/ebft_opencode.py

@@ -0,0 +1,28 @@
+"""
+Dataset transform for nvidia/OpenCodeInstruct with EBFT.
+
+Maps the dataset's `input` (prompt) and `output` (code solution) fields
+to the format expected by the EBFT trainer.
+"""
+
+
+def transform(cfg, *args, **kwargs):
+    def transform_fn(example, tokenizer=None):
+        return {
+            "prompt": [
+                {"role": "user", "content": example["input"]},
+            ],
+            "ground_truth": example["output"],
+        }
+
+    return transform_fn, {
+        "remove_columns": [
+            "id",
+            "domain",
+            "generation_algorithm",
+            "llm_judgement",
+            "unit_tests",
+            "tests_execution_status",
+            "average_test_score",
+        ]
+    }

examples/ebft/ebft_pretrain.py

@@ -0,0 +1,31 @@
+"""
+Dataset transform for unstructured text data with strided EBFT.
+
+Tokenizes raw text into fixed-length input_ids for the strided trainer.
+Sequences are padded to sequence_len for uniform batching.
+"""
+
+
+def transform(cfg, *args, **kwargs):
+    seq_len = cfg.sequence_len
+
+    def transform_fn(example, tokenizer=None):
+        text = example.get("question", example.get("text", ""))
+        if tokenizer is None:
+            return {"prompt": text}
+
+        encoded = tokenizer(
+            text,
+            truncation=True,
+            max_length=seq_len,
+            padding="max_length",
+            add_special_tokens=True,
+            return_tensors=None,
+        )
+        return {
+            "input_ids": encoded["input_ids"],
+            "attention_mask": encoded["attention_mask"],
+            "labels": list(encoded["input_ids"]),
+        }
+
+    return transform_fn, {"remove_columns": ["question", "answer"]}

examples/ebft/ebft_strided_structured.py

@@ -0,0 +1,80 @@
+"""
+Dataset transform for structured (prompt, completion) data with strided EBFT.
+
+Tokenizes prompt and completion separately, concatenates into a single
+input_ids sequence, and marks prompt tokens with labels=-100 so the
+strided trainer knows where to place anchors (completion span only).
+
+Works with datasets that have chat-style fields (e.g., nvidia/OpenCodeInstruct).
+"""
+
+
+def transform(cfg, *args, **kwargs):
+    seq_len = cfg.sequence_len
+
+    def transform_fn(example, tokenizer=None):
+        # Extract prompt and completion from the example
+        prompt_text = example.get(
+            "input", example.get("prompt", example.get("question", ""))
+        )
+        completion_text = example.get(
+            "output", example.get("completion", example.get("answer", ""))
+        )
+
+        if tokenizer is None:
+            return {"prompt": prompt_text}
+
+        pad_id = tokenizer.pad_token_id or tokenizer.eos_token_id
+
+        # Tokenize prompt and completion separately
+        prompt_enc = tokenizer(
+            prompt_text,
+            truncation=False,
+            add_special_tokens=True,
+            return_tensors=None,
+        )
+        completion_enc = tokenizer(
+            completion_text,
+            truncation=False,
+            add_special_tokens=False,
+            return_tensors=None,
+        )
+
+        prompt_ids = prompt_enc["input_ids"]
+        completion_ids = completion_enc["input_ids"]
+
+        # Truncate to fit within seq_len (prioritize keeping prompt + some completion)
+        total_len = len(prompt_ids) + len(completion_ids)
+        if total_len > seq_len:
+            # Truncate completion first, then prompt if needed
+            max_completion = seq_len - len(prompt_ids)
+            if max_completion < 1:
+                # Prompt alone exceeds seq_len — truncate prompt, keep at least 1 completion token
+                prompt_ids = prompt_ids[: seq_len - 1]
+                completion_ids = completion_ids[:1]
+            else:
+                completion_ids = completion_ids[:max_completion]
+
+        input_ids = prompt_ids + completion_ids
+        prompt_length = len(prompt_ids)
+
+        # Labels: -100 for prompt tokens, input_ids for completion tokens
+        labels = [-100] * prompt_length + completion_ids
+
+        # Pad to seq_len
+        pad_len = seq_len - len(input_ids)
+        attention_mask = [1] * len(input_ids) + [0] * pad_len
+        labels = labels + [-100] * pad_len
+        input_ids = input_ids + [pad_id] * pad_len
+
+        return {
+            "input_ids": input_ids,
+            "attention_mask": attention_mask,
+            "labels": labels,
+            "prompt_length": prompt_length,
+        }
+
+    # Signal to remove all original columns (filtered to existing ones at map time)
+    return transform_fn, {
+        "remove_columns": "__all__",
+    }

examples/ebft/llama-1b-ebft-opencode-novllm.yaml

@@ -0,0 +1,64 @@
+# EBFT validation config — no vLLM, uses HF generate for simplicity
+# Run: CUDA_VISIBLE_DEVICES=0 axolotl train examples/ebft/llama-1b-ebft-opencode-novllm.yaml
+
+base_model: meta-llama/Llama-3.2-1B
+chat_template: llama3
+rl: ebft
+
+ebft:
+  feature_layers: [0.25, 0.5, 0.75]
+  embed_method: last_token
+  use_whitening: false
+  alignment_coef: 1.0
+  diversity_coef: 1.0
+  ce_coef: 0.0
+
+trl:
+  num_generations: 4
+  max_completion_length: 128
+  temperature: 1.0
+  use_vllm: false
+  scale_rewards: true
+  loss_type: grpo
+  epsilon: 0.2
+
+datasets:
+  - path: nvidia/OpenCodeInstruct
+    type: ebft_opencode.transform
+    split: train[:1%]
+
+sequence_len: 512
+micro_batch_size: 2
+gradient_accumulation_steps: 2
+num_epochs: 1
+max_steps: 10
+
+learning_rate: 1.0e-5
+optimizer: adamw_torch_fused
+lr_scheduler: cosine
+warmup_steps: 2
+weight_decay: 0.01
+
+adapter: lora
+lora_r: 16
+lora_alpha: 32
+lora_dropout: 0.05
+lora_target_linear: true
+
+bf16: auto
+flash_attention: true
+gradient_checkpointing: true
+
+special_tokens:
+  pad_token: "<|end_of_text|>"
+
+val_set_size: 0.0
+output_dir: ./outputs/ebft-validation
+
+wandb_project: ebft
+wandb_run_id:
+wandb_watch:
+wandb_log_model:
+
+logging_steps: 1
+save_steps: 100

examples/ebft/llama-1b-ebft-opencode.yaml

@@ -0,0 +1,81 @@
+# EBFT: Energy-Based Fine-Tuning with Llama-3.2-1B on OpenCodeInstruct
+#
+# Paper: "Matching Features, Not Tokens" (Jelassi et al., 2026)
+# https://arxiv.org/abs/2603.12248
+#
+# Prerequisites:
+#   1. Start vLLM server on a separate GPU:
+#      CUDA_VISIBLE_DEVICES=1 python -m trl.scripts.vllm_serve \
+#          --model meta-llama/Llama-3.2-1B \
+#          --host 0.0.0.0 --port 8000 \
+#          --gpu-memory-utilization 0.4 --dtype bfloat16
+#
+#   2. Run training:
+#      CUDA_VISIBLE_DEVICES=0 axolotl train examples/ebft/llama-1b-ebft-opencode.yaml
+
+base_model: meta-llama/Llama-3.2-1B
+chat_template: llama3
+
+# --- Training method ---
+rl: ebft
+
+# --- EBFT configuration ---
+ebft:
+  feature_layers: [0.25, 0.5, 0.75]  # extract hidden states at 25%, 50%, 75% depth
+  embed_method: last_token            # pool to sequence embedding via last token
+  use_whitening: false                # SVD whitening (disable for speed in small runs)
+  alignment_coef: 1.0                 # cosine similarity with ground-truth features
+  diversity_coef: 1.0                 # pairwise similarity penalty
+  ce_coef: 0.0                        # cross-entropy on ground-truth (0 = pure feature matching)
+
+# --- Generation settings (via TRL/GRPO infrastructure) ---
+trl:
+  num_generations: 4                  # samples per prompt for RLOO
+  max_completion_length: 256          # max generated tokens
+  temperature: 1.0
+  use_vllm: true
+  scale_rewards: true
+  loss_type: grpo
+  epsilon: 0.2
+
+# --- Dataset ---
+datasets:
+  - path: nvidia/OpenCodeInstruct
+    type: ebft_opencode.transform
+    split: train[:1%]        # first 1% for validation runs
+
+# --- Training hyperparameters ---
+sequence_len: 1024
+micro_batch_size: 2
+gradient_accumulation_steps: 4
+num_epochs: 1
+max_steps: 50
+
+learning_rate: 1.0e-5
+optimizer: adamw_torch_fused
+lr_scheduler: cosine
+warmup_steps: 5
+weight_decay: 0.01
+
+# --- LoRA (recommended to reduce memory with frozen feature network) ---
+adapter: lora
+lora_r: 16
+lora_alpha: 32
+lora_dropout: 0.05
+lora_target_linear: true
+
+# --- Hardware ---
+bf16: auto
+flash_attention: true
+gradient_checkpointing: true
+
+special_tokens:
+  pad_token: "<|end_of_text|>"
+
+val_set_size: 0.0
+output_dir: ./outputs/ebft-llama-1b-opencode
+
+# --- Logging ---
+use_tensorboard: true
+logging_steps: 1
+save_steps: 25