avoid crash/oom on train end

* offload activations to disk instead of CPU RAM

* add prefetch

* Disco :dance:

* include offload_disk in e2e test for AC

* document and make sure to cleanup

* fix annotation to match docs

* fix docs build

* address PR feedback

.github/workflows/base.yml

@@ -22,12 +22,6 @@ jobs:
       fail-fast: false
       matrix:
         include:
-          - cuda: "124"
-            cuda_version: 12.4.1
-            cudnn_version: ""
-            python_version: "3.11"
-            pytorch: 2.4.1
-            torch_cuda_arch_list: "7.0 7.5 8.0 8.6 8.7 8.9 9.0+PTX"
           - cuda: "124"
             cuda_version: 12.4.1
             cudnn_version: ""

.github/workflows/main.yml

@@ -15,16 +15,11 @@ jobs:
       fail-fast: false
       matrix:
         include:
-          - cuda: 124
-            cuda_version: 12.4.1
-            python_version: "3.11"
-            pytorch: 2.4.1
-            axolotl_extras:
           - cuda: 124
             cuda_version: 12.4.1
             python_version: "3.11"
             pytorch: 2.5.1
-            axolotl_extras: vllm
+            axolotl_extras:
           - cuda: 124
             cuda_version: 12.4.1
             python_version: "3.11"
@@ -35,7 +30,7 @@ jobs:
             cuda_version: 12.6.3
             python_version: "3.11"
             pytorch: 2.7.0
-            axolotl_extras: vllm
+            axolotl_extras:
     runs-on: axolotl-gpu-runner
     steps:
       - name: Checkout
@@ -67,6 +62,7 @@ jobs:
             CUDA=${{ matrix.cuda }}
             PYTORCH_VERSION=${{ matrix.pytorch }}
             AXOLOTL_ARGS=${{ matrix.axolotl_args }}
+            AXOLOTL_EXTRAS=${{ matrix.axolotl_extras}}
           file: ./docker/Dockerfile
           push: ${{ github.event_name != 'pull_request' }}
           tags: |
@@ -82,11 +78,6 @@ jobs:
     strategy:
       matrix:
         include:
-          - cuda: 124
-            cuda_version: 12.4.1
-            python_version: "3.11"
-            pytorch: 2.4.1
-            axolotl_extras:
           - cuda: 124
             cuda_version: 12.4.1
             python_version: "3.11"

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

@@ -3,12 +3,13 @@ name: docker-multigpu-tests-biweekly
 on:
   pull_request:
     paths:
-      - 'tests/e2e/multigpu/*.py'
+      - 'tests/e2e/multigpu/**.py'
       - 'requirements.txt'
       - 'setup.py'
       - 'pyproject.toml'
       - '.github/workflows/multi-gpu-e2e.yml'
       - 'src/axolotl/core/trainers/mixins/sequence_parallel.py'
+      - 'src/axolotl/utils/distributed.py'
   workflow_dispatch:
   schedule:
     - cron: '0 0 * * 1,4'  # Runs at 00:00 UTC every monday & thursday
@@ -32,18 +33,11 @@ jobs:
             axolotl_extras: vllm
             num_gpus: 2
             nightly_build: "true"
-          - cuda: 124
-            cuda_version: 12.4.1
-            python_version: "3.11"
-            pytorch: 2.4.1
-            axolotl_extras:  # no vllm support for 2.4.1
-            num_gpus: 2
-            nightly_build: "true"
           - cuda: 124
             cuda_version: 12.4.1
             python_version: "3.11"
             pytorch: 2.5.1
-            axolotl_extras: vllm
+            axolotl_extras:
             num_gpus: 2
             nightly_build: "true"
           - cuda: 126

.github/workflows/nightlies.yml

@@ -12,11 +12,6 @@ jobs:
       fail-fast: false
       matrix:
         include:
-          - cuda: 124
-            cuda_version: 12.4.1
-            python_version: "3.11"
-            pytorch: 2.4.1
-            axolotl_extras:
           - cuda: 124
             cuda_version: 12.4.1
             python_version: "3.11"
@@ -70,11 +65,6 @@ jobs:
     strategy:
       matrix:
         include:
-          - cuda: 124
-            cuda_version: 12.4.1
-            python_version: "3.11"
-            pytorch: 2.4.1
-            axolotl_extras:
           - cuda: 124
             cuda_version: 12.4.1
             python_version: "3.11"

.github/workflows/preview-docs.yml

@@ -0,0 +1,61 @@
+name: Preview
+on:
+  workflow_dispatch:
+  pull_request:
+    types: [opened, synchronize, reopened]
+
+    # Run the workflow only when one of these files changes
+    paths:
+      - '**/*.md'      # any Markdown file
+      - '**/*.qmd'     # any Quarto file
+      - '_quarto.yaml'
+
+permissions:
+  checks: write
+  contents: write
+  deployments: write
+  issues: write
+  discussions: write
+  pages: write
+  pull-requests: write
+  statuses: write
+
+jobs:
+  preview:
+    runs-on: ubuntu-latest
+    steps:
+      - name: Check out repository
+        uses: actions/checkout@v4
+
+      - name: Set up Quarto
+        uses: quarto-dev/quarto-actions/setup@v2
+
+      - name: Setup Python
+        uses: actions/setup-python@v5
+        with:
+          python-version: '3.11'
+
+      - name: Install dependencies
+        run: |
+          python3 -m pip install jupyter quartodoc
+          python3 -m pip install -e . --no-deps
+
+      - name: Build autodoc
+        run: quartodoc build
+
+      - name: Quarto render
+        run: quarto render
+
+      - name: Netlify Publish
+        uses: nwtgck/actions-netlify@v3.0
+        with:
+          publish-dir: './_site'
+          enable-pull-request-comment: true
+          enable-github-deployment: true
+          github-token: ${{ secrets.GITHUB_TOKEN }}
+          deploy-message: "Deployed On Netlify"
+          github-deployment-environment: 'preview'
+          github-deployment-description: 'Preview Deployment'
+        env:
+          NETLIFY_AUTH_TOKEN: ${{ secrets.NETLIFY_AUTH_TOKEN }}
+          NETLIFY_SITE_ID: ${{ secrets.NETLIFY_SITE_ID }}

.github/workflows/tests-nightly.yml

@@ -18,15 +18,102 @@ jobs:
         env:
           SKIP: no-commit-to-branch
 
+  preload-cache:
+    name: Preload HF cache
+    runs-on: ubuntu-latest
+    strategy:
+      fail-fast: false
+      matrix:
+        python_version: ["3.11"]
+        pytorch_version: ["2.6.0"]
+    timeout-minutes: 20
+
+    env:
+      AXOLOTL_IS_CI_CACHE_PRELOAD: "1"
+
+    steps:
+      - name: Check out repository code
+        uses: actions/checkout@v4
+
+      - name: Restore HF cache
+        id: hf-cache-restore
+        uses: actions/cache/restore@v4
+        with:
+          path: |
+            /home/runner/.cache/huggingface/hub/datasets--*
+            /home/runner/.cache/huggingface/hub/models--*
+          key: ${{ runner.os }}-hf-hub-cache-v2
+
+      - name: Setup Python
+        uses: actions/setup-python@v5
+        with:
+          python-version: ${{ matrix.python_version }}
+          cache: 'pip' # caching pip dependencies
+
+      - name: upgrade pip
+        run: |
+          pip3 install --upgrade pip
+          pip3 install --upgrade packaging==23.2 setuptools==75.8.0 wheel
+
+      - name: Install PyTorch
+        run: |
+          pip3 install torch==${{ matrix.pytorch_version }}
+
+      - name: Install dependencies
+        run: |
+          pip3 show torch
+          pip3 install --no-build-isolation -U -e .
+          python scripts/unsloth_install.py | sh
+          python scripts/cutcrossentropy_install.py | sh
+          pip3 install -r requirements-dev.txt -r requirements-tests.txt
+
+      - name: Make sure PyTorch version wasn't clobbered
+        run: |
+          python -c "import torch; assert '${{ matrix.pytorch_version }}' in torch.__version__"
+
+      - name: Ensure axolotl CLI was installed
+        run: |
+          axolotl --help
+
+      - name: Pre-Download dataset fixture
+        run: |
+          huggingface-cli download --repo-type=dataset axolotl-ai-internal/axolotl-oss-dataset-fixtures
+
+      - name: Run tests
+        run: |
+          pytest -v tests/conftest.py
+
+      - name: Upload coverage to Codecov
+        uses: codecov/codecov-action@v5
+        with:
+          token: ${{ secrets.CODECOV_TOKEN }}
+          files: ./coverage.xml
+          flags: unittests,pytorch-${{ matrix.pytorch_version }}
+          fail_ci_if_error: false
+
+      - name: cleanup pip cache
+        run: |
+          find "$(pip cache dir)/http-v2" -type f -mtime +14 -exec rm {} \;
+
+      - name: Save HF cache
+        id: hf-cache
+        uses: actions/cache/save@v4
+        with:
+          path: |
+            /home/runner/.cache/huggingface/hub/datasets--*
+            /home/runner/.cache/huggingface/hub/models--*
+          key: ${{ steps.hf-cache-restore.outputs.cache-primary-key }}
+
   pytest:
     name: PyTest
     runs-on: ubuntu-latest
+    needs: [preload-cache]
     strategy:
       fail-fast: false
       max-parallel: 2
       matrix:
         python_version: ["3.11"]
-        pytorch_version: ["2.4.1", "2.5.1", "2.6.0"]
+        pytorch_version: ["2.5.1", "2.6.0", "2.7.0"]
     timeout-minutes: 20
 
     steps:
@@ -106,13 +193,6 @@ jobs:
       fail-fast: false
       matrix:
         include:
-          - cuda: 124
-            cuda_version: 12.4.1
-            python_version: "3.11"
-            pytorch: 2.4.1
-            num_gpus: 1
-            axolotl_extras:
-            nightly_build: "true"
           - cuda: 124
             cuda_version: 12.4.1
             python_version: "3.11"

.github/workflows/tests.yml

@@ -27,6 +27,9 @@ concurrency:
   group: ${{ github.workflow }}-${{ github.ref }}
   cancel-in-progress: ${{ github.ref != 'refs/heads/main' }}
 
+env:
+  TRANSFORMERS_IS_CI: "yes"
+
 jobs:
   pre-commit:
     name: pre-commit
@@ -41,29 +44,127 @@ jobs:
         env:
           SKIP: no-commit-to-branch
 
+#  preload-cache:
+#    name: Preload HF cache
+#    runs-on: ubuntu-latest
+#    strategy:
+#      fail-fast: false
+#      matrix:
+#        python_version: ["3.11"]
+#        pytorch_version: ["2.6.0"]
+#    timeout-minutes: 20
+#
+#    env:
+#      AXOLOTL_IS_CI_CACHE_PRELOAD: "1"
+#
+#    steps:
+#      - name: Check out repository code
+#        uses: actions/checkout@v4
+#
+#      - name: Restore HF cache
+#        id: hf-cache-restore
+#        uses: actions/cache/restore@v4
+#        with:
+#          path: |
+#            /home/runner/.cache/huggingface/hub/datasets--*
+#            /home/runner/.cache/huggingface/hub/models--*
+#          key: ${{ runner.os }}-hf-hub-cache-v2
+#
+#      - name: Restore Cache from S3
+#        id: hf-cache-restore-s3
+#        run: |
+#          mkdir -p /home/runner/.cache/huggingface/hub
+#          curl -L https://d1dttdx32dkk5p.cloudfront.net/hf-cache.tar.zst | tar -xf - -C /home/runner/.cache/huggingface/hub/  --use-compress-program unzstd
+#
+#      - name: Setup Python
+#        uses: actions/setup-python@v5
+#        with:
+#          python-version: ${{ matrix.python_version }}
+#          cache: 'pip' # caching pip dependencies
+#
+#      - name: upgrade pip
+#        run: |
+#          pip3 install --upgrade pip
+#          pip3 install --upgrade packaging==23.2 setuptools==75.8.0 wheel
+#
+#      - name: Install PyTorch
+#        run: |
+#          pip3 install torch==${{ matrix.pytorch_version }}
+#
+#      - name: Install dependencies
+#        run: |
+#          pip3 show torch
+#          pip3 install --no-build-isolation -U -e .
+#          python scripts/unsloth_install.py | sh
+#          python scripts/cutcrossentropy_install.py | sh
+#          pip3 install -r requirements-dev.txt -r requirements-tests.txt
+#
+#      - name: Make sure PyTorch version wasn't clobbered
+#        run: |
+#          python -c "import torch; assert '${{ matrix.pytorch_version }}' in torch.__version__"
+#
+#      - name: Ensure axolotl CLI was installed
+#        run: |
+#          axolotl --help
+#
+#      - name: Pre-Download dataset fixture
+#        run: |
+#          huggingface-cli download --repo-type=dataset axolotl-ai-internal/axolotl-oss-dataset-fixtures
+#
+#      - name: Run tests
+#        run: |
+#          pytest -v tests/conftest.py
+#
+#      - name: Upload coverage to Codecov
+#        uses: codecov/codecov-action@v5
+#        with:
+#          token: ${{ secrets.CODECOV_TOKEN }}
+#          files: ./coverage.xml
+#          flags: unittests,pytorch-${{ matrix.pytorch_version }}
+#          fail_ci_if_error: false
+#
+#      - name: cleanup pip cache
+#        run: |
+#          find "$(pip cache dir)/http-v2" -type f -mtime +14 -exec rm {} \;
+#
+#      - name: Save HF cache
+#        id: hf-cache
+#        uses: actions/cache/save@v4
+#        with:
+#          path: |
+#            /home/runner/.cache/huggingface/hub/datasets--*
+#            /home/runner/.cache/huggingface/hub/models--*
+#          key: ${{ steps.hf-cache-restore.outputs.cache-primary-key }}
+
   pytest:
     name: PyTest
     runs-on: ubuntu-latest
+#    needs: [preload-cache]
     strategy:
       fail-fast: false
-      max-parallel: 2
       matrix:
         python_version: ["3.11"]
-        pytorch_version: ["2.4.1", "2.5.1", "2.6.0", "2.7.0"]
+        pytorch_version: ["2.5.1", "2.6.0", "2.7.0"]
     timeout-minutes: 20
 
     steps:
       - name: Check out repository code
         uses: actions/checkout@v4
 
-      - name: Restore HF cache
-        id: hf-cache-restore
-        uses: actions/cache/restore@v4
-        with:
-          path: |
-            /home/runner/.cache/huggingface/hub/datasets--*
-            /home/runner/.cache/huggingface/hub/models--*
-          key: ${{ runner.os }}-hf-hub-cache-v2
+#      - name: Restore HF cache
+#        id: hf-cache-restore
+#        uses: actions/cache/restore@v4
+#        with:
+#          path: |
+#            /home/runner/.cache/huggingface/hub/datasets--*
+#            /home/runner/.cache/huggingface/hub/models--*
+#          key: ${{ runner.os }}-hf-hub-cache-v2
+
+      - name: Restore Cache from S3
+        id: hf-cache-restore-s3
+        run: |
+          mkdir -p /home/runner/.cache/huggingface/hub
+          curl -L https://d1dttdx32dkk5p.cloudfront.net/hf-cache.tar.zst | tar -xf - -C /home/runner/.cache/huggingface/hub/  --use-compress-program unzstd
 
       - name: Setup Python
         uses: actions/setup-python@v5
@@ -118,38 +219,35 @@ jobs:
         run: |
           find "$(pip cache dir)/http-v2" -type f -mtime +14 -exec rm {} \;
 
-      - name: Save HF cache
-        id: hf-cache
-        uses: actions/cache/save@v4
-        with:
-          path: |
-            /home/runner/.cache/huggingface/hub/datasets--*
-            /home/runner/.cache/huggingface/hub/models--*
-          key: ${{ steps.hf-cache-restore.outputs.cache-primary-key }}
-
   pytest-sdist:
     name: PyTest from Source Dist
     runs-on: ubuntu-latest
+#    needs: [preload-cache]
     strategy:
       fail-fast: false
-      max-parallel: 1
       matrix:
         python_version: ["3.11"]
-        pytorch_version: ["2.4.1", "2.5.1", "2.6.0"]
+        pytorch_version: ["2.5.1", "2.6.0", "2.7.0"]
     timeout-minutes: 20
 
     steps:
       - name: Check out repository code
         uses: actions/checkout@v4
 
-      - name: Restore HF cache
-        id: hf-cache-restore
-        uses: actions/cache/restore@v4
-        with:
-          path: |
-            /home/runner/.cache/huggingface/hub/datasets--*
-            /home/runner/.cache/huggingface/hub/models--*
-          key: ${{ runner.os }}-hf-hub-cache-v2
+#      - name: Restore HF cache
+#        id: hf-cache-restore
+#        uses: actions/cache/restore@v4
+#        with:
+#          path: |
+#            /home/runner/.cache/huggingface/hub/datasets--*
+#            /home/runner/.cache/huggingface/hub/models--*
+#          key: ${{ runner.os }}-hf-hub-cache-v2
+
+      - name: Restore Cache from S3
+        id: hf-cache-restore-s3
+        run: |
+          mkdir -p /home/runner/.cache/huggingface/hub
+          curl -L https://d1dttdx32dkk5p.cloudfront.net/hf-cache.tar.zst | tar -xf - -C /home/runner/.cache/huggingface/hub/  --use-compress-program unzstd
 
       - name: Setup Python
         uses: actions/setup-python@v5
@@ -196,15 +294,6 @@ jobs:
         run: |
           find "$(pip cache dir)/http-v2" -type f -mtime +14 -exec rm {} \;
 
-      - name: Save HF cache
-        id: hf-cache
-        uses: actions/cache/save@v4
-        with:
-          path: |
-            /home/runner/.cache/huggingface/hub/datasets--*
-            /home/runner/.cache/huggingface/hub/models--*
-          key: ${{ steps.hf-cache-restore.outputs.cache-primary-key }}
-
   docker-e2e-tests-1st:
     if: ${{ ! contains(github.event.commits[0].message, '[skip e2e]') && github.repository_owner == 'axolotl-ai-cloud' }}
     # this job needs to be run on self-hosted GPU runners...
@@ -261,15 +350,15 @@ jobs:
           - cuda: 124
             cuda_version: 12.4.1
             python_version: "3.11"
-            pytorch: 2.4.1
+            pytorch: 2.6.0
             num_gpus: 1
-            axolotl_extras:
+            axolotl_extras: llmcompressor
           - cuda: 124
             cuda_version: 12.4.1
             python_version: "3.11"
             pytorch: 2.5.1
             num_gpus: 1
-            axolotl_extras: vllm
+            axolotl_extras:
           - cuda: 126
             cuda_version: 12.6.3
             python_version: "3.11"
@@ -300,3 +389,43 @@ jobs:
       - name: Run tests job on Modal
         run: |
           modal run cicd.e2e_tests
+
+  docker-e2e-cleanup:
+    runs-on: [self-hosted, modal]
+    timeout-minutes: 90
+    needs: [docker-e2e-tests]
+
+    strategy:
+      fail-fast: false
+      matrix:
+        include:
+          - cuda: 124
+            cuda_version: 12.4.1
+            python_version: "3.11"
+            pytorch: 2.6.0
+            num_gpus: 1
+            axolotl_extras: vllm
+    steps:
+      - name: Checkout
+        uses: actions/checkout@v4
+      - name: Install Python
+        uses: actions/setup-python@v5
+        with:
+          python-version: "3.11"
+      - name: Install Modal
+        run: |
+          python -m pip install --upgrade pip
+          pip install modal==0.71.8 jinja2
+      - name: Update env vars
+        run: |
+          echo "BASE_TAG=main-base-py${{ matrix.python_version }}-cu${{ matrix.cuda }}-${{ matrix.pytorch }}" >> $GITHUB_ENV
+          echo "PYTORCH_VERSION=${{ matrix.pytorch}}" >> $GITHUB_ENV
+          echo "AXOLOTL_ARGS=${{ matrix.axolotl_args}}" >> $GITHUB_ENV
+          echo "AXOLOTL_EXTRAS=${{ matrix.axolotl_extras}}" >> $GITHUB_ENV
+          echo "CUDA=${{ matrix.cuda }}" >> $GITHUB_ENV
+          echo "MODAL_IMAGE_BUILDER_VERSION=2024.10" >> $GITHUB_ENV
+          echo "N_GPUS=${{ matrix.num_gpus }}" >> $GITHUB_ENV
+          echo "CODECOV_TOKEN=${{ secrets.CODECOV_TOKEN }}" >> $GITHUB_ENV
+      - name: Run tests job on Modal
+        run: |
+          modal run cicd.cleanup

.runpod/Dockerfile

@@ -1,11 +1,10 @@
-FROM runpod/pytorch:3.10-2.0.0-117
+FROM axolotlai/axolotl-cloud:main-py3.11-cu124-2.6.0
 
 COPY .runpod/requirements.txt /requirements.txt
 RUN --mount=type=cache,target=/root/.cache/pip \
     python3 -m pip install --upgrade pip && \
     python3 -m pip install --upgrade -r /requirements.txt
 
-
 # Environment settings
 ARG BASE_VOLUME="/runpod-volume"
 ENV BASE_VOLUME=$BASE_VOLUME
@@ -15,4 +14,5 @@ ENV TRANSFORMERS_CACHE="${BASE_VOLUME}/huggingface-cache/hub"
 
 COPY .runpod/src /src
 
+WORKDIR /src
 CMD ["python3", "/src/handler.py"]

.runpod/requirements.txt

@@ -5,11 +5,3 @@
 # git+https://github.com/runpod/runpod-python.git
 # To learn more, see https://pip.pypa.io/en/stable/reference/requirements-file-format/
 runpod~=1.7.0
-huggingface_hub
-typing-extensions
-pydantic
-pydantic-settings
-hf-transfer
-setuptools
-numpy==2.0.0
-axolotl[flash-attn,deepspeed]

.runpod/src/handler.py

@@ -57,8 +57,10 @@ async def handler(job):
     logger.info("Training Complete.")
 
     # Cleanup
-    del os.environ["WANDB_API_KEY"]
-    del os.environ["HF_TOKEN"]
+    if "WANDB_API_KEY" in os.environ:
+        del os.environ["WANDB_API_KEY"]
+    if "HF_TOKEN" in os.environ:
+        del os.environ["HF_TOKEN"]
 
 
 runpod.serverless.start({"handler": handler, "return_aggregate_stream": True})

.runpod/test-input.json

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

.runpod/tests.json

@@ -11,43 +11,43 @@
           "hf_token": ""
         },
         "args": {
-          "base_model": "NousResearch/Meta-Llama-3-8B",
-          "model_type": "LlamaForCausalLM",
+          "base_model": "HuggingFaceTB/SmolLM2-135M",
+          "model_type": "AutoModelForCausalLM",
           "tokenizer_type": "AutoTokenizer",
-          "load_in_8bit": true,
-          "load_in_4bit": false,
+          "load_in_4bit": true,
           "strict": false,
           "datasets": [
             {
               "path": "mhenrichsen/alpaca_2k_test",
-              "type": "alpaca"
+              "type": "alpaca",
+              "split": "train[:10%]"
             }
           ],
-          "val_set_size": 0.05,
+          "val_set_size": 0.02,
           "output_dir": "./outputs/lora-out",
           "sequence_len": 4096,
           "sample_packing": true,
           "eval_sample_packing": false,
           "pad_to_sequence_len": true,
-          "adapter": "lora",
+          "adapter": "qlora",
           "lora_r": 32,
-          "lora_alpha": 16,
+          "lora_alpha": 64,
           "lora_dropout": 0.05,
           "lora_target_linear": true,
           "lora_modules_to_save": [
             "embed_tokens",
             "lm_head"
           ],
-          "gradient_accumulation_steps": 4,
-          "micro_batch_size": 2,
+          "gradient_accumulation_steps": 2,
+          "micro_batch_size": 1,
           "num_epochs": 1,
-          "optimizer": "adamw_bnb_8bit",
+          "optimizer": "adamw_torch_fused",
           "lr_scheduler": "cosine",
           "learning_rate": 0.0002,
           "train_on_inputs": false,
           "group_by_length": false,
           "bf16": "auto",
-          "tf32": false,
+          "tf32": true,
           "gradient_checkpointing": true,
           "logging_steps": 1,
           "flash_attention": true,
@@ -57,8 +57,9 @@
           "saves_per_epoch": 1,
           "weight_decay": 0.0,
           "special_tokens": {
-            "pad_token": "<|end_of_text|>"
-          }
+            "pad_token": "<|endoftext|>"
+          },
+          "max_steps": 20
         }
       },
       "timeout": 100000

_quarto.yml

@@ -48,8 +48,23 @@ quartodoc:
       contents:
         - core.trainers.base
         - core.trainers.trl
+        - core.trainers.mamba
+        - core.trainers.relora
         - core.trainers.dpo.trainer
         - core.trainers.grpo.trainer
+        - core.trainers.grpo.sampler
+        - core.trainers.utils
+    - title: Mixins
+      desc: Mixin classes for augmenting trainers
+      contents:
+        - core.trainers.mixins.optimizer
+        - core.trainers.mixins.rng_state_loader
+        - core.trainers.mixins.scheduler
+        - core.trainers.mixins.sequence_parallel
+    - title: Context Managers
+      desc: Context managers for altering trainer behaviors
+      contents:
+        - utils.ctx_managers.sequence_parallel
     - title: Prompt Strategies
       desc: Prompt formatting strategies
       contents:
@@ -86,7 +101,7 @@ quartodoc:
         - kernels.swiglu
         - kernels.quantize
         - kernels.utils
-    - title: MonkeyPatches
+    - title: Monkey Patches
       desc: Runtime patches for model optimizations
       contents:
         - monkeypatch.llama_attn_hijack_flash
@@ -124,7 +139,8 @@ quartodoc:
         - utils.optimizers.adopt
         - utils.data.pretraining
         - utils.data.sft
-        - utils.gradient_checkpointing.unsloth
+        - utils.gradient_checkpointing.offload_cpu
+        - utils.gradient_checkpointing.offload_disk
     - title: Schemas
       desc: Pydantic data models for Axolotl config
       contents:

cicd/cicd.sh

@@ -18,7 +18,7 @@ pytest -v --durations=10 \
   --cov-append
 
 # Run patched tests excluding lora kernels with coverage append
-pytest -v --durations=10 \
+pytest --full-trace -vvv --durations=10 \
   --ignore=tests/e2e/patched/lora_kernels \
   /workspace/axolotl/tests/e2e/patched \
   --cov=axolotl \

cicd/cleanup.py

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

cicd/cleanup.sh

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

cicd/e2e_tests.py

@@ -1,75 +1,12 @@
 """Modal app to run axolotl GPU tests"""
 
-# pylint: disable=duplicate-code
-
-import os
-import pathlib
-import tempfile
-
-import jinja2
-import modal
-from jinja2 import select_autoescape
-from modal import App, Image
-
-cicd_path = pathlib.Path(__file__).parent.resolve()
-
-template_loader = jinja2.FileSystemLoader(searchpath=cicd_path)
-template_env = jinja2.Environment(
-    loader=template_loader, autoescape=select_autoescape()
-)
-df_template = template_env.get_template("Dockerfile.jinja")
-
-df_args = {
-    "AXOLOTL_EXTRAS": os.environ.get("AXOLOTL_EXTRAS", ""),
-    "AXOLOTL_ARGS": os.environ.get("AXOLOTL_ARGS", ""),
-    "PYTORCH_VERSION": os.environ.get("PYTORCH_VERSION", "2.4.1"),
-    "BASE_TAG": os.environ.get("BASE_TAG", "main-base-py3.11-cu121-2.4.1"),
-    "CUDA": os.environ.get("CUDA", "121"),
-    "GITHUB_REF": os.environ.get("GITHUB_REF", "refs/heads/main"),
-    "GITHUB_SHA": os.environ.get("GITHUB_SHA", ""),
-    "NIGHTLY_BUILD": os.environ.get("NIGHTLY_BUILD", ""),
-    "CODECOV_TOKEN": os.environ.get("CODECOV_TOKEN", ""),
-    "HF_HOME": "/workspace/data/huggingface-cache/hub",
-}
-
-dockerfile_contents = df_template.render(**df_args)
-
-temp_dir = tempfile.mkdtemp()
-with open(pathlib.Path(temp_dir) / "Dockerfile", "w", encoding="utf-8") as f:
-    f.write(dockerfile_contents)
-
-cicd_image = Image.from_dockerfile(
-    pathlib.Path(temp_dir) / "Dockerfile",
-    context_mount=None,
-    force_build=True,
-    gpu="A10G",
-).env(df_args)
-
-app = App("Axolotl CI/CD", secrets=[])
-
-hf_cache_volume = modal.Volume.from_name(
-    "axolotl-ci-hf-hub-cache", create_if_missing=True
-)
-VOLUME_CONFIG = {
-    "/workspace/data/huggingface-cache/hub": hf_cache_volume,
-}
-
-N_GPUS = int(os.environ.get("N_GPUS", 1))
-GPU_CONFIG = modal.gpu.L40S(count=N_GPUS)
-
-
-def run_cmd(cmd: str, run_folder: str):
-    import subprocess  # nosec
-
-    # Propagate errors from subprocess.
-    if exit_code := subprocess.call(cmd.split(), cwd=run_folder):  # nosec
-        exit(exit_code)  # pylint: disable=consider-using-sys-exit
+from .single_gpu import GPU_CONFIG, VOLUME_CONFIG, app, cicd_image, run_cmd
 
 
 @app.function(
     image=cicd_image,
     gpu=GPU_CONFIG,
-    timeout=60 * 60,
+    timeout=90 * 60,  # 90 min
     cpu=8.0,
     memory=131072,
     volumes=VOLUME_CONFIG,

cicd/multigpu.sh

@@ -20,4 +20,4 @@ pytest -v  --durations=10 -n1 /workspace/axolotl/tests/e2e/multigpu/patched/ \
   --cov-report=xml:multigpu-coverage.xml
 
 # Upload coverage to Codecov
-codecov upload-process -t $CODECOV_TOKEN -f multigpu-coverage.xml -F multigpu,docker-tests,pytorch-${PYTORCH_VERSION}
+codecov upload-process -t "${CODECOV_TOKEN}" -f multigpu-coverage.xml -F multigpu,docker-tests,pytorch-${PYTORCH_VERSION} || true

cicd/single_gpu.py

@@ -0,0 +1,66 @@
+"""Modal app to run axolotl GPU tests"""
+
+# pylint: disable=duplicate-code
+
+import os
+import pathlib
+import tempfile
+
+import jinja2
+import modal
+from jinja2 import select_autoescape
+from modal import App, Image
+
+cicd_path = pathlib.Path(__file__).parent.resolve()
+
+template_loader = jinja2.FileSystemLoader(searchpath=cicd_path)
+template_env = jinja2.Environment(
+    loader=template_loader, autoescape=select_autoescape()
+)
+df_template = template_env.get_template("Dockerfile.jinja")
+
+df_args = {
+    "AXOLOTL_EXTRAS": os.environ.get("AXOLOTL_EXTRAS", ""),
+    "AXOLOTL_ARGS": os.environ.get("AXOLOTL_ARGS", ""),
+    "PYTORCH_VERSION": os.environ.get("PYTORCH_VERSION", "2.4.1"),
+    "BASE_TAG": os.environ.get("BASE_TAG", "main-base-py3.11-cu121-2.4.1"),
+    "CUDA": os.environ.get("CUDA", "121"),
+    "GITHUB_REF": os.environ.get("GITHUB_REF", "refs/heads/main"),
+    "GITHUB_SHA": os.environ.get("GITHUB_SHA", ""),
+    "NIGHTLY_BUILD": os.environ.get("NIGHTLY_BUILD", ""),
+    "CODECOV_TOKEN": os.environ.get("CODECOV_TOKEN", ""),
+    "HF_HOME": "/workspace/data/huggingface-cache/hub",
+}
+
+dockerfile_contents = df_template.render(**df_args)
+
+temp_dir = tempfile.mkdtemp()
+with open(pathlib.Path(temp_dir) / "Dockerfile", "w", encoding="utf-8") as f:
+    f.write(dockerfile_contents)
+
+cicd_image = Image.from_dockerfile(
+    pathlib.Path(temp_dir) / "Dockerfile",
+    context_mount=None,
+    force_build=True,
+    gpu="A10G",
+).env(df_args)
+
+app = App("Axolotl CI/CD", secrets=[])
+
+hf_cache_volume = modal.Volume.from_name(
+    "axolotl-ci-hf-hub-cache", create_if_missing=True
+)
+VOLUME_CONFIG = {
+    "/workspace/data/huggingface-cache/hub": hf_cache_volume,
+}
+
+N_GPUS = int(os.environ.get("N_GPUS", 1))
+GPU_CONFIG = modal.gpu.L40S(count=N_GPUS)
+
+
+def run_cmd(cmd: str, run_folder: str):
+    import subprocess  # nosec
+
+    # Propagate errors from subprocess.
+    if exit_code := subprocess.call(cmd.split(), cwd=run_folder):  # nosec
+        exit(exit_code)  # pylint: disable=consider-using-sys-exit

codecov.yml

@@ -19,7 +19,7 @@ coverage:
         if_no_uploads: error
         if_not_found: success
         if_ci_failed: error
-        only_pulls: false
+        only_pulls: true
         flags: null
         paths: null
     patch:

docs/config.qmd

@@ -32,6 +32,8 @@ tokenizer_legacy:
 resize_token_embeddings_to_32x:
 # Optional[bool] Whether to shrink the embeddings to len(tokenizer). By default, we won't shrink.
 shrink_embeddings:
+# Optional[bool] Don't upcast the embeddings to float32 when using PEFT. Useful for low-VRAM GPUs
+embeddings_skip_upcast:
 # Whether to load the model with randomly initialized weights. Useful for
 # pre-training a model from scratch or debugging purposes.
 random_init_weights:
@@ -73,11 +75,12 @@ load_in_8bit: true
 load_in_4bit:
 
 # Use CUDA bf16
-bf16: true # bool or 'full' for `bf16_full_eval`. require >=ampere
+bf16: true # bool or 'full' for `bf16_full_eval`, or 'auto' for automatic detection. require >=ampere
 # Use CUDA fp16
 fp16: true
 # Use CUDA tf32
 tf32: true # require >=ampere
+# Note: if bf16 is set to 'auto', and fp16 is set to true, we will prefer the explict fp16 setting
 
 # No AMP (automatic mixed precision)
 bfloat16: true # require >=ampere
@@ -154,6 +157,10 @@ datasets:
     # Key containing the messages (default: "messages")
     field_messages: messages
 
+    # Key containing the system message (default: "system")
+    # If the system message is not present in the dataset sample, it will be loaded from the field_system property.
+    field_system: system
+
     # Mapping of properties from the input dataset to the chat template.
     # (default: message_property_mappings={'role':'role', 'content':'content'})
     # If a property exists in the template but not in this mapping, the system will attempt
@@ -180,10 +187,14 @@ datasets:
     # adding a system turn with empty content.
     drop_system_message:
 
+    # Optional[bool]. (for Qwen3 template only) Whether to split the assistant content based on a reasoning trace inside delimited tags
+    # See example at `docs/dataset-formats/conversation.qmd`
+    split_thinking:
+
     # IMPORTANT: The following fields determine which parts of the conversation to train on.
     # Priority order: message_field_training > message_field_training_detail > train_on_inputs or role in roles_to_train
     # See examples at `docs/dataset-formats/conversation.qmd`
-    # Note: If the below 4 fields are set to empty, defaults to training only on the last message.
+    # Note: If the below 5 fields are empty, defaults to training only on the last message.
 
     # Optional[List[str]]. Roles to train on. The tokens from these roles will be considered for the loss.
     roles_to_train: ["assistant"]  # default
@@ -192,7 +203,13 @@ datasets:
     # - turn (default): train on the EOS token at the end of each trainable turn
     # - last: train on the last EOS token in the conversation
     # TIP: Please make sure that your `tokenizer.eos_token` is same as EOS/EOT token in template. Otherwise, set `eos_token` under `special_tokens`.
-    train_on_eos: last
+    train_on_eos: turn
+    # Optional[str]. Which EOT (End-of-Turn) tokens to train on in the conversation. Possible values are:
+    # - all: train on all EOT tokens
+    # - turn: train on the EOT token at the end of each trainable turn
+    # - last: train on the last EOT token in the conversation
+    # If not specified, defaults to the value of train_on_eos for backward compatibility.
+    train_on_eot:
     # The key in the message turn that indicates via boolean whether tokens of a turn should be considered for training. Useful to selectively train on certain turns besides the `roles_to_train`.
     message_field_training: training
     # The key in the message turn that contains the training details. Useful to selectively train on certain tokens in a turn.
@@ -275,8 +292,17 @@ process_reward_model:
 chat_template: tokenizer_default
 # custom jinja template for chat template. This will be only used if chat_template is set to `jinja` or `null` (in which case chat_template is automatically set to `jinja`). Default is null.
 chat_template_jinja: null
-# Changes the default system message. Currently only supports chatml.
-default_system_message: You are a helpful assistant. Please give a long and detailed answer.
+# Optional[List[str]]. Custom EOT (End-of-Turn) tokens to mask/unmask during training.
+# These tokens mark the boundaries between conversation turns.
+# For example: ["/INST", "</s>", "[/SYSTEM_PROMPT]"]
+# If not specified, defaults to just the model's eos_token.
+# This is useful for templates that use multiple delimiter tokens.
+eot_tokens:
+  # - "</s>"
+  # - "[/INST]"
+  # - "[/SYSTEM_PROMPT]"
+# Changes the default system message
+default_system_message: You are a helpful assistant. Please give a long and detailed answer. # Currently only supports chatml.
 # Axolotl attempts to save the dataset as an arrow after packing the data together so
 # subsequent training attempts load faster, relative path
 dataset_prepared_path: data/last_run_prepared
@@ -479,6 +505,7 @@ save_strategy: # Set to `"no"` to skip checkpoint saves, `"epoch"` at end of eac
 save_steps: # Leave empty to save at each epoch, integer for every N steps. float for fraction of total steps
 saves_per_epoch: # number of times per epoch to save a checkpoint, mutually exclusive with save_steps
 save_total_limit: # Checkpoints saved at a time
+save_only_model: # Save only the model weights, skipping the optimizer. Using this means you can't resume from checkpoints.
 # Maximum number of iterations to train for. It precedes num_epochs which means that
 # if both are set, num_epochs will not be guaranteed.
 # e.g., when 1 epoch is 1000 steps => `num_epochs: 2` and `max_steps: 100` will train for 100 steps
@@ -512,7 +539,7 @@ train_on_inputs: false
 # Note that training loss may have an oscillating pattern with this enabled.
 group_by_length: false
 
-# Whether to use gradient checkpointing. Available options are: true, false, "offload".
+# Whether to use gradient checkpointing. Available options are: true, false, "offload", "offload_disk".
 # https://huggingface.co/docs/transformers/v4.18.0/en/performance#gradient-checkpointing
 gradient_checkpointing: false
 # additional kwargs to pass to the trainer for gradient checkpointing
@@ -524,7 +551,7 @@ gradient_checkpointing: false
 early_stopping_patience: 3
 
 # Specify a scheduler and kwargs to use with the optimizer
-lr_scheduler: # 'one_cycle' | 'rex' | 'log_sweep' | empty for cosine
+lr_scheduler: # 'one_cycle' | 'rex' | 'log_sweep' | 'linear' | 'cosine_with_restarts' | 'polynomial' | 'constant' | 'constant_with_warmup' | 'inverse_sqrt' | 'reduce_lr_on_plateau' | 'cosine_with_min_lr' | 'warmup_stable_decay' | empty for cosine
 lr_scheduler_kwargs:
 cosine_min_lr_ratio: # decay lr to some percentage of the peak lr, e.g. cosine_min_lr_ratio=0.1 for 10% of peak lr
 cosine_constant_lr_ratio: # freeze lr at some percentage of the step, e.g. cosine_constant_lr_ratio=0.8 means start cosine_min_lr at 80% of training step (https://arxiv.org/pdf/2308.04014.pdf)
@@ -586,6 +613,7 @@ lr_div_factor: # Learning rate div factor
 # - optimi_adamw
 # - ao_adamw_8bit
 # - ao_adamw_fp8
+# - came_pytorch
 optimizer:
 # Dictionary of arguments to pass to the optimizer
 optim_args:
@@ -661,8 +689,10 @@ special_tokens:
   # unk_token: "<unk>"
   # pad_token: "[PAD]"
 
-# Add extra tokens.
+# Optional[list[str]]. Add extra tokens to the tokenizer.
 tokens:
+  # - "<|startoftext|>"
+  # - "<|endoftext|>"
 
 # Mapping token_id to new_token_string to override reserved added_tokens in the tokenizer.
 # Only works for tokens that are not part of the base vocab (aka are added_tokens).

docs/custom_integrations.qmd

@@ -49,7 +49,8 @@ sections = [
     ("Knowledge Distillation (KD)", "kd"),
     ("Liger Kernels", "liger"),
     ("Language Model Evaluation Harness (LM Eval)", "lm_eval"),
-    ("Spectrum", "spectrum")
+    ("Spectrum", "spectrum"),
+    ("LLMCompressor", "llm_compressor")
 ]
 
 for section_name, folder_name in sections:

docs/dataset-formats/conversation.qmd

@@ -4,18 +4,6 @@ description: Conversation format for supervised fine-tuning.
 order: 3
 ---
 
-## sharegpt
-
-::: {.callout-important}
-ShareGPT is deprecated!. Please see [chat_template](#chat_template) section below.
-:::
-
-## pygmalion
-
-```{.json filename="data.jsonl"}
-{"conversations": [{"role": "...", "value": "..."}]}
-```
-
 ## chat_template
 
 Chat Template strategy uses a jinja2 template that converts a list of messages into a prompt. Support using tokenizer's template, a supported template, or custom jinja2.
@@ -64,7 +52,7 @@ We recommend checking the below examples for other usecases.
 
 ### Examples
 
-1. Using the default chat template in the tokenizer_config.json on OpenAI messages format, training on only last message.
+1. (Legacy) Using the default chat template in the tokenizer_config.json on OpenAI messages format, training on only last message.
 
 ```yaml
 datasets:
@@ -109,10 +97,55 @@ datasets:
 ```
 
 ::: {.callout-important}
-Please make sure that your `tokenizer.eos_token` is same as EOS/EOT token in template. Otherwise, set `eos_token` under `special_tokens`.
+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: `.
 :::
 
-5. (Advanced) Using fine-grained control over tokens and turns to train in a conversation
+5. If you are using a template that has a different EOT (End-of-Turn) token from EOS token or multiple EOT tokens (like Mistral V7 Tekken), set the `eot_tokens: ` config. The handling of EOT tokens follows `train_on_eos: ` which defaults to turn.
+
+```yaml
+eot_tokens:
+  - "[/INST]"
+  # - "[/SYSTEM_PROMPT]"
+
+datasets:
+  - path: ...
+    type: chat_template
+
+    # optional
+    train_on_eot: turn  # defaults read from train_on_eos (which defaults to turn)
+```
+
+::: {.callout-tip}
+See [config documentation](../config.qmd) for detailed explanations of "turn", "last", and "all" options for training on tokens.
+:::
+
+::: {.callout-note}
+Using `eot_tokens` requires each token that exists in `chat_template` to be a single token in the tokenizer. Otherwise, the tokenizer will split the token and cause unexpected behavior.
+
+You can add those tokens as new tokens under `tokens: ` or (recommended) override unused added_tokens via `added_tokens_overrides: `. See [config](../config.qmd) for more details.
+:::
+
+6. Continuing from the previous example, if you want to train on all EOT token trainable turns but only last EOS token, set `train_on_eos: last`.
+
+```yaml
+eot_tokens:
+  - "[/INST]"
+  # ...
+
+datasets:
+  - path: ...
+    type: chat_template
+
+    train_on_eos: last
+    train_on_eot: turn
+```
+
+::: {.callout-tip}
+If EOS token only appears at the end of a prompt, `train_on_eos: last` is equivalent to `train_on_eos: turn`. Therefore, generally, you can leave them to their defaults and omit them.
+:::
+
+
+7. (Advanced) Using fine-grained control over tokens and turns to train in a conversation
 
 For a data sample that looks like:
 
@@ -162,3 +195,43 @@ datasets:
 ::: {.callout-tip}
 It is not necessary to set both `message_field_training` and `message_field_training_detail` at once.
 :::
+
+8. (For Qwen3 template only) Enable reasoning split, where the reasoning is split from the content and passed as a separate field into the template.
+
+```yaml
+datasets:
+  - path: ...
+    type: chat_template
+    chat_template: qwen3
+    split_thinking: true
+```
+
+For example, a content can look like:
+
+```json
+{
+  "content": "<think>Some thinking outputs</think>Output after thinking."
+}
+```
+
+After split, it will look like:
+
+```json
+{
+  "reasoning_content": "Some thinking outputs",
+  "content": "Output after thinking..."
+}
+```
+
+
+## sharegpt
+
+::: {.callout-important}
+ShareGPT is deprecated!. Please see [chat_template](#chat_template) section.
+:::
+
+## pygmalion
+
+```{.json filename="data.jsonl"}
+{"conversations": [{"role": "...", "value": "..."}]}
+```

docs/faq.qmd

@@ -73,10 +73,40 @@ description: Frequently asked questions
 
 > A: This is likely an empty turn.
 
-**Q: The EOS/EOT token is incorrectly being masked or not being masked.**
+**Q: The EOS token is incorrectly being masked or not being masked / `EOS token __ not found in chat template`.**
 
-> A: This is because of the mismatch between `tokenizer.eos_token` and EOS/EOT token in template. Please make sure to set `eos_token` under `special_tokens` to the same EOS/EOT token as in template.
+> A: There can be two reasons:
+
+> 1. This is because of the mismatch between `tokenizer.eos_token` and EOS token in template. Please make sure to set `eos_token: ` under `special_tokens: ` to the same EOS token as in template.
+
+> 2. The EOS token is not in the template. Please check if your template is correct. As an example, `phi_35` template does not use its dedicated EOS token `<|endoftext|>` at the end.
 
 **Q: "`chat_template` choice is `tokenizer_default` but tokenizer's `chat_template` is null. Please add a `chat_template` in tokenizer config"**
 
 > A: This is because the tokenizer does not have a chat template. Please add a chat template in the tokenizer config. See [chat_template](dataset-formats/conversation.qmd#chat-template) for more details.
+
+**Q: The EOT token(s) are incorrectly being masked or not being masked / `EOT token __ not found in chat template`.**
+
+> A: There can be two reasons:
+
+> 1. The EOT token is different from the EOS token and was not specified under `eot_tokens: `. Please set `eot_tokens: ` to the same EOT token(s) as in template.
+
+> 2. There is more than one EOT token per turn in the template. Please raise an issue with examples as we recognize this as an edge case.
+
+**Q: `EOT token encoding failed. Please check if the token is valid and can be encoded.`**
+
+> A: There could be some issue with the tokenizer or unicode encoding. Please raise an issue with examples with the EOT token & tokenizer causing the issue.
+
+**Q: `EOT token __ is encoded as multiple tokens.`**
+
+> A: This is because the EOT token is encoded as multiple tokens which can cause unexpected behavior. Please add it under `tokens: ` or (recommended) override unused added_tokens via `added_tokens_overrides: `.
+
+**Q: `Conflict between train_on_eos and train_on_eot. eos_token is in eot_tokens and train_on_eos != train_on_eot`**
+
+> A: This is because the EOS token is in the `eot_tokens: ` while mismatch between `train_on_eos: ` and `train_on_eot: `. This will cause one to override the other. Please ensure that `train_on_eos: ` and `train_on_eot: ` are the same or remove the EOS token from `eot_tokens: `.
+
+**Q: If `eot_tokens: ` is not provided, what happens?**
+
+> A: If `eot_tokens: ` is not provided, the default behavior is the same as before. EOS tokens used to delimit turns are masked/unmasked depending on whether the turn is trainable.
+
+> Internally, `eot_tokens: tokenizer.eos_token` and `train_on_eot: train_on_eos` (which defaults to `turn`). This transition helps clarify the naming and behavior of EOT/EOS tokens.

docs/multimodal.qmd

@@ -164,7 +164,7 @@ Here is an example of a multi-modal dataset:
         {
             "role": "user",
             "content": [
-                {"type": "image", "image": "https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/bee.jpg"},
+                {"type": "image", "url": "https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/bee.jpg"},
                 {"type": "text", "text": "Describe this image in detail."}
             ]
         },

docs/rlhf.qmd

@@ -502,9 +502,7 @@ The input format is a simple JSON input with customizable fields based on the ab
 Check out our [GRPO cookbook](https://github.com/axolotl-ai-cloud/axolotl-cookbook/tree/main/grpo#training-an-r1-style-large-language-model-using-grpo).
 :::
 
-If you have multiple GPUs available, we reccomend using `vLLM` with the `GRPOTrainer` to significantly speedup trajectory generation during training.
-First, launch a `vLLM` server using `trl vllm-serve` - you may use a config file or CLI overrides to configure your vLLM server. In this example, we're
-using 4 GPUs - 2 for training, and 2 for vLLM:
+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:
 
 ::: {.callout-important}
 Make sure you've installed the correct version of vLLM by including it as an extra when installing axolotl, e.g. `pip install axolotl[vllm]`.
@@ -539,6 +537,10 @@ Your `vLLM` instance will now attempt to spin up, and it's time to kick off trai
 CUDA_VISIBLE_DEVICES=0,1 axolotl train grpo.yaml --num-processes 2
 ```
 
+::: {.callout-note}
+Due to TRL's implementation with vLLM, the vLLM instance must use the last N GPUs instead of the first N GPUs. This is why in the example above, we use `CUDA_VISIBLE_DEVICES=2,3` for the vLLM instance.
+:::
+
 #### Reward functions
 
 GRPO uses custom reward functions and transformations. Please have them ready locally.

docs/sequence_parallelism.qmd

@@ -3,8 +3,6 @@ title: Sequence Parallelism
 description: Train with long sequences split across multiple GPUs.
 ---
 
-# Sequence Parallelism
-
 Sequence parallelism is a technique that splits sequences across multiple GPUs,
 allowing you to train with very long sequences that wouldn't fit on a single GPU. Each
 GPU processes a different portion of the sequence, and the results are aggregated
@@ -27,7 +25,7 @@ To enable sequence parallelism, add the following to your configuration file:
 sequence_parallel_degree: 4  # Split sequences across 4 GPUs
 # Optional; strides across the key dimension. Larger values use more memory but should make training faster.
 heads_k_stride: 1
-# Optional; one of "varlen_llama3", "batch_ring", "batch_zigzag", "batch_stripe". Defaults to
+# Optional; one of "varlen_llama3" or "batch_ring". Defaults to
 # "varlen_llama3" when `sample_packing: true`, and "batch_ring" otherwise.
 ring_attn_func:
 ```

examples/llama-3/sparse-finetuning.yaml

@@ -0,0 +1,77 @@
+base_model: neuralmagic/Sparse-Llama-3.1-8B-2of4
+
+plugins:
+  - axolotl.integrations.llm_compressor.LLMCompressorPlugin
+
+load_in_8bit: false
+load_in_4bit: false
+strict: false
+
+datasets:
+  - path: tatsu-lab/alpaca
+    type: alpaca
+dataset_prepared_path: last_run_prepared
+val_set_size: 0.05
+output_dir: ./outputs/out
+
+sequence_len: 4096
+sample_packing: true
+pad_to_sequence_len: true
+eval_sample_packing: false
+
+wandb_project:
+wandb_entity:
+wandb_watch:
+wandb_name:
+wandb_log_model:
+
+gradient_accumulation_steps: 8
+micro_batch_size: 1
+num_epochs: 1
+optimizer: paged_adamw_8bit
+lr_scheduler: cosine
+learning_rate: 2e-5
+
+train_on_inputs: false
+group_by_length: false
+bf16: auto
+fp16:
+tf32: false
+
+gradient_checkpointing: true
+gradient_checkpointing_kwargs:
+  use_reentrant: false
+early_stopping_patience:
+resume_from_checkpoint:
+logging_steps: 1
+xformers_attention:
+flash_attention: true
+
+warmup_steps: 100
+evals_per_epoch: 2
+eval_table_size:
+saves_per_epoch: 1
+debug:
+deepspeed:
+weight_decay: 0.0
+fsdp:
+fsdp_config:
+special_tokens:
+  pad_token: <|end_of_text|>
+
+llmcompressor:
+  recipe:
+    finetuning_stage:
+      finetuning_modifiers:
+        ConstantPruningModifier:
+          targets: [
+            're:.*q_proj.weight',
+            're:.*k_proj.weight',
+            're:.*v_proj.weight',
+            're:.*o_proj.weight',
+            're:.*gate_proj.weight',
+            're:.*up_proj.weight',
+            're:.*down_proj.weight',
+          ]
+          start: 0
+  save_compressed: true

examples/llama-4/README.md

@@ -34,3 +34,5 @@ We provide a script to delinearize Llama 4 linearized models into regular Huggin
 ```bash
 axolotl delinearize-llama4 --model path/to/model_dir --output path/to/output_dir
 ```
+
+Note: This only works with the non-quantized linearized model. If you have an adapter, merge it with the *non-quantized linearized* model before delinearizing.

examples/orpheus/README.md

@@ -0,0 +1,341 @@
+# Finetuning LLMs to output audio
+
+In this example, we finetune Orpcanopylabs/orpheus-tts-0.1-pretrained (a LLaMA 3.2 3b model) to output audio.
+
+The `finetune.yml` withe current settings will run on any Nvidia GPU with 45GB VRAM or more. If you adjust the batch size it can easily run on any GPU under 24GB.
+
+## Dataset pre-processing for pre-training
+If you are adding another voice in English, please jump ahead to finetuning pre-processing.
+
+For this to work, we need to preprocess our dataset. Since we are expecting to output audio, we will need to add tokens to the tokenizer.
+
+Using this code, it will download the SNAC model and add the correct tokens and upload the final dataset.
+
+```python
+import torch
+from snac import SNAC
+from datasets import load_dataset
+from huggingface_hub import snapshot_download
+from datasets import load_dataset
+import random
+import torchaudio.transforms as T
+from transformers import AutoTokenizer
+import os
+
+my_original_dataset_name = "<huggingface-id-of-dataset-that-we-want-to-preprocess>"
+name_to_push_dataset_to = "<huggingface-id-of-where-to-save-dataset>"
+
+dsn = my_original_dataset_name
+
+snapshot_download(
+    repo_id=dsn,
+    repo_type="dataset",
+    revision="main",
+    max_workers=64,
+)
+
+
+ds = load_dataset(dsn, split="train")
+ds_sample_rate = ds[0]["audio"]["sampling_rate"]
+
+model = SNAC.from_pretrained("hubertsiuzdak/snac_24khz")
+model = model.to("mps")
+
+def tokenise_audio(waveform):
+  waveform = torch.from_numpy(waveform).unsqueeze(0)
+  waveform = waveform.to(dtype=torch.float32)
+  resample_transform = T.Resample(orig_freq=ds_sample_rate, new_freq=24000)
+  waveform = resample_transform(waveform)
+
+  waveform = waveform.unsqueeze(0).to("cuda")
+
+  #generate the codes from snac
+  with torch.inference_mode():
+    codes = model.encode(waveform)
+
+  all_codes = []
+  for i in range(codes[0].shape[1]):
+    all_codes.append(codes[0][0][i].item()+128266)
+    all_codes.append(codes[1][0][2*i].item()+128266+4096)
+    all_codes.append(codes[2][0][4*i].item()+128266+(2*4096))
+    all_codes.append(codes[2][0][(4*i)+1].item()+128266+(3*4096))
+    all_codes.append(codes[1][0][(2*i)+1].item()+128266+(4*4096))
+    all_codes.append(codes[2][0][(4*i)+2].item()+128266+(5*4096))
+    all_codes.append(codes[2][0][(4*i)+3].item()+128266+(6*4096))
+
+
+  return all_codes
+
+def add_codes(example):
+    # Always initialize codes_list to None
+    codes_list = None
+
+    try:
+        answer_audio = example.get("audio")
+        # If there's a valid audio array, tokenise it
+        if answer_audio and "array" in answer_audio:
+            audio_array = answer_audio["array"]
+            codes_list = tokenise_audio(audio_array)
+    except Exception as e:
+        print(f"Skipping row due to error: {e}")
+        # Keep codes_list as None if we fail
+    example["codes_list"] = codes_list
+
+    return example
+
+ds = ds.map(add_codes, remove_columns=["audio"])
+
+#@title Load Tokenizer
+tokeniser_length = 128256
+start_of_text = 128000
+end_of_text = 128009
+
+start_of_speech = tokeniser_length + 1
+end_of_speech = tokeniser_length + 2
+
+start_of_human = tokeniser_length + 3
+end_of_human = tokeniser_length + 4
+
+start_of_ai = tokeniser_length + 5
+end_of_ai =  tokeniser_length + 6
+pad_token = tokeniser_length + 7
+
+audio_tokens_start = tokeniser_length + 10
+
+tokenizer_name = "canopylabs/orpheus-3b-0.1-pretrained"
+
+
+tokenizer = AutoTokenizer.from_pretrained(tokenizer_name)
+num_proc = os.cpu_count() - 2
+
+ds = ds.filter(lambda x: x["codes_list"] is not None)
+ds = ds.filter(lambda x: len(x["codes_list"]) > 0)
+
+#@title Create Input Ids
+def remove_duplicate_frames(example):
+    vals = example["codes_list"]
+    if len(vals) % 7 != 0:
+        raise ValueError("Input list length must be divisible by 7")
+
+    result = vals[:7]
+
+    removed_frames = 0
+
+    for i in range(7, len(vals), 7):
+        current_first = vals[i]
+        previous_first = result[-7]
+
+        if current_first != previous_first:
+            result.extend(vals[i:i+7])
+        else:
+            removed_frames += 1
+
+    example["codes_list"] = result
+
+    return example
+
+ds = ds.map(remove_duplicate_frames, num_proc=num_proc)
+
+
+def create_input_ids(example):
+    text_ids = tokenizer.encode({example['text']},  add_special_tokens=True)
+    text_ids.append(end_of_text)
+    example["text_tokens"] = text_ids
+    input_ids = (
+        [start_of_human]
+        + example["text_tokens"]
+        + [end_of_human]
+        + [start_of_ai]
+        + [start_of_speech]
+        + example["codes_list"]
+        + [end_of_speech]
+        + [end_of_ai]
+    )
+    example["input_ids"] = input_ids
+    example["labels"] = input_ids
+    example["attention_mask"] = [1] * len(input_ids)
+
+    return example
+
+ds = ds.map(create_input_ids, num_proc=num_proc, remove_columns=["text", "codes_list"])
+
+#@title Remove unnecessary columns
+columns_to_keep = ["input_ids", "labels", "attention_mask"]
+columns_to_remove = [col for col in ds.column_names if col not in columns_to_keep]
+
+ds = ds.remove_columns(columns_to_remove)
+
+ds.push_to_hub(name_to_push_dataset_to)
+```
+
+
+## Finetune pre-processing
+Use this code to add a new voice.
+
+```python
+import torch
+from snac import SNAC
+from datasets import load_dataset
+from huggingface_hub import snapshot_download
+from datasets import load_dataset
+import random
+import torchaudio.transforms as T
+from transformers import AutoTokenizer
+import os
+
+my_original_dataset_name = "<huggingface-id-of-dataset-that-we-want-to-preprocess>"
+name_to_push_dataset_to = "<huggingface-id-of-where-to-save-dataset>"
+
+dsn = my_original_dataset_name
+
+snapshot_download(
+    repo_id=dsn,
+    repo_type="dataset",
+    revision="main",
+    max_workers=64,
+)
+
+
+ds = load_dataset(dsn, split="train")
+ds_sample_rate = ds[0]["audio"]["sampling_rate"]
+
+model = SNAC.from_pretrained("hubertsiuzdak/snac_24khz")
+model = model.to("mps")
+
+def tokenise_audio(waveform):
+  waveform = torch.from_numpy(waveform).unsqueeze(0)
+  waveform = waveform.to(dtype=torch.float32)
+  resample_transform = T.Resample(orig_freq=ds_sample_rate, new_freq=24000)
+  waveform = resample_transform(waveform)
+
+  waveform = waveform.unsqueeze(0).to("cuda")
+
+  #generate the codes from snac
+  with torch.inference_mode():
+    codes = model.encode(waveform)
+
+  all_codes = []
+  for i in range(codes[0].shape[1]):
+    all_codes.append(codes[0][0][i].item()+128266)
+    all_codes.append(codes[1][0][2*i].item()+128266+4096)
+    all_codes.append(codes[2][0][4*i].item()+128266+(2*4096))
+    all_codes.append(codes[2][0][(4*i)+1].item()+128266+(3*4096))
+    all_codes.append(codes[1][0][(2*i)+1].item()+128266+(4*4096))
+    all_codes.append(codes[2][0][(4*i)+2].item()+128266+(5*4096))
+    all_codes.append(codes[2][0][(4*i)+3].item()+128266+(6*4096))
+
+
+  return all_codes
+
+def add_codes(example):
+    # Always initialize codes_list to None
+    codes_list = None
+
+    try:
+        answer_audio = example.get("audio")
+        # If there's a valid audio array, tokenise it
+        if answer_audio and "array" in answer_audio:
+            audio_array = answer_audio["array"]
+            codes_list = tokenise_audio(audio_array)
+    except Exception as e:
+        print(f"Skipping row due to error: {e}")
+        # Keep codes_list as None if we fail
+    example["codes_list"] = codes_list
+
+    return example
+
+ds = ds.map(add_codes, remove_columns=["audio"])
+
+#@title Load Tokenizer
+tokeniser_length = 128256
+start_of_text = 128000
+end_of_text = 128009
+
+start_of_speech = tokeniser_length + 1
+end_of_speech = tokeniser_length + 2
+
+start_of_human = tokeniser_length + 3
+end_of_human = tokeniser_length + 4
+
+start_of_ai = tokeniser_length + 5
+end_of_ai =  tokeniser_length + 6
+pad_token = tokeniser_length + 7
+
+audio_tokens_start = tokeniser_length + 10
+
+tokenizer_name = "canopylabs/orpheus-3b-0.1-pretrained"
+
+
+tokenizer = AutoTokenizer.from_pretrained(tokenizer_name)
+num_proc = os.cpu_count() - 2
+
+ds = ds.filter(lambda x: x["codes_list"] is not None)
+ds = ds.filter(lambda x: len(x["codes_list"]) > 0)
+
+#@title Create Input Ids
+def remove_duplicate_frames(example):
+    vals = example["codes_list"]
+    if len(vals) % 7 != 0:
+        raise ValueError("Input list length must be divisible by 7")
+
+    result = vals[:7]
+
+    removed_frames = 0
+
+    for i in range(7, len(vals), 7):
+        current_first = vals[i]
+        previous_first = result[-7]
+
+        if current_first != previous_first:
+            result.extend(vals[i:i+7])
+        else:
+            removed_frames += 1
+
+    example["codes_list"] = result
+
+    return example
+
+ds = ds.map(remove_duplicate_frames, num_proc=num_proc)
+
+tok_info = '''*** HERE you can modify the text prompt
+i.e. if you wanted a multispeaker model like canopylabs/orpheus-3b-0.1-ft, you can pass:
+f"{example["source"]}:  {example["text"]}", as is passed.
+'''
+print(tok_info)
+
+def create_input_ids(example):
+    text_ids = tokenizer.encode(f"{example['speaker_id']}: {example['text']}",  add_special_tokens=True)
+    text_ids.append(end_of_text)
+    example["text_tokens"] = text_ids
+    input_ids = (
+        [start_of_human]
+        + example["text_tokens"]
+        + [end_of_human]
+        + [start_of_ai]
+        + [start_of_speech]
+        + example["codes_list"]
+        + [end_of_speech]
+        + [end_of_ai]
+    )
+    example["input_ids"] = input_ids
+    example["labels"] = input_ids
+    example["attention_mask"] = [1] * len(input_ids)
+
+    return example
+
+ds = ds.map(create_input_ids, num_proc=num_proc, remove_columns=["text", "codes_list"])
+
+#@title Remove unnecessary columns
+columns_to_keep = ["input_ids", "labels", "attention_mask"]
+columns_to_remove = [col for col in ds.column_names if col not in columns_to_keep]
+
+ds = ds.remove_columns(columns_to_remove)
+
+ds.push_to_hub(name_to_push_dataset_to)
+```
+
+## Training
+After preprocessing is done, fill out the blanks in finetune.yml and simply run `axolotl train finetune.yml`
+
+## Inference
+For inference, please refer to the original [orpheus github](https://github.com/canopyai/Orpheus-TTS/tree/main).

examples/orpheus/finetune.yml

@@ -0,0 +1,52 @@
+base_model: canopylabs/orpheus-3b-0.1-pretrained
+
+hub_model_id: <your-hub-model-id>
+
+plugins:
+  - axolotl.integrations.liger.LigerPlugin
+liger_rope: true
+liger_rms_norm: true
+liger_glu_activation: true
+liger_fused_linear_cross_entropy: true
+
+datasets:
+  - path: <your-hf-dataset-id>
+    type:  # leave empty to load pre-tokenized
+dataset_prepared_path: last_run_prepared
+val_set_size: 0.01
+output_dir: ./outputs/out
+
+sequence_len: 8192
+sample_packing: true
+pad_to_sequence_len: true
+
+wandb_project:
+wandb_entity:
+wandb_watch:
+wandb_name:
+wandb_log_model:
+
+gradient_accumulation_steps: 8
+micro_batch_size: 4
+num_epochs: 3
+optimizer: adamw_torch_fused
+lr_scheduler: cosine
+learning_rate: 2e-5
+
+bf16: auto
+tf32: false
+
+gradient_checkpointing: true
+gradient_checkpointing_kwargs:
+  use_reentrant: false
+resume_from_checkpoint:
+logging_steps: 1
+flash_attention: true
+
+warmup_steps: 20
+evals_per_epoch: 5
+saves_per_epoch: 5
+weight_decay: 0.05
+
+special_tokens:
+  pad_token: <custom_token_7>

examples/qwen3/32b-qlora.yaml

@@ -0,0 +1,69 @@
+base_model: Qwen/Qwen3-32B
+# Automatically upload checkpoint and final model to HF
+# hub_model_id: username/custom_model_name
+
+plugins:
+  - axolotl.integrations.cut_cross_entropy.CutCrossEntropyPlugin
+strict: false
+
+chat_template: qwen3
+datasets:
+  - path: mlabonne/FineTome-100k
+    type: chat_template
+    split: train[:20%]
+    field_messages: conversations
+    message_property_mappings:
+      role: from
+      content: value
+val_set_size: 0.0
+output_dir: ./outputs/out
+dataset_prepared_path: last_run_prepared
+
+sequence_len: 2048
+sample_packing: true
+eval_sample_packing: true
+pad_to_sequence_len: true
+
+load_in_4bit: true
+adapter: qlora
+lora_r: 16
+lora_alpha: 32
+lora_target_modules:
+  - q_proj
+  - k_proj
+  - v_proj
+  - o_proj
+  - down_proj
+  - up_proj
+lora_mlp_kernel: true
+lora_qkv_kernel: true
+lora_o_kernel: true
+
+wandb_project:
+wandb_entity:
+wandb_watch:
+wandb_name:
+wandb_log_model:
+
+gradient_accumulation_steps: 2
+micro_batch_size: 1
+num_epochs: 1
+optimizer: adamw_torch_4bit
+lr_scheduler: cosine
+learning_rate: 0.0002
+
+bf16: auto
+tf32: true
+
+gradient_checkpointing: offload
+gradient_checkpointing_kwargs:
+  use_reentrant: false
+resume_from_checkpoint:
+logging_steps: 1
+flash_attention: true
+
+warmup_steps: 10
+evals_per_epoch: 4
+saves_per_epoch: 1
+weight_decay: 0.0
+special_tokens:

examples/qwen3/qlora-fsdp.yaml

@@ -0,0 +1,68 @@
+base_model: Qwen/Qwen3-8B
+# Automatically upload checkpoint and final model to HF
+# hub_model_id: username/custom_model_name
+
+load_in_8bit: false
+load_in_4bit: true
+strict: false
+
+datasets:
+  - path: tatsu-lab/alpaca
+    type: alpaca
+dataset_prepared_path:
+val_set_size: 0.05
+output_dir: ./outputs/out
+
+sequence_len: 2048
+sample_packing: true
+eval_sample_packing: true
+pad_to_sequence_len: true
+
+adapter: qlora
+lora_model_dir:
+lora_r: 32
+lora_alpha: 64
+lora_dropout: 0.05
+lora_target_linear: true
+
+wandb_project:
+wandb_entity:
+wandb_watch:
+wandb_name:
+wandb_log_model:
+
+gradient_accumulation_steps: 4
+micro_batch_size: 1
+num_epochs: 1
+optimizer: adamw_torch_fused
+lr_scheduler: cosine
+learning_rate: 0.0002
+
+bf16: auto
+tf32: true
+
+gradient_checkpointing: true
+gradient_checkpointing_kwargs:
+  use_reentrant: false
+resume_from_checkpoint:
+logging_steps: 1
+flash_attention: true
+
+warmup_steps: 10
+evals_per_epoch: 4
+saves_per_epoch: 1
+weight_decay: 0.0
+fsdp:
+  - full_shard
+  - auto_wrap
+fsdp_config:
+  fsdp_limit_all_gathers: true
+  fsdp_sync_module_states: true
+  fsdp_offload_params: true
+  fsdp_use_orig_params: false
+  fsdp_cpu_ram_efficient_loading: true
+  fsdp_auto_wrap_policy: TRANSFORMER_BASED_WRAP
+  fsdp_transformer_layer_cls_to_wrap: Qwen3DecoderLayer
+  fsdp_state_dict_type: FULL_STATE_DICT
+  fsdp_sharding_strategy: FULL_SHARD
+special_tokens:

requirements.txt

@@ -6,19 +6,20 @@ triton>=3.0.0
 mamba-ssm==1.2.0.post1
 xformers>=0.0.23.post1
 autoawq==0.2.7.post3
-liger-kernel==0.5.8
+liger-kernel==0.5.9
 # END section
 
 packaging==23.2
 
-peft==0.15.1
+huggingface_hub==0.31.0
+peft==0.15.2
 transformers==4.51.3
 tokenizers>=0.21.1
 accelerate==1.6.0
-datasets==3.5.0
+datasets==3.5.1
 deepspeed>=0.15.4
-trl==0.16.1
-hf_xet==1.0.0
+trl==0.17.0
+hf_xet==1.1.0
 hqq==0.2.5
 
 optimum==1.16.2

setup.py

@@ -67,13 +67,13 @@ def parse_requirements(extras_require_map):
             if (major, minor) >= (2, 7):
                 _install_requires.pop(_install_requires.index(xformers_version))
                 # _install_requires.append("xformers==0.0.29.post3")  # xformers seems to be hard pinned to 2.6.0
-                extras_require_map["vllm"] = ["vllm==0.8.3"]
+                extras_require_map["vllm"] = ["vllm==0.8.5.post1"]
             elif (major, minor) >= (2, 6):
                 _install_requires.pop(_install_requires.index(xformers_version))
                 _install_requires.append(
                     "xformers==0.0.29.post2"
                 )  # vllm needs post2 w torch 2.6
-                extras_require_map["vllm"] = ["vllm==0.8.3"]
+                extras_require_map["vllm"] = ["vllm==0.8.5.post1"]
             elif (major, minor) >= (2, 5):
                 _install_requires.pop(_install_requires.index(xformers_version))
                 if patch == 0:
@@ -142,6 +142,7 @@ extras_require = {
         "apollo-torch",
         "lomo-optim==0.1.1",
         "torch-optimi==0.2.1",
+        "came_pytorch==0.1.3",
     ],
     "ray": [
         "ray[train]",
@@ -149,6 +150,9 @@ extras_require = {
     "vllm": [
         "vllm==0.7.2",
     ],
+    "llmcompressor": [
+        "llmcompressor==0.5.1",
+    ],
 }
 
 install_requires, dependency_links, extras_require_build = parse_requirements(

src/axolotl/__init__.py

@@ -4,4 +4,4 @@ import pkgutil
 
 __path__ = pkgutil.extend_path(__path__, __name__)  # Make this a namespace package
 
-__version__ = "0.8.0"
+__version__ = "0.10.0.dev0"

src/axolotl/cli/__init__.py

@@ -2,4 +2,7 @@
 
 import os
 
+from axolotl.logging_config import configure_logging
+
 os.environ["HF_HUB_ENABLE_HF_TRANSFER"] = "1"
+configure_logging()

src/axolotl/cli/args.py

@@ -82,6 +82,12 @@ class VllmServeCliArgs:
             "hardware support this feature."
         },
     )
+    serve_module: Optional[str] = field(
+        default=None,
+        metadata={
+            "help": "Module to serve. If not set, the default module will be used."
+        },
+    )
 
 
 @dataclass

src/axolotl/cli/art.py

@@ -16,8 +16,15 @@ AXOLOTL_LOGO = """
     @@@@  @@@@@@@@@@@@@@@@
 """
 
+HAS_PRINTED_LOGO = False
+
 
 def print_axolotl_text_art():
     """Prints axolotl ASCII art."""
+
+    global HAS_PRINTED_LOGO  # pylint: disable=global-statement
+    if HAS_PRINTED_LOGO:
+        return
     if is_main_process():
+        HAS_PRINTED_LOGO = True
         print(AXOLOTL_LOGO)

src/axolotl/cli/checks.py

@@ -8,9 +8,6 @@ from accelerate.commands.config import config_args
 from huggingface_hub import HfApi
 from huggingface_hub.utils import LocalTokenNotFoundError
 
-from axolotl.logging_config import configure_logging
-
-configure_logging()
 LOG = logging.getLogger(__name__)
 
 

src/axolotl/cli/config.py

@@ -5,6 +5,7 @@ import logging
 import os
 import tempfile
 from pathlib import Path
+from tempfile import NamedTemporaryFile
 from typing import Union
 from urllib.parse import urlparse
 
@@ -152,7 +153,15 @@ def prepare_plugins(cfg: DictDefault):
             plugin_manager.register(plugin_name)
 
 
-def load_cfg(config: Union[str, Path] = Path("examples/"), **kwargs) -> DictDefault:
+def plugin_set_cfg(cfg: DictDefault):
+    if cfg.get("plugins"):
+        plugin_manager = PluginManager.get_instance()
+        plugin_manager.cfg = cfg
+
+
+def load_cfg(
+    config: str | Path | DictDefault = Path("examples/"), **kwargs
+) -> DictDefault:
     """
     Loads the `axolotl` configuration stored at `config`, validates it, and performs
     various setup.
@@ -164,13 +173,24 @@ def load_cfg(config: Union[str, Path] = Path("examples/"), **kwargs) -> DictDefa
     Returns:
         `DictDefault` mapping configuration keys to values.
     """
-    config = check_remote_config(config)
-    if Path(config).is_dir():
-        config = choose_config(Path(config))
+    if isinstance(config, (str, Path)):
+        config = check_remote_config(config)
+        if Path(config).is_dir():
+            config = choose_config(Path(config))
 
-    # Load the config from the yaml file
-    with open(config, encoding="utf-8") as file:
-        cfg: DictDefault = DictDefault(yaml.safe_load(file))
+        # Load the config from the yaml file
+        with open(config, encoding="utf-8") as file:
+            cfg: DictDefault = DictDefault(yaml.safe_load(file))
+
+        cfg.axolotl_config_path = config
+    else:
+        cfg = config
+        with NamedTemporaryFile(
+            mode="w", delete=False, suffix=".yml", prefix="axolotl_config_"
+        ) as temp_file:
+            temp_file.write(yaml.dump(config.to_dict()))
+            temp_file.close()
+        cfg.axolotl_config_path = temp_file.name
 
     # If there are any options passed in the cli, if it is something that seems valid
     # from the yaml, then overwrite the value
@@ -184,8 +204,6 @@ def load_cfg(config: Union[str, Path] = Path("examples/"), **kwargs) -> DictDefa
             else:
                 cfg[k] = kwargs[k]
 
-    cfg.axolotl_config_path = config
-
     try:
         device_props = torch.cuda.get_device_properties("cuda")
         gpu_version = "sm_" + str(device_props.major) + str(device_props.minor)
@@ -213,5 +231,6 @@ def load_cfg(config: Union[str, Path] = Path("examples/"), **kwargs) -> DictDefa
     setup_wandb_env_vars(cfg)
     setup_mlflow_env_vars(cfg)
     setup_comet_env_vars(cfg)
+    plugin_set_cfg(cfg)
 
     return cfg

src/axolotl/cli/evaluate.py

@@ -1,6 +1,7 @@
 """CLI to run evaluation on a model."""
 
 import logging
+import os
 from pathlib import Path
 from typing import Union
 
@@ -14,6 +15,7 @@ from axolotl.cli.checks import check_accelerate_default_config, check_user_token
 from axolotl.cli.config import load_cfg
 from axolotl.common.datasets import load_datasets, load_preference_datasets
 from axolotl.evaluate import evaluate
+from axolotl.utils import patch_optimized_env
 from axolotl.utils.dict import DictDefault
 
 LOG = logging.getLogger(__name__)
@@ -29,10 +31,14 @@ def do_evaluate(cfg: DictDefault, cli_args: TrainerCliArgs) -> None:
         cfg: Dictionary mapping `axolotl` config keys to values.
         cli_args: CLI arguments.
     """
+    # Enable expandable segments for cuda allocation to improve VRAM usage
+    patch_optimized_env()
+
     # pylint: disable=duplicate-code
     print_axolotl_text_art()
     check_accelerate_default_config()
-    check_user_token()
+    if int(os.getenv("LOCAL_RANK", "0")) == 0:
+        check_user_token()
 
     if cfg.rl:
         dataset_meta = load_preference_datasets(cfg=cfg, cli_args=cli_args)

src/axolotl/cli/main.py

@@ -28,9 +28,8 @@ from axolotl.cli.utils import (
     fetch_from_github,
     filter_none_kwargs,
 )
-from axolotl.cli.vllm_serve import do_vllm_serve
 from axolotl.integrations.lm_eval.cli import lm_eval
-from axolotl.utils import set_pytorch_cuda_alloc_conf
+from axolotl.utils import patch_optimized_env
 from axolotl.utils.schemas.config import AxolotlInputConfig
 
 
@@ -56,6 +55,8 @@ def preprocess(config: str, cloud: Optional[str] = None, **kwargs) -> None:
         kwargs: Additional keyword arguments which correspond to CLI args or `axolotl`
             config options.
     """
+    patch_optimized_env()
+
     if cloud:
         from axolotl.cli.cloud import do_cli_preprocess
 
@@ -101,7 +102,7 @@ def train(
             config options.
     """
     # Enable expandable segments for cuda allocation to improve VRAM usage
-    set_pytorch_cuda_alloc_conf()
+    patch_optimized_env()
 
     if "use_ray" in kwargs and kwargs["use_ray"]:
         accelerate = False
@@ -327,6 +328,8 @@ def fetch(directory: str, dest: Optional[str]) -> None:
 @add_options_from_dataclass(VllmServeCliArgs)
 @filter_none_kwargs
 def vllm_serve(config: str, **cli_args: VllmServeCliArgs):
+    from axolotl.cli.vllm_serve import do_vllm_serve
+
     do_vllm_serve(config, cli_args)
 
 

src/axolotl/cli/preprocess.py

@@ -18,6 +18,7 @@ from axolotl.cli.checks import check_accelerate_default_config, check_user_token
 from axolotl.cli.config import load_cfg
 from axolotl.common.const import DEFAULT_DATASET_PREPARED_PATH
 from axolotl.common.datasets import load_datasets, load_preference_datasets
+from axolotl.integrations.base import PluginManager
 from axolotl.utils.dict import DictDefault
 from axolotl.utils.trainer import disable_datasets_caching
 
@@ -47,7 +48,10 @@ def do_preprocess(cfg: DictDefault, cli_args: PreprocessCliArgs) -> None:
         cfg.dataset_prepared_path = DEFAULT_DATASET_PREPARED_PATH
 
     with disable_datasets_caching():
-        if cfg.rl:
+        plugin_manager = PluginManager.get_instance()
+        if plugin_manager.load_datasets(cfg, preprocess=True):
+            pass
+        elif cfg.rl:
             load_preference_datasets(cfg=cfg, cli_args=cli_args)
         else:
             load_datasets(cfg=cfg, cli_args=cli_args)

src/axolotl/cli/train.py

@@ -1,5 +1,6 @@
 """CLI to run training on a model."""
 
+import gc
 import logging
 import os
 from pathlib import Path
@@ -17,7 +18,7 @@ from axolotl.cli.config import load_cfg
 from axolotl.common.datasets import load_datasets, load_preference_datasets
 from axolotl.integrations.base import PluginManager
 from axolotl.train import train
-from axolotl.utils import set_pytorch_cuda_alloc_conf
+from axolotl.utils import patch_optimized_env
 from axolotl.utils.config import normalize_config, resolve_dtype
 from axolotl.utils.dict import DictDefault
 
@@ -35,21 +36,27 @@ def do_train(cfg: DictDefault, cli_args: TrainerCliArgs):
         cli_args: Training-specific CLI arguments.
     """
     # Enable expandable segments for cuda allocation to improve VRAM usage
-    set_pytorch_cuda_alloc_conf()
+    patch_optimized_env()
 
     print_axolotl_text_art()
     check_accelerate_default_config()
     if int(os.getenv("LOCAL_RANK", "0")) == 0:
         check_user_token()
 
-    if cfg.rl:
-        dataset_meta = load_preference_datasets(cfg=cfg, cli_args=cli_args)
-    else:
-        dataset_meta = load_datasets(cfg=cfg, cli_args=cli_args)
+    plugin_manager = PluginManager.get_instance()
+    dataset_meta = plugin_manager.load_datasets(cfg, preprocess=False)
+    if not dataset_meta:
+        if cfg.rl:
+            dataset_meta = load_preference_datasets(cfg=cfg, cli_args=cli_args)
+        else:
+            dataset_meta = load_datasets(cfg=cfg, cli_args=cli_args)
 
     model, tokenizer, trainer = train(cfg=cfg, dataset_meta=dataset_meta)
+
     del model, tokenizer, trainer
 
+    gc.collect()
+
     plugin_manager = PluginManager.get_instance()
     plugin_manager.post_train_unload(cfg)
 

src/axolotl/cli/utils.py

@@ -20,11 +20,9 @@ from transformers import (
     ProcessorMixin,
 )
 
-from axolotl.logging_config import configure_logging
 from axolotl.utils.dict import DictDefault
 from axolotl.utils.models import load_model, load_processor, load_tokenizer
 
-configure_logging()
 LOG = logging.getLogger(__name__)
 
 

src/axolotl/cli/vllm_serve.py

@@ -6,7 +6,6 @@ from pathlib import Path
 from typing import Union
 
 from trl.scripts.vllm_serve import ScriptArguments
-from trl.scripts.vllm_serve import main as vllm_serve_main
 
 from axolotl.cli.config import load_cfg
 
@@ -28,6 +27,9 @@ def do_vllm_serve(
     cfg = load_cfg(config)
     model = cfg.base_model
 
+    serve_module = cli_args.get("serve_module", "trl.scripts.vllm_serve")
+    vllm_serve_main = getattr(__import__(serve_module, fromlist=["main"]), "main")
+
     tensor_parallel_size = (
         cli_args.get("tensor_parallel_size") or cfg.vllm.tensor_parallel_size
     )

src/axolotl/common/architectures.py

@@ -11,5 +11,6 @@ MOE_ARCH_BLOCK = {
     ],
     "mixtral": "MixtralSparseMoeBlock",
     "qwen2_moe": "Qwen2MoeSparseMoeBlock",
+    "qwen3_moe": "Qwen3MoeSparseMoeBlock",
     "deepseek_v2": "DeepseekV2MoE",
 }

src/axolotl/common/datasets.py

@@ -14,6 +14,7 @@ from axolotl.utils.data import prepare_dataset
 from axolotl.utils.data.rl import load_prepare_preference_datasets
 from axolotl.utils.dict import DictDefault
 from axolotl.utils.models import load_processor, load_tokenizer
+from axolotl.utils.schemas.enums import RLType
 from axolotl.utils.tokenization import check_dataset_labels
 
 LOG = logging.getLogger(__name__)
@@ -47,7 +48,8 @@ def sample_dataset(dataset: Dataset, num_samples: int) -> Dataset:
 def load_datasets(
     *,
     cfg: DictDefault,
-    cli_args: Union[PreprocessCliArgs, TrainerCliArgs],
+    cli_args: PreprocessCliArgs | TrainerCliArgs | None = None,
+    debug: bool = False,
 ) -> TrainDatasetMeta:
     """
     Loads one or more training or evaluation datasets, calling
@@ -56,6 +58,7 @@ def load_datasets(
     Args:
         cfg: Dictionary mapping `axolotl` config keys to values.
         cli_args: Command-specific CLI arguments.
+        debug: Whether to print out tokenization of sample
 
     Returns:
         Dataclass with fields for training and evaluation datasets and the computed
@@ -64,7 +67,8 @@ def load_datasets(
     tokenizer = load_tokenizer(cfg)
     processor = load_processor(cfg, tokenizer=tokenizer) if cfg.processor_type else None
     preprocess_iterable = (
-        hasattr(cli_args, "iterable")
+        cli_args
+        and hasattr(cli_args, "iterable")
         and cli_args.iterable is not None
         and cli_args.iterable
     )
@@ -76,20 +80,25 @@ def load_datasets(
         preprocess_iterable=preprocess_iterable,
     )
 
-    if (
-        cli_args.debug
-        or cfg.debug
-        or cli_args.debug_text_only
-        or int(cli_args.debug_num_examples) > 0
-    ):
+    if (  # pylint: disable=too-many-boolean-expressions
+        cli_args
+        and (
+            cli_args.debug
+            or cfg.debug
+            or cli_args.debug_text_only
+            or int(cli_args.debug_num_examples) > 0
+        )
+    ) or debug:
         LOG.info("check_dataset_labels...")
 
-        train_samples = sample_dataset(train_dataset, cli_args.debug_num_examples)
+        num_examples = cli_args.debug_num_examples if cli_args else 1
+        text_only = cli_args.debug_text_only if cli_args else False
+        train_samples = sample_dataset(train_dataset, num_examples)
         check_dataset_labels(
             train_samples,
             tokenizer,
-            num_examples=cli_args.debug_num_examples,
-            text_only=cli_args.debug_text_only,
+            num_examples=num_examples,
+            text_only=text_only,
         )
 
         LOG.info("printing prompters...")
@@ -125,7 +134,7 @@ def load_preference_datasets(
     total_num_steps: Optional[int] = int(
         math.ceil(len(train_dataset) * cfg.num_epochs / cfg.batch_size)
     )
-    if cfg.rl == "grpo":
+    if cfg.rl is RLType.GRPO:
         total_num_steps = None
 
     if cli_args.debug or cfg.debug:

src/axolotl/core/trainer_builder.py

@@ -21,6 +21,7 @@ import importlib.util
 import inspect
 import logging
 import math
+import os
 import sys
 from abc import abstractmethod
 from pathlib import Path
@@ -60,6 +61,7 @@ from axolotl.core.training_args import (
 from axolotl.integrations.base import PluginManager
 from axolotl.monkeypatch.multipack import SUPPORTED_MULTIPACK_MODEL_TYPES
 from axolotl.monkeypatch.relora import ReLoRACallback
+from axolotl.monkeypatch.trainer.lr import patch_trainer_get_lr
 from axolotl.processing_strategies import get_processing_strategy
 from axolotl.utils import is_comet_available, is_mlflow_available
 from axolotl.utils.callbacks import (
@@ -71,6 +73,7 @@ from axolotl.utils.callbacks import (
     SaveBetterTransformerModelCallback,
     bench_eval_callback_factory,
     causal_lm_bench_eval_callback_factory,
+    colab_inference_post_train_callback,
     log_prediction_callback_factory,
 )
 from axolotl.utils.callbacks.lisa import lisa_callback_factory
@@ -84,7 +87,7 @@ from axolotl.utils.collators import (
 )
 from axolotl.utils.collators.mm_chat import MultiModalChatDataCollator
 from axolotl.utils.models import ensure_dtype
-from axolotl.utils.schemas.enums import CustomSupportedOptimizers
+from axolotl.utils.schemas.enums import CustomSupportedOptimizers, RLType
 
 try:
     import torch._dynamo  # pylint: disable=ungrouped-imports
@@ -114,6 +117,8 @@ class TrainerBuilderBase(abc.ABC):
         if hasattr(model, "add_model_tags"):
             model.add_model_tags(["axolotl"])
 
+        patch_trainer_get_lr()
+
     @property
     def model_ref(self):
         return self._model_ref
@@ -165,6 +170,9 @@ class TrainerBuilderBase(abc.ABC):
                 )
             )
 
+        if self.cfg.gc_steps:
+            callbacks.append(GCCallback(gc_steps=self.cfg.gc_steps))
+
         if self.cfg.use_wandb:
             callbacks.append(
                 SaveAxolotlConfigtoWandBCallback(self.cfg.axolotl_config_path)
@@ -246,9 +254,6 @@ class HFCausalTrainerBuilder(TrainerBuilderBase):
         if self.cfg.loss_watchdog_threshold is not None:
             callbacks.append(LossWatchDogCallback(self.cfg))
 
-        if self.cfg.gc_steps:
-            callbacks.append(GCCallback(gc_steps=self.cfg.gc_steps))
-
         return callbacks
 
     def get_post_trainer_create_callbacks(self, trainer):
@@ -290,6 +295,10 @@ class HFCausalTrainerBuilder(TrainerBuilderBase):
         if self.cfg.lisa_step_interval and self.cfg.lisa_n_layers:
             callbacks.append(lisa_callback_factory(trainer))
 
+        if any("COLAB_" in key for key in os.environ):
+            ColabCallback = colab_inference_post_train_callback(trainer)
+            callbacks.append(ColabCallback(self.cfg))
+
         callbacks.extend(super().get_post_trainer_create_callbacks(trainer=trainer))
         return callbacks
 
@@ -344,7 +353,7 @@ class HFCausalTrainerBuilder(TrainerBuilderBase):
         training_arguments_kwargs["warmup_steps"] = warmup_steps
         training_arguments_kwargs["logging_steps"] = logging_steps
 
-        if self.cfg.seed:
+        if self.cfg.seed is not None:
             training_arguments_kwargs["seed"] = self.cfg.seed
 
         if self.cfg.gradient_checkpointing:
@@ -485,7 +494,7 @@ class HFCausalTrainerBuilder(TrainerBuilderBase):
 
         # these are all the "standard" kwargs that are def used
         training_arguments_kwargs["max_steps"] = (
-            total_num_steps if self.cfg.max_steps else -1
+            self.cfg.max_steps if self.cfg.max_steps else -1
         )
         training_arguments_kwargs["max_seq_length"] = self.cfg.sequence_len
         training_arguments_kwargs["per_device_train_batch_size"] = (
@@ -538,8 +547,6 @@ class HFCausalTrainerBuilder(TrainerBuilderBase):
         report_to = []
         if self.cfg.use_wandb:
             report_to.append("wandb")
-            if self.cfg.wandb_name:
-                training_arguments_kwargs["run_name"] = self.cfg.wandb_name
         if self.cfg.use_mlflow:
             report_to.append("mlflow")
         if self.cfg.use_tensorboard:
@@ -699,6 +706,20 @@ class HFCausalTrainerBuilder(TrainerBuilderBase):
                 optimizer_cls = ADOPT
                 adam_kwargs["decouple"] = True
                 optimizer_kwargs.update(adam_kwargs)
+            elif self.cfg.optimizer == "came_pytorch":
+                from came_pytorch import CAME
+
+                optimizer_cls = CAME
+
+                beta1 = training_arguments_kwargs.get("adam_beta1", 0.9)
+                beta2 = training_arguments_kwargs.get("adam_beta2", 0.999)
+                beta3 = training_arguments_kwargs.get("adam_beta2", 0.9999)
+                eps1 = training_arguments_kwargs.get("adam_epsilon", 1e-30)
+                eps2 = training_arguments_kwargs.get("adam_epsilon2", 1e-16)
+                adam_kwargs["betas"] = (beta1, beta2, beta3)
+                adam_kwargs["eps"] = (eps1, eps2)
+
+                optimizer_kwargs.update(adam_kwargs)
 
             # Parse any additional optimizer args from config
             if self.cfg.optim_args:
@@ -798,14 +819,15 @@ class HFCausalTrainerBuilder(TrainerBuilderBase):
         data_collator_kwargs = {
             "padding": True,  # True/"longest" is the default
         }
+        multiple = 64
         if self.cfg.pad_to_sequence_len:
-            data_collator_kwargs["pad_to_multiple_of"] = 64 * math.ceil(
-                self.cfg.sequence_len / 64
+            data_collator_kwargs["pad_to_multiple_of"] = multiple * math.ceil(
+                self.cfg.sequence_len / multiple
             )
         else:
             # A100 is best at 64, while others at 8. Let's use the larger so we don't have to check
             # https://docs.nvidia.com/deeplearning/performance/dl-performance-matrix-multiplication/index.html
-            data_collator_kwargs["pad_to_multiple_of"] = 64
+            data_collator_kwargs["pad_to_multiple_of"] = multiple
 
         if self.cfg.reward_model:
             data_collator_kwargs["max_length"] = self.cfg.sequence_len
@@ -1011,6 +1033,10 @@ class HFRLTrainerBuilder(TrainerBuilderBase):
             training_args_kwargs["dataloader_prefetch_factor"] = (
                 self.cfg.dataloader_prefetch_factor
             )
+
+        if self.cfg.seed is not None:
+            training_args_kwargs["seed"] = self.cfg.seed
+
         if self.cfg.gradient_checkpointing:
             training_args_kwargs["gradient_checkpointing"] = (
                 self.cfg.gradient_checkpointing
@@ -1034,6 +1060,8 @@ class HFRLTrainerBuilder(TrainerBuilderBase):
             # default to saving each epoch if not defined
             training_args_kwargs["save_strategy"] = "epoch"
 
+        training_args_kwargs["save_only_model"] = self.cfg.save_only_model
+
         if self.cfg.dataset_processes:
             training_args_kwargs["dataset_num_proc"] = self.cfg.dataset_processes
 
@@ -1051,9 +1079,13 @@ class HFRLTrainerBuilder(TrainerBuilderBase):
         if self.cfg.use_wandb:
             training_args_kwargs["run_name"] = self.cfg.wandb_name
 
+        training_args_kwargs["sequence_parallel_degree"] = (
+            self.cfg.sequence_parallel_degree
+        )
+
         training_args_cls = None
         blocklist_args_kwargs = []
-        if self.cfg.rl == "simpo":
+        if self.cfg.rl is RLType.SIMPO:
             training_args_cls = AxolotlCPOConfig
             training_args_kwargs["loss_type"] = "simpo"
             training_args_kwargs["max_length"] = self.cfg.sequence_len
@@ -1061,13 +1093,13 @@ class HFRLTrainerBuilder(TrainerBuilderBase):
             if self.cfg.cpo_alpha is not None:
                 training_args_kwargs["cpo_alpha"] = self.cfg.cpo_alpha
 
-        elif self.cfg.rl == "orpo":
+        elif self.cfg.rl is RLType.ORPO:
             training_args_cls = AxolotlORPOConfig
             training_args_kwargs["max_length"] = self.cfg.sequence_len
             if self.cfg.max_prompt_len:
                 training_args_kwargs["max_prompt_length"] = self.cfg.max_prompt_len
 
-        elif self.cfg.rl == "kto":
+        elif self.cfg.rl is RLType.KTO:
             training_args_cls = AxolotlKTOConfig
 
             training_args_kwargs["desirable_weight"] = (
@@ -1081,14 +1113,14 @@ class HFRLTrainerBuilder(TrainerBuilderBase):
             if self.cfg.max_prompt_len:
                 training_args_kwargs["max_prompt_length"] = self.cfg.max_prompt_len
 
-        elif self.cfg.rl == "grpo":
+        elif self.cfg.rl is RLType.GRPO:
             training_args_cls = GRPOStrategy.get_training_args_class()
             training_args_kwargs.update(GRPOStrategy.set_training_args_kwargs(self.cfg))
             blocklist_args_kwargs = GRPOStrategy.get_blocklist_args_kwargs()
 
         else:
             training_args_cls = AxolotlDPOConfig
-            if self.cfg.rl == "ipo":
+            if self.cfg.rl is RLType.IPO:
                 training_args_kwargs["loss_type"] = "ipo"
             training_args_kwargs["max_length"] = self.cfg.sequence_len
             training_args_kwargs["max_completion_length"] = None
@@ -1131,67 +1163,73 @@ class HFRLTrainerBuilder(TrainerBuilderBase):
 
     def build(self, total_num_steps):
         training_args = self.build_training_arguments(total_num_steps)
-        dpo_trainer_kwargs = {}
-        if self.cfg.rl == "ipo":
+        trainer_kwargs = {}
+        if self.cfg.rl is RLType.IPO:
             if self.cfg.dpo_label_smoothing:
-                dpo_trainer_kwargs["label_smoothing"] = self.cfg.dpo_label_smoothing
+                trainer_kwargs["label_smoothing"] = self.cfg.dpo_label_smoothing
         if self.eval_dataset:
-            dpo_trainer_kwargs["eval_dataset"] = self.eval_dataset
+            trainer_kwargs["eval_dataset"] = self.eval_dataset
         if self.cfg.adapter and self.peft_config:
-            dpo_trainer_kwargs["peft_config"] = self.peft_config
+            trainer_kwargs["peft_config"] = self.peft_config
         if self.cfg.precompute_ref_log_probs is not None:
-            dpo_trainer_kwargs["precompute_ref_log_probs"] = (
+            trainer_kwargs["precompute_ref_log_probs"] = (
                 self.cfg.precompute_ref_log_probs
             )
-        if self.cfg.rl == "grpo":
-            trainer_cls = GRPOStrategy.get_trainer_class()
+        if self.cfg.rl is RLType.GRPO:
+            trainer_cls = GRPOStrategy.get_trainer_class(
+                sequence_parallel=self.cfg.sequence_parallel_degree > 1
+            )
             trainer_cls_args = [self.model]
             trainer_cls_args.extend(GRPOStrategy.set_trainer_args(self.cfg))
-            dpo_trainer_kwargs.update(GRPOStrategy.set_trainer_kwargs(self.cfg))
-        elif self.cfg.rl in ["dpo", "ipo"]:
+            trainer_kwargs.update(GRPOStrategy.set_trainer_kwargs(self.cfg))
+        elif self.cfg.rl in [RLType.DPO, RLType.IPO]:
             trainer_cls = DPOStrategy.get_trainer_class()
             trainer_cls_args = [self.model, self.model_ref]
-        elif self.cfg.rl == "orpo":
+        elif self.cfg.rl is RLType.ORPO:
             trainer_cls = AxolotlORPOTrainer
             trainer_cls_args = [self.model]
-        elif self.cfg.rl in ["kto"]:
+        elif self.cfg.rl is RLType.KTO:
             trainer_cls = AxolotlKTOTrainer
             trainer_cls_args = [self.model]
-        elif self.cfg.rl in ["simpo"]:
+        elif self.cfg.rl is RLType.SIMPO:
             trainer_cls = AxolotlCPOTrainer
             trainer_cls_args = [self.model]
         else:
             raise ValueError(f"Unsupported RL: {self.cfg.rl}")
 
+        if self.cfg.plugins:
+            plugin_manager = PluginManager.get_instance()
+            trainer_cls = plugin_manager.get_trainer_cls(self.cfg)
+
         sig = inspect.signature(trainer_cls)
         if "tokenizer" in sig.parameters.keys():
-            dpo_trainer_kwargs["tokenizer"] = self.tokenizer
+            trainer_kwargs["tokenizer"] = self.tokenizer
         else:
-            dpo_trainer_kwargs["processing_class"] = self.tokenizer
+            trainer_kwargs["processing_class"] = self.tokenizer
 
         if self.cfg.datasets is not None and (
             trainer_cls is DPOStrategy.get_trainer_class()
         ):
-            dpo_trainer_kwargs["dataset_tags"] = [
+            trainer_kwargs["dataset_tags"] = [
                 d["path"] for d in self.cfg.datasets if not Path(d["path"]).is_dir()
             ]
-        dpo_trainer = trainer_cls(
+        trainer = trainer_cls(
             *trainer_cls_args,
             args=training_args,
             train_dataset=self.train_dataset,
             callbacks=self.get_callbacks(),
-            **dpo_trainer_kwargs,
+            **trainer_kwargs,
         )
         if self.cfg.fsdp:
-            ensure_dtype(dpo_trainer.model, dtype=self.cfg.torch_dtype)
-            if self.cfg.rl in ["dpo", "ipo"] and dpo_trainer.ref_model:
-                ensure_dtype(dpo_trainer.ref_model, dtype=self.cfg.torch_dtype)
+            ensure_dtype(trainer.model, dtype=self.cfg.torch_dtype)
+            if self.cfg.rl in [RLType.DPO, RLType.IPO] and trainer.ref_model:
+                ensure_dtype(trainer.ref_model, dtype=self.cfg.torch_dtype)
 
-        dpo_trainer = self.hook_post_create_trainer(dpo_trainer)
-        for callback in self.get_post_trainer_create_callbacks(dpo_trainer):
-            dpo_trainer.add_callback(callback)
+        trainer = self.hook_post_create_trainer(trainer)
+        for callback in self.get_post_trainer_create_callbacks(trainer):
+            trainer.add_callback(callback)
 
-        return dpo_trainer
+        return trainer
 
 
 class HFPPOTrainerBuilder(TrainerBuilderBase):

src/axolotl/core/trainers/__init__.py

@@ -5,7 +5,7 @@
 
 from .base import AxolotlTrainer
 from .dpo.trainer import AxolotlDPOTrainer
-from .grpo.trainer import AxolotlGRPOTrainer
+from .grpo.trainer import AxolotlGRPOSequenceParallelTrainer, AxolotlGRPOTrainer
 from .mamba import AxolotlMambaTrainer
 from .relora import ReLoRATrainer
 from .trl import (

src/axolotl/core/trainers/base.py

@@ -114,6 +114,8 @@ class AxolotlTrainer(
             packing_efficiency_estimate=self.args.sample_packing_efficiency,
             batch_max_len=batch_max_len,
             batch_size=batch_size,
+            group_size=self.args.sample_packing_group_size,
+            bin_size=self.args.sample_packing_bin_size,
             sequential=self.args.sample_packing_sequentially,
             drop_last=True,
         )
@@ -371,15 +373,13 @@ class AxolotlTrainer(
                 num_items_in_batch=num_items_in_batch,
             )
 
-        loss = super().compute_loss(
+        return super().compute_loss(
             model,
             inputs,
             return_outputs=return_outputs,
             num_items_in_batch=num_items_in_batch,
         )
 
-        return loss
-
     @staticmethod
     def orpo_concatenate_inputs(inputs, label_pad_token=-100, pad_token=0, device=None):
         concatenated_batch = {}

src/axolotl/core/trainers/dpo/__init__.py

@@ -1,14 +1,11 @@
-"""
-DPO Specific Strategy for training
-"""
+"""DPO Specific Strategy for training"""
 
 from axolotl.core.trainers.dpo.trainer import AxolotlDPOTrainer
+from axolotl.utils.schemas.enums import RLType
 
 
 class DPOStrategy:
-    """
-    Strategy for DPO training
-    """
+    """Strategy for DPO training"""
 
     @classmethod
     def get_trainer_class(cls):
@@ -23,7 +20,7 @@ class DPOStrategy:
     @classmethod
     def set_training_args_kwargs(cls, cfg):
         training_args_kwargs = {}
-        if cfg.rl == "ipo":
+        if cfg.rl is RLType.IPO:
             training_args_kwargs["loss_type"] = "ipo"
         training_args_kwargs["max_length"] = cfg.sequence_len
         training_args_kwargs["max_completion_length"] = None

src/axolotl/core/trainers/dpo/trainer.py

@@ -3,15 +3,29 @@ DPO trainer for axolotl
 """
 
 import gc
+import random
 from functools import wraps
-from typing import Any, Dict, Union
+from typing import Any, Dict, Optional, Union
 
+import pandas as pd
 import torch
+import wandb
+from accelerate import PartialState
+from datasets import Dataset, IterableDataset
 from peft.optimizers import create_loraplus_optimizer
 from torch import nn
-from transformers import Trainer
+from torch.utils.data import DataLoader
+from transformers import (
+    BaseImageProcessor,
+    FeatureExtractionMixin,
+    PreTrainedTokenizerBase,
+    ProcessorMixin,
+    Trainer,
+)
+from transformers.trainer_utils import EvalLoopOutput
 from transformers.utils import is_sagemaker_mp_enabled
-from trl import DPOTrainer
+from trl import DPOConfig, DPOTrainer, maybe_apply_chat_template, maybe_extract_prompt
+from trl.trainer.utils import log_table_to_comet_experiment
 
 from axolotl.core.trainers.mixins import RngLoaderMixin, SchedulerMixin
 from axolotl.core.trainers.utils import (
@@ -81,6 +95,64 @@ class AxolotlDPOTrainer(RngLoaderMixin, SchedulerMixin, DPOTrainer):
 
         return super().push_to_hub(*args, **kwargs)
 
+    # TODO: remove this once https://github.com/huggingface/trl/pull/3377 is in a release
+    def _prepare_dataset(
+        self,
+        dataset: Union[Dataset, IterableDataset],
+        processing_class: Union[
+            PreTrainedTokenizerBase,
+            BaseImageProcessor,
+            FeatureExtractionMixin,
+            ProcessorMixin,
+        ],
+        args: DPOConfig,
+        dataset_name: str,
+    ) -> Union[Dataset, IterableDataset]:
+        # Build the kwargs for the `map` function
+        map_kwargs: Dict[str, Any] = {"writer_batch_size": 10}
+        if isinstance(dataset, Dataset):  # IterableDataset does not support num_proc
+            map_kwargs["num_proc"] = args.dataset_num_proc
+
+        with PartialState().main_process_first():
+            # Extract prompt if needed
+            if isinstance(
+                dataset, Dataset
+            ):  # `IterableDataset.map` does not support `desc`
+                map_kwargs["desc"] = f"Extracting prompt in {dataset_name} dataset"
+            dataset = dataset.map(maybe_extract_prompt, **map_kwargs)
+
+            # Apply the chat template if needed
+            if isinstance(
+                dataset, Dataset
+            ):  # `IterableDataset.map` does not support `desc`
+                map_kwargs["desc"] = f"Applying chat template to {dataset_name} dataset"
+            dataset = dataset.map(
+                maybe_apply_chat_template,
+                fn_kwargs={"tokenizer": processing_class, "tools": args.tools},
+                **map_kwargs,
+            )
+
+            # Tokenize the dataset
+            if isinstance(
+                dataset, Dataset
+            ):  # `IterableDataset.map` does not support `desc`
+                map_kwargs["desc"] = f"Tokenizing {dataset_name} dataset"
+
+            dataset = dataset.map(
+                self.tokenize_row if not self.is_vision_model else self.process_row,
+                remove_columns=["chosen", "rejected"],
+                fn_kwargs={
+                    "processing_class": processing_class,
+                    "max_prompt_length": args.max_prompt_length,
+                    "max_completion_length": args.max_completion_length,
+                    # for enc-dec, we add the special tokens ([bos_token] + prompt + [eos_token]; completion + [eos_token])
+                    "add_special_tokens": False,
+                },
+                **map_kwargs,
+            )
+
+        return dataset
+
     @staticmethod
     def tokenize_row(
         features,
@@ -105,12 +177,8 @@ class AxolotlDPOTrainer(RngLoaderMixin, SchedulerMixin, DPOTrainer):
             # dpo trainer may incorrectly prepend the bos_token_id to the dpo outputs
             if res["chosen_input_ids"][0] == processing_class.bos_token_id:
                 res["chosen_input_ids"] = res["chosen_input_ids"][1:]
-                res["chosen_labels"] = res["chosen_labels"][1:]
-                res["chosen_attention_mask"] = res["chosen_attention_mask"][1:]
             if res["rejected_input_ids"][0] == processing_class.bos_token_id:
                 res["rejected_input_ids"] = res["rejected_input_ids"][1:]
-                res["rejected_labels"] = res["rejected_labels"][1:]
-                res["rejected_attention_mask"] = res["rejected_attention_mask"][1:]
 
         return res
 
@@ -124,3 +192,69 @@ class AxolotlDPOTrainer(RngLoaderMixin, SchedulerMixin, DPOTrainer):
         gc.collect()
         torch.cuda.empty_cache()
         return loss
+
+    # TODO: remove this once https://github.com/huggingface/trl/pull/3377 is in a release
+    def evaluation_loop(
+        self,
+        dataloader: DataLoader,
+        description: str,
+        prediction_loss_only: Optional[bool] = None,
+        ignore_keys: Optional[list[str]] = None,
+        metric_key_prefix: str = "eval",
+    ) -> EvalLoopOutput:
+        """
+        Overriding built-in evaluation loop to store metrics for each batch.
+        Prediction/evaluation loop, shared by `Trainer.evaluate()` and `Trainer.predict()`.
+
+        Works both with or without labels.
+        """
+
+        # Sample and save to game log if requested (for one batch to save time)
+        if self.generate_during_eval:
+            # Generate random indices within the range of the total number of samples
+            num_samples = len(dataloader.dataset)
+            random_indices = random.sample(
+                range(num_samples), k=self.args.eval_batch_size
+            )
+
+            # Use dataloader.dataset.select to get the random batch without iterating over the DataLoader
+            random_batch_dataset = dataloader.dataset.select(random_indices)
+            random_batch = self.data_collator(random_batch_dataset)
+            random_batch = self._prepare_inputs(random_batch)
+
+            policy_output_decoded, ref_output_decoded = (
+                self.generate_from_model_and_ref(self.model, random_batch)
+            )
+
+            table = pd.DataFrame(
+                columns=["Prompt", "Policy", "Ref Model"],
+                data=[
+                    [prompt, pol[len(prompt) :], ref[len(prompt) :]]
+                    for prompt, pol, ref in zip(
+                        random_batch_dataset["prompt"],
+                        policy_output_decoded,
+                        ref_output_decoded,
+                    )
+                ],
+            )
+            if "wandb" in self.args.report_to and self.accelerator.is_main_process:
+                wandb.log({"game_log": wandb.Table(data=table)})
+
+            if "comet_ml" in self.args.report_to:
+                log_table_to_comet_experiment(
+                    name="game_log.csv",
+                    table=table,
+                )
+
+        # Base evaluation
+        initial_output = super(  # pylint: disable=bad-super-call
+            DPOTrainer, self
+        ).evaluation_loop(
+            dataloader,
+            description,
+            prediction_loss_only,
+            ignore_keys,
+            metric_key_prefix,
+        )
+
+        return initial_output

src/axolotl/core/trainers/grpo/__init__.py

@@ -1,37 +1,41 @@
-"""
-GRPO Specific Strategy for training
-"""
+"""GRPO Specific Strategy for training"""
 
 import importlib
 import inspect
 import logging
+from typing import Any
 
 from trl.trainer.grpo_trainer import RewardFunc
 
-from axolotl.core.trainers.grpo.trainer import AxolotlGRPOTrainer
+from axolotl.core.trainers.grpo.args import AxolotlGRPOConfig
+from axolotl.core.trainers.grpo.trainer import (
+    AxolotlGRPOSequenceParallelTrainer,
+    AxolotlGRPOTrainer,
+)
+from axolotl.utils.dict import DictDefault
 from axolotl.utils.schemas.trl import TRLConfig
 
-LOG = logging.getLogger("axolotl")
+LOG = logging.getLogger(__name__)
 
 
 class GRPOStrategy:
-    """
-    Strategy for GRPO training
-    """
+    """Strategy for GRPO training"""
 
     @classmethod
-    def get_trainer_class(cls):
+    def get_trainer_class(
+        cls, sequence_parallel: bool
+    ) -> type[AxolotlGRPOTrainer] | type[AxolotlGRPOSequenceParallelTrainer]:
+        if sequence_parallel:
+            return AxolotlGRPOSequenceParallelTrainer
         return AxolotlGRPOTrainer
 
     @classmethod
-    def get_training_args_class(cls):
-        from axolotl.core.trainers.grpo.args import AxolotlGRPOConfig
-
+    def get_training_args_class(cls) -> type[AxolotlGRPOConfig]:
         return AxolotlGRPOConfig
 
     @classmethod
-    def set_training_args_kwargs(cls, cfg):
-        grpo_args_kwargs = {}
+    def set_training_args_kwargs(cls, cfg: DictDefault) -> dict[str, Any]:
+        grpo_args_kwargs: dict[str, Any] = {}
 
         if not hasattr(cfg, "trl") or not cfg.trl:
             return grpo_args_kwargs
@@ -40,8 +44,8 @@ class GRPOStrategy:
 
         if trl.use_vllm:
             grpo_args_kwargs["use_vllm"] = trl.use_vllm
-            grpo_args_kwargs["vllm_server_host"] = trl.vllm_server_host or trl.vllm.host
-            grpo_args_kwargs["vllm_server_port"] = trl.vllm_server_port or trl.vllm.port
+            grpo_args_kwargs["vllm_server_host"] = trl.vllm_server_host or trl.vllm.host  # type: ignore[attr-defined]
+            grpo_args_kwargs["vllm_server_port"] = trl.vllm_server_port or trl.vllm.port  # type: ignore[attr-defined]
             if trl.vllm_server_timeout:
                 grpo_args_kwargs["vllm_server_timeout"] = trl.vllm_server_timeout
             if trl.vllm_guided_decoding_regex:
@@ -63,6 +67,7 @@ class GRPOStrategy:
 
         grpo_args_kwargs["max_completion_length"] = trl.max_completion_length
         grpo_args_kwargs["log_completions"] = trl.log_completions
+        grpo_args_kwargs["num_completions_to_print"] = trl.num_completions_to_print
 
         if trl.reward_weights:
             grpo_args_kwargs["reward_weights"] = trl.reward_weights
@@ -70,6 +75,13 @@ class GRPOStrategy:
         if trl.scale_rewards is not None:
             grpo_args_kwargs["scale_rewards"] = trl.scale_rewards
 
+        if trl.loss_type is not None:
+            grpo_args_kwargs["loss_type"] = trl.loss_type
+        if trl.mask_truncated_completions is not None:
+            grpo_args_kwargs["mask_truncated_completions"] = (
+                trl.mask_truncated_completions
+            )
+
         if trl.temperature is not None:
             grpo_args_kwargs["temperature"] = trl.temperature
         if trl.top_p is not None:
@@ -85,21 +97,27 @@ class GRPOStrategy:
             grpo_args_kwargs["num_iterations"] = trl.num_iterations
         if trl.epsilon is not None:
             grpo_args_kwargs["epsilon"] = trl.epsilon
+        if trl.epsilon_high is not None:
+            grpo_args_kwargs["epsilon_high"] = trl.epsilon_high
+
+        if trl.use_liger_loss is not None:
+            grpo_args_kwargs["use_liger_loss"] = trl.use_liger_loss
 
         return grpo_args_kwargs
 
     @classmethod
-    def set_trainer_args(cls, cfg):
+    def set_trainer_args(cls, cfg: DictDefault) -> list[Any]:
         trainer_args = []
         if cfg.trl and cfg.trl.reward_funcs:
             reward_funcs = []
             for reward_func_fqn in cfg.trl.reward_funcs:
                 reward_funcs.append(cls.get_reward_func(reward_func_fqn))
             trainer_args.append(reward_funcs)
+
         return trainer_args
 
     @classmethod
-    def set_trainer_kwargs(cls, cfg):
+    def set_trainer_kwargs(cls, cfg: DictDefault) -> dict[str, Any]:
         trainer_kwargs = {}
         if cfg.trl and cfg.trl.reward_processing_classes:
             trainer_kwargs["reward_processing_classes"] = (
@@ -113,7 +131,7 @@ class GRPOStrategy:
         return None
 
     @classmethod
-    def get_blocklist_args_kwargs(cls):
+    def get_blocklist_args_kwargs(cls) -> list[str]:
         return ["dataset_num_proc"]
 
     @classmethod
@@ -124,18 +142,20 @@ class GRPOStrategy:
         Args:
             reward_func_fqn (str): Fully qualified name of the reward function (e.g. r1_grpo.gsm8k_transform),
                 or a HF hub path to the reward model.
-        Raises:
-            ValueError: If the reward function does not accept at least two arguments.
 
         Returns:
             RewardFunc: A callable that accepts prompts and completions and returns rewards,
                 or a path to a reward model.
 
+        Raises:
+            ValueError: If the reward function does not accept at least two arguments.
         """
         try:
             # use importlib to dynamically load the reward function from the module
             reward_func_module_name = reward_func_fqn.split(".")[-1]
-            reward_func_module = importlib.import_module(reward_func_fqn.split(".")[-2])
+            reward_func_module = importlib.import_module(
+                ".".join(reward_func_fqn.split(".")[:-1])
+            )
             reward_func = getattr(reward_func_module, reward_func_module_name)
             if not len(inspect.signature(reward_func).parameters) >= 2:
                 raise ValueError(

src/axolotl/core/trainers/grpo/args.py

@@ -11,6 +11,4 @@ from axolotl.core.training_args import AxolotlTrainingMixins
 
 @dataclass
 class AxolotlGRPOConfig(AxolotlTrainingMixins, GRPOConfig):
-    """
-    Axolotl GRPO Config for GRPO training
-    """
+    """Axolotl GRPO Config for GRPO training"""

src/axolotl/core/trainers/grpo/sampler.py

@@ -0,0 +1,172 @@
+"""Repeat random sampler (similar to the one implemented in
+https://github.com/huggingface/trl/blob/main/trl/trainer/grpo_trainer.py) that adds
+sequence parallelism functionality; i.e., duplicating data across ranks in the same
+sequence parallel group.
+"""
+
+from typing import Iterator, Sized
+
+import torch
+from torch.utils.data import Sampler
+
+
+class SequenceParallelRepeatRandomSampler(Sampler):
+    """Sampler for GRPO training with sequence parallelism.
+
+    This sampler ensures:
+    - Ranks in the same sequence parallel (SP) group receive identical data.
+    - Each index is repeated multiple times for sampling different completions.
+    - Entire batches are repeated for reuse in multiple updates.
+    - Data is properly distributed across SP groups.
+
+    In the table below, the values represent dataset indices. Each SP group has
+    `sequence_parallel_degree = 2` GPUs working together on the same data. There are 2
+    SP groups (SP0 and SP1), with `world_size = 4` total GPUs.
+
+                                               Sequence Parallel Groups
+                                        |       SP0        |       SP1        |
+                                        |  GPU 0  |  GPU 1 |  GPU 2  |  GPU 3 |
+                    global_step  step    <---> mini_repeat_count=3
+                                            <----------> batch_size=2 per SP group
+    grad_accum=2   ▲  ▲  0       0         [0 0 0  1 1 1]     [2 2 2  3 3 3]   <- SP groups get different data
+                   ▼  |  0       1         [0 0 0  1 1 1]     [2 2 2  3 3 3]   <- Same data for each SP group GPU
+                      |
+                      |  1       2         [0 0 0  1 1 1]     [2 2 2  3 3 3]   <- Repeat same indices for iterations
+    num_iterations=2  ▼  1       3         [0 0 0  1 1 1]     [2 2 2  3 3 3]   <- When using gradient accumulation
+
+                         2       4         [4 4 4  5 5 5]     [6 6 6  7 7 7]   <- New batch of data indices
+                         2       5         [4 4 4  5 5 5]     [6 6 6  7 7 7]
+                                            ...
+
+    Args:
+        dataset: Dataset to sample from.
+        mini_repeat_count: How many times to repeat each sample immediately.
+        world_size: Total number of processes.
+        rank: Rank of current process.
+        batch_size: Number of samples per batch.
+        repeat_count: How many times to repeat the full sampling process.
+        sequence_parallel_degree: Number of ranks in a sequence parallel group.
+        shuffle: Whether to shuffle the dataset.
+        seed: Random seed for shuffling.
+        drop_last: Whether to drop the last incomplete batch.
+    """
+
+    def __init__(
+        self,
+        dataset: Sized,
+        mini_repeat_count: int,
+        world_size: int,
+        rank: int,
+        batch_size: int = 1,
+        repeat_count: int = 1,
+        sequence_parallel_degree: int = 1,
+        shuffle: bool = True,
+        seed: int = 0,
+        drop_last: bool = False,
+    ):
+        self.dataset = dataset
+        self.mini_repeat_count = mini_repeat_count
+        self.batch_size = batch_size
+        self.repeat_count = repeat_count
+        self.shuffle = shuffle
+        self.seed = seed
+        self.drop_last = drop_last
+        self.epoch = 0
+
+        self.world_size = world_size
+        self.rank = rank
+
+        # Sequence parallelism parameters
+        self.sequence_parallel_degree = sequence_parallel_degree
+        self.num_sp_groups = world_size // sequence_parallel_degree
+        self.sp_group_id = rank // sequence_parallel_degree
+
+        # Adjust dataset size for distributed sampling
+        self.num_samples = len(self.dataset)
+        self.total_size = self.num_samples
+
+        # Calculate effective number of samples per SP group
+        if (
+            self.drop_last
+            and self.total_size % (self.num_sp_groups * self.batch_size) != 0
+        ):
+            # Drop last incomplete batch if drop_last is True
+            self.num_samples_per_sp_group = (
+                self.total_size // self.batch_size // self.num_sp_groups
+            ) * self.batch_size
+        else:
+            # Round up to include last batch if drop_last is False
+            self.num_samples_per_sp_group = (
+                (self.total_size + self.batch_size * self.num_sp_groups - 1)
+                // (self.batch_size * self.num_sp_groups)
+                * self.batch_size
+            )
+
+        if shuffle:
+            self.generator = torch.Generator()
+            self.generator.manual_seed(seed)
+
+    def __iter__(self) -> Iterator[int]:
+        """Creates iterator over dataset indices.
+
+        Returns:
+            Iterator that yields indices into the dataset.
+        """
+        # Deterministically shuffle based on epoch and seed
+        if self.shuffle:
+            indices = torch.randperm(
+                self.num_samples, generator=self.generator
+            ).tolist()
+        else:
+            indices = list(range(self.num_samples))
+
+        # Add extra samples to make it evenly divisible by batch_size
+        if len(indices) % self.batch_size != 0:
+            padding = indices[: self.batch_size - len(indices) % self.batch_size]
+            indices += padding
+
+        # Subsample based on SP group ID
+        # Each SP group gets distinct batches of data
+        batch_indices = []
+        for i in range(0, len(indices), self.batch_size * self.num_sp_groups):
+            start_idx = i + self.sp_group_id * self.batch_size
+            end_idx = min(start_idx + self.batch_size, len(indices))
+            if start_idx < len(indices):
+                for j in range(self.batch_size):
+                    if start_idx + j < end_idx:
+                        batch_indices.append(indices[start_idx + j])
+
+        # Make sure batch_indices is exactly batch_size * num_batches_per_sp_group
+        if self.drop_last:
+            num_batches_per_sp_group = self.num_samples_per_sp_group // self.batch_size
+            target_len = self.batch_size * num_batches_per_sp_group
+            if len(batch_indices) > target_len:
+                batch_indices = batch_indices[:target_len]
+
+        # Apply the GRPO repeat pattern
+        final_indices = []
+        for _ in range(self.repeat_count):
+            for idx in batch_indices:
+                for _ in range(self.mini_repeat_count):
+                    final_indices.append(idx)
+
+        return iter(final_indices)
+
+    def __len__(self) -> int:
+        """Returns the total length of the iterable including repetitions.
+
+        Returns:
+            Total number of samples.
+        """
+        # Total length including all repetitions
+        return (
+            self.num_samples_per_sp_group * self.mini_repeat_count * self.repeat_count
+        )
+
+    def set_epoch(self, epoch: int) -> None:
+        """Sets the epoch for this sampler.
+
+        Args:
+            epoch: Epoch number to use for shuffling.
+        """
+        self.epoch = epoch

src/axolotl/core/trainers/grpo/trainer.py

@@ -1,23 +1,63 @@
-"""
-Axolotl GRPO trainer
-"""
+"""Axolotl GRPO trainers (with and without sequence parallelism handling)"""
 
+# pylint: disable=too-many-lines,duplicate-code,protected-access,no-member
+
+import warnings
 from contextlib import nullcontext
+from typing import Any
 
-from accelerate.utils import is_deepspeed_available, is_peft_model
+import datasets
+import torch
+import torch.distributed as dist
+import torch.utils.data
+from accelerate.utils import (
+    broadcast_object_list,
+    gather,
+    gather_object,
+    is_peft_model,
+)
+from datasets import Dataset, IterableDataset
+from torch import nn
+from torch.utils.data import (
+    BatchSampler,
+    DataLoader,
+    Sampler,
+)
+from transformers import (
+    PreTrainedModel,
+    PreTrainedTokenizerBase,
+    Trainer,
+    TrainerCallback,
+)
+from transformers.trainer_utils import seed_worker
+from transformers.utils import is_peft_available
 from trl import GRPOTrainer
-from trl.extras.profiling import profiling_decorator
+from trl.data_utils import (
+    apply_chat_template,
+    is_conversational,
+    maybe_apply_chat_template,
+)
+from trl.extras.profiling import profiling_context, profiling_decorator
+from trl.import_utils import is_deepspeed_available
+from trl.models import unwrap_model_for_generation
+from trl.trainer.grpo_config import GRPOConfig
+from trl.trainer.grpo_trainer import RewardFunc, nanstd
+from trl.trainer.utils import pad
 
+from axolotl.core.trainers.grpo.sampler import SequenceParallelRepeatRandomSampler
 from axolotl.core.trainers.mixins import RngLoaderMixin, SchedulerMixin
+from axolotl.monkeypatch.attention.ring_attn.patch import get_ring_attn_group
+
+if is_peft_available():
+    # pylint: disable=unused-import
+    from peft import PeftConfig
 
 if is_deepspeed_available():
     import deepspeed
 
 
 class AxolotlGRPOTrainer(RngLoaderMixin, SchedulerMixin, GRPOTrainer):
-    """
-    Extend the base GRPOTrainer for axolotl helpers
-    """
+    """Extend the base GRPOTrainer for axolotl helpers"""
 
     _tag_names = ["trl", "grpo", "axolotl"]
 
@@ -67,3 +107,600 @@ class AxolotlGRPOTrainer(RngLoaderMixin, SchedulerMixin, GRPOTrainer):
         # Reset cache on main process
         if self.accelerator.is_main_process:
             self.vllm_client.reset_prefix_cache()
+
+
+class AxolotlGRPOSequenceParallelTrainer(AxolotlGRPOTrainer):
+    """Extend the base GRPOTrainer for sequence parallelism handling"""
+
+    def __init__(
+        self,
+        model: str | PreTrainedModel,
+        reward_funcs: RewardFunc | list[RewardFunc],
+        args: GRPOConfig | None = None,
+        train_dataset: Dataset | IterableDataset | None = None,
+        eval_dataset: (
+            Dataset | IterableDataset | dict[str, Dataset | IterableDataset] | None
+        ) = None,
+        processing_class: PreTrainedTokenizerBase | None = None,
+        reward_processing_classes: (
+            PreTrainedTokenizerBase | list[PreTrainedTokenizerBase] | None
+        ) = None,
+        callbacks: list[TrainerCallback] | None = None,
+        optimizers: tuple[
+            torch.optim.Optimizer | None, torch.optim.lr_scheduler.LambdaLR | None
+        ] = (None, None),
+        peft_config: "PeftConfig | None" = None,
+    ):
+        # First call the superclass constructor with all arguments
+        super().__init__(
+            model=model,
+            reward_funcs=reward_funcs,
+            args=args,
+            train_dataset=train_dataset,
+            eval_dataset=eval_dataset,
+            processing_class=processing_class,
+            reward_processing_classes=reward_processing_classes,
+            callbacks=callbacks,
+            optimizers=optimizers,
+            peft_config=peft_config,
+        )
+
+        # Get number of SP groups (number of processes divided by SP degree)
+        num_processes = self.accelerator.num_processes
+        num_sp_groups = num_processes // self.args.sequence_parallel_degree
+
+        # Calculate batch size per SP group (not per process)
+        sp_group_batch_size = self.args.per_device_train_batch_size * num_sp_groups
+        possible_values = [
+            n_gen
+            for n_gen in range(2, sp_group_batch_size + 1)
+            if (sp_group_batch_size) % n_gen == 0
+        ]
+
+        if self.num_generations not in possible_values:
+            raise ValueError(
+                f"The batch size per SP group ({num_sp_groups} x "
+                f"{self.args.per_device_train_batch_size}) must be evenly divisible by "
+                f"the number of generations per prompt ({self.num_generations}). Given "
+                "the current configuration, the valid values for the number of "
+                f"generations are: {possible_values}."
+            )
+
+        if self.args.eval_strategy != "no":
+            # If sequence parallelism is enabled, calculate batch size per SP group
+            sp_group_eval_batch_size = args.per_device_eval_batch_size * num_sp_groups  # type: ignore[union-attr]
+            possible_values = [
+                n_gen
+                for n_gen in range(2, sp_group_eval_batch_size + 1)
+                if (sp_group_eval_batch_size) % n_gen == 0
+            ]
+
+            if self.num_generations not in possible_values:
+                raise ValueError(
+                    f"With sequence parallelism (degree {self.args.sequence_parallel_degree}), "
+                    f"the eval batch size per SP group ({num_sp_groups} x {self.args.per_device_eval_batch_size}) "
+                    f"must be evenly divisible by the number of generations per prompt "
+                    f"({self.num_generations}). Given the current eval batch size, "
+                    f"the valid values for the number of generations are: {possible_values}."
+                )
+
+        # Initialize the SP group
+        self.sp_group = get_ring_attn_group()
+        self.rank = dist.get_rank()
+        self.world_size = dist.get_world_size()
+        self.local_rank = dist.get_rank(group=self.sp_group)
+        self.local_world_size = dist.get_world_size(group=self.sp_group)
+
+    def _get_train_sampler(self) -> Sampler:
+        effective_batch_size = (
+            self.args.per_device_train_batch_size
+            * self.world_size
+            * self.args.gradient_accumulation_steps
+        )
+
+        return SequenceParallelRepeatRandomSampler(
+            dataset=self.train_dataset,
+            mini_repeat_count=self.num_generations,
+            world_size=self.world_size,
+            rank=self.rank,
+            batch_size=effective_batch_size
+            // self.num_generations
+            // self.args.sequence_parallel_degree,
+            repeat_count=self.num_iterations * self.args.gradient_accumulation_steps,
+            sequence_parallel_degree=self.args.sequence_parallel_degree,
+            shuffle=True,
+            seed=self.args.seed,
+            drop_last=True,
+        )
+
+    def _create_dataloader_params(self, is_eval=False, custom_batch_size=None):
+        """Create common dataloader parameters for train or eval."""
+        batch_size = custom_batch_size or (
+            self.args.eval_batch_size if is_eval else self._train_batch_size
+        )
+
+        params = {
+            "batch_size": batch_size,
+            "collate_fn": self.data_collator,
+            "num_workers": self.args.dataloader_num_workers,
+            "pin_memory": self.args.dataloader_pin_memory,
+        }
+
+        # Add persistent workers only for training
+        if not is_eval and hasattr(self.args, "dataloader_persistent_workers"):
+            params["persistent_workers"] = self.args.dataloader_persistent_workers
+
+        # Add prefetch factor if specified
+        if self.args.dataloader_prefetch_factor:
+            params["prefetch_factor"] = self.args.dataloader_prefetch_factor
+
+        return params
+
+    def _prepare_dataloader(
+        self, dataset, sampler, is_eval=False, custom_batch_size=None
+    ):
+        """Prepare a dataloader with the given dataset and sampler."""
+        # Get base parameters
+        dataloader_params = self._create_dataloader_params(is_eval, custom_batch_size)
+
+        # Add sampler configuration
+        if not isinstance(dataset, torch.utils.data.IterableDataset):
+            if isinstance(sampler, BatchSampler):
+                # batch_size and batch_sampler are mutually exclusive
+                dataloader_params["batch_sampler"] = sampler
+                del dataloader_params["batch_size"]
+            else:
+                dataloader_params["sampler"] = sampler
+                dataloader_params["drop_last"] = self.args.dataloader_drop_last
+
+            if not is_eval:
+                dataloader_params["worker_init_fn"] = seed_worker
+
+        # Create the dataloader
+        dataloader = DataLoader(dataset, **dataloader_params)
+
+        if self.args.sample_packing and (
+            (not is_eval and not self.args.pretraining)
+            or (is_eval and self.args.eval_sample_packing is not False)
+        ):
+            self.accelerator.even_batches = False
+
+        # Return unprepared dataloader if using sequence parallelism
+        # TODO(djsaunde): We might be able to use `accelerate`'s dataloader preparation
+        # if we use `dispatch_batches` and `slice_fn_for_dispatch` properly (i.e.,
+        # slice each batch along the sequence dimension).
+        if self.args.sequence_parallel_degree > 1:
+            return dataloader
+
+        # Otherwise prepare with accelerator
+        return self.accelerator.prepare_data_loader(dataloader)
+
+    def get_train_dataloader(self) -> DataLoader:
+        """Get dataloader for training"""
+        train_dataset = self.train_dataset
+        # pylint: disable=access-member-before-definition
+        data_collator = self.data_collator  # type: ignore
+
+        # Handle dataset preprocessing
+        if isinstance(train_dataset, datasets.Dataset):
+            # Add debug print before any modifications
+            if self.args.sample_packing and not self.args.pretraining:
+                train_dataset = train_dataset.remove_columns(["length"])
+            if not self.args.sample_packing or self.args.pretraining:
+                train_dataset = self._remove_unused_columns(
+                    train_dataset, description="training"
+                )
+        else:
+            self.data_collator = self._get_collator_with_removed_columns(  # pylint: disable=attribute-defined-outside-init
+                data_collator,
+                description="training",
+            )
+
+        # Get sampler and create dataloader
+        sampler = self._get_train_sampler()
+        dataloader = self._prepare_dataloader(train_dataset, sampler, is_eval=False)
+
+        return dataloader
+
+    def _generate_and_score_completions(
+        self, inputs: list[dict[str, torch.Tensor | Any]]
+    ) -> dict[str, torch.Tensor | Any]:
+        device = self.accelerator.device
+        mode = "eval" if self.control.should_evaluate else "train"
+
+        prompts = [x["prompt"] for x in inputs]
+        prompts_text = [
+            maybe_apply_chat_template(example, self.processing_class)["prompt"]
+            for example in inputs
+        ]
+        prompt_inputs = self.processing_class(
+            text=prompts_text,
+            return_tensors="pt",
+            padding=True,
+            padding_side="left",
+            add_special_tokens=False,
+        )
+        prompt_inputs = Trainer._prepare_inputs(self, prompt_inputs)
+        prompt_ids, prompt_mask = (
+            prompt_inputs["input_ids"],
+            prompt_inputs["attention_mask"],
+        )
+
+        if self.max_prompt_length is not None:
+            prompt_ids = prompt_ids[:, -self.max_prompt_length :]
+            prompt_mask = prompt_mask[:, -self.max_prompt_length :]
+
+        # Generate completions using either vLLM or regular generation
+        if self.args.use_vllm:
+            # First, have main process load weights if needed
+            # pylint: disable=access-member-before-definition
+            if self.state.global_step != self._last_loaded_step:  # type: ignore[has-type]
+                self._move_model_to_vllm()
+                # pylint: disable=attribute-defined-outside-init
+                self._last_loaded_step = self.state.global_step
+
+            # Generate completions using vLLM: gather all prompts and use them in a single call in the main process
+            all_prompts_text = gather_object(prompts_text)
+            if self.accelerator.is_main_process:
+                if self.args.sequence_parallel_degree > 1:
+                    # Calculate sequence parallel group information
+                    world_size = self.accelerator.num_processes
+                    sequence_parallel_degree = self.args.sequence_parallel_degree
+                    num_sp_groups = world_size // sequence_parallel_degree
+
+                    # Since processes in the same SP group have the same prompts, we need to ensure
+                    # we only take one copy of each prompt from each SP group
+                    ordered_set_of_prompts = []
+                    for sp_group_id in range(num_sp_groups):
+                        # Get the first process from each SP group (typically the group leader)
+                        group_leader_rank = sp_group_id * sequence_parallel_degree
+
+                        # Extract prompts from this SP group, accounting for num_generations duplicates
+                        # We only need prompts from one rank in each SP group
+                        group_prompts = all_prompts_text[
+                            group_leader_rank
+                            * len(prompts_text) : (group_leader_rank + 1)
+                            * len(prompts_text) : self.num_generations
+                        ]
+
+                        ordered_set_of_prompts.extend(group_prompts)
+                else:
+                    # Since 'prompts' contains 'num_generations' duplicates, we first take unique prompts, and generate
+                    # num_generations outputs for each one. This is faster than generating outputs for each duplicate
+                    # prompt individually.
+                    ordered_set_of_prompts = all_prompts_text[
+                        :: self.num_generations * self.args.sequence_parallel_degree
+                    ]
+
+                with profiling_context(self, "vLLM.generate"):
+                    completion_ids = self.vllm_client.generate(
+                        prompts=ordered_set_of_prompts,
+                        n=self.num_generations,
+                        repetition_penalty=self.repetition_penalty,
+                        temperature=self.temperature,
+                        top_p=self.top_p,
+                        top_k=-1 if self.top_k is None else self.top_k,
+                        min_p=0.0 if self.min_p is None else self.min_p,
+                        max_tokens=self.max_completion_length,
+                        guided_decoding_regex=self.guided_decoding_regex,
+                    )
+            else:
+                completion_ids = [None] * (
+                    len(all_prompts_text) // self.args.sequence_parallel_degree
+                )
+
+            # Broadcast the completions from the main process to all processes
+            completion_ids = broadcast_object_list(completion_ids, from_process=0)
+
+            # Determine the appropriate slice based on sequence parallelism
+            if self.args.sequence_parallel_degree > 1:
+                # Calculate SP group ID (which group of ranks this rank belongs to)
+                sp_group_id = self.accelerator.process_index // self.local_world_size
+
+                # Calculate the start index for this SP group
+                sp_group_start = sp_group_id * len(prompts) * self.local_world_size
+
+                # All ranks in the same SP group get the same data slice
+                process_slice = slice(
+                    sp_group_start,
+                    sp_group_start + len(prompts),
+                )
+                completion_ids = completion_ids[process_slice]
+            else:
+                # Original behavior for non-sequence parallel case
+                process_slice = slice(
+                    self.accelerator.process_index * len(prompts),
+                    (self.accelerator.process_index + 1) * len(prompts),
+                )
+                completion_ids = completion_ids[process_slice]
+
+            # Pad the completions, and concatenate them with the prompts
+            completion_ids = [
+                torch.tensor(ids, device=device) for ids in completion_ids
+            ]
+            completion_ids = pad(
+                completion_ids, padding_value=self.processing_class.pad_token_id
+            )
+            prompt_completion_ids = torch.cat([prompt_ids, completion_ids], dim=1)
+        else:
+            # Regular generation path
+            with unwrap_model_for_generation(
+                self.model_wrapped,
+                self.accelerator,
+                gather_deepspeed3_params=self.args.ds3_gather_for_generation,
+            ) as unwrapped_model:
+                prompt_completion_ids = unwrapped_model.generate(
+                    prompt_ids,
+                    attention_mask=prompt_mask,
+                    generation_config=self.generation_config,
+                )
+
+            # Compute prompt length and extract completion ids
+            prompt_length = prompt_ids.size(1)
+            prompt_ids = prompt_completion_ids[:, :prompt_length]
+            completion_ids = prompt_completion_ids[:, prompt_length:]
+
+        # Mask everything after the first EOS token
+        is_eos = completion_ids == self.processing_class.eos_token_id
+        eos_idx = torch.full(
+            (is_eos.size(0),), is_eos.size(1), dtype=torch.long, device=device
+        )
+        eos_idx[is_eos.any(dim=1)] = is_eos.int().argmax(dim=1)[is_eos.any(dim=1)]
+        sequence_indices = torch.arange(is_eos.size(1), device=device).expand(
+            is_eos.size(0), -1
+        )
+        completion_mask = (sequence_indices <= eos_idx.unsqueeze(1)).int()
+
+        # If mask_truncated_completions is enabled, zero out truncated completions in completion_mask
+        if self.args.mask_truncated_completions:
+            truncated_completions = ~is_eos.any(dim=1)
+            completion_mask = (
+                completion_mask * (~truncated_completions).unsqueeze(1).int()
+            )
+
+        # Concatenate prompt_mask with completion_mask for logit computation
+        attention_mask = torch.cat([prompt_mask, completion_mask], dim=1)  # (B, P+C)
+
+        logits_to_keep = completion_ids.size(
+            1
+        )  # we only need to compute the logits for the completion tokens
+        batch_size = (
+            self.args.per_device_train_batch_size
+            if mode == "train"
+            else self.args.per_device_eval_batch_size
+        )
+
+        with torch.no_grad():
+            # When using num_iterations == 1, old_per_token_logps == per_token_logps, so we can skip it's
+            # computation here, and use per_token_logps.detach() instead.
+            if self.num_iterations > 1:
+                old_per_token_logps = self._get_per_token_logps(
+                    self.model,
+                    prompt_completion_ids,
+                    attention_mask,
+                    logits_to_keep,
+                    batch_size,
+                )
+            else:
+                old_per_token_logps = None
+
+            if self.beta == 0.0:
+                ref_per_token_logps = None
+            elif self.ref_model is not None:
+                ref_per_token_logps = self._get_per_token_logps(
+                    self.ref_model,
+                    prompt_completion_ids,
+                    attention_mask,
+                    logits_to_keep,
+                    batch_size,
+                )
+            else:
+                with self.accelerator.unwrap_model(self.model).disable_adapter():
+                    ref_per_token_logps = self._get_per_token_logps(
+                        self.model,
+                        prompt_completion_ids,
+                        attention_mask,
+                        logits_to_keep,
+                        batch_size,
+                    )
+
+        # Decode the generated completions
+        completions_text = self.processing_class.batch_decode(
+            completion_ids, skip_special_tokens=True
+        )
+        if is_conversational(inputs[0]):
+            completions = []
+            for prompt, completion in zip(prompts, completions_text):
+                bootstrap = (
+                    prompt.pop()["content"] if prompt[-1]["role"] == "assistant" else ""
+                )
+                completions.append(
+                    [{"role": "assistant", "content": bootstrap + completion}]
+                )
+        else:
+            completions = completions_text
+
+        rewards_per_func = torch.zeros(
+            len(prompts), len(self.reward_funcs), device=device
+        )
+        for i, (reward_func, reward_processing_class, reward_func_name) in enumerate(
+            zip(
+                self.reward_funcs,
+                self.reward_processing_classes,
+                self.reward_func_names,
+            )
+        ):
+            with profiling_context(self, reward_func_name):
+                if isinstance(
+                    reward_func, nn.Module
+                ):  # Module instead of PretrainedModel for compat with compiled models
+                    if is_conversational(inputs[0]):
+                        messages = [
+                            {"messages": p + c} for p, c in zip(prompts, completions)
+                        ]
+                        texts = [
+                            apply_chat_template(x, reward_processing_class)["text"]
+                            for x in messages
+                        ]
+                    else:
+                        texts = [p + c for p, c in zip(prompts, completions)]
+                    reward_inputs = reward_processing_class(
+                        text=texts,
+                        return_tensors="pt",
+                        padding=True,
+                        padding_side="right",
+                        add_special_tokens=False,
+                    )
+                    reward_inputs = Trainer._prepare_inputs(self, reward_inputs)
+                    with torch.inference_mode():
+                        rewards_per_func[:, i] = reward_func(**reward_inputs).logits[
+                            :, 0
+                        ]  # Shape (B*G,)
+                else:
+                    # Repeat all input columns (but "prompt" and "completion") to match the number of generations
+                    keys = [
+                        key for key in inputs[0] if key not in ["prompt", "completion"]
+                    ]
+                    reward_kwargs = {
+                        key: [example[key] for example in inputs] for key in keys
+                    }
+                    output_reward_func = reward_func(
+                        prompts=prompts, completions=completions, **reward_kwargs
+                    )
+                    # Convert None values to NaN
+                    output_reward_func = [
+                        reward if reward is not None else torch.nan
+                        for reward in output_reward_func
+                    ]
+
+                    rewards_per_func[:, i] = torch.tensor(
+                        output_reward_func, dtype=torch.float32, device=device
+                    )
+
+        # If all reward functions return None for a given row, issue a detailed warning
+        if torch.isnan(rewards_per_func).all(dim=1).any():
+            nan_row_idx = (
+                torch.isnan(rewards_per_func).all(dim=1).nonzero(as_tuple=True)[0][0]
+            )
+            row_reward_kwargs = {
+                key: value[nan_row_idx] for key, value in reward_kwargs.items()
+            }
+            row_reward_kwargs["prompt"] = prompts[nan_row_idx]
+            row_reward_kwargs["completion"] = completions[nan_row_idx]
+            warnings.warn(
+                f"All reward functions returned None for the following kwargs: {row_reward_kwargs}. "
+                "Please ensure that at least one reward function returns a valid reward."
+            )
+
+        # Gather the reward per function: this part is crucial, because the rewards are normalized per group and the
+        # completions may be distributed across processes
+        rewards_per_func = gather(rewards_per_func)
+
+        # Apply weights to each reward function's output and sum
+        rewards = (
+            rewards_per_func * self.reward_weights.to(device).unsqueeze(0)
+        ).nansum(dim=1)
+
+        # Compute grouped-wise rewards
+        mean_grouped_rewards = rewards.view(-1, self.num_generations).mean(dim=1)
+        std_grouped_rewards = rewards.view(-1, self.num_generations).std(dim=1)
+
+        # Normalize the rewards to compute the advantages
+        mean_grouped_rewards = mean_grouped_rewards.repeat_interleave(
+            self.num_generations, dim=0
+        )
+        std_grouped_rewards = std_grouped_rewards.repeat_interleave(
+            self.num_generations, dim=0
+        )
+        advantages = rewards - mean_grouped_rewards
+        if self.args.scale_rewards:
+            advantages = advantages / (std_grouped_rewards + 1e-4)
+
+        # Slice to keep only the local part of the data
+        if self.args.sequence_parallel_degree > 1:
+            # Calculate SP group ID (which group of ranks this rank belongs to)
+            sp_group_id = self.accelerator.process_index // self.local_world_size
+
+            # Calculate the start index for this SP group
+            sp_group_start = sp_group_id * len(prompts) * self.local_world_size
+
+            # All ranks in the same SP group get the same data slice
+            process_slice = slice(
+                sp_group_start,
+                sp_group_start + len(prompts),
+            )
+        else:
+            # Original behavior for non-sequence parallel case
+            process_slice = slice(
+                self.accelerator.process_index * len(prompts),
+                (self.accelerator.process_index + 1) * len(prompts),
+            )
+        advantages = advantages[process_slice]
+
+        # Log the metrics
+        if mode == "train":
+            self._total_train_tokens += (
+                self.accelerator.gather_for_metrics(attention_mask.sum()).sum().item()
+            )
+        self._metrics[mode]["num_tokens"] = [self._total_train_tokens]
+
+        # log completion lengths, mean, min, max
+        agg_completion_mask = self.accelerator.gather_for_metrics(
+            completion_mask.sum(1)
+        )
+        self._metrics[mode]["completions/mean_length"].append(
+            agg_completion_mask.float().mean().item()
+        )
+        self._metrics[mode]["completions/min_length"].append(
+            agg_completion_mask.float().min().item()
+        )
+        self._metrics[mode]["completions/max_length"].append(
+            agg_completion_mask.float().max().item()
+        )
+
+        # identify sequences that terminated with EOS and log their lengths
+        agg_terminated_with_eos = self.accelerator.gather_for_metrics(is_eos.any(dim=1))
+        term_completion_mask = agg_completion_mask[agg_terminated_with_eos]
+        clipped_completions_ratio = 1 - len(term_completion_mask) / len(
+            agg_completion_mask
+        )
+        self._metrics[mode]["completions/clipped_ratio"].append(
+            clipped_completions_ratio
+        )
+        if len(term_completion_mask) == 0:
+            # edge case where no completed sequences are found
+            term_completion_mask = torch.zeros(1, device=device)
+        self._metrics[mode]["completions/mean_terminated_length"].append(
+            term_completion_mask.float().mean().item()
+        )
+        self._metrics[mode]["completions/min_terminated_length"].append(
+            term_completion_mask.float().min().item()
+        )
+        self._metrics[mode]["completions/max_terminated_length"].append(
+            term_completion_mask.float().max().item()
+        )
+
+        # Calculate mean reward per function, but only for samples where the function was applied (non-NaN values)
+        for i, reward_func_name in enumerate(self.reward_func_names):
+            mean_rewards = torch.nanmean(rewards_per_func[:, i]).item()
+            self._metrics[mode][f"rewards/{reward_func_name}/mean"].append(mean_rewards)
+            std_rewards = nanstd(rewards_per_func[:, i]).item()
+            self._metrics[mode][f"rewards/{reward_func_name}/std"].append(std_rewards)
+        self._metrics[mode]["reward"].append(mean_grouped_rewards.mean().item())
+        self._metrics[mode]["reward_std"].append(std_grouped_rewards.mean().item())
+
+        # Log prompt and completion texts
+        self._textual_logs["prompt"].extend(gather_object(prompts_text))
+        self._textual_logs["completion"].extend(gather_object(completions_text))
+        for i, name in enumerate(self.reward_func_names):
+            self._textual_logs["rewards"][name].extend(rewards_per_func[:, i].tolist())
+
+        return {
+            "prompt_ids": prompt_ids,
+            "prompt_mask": prompt_mask,
+            "completion_ids": completion_ids,
+            "completion_mask": completion_mask,
+            "advantages": advantages,
+            "old_per_token_logps": old_per_token_logps,
+            "ref_per_token_logps": ref_per_token_logps,
+        }

src/axolotl/core/trainers/mixins/__init__.py

@@ -6,4 +6,4 @@
 from .optimizer import OptimizerMixin
 from .rng_state_loader import RngLoaderMixin
 from .scheduler import SchedulerMixin
-from .sequence_parallel import SequenceParallelContextManager, SequenceParallelMixin
+from .sequence_parallel import SequenceParallelMixin

src/axolotl/core/trainers/mixins/scheduler.py

@@ -3,9 +3,10 @@
 import logging
 
 import torch
-from torch.optim.lr_scheduler import OneCycleLR
+from torch.optim.lr_scheduler import LRScheduler, OneCycleLR
 from transformers.trainer import Trainer
 
+from axolotl.integrations.base import PluginManager
 from axolotl.utils.schedulers import (
     RexLR,
     get_cosine_schedule_with_min_lr,
@@ -25,9 +26,9 @@ class SchedulerMixin(Trainer):
 
     def create_scheduler(
         self, num_training_steps: int, optimizer: torch.optim.Optimizer = None
-    ):
+    ) -> LRScheduler:
         """
-        Setup the scheduler. The optimizer of the trainer must have been set up either before this method is called or
+        Set up the scheduler. The optimizer of the trainer must have been set up either before this method is called or
         passed as an argument.
 
         Args:
@@ -47,7 +48,16 @@ class SchedulerMixin(Trainer):
         # fmt: off
         if self.lr_scheduler is None:  # type: ignore  # pylint: disable=access-member-before-definition
             # fmt: on
-            if self.args.alternate_lr_scheduler_type == "one_cycle":
+            plugin_manager = PluginManager.get_instance()
+            lr_scheduler: LRScheduler | None = plugin_manager.create_lr_scheduler(
+                trainer=self,
+                optimizer=optimizer,
+                num_training_steps=num_training_steps
+            )
+            if lr_scheduler is not None:
+                LOG.info(f"Using plugin-created lr_scheduler: {lr_scheduler}")
+                self.lr_scheduler = lr_scheduler
+            elif self.args.alternate_lr_scheduler_type == "one_cycle":
                 num_warmup_steps = self.args.get_warmup_steps(num_training_steps)
                 pct_start = num_warmup_steps / num_training_steps
                 extra_lr_kwargs = {}
@@ -110,4 +120,4 @@ class SchedulerMixin(Trainer):
             if use_cosine_min_lr:
                 LOG.warning("axolotl's cosine scheduler with min lr not used (e.g., because of deepspeed).")
 
-        return self.lr_scheduler
+        return self.lr_scheduler  # type: ignore

src/axolotl/core/trainers/mixins/sequence_parallel.py

@@ -1,85 +1,13 @@
-"""
-Module for Axolotl trainer sequence parallelism mixin and training context manager
-"""
+"""Module for Axolotl trainer sequence parallelism mixin"""
 
-import functools
-import logging
-
-import torch
 import torch.distributed as dist
 from datasets import Dataset
-from torch import nn
 from torch.utils.data import DistributedSampler, Sampler
-from torch.utils.hooks import RemovableHandle
 
 from axolotl.monkeypatch.attention.ring_attn import (
-    RingAttnFunc,
     get_ring_attn_group,
-    update_ring_attn_params,
 )
 
-LOG = logging.getLogger(__name__)
-
-
-def apply_sequence_parallelism(
-    batch: dict[str, torch.Tensor],
-    local_rank: int,
-    local_world_size: int,
-    ring_attn_func: RingAttnFunc,
-) -> dict[str, torch.Tensor]:
-    """
-    Apply sequence parallelism slicing to a batch.
-
-    Args:
-        batch: Batch dictionary (e.g., input_ids, attention_mask, etc.)
-        local_rank: Local rank in the sequence parallel group
-        local_world_size: World size of the sequence parallel group
-        ring_attn_func: The ring attention function to use
-
-    Returns:
-        Sliced batch dictionary.
-    """
-    # Update ring attention params if needed
-    if batch.get("position_ids") is not None:
-        update_ring_attn_params(position_ids=batch["position_ids"])
-
-    # Slice batch for sequence parallel processing
-    total_seq_len = batch["input_ids"].size(1)
-    for key in batch:
-        if (
-            key in batch
-            and isinstance(batch[key], torch.Tensor)
-            and batch[key].dim() > 1
-            and batch[key].size(1) == total_seq_len
-        ):
-
-            if ring_attn_func in [
-                RingAttnFunc.VARLEN_LLAMA3,
-                RingAttnFunc.BATCH_RING,
-            ]:
-                # Split in sequential fashion and grab this rank's chunk
-                batch[key] = (
-                    batch[key].chunk(local_world_size, dim=1)[local_rank].contiguous()
-                )
-            elif ring_attn_func is RingAttnFunc.BATCH_ZIGZAG:
-                chunks = batch[key].chunk(2 * local_world_size, dim=1)
-
-                # Take rank's chunk and opposing chunk for zigzag pattern
-                selected_chunks = [
-                    chunks[local_rank],
-                    chunks[2 * local_world_size - local_rank - 1],
-                ]
-                batch[key] = torch.cat(selected_chunks, dim=1).contiguous()
-            elif ring_attn_func is RingAttnFunc.BATCH_STRIPE:
-                # Split into striped data and stack
-                tensor = torch.stack(
-                    batch[key].split(local_world_size, dim=1),
-                    dim=1,
-                ).transpose(1, 2)
-                batch[key] = tensor[:, local_rank].contiguous()
-
-    return batch
-
 
 class SequenceParallelMixin:
     """
@@ -157,157 +85,3 @@ class SequenceParallelMixin:
         return self._create_sequence_parallel_sampler(
             eval_dataset, shuffle=False, is_eval=True
         )
-
-
-class SequenceParallelContextManager:
-    """
-    Context manager for sequence parallelism operations.
-
-    This class provides a context that will automatically apply sequence parallelism
-    during model forward passes using a pre-forward hook, and gather outputs from
-    across the sequence parallelism group using a post-forward hook.
-    """
-
-    def __init__(
-        self,
-        model: nn.Module,
-        sequence_parallel_degree: int,
-        ring_attn_func: RingAttnFunc,
-    ):
-        self.model = model
-        self.sequence_parallel_degree = sequence_parallel_degree
-        self.ring_attn_func = ring_attn_func
-        self.process_group = get_ring_attn_group()
-
-        # Initialize sequence parallel group details
-        self.local_rank = dist.get_rank(self.process_group)
-        self.local_world_size = dist.get_world_size(self.process_group)
-
-        # Will store hook handles for removal
-        self.hook_handles: list[RemovableHandle] = []
-
-        # Create a partially applied version of the apply_sequence_parallelism function
-        # with pre-configured params
-        self.apply_sequence_parallelism = functools.partial(
-            apply_sequence_parallelism,
-            local_rank=self.local_rank,
-            local_world_size=self.local_world_size,
-            ring_attn_func=self.ring_attn_func,
-        )
-
-    def __enter__(self):
-        # Forward pre-hook to apply sequence parallelism
-        def sequence_parallel_pre_hook(_, args, kwargs):
-            # Apply sequence parallelism to kwargs
-            kwargs = self.apply_sequence_parallelism(batch=kwargs)
-            return args, kwargs
-
-        # Forward post-hook to gather outputs
-        def sequence_parallel_post_hook(_, __, output):
-            # Gather the sharded outputs
-            return self.gather_outputs(output)
-
-        # Register both hooks
-        self.hook_handles.append(
-            self.model.register_forward_pre_hook(
-                sequence_parallel_pre_hook, with_kwargs=True
-            )
-        )
-        self.hook_handles.append(
-            self.model.register_forward_hook(sequence_parallel_post_hook)
-        )
-
-        return self
-
-    def __exit__(self, exc_type, exc_val, exc_tb):
-        # Remove all hooks
-        for handle in self.hook_handles:
-            handle.remove()
-        self.hook_handles = []
-
-    def gather_outputs(self, output):
-        """Gather sharded outputs from all ranks and reconstruct the full tensor."""
-        # Handle different output formats (dict, tensor, etc.)
-        if isinstance(output, dict):
-            gathered_output = {}
-            for key, value in output.items():
-                if isinstance(value, torch.Tensor) and value.dim() > 1:
-                    # Gather logits or other sequence-sharded tensors
-                    gathered_value = self.gather_tensor(value)
-                    gathered_output[key] = gathered_value
-                else:
-                    gathered_value = value.clone()
-                    dist.all_reduce(
-                        gathered_value, op=dist.ReduceOp.SUM, group=self.process_group
-                    )
-                    gathered_output[key] = gathered_value
-            return gathered_output
-        if isinstance(output, torch.Tensor):
-            return self.gather_tensor(output)
-
-        return output
-
-    def gather_tensor(self, tensor):
-        """Gather a sharded tensor from all ranks."""
-        # Prepare tensors for all_gather
-        world_size = self.local_world_size
-
-        # Create list to store tensors from all ranks
-        gathered_tensors = [torch.zeros_like(tensor) for _ in range(world_size)]
-
-        # All-gather operation
-        dist.all_gather(gathered_tensors, tensor, group=self.process_group)
-
-        # Concatenate along sequence dimension (typically dim=1)
-        if self.ring_attn_func in [RingAttnFunc.VARLEN_LLAMA3, RingAttnFunc.BATCH_RING]:
-            # Simple concatenation for standard sharding
-            return torch.cat(gathered_tensors, dim=1)
-
-        if self.ring_attn_func is RingAttnFunc.BATCH_ZIGZAG:
-            # Each rank has a pattern of (rank, world_size*2-rank-1)
-            reconstituted_tensors = [None] * (world_size * 2)
-
-            # First, split each gathered tensor into its two chunks
-            for rank, gathered_tensor in enumerate(gathered_tensors):
-                # Each tensor contains two chunks in the sequence dimension
-                chunk_size = gathered_tensor.size(1) // 2
-                chunk1, chunk2 = gathered_tensor.split(chunk_size, dim=1)
-
-                # Place chunks in their original positions
-                reconstituted_tensors[rank] = chunk1
-                reconstituted_tensors[world_size * 2 - rank - 1] = chunk2
-
-            # Concatenate the reconstituted tensors in the correct order
-            return torch.cat(reconstituted_tensors, dim=1)
-
-        # Otherwise, RingAttnFunc.BATCH_STRIPE
-        # In striping, each rank has every world_size-th slice
-        batch_size = tensor.size(0)
-        hidden_dim = tensor.size(-1)
-
-        # First, determine the full sequence length
-        total_seq_len = 0
-        for t in gathered_tensors:
-            total_seq_len += t.size(1)
-
-        # Create a tensor to hold the unstriped result
-        result = torch.zeros(
-            batch_size,
-            total_seq_len,
-            hidden_dim,
-            dtype=tensor.dtype,
-            device=tensor.device,
-        )
-
-        # For each rank's tensor, distribute its slices to the correct positions
-        for rank, gathered_tensor in enumerate(gathered_tensors):
-            # The rank's tensor contains every world_size-th slice
-            # starting from its rank position
-            seq_len = gathered_tensor.size(1)
-            for i in range(seq_len):
-                # Calculate the position in the full tensor
-                pos = i * world_size + rank
-                if pos < total_seq_len:
-                    result[:, pos] = gathered_tensor[:, i]
-
-        return result

src/axolotl/core/trainers/relora.py

@@ -1,6 +1,7 @@
 """Module for ReLoRA trainer"""
 
 import torch
+from torch.optim.lr_scheduler import LRScheduler
 
 from axolotl.core.trainers.base import AxolotlTrainer
 from axolotl.monkeypatch.relora import ReLoRAScheduler
@@ -19,9 +20,11 @@ class ReLoRATrainer(AxolotlTrainer):
         self,
         num_training_steps: int,
         optimizer: torch.optim.Optimizer | None = None,
-    ):
+    ) -> LRScheduler:
         optimizer = self.optimizer if optimizer is None else optimizer
-        lr_scheduler = super().create_scheduler(num_training_steps, optimizer)
+        lr_scheduler: LRScheduler = super().create_scheduler(
+            num_training_steps, optimizer
+        )
 
         if self.args.relora_steps:
             warmup_steps = (
@@ -30,7 +33,7 @@ class ReLoRATrainer(AxolotlTrainer):
             anneal_steps = (
                 self.args.relora_anneal_steps if self.args.relora_anneal_steps else 1
             )
-            self.lr_scheduler = ReLoRAScheduler(
+            self.lr_scheduler = ReLoRAScheduler(  # type: ignore
                 optimizer,
                 lr_scheduler,
                 self.args.relora_steps,
@@ -38,6 +41,6 @@ class ReLoRATrainer(AxolotlTrainer):
                 warmup_steps,
             )
         else:
-            self.lr_scheduler = lr_scheduler
+            self.lr_scheduler = lr_scheduler  # type: ignore
 
-        return self.lr_scheduler
+        return self.lr_scheduler  # type: ignore

src/axolotl/core/training_args.py

@@ -9,7 +9,7 @@ from PIL.Image import Resampling
 from transformers import TrainingArguments
 from trl import CPOConfig, KTOConfig, ORPOConfig, PRMConfig, RewardConfig
 
-from axolotl.monkeypatch.attention.ring_attn.patch import RingAttnFunc
+from axolotl.utils.schemas.enums import RingAttnFunc
 
 
 @dataclass

src/axolotl/evaluate.py

@@ -11,20 +11,19 @@ from accelerate.logging import get_logger
 from datasets import Dataset
 from transformers.trainer import Trainer
 
-from axolotl.logging_config import configure_logging
-from axolotl.train import TrainDatasetMeta
-from axolotl.utils import set_pytorch_cuda_alloc_conf
+from axolotl.train import (
+    TrainDatasetMeta,
+    setup_model_and_tokenizer,
+)
 from axolotl.utils.dict import DictDefault
 from axolotl.utils.distributed import cleanup_distributed
-from axolotl.utils.models import load_model, load_processor, load_tokenizer
 from axolotl.utils.trainer import setup_trainer
 
 project_root = os.path.abspath(os.path.join(os.path.dirname(__file__), ".."))
 src_dir = os.path.join(project_root, "src")
 sys.path.insert(0, src_dir)
 
-configure_logging()
-LOG = get_logger("axolotl.evaluate")
+LOG = get_logger(__name__)
 
 
 def evaluate_dataset(
@@ -75,37 +74,22 @@ def evaluate(*, cfg: DictDefault, dataset_meta: TrainDatasetMeta) -> Dict[str, f
     Returns:
         Dictionary mapping metric names to their values.
     """
-    # pylint: disable=duplicate-code
-    # Enable expandable segments for cuda allocation to improve VRAM usage
-    set_pytorch_cuda_alloc_conf()
-
-    # Load tokenizer
-    LOG.debug(
-        f"loading tokenizer... {cfg.tokenizer_config or cfg.base_model_config}",
-        main_process_only=True,
-    )
-    tokenizer = load_tokenizer(cfg)
-
-    # Load processor for multimodal models if needed
-    processor = None
-    if cfg.is_multimodal:
-        processor = load_processor(cfg, tokenizer)
+    # Load tokenizer, processor and model
+    LOG.debug("loading model for evaluation...")
+    model, tokenizer, _, processor = setup_model_and_tokenizer(cfg)
 
     # Get datasets
+    # pylint: disable=duplicate-code
     train_dataset = dataset_meta.train_dataset
     eval_dataset = dataset_meta.eval_dataset
     total_num_steps = dataset_meta.total_num_steps
 
-    # Load model
-    LOG.debug("loading model for evaluation...")
-    model, _ = load_model(cfg, tokenizer, processor=processor)
-
     # Set up trainer
     trainer = setup_trainer(
-        cfg,
+        cfg=cfg,
         train_dataset=train_dataset,
         eval_dataset=eval_dataset,
-        model=(model, None, None),  # No need for model_ref or peft_config
+        model=model,
         tokenizer=tokenizer,
         processor=processor,
         total_num_steps=total_num_steps,

src/axolotl/integrations/base.py

@@ -24,6 +24,9 @@ import logging
 from typing import OrderedDict
 
 import torch
+from torch.optim.lr_scheduler import LRScheduler
+
+from axolotl.utils.dict import DictDefault
 
 
 class BasePlugin:
@@ -35,12 +38,15 @@ class BasePlugin:
 
     Methods:
     register(cfg): Registers the plugin with the given configuration.
+    load_datasets(cfg): Loads and preprocesses the dataset for training.
     pre_model_load(cfg): Performs actions before the model is loaded.
-    post_model_load(cfg, model): Performs actions after the model is loaded.
+    post_model_build(cfg, model): Performs actions after the model is loaded, but before LoRA adapters are applied.
     pre_lora_load(cfg, model): Performs actions before LoRA weights are loaded.
     post_lora_load(cfg, model): Performs actions after LoRA weights are loaded.
+    post_model_load(cfg, model): Performs actions after the model is loaded, inclusive of any adapters.
+    post_trainer_create(cfg, trainer): Performs actions after the trainer is created.
     create_optimizer(cfg, trainer): Creates and returns an optimizer for training.
-    create_lr_scheduler(cfg, trainer, optimizer): Creates and returns a learning rate scheduler.
+    create_lr_scheduler(cfg, trainer, optimizer, num_training_steps): Creates and returns a learning rate scheduler.
     add_callbacks_pre_trainer(cfg, model): Adds callbacks to the trainer before training.
     add_callbacks_post_trainer(cfg, trainer): Adds callbacks to the trainer after training.
     """
@@ -61,106 +67,139 @@ class BasePlugin:
         None
         """
 
-    def get_input_args(self):
+    def get_input_args(self) -> str | None:
         """
         Returns a pydantic model for the plugin's input arguments.
         """
 
+    def load_datasets(self, cfg: DictDefault, preprocess: bool = False):
+        """
+        Loads and preprocesses the dataset for training.
+
+        Args:
+            cfg: The configuration for the plugin.
+            preprocess: Whether this is the preprocess step of the datasets.
+
+        Returns:
+            dataset_meta: The metadata for the training dataset.
+        """
+
     def pre_model_load(self, cfg):  # pylint: disable=unused-argument
         """
         Performs actions before the model is loaded.
 
-        Parameters:
-        cfg (dict): The configuration for the plugin.
+        Args:
+            cfg (dict): The configuration for the plugin.
 
         Returns:
-        None
+            None
+        """
+
+    def post_model_build(self, cfg, model):  # pylint: disable=unused-argument
+        """
+        Performs actions after the model is built/loaded, but before any adapters are applied.
+
+        Args:
+            cfg (dict): The configuration for the plugin.
         """
 
     def post_model_load(self, cfg, model):  # pylint: disable=unused-argument
         """
         Performs actions after the model is loaded.
 
-        Parameters:
-        cfg (dict): The configuration for the plugin.
-        model (object): The loaded model.
+        Args:
+            cfg (dict): The configuration for the plugin.
+            model (object): The loaded model.
 
         Returns:
-        None
+            None
         """
 
     def pre_lora_load(self, cfg, model):  # pylint: disable=unused-argument
         """
         Performs actions before LoRA weights are loaded.
 
-        Parameters:
-        cfg (dict): The configuration for the plugin.
-        model (object): The loaded model.
+        Args:
+            cfg (dict): The configuration for the plugin.
+            model (object): The loaded model.
 
         Returns:
-        None
+            None
         """
 
     def post_lora_load(self, cfg, model):  # pylint: disable=unused-argument
         """
         Performs actions after LoRA weights are loaded.
 
-        Parameters:
-        cfg (dict): The configuration for the plugin.
-        model (object): The loaded model.
+        Args:
+            cfg (dict): The configuration for the plugin.
+            model (object): The loaded model.
 
         Returns:
-        None
+            None
         """
 
     def get_trainer_cls(self, cfg):  # pylint: disable=unused-argument):
         """
         Returns a custom class for the trainer.
 
-        Parameters:
-        cfg (dict): The global axolotl configuration.
+        Args:
+            cfg (dict): The global axolotl configuration.
 
         Returns:
-        class: The class for the trainer.
+            class: The class for the trainer.
+        """
+
+    def post_trainer_create(self, cfg, trainer):  # pylint: disable=unused-argument
+        """
+        Performs actions after the trainer is created.
+
+        Args:
+            cfg (dict): The configuration for the plugin.
+            trainer (object): The trainer object for training.
+
+        Returns:
+            None
         """
 
     def create_optimizer(self, cfg, trainer):  # pylint: disable=unused-argument
         """
         Creates and returns an optimizer for training.
 
-        Parameters:
-        cfg (dict): The configuration for the plugin.
-        trainer (object): The trainer object for training.
+        Args:
+            cfg (dict): The configuration for the plugin.
+            trainer (object): The trainer object for training.
 
         Returns:
-        object: The created optimizer.
+            object: The created optimizer.
         """
 
     def create_lr_scheduler(
-        self, cfg, trainer, optimizer
-    ):  # pylint: disable=unused-argument
+        self, cfg, trainer, optimizer, num_training_steps
+    ) -> LRScheduler | None:  # pylint: disable=unused-argument
         """
         Creates and returns a learning rate scheduler.
 
-        Parameters:
-        cfg (dict): The configuration for the plugin.
-        trainer (object): The trainer object for training.
-        optimizer (object): The optimizer for training.
+        Args:
+            cfg (dict): The configuration for the plugin.
+            trainer (object): The trainer object for training.
+            optimizer (object): The optimizer for training.
+            num_training_steps (int): Total number of training steps
 
         Returns:
-        object: The created learning rate scheduler.
+            object (LRScheduler): The created learning rate scheduler.
         """
 
     def add_callbacks_pre_trainer(self, cfg, model):  # pylint: disable=unused-argument
         """
         setup callbacks before creating the trainer.
 
-        Parameters:
-        cfg (dict): The configuration for the plugin.
-        model (object): The loaded model.
+        Args:
+            cfg (dict): The configuration for the plugin.
+            model (object): The loaded model.
 
         Returns:
-        List[callable]: A list of callback functions to be added to the TrainingArgs
+            List[callable]: A list of callback functions to be added to the TrainingArgs
         """
         return []
 
@@ -171,12 +210,12 @@ class BasePlugin:
         Adds callbacks to the trainer after creating the trainer.
         This is useful for callbacks that require access to the model or trainer.
 
-        Parameters:
-        cfg (dict): The configuration for the plugin.
-        trainer (object): The trainer object for training.
+        Args:
+            cfg (dict): The configuration for the plugin.
+            trainer (object): The trainer object for training.
 
         Returns:
-        List[callable]: A list of callback functions to be added
+            List[callable]: A list of callback functions to be added
         """
         return []
 
@@ -184,23 +223,23 @@ class BasePlugin:
         """
         Performs actions after training is complete.
 
-        Parameters:
-        cfg (dict): The axolotl configuration
-        model (object): The loaded model.
+        Args:
+            cfg (dict): The axolotl configuration
+            model (object): The loaded model.
 
         Returns:
-        None
+            None
         """
 
     def post_train_unload(self, cfg):  # pylint: disable=unused-argument
         """
         Performs actions after training is complete and the model is unloaded.
 
-        Parameters:
-        cfg (dict): The configuration for the plugin.
+        Args:
+            cfg (dict): The configuration for the plugin.
 
         Returns:
-        None
+            None
         """
 
 
@@ -261,6 +300,7 @@ class PluginManager:
     plugins: OrderedDict[str, BasePlugin] = collections.OrderedDict()
 
     _instance = None
+    _cfg = None
 
     def __new__(cls):
         """
@@ -268,7 +308,9 @@ class PluginManager:
         """
         if cls._instance is None:
             cls._instance = super(PluginManager, cls).__new__(cls)
-            cls._instance.plugins = collections.OrderedDict()
+            cls._instance.plugins: OrderedDict[str, BasePlugin] = (
+                collections.OrderedDict()
+            )
         return cls._instance
 
     @staticmethod
@@ -281,6 +323,14 @@ class PluginManager:
             PluginManager()
         return PluginManager._instance  # type: ignore
 
+    @property
+    def cfg(self):
+        return self._cfg
+
+    @cfg.setter
+    def cfg(self, cfg):
+        self._cfg = cfg
+
     def register(self, plugin_name: str):
         """
         Registers a new plugin by its name.
@@ -316,6 +366,27 @@ class PluginManager:
                 input_args.append(input_args_from_plugin)
         return input_args
 
+    def load_datasets(self, cfg, preprocess: bool = False):
+        """
+        Calls the load_datasets method of each registered plugin.
+
+        Args:
+            cfg: The configuration for the plugins.
+            preprocess : Whether this is preprocess step of the datasets.
+
+        Returns:
+            dataset_meta: The dataset metadata loaded from all registered plugins.
+        """
+        return_ds_meta = None
+        for plugin in self.plugins.values():
+            dataset_meta = plugin.load_datasets(cfg, preprocess)
+            if dataset_meta is not None:
+                if return_ds_meta is None:
+                    return_ds_meta = dataset_meta
+                else:
+                    raise RuntimeError("Multiple plugins loaded datasets")
+        return return_ds_meta
+
     def pre_model_load(self, cfg):
         """
         Calls the pre_model_load method of all registered plugins.
@@ -329,9 +400,22 @@ class PluginManager:
         for plugin in self.plugins.values():
             plugin.pre_model_load(cfg)
 
+    def post_model_build(self, cfg, model):
+        """
+        Calls the post_model_build method of all registered plugins after the model has been built/loaded,
+        but before any adapters have been applied.
+
+        Args:
+            cfg (dict): The configuration for the plugins.
+            model (object): The loaded model.
+        """
+        for plugin in self.plugins.values():
+            plugin.post_model_build(cfg, model)
+
     def post_model_load(self, cfg, model):
         """
-        Calls the post_model_load method of all registered plugins.
+        Calls the post_model_load method of all registered plugins after the model has been loaded
+        inclusive of any adapters
 
         Parameters:
         cfg (dict): The configuration for the plugins.
@@ -387,29 +471,43 @@ class PluginManager:
                 return trainer_cls
         return None
 
-    def create_optimizer(self, cfg, trainer):
+    def post_trainer_create(self, cfg, trainer):
         """
-        Calls the create_optimizer method of all registered plugins and returns the first non-None optimizer.
+        Calls the post_trainer_create method of all registered plugins.
 
         Parameters:
         cfg (dict): The configuration for the plugins.
         trainer (object): The trainer object for training.
 
+        Returns:
+        None
+        """
+        for plugin in self.plugins.values():
+            plugin.post_trainer_create(cfg, trainer)
+
+    def create_optimizer(self, trainer):
+        """
+        Calls the create_optimizer method of all registered plugins and returns the first non-None optimizer.
+
+        Parameters:
+        trainer (object): The trainer object for training.
+
         Returns:
         object: The created optimizer, or None if none was found.
         """
         for plugin in self.plugins.values():
-            optimizer = plugin.create_optimizer(cfg, trainer)
+            optimizer = plugin.create_optimizer(self.cfg, trainer)
             if optimizer is not None:
                 return optimizer
         return None
 
-    def create_lr_scheduler(self, cfg, trainer, optimizer):
+    def create_lr_scheduler(
+        self, trainer, optimizer, num_training_steps
+    ) -> LRScheduler | None:
         """
         Calls the create_lr_scheduler method of all registered plugins and returns the first non-None scheduler.
 
         Parameters:
-        cfg (dict): The configuration for the plugins.
         trainer (object): The trainer object for training.
         optimizer (object): The optimizer for training.
 
@@ -417,7 +515,12 @@ class PluginManager:
         object: The created learning rate scheduler, or None if none was found.
         """
         for plugin in self.plugins.values():
-            scheduler = plugin.create_lr_scheduler(cfg, trainer, optimizer)
+            scheduler: LRScheduler | None = plugin.create_lr_scheduler(
+                self.cfg,
+                trainer=trainer,
+                optimizer=optimizer,
+                num_training_steps=num_training_steps,
+            )
             if scheduler is not None:
                 return scheduler
         return None
@@ -458,6 +561,20 @@ class PluginManager:
                 callbacks.extend(plugin_callbacks)
         return callbacks
 
+    def post_train(self, cfg, model):
+        """
+        Calls the post_train method of all registered plugins.
+
+        Parameters:
+        cfg (dict): The configuration for the plugins.
+        model (object): The loaded model.
+
+        Returns:
+        None
+        """
+        for plugin in self.plugins.values():
+            plugin.post_train(cfg, model)
+
     def post_train_unload(self, cfg):
         """
         Calls the post_train_unload method of all registered plugins.

src/axolotl/integrations/cut_cross_entropy/README.md

@@ -32,8 +32,8 @@ plugins:
 ## Supported Models
 
 - llama
-- llama4_text
 - llama4
+- llama4_text
 - mllama
 - phi3
 - gemma
@@ -43,6 +43,11 @@ plugins:
 - mistral
 - mistral3
 - qwen2
+- qwen2_moe
+- qwen2_vl
+- qwen2_5_vl
+- qwen3
+- qwen3_moe
 - cohere
 - cohere2
 - glm

src/axolotl/integrations/cut_cross_entropy/__init__.py

@@ -25,7 +25,7 @@ import torch
 
 from axolotl.integrations.base import BasePlugin
 from axolotl.utils import get_pytorch_version
-from axolotl.utils.distributed import zero_only
+from axolotl.utils.distributed import is_main_process
 
 from .args import CutCrossEntropyArgs  # pylint: disable=unused-import. # noqa: F401
 
@@ -76,7 +76,7 @@ class CutCrossEntropyPlugin(BasePlugin):
                 cce_patch,
             )
 
-            with zero_only():
+            if is_main_process(use_environ=True):
                 LOG.info(
                     f"Applying Cut Cross Entropy to model type: {cfg.model_config_type}"
                 )

src/axolotl/integrations/cut_cross_entropy/monkeypatch/llama.py

@@ -0,0 +1,174 @@
+"""Llama CCE patch. Adapted from transformers v4.51.2"""
+
+# pylint: disable=duplicate-code
+
+
+from types import MethodType
+from typing import Optional, Union
+
+import torch
+import transformers
+from cut_cross_entropy.transformers.utils import (
+    PatchOptions,
+    TransformersModelT,
+    apply_lce,
+)
+from transformers.cache_utils import Cache
+from transformers.modeling_outputs import (
+    BaseModelOutputWithPast,
+    CausalLMOutputWithPast,
+)
+from transformers.models.llama.modeling_llama import (
+    _CONFIG_FOR_DOC,
+    LLAMA_INPUTS_DOCSTRING,
+    KwargsForCausalLM,
+)
+from transformers.processing_utils import Unpack
+from transformers.utils import (
+    add_start_docstrings_to_model_forward,
+    replace_return_docstrings,
+)
+from transformers.utils.deprecation import deprecate_kwarg
+from transformers.utils.generic import can_return_tuple
+
+_PATCH_OPTS: PatchOptions | None = None
+
+
+@can_return_tuple
+@deprecate_kwarg("num_logits_to_keep", version="4.50", new_name="logits_to_keep")
+@add_start_docstrings_to_model_forward(LLAMA_INPUTS_DOCSTRING)
+@replace_return_docstrings(
+    output_type=CausalLMOutputWithPast, config_class=_CONFIG_FOR_DOC
+)
+def cce_forward(
+    self,
+    input_ids: Optional[torch.LongTensor] = None,
+    attention_mask: Optional[torch.Tensor] = None,
+    position_ids: Optional[torch.LongTensor] = None,
+    past_key_values: Optional[Cache] = None,
+    inputs_embeds: Optional[torch.FloatTensor] = None,
+    labels: Optional[torch.LongTensor] = None,
+    use_cache: Optional[bool] = None,
+    output_attentions: Optional[bool] = None,
+    output_hidden_states: Optional[bool] = None,
+    cache_position: Optional[torch.LongTensor] = None,
+    logits_to_keep: Union[int, torch.Tensor] = 0,
+    **kwargs: Unpack[KwargsForCausalLM],
+) -> CausalLMOutputWithPast:
+    r"""
+        labels (`torch.LongTensor` of shape `(batch_size, sequence_length)`, *optional*):
+            Labels for computing the masked language modeling loss. Indices should either be in `[0, ...,
+            config.vocab_size]` or -100 (see `input_ids` docstring). Tokens with indices set to `-100` are ignored
+            (masked), the loss is only computed for the tokens with labels in `[0, ..., config.vocab_size]`.
+
+        logits_to_keep (`int` or `torch.Tensor`, *optional*):
+            If an `int`, compute logits for the last `logits_to_keep` tokens. If `0`, calculate logits for all
+            `input_ids` (special case). Only last token logits are needed for generation, and calculating them only for that
+            token can save memory, which becomes pretty significant for long sequences or large vocabulary size.
+            If a `torch.Tensor`, must be 1D corresponding to the indices to keep in the sequence length dimension.
+            This is useful when using packed tensor format (single dimension for batch and sequence length).
+
+    Returns:
+
+    Example:
+
+    ```python
+    >>> from transformers import AutoTokenizer, LlamaForCausalLM
+
+    >>> model = LlamaForCausalLM.from_pretrained("meta-llama/Llama-2-7b-hf")
+    >>> tokenizer = AutoTokenizer.from_pretrained("meta-llama/Llama-2-7b-hf")
+
+    >>> prompt = "Hey, are you conscious? Can you talk to me?"
+    >>> inputs = tokenizer(prompt, return_tensors="pt")
+
+    >>> # Generate
+    >>> generate_ids = model.generate(inputs.input_ids, max_length=30)
+    >>> tokenizer.batch_decode(generate_ids, skip_special_tokens=True, clean_up_tokenization_spaces=False)[0]
+    "Hey, are you conscious? Can you talk to me?\nI'm not conscious, but I can talk to you."
+    ```"""
+    output_attentions = (
+        output_attentions
+        if output_attentions is not None
+        else self.config.output_attentions
+    )
+    output_hidden_states = (
+        output_hidden_states
+        if output_hidden_states is not None
+        else self.config.output_hidden_states
+    )
+
+    # decoder outputs consists of (dec_features, layer_state, dec_hidden, dec_attn)
+    outputs: BaseModelOutputWithPast = self.model(
+        input_ids=input_ids,
+        attention_mask=attention_mask,
+        position_ids=position_ids,
+        past_key_values=past_key_values,
+        inputs_embeds=inputs_embeds,
+        use_cache=use_cache,
+        output_attentions=output_attentions,
+        output_hidden_states=output_hidden_states,
+        cache_position=cache_position,
+        **kwargs,
+    )
+
+    hidden_states = outputs.last_hidden_state
+    if hidden_states is None:
+        raise ValueError("hidden_states is None")
+
+    loss = None
+    logits = None
+
+    # Only compute necessary logits, and do not upcast them to float if we are not computing the loss
+    slice_indices = (
+        slice(-logits_to_keep, None)
+        if isinstance(logits_to_keep, int)
+        else logits_to_keep
+    )
+    if _PATCH_OPTS is not None and _PATCH_OPTS.use_lce(labels, self.training):
+        assert labels is not None
+        loss = apply_lce(
+            hidden_states[:, slice_indices, :],
+            self.lm_head.weight,
+            labels,
+            _PATCH_OPTS,
+            **kwargs,
+        )
+    else:
+        logits = self.lm_head(hidden_states[:, slice_indices, :])
+
+        if labels is not None:
+            loss = self.loss_function(
+                logits=logits,
+                labels=labels,
+                vocab_size=self.config.vocab_size,
+                **kwargs,
+            )
+
+    return CausalLMOutputWithPast(
+        loss=loss,
+        logits=logits,
+        past_key_values=outputs.past_key_values,
+        hidden_states=outputs.hidden_states,
+        attentions=outputs.attentions,
+    )
+
+
+def patch_llama(
+    maybe_model: TransformersModelT | str | transformers.PretrainedConfig,
+    patch_options: PatchOptions,
+) -> TransformersModelT | None:
+    """Patch Llama for CCE."""
+    global _PATCH_OPTS  # pylint: disable=global-statement
+    from transformers.models.llama import modeling_llama
+
+    _PATCH_OPTS = patch_options
+
+    if isinstance(maybe_model, transformers.PreTrainedModel):
+        assert isinstance(
+            maybe_model, modeling_llama.LlamaForCausalLM
+        ), f"Expected a LlamaForCausalLM model. Got {type(maybe_model)}."
+        maybe_model.forward = MethodType(cce_forward, maybe_model)
+        return maybe_model
+
+    modeling_llama.LlamaForCausalLM.forward = cce_forward
+    return None

src/axolotl/integrations/cut_cross_entropy/monkeypatch/patch.py

@@ -5,9 +5,7 @@
 import transformers
 from cut_cross_entropy.cce_utils import LinearCrossEntropyImpl
 from cut_cross_entropy.linear_cross_entropy import LCE_IMPL_DEFAULT
-from cut_cross_entropy.transformers.llama import patch_llama
 from cut_cross_entropy.transformers.phi3 import patch_phi3
-from cut_cross_entropy.transformers.qwen2 import patch_qwen2
 from cut_cross_entropy.transformers.utils import PatchOptions, TransformersModelT
 
 from axolotl.integrations.cut_cross_entropy.monkeypatch.cohere import (
@@ -24,6 +22,9 @@ from axolotl.integrations.cut_cross_entropy.monkeypatch.glm4 import (
     patch_glm,
     patch_glm4,
 )
+from axolotl.integrations.cut_cross_entropy.monkeypatch.llama import (
+    patch_llama,
+)
 from axolotl.integrations.cut_cross_entropy.monkeypatch.llama4 import (
     patch_llama4,
     patch_llama4_text,
@@ -33,6 +34,22 @@ from axolotl.integrations.cut_cross_entropy.monkeypatch.mistral3 import (
     patch_mistral3,
 )
 from axolotl.integrations.cut_cross_entropy.monkeypatch.mllama import patch_mllama
+from axolotl.integrations.cut_cross_entropy.monkeypatch.qwen2 import (
+    patch_qwen2,
+)
+from axolotl.integrations.cut_cross_entropy.monkeypatch.qwen2_5_vl import (
+    patch_qwen2_5_vl,
+)
+from axolotl.integrations.cut_cross_entropy.monkeypatch.qwen2_moe import (
+    patch_qwen2_moe,
+)
+from axolotl.integrations.cut_cross_entropy.monkeypatch.qwen2_vl import (
+    patch_qwen2_vl,
+)
+from axolotl.integrations.cut_cross_entropy.monkeypatch.qwen3 import patch_qwen3
+from axolotl.integrations.cut_cross_entropy.monkeypatch.qwen3_moe import (
+    patch_qwen3_moe,
+)
 
 CUT_CROSS_ENTROPY_MODEL_MAPPING = {
     "llama": patch_llama,
@@ -47,6 +64,11 @@ CUT_CROSS_ENTROPY_MODEL_MAPPING = {
     "mistral": patch_mistral,
     "mistral3": patch_mistral3,
     "qwen2": patch_qwen2,
+    "qwen2_moe": patch_qwen2_moe,
+    "qwen2_vl": patch_qwen2_vl,
+    "qwen2_5_vl": patch_qwen2_5_vl,
+    "qwen3": patch_qwen3,
+    "qwen3_moe": patch_qwen3_moe,
     "cohere": patch_cohere,
     "cohere2": patch_cohere2,
     "glm": patch_glm,

src/axolotl/integrations/cut_cross_entropy/monkeypatch/qwen2.py

@@ -0,0 +1,37 @@
+"""Qwen2 CCE patch. The model inherits Llama's modeling code and uses the same forward method."""
+
+# pylint: disable=duplicate-code
+
+from types import MethodType
+
+import transformers
+from cut_cross_entropy.transformers.utils import (
+    PatchOptions,
+    TransformersModelT,
+)
+
+
+def patch_qwen2(
+    maybe_model: TransformersModelT | str | transformers.PretrainedConfig,
+    patch_options: PatchOptions,
+) -> TransformersModelT | None:
+    from transformers.models.qwen2 import modeling_qwen2
+
+    # Set the _PATCH_OPTS in the llama patch file
+    import axolotl.integrations.cut_cross_entropy.monkeypatch.llama as llama_patch
+
+    llama_patch._PATCH_OPTS = patch_options  # pylint: disable=protected-access
+
+    from axolotl.integrations.cut_cross_entropy.monkeypatch.llama import (
+        cce_forward,
+    )
+
+    if isinstance(maybe_model, transformers.PreTrainedModel):
+        assert isinstance(
+            maybe_model, modeling_qwen2.Qwen2ForCausalLM
+        ), f"Expected a Qwen2ForCausalLM model. Got {type(maybe_model)}."
+        maybe_model.forward = MethodType(cce_forward, maybe_model)
+        return maybe_model
+
+    modeling_qwen2.Qwen2ForCausalLM.forward = cce_forward
+    return None

src/axolotl/integrations/cut_cross_entropy/monkeypatch/qwen2_5_vl.py

@@ -0,0 +1,246 @@
+"""Qwen2.5 VL CCE patch. Adapted from transformers v4.51.2"""
+
+# pylint: disable=duplicate-code
+
+
+from types import MethodType
+from typing import Optional, Tuple, Union
+
+import torch
+import transformers
+from cut_cross_entropy.transformers.utils import (
+    PatchOptions,
+    TransformersModelT,
+    apply_lce,
+)
+from torch.nn import CrossEntropyLoss
+from transformers.models.qwen2_5_vl.modeling_qwen2_5_vl import (
+    Qwen2_5_VLCausalLMOutputWithPast,
+)
+
+_PATCH_OPTS: PatchOptions | None = None
+
+
+def cce_forward_multimodal(
+    self,
+    input_ids: Optional[torch.LongTensor] = None,
+    attention_mask: Optional[torch.Tensor] = None,
+    position_ids: Optional[torch.LongTensor] = None,
+    past_key_values: Optional[list[torch.FloatTensor]] = None,
+    inputs_embeds: Optional[torch.FloatTensor] = None,
+    labels: Optional[torch.LongTensor] = None,
+    use_cache: Optional[bool] = None,
+    output_attentions: Optional[bool] = None,
+    output_hidden_states: Optional[bool] = None,
+    return_dict: Optional[bool] = None,
+    pixel_values: Optional[torch.Tensor] = None,
+    pixel_values_videos: Optional[torch.FloatTensor] = None,
+    image_grid_thw: Optional[torch.LongTensor] = None,
+    video_grid_thw: Optional[torch.LongTensor] = None,
+    rope_deltas: Optional[torch.LongTensor] = None,
+    cache_position: Optional[torch.LongTensor] = None,
+    second_per_grid_ts: Optional[torch.Tensor] = None,
+) -> Union[Tuple, Qwen2_5_VLCausalLMOutputWithPast]:
+    r"""
+        labels (`torch.LongTensor` of shape `(batch_size, sequence_length)`, *optional*):
+            Labels for computing the masked language modeling loss. Indices should either be in `[0, ...,
+            config.vocab_size]` or -100 (see `input_ids` docstring). Tokens with indices set to `-100` are ignored
+            (masked), the loss is only computed for the tokens with labels in `[0, ..., config.vocab_size]`.
+
+    Returns:
+
+    Example:
+
+    ```python
+    >>> from PIL import Image
+    >>> import requests
+    >>> from transformers import AutoProcessor, Qwen2_5_VLForConditionalGeneration
+
+    >>> model = Qwen2_5_VLForConditionalGeneration.from_pretrained("Qwen/Qwen2.5-VL-7B-Instruct")
+    >>> processor = AutoProcessor.from_pretrained("Qwen/Qwen2.5-VL-7B-Instruct")
+
+    >>> messages = [
+        {
+            "role": "user",
+            "content": [
+                {"type": "image"},
+                {"type": "text", "text": "What is shown in this image?"},
+            ],
+        },
+    ]
+    >>> url = "https://www.ilankelman.org/stopsigns/australia.jpg"
+    >>> image = Image.open(requests.get(url, stream=True).raw)
+
+    >>> text = processor.apply_chat_template(messages, tokenize=False, add_generation_prompt=True)
+    >>> inputs = processor(text=[text], images=[image], vision_infos=[vision_infos])
+
+    >>> # Generate
+    >>> generate_ids = model.generate(inputs.input_ids, max_length=30)
+    >>> tokenizer.batch_decode(generate_ids, skip_special_tokens=True, clean_up_tokenization_spaces=False)[0]
+    "The image shows a street scene with a red stop sign in the foreground. In the background, there is a large red gate with Chinese characters ..."
+    ```"""
+
+    output_attentions = (
+        output_attentions
+        if output_attentions is not None
+        else self.config.output_attentions
+    )
+    output_hidden_states = (
+        output_hidden_states
+        if output_hidden_states is not None
+        else self.config.output_hidden_states
+    )
+    return_dict = (
+        return_dict if return_dict is not None else self.config.use_return_dict
+    )
+
+    if inputs_embeds is None:
+        inputs_embeds = self.model.embed_tokens(input_ids)
+        if pixel_values is not None:
+            pixel_values = pixel_values.type(self.visual.dtype)
+            image_embeds = self.visual(pixel_values, grid_thw=image_grid_thw)
+            n_image_tokens = (input_ids == self.config.image_token_id).sum().item()
+            n_image_features = image_embeds.shape[0]
+            if n_image_tokens != n_image_features:
+                raise ValueError(
+                    f"Image features and image tokens do not match: tokens: {n_image_tokens}, features {n_image_features}"
+                )
+
+            mask = input_ids == self.config.image_token_id
+            mask_unsqueezed = mask.unsqueeze(-1)
+            mask_expanded = mask_unsqueezed.expand_as(inputs_embeds)
+            image_mask = mask_expanded.to(inputs_embeds.device)
+
+            image_embeds = image_embeds.to(inputs_embeds.device, inputs_embeds.dtype)
+            inputs_embeds = inputs_embeds.masked_scatter(image_mask, image_embeds)  # type: ignore
+
+        if pixel_values_videos is not None:
+            pixel_values_videos = pixel_values_videos.type(self.visual.dtype)
+            video_embeds = self.visual(pixel_values_videos, grid_thw=video_grid_thw)
+            n_video_tokens = (input_ids == self.config.video_token_id).sum().item()
+            n_video_features = video_embeds.shape[0]
+            if n_video_tokens != n_video_features:
+                raise ValueError(
+                    f"Video features and video tokens do not match: tokens: {n_video_tokens}, features {n_video_features}"
+                )
+
+            mask = input_ids == self.config.video_token_id
+            mask_unsqueezed = mask.unsqueeze(-1)
+            mask_expanded = mask_unsqueezed.expand_as(inputs_embeds)
+            video_mask = mask_expanded.to(inputs_embeds.device)
+
+            video_embeds = video_embeds.to(inputs_embeds.device, inputs_embeds.dtype)
+            inputs_embeds = inputs_embeds.masked_scatter(video_mask, video_embeds)  # type: ignore
+
+        if attention_mask is not None:
+            attention_mask = attention_mask.to(inputs_embeds.device)
+
+    # if we get 4D attention mask we cannot calculate rope deltas anymore. TODO @raushan fixme
+    if position_ids is None and (attention_mask is None or attention_mask.ndim == 2):
+        # calculate RoPE index once per generation in the pre-fill stage only
+        if (
+            (cache_position is not None and cache_position[0] == 0)
+            or self.rope_deltas is None
+            or (past_key_values is None or past_key_values.get_seq_length() == 0)  # type: ignore
+        ):
+            position_ids, rope_deltas = self.get_rope_index(
+                input_ids,
+                image_grid_thw,
+                video_grid_thw,
+                second_per_grid_ts,
+                attention_mask,
+            )
+            self.rope_deltas = rope_deltas
+        # then use the prev pre-calculated rope-deltas to get the correct position ids
+        else:
+            batch_size, seq_length, _ = inputs_embeds.shape
+            delta = (
+                (cache_position[0] + self.rope_deltas).to(inputs_embeds.device)
+                if cache_position is not None
+                else 0
+            )
+            position_ids = torch.arange(seq_length, device=inputs_embeds.device)  # type: ignore
+            position_ids = position_ids.view(1, -1).expand(batch_size, -1)  # type: ignore
+            if cache_position is not None:  # otherwise `deltas` is an int `0`
+                delta = delta.repeat_interleave(batch_size // delta.shape[0], dim=0)  # type: ignore
+            position_ids = position_ids.add(delta)  # type: ignore
+            position_ids = position_ids.unsqueeze(0).expand(3, -1, -1)  # type: ignore
+
+    outputs = self.model(
+        input_ids=None,
+        position_ids=position_ids,
+        attention_mask=attention_mask,
+        past_key_values=past_key_values,
+        inputs_embeds=inputs_embeds,
+        use_cache=use_cache,
+        output_attentions=output_attentions,
+        output_hidden_states=output_hidden_states,
+        return_dict=return_dict,
+        cache_position=cache_position,
+    )
+
+    hidden_states = outputs[0]
+    logits = None
+    loss = None
+
+    if _PATCH_OPTS is not None and _PATCH_OPTS.use_lce(labels, self.training):
+        assert labels is not None
+        loss = apply_lce(
+            hidden_states,
+            self.lm_head.weight,
+            labels,
+            _PATCH_OPTS,
+        )
+    else:
+        logits = self.lm_head(hidden_states)
+
+        if labels is not None:
+            # Upcast to float if we need to compute the loss to avoid potential precision issues
+            logits = logits.float()
+            # Shift so that tokens < n predict n
+            shift_logits = logits[..., :-1, :].contiguous()
+            shift_labels = labels[..., 1:].contiguous()
+            # Flatten the tokens
+            loss_fct = CrossEntropyLoss()
+            shift_logits = shift_logits.view(-1, self.config.vocab_size)
+            shift_labels = shift_labels.view(-1)
+            # Enable model parallelism
+            shift_labels = shift_labels.to(shift_logits.device)
+            loss = loss_fct(shift_logits, shift_labels)
+
+    if not return_dict:
+        output = (logits,) + outputs[1:]
+        return (loss,) + output if loss is not None else output
+
+    return Qwen2_5_VLCausalLMOutputWithPast(
+        loss=loss,
+        logits=logits,
+        past_key_values=outputs.past_key_values,
+        hidden_states=outputs.hidden_states,
+        attentions=outputs.attentions,
+        rope_deltas=self.rope_deltas,
+    )
+
+
+def patch_qwen2_5_vl(
+    maybe_model: TransformersModelT | str | transformers.PretrainedConfig,
+    patch_options: PatchOptions,
+) -> TransformersModelT | None:
+    global _PATCH_OPTS  # pylint: disable=global-statement
+
+    from transformers.models.qwen2_5_vl import modeling_qwen2_5_vl
+
+    _PATCH_OPTS = patch_options
+
+    if isinstance(maybe_model, transformers.PreTrainedModel):
+        assert isinstance(
+            maybe_model, modeling_qwen2_5_vl.Qwen2_5_VLForConditionalGeneration
+        ), f"Expected a Qwen2_5_VLForConditionalGeneration model. Got {type(maybe_model)}."
+        maybe_model.forward = MethodType(cce_forward_multimodal, maybe_model)
+
+        return maybe_model
+
+    modeling_qwen2_5_vl.Qwen2_5_VLForConditionalGeneration.forward = (
+        cce_forward_multimodal
+    )
+    return None

src/axolotl/integrations/cut_cross_entropy/monkeypatch/qwen2_moe.py

@@ -0,0 +1,188 @@
+"""Qwen2 MoE CCE patch. Adapted from transformers v4.51.2"""
+
+# pylint: disable=duplicate-code
+
+from types import MethodType
+from typing import Optional, Union
+
+import torch
+import transformers
+from cut_cross_entropy.transformers.utils import (
+    PatchOptions,
+    TransformersModelT,
+    apply_lce,
+)
+from transformers.models.qwen2_moe.modeling_qwen2_moe import (
+    _CONFIG_FOR_DOC,
+    QWEN2MOE_INPUTS_DOCSTRING,
+    MoeCausalLMOutputWithPast,
+    MoeModelOutputWithPast,
+    load_balancing_loss_func,
+)
+from transformers.utils import (
+    add_start_docstrings_to_model_forward,
+    replace_return_docstrings,
+)
+from transformers.utils.deprecation import deprecate_kwarg
+from transformers.utils.generic import can_return_tuple
+
+_PATCH_OPTS: PatchOptions | None = None
+
+
+@can_return_tuple
+@deprecate_kwarg("num_logits_to_keep", version="4.50", new_name="logits_to_keep")
+@add_start_docstrings_to_model_forward(QWEN2MOE_INPUTS_DOCSTRING)
+@replace_return_docstrings(
+    output_type=MoeCausalLMOutputWithPast, config_class=_CONFIG_FOR_DOC
+)
+def forward(
+    self,
+    input_ids: Optional[torch.LongTensor] = None,
+    attention_mask: Optional[torch.Tensor] = None,
+    position_ids: Optional[torch.LongTensor] = None,
+    past_key_values: Optional[list[torch.FloatTensor]] = None,
+    inputs_embeds: Optional[torch.FloatTensor] = None,
+    labels: Optional[torch.LongTensor] = None,
+    use_cache: Optional[bool] = None,
+    output_attentions: Optional[bool] = None,
+    output_hidden_states: Optional[bool] = None,
+    output_router_logits: Optional[bool] = None,
+    cache_position: Optional[torch.LongTensor] = None,
+    logits_to_keep: Union[int, torch.Tensor] = 0,
+    **loss_kwargs,
+) -> MoeCausalLMOutputWithPast:
+    r"""
+        labels (`torch.LongTensor` of shape `(batch_size, sequence_length)`, *optional*):
+            Labels for computing the masked language modeling loss. Indices should either be in `[0, ...,
+            config.vocab_size]` or -100 (see `input_ids` docstring). Tokens with indices set to `-100` are ignored
+            (masked), the loss is only computed for the tokens with labels in `[0, ..., config.vocab_size]`.
+
+        logits_to_keep (`int` or `torch.Tensor`, *optional*):
+            If an `int`, compute logits for the last `logits_to_keep` tokens. If `0`, calculate logits for all
+            `input_ids` (special case). Only last token logits are needed for generation, and calculating them only for that
+            token can save memory, which becomes pretty significant for long sequences or large vocabulary size.
+            If a `torch.Tensor`, must be 1D corresponding to the indices to keep in the sequence length dimension.
+            This is useful when using packed tensor format (single dimension for batch and sequence length).
+
+    Returns:
+
+    Example:
+
+    ```python
+    >>> from transformers import AutoTokenizer, Qwen2MoeForCausalLM
+
+    >>> model = Qwen2MoeForCausalLM.from_pretrained(PATH_TO_CONVERTED_WEIGHTS)
+    >>> tokenizer = AutoTokenizer.from_pretrained(PATH_TO_CONVERTED_TOKENIZER)
+
+    >>> prompt = "Hey, are you conscious? Can you talk to me?"
+    >>> inputs = tokenizer(prompt, return_tensors="pt")
+
+    >>> # Generate
+    >>> generate_ids = model.generate(inputs.input_ids, max_length=30)
+    >>> tokenizer.batch_decode(generate_ids, skip_special_tokens=True, clean_up_tokenization_spaces=False)[0]
+    "Hey, are you conscious? Can you talk to me?\nI'm not conscious, but I can talk to you."
+    ```"""
+
+    output_attentions = (
+        output_attentions
+        if output_attentions is not None
+        else self.config.output_attentions
+    )
+    output_router_logits = (
+        output_router_logits
+        if output_router_logits is not None
+        else self.config.output_router_logits
+    )
+    output_hidden_states = (
+        output_hidden_states
+        if output_hidden_states is not None
+        else self.config.output_hidden_states
+    )
+
+    # decoder outputs consists of (dec_features, layer_state, dec_hidden, dec_attn)
+    outputs: MoeModelOutputWithPast = self.model(
+        input_ids=input_ids,
+        attention_mask=attention_mask,
+        position_ids=position_ids,
+        past_key_values=past_key_values,
+        inputs_embeds=inputs_embeds,
+        use_cache=use_cache,
+        output_attentions=output_attentions,
+        output_hidden_states=output_hidden_states,
+        output_router_logits=output_router_logits,
+        cache_position=cache_position,
+    )
+
+    hidden_states = outputs.last_hidden_state
+    loss = None
+    logits = None
+
+    if hidden_states is None:
+        raise ValueError("hidden_states is None")
+
+    # Only compute necessary logits, and do not upcast them to float if we are not computing the loss
+    slice_indices = (
+        slice(-logits_to_keep, None)
+        if isinstance(logits_to_keep, int)
+        else logits_to_keep
+    )
+
+    if _PATCH_OPTS is not None and _PATCH_OPTS.use_lce(labels, self.training):
+        assert labels is not None
+        loss = apply_lce(
+            hidden_states[:, slice_indices, :],
+            self.lm_head.weight,
+            labels,
+            _PATCH_OPTS,
+            **loss_kwargs,
+        )
+    else:
+        logits = self.lm_head(hidden_states[:, slice_indices, :])
+
+        if labels is not None:
+            loss = self.loss_function(logits, labels, self.vocab_size, **loss_kwargs)
+
+    aux_loss = None
+    if output_router_logits:
+        aux_loss = load_balancing_loss_func(
+            outputs.router_logits,
+            self.num_experts,
+            self.num_experts_per_tok,
+            attention_mask,
+        )
+        if labels is not None:
+            loss += self.router_aux_loss_coef * aux_loss.to(  # type: ignore
+                loss.device  # type: ignore
+            )  # make sure to reside in the same device
+
+    return MoeCausalLMOutputWithPast(
+        loss=loss,
+        aux_loss=aux_loss,  # type: ignore
+        logits=logits,
+        past_key_values=outputs.past_key_values,
+        hidden_states=outputs.hidden_states,
+        attentions=outputs.attentions,
+        router_logits=outputs.router_logits,
+    )
+
+
+def patch_qwen2_moe(
+    maybe_model: TransformersModelT | str | transformers.PretrainedConfig,
+    patch_options: PatchOptions,
+) -> TransformersModelT | None:
+    global _PATCH_OPTS  # pylint: disable=global-statement
+
+    from transformers.models.qwen2_moe import modeling_qwen2_moe
+
+    _PATCH_OPTS = patch_options
+
+    if isinstance(maybe_model, transformers.PreTrainedModel):
+        assert isinstance(
+            maybe_model, modeling_qwen2_moe.Qwen2MoeForCausalLM
+        ), f"Expected a Qwen3MoeForCausalLM model. Got {type(maybe_model)}."
+        maybe_model.forward = MethodType(forward, maybe_model)
+
+        return maybe_model
+
+    modeling_qwen2_moe.Qwen2MoeForCausalLM.forward = forward
+    return None

src/axolotl/integrations/cut_cross_entropy/monkeypatch/qwen2_vl.py

@@ -0,0 +1,249 @@
+"""Qwen2 VL CCE patch. Adapted from transformers v4.51.2"""
+
+# pylint: disable=duplicate-code
+
+from types import MethodType
+from typing import Optional, Tuple, Union
+
+import torch
+import transformers
+from cut_cross_entropy.transformers.utils import (
+    PatchOptions,
+    TransformersModelT,
+    apply_lce,
+)
+from torch.nn import CrossEntropyLoss
+from transformers.models.qwen2_vl.modeling_qwen2_vl import (
+    _CONFIG_FOR_DOC,
+    QWEN2_VL_INPUTS_DOCSTRING,
+    Qwen2VLCausalLMOutputWithPast,
+)
+from transformers.utils import (
+    add_start_docstrings_to_model_forward,
+    replace_return_docstrings,
+)
+
+_PATCH_OPTS: PatchOptions | None = None
+
+
+@add_start_docstrings_to_model_forward(QWEN2_VL_INPUTS_DOCSTRING)
+@replace_return_docstrings(
+    output_type=Qwen2VLCausalLMOutputWithPast, config_class=_CONFIG_FOR_DOC
+)
+def cce_forward_multimodal(
+    self,
+    input_ids: Optional[torch.LongTensor] = None,
+    attention_mask: Optional[torch.Tensor] = None,
+    position_ids: Optional[torch.LongTensor] = None,
+    past_key_values: Optional[list[torch.FloatTensor]] = None,
+    inputs_embeds: Optional[torch.FloatTensor] = None,
+    labels: Optional[torch.LongTensor] = None,
+    use_cache: Optional[bool] = None,
+    output_attentions: Optional[bool] = None,
+    output_hidden_states: Optional[bool] = None,
+    return_dict: Optional[bool] = None,
+    pixel_values: Optional[torch.Tensor] = None,
+    pixel_values_videos: Optional[torch.FloatTensor] = None,
+    image_grid_thw: Optional[torch.LongTensor] = None,
+    video_grid_thw: Optional[torch.LongTensor] = None,
+    rope_deltas: Optional[torch.LongTensor] = None,
+    cache_position: Optional[torch.LongTensor] = None,
+) -> Union[Tuple, Qwen2VLCausalLMOutputWithPast]:
+    r"""
+        labels (`torch.LongTensor` of shape `(batch_size, sequence_length)`, *optional*):
+            Labels for computing the masked language modeling loss. Indices should either be in `[0, ...,
+            config.vocab_size]` or -100 (see `input_ids` docstring). Tokens with indices set to `-100` are ignored
+            (masked), the loss is only computed for the tokens with labels in `[0, ..., config.vocab_size]`.
+
+    Returns:
+
+    Example:
+
+    ```python
+    >>> from PIL import Image
+    >>> import requests
+    >>> from transformers import AutoProcessor, Qwen2VLForConditionalGeneration
+
+    >>> model = Qwen2VLForConditionalGeneration.from_pretrained("Qwen/Qwen2-VL-7B-Instruct")
+    >>> processor = AutoProcessor.from_pretrained("Qwen/Qwen2-VL-7B-Instruct")
+
+    >>> messages = [
+        {
+            "role": "user",
+            "content": [
+                {"type": "image"},
+                {"type": "text", "text": "What is shown in this image?"},
+            ],
+        },
+    ]
+    >>> url = "https://www.ilankelman.org/stopsigns/australia.jpg"
+    >>> image = Image.open(requests.get(url, stream=True).raw)
+
+    >>> text = processor.apply_chat_template(messages, tokenize=False, add_generation_prompt=True)
+    >>> inputs = processor(text=[text], images=[image], vision_infos=[vision_infos])
+
+    >>> # Generate
+    >>> generate_ids = model.generate(inputs.input_ids, max_length=30)
+    >>> tokenizer.batch_decode(generate_ids, skip_special_tokens=True, clean_up_tokenization_spaces=False)[0]
+    "The image shows a street scene with a red stop sign in the foreground. In the background, there is a large red gate with Chinese characters ..."
+    ```"""
+
+    output_attentions = (
+        output_attentions
+        if output_attentions is not None
+        else self.config.output_attentions
+    )
+    output_hidden_states = (
+        output_hidden_states
+        if output_hidden_states is not None
+        else self.config.output_hidden_states
+    )
+    return_dict = (
+        return_dict if return_dict is not None else self.config.use_return_dict
+    )
+
+    if inputs_embeds is None:
+        inputs_embeds = self.model.embed_tokens(input_ids)
+        if pixel_values is not None:
+            pixel_values = pixel_values.type(self.visual.get_dtype())
+            image_embeds = self.visual(pixel_values, grid_thw=image_grid_thw)
+            n_image_tokens = (input_ids == self.config.image_token_id).sum().item()
+            n_image_features = image_embeds.shape[0]
+            if n_image_tokens != n_image_features:
+                raise ValueError(
+                    f"Image features and image tokens do not match: tokens: {n_image_tokens}, features {n_image_features}"
+                )
+            image_mask = (
+                (input_ids == self.config.image_token_id)
+                .unsqueeze(-1)
+                .expand_as(inputs_embeds)
+                .to(inputs_embeds.device)
+            )
+            image_embeds = image_embeds.to(inputs_embeds.device, inputs_embeds.dtype)
+            inputs_embeds = inputs_embeds.masked_scatter(image_mask, image_embeds)  # type: ignore
+
+        if pixel_values_videos is not None:
+            pixel_values_videos = pixel_values_videos.type(self.visual.get_dtype())
+            video_embeds = self.visual(pixel_values_videos, grid_thw=video_grid_thw)
+            n_video_tokens = (input_ids == self.config.video_token_id).sum().item()
+            n_video_features = video_embeds.shape[0]
+            if n_video_tokens != n_video_features:
+                raise ValueError(
+                    f"Video features and video tokens do not match: tokens: {n_video_tokens}, features {n_video_features}"
+                )
+            video_mask = (
+                (input_ids == self.config.video_token_id)
+                .unsqueeze(-1)
+                .expand_as(inputs_embeds)
+                .to(inputs_embeds.device)
+            )
+            video_embeds = video_embeds.to(inputs_embeds.device, inputs_embeds.dtype)
+            inputs_embeds = inputs_embeds.masked_scatter(video_mask, video_embeds)  # type: ignore
+
+        if attention_mask is not None:
+            attention_mask = attention_mask.to(inputs_embeds.device)
+
+    # if we get 4D attention mask we cannot calculate rope deltas anymore. TODO @raushan fixme
+    if position_ids is None and (attention_mask is None or attention_mask.ndim == 2):
+        # calculate RoPE index once per generation in the pre-fill stage only
+        if (
+            (cache_position is not None and cache_position[0] == 0)
+            or self.rope_deltas is None
+            or (past_key_values is None or past_key_values.get_seq_length() == 0)  # type: ignore
+        ):
+            position_ids, rope_deltas = self.get_rope_index(
+                input_ids, image_grid_thw, video_grid_thw, attention_mask
+            )
+            self.rope_deltas = rope_deltas
+        # then use the prev pre-calculated rope-deltas to get the correct position ids
+        else:
+            batch_size, seq_length, _ = inputs_embeds.shape
+            delta = (
+                cache_position[0] + self.rope_deltas
+                if cache_position is not None
+                else 0
+            )
+            position_ids = torch.arange(seq_length, device=inputs_embeds.device)  # type: ignore
+            position_ids = position_ids.view(1, -1).expand(batch_size, -1)  # type: ignore
+            if cache_position is not None:  # otherwise `deltas` is an int `0`
+                delta = delta.repeat_interleave(batch_size // delta.shape[0], dim=0)  # type: ignore
+                delta = delta.to(position_ids.device)  # type: ignore
+            position_ids = position_ids.add(delta)  # type: ignore
+            position_ids = position_ids.unsqueeze(0).expand(3, -1, -1)  # type: ignore
+
+    outputs = self.model(
+        input_ids=None,
+        position_ids=position_ids,
+        attention_mask=attention_mask,
+        past_key_values=past_key_values,
+        inputs_embeds=inputs_embeds,
+        use_cache=use_cache,
+        output_attentions=output_attentions,
+        output_hidden_states=output_hidden_states,
+        return_dict=return_dict,
+        cache_position=cache_position,
+    )
+
+    hidden_states = outputs[0]
+    logits = None
+    loss = None
+
+    if _PATCH_OPTS is not None and _PATCH_OPTS.use_lce(labels, self.training):
+        assert labels is not None
+        loss = apply_lce(
+            hidden_states,
+            self.lm_head.weight,
+            labels,
+            _PATCH_OPTS,
+        )
+    else:
+        logits = self.lm_head(hidden_states)
+
+        if labels is not None:
+            # Upcast to float if we need to compute the loss to avoid potential precision issues
+            logits = logits.float()
+            # Shift so that tokens < n predict n
+            shift_logits = logits[..., :-1, :].contiguous()
+            shift_labels = labels[..., 1:].contiguous()
+            # Flatten the tokens
+            loss_fct = CrossEntropyLoss()
+            shift_logits = shift_logits.view(-1, self.config.vocab_size)
+            shift_labels = shift_labels.view(-1)
+            # Enable model parallelism
+            shift_labels = shift_labels.to(shift_logits.device)
+            loss = loss_fct(shift_logits, shift_labels)
+
+    if not return_dict:
+        output = (logits,) + outputs[1:]
+        return (loss,) + output if loss is not None else output
+
+    return Qwen2VLCausalLMOutputWithPast(
+        loss=loss,
+        logits=logits,
+        past_key_values=outputs.past_key_values,
+        hidden_states=outputs.hidden_states,
+        attentions=outputs.attentions,
+        rope_deltas=self.rope_deltas,
+    )
+
+
+def patch_qwen2_vl(
+    maybe_model: TransformersModelT | str | transformers.PretrainedConfig,
+    patch_options: PatchOptions,
+) -> TransformersModelT | None:
+    global _PATCH_OPTS  # pylint: disable=global-statement
+
+    from transformers.models.qwen2_vl import modeling_qwen2_vl
+
+    _PATCH_OPTS = patch_options
+
+    if isinstance(maybe_model, transformers.PreTrainedModel):
+        assert isinstance(
+            maybe_model, modeling_qwen2_vl.Qwen2VLForConditionalGeneration
+        ), f"Expected a Qwen2VLForConditionalGeneration model. Got {type(maybe_model)}."
+        maybe_model.forward = MethodType(cce_forward_multimodal, maybe_model)
+
+        return maybe_model
+
+    modeling_qwen2_vl.Qwen2VLForConditionalGeneration.forward = cce_forward_multimodal
+    return None

src/axolotl/integrations/cut_cross_entropy/monkeypatch/qwen3.py

@@ -0,0 +1,35 @@
+"""Qwen3 CCE patch. The model inherits Llama's modeling code and uses the same forward method."""
+
+# pylint: disable=duplicate-code
+
+from types import MethodType
+
+import transformers
+from cut_cross_entropy.transformers.utils import (
+    PatchOptions,
+    TransformersModelT,
+)
+
+
+def patch_qwen3(
+    maybe_model: TransformersModelT | str | transformers.PretrainedConfig,
+    patch_options: PatchOptions,
+) -> TransformersModelT | None:
+    from transformers.models.qwen3 import modeling_qwen3
+
+    # Set the _PATCH_OPTS in the llama patch file
+    import axolotl.integrations.cut_cross_entropy.monkeypatch.llama as llama_patch
+
+    llama_patch._PATCH_OPTS = patch_options  # pylint: disable=protected-access
+
+    from axolotl.integrations.cut_cross_entropy.monkeypatch.llama import cce_forward
+
+    if isinstance(maybe_model, transformers.PreTrainedModel):
+        assert isinstance(
+            maybe_model, modeling_qwen3.Qwen3ForCausalLM
+        ), f"Expected a Qwen3ForCausalLM model. Got {type(maybe_model)}."
+        maybe_model.forward = MethodType(cce_forward, maybe_model)
+        return maybe_model
+
+    modeling_qwen3.Qwen3ForCausalLM.forward = cce_forward
+    return None

src/axolotl/integrations/cut_cross_entropy/monkeypatch/qwen3_moe.py

@@ -0,0 +1,194 @@
+"""Qwen3 MoE CCE patch. Adapted from transformers v4.51.2"""
+
+# pylint: disable=duplicate-code
+
+from types import MethodType
+from typing import Optional, Union
+
+import torch
+import transformers
+from cut_cross_entropy.transformers.utils import (
+    PatchOptions,
+    TransformersModelT,
+    apply_lce,
+)
+from transformers.modeling_outputs import CausalLMOutputWithPast
+from transformers.models.qwen3_moe.modeling_qwen3_moe import (
+    _CONFIG_FOR_DOC,
+    QWEN3_MOE_INPUTS_DOCSTRING,
+    KwargsForCausalLM,
+    MoeCausalLMOutputWithPast,
+    MoeModelOutputWithPast,
+    load_balancing_loss_func,
+)
+from transformers.processing_utils import Unpack
+from transformers.utils import (
+    add_start_docstrings_to_model_forward,
+    replace_return_docstrings,
+)
+from transformers.utils.deprecation import deprecate_kwarg
+from transformers.utils.generic import can_return_tuple
+
+_PATCH_OPTS: PatchOptions | None = None
+
+
+@can_return_tuple
+@deprecate_kwarg("num_logits_to_keep", version="4.50", new_name="logits_to_keep")
+@add_start_docstrings_to_model_forward(QWEN3_MOE_INPUTS_DOCSTRING)
+@replace_return_docstrings(
+    output_type=CausalLMOutputWithPast, config_class=_CONFIG_FOR_DOC
+)
+def forward(
+    self,
+    input_ids: Optional[torch.LongTensor] = None,
+    attention_mask: Optional[torch.Tensor] = None,
+    position_ids: Optional[torch.LongTensor] = None,
+    past_key_values: Optional[list[torch.FloatTensor]] = None,
+    inputs_embeds: Optional[torch.FloatTensor] = None,
+    labels: Optional[torch.LongTensor] = None,
+    use_cache: Optional[bool] = None,
+    output_attentions: Optional[bool] = None,
+    output_hidden_states: Optional[bool] = None,
+    output_router_logits: Optional[bool] = None,
+    cache_position: Optional[torch.LongTensor] = None,
+    logits_to_keep: Union[int, torch.Tensor] = 0,
+    **kwargs: Unpack[KwargsForCausalLM],
+) -> MoeCausalLMOutputWithPast:
+    r"""
+        labels (`torch.LongTensor` of shape `(batch_size, sequence_length)`, *optional*):
+            Labels for computing the masked language modeling loss. Indices should either be in `[0, ...,
+            config.vocab_size]` or -100 (see `input_ids` docstring). Tokens with indices set to `-100` are ignored
+            (masked), the loss is only computed for the tokens with labels in `[0, ..., config.vocab_size]`.
+
+        logits_to_keep (`int` or `torch.Tensor`, *optional*):
+            If an `int`, compute logits for the last `logits_to_keep` tokens. If `0`, calculate logits for all
+            `input_ids` (special case). Only last token logits are needed for generation, and calculating them only for that
+            token can save memory, which becomes pretty significant for long sequences or large vocabulary size.
+            If a `torch.Tensor`, must be 1D corresponding to the indices to keep in the sequence length dimension.
+            This is useful when using packed tensor format (single dimension for batch and sequence length).
+
+    Returns:
+
+    Example:
+
+    ```python
+    >>> from transformers import AutoTokenizer, Qwen3MoeForCausalLM
+
+    >>> model = Qwen3MoeForCausalLM.from_pretrained("Qwen/Qwen3-MoE-15B-A2B")
+    >>> tokenizer = AutoTokenizer.from_pretrained("Qwen/Qwen3-MoE-15B-A2B")
+
+    >>> prompt = "Hey, are you conscious? Can you talk to me?"
+    >>> inputs = tokenizer(prompt, return_tensors="pt")
+
+    >>> # Generate
+    >>> generate_ids = model.generate(inputs.input_ids, max_length=30)
+    >>> tokenizer.batch_decode(generate_ids, skip_special_tokens=True, clean_up_tokenization_spaces=False)[0]
+    "Hey, are you conscious? Can you talk to me?\nI'm not conscious, but I can talk to you."
+    ```"""
+
+    output_attentions = (
+        output_attentions
+        if output_attentions is not None
+        else self.config.output_attentions
+    )
+    output_router_logits = (
+        output_router_logits
+        if output_router_logits is not None
+        else self.config.output_router_logits
+    )
+
+    output_hidden_states = (
+        output_hidden_states
+        if output_hidden_states is not None
+        else self.config.output_hidden_states
+    )
+
+    # decoder outputs consists of (dec_features, layer_state, dec_hidden, dec_attn)
+    outputs: MoeModelOutputWithPast = self.model(
+        input_ids=input_ids,
+        attention_mask=attention_mask,
+        position_ids=position_ids,
+        past_key_values=past_key_values,
+        inputs_embeds=inputs_embeds,
+        use_cache=use_cache,
+        output_attentions=output_attentions,
+        output_hidden_states=output_hidden_states,
+        output_router_logits=output_router_logits,
+        cache_position=cache_position,
+        **kwargs,
+    )
+
+    hidden_states = outputs.last_hidden_state
+
+    if hidden_states is None:
+        raise ValueError("hidden_states is None")
+
+    loss = None
+    logits = None
+
+    # Only compute necessary logits, and do not upcast them to float if we are not computing the loss
+    slice_indices = (
+        slice(-logits_to_keep, None)
+        if isinstance(logits_to_keep, int)
+        else logits_to_keep
+    )
+
+    if _PATCH_OPTS is not None and _PATCH_OPTS.use_lce(labels, self.training):
+        assert labels is not None
+        loss = apply_lce(
+            hidden_states[:, slice_indices, :],
+            self.lm_head.weight,
+            labels,
+            _PATCH_OPTS,
+            **kwargs,
+        )
+    else:
+        logits = self.lm_head(hidden_states[:, slice_indices, :])
+
+        if labels is not None:
+            loss = self.loss_function(logits, labels, self.vocab_size, **kwargs)
+
+    aux_loss = None
+    if output_router_logits:
+        aux_loss = load_balancing_loss_func(
+            outputs.router_logits,
+            self.num_experts,
+            self.num_experts_per_tok,
+            attention_mask,
+        )
+        if labels is not None:
+            loss += self.router_aux_loss_coef * aux_loss.to(  # type: ignore
+                loss.device  # type: ignore
+            )  # make sure to reside in the same device
+
+    return MoeCausalLMOutputWithPast(
+        loss=loss,
+        aux_loss=aux_loss,  # type: ignore
+        logits=logits,
+        past_key_values=outputs.past_key_values,
+        hidden_states=outputs.hidden_states,
+        attentions=outputs.attentions,
+        router_logits=outputs.router_logits,
+    )
+
+
+def patch_qwen3_moe(
+    maybe_model: TransformersModelT | str | transformers.PretrainedConfig,
+    patch_options: PatchOptions,
+) -> TransformersModelT | None:
+    global _PATCH_OPTS  # pylint: disable=global-statement
+
+    from transformers.models.qwen3_moe import modeling_qwen3_moe
+
+    _PATCH_OPTS = patch_options
+
+    if isinstance(maybe_model, transformers.PreTrainedModel):
+        assert isinstance(
+            maybe_model, modeling_qwen3_moe.Qwen3MoeForCausalLM
+        ), f"Expected a Qwen3MoeForCausalLM model. Got {type(maybe_model)}."
+        maybe_model.forward = MethodType(forward, maybe_model)
+
+        return maybe_model
+
+    modeling_qwen3_moe.Qwen3MoeForCausalLM.forward = forward
+    return None

src/axolotl/integrations/kd/chat_template.py

@@ -35,6 +35,9 @@ class ChatTemplateStrategyWithKD(ChatTemplateStrategy):
         sequence_len,
         roles_to_train=None,
         train_on_eos=None,
+        train_on_eot=None,
+        eot_tokens=None,
+        split_thinking: bool | None = False,
         logprobs_field="logprobs",
         gen_temperature=1.0,
         kd_temperature=1.0,
@@ -50,6 +53,9 @@ class ChatTemplateStrategyWithKD(ChatTemplateStrategy):
             sequence_len,
             roles_to_train=roles_to_train,
             train_on_eos=train_on_eos,
+            train_on_eot=train_on_eot,
+            eot_tokens=eot_tokens,
+            split_thinking=split_thinking,
         )
 
     @property

src/axolotl/integrations/liger/__init__.py

@@ -23,8 +23,8 @@ import logging
 import sys
 
 from axolotl.integrations.base import BasePlugin
+from axolotl.utils.distributed import is_main_process
 
-from ...utils.distributed import zero_only
 from .args import LigerArgs  # pylint: disable=unused-import. # noqa: F401
 from .utils import patch_with_compile_disable
 
@@ -85,7 +85,7 @@ class LigerPlugin(BasePlugin):
                 kwargs["geglu"] = cfg.liger_glu_activation
             elif "swiglu" in liger_fn_sig.parameters:
                 kwargs["swiglu"] = cfg.liger_glu_activation
-            with zero_only():
+            if is_main_process(use_environ=True):
                 LOG.info(
                     f"Applying LIGER to {cfg.model_config_type} with kwargs: {kwargs}"
                 )
@@ -151,6 +151,30 @@ class LigerPlugin(BasePlugin):
                 rms_norm=cfg.liger_rms_norm,
                 layer_norm=cfg.liger_layer_norm,
             )
+        elif cfg.model_config_type == "qwen3":
+            from axolotl.integrations.liger.models.qwen3 import (
+                apply_liger_kernel_to_qwen3,
+            )
+
+            apply_liger_kernel_to_qwen3(
+                cross_entropy=cfg.liger_cross_entropy,
+                fused_linear_cross_entropy=cfg.liger_fused_linear_cross_entropy,
+                glu_activation=cfg.liger_glu_activation,
+                rms_norm=cfg.liger_rms_norm,
+                layer_norm=cfg.liger_layer_norm,
+            )
+        elif cfg.model_config_type == "qwen3_moe":
+            from axolotl.integrations.liger.models.qwen3_moe import (
+                apply_liger_kernel_to_qwen3_moe,
+            )
+
+            apply_liger_kernel_to_qwen3_moe(
+                cross_entropy=cfg.liger_cross_entropy,
+                fused_linear_cross_entropy=cfg.liger_fused_linear_cross_entropy,
+                glu_activation=cfg.liger_glu_activation,
+                rms_norm=cfg.liger_rms_norm,
+                layer_norm=cfg.liger_layer_norm,
+            )
         else:
             logging.warning(
                 f"Unsupported model config type: {cfg.model_config_type}. Liger not applied."

src/axolotl/integrations/liger/models/qwen3.py

@@ -0,0 +1,160 @@
+"""
+Liger FLCE for Qwen3. Based on transformers v4.51.3.
+"""
+
+import sys
+from typing import Optional, Tuple, Union
+
+import torch
+from liger_kernel.transformers.model.loss_utils import LigerForCausalLMLoss
+from transformers.cache_utils import Cache
+from transformers.modeling_outputs import CausalLMOutputWithPast
+
+
+def lce_forward(
+    self,
+    input_ids: Optional[torch.LongTensor] = None,
+    attention_mask: Optional[torch.Tensor] = None,
+    position_ids: Optional[torch.LongTensor] = None,
+    past_key_values: Optional[Cache] = None,
+    inputs_embeds: Optional[torch.FloatTensor] = None,
+    labels: Optional[torch.LongTensor] = None,
+    use_cache: Optional[bool] = None,
+    output_attentions: Optional[bool] = None,
+    output_hidden_states: Optional[bool] = None,
+    cache_position: Optional[torch.LongTensor] = None,
+    logits_to_keep: Union[int, torch.Tensor] = 0,
+    **kwargs,
+) -> Union[Tuple, CausalLMOutputWithPast]:
+    r"""
+    Args:
+        labels (`torch.LongTensor` of shape `(batch_size, sequence_length)`, *optional*):
+            Labels for computing the masked language modeling loss. Indices should either be in `[0, ...,
+            config.vocab_size]` or -100 (see `input_ids` docstring). Tokens with indices set to `-100` are ignored
+            (masked), the loss is only computed for the tokens with labels in `[0, ..., config.vocab_size]`.
+
+        logits_to_keep (`int` or `torch.Tensor`, *optional*):
+            If an `int`, compute logits for the last `logits_to_keep` tokens. If `0`, calculate logits for all
+            `input_ids` (special case). Only last token logits are needed for generation, and calculating them only for that
+            token can save memory, which becomes pretty significant for long sequences or large vocabulary size.
+            If a `torch.Tensor`, must be 1D corresponding to the indices to keep in the sequence length dimension.
+            This is useful when using packed tensor format (single dimension for batch and sequence length).
+
+    Returns:
+    """
+
+    # pylint: disable=duplicate-code
+    output_attentions = (
+        output_attentions
+        if output_attentions is not None
+        else self.config.output_attentions
+    )
+    output_hidden_states = (
+        output_hidden_states
+        if output_hidden_states is not None
+        else self.config.output_hidden_states
+    )
+
+    # decoder outputs consists of (dec_features, layer_state, dec_hidden, dec_attn)
+    outputs = self.model(
+        input_ids=input_ids,
+        attention_mask=attention_mask,
+        position_ids=position_ids,
+        past_key_values=past_key_values,
+        inputs_embeds=inputs_embeds,
+        use_cache=use_cache,
+        output_attentions=output_attentions,
+        output_hidden_states=output_hidden_states,
+        cache_position=cache_position,
+        **kwargs,
+    )
+
+    hidden_states = outputs[0]
+
+    logits = None
+    loss = None
+    # if in training mode, don't materialize logits
+    if self.training and (labels is not None):
+        loss = LigerForCausalLMLoss(
+            hidden_states=hidden_states,
+            lm_head_weight=self.lm_head.weight,
+            labels=labels,
+            hidden_size=self.config.hidden_size,
+            **kwargs,
+        )
+
+    else:  # if in inference mode materialize logits
+        slice_indices = (
+            slice(-logits_to_keep, None)
+            if isinstance(logits_to_keep, int)
+            else logits_to_keep
+        )
+        logits = self.lm_head(hidden_states[:, slice_indices, :])
+        if labels is not None:
+            loss = self.loss_function(
+                logits=logits,
+                labels=labels,
+                vocab_size=self.config.vocab_size,
+                **kwargs,
+            )
+
+    return CausalLMOutputWithPast(
+        loss=loss,
+        logits=logits,
+        past_key_values=outputs.past_key_values,
+        hidden_states=outputs.hidden_states,
+        attentions=outputs.attentions,
+    )
+
+
+def apply_liger_kernel_to_qwen3(
+    cross_entropy: bool = False,
+    fused_linear_cross_entropy: bool = False,
+    rms_norm: bool = False,
+    glu_activation: bool = False,
+    layer_norm: bool = False,
+    **kwargs,  # pylint: disable=unused-argument
+) -> None:
+    # pylint: disable=duplicate-code
+    """
+    Apply Liger kernels to replace original implementation in HuggingFace Llama models (2 and 3)
+
+    Args:
+        cross_entropy (bool): Whether to apply Liger's cross entropy loss. Default is False.
+        fused_linear_cross_entropy (bool):
+            Whether to apply Liger's fused linear cross entropy loss. Default is False.
+            `cross_entropy` and `fused_linear_cross_entropy` cannot both be False.
+            If `fused_linear_cross_entropy` is True, the logits will not be materialized but more memory efficient.
+        rms_norm (bool): Whether to apply Liger's RMSNorm. Default is False.
+        glu_activation (bool): Whether to apply Liger's SwiGLU MLP. Default is False.
+        layer_norm (bool): Whether to apply Liger's LayerNorm. Default is False.
+    """
+
+    import transformers.models.qwen3.modeling_qwen3  # noqa: F401  # pylint: disable=unused-import
+    from liger_kernel.transformers.functional import liger_cross_entropy
+    from liger_kernel.transformers.layer_norm import LigerLayerNorm
+    from liger_kernel.transformers.rms_norm import LigerRMSNorm
+    from liger_kernel.transformers.swiglu import LigerSwiGLUMLP
+
+    assert not (
+        cross_entropy and fused_linear_cross_entropy
+    ), "cross_entropy and fused_linear_cross_entropy cannot both be True."
+
+    modeling_qwen3 = sys.modules["transformers.models.qwen3.modeling_qwen3"]
+
+    if rms_norm:
+        modeling_qwen3.Qwen3RMSNorm = LigerRMSNorm
+
+    if glu_activation:
+        modeling_qwen3.Qwen3MLP = LigerSwiGLUMLP
+
+    if layer_norm:
+        modeling_qwen3.nn.LayerNorm = LigerLayerNorm
+
+    if cross_entropy:
+        from transformers.loss.loss_utils import nn
+
+        nn.functional.cross_entropy = liger_cross_entropy
+
+    if fused_linear_cross_entropy:
+        modeling_qwen3.Qwen3ForCausalLM.forward = lce_forward

src/axolotl/integrations/liger/models/qwen3_moe.py

@@ -0,0 +1,191 @@
+"""
+Liger FLCE for Qwen3 MoE. Based on transformers v4.51.3.
+"""
+
+import sys
+from copy import deepcopy
+from typing import List, Optional, Union
+
+import torch
+from liger_kernel.transformers.model.loss_utils import LigerForCausalLMLoss
+from transformers.modeling_outputs import MoeCausalLMOutputWithPast
+from transformers.models.qwen3_moe.modeling_qwen3_moe import load_balancing_loss_func
+
+
+def lce_forward(
+    self,
+    input_ids: Optional[torch.LongTensor] = None,
+    attention_mask: Optional[torch.Tensor] = None,
+    position_ids: Optional[torch.LongTensor] = None,
+    past_key_values: Optional[List[torch.FloatTensor]] = None,
+    inputs_embeds: Optional[torch.FloatTensor] = None,
+    labels: Optional[torch.LongTensor] = None,
+    use_cache: Optional[bool] = None,
+    output_attentions: Optional[bool] = None,
+    output_hidden_states: Optional[bool] = None,
+    output_router_logits: Optional[bool] = None,
+    cache_position: Optional[torch.LongTensor] = None,
+    logits_to_keep: Union[int, torch.Tensor] = 0,
+    **kwargs,
+) -> MoeCausalLMOutputWithPast:
+    r"""
+    Args:
+        labels (`torch.LongTensor` of shape `(batch_size, sequence_length)`, *optional*):
+            Labels for computing the masked language modeling loss. Indices should either be in `[0, ...,
+            config.vocab_size]` or -100 (see `input_ids` docstring). Tokens with indices set to `-100` are ignored
+            (masked), the loss is only computed for the tokens with labels in `[0, ..., config.vocab_size]`.
+
+        logits_to_keep (`int` or `torch.Tensor`, *optional*):
+            If an `int`, compute logits for the last `logits_to_keep` tokens. If `0`, calculate logits for all
+            `input_ids` (special case). Only last token logits are needed for generation, and calculating them only for that
+            token can save memory, which becomes pretty significant for long sequences or large vocabulary size.
+            If a `torch.Tensor`, must be 1D corresponding to the indices to keep in the sequence length dimension.
+            This is useful when using packed tensor format (single dimension for batch and sequence length).
+
+    Returns:
+    """
+
+    # pylint: disable=duplicate-code
+    output_attentions = (
+        output_attentions
+        if output_attentions is not None
+        else self.config.output_attentions
+    )
+    output_router_logits = (
+        output_router_logits
+        if output_router_logits is not None
+        else self.config.output_router_logits
+    )
+    output_hidden_states = (
+        output_hidden_states
+        if output_hidden_states is not None
+        else self.config.output_hidden_states
+    )
+
+    # decoder outputs consists of (dec_features, layer_state, dec_hidden, dec_attn)
+    outputs = self.model(
+        input_ids=input_ids,
+        attention_mask=attention_mask,
+        position_ids=position_ids,
+        past_key_values=past_key_values,
+        inputs_embeds=inputs_embeds,
+        use_cache=use_cache,
+        output_attentions=output_attentions,
+        output_hidden_states=output_hidden_states,
+        output_router_logits=output_router_logits,
+        cache_position=cache_position,
+        **kwargs,
+    )
+
+    hidden_states = outputs[0]
+
+    logits = None
+    loss = None
+    # if in training mode, don't materialize logits
+    if self.training and (labels is not None):
+        loss = LigerForCausalLMLoss(
+            hidden_states=hidden_states,
+            lm_head_weight=self.lm_head.weight,
+            labels=labels,
+            hidden_size=self.config.hidden_size,
+            **kwargs,
+        )
+
+    else:  # if in inference mode materialize logits
+        slice_indices = (
+            slice(-logits_to_keep, None)
+            if isinstance(logits_to_keep, int)
+            else logits_to_keep
+        )
+        logits = self.lm_head(hidden_states[:, slice_indices, :])
+        if labels is not None:
+            loss = self.loss_function(
+                logits=logits,
+                labels=labels,
+                vocab_size=self.config.vocab_size,
+                **kwargs,
+            )
+
+    aux_loss = None
+    if output_router_logits:
+        aux_loss = load_balancing_loss_func(
+            outputs.router_logits,
+            self.num_experts,
+            self.num_experts_per_tok,
+            attention_mask,
+        )
+        if labels is not None:
+            loss += self.router_aux_loss_coef * aux_loss.to(
+                loss.device
+            )  # make sure to reside in the same device
+
+    return MoeCausalLMOutputWithPast(
+        loss=loss,
+        aux_loss=aux_loss,
+        logits=logits,
+        past_key_values=outputs.past_key_values,
+        hidden_states=outputs.hidden_states,
+        attentions=outputs.attentions,
+    )
+
+
+def apply_liger_kernel_to_qwen3_moe(
+    cross_entropy: bool = False,
+    fused_linear_cross_entropy: bool = False,
+    rms_norm: bool = False,
+    glu_activation: bool = False,
+    layer_norm: bool = False,
+    **kwargs,  # pylint: disable=unused-argument
+) -> None:
+    # pylint: disable=duplicate-code
+    """
+    Apply Liger kernels to replace original implementation in HuggingFace Llama models (2 and 3)
+
+    Args:
+        cross_entropy (bool): Whether to apply Liger's cross entropy loss. Default is False.
+        fused_linear_cross_entropy (bool):
+            Whether to apply Liger's fused linear cross entropy loss. Default is False.
+            `cross_entropy` and `fused_linear_cross_entropy` cannot both be False.
+            If `fused_linear_cross_entropy` is True, the logits will not be materialized but more memory efficient.
+        rms_norm (bool): Whether to apply Liger's RMSNorm. Default is False.
+        glu_activation (bool): Whether to apply Liger's SwiGLU MLP. Default is False.
+        layer_norm (bool): Whether to apply Liger's LayerNorm. Default is False.
+    """
+
+    import transformers.models.qwen3_moe.modeling_qwen3_moe  # noqa: F401  # pylint: disable=unused-import
+    from liger_kernel.transformers.functional import liger_cross_entropy
+    from liger_kernel.transformers.layer_norm import LigerLayerNorm
+    from liger_kernel.transformers.rms_norm import LigerRMSNorm
+    from liger_kernel.transformers.swiglu import LigerSwiGLUMLP
+
+    assert not (
+        cross_entropy and fused_linear_cross_entropy
+    ), "cross_entropy and fused_linear_cross_entropy cannot both be True."
+
+    modeling_qwen3_moe = sys.modules["transformers.models.qwen3_moe.modeling_qwen3_moe"]
+
+    if rms_norm:
+        modeling_qwen3_moe.Qwen3MoeRMSNorm = LigerRMSNorm
+
+    if glu_activation:
+
+        def _liger_swiglu_mlp_wrapper(config, intermediate_size=None, **kwargs):
+            "Accepts intermediate_size to pass to LigerSwiGLUMLP"
+            # clone config to avoid modifying the original
+            config = deepcopy(config)
+            if intermediate_size:
+                setattr(config, "intermediate_size", intermediate_size)
+            return LigerSwiGLUMLP(config, **kwargs)
+
+        modeling_qwen3_moe.Qwen3MoeMLP = _liger_swiglu_mlp_wrapper
+
+    if layer_norm:
+        modeling_qwen3_moe.nn.LayerNorm = LigerLayerNorm
+
+    if cross_entropy:
+        from transformers.loss.loss_utils import nn
+
+        nn.functional.cross_entropy = liger_cross_entropy
+
+    if fused_linear_cross_entropy:
+        modeling_qwen3_moe.Qwen3MoeForCausalLM.forward = lce_forward

src/axolotl/integrations/llm_compressor/README.md

@@ -0,0 +1,108 @@
+# LLMCompressor Integration
+
+Fine-tune sparsified models in Axolotl using Neural Magic's [LLMCompressor](https://github.com/vllm-project/llm-compressor).
+
+This integration enables fine-tuning of models sparsified using LLMCompressor within the Axolotl training framework. By combining LLMCompressor's model compression capabilities with Axolotl's distributed training pipelines, users can efficiently fine-tune sparse models at scale.
+
+It uses Axolotl’s plugin system to hook into the fine-tuning flows while maintaining sparsity throughout training.
+
+---
+
+## Requirements
+
+- Axolotl with `llmcompressor` extras:
+
+  ```bash
+  pip install "axolotl[llmcompressor]"
+  ```
+
+- Requires `llmcompressor >= 0.5.1`
+
+This will install all necessary dependencies to fine-tune sparsified models using the integration.
+
+---
+
+## Usage
+
+To enable sparse fine-tuning with this integration, include the plugin in your Axolotl config:
+
+```yaml
+plugins:
+  - axolotl.integrations.llm_compressor.LLMCompressorPlugin
+
+llmcompressor:
+  recipe:
+    finetuning_stage:
+      finetuning_modifiers:
+        ConstantPruningModifier:
+          targets: [
+            're:.*q_proj.weight',
+            're:.*k_proj.weight',
+            're:.*v_proj.weight',
+            're:.*o_proj.weight',
+            're:.*gate_proj.weight',
+            're:.*up_proj.weight',
+            're:.*down_proj.weight',
+          ]
+          start: 0
+  save_compressed: true
+# ... (other training arguments)
+```
+
+This plugin **does not apply pruning or sparsification itself** — it is intended for **fine-tuning models that have already been sparsified**.
+
+Pre-sparsified checkpoints can be:
+- Generated using [LLMCompressor](https://github.com/vllm-project/llm-compressor)
+- Downloaded from [Neural Magic's Hugging Face page](https://huggingface.co/neuralmagic)
+- Any custom LLM with compatible sparsity patterns that you've created yourself
+
+To learn more about writing and customizing LLMCompressor recipes, refer to the official documentation:
+[https://github.com/vllm-project/llm-compressor/blob/main/README.md](https://github.com/vllm-project/llm-compressor/blob/main/README.md)
+
+### Storage Optimization with save_compressed
+
+Setting `save_compressed: true` in your configuration enables saving models in a compressed format, which:
+- Reduces disk space usage by approximately 40%
+- Maintains compatibility with vLLM for accelerated inference
+- Maintains compatibility with llmcompressor for further optimization (example: quantization)
+
+This option is highly recommended when working with sparse models to maximize the benefits of model compression.
+
+### Example Config
+
+See [`examples/llama-3/sparse-finetuning.yaml`](examples/llama-3/sparse-finetuning.yaml) for a complete example.
+
+---
+
+## Inference with vLLM
+
+After fine-tuning your sparse model, you can leverage vLLM for efficient inference.
+You can also use LLMCompressor to apply additional quantization to your fine-tuned
+sparse model before inference for even greater performance benefits.:
+
+```python
+from vllm import LLM, SamplingParams
+
+prompts = [
+    "Hello, my name is",
+    "The president of the United States is",
+    "The capital of France is",
+    "The future of AI is",
+]
+sampling_params = SamplingParams(temperature=0.8, top_p=0.95)
+llm = LLM("path/to/your/sparse/model")
+outputs = llm.generate(prompts, sampling_params)
+
+for output in outputs:
+    prompt = output.prompt
+    generated_text = output.outputs[0].text
+    print(f"Prompt: {prompt!r}, Generated text: {generated_text!r}")
+```
+
+For more details on vLLM's capabilities and advanced configuration options, see the [official vLLM documentation](https://docs.vllm.ai/).
+
+## Learn More
+
+For details on available sparsity and quantization schemes, fine-tuning recipes, and usage examples, visit the official LLMCompressor repository:
+
+[https://github.com/vllm-project/llm-compressor](https://github.com/vllm-project/llm-compressor)

src/axolotl/integrations/llm_compressor/__init__.py

@@ -0,0 +1,5 @@
+"""Integration entry point for the LLMCompressor plugin."""
+
+from .plugin import LLMCompressorPlugin
+
+__all__ = ["LLMCompressorPlugin"]

src/axolotl/integrations/llm_compressor/args.py

@@ -0,0 +1,40 @@
+"""
+LLMCompressor and Sparse Finetuning config models.
+"""
+
+from typing import Any
+
+from pydantic import BaseModel, Field
+from typing_extensions import Annotated
+
+
+class CompressionArgs(BaseModel):
+    """Sparse Finetuning config for LLMCompressor."""
+
+    # Typing for recipe is set to Any due to:
+    # https://github.com/vllm-project/llm-compressor/issues/1319
+    recipe: Annotated[
+        Any,
+        Field(
+            description="The recipe containing the compression algorithms and hyperparameters to apply."
+        ),
+    ]
+
+    save_compressed: Annotated[
+        bool,
+        Field(
+            default=False,
+            description="Whether to save the compressed model after training.",
+        ),
+    ]
+
+
+class LLMCompressorArgs(BaseModel):
+    """LLMCompressor configuration BaseModel."""
+
+    llmcompressor: Annotated[
+        CompressionArgs,
+        Field(
+            description="Arguments enabling compression pathways through the LLM Compressor plugins"
+        ),
+    ]

src/axolotl/integrations/llm_compressor/plugin.py

@@ -0,0 +1,171 @@
+"""
+Sparse Finetuning plugin for Axolotl — enables handling of sparse neural networks
+by maintaining masks for zero weights during training.
+"""
+
+import logging
+from functools import wraps
+from typing import Any, Callable, Concatenate, ParamSpec, TypeVar
+
+from llmcompressor import active_session, create_session
+from llmcompressor.core import callbacks as session_callbacks
+from llmcompressor.recipe import Recipe
+from torch.nn import Module
+from transformers.trainer import Trainer
+from transformers.trainer_callback import TrainerCallback, TrainerControl, TrainerState
+from transformers.training_args import TrainingArguments
+
+from axolotl.integrations.base import BasePlugin
+
+P = ParamSpec("P")  # Params for generic function signatures
+R = TypeVar("R")  # Return type for generic function signatures
+
+LOG = logging.getLogger("axolotl.integrations.llm_compressor")
+
+
+class LLMCompressorCallbackHandler(TrainerCallback):
+    """
+    Trainer callback for Sparse Finetuning.
+    Maintains sparsity patterns during training by applying masks after optimization steps,
+    ensuring zero-weight updates are canceled out.
+    """
+
+    def __init__(self, trainer: Trainer, recipe: Any):
+        """
+        Initialize the Sparse Finetuning callback handler.
+
+        Args:
+            trainer (Trainer): Huggingface Trainer instance.
+            recipe (Recipe | dict): Sparse finetuning recipe to apply.
+        """
+        super().__init__()
+        self.trainer = trainer
+        self.recipe = (
+            Recipe.model_validate(recipe) if not isinstance(recipe, Recipe) else recipe
+        )
+        self.original_compute_loss = trainer.compute_loss
+        self.trainer.compute_loss = compute_loss_wrapper(self.trainer.compute_loss)
+        create_session()
+
+    def on_train_begin(
+        self,
+        args: TrainingArguments,
+        state: TrainerState,
+        control: TrainerControl,
+        **kwargs,
+    ) -> None:
+        """
+        Called at the beginning of training. Initializes the compression session.
+
+        Args:
+            args (TrainingArguments): Training arguments.
+            state (TrainerState): Trainer state.
+            control (TrainerControl): Trainer control.
+        """
+        super().on_train_begin(args, state, control, **kwargs)
+        self.trainer.accelerator.wait_for_everyone()
+        active_session().initialize(
+            model=self.trainer.model,
+            optimizer=self.trainer.optimizer,
+            start=state.epoch,
+            recipe=self.recipe,
+        )
+        self.trainer.accelerator.wait_for_everyone()
+
+    def on_step_begin(
+        self,
+        args: TrainingArguments,
+        state: TrainerState,
+        control: TrainerControl,
+        **kwargs,
+    ) -> None:
+        """
+        Called at the beginning of a training step. Triggers batch_start callback.
+        """
+        super().on_step_begin(args, state, control, **kwargs)
+        session_callbacks.batch_start()
+
+    def on_step_end(
+        self,
+        args: TrainingArguments,
+        state: TrainerState,
+        control: TrainerControl,
+        **kwargs,
+    ) -> None:
+        """
+        Called at the end of a training step. Triggers optimizer and batch_end callbacks.
+        """
+        super().on_step_end(args, state, control, **kwargs)
+        session_callbacks.optim_pre_step()
+        session_callbacks.optim_post_step()
+        session_callbacks.batch_end()
+
+    def on_train_end(
+        self,
+        args: TrainingArguments,
+        state: TrainerState,
+        control: TrainerControl,
+        **kwargs,
+    ) -> None:
+        """
+        Called at the end of training. Finalizes the compression session.
+        """
+        super().on_train_end(args, state, control, **kwargs)
+        active_session().finalize()
+        self.trainer.compute_loss_func = self.original_compute_loss
+
+
+class LLMCompressorPlugin(BasePlugin):
+    """
+    Sparse Finetuning plugin for Axolotl integration.
+    """
+
+    def get_input_args(self) -> str:
+        """
+        Returns the path to the plugin's argument definition.
+
+        Returns:
+            str: Dotted path to the LLMCompressorArgs class.
+        """
+        return "axolotl.integrations.llm_compressor.args.LLMCompressorArgs"
+
+    def add_callbacks_post_trainer(self, cfg: Any, trainer: Trainer) -> list:
+        """
+        Adds Sparse Finetuning callback to the Trainer instance.
+
+        Args:
+            cfg (Any): Configuration object containing the sparse recipe.
+            trainer (Trainer): Huggingface Trainer instance.
+
+        Returns:
+            list: List containing the configured callback instances.
+        """
+        LOG.info("Adding Sparse Finetuning callback to the trainer")
+        callback = LLMCompressorCallbackHandler(
+            trainer=trainer,
+            recipe=cfg.llmcompressor.recipe,
+        )
+        return [callback]
+
+
+def compute_loss_wrapper(
+    compute_loss_func: Callable[Concatenate[Module, P], R],
+) -> Callable[Concatenate[Module, P], R]:
+    """
+    Wraps the loss computation function to trigger the loss_calculated callback.
+
+    Args:
+        compute_loss_func (Callable): Original loss computation function.
+
+    Returns:
+        Callable: Wrapped function that also invokes the loss_calculated callback.
+    """
+
+    @wraps(compute_loss_func)
+    def compute_and_notify(model: Module, *args: P.args, **kwargs: P.kwargs) -> R:
+        loss = compute_loss_func(model, *args, **kwargs)
+        if active_session().lifecycle.initialized_ and model.training:
+            session_callbacks.loss_calculated(loss=loss)
+        return loss
+
+    return compute_and_notify

src/axolotl/integrations/llm_compressor/utils.py

@@ -0,0 +1,40 @@
+"""Utilities for llmcompressor integration with axolotl."""
+
+from typing import Union
+
+from llmcompressor.transformers.sparsification.compressed_tensors_utils import (
+    modify_save_pretrained,
+)
+from transformers import PreTrainedModel, Trainer
+
+
+def save_compressed_model(
+    model: PreTrainedModel,
+    output_dir: Union[str, bytes],
+    trainer: Trainer,
+    safe_serialization: bool = False,
+    save_compressed: bool = False,
+) -> None:
+    """
+    Synchronize processes, apply compression hooks, and save the model.
+
+    Args:
+        model (PreTrainedModel): The model to be saved.
+        output_dir (str or bytes): Path where the model files will be written.
+        trainer (Trainer): Hugging Face Trainer for process synchronization.
+        safe_serialization (bool): Use safe serialization if True.
+        save_compressed (bool): Write compressed tensors if True.
+    """
+    trainer.accelerator.wait_for_everyone()
+
+    # Only the main process writes the files
+    if not trainer.accelerator.is_main_process:
+        return
+
+    modify_save_pretrained(model)
+    model.save_pretrained(
+        output_dir,
+        safe_serialization=safe_serialization,
+        save_compressed=save_compressed,
+        skip_sparsity_compression_stats=not save_compressed,
+    )

src/axolotl/monkeypatch/attention/__init__.py

@@ -0,0 +1,19 @@
+"""
+attention module for attention monkeypatches
+"""
+
+from transformers.integrations.flash_attention import flash_attention_forward
+
+
+def patch_xformers_attn_over_fa2():
+    from transformers.modeling_utils import ALL_ATTENTION_FUNCTIONS
+
+    from .xformers import xformers_attention_forward
+
+    ALL_ATTENTION_FUNCTIONS["flash_attention_2"] = xformers_attention_forward
+
+
+def unpatch_xformers_attn_over_fa2():
+    from transformers.modeling_utils import ALL_ATTENTION_FUNCTIONS
+
+    ALL_ATTENTION_FUNCTIONS["flash_attention_2"] = flash_attention_forward()

src/axolotl/monkeypatch/attention/ring_attn/__init__.py

@@ -4,7 +4,6 @@
 # flake8: noqa
 
 from .patch import (
-    RingAttnFunc,
     get_ring_attn_group,
     register_ring_attn,
     set_ring_attn_group,

src/axolotl/monkeypatch/attention/ring_attn/adapters/batch.py

@@ -16,11 +16,7 @@ import torch
 import torch.distributed as dist
 import transformers
 import transformers.modeling_flash_attention_utils
-from ring_flash_attn import (
-    ring_flash_attn_func,
-    stripe_flash_attn_func,
-    zigzag_ring_flash_attn_func,
-)
+from ring_flash_attn import ring_flash_attn_func
 from ring_flash_attn.adapters.hf_adapter import check_params
 from transformers.modeling_flash_attention_utils import (
     _flash_supports_window_size,
@@ -28,12 +24,12 @@ from transformers.modeling_flash_attention_utils import (
 )
 from transformers.modeling_utils import ALL_ATTENTION_FUNCTIONS
 
-from axolotl.monkeypatch.attention.ring_attn.patch import RingAttnFunc
+from axolotl.utils.schemas.enums import RingAttnFunc
 
 RING_ATTN_FUNC_MAPPING = {
-    RingAttnFunc.BATCH_RING: ring_flash_attn_func,
-    RingAttnFunc.BATCH_ZIGZAG: zigzag_ring_flash_attn_func,
-    RingAttnFunc.BATCH_STRIPE: stripe_flash_attn_func,
+    RingAttnFunc.BATCH_RING: torch.compile(ring_flash_attn_func),
+    # RingAttnFunc.BATCH_ZIGZAG: torch.compile(zigzag_ring_flash_attn_func),
+    # RingAttnFunc.BATCH_STRIPE: torch.compile(stripe_flash_attn_func),
 }
 
 

src/axolotl/monkeypatch/attention/ring_attn/patch.py

@@ -6,16 +6,13 @@ package, specifically the `hf_adapter.substitute_hf_flash_attn` function to patc
 their sequence parallel version of Flash Attention 2.
 """
 
-from enum import Enum
-
 import torch
 import torch.distributed as dist
 from accelerate.logging import get_logger
 
-from axolotl.logging_config import configure_logging
 from axolotl.monkeypatch.utils import get_cu_seqlens_from_pos_ids
+from axolotl.utils.schemas.enums import RingAttnFunc
 
-configure_logging()
 LOG = get_logger(__name__)
 
 
@@ -43,17 +40,6 @@ def set_ring_attn_group(ring_attn_group: dist.ProcessGroup | None):
     RING_ATTN_GROUP = ring_attn_group
 
 
-class RingAttnFunc(str, Enum):
-    """Enum class for supported `ring-flash-attn` implementations"""
-
-    # VARLEN_RING = "varlen_ring"
-    # VARLEN_ZIGZAG = "varlen_zigzag"
-    VARLEN_LLAMA3 = "varlen_llama3"
-    BATCH_RING = "batch_ring"
-    BATCH_ZIGZAG = "batch_zigzag"
-    BATCH_STRIPE = "batch_stripe"
-
-
 def register_ring_attn(
     sequence_parallel_degree: int,
     heads_k_stride: int | None,
@@ -119,11 +105,7 @@ def register_ring_attn(
         substitute_hf_flash_attn(
             process_group=get_ring_attn_group(), heads_k_stride=heads_k_stride or 1
         )
-    elif ring_attn_func in [
-        RingAttnFunc.BATCH_RING,
-        RingAttnFunc.BATCH_ZIGZAG,
-        RingAttnFunc.BATCH_STRIPE,
-    ]:
+    elif ring_attn_func is RingAttnFunc.BATCH_RING:
         from axolotl.monkeypatch.attention.ring_attn.adapters.batch import (
             substitute_hf_flash_attn,
         )

src/axolotl/monkeypatch/attention/xformers.py

@@ -0,0 +1,160 @@
+"""
+xformers attention implementation for packing
+"""
+
+from typing import Optional
+
+import torch
+import xformers
+import xformers.ops.fmha
+from transformers.modeling_flash_attention_utils import (
+    _upad_input,
+)
+
+from axolotl.monkeypatch.utils import get_cu_seqlens_from_pos_ids
+
+xformers_attention = xformers.ops.fmha.memory_efficient_attention
+
+
+def xformers_attention_forward(
+    module: torch.nn.Module,
+    query: torch.Tensor,
+    key: torch.Tensor,
+    value: torch.Tensor,
+    attention_mask: Optional[torch.Tensor] = None,
+    position_ids: Optional[torch.LongTensor] = None,
+    dropout: float = 0.0,  # pylint: disable=unused-argument
+    scaling: Optional[float] = None,  # pylint: disable=unused-argument
+    sliding_window: Optional[int] = None,  # pylint: disable=unused-argument
+    softcap: Optional[float] = None,  # pylint: disable=unused-argument
+    cu_seq_lens_q: Optional[torch.LongTensor] = None,
+    cu_seq_lens_k: Optional[torch.LongTensor] = None,
+    max_length_q: Optional[int] = None,
+    max_length_k: Optional[int] = None,  # pylint: disable=unused-argument
+    **kwargs,  # pylint: disable=unused-argument
+):
+    # Get dimensions
+    # query: [batch, heads, seq_len, hidden_dim]
+    batch_size = query.size(0)
+    query_length = query.shape[2]
+    key_length = key.shape[2]
+
+    # Default causal mask
+    attn_bias = xformers.ops.LowerTriangularMask()
+
+    # Check if we have sliding window attention
+    has_sliding_window = sliding_window is not None and sliding_window < query_length
+
+    # Transpose dimensions for xformers (Q: [b, h, s, d] -> [b, s, h, d])
+    query = query.transpose(1, 2)
+    key = key.transpose(1, 2)
+    value = value.transpose(1, 2)
+
+    # Get GQA parameters
+    num_attention_heads = module.config.num_attention_heads
+    num_key_value_heads = module.config.num_key_value_heads
+    head_dim = query.size(-1)
+    is_gqa = num_attention_heads != num_key_value_heads
+    n_groups = num_attention_heads // num_key_value_heads if is_gqa else 1
+
+    # If position_ids is provided and check all examples do not contain only 1 sequence, If tensor in increasing
+    # then we probably have one sequence, otherwise it is packed. Additionally check we are in pre-fill/training stage.
+    # Use `flash_attn_varlen_func` to prevent cross-example attention and also allow padding free approach
+    if position_ids is not None and (
+        max_length_q is not None
+        or (query_length != 1 and not (torch.diff(position_ids, dim=-1) >= 0).all())
+    ):
+        if cu_seq_lens_q is None or cu_seq_lens_k is None:
+            cu_seq_lens_q = get_cu_seqlens_from_pos_ids(position_ids)[0]
+            cu_seq_lens_q = cu_seq_lens_q.squeeze()
+            seq_lengths = cu_seq_lens_q[1:] - cu_seq_lens_q[:-1]
+            attn_bias = (
+                xformers.ops.fmha.attn_bias.BlockDiagonalCausalMask.from_seqlens(
+                    q_seqlen=seq_lengths.tolist(),
+                )
+            )
+        else:
+            query = query.reshape(-1, query.size(-2), query.size(-1))
+            key = key.reshape(-1, key.size(-2), key.size(-1))
+            value = value.reshape(-1, value.size(-2), value.size(-1))
+
+        # Handle GQA
+        if is_gqa:
+            key = key.repeat_interleave(n_groups, dim=2)
+            value = value.repeat_interleave(n_groups, dim=2)
+
+    elif attention_mask is not None:
+        query, key, value, _, cu_seq_lens, _ = _upad_input(
+            query, key, value, attention_mask, query_length
+        )
+        cu_seq_lens_q, cu_seq_lens_k = cu_seq_lens
+        seq_lengths = []
+        for i in range(len(cu_seq_lens_q) - 1):
+            seq_lengths.append(cu_seq_lens_q[i + 1] - cu_seq_lens_q[i])
+        attn_bias = xformers.ops.fmha.attn_bias.BlockDiagonalCausalMask.from_seqlens(
+            q_seqlen=seq_lengths,
+            kv_seqlen=seq_lengths,
+        )
+
+        # Handle GQA
+        if is_gqa:
+            key = key.repeat_interleave(n_groups, dim=2)
+            value = value.repeat_interleave(n_groups, dim=2)
+    else:
+        # Handle Group Query Attention (GQA) using view/expand approach from reference
+        key = key.view(batch_size, key_length, num_key_value_heads, 1, head_dim)
+        value = value.view(batch_size, key_length, num_key_value_heads, 1, head_dim)
+        key = key.expand(
+            batch_size, key_length, num_key_value_heads, n_groups, head_dim
+        )
+        value = value.expand(
+            batch_size, key_length, num_key_value_heads, n_groups, head_dim
+        )
+
+        if module.training:
+            key = key.reshape(batch_size, key_length, num_attention_heads, head_dim)
+            value = value.reshape(batch_size, key_length, num_attention_heads, head_dim)
+
+            if has_sliding_window:
+                query = query.view(
+                    1, batch_size * query_length, num_attention_heads, head_dim
+                )
+                key = key.view(
+                    1, batch_size * key_length, num_attention_heads, head_dim
+                )
+                value = value.view(
+                    1, batch_size * key_length, num_attention_heads, head_dim
+                )
+        else:
+            query = query.view(
+                batch_size, query_length, num_key_value_heads, n_groups, head_dim
+            )
+
+            # If we need a sliding window attention
+            if has_sliding_window:
+                query = query.view(
+                    1,
+                    batch_size * query_length,
+                    num_key_value_heads,
+                    n_groups,
+                    head_dim,
+                )
+                key = key.view(
+                    1, batch_size * key_length, num_key_value_heads, n_groups, head_dim
+                )
+                value = value.view(
+                    1, batch_size * key_length, num_key_value_heads, n_groups, head_dim
+                )
+
+    # Run the xformers attention
+    attn_output = xformers_attention(
+        query,
+        key,
+        value,
+        attn_bias=attn_bias,
+    )
+
+    attn_output = attn_output.view(
+        batch_size, -1, attn_output.size(-2), attn_output.size(-1)
+    )
+    return attn_output, None

src/axolotl/monkeypatch/lora_kernels.py

@@ -23,22 +23,42 @@ from axolotl.utils.dict import DictDefault
 
 LOG = get_logger(__name__)
 
-ORIGINAL_QKV_CODE = """
+QKV_PATCHES = [
+    (
+        """
     query_states = self.q_proj(hidden_states).view(hidden_shape).transpose(1, 2)
     key_states = self.k_proj(hidden_states).view(hidden_shape).transpose(1, 2)
     value_states = self.v_proj(hidden_states).view(hidden_shape).transpose(1, 2)
 """.lstrip(
-    "\n"
-)
-
-PATCHED_QKV_CODE = """
+            "\n"
+        ),
+        """
     query_states, key_states, value_states = self.apply_qkv(hidden_states)
     query_states = query_states.view(hidden_shape).transpose(1, 2)
     key_states = key_states.view(hidden_shape).transpose(1, 2)
     value_states = value_states.view(hidden_shape).transpose(1, 2)
 """.lstrip(
-    "\n"
-)
+            "\n"
+        ),
+    ),
+    (
+        """
+    query_states = self.q_norm(self.q_proj(hidden_states).view(hidden_shape)).transpose(1, 2)
+    key_states = self.k_norm(self.k_proj(hidden_states).view(hidden_shape)).transpose(1, 2)
+    value_states = self.v_proj(hidden_states).view(hidden_shape).transpose(1, 2)
+""".lstrip(
+            "\n"
+        ),
+        """
+    query_states, key_states, value_states = self.apply_qkv(hidden_states)
+    query_states = self.q_norm(query_states.view(hidden_shape)).transpose(1, 2)
+    key_states = self.k_norm(key_states.view(hidden_shape)).transpose(1, 2)
+    value_states = value_states.view(hidden_shape).transpose(1, 2)
+""".lstrip(
+            "\n"
+        ),
+    ),
+]
 
 ORIGINAL_O_CODE = """
     attn_output = self.o_proj(attn_output)
@@ -128,10 +148,11 @@ def get_attention_cls_from_config(cfg: DictDefault) -> Type[nn.Module]:
     try:
         # Dynamically import the module and attention class
         module_path = f"transformers.models.{model_type}.modeling_{model_type}"
-        module = __import__(
-            module_path, fromlist=[f"{model_type.capitalize()}Attention"]
+        model_cls_prefix = "".join(
+            [part.capitalize() for part in model_type.split("_")]
         )
-        attention_cls = getattr(module, f"{model_type.capitalize()}Attention")
+        module = __import__(module_path, fromlist=[f"{model_cls_prefix}Attention"])
+        attention_cls = getattr(module, f"{model_cls_prefix}Attention")
 
         return attention_cls
     except (ImportError, AttributeError) as e:
@@ -168,10 +189,18 @@ def patch_self_attn_lora(cfg: DictDefault):
     attention_cls._original_forward = self_attn_forward
     self_attn_forward, _ = detab_code(self_attn_forward)
 
-    assert ORIGINAL_QKV_CODE in self_attn_forward, "Original QKV code not found"
+    assert any(
+        qkv_options[0] in self_attn_forward for qkv_options in QKV_PATCHES
+    ), "Original QKV code not found"
     assert ORIGINAL_O_CODE in self_attn_forward, "Original O code not found"
 
-    self_attn_forward = self_attn_forward.replace(ORIGINAL_QKV_CODE, PATCHED_QKV_CODE)
+    for qkv_orig, qkv_patched in QKV_PATCHES:
+        if qkv_orig in self_attn_forward:
+            self_attn_forward = self_attn_forward.replace(
+                qkv_orig,
+                qkv_patched,
+            )
+            break
     self_attn_forward = self_attn_forward.replace(ORIGINAL_O_CODE, PATCHED_O_CODE)
     self_attn_forward = self_attn_forward.replace(
         "def forward(",

src/axolotl/monkeypatch/multipack.py

@@ -18,6 +18,8 @@ SUPPORTED_MULTIPACK_MODEL_TYPES = [
     "mixtral",
     "qwen2",
     "qwen2_moe",
+    "qwen3",
+    "qwen3_moe",
     "falcon",
     "phi",
     "phi3",

src/axolotl/monkeypatch/peft/utils.py

@@ -0,0 +1,78 @@
+"""
+Patch prepare_model_for_kbit_training to not upcast everything
+"""
+
+import inspect
+import logging
+
+import peft
+
+import axolotl
+from axolotl.monkeypatch.utils import detab_code
+
+LOG = logging.getLogger(__name__)
+
+ORIGINAL_PREPARE_CODE = """
+        for param in model.parameters():
+            if (
+                (param.dtype == torch.float16) or (param.dtype == torch.bfloat16)
+            ) and param.__class__.__name__ != "Params4bit":
+                param.data = param.data.to(torch.float32)
+"""
+
+PATCHED_PREPARE_CODE = """
+        for name, param in model.named_parameters():
+            if (
+                (param.dtype == torch.float16) or (param.dtype == torch.bfloat16)
+            ) and param.__class__.__name__ != "Params4bit" and all(embed_name not in name for embed_name in ["embed_tokens", "lm_head"]):
+                param.data = param.data.to(torch.float32)
+"""
+
+
+def get_peft_prep_code() -> str:
+    prepare = inspect.getsource(peft.utils.other.prepare_model_for_kbit_training)
+    return prepare
+
+
+def check_peft_prep_code_is_patchable() -> bool:
+    prep_code = get_peft_prep_code()
+    prep_code, _ = detab_code(prep_code)
+    return ORIGINAL_PREPARE_CODE in prep_code
+
+
+def patch_peft_prep_code():
+    """
+    monkeypatch create_accelerator_and_postprocess so it checks for additional kwargs
+    """
+
+    try:
+        prep_code = get_peft_prep_code()
+    except OSError:
+        return
+    peft.utils.other._original_create_accelerator_and_postprocess = (  # pylint: disable=protected-access
+        prep_code
+    )
+    prep_code, _ = detab_code(prep_code)
+    if ORIGINAL_PREPARE_CODE not in prep_code:
+        return
+
+    prep_code = prep_code.replace(ORIGINAL_PREPARE_CODE, PATCHED_PREPARE_CODE)
+    prep_code = prep_code.replace(
+        "def prepare_model_for_kbit_training(",
+        "def fixed_prepare_model_for_kbit_training(",
+        1,
+    )
+
+    items_to_import = []
+    for item in dir(peft.utils.other):
+        if item in prep_code:
+            items_to_import.append(item)
+
+    exec(  # pylint: disable=exec-used  # nosec B102
+        "from peft.utils.other import (" + ", ".join(x for x in items_to_import) + ")",
+        globals(),
+    )
+    exec(prep_code, globals())  # pylint: disable=exec-used  # nosec B102
+    LOG.info("patching prepare_model_for_kbit_training to allow for overrides")
+    peft.utils.other.prepare_model_for_kbit_training = fixed_prepare_model_for_kbit_training  # pylint: disable=protected-access  # pylint: disable=undefined-variable  # noqa: F821
+    axolotl.utils.models.prepare_model_for_kbit_training = fixed_prepare_model_for_kbit_training  # pylint: disable=protected-access  # pylint: disable=undefined-variable  # noqa: F821

src/axolotl/monkeypatch/trainer/lr.py

@@ -0,0 +1,42 @@
+"""
+monkeypatch for Trainer _get_learning_rate method
+"""
+
+import logging
+
+import torch
+
+LOG = logging.getLogger(__name__)
+
+
+# TODO remove this patch once https://github.com/huggingface/transformers/pull/37881 is included in a release
+def _get_learning_rate(self):
+    if self.is_deepspeed_enabled:
+        # with deepspeed's fp16 and dynamic loss scale enabled the optimizer/scheduler steps may
+        # not run for the first few dozen steps while loss scale is too large, and thus during
+        # that time `get_last_lr` will fail if called during that warm up stage, so work around it:
+        try:
+            last_lr = self.lr_scheduler.get_last_lr()[0]
+        except AssertionError as e:
+            if "need to call step" in str(e):
+                LOG.warning(
+                    "tried to get lr value before scheduler/optimizer started stepping, returning lr=0"
+                )
+                last_lr = 0
+            else:
+                raise
+    else:
+        if isinstance(self.lr_scheduler, torch.optim.lr_scheduler.ReduceLROnPlateau):
+            last_lr = self.optimizer.param_groups[0]["lr"]
+        else:
+            last_lr = self.lr_scheduler.get_last_lr()[0]
+
+    if torch.is_tensor(last_lr):
+        last_lr = last_lr.item()
+    return last_lr
+
+
+def patch_trainer_get_lr():
+    from transformers.trainer import Trainer
+
+    Trainer._get_learning_rate = _get_learning_rate  # pylint: disable=protected-access

src/axolotl/prompt_strategies/chat_template.py

@@ -4,7 +4,7 @@ HF Chat Templates prompt strategy
 
 import logging
 from collections import defaultdict
-from typing import Any, Dict, List, Optional, Set, Union
+from typing import Any, Dict, List, Set, Union
 
 from pydantic import BaseModel
 from transformers import ProcessorMixin
@@ -29,11 +29,12 @@ class ChatTemplatePrompter(Prompter):
         chat_template: str,
         processor=None,
         max_length=2048,
-        message_property_mappings: Optional[Dict[str, str]] = None,
-        message_field_training: Optional[str] = None,
-        message_field_training_detail: Optional[str] = None,
+        message_property_mappings: Dict[str, str] | None = None,
+        message_field_training: str | None = None,
+        message_field_training_detail: str | None = None,
         field_messages: str = "messages",
-        roles: Optional[Dict[str, List[str]]] = None,
+        field_system: str = "system",
+        roles: Dict[str, List[str]] | None = None,
         drop_system_message: bool = False,
     ):
         # check if message_property_mappings is None or empty dict
@@ -41,6 +42,7 @@ class ChatTemplatePrompter(Prompter):
             message_property_mappings = {
                 "role": "role",
                 "content": "content",
+                "reasoning_content": "reasoning_content",
             }
 
         if roles:
@@ -62,8 +64,9 @@ class ChatTemplatePrompter(Prompter):
         self.message_field_training = message_field_training
         self.message_field_training_detail = message_field_training_detail
         self.field_messages = field_messages
+        self.field_system = field_system
         self.tokenizer = tokenizer
-        self.processor: Optional[ProcessorMixin] = processor
+        self.processor: ProcessorMixin | None = processor
         self.chat_template = chat_template
         self.max_length = max_length
         self.drop_system_message = drop_system_message
@@ -220,10 +223,13 @@ class ChatTemplateStrategy(PromptTokenizingStrategy):
         self,
         prompter: "ChatTemplatePrompter",
         tokenizer,
-        train_on_inputs,
-        sequence_len,
-        roles_to_train=None,
-        train_on_eos=None,
+        train_on_inputs: bool,
+        sequence_len: int,
+        roles_to_train: list[str] | None = None,
+        train_on_eos: str | None = None,
+        train_on_eot: str | None = None,
+        eot_tokens: list[str] | None = None,
+        split_thinking: bool | None = False,
     ):
         super().__init__(prompter, tokenizer, train_on_inputs, sequence_len)
         self.prompter: ChatTemplatePrompter = prompter
@@ -236,12 +242,88 @@ class ChatTemplateStrategy(PromptTokenizingStrategy):
             ]
 
         self.train_on_eos = train_on_eos
+        # Backward compatibility, load from train_on_eos
+        self.train_on_eot = train_on_eot if train_on_eot is not None else train_on_eos
+
+        # Default to eos_token if eot_tokens not provided
+        self.eot_tokens = (
+            eot_tokens if eot_tokens is not None else [self.tokenizer.eos_token]
+        )
+        self.split_thinking = split_thinking
+
         self.images = "images"
 
         LOG.debug(
             f"The chat template uses the following properites on the message: {self.prompter.chat_template_msg_variables}"
         )
 
+        self._validate_eot_and_eos_tokens()
+
+    def _validate_eot_and_eos_tokens(self):
+        """
+        - Validates that EOT tokens (or eos_token) are in the chat_template
+        - Checks if EOT tokens are encoded as multiple tokens in the tokenizer.
+        - Checks for potential conflicts between train_on_eos and train_on_eot.
+        """
+        if self.prompter.chat_template is None:
+            # Usually this should not happen
+            LOG.warning(
+                "No chat template provided, skipping EOT and EOS token validation"
+            )
+            return
+
+        # If the EOT token is the same as the EOS token, we need to check differently
+        if len(self.eot_tokens) == 1 and self.eot_tokens[0] == self.tokenizer.eos_token:
+            # Check if the eos_token is in the chat_template or as a variable `eos_token`
+            # Note: we check for `eos_token` in the string, but it could possibly not be a variable
+            if (
+                self.tokenizer.eos_token not in self.prompter.chat_template
+                and "eos_token" not in self.prompter.chat_template
+            ):
+                LOG.warning(
+                    f"EOS token '{self.tokenizer.eos_token}' not found in chat_template. Please check if your template/EOS token is correct."
+                )
+            return
+
+        # Create a new list to store tokens that should be kept
+        valid_eot_tokens = []
+        for token in self.eot_tokens:
+            # Check if EOT token is in the chat_template
+            if token not in self.prompter.chat_template:
+                LOG.warning(f"EOT token '{token}' not found in chat_template.")
+                # Don't add to the valid tokens list
+                continue
+
+            valid_eot_tokens.append(token)
+
+        # Replace the original list with the filtered one
+        self.eot_tokens = valid_eot_tokens
+
+        for token in self.eot_tokens:
+            # If token in template, check if EOT token is in tokenizer and not encoded as multiple tokens
+            token_ids = self.tokenizer.encode(token, add_special_tokens=False)
+            if not token_ids:
+                raise ValueError(
+                    "EOT token encoding failed. Please check if the token is valid and can be encoded."
+                )
+            if token_ids and len(token_ids) > 1:
+                raise ValueError(
+                    f"EOT token '{token}' is encoded as multiple tokens: {token_ids}. Please add it under `tokens: ` in the config "
+                    "or (recommended) override unused added_tokens via `added_tokens_overrides: `."
+                )
+
+        # If eos_token is in eot_tokens and conflict between train_on_eos and train_on_eot, raise an error
+        if (
+            self.tokenizer.eos_token in self.eot_tokens
+            and self.train_on_eos != self.train_on_eot
+        ):
+            raise ValueError(
+                "Conflict between train_on_eos and train_on_eot. eos_token is in eot_tokens and train_on_eos != train_on_eot"
+                f"train_on_eos: {self.train_on_eos}, train_on_eot: {self.train_on_eot}"
+                f"eot_tokens: {self.eot_tokens}"
+                f"eos_token: {self.tokenizer.eos_token}"
+            )
+
     @property
     def supports_batched(self) -> bool:
         # Let calling code know we can handle lists of examples
@@ -285,6 +367,7 @@ class ChatTemplateStrategy(PromptTokenizingStrategy):
         if (
             not self.roles_to_train
             and not self.train_on_eos
+            and not self.train_on_eot
             and not self.prompter.message_field_training  # type: ignore
             and not self.prompter.message_field_training_detail  # type: ignore
         ):
@@ -320,6 +403,7 @@ class ChatTemplateStrategy(PromptTokenizingStrategy):
         labels = [IGNORE_TOKEN_ID] * len(input_ids)
 
         last_eos_idx = -1
+        last_eot_idx = -1
         for index, turn in enumerate(turns):
             role = turn.get("role")
             content = turn.get("content")
@@ -368,24 +452,45 @@ class ChatTemplateStrategy(PromptTokenizingStrategy):
 
                 LOG.debug(f"Labels after processing turn {index}: {labels}")
 
-            # Handle EOS token
-            eos_idx = self.find_first_eos_token(input_ids, start_idx=turn_end_idx)
-            if abs(eos_idx - turn_end_idx) <= 3:  # Allow for some template padding
-                last_eos_idx = eos_idx
-                if self.train_on_eos == "all" or (
-                    self.train_on_eos == "turn" and should_train
-                ):
-                    labels[eos_idx] = input_ids[eos_idx]
-                    LOG.debug(f"EOS token set for training at index {eos_idx}")
-            else:
-                LOG.debug(
-                    f"EOS token missing after turn {turn}. eos_idx: {eos_idx}, turn_end_idx: {turn_end_idx}"
-                )
+            # Handle special tokens (EOT and EOS)
+            for token_type, find_func, train_option in [
+                ("EOT", self.find_first_eot_token, self.train_on_eot),
+                ("EOS", self.find_first_eos_token, self.train_on_eos),
+            ]:
+                token_idx = find_func(input_ids, start_idx=turn_end_idx)
 
-        # Handle 'last' option for train_on_eos
-        if self.train_on_eos == "last" and last_eos_idx != -1:
-            labels[last_eos_idx] = input_ids[last_eos_idx]
-            LOG.debug(f"Last EOS token set for training at index {last_eos_idx}")
+                if (
+                    token_idx != -1 and abs(token_idx - turn_end_idx) <= 3
+                ):  # Allow for some template padding
+                    # Update the last token index
+                    if token_type == "EOT":  # nosec B105
+                        last_eot_idx = token_idx
+                    else:
+                        last_eos_idx = token_idx
+
+                    # Set labels if needed for this turn
+                    if train_option == "all" or (
+                        train_option == "turn" and should_train
+                    ):
+                        labels[token_idx] = input_ids[token_idx]
+                        LOG.debug(
+                            f"{token_type} token set for training at index {token_idx}"
+                        )
+                else:
+                    LOG.debug(
+                        f"{token_type} token missing after turn {turn}. {token_type.lower()}_idx: {token_idx}, turn_end_idx: {turn_end_idx}"
+                    )
+
+        # Handle 'last' option for special tokens
+        for token_type, last_idx, train_option in [
+            ("EOT", last_eot_idx, self.train_on_eot),
+            ("EOS", last_eos_idx, self.train_on_eos),
+        ]:
+            if train_option == "last" and last_idx != -1:
+                labels[last_idx] = input_ids[last_idx]
+                LOG.debug(
+                    f"Last {token_type} token set for training at index {last_idx}"
+                )
 
         LOG.debug(f"Final labels: {labels}")
 
@@ -402,6 +507,25 @@ class ChatTemplateStrategy(PromptTokenizingStrategy):
                 return i
         return -1
 
+    def find_first_eot_token(self, input_ids, start_idx):
+        """Find the first EOT token in the input_ids starting from start_idx."""
+        # Get token IDs for all EOT tokens
+        eot_token_ids = []
+        for token in self.eot_tokens:
+            token_ids = self.tokenizer.encode(token, add_special_tokens=False)
+            if len(token_ids) != 1:
+                raise ValueError(
+                    f"EOT token '{token}' is encoded as multiple tokens: {token_ids}. Please add it under `tokens: ` in the config."
+                )
+
+            eot_token_ids.append(token_ids[0])  # Use the last token ID if multiple
+
+        # Search for any of the EOT token IDs
+        for i in range(start_idx, len(input_ids)):
+            if input_ids[i] in eot_token_ids:
+                return i
+        return -1
+
     def find_turn(self, turns: list[dict], turn_idx: int):
         """
         Locate the starting and ending indices of the specified turn in a conversation.
@@ -488,6 +612,17 @@ class ChatTemplateStrategy(PromptTokenizingStrategy):
 
     def get_conversation_thread(self, prompt):
         turns = []
+
+        possible_sys_turn = self.transform_message(
+            prompt[self.prompter.field_messages][0]
+        )
+        if (
+            possible_sys_turn["role"] != "system"
+            and self.prompter.field_system in prompt
+        ):
+            turn = {"role": "system", "content": prompt[self.prompter.field_system]}
+            turns.append(turn)
+
         for message in prompt[self.prompter.field_messages]:
             transformed_message = self.transform_message(message)
 
@@ -523,6 +658,52 @@ class ChatTemplateStrategy(PromptTokenizingStrategy):
                 transformed_message["role"], transformed_message["role"]
             )
 
+        # TODO handle reasoning_content with split_thinking
+        # if the role is assistant that we want to use reasoning_content
+        if self.split_thinking and transformed_message["role"] == "assistant":
+            content = transformed_message["content"]
+            thinking_pairs = [
+                ("<think>", "</think>"),
+                ("<reasoning>", "</reasoning>"),
+                ("<|begin_of_thought|>", "<|end_of_thought|>"),
+            ]
+            content_pairs = [("<|begin_of_solution|>", "<|end_of_solution|>")]
+            for tpair in thinking_pairs:
+                # check if the thinking pair is in the content
+                if tpair[0] in content and tpair[1] in content:
+                    # find the start and end index of the thinking pair
+                    t_start_idx = content.find(tpair[0])
+                    t_end_idx = content.find(tpair[1])
+
+                    # get the thinking content
+                    thinking_content = content[t_start_idx + len(tpair[0]) : t_end_idx]
+                    transformed_message["reasoning_content"] = thinking_content.strip()
+
+                    # take remainder of the content
+                    # strip whitespace from beginning of the remainder (thinking tokens)
+                    remainder = content[t_end_idx + len(tpair[1]) :].lstrip()
+
+                    # check if the content pair is in the remainder
+                    cpair_found = False
+                    for cpair in content_pairs:
+                        if cpair[0] in remainder and cpair[1] in remainder:
+                            # find the start and end index of the content pair
+                            c_start_idx = remainder.find(cpair[0])
+                            c_end_idx = remainder.find(cpair[1])
+
+                            # get the content content
+                            content_content = remainder[
+                                c_start_idx + len(cpair[0]) : c_end_idx
+                            ]
+                            transformed_message["content"] = content_content.strip()
+                            cpair_found = True
+                            break
+
+                    # else, the content is the remainder
+                    if not cpair_found:
+                        transformed_message["content"] = remainder
+                    break
+
         # Determine which keys in the original message were not mapped
         mapped_values = set(self.prompter.message_property_mappings.values())
         remaining_keys = set(message) - mapped_values
@@ -555,13 +736,16 @@ class StrategyLoader:
             "sequence_len": cfg.sequence_len,
             "roles_to_train": ds_cfg.get("roles_to_train", ["assistant"]),
             "train_on_eos": ds_cfg.get("train_on_eos", "turn"),
+            "train_on_eot": ds_cfg.get("train_on_eot", None),
+            "eot_tokens": cfg.get("eot_tokens", None),  # loads from cfg, not ds_cfg
+            "split_thinking": ds_cfg.get("split_thinking", False),
         }
 
     def __call__(
         self,
         tokenizer,
         cfg,
-        ds_cfg: Optional[Union[Dict[str, Any], DatasetConfig]] = None,
+        ds_cfg: Union[Dict[str, Any], DatasetConfig] | None = None,
         processor=None,
     ):
         if ds_cfg is None:

src/axolotl/train.py

@@ -2,17 +2,17 @@
 
 import importlib
 import inspect
+import logging
 import os
 import signal
 import sys
 import weakref
-from contextlib import nullcontext
+from contextlib import ExitStack
 from pathlib import Path
 from typing import Any, Dict
 
 import torch
 import transformers.modelcard
-from accelerate.logging import get_logger
 from accelerate.utils import save_fsdp_model
 from datasets import Dataset
 from huggingface_hub.errors import OfflineModeIsEnabled
@@ -21,19 +21,19 @@ from transformers import PreTrainedModel, PreTrainedTokenizer, ProcessorMixin
 from transformers.integrations.deepspeed import is_deepspeed_zero3_enabled
 from transformers.trainer import Trainer
 
+from axolotl.cli.art import print_axolotl_text_art
 from axolotl.common.datasets import TrainDatasetMeta
 from axolotl.contribs.lgpl import (  # pylint: disable = no-name-in-module
     fix_untrained_tokens,
 )
 from axolotl.core.trainer_builder import HFCausalTrainerBuilder, HFRLTrainerBuilder
-from axolotl.core.trainers.mixins.sequence_parallel import (
-    SequenceParallelContextManager,
-)
-from axolotl.logging_config import configure_logging
+from axolotl.integrations.base import PluginManager
+from axolotl.utils.ctx_managers.sequence_parallel import SequenceParallelContextManager
 from axolotl.utils.dict import DictDefault
 from axolotl.utils.distributed import cleanup_distributed
 from axolotl.utils.freeze import freeze_layers_except
 from axolotl.utils.models import load_model, load_processor, load_tokenizer
+from axolotl.utils.schemas.enums import RLType
 from axolotl.utils.trainer import setup_trainer
 
 try:
@@ -41,8 +41,7 @@ try:
 except ImportError:
     BetterTransformer = None
 
-configure_logging()
-LOG = get_logger(__name__)
+LOG = logging.getLogger(__name__)
 
 
 def setup_model_and_tokenizer(
@@ -63,7 +62,6 @@ def setup_model_and_tokenizer(
     # Load tokenizer
     LOG.debug(
         f"loading tokenizer... {cfg.tokenizer_config or cfg.base_model_config}",
-        main_process_only=True,
     )
     tokenizer = load_tokenizer(cfg)
 
@@ -108,7 +106,7 @@ def setup_reference_model(
         Reference model if needed for RL training, `None` otherwise.
     """
     model_ref = None
-    if cfg.rl and cfg.rl != "orpo":
+    if cfg.rl and cfg.rl != RLType.ORPO:
         if cfg.adapter and not cfg.rl_adapter_ref_model:
             # use built-in trl autounwrap
             LOG.debug("Passing model_ref: None to RL trainer")
@@ -189,28 +187,32 @@ def execute_training(
         trainer: The configured trainer object.
         resume_from_checkpoint: Path to checkpoint to resume from, if applicable.
     """
-    # Define the context managers to use
-    flash_context = (
-        torch.backends.cuda.sdp_kernel(
-            enable_flash=True,
-            enable_math=True,
-            enable_mem_efficient=True,
-        )
-        if cfg.flash_optimum
-        else nullcontext()
-    )
-    sequence_parallel_context = (
-        SequenceParallelContextManager(
-            model=trainer.model,
-            sequence_parallel_degree=cfg.sequence_parallel_degree,
-            ring_attn_func=cfg.ring_attn_func,
-        )
-        if cfg.sequence_parallel_degree > 1
-        else nullcontext()
-    )
+    with ExitStack() as stack:
+        # Define the context managers to use
+        if cfg.flash_optimum:
+            stack.enter_context(
+                torch.backends.cuda.sdp_kernel(
+                    enable_flash=True,
+                    enable_math=True,
+                    enable_mem_efficient=True,
+                )
+            )
 
-    LOG.info("Starting trainer...")
-    with flash_context, sequence_parallel_context:
+        if cfg.sequence_parallel_degree > 1:
+            models = [trainer.model]
+            if hasattr(trainer, "ref_model"):
+                models.append(trainer.ref_model)
+
+            stack.enter_context(
+                SequenceParallelContextManager(
+                    models=models,
+                    sequence_parallel_degree=cfg.sequence_parallel_degree,
+                    gradient_accumulation_steps=cfg.gradient_accumulation_steps,
+                    ring_attn_func=cfg.ring_attn_func,
+                )
+            )
+
+        LOG.info("Starting trainer...")
         trainer.train(resume_from_checkpoint=resume_from_checkpoint)
 
 
@@ -287,15 +289,32 @@ def save_trained_model(
                 os.remove(os.path.join(cfg.output_dir, "model.safetensors"))
             except FileNotFoundError:
                 pass
-    elif cfg.local_rank == 0:
-        if cfg.flash_optimum and BetterTransformer:
-            model = BetterTransformer.reverse(model)
+    else:
+        if cfg.local_rank == 0:
+            if cfg.flash_optimum and BetterTransformer:
+                model = BetterTransformer.reverse(model)
 
-        if cfg.rl and cfg.adapter and not cfg.rl_adapter_ref_model:
-            trainer.model.save_pretrained(
-                cfg.output_dir, safe_serialization=safe_serialization
-            )
-        model.save_pretrained(cfg.output_dir, safe_serialization=safe_serialization)
+            if cfg.rl and cfg.adapter and not cfg.rl_adapter_ref_model:
+                trainer.model.save_pretrained(
+                    cfg.output_dir, safe_serialization=safe_serialization
+                )
+
+            model.save_pretrained(cfg.output_dir, safe_serialization=safe_serialization)
+        trainer.accelerator.wait_for_everyone()
+
+    if hasattr(cfg, "llmcompressor") and cfg.llmcompressor:
+        # TODO: add integration support so this can be implemented completely within the plugin
+        from axolotl.integrations.llm_compressor.utils import (
+            save_compressed_model,
+        )
+
+        save_compressed_model(
+            model=model,
+            output_dir=cfg.output_dir,
+            trainer=trainer,
+            safe_serialization=safe_serialization,
+            save_compressed=cfg.llmcompressor.save_compressed,
+        )
 
 
 def create_model_card(cfg: DictDefault, trainer: Trainer):
@@ -502,6 +521,8 @@ def train(
     Returns:
         Tuple of (model, tokenizer) after training
     """
+    print_axolotl_text_art()
+
     # Setup model, tokenizer, (causal or RLHF) trainer, etc.
     (
         trainer,
@@ -511,6 +532,9 @@ def train(
         processor,
     ) = setup_model_and_trainer(cfg, dataset_meta)
 
+    plugin_manager = PluginManager.get_instance()
+    plugin_manager.post_trainer_create(cfg, trainer)
+
     # Handle untrained tokens if configured
     safe_serialization = cfg.save_safetensors is True
     train_dataset = dataset_meta.train_dataset
@@ -533,4 +557,6 @@ def train(
     if not cfg.use_ray:
         cleanup_distributed()
 
+    plugin_manager.post_train(cfg, model)
+
     return model, tokenizer, trainer

src/axolotl/utils/__init__.py

@@ -43,3 +43,12 @@ def set_pytorch_cuda_alloc_conf():
             os.environ["PYTORCH_CUDA_ALLOC_CONF"] = (
                 "expandable_segments:True,roundup_power2_divisions:16"
             )
+
+
+def patch_optimized_env():
+    """
+    Patch environment variables to improve VRAM usage and increase download speed
+    """
+    if os.getenv("HF_HUB_ENABLE_HF_TRANSFER") is None:
+        os.environ["HF_HUB_ENABLE_HF_TRANSFER"] = "1"
+    set_pytorch_cuda_alloc_conf()