nit [skip-e2e]

.bandit

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

.coderabbit.yaml

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

.flake8

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

.github/CONTRIBUTING.md

@@ -62,7 +62,7 @@ We welcome ideas for improvements and new features. To suggest an enhancement, o
 You can skip certain CI checks by including specific keywords in your commit messages:
 
 - `[skip ci]` or `skip ci` - Skips all CI checks for that commit
-- `[skip-e2e]` or `skip-e2e` - Skips only end-to-end tests while running other CI checks. You may also include this in the title of your PR to disable end-to-end tests for the entire PR.
+- `[skip-e2e]` or `skip-e2e` - Skips only end-to-end tests while running other CI checks
 
 ## Style Guidelines
 

.github/workflows/main.yml

@@ -36,11 +36,6 @@ jobs:
             python_version: "3.11"
             pytorch: 2.7.1
             axolotl_extras:
-          - cuda: 128
-            cuda_version: 12.8.1
-            python_version: "3.11"
-            pytorch: 2.8.0
-            axolotl_extras:
     runs-on: axolotl-gpu-runner
     steps:
       - name: Checkout
@@ -115,11 +110,6 @@ jobs:
             python_version: "3.11"
             pytorch: 2.7.1
             axolotl_extras:
-          - cuda: 128
-            cuda_version: 12.8.1
-            python_version: "3.11"
-            pytorch: 2.8.0
-            axolotl_extras:
     runs-on: axolotl-gpu-runner
     steps:
       - name: Checkout
@@ -179,12 +169,6 @@ jobs:
             pytorch: 2.7.1
             axolotl_extras: vllm
             is_latest: true
-          - cuda: 128
-            cuda_version: 12.8.1
-            python_version: "3.11"
-            pytorch: 2.8.0
-            axolotl_extras:
-            is_latest:
     runs-on: axolotl-gpu-runner
     steps:
       - name: Checkout

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

@@ -36,15 +36,15 @@ jobs:
           - cuda: 126
             cuda_version: 12.6.3
             python_version: "3.11"
-            pytorch: 2.7.1
+            pytorch: 2.7.0
-            axolotl_extras: vllm
+            axolotl_extras:
             num_gpus: 2
             nightly_build: "true"
-          - cuda: 128
+          - cuda: 126
-            cuda_version: 12.8.1
+            cuda_version: 12.6.3
             python_version: "3.11"
-            pytorch: 2.8.0
+            pytorch: 2.7.1
-            axolotl_extras: fbgemm-gpu
+            axolotl_extras: vllm
             num_gpus: 2
             nightly_build: "true"
     runs-on: [self-hosted, modal]

.github/workflows/tests.yml

@@ -55,7 +55,7 @@ jobs:
       fail-fast: false
       matrix:
         python_version: ["3.11"]
-        pytorch_version: ["2.6.0", "2.7.1", "2.8.0"]
+        pytorch_version: ["2.6.0", "2.7.0", "2.7.1"]
     timeout-minutes: 20
 
     steps:
@@ -130,7 +130,7 @@ jobs:
       fail-fast: false
       matrix:
         python_version: ["3.11"]
-        pytorch_version: ["2.6.0", "2.7.1", "2.8.0"]
+        pytorch_version: ["2.6.0", "2.7.0", "2.7.1"]
     timeout-minutes: 20
 
     steps:
@@ -188,44 +188,13 @@ jobs:
         run: |
           find "$(pip cache dir)/http-v2" -type f -mtime +14 -exec rm {} \;
 
-  gate-skip-e2e:
-    needs: [pre-commit, pytest, pytest-sdist]
-    runs-on: ubuntu-latest
-    outputs:
-      skip: ${{ steps.compute.outputs.skip }}
-    steps:
-      - uses: actions/github-script@v7
-        id: compute
-        with:
-          script: |
-            const token = /\[skip-e2e\]/i;
-            let msg = '';
-            if (context.eventName === 'push') {
-              msg = context.payload.head_commit?.message || '';
-            } else if (context.eventName === 'pull_request') {
-              const { owner, repo } = context.repo;
-              const prNumber = context.payload.pull_request.number;
-              const commits = await github.paginate(
-                github.rest.pulls.listCommits,
-                { owner, repo, pull_number: prNumber, per_page: 100 }
-              );
-              msg = commits.at(-1)?.commit?.message || '';
-            }
-            const title = context.payload.pull_request?.title || '';
-            const body  = context.payload.pull_request?.body  || '';
-            const skip = token.test(msg) || token.test(title) || token.test(body);
-            core.setOutput('skip', String(skip));
-
   docker-e2e-tests-1st:
     # Run this job first as a gate for running the remainder of the test matrix
-    if: >
+    if: ${{ ! contains(github.event.commits[0].message, '[skip e2e]') && github.repository_owner == 'axolotl-ai-cloud' && !github.event.pull_request.draft }}
-      github.repository_owner == 'axolotl-ai-cloud' &&
-      (github.event_name != 'pull_request' || !github.event.pull_request.draft) &&
-      needs.gate-skip-e2e.outputs.skip != 'true'
     # this job needs to be run on self-hosted GPU runners...
     runs-on: [self-hosted, modal]
     timeout-minutes: 120
-    needs: [pre-commit, pytest, pytest-sdist, gate-skip-e2e]
+    needs: [pre-commit, pytest, pytest-sdist]
 
     strategy:
       fail-fast: false
@@ -240,7 +209,7 @@ jobs:
           - cuda: 126
             cuda_version: 12.6.3
             python_version: "3.11"
-            pytorch: 2.7.1
+            pytorch: 2.6.0
             num_gpus: 1
             axolotl_extras:
             dockerfile: "Dockerfile-uv.jinja"
@@ -271,16 +240,13 @@ jobs:
           modal run cicd.e2e_tests
 
   docker-e2e-tests:
-    if: >
+    if: ${{ github.repository_owner == 'axolotl-ai-cloud' && !github.event.pull_request.draft }}
-      github.repository_owner == 'axolotl-ai-cloud' &&
-      (github.event_name != 'pull_request' || !github.event.pull_request.draft) &&
-      needs.gate-skip-e2e.outputs.skip != 'true'
     # this job needs to be run on self-hosted GPU runners...
     runs-on: [self-hosted, modal]
     timeout-minutes: 120
     # Only run the remainder of the matrix if the first e2e check passed;
     # this is to save on wasted compute costs for known failures that get caught in the first run
-    needs: [pre-commit, pytest, gate-skip-e2e, docker-e2e-tests-1st]
+    needs: [pre-commit, pytest, docker-e2e-tests-1st]
 
     strategy:
       fail-fast: false
@@ -298,13 +264,6 @@ jobs:
             pytorch: 2.7.1
             num_gpus: 1
             axolotl_extras:
-          - cuda: 128
-            cuda_version: 12.8.1
-            python_version: "3.11"
-            pytorch: 2.8.0
-            num_gpus: 1
-            gpu_type: "B200"
-            axolotl_extras: fbgemm-gpu
     steps:
       - name: Checkout
         uses: actions/checkout@v4
@@ -325,7 +284,6 @@ jobs:
           echo "CUDA=${{ matrix.cuda }}" >> $GITHUB_ENV
           echo "MODAL_IMAGE_BUILDER_VERSION=2024.10" >> $GITHUB_ENV
           echo "N_GPUS=${{ matrix.num_gpus }}" >> $GITHUB_ENV
-          echo "GPU_TYPE=${{ matrix.gpu_type || 'L40S'}}" >> $GITHUB_ENV
           echo "CODECOV_TOKEN=${{ secrets.CODECOV_TOKEN }}" >> $GITHUB_ENV
           echo "E2E_DOCKERFILE=${{ matrix.dockerfile || 'Dockerfile.jinja'}}" >> $GITHUB_ENV
       - name: Run tests job on Modal
@@ -342,10 +300,10 @@ jobs:
       fail-fast: false
       matrix:
         include:
-          - cuda: 126
+          - cuda: 124
-            cuda_version: 12.6.3
+            cuda_version: 12.4.1
             python_version: "3.11"
-            pytorch: 2.7.1
+            pytorch: 2.6.0
             num_gpus: 1
             axolotl_extras:
     steps:

.isort.cfg

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

.pre-commit-config.yaml

@@ -10,12 +10,22 @@ repos:
     -   id: trailing-whitespace
     -   id: no-commit-to-branch
         args: ['--branch', 'main']
--   repo: https://github.com/astral-sh/ruff-pre-commit
+-   repo: https://github.com/psf/black
-    rev: v0.12.12
+    rev: 25.1.0
     hooks:
-    -   id: ruff
+    -   id: black
-        args: [--fix, --select, I]
+-   repo: https://github.com/pycqa/isort
-    -   id: ruff-format
+    rev: 6.0.1
+    hooks:
+      - id: isort
+-   repo: https://github.com/PyCQA/flake8
+    rev: 7.3.0
+    hooks:
+    - id: flake8
+-   repo: https://github.com/pylint-dev/pylint
+    rev: v3.3.8
+    hooks:
+    - id: pylint
 -   repo: https://github.com/pre-commit/mirrors-mypy
     rev: v1.17.1
     hooks:

.pylintrc

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

CITATION.cff

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

README.md

@@ -5,9 +5,6 @@
         <img alt="Axolotl" src="https://raw.githubusercontent.com/axolotl-ai-cloud/axolotl/887513285d98132142bf5db2a74eb5e0928787f1/image/axolotl_logo_digital_black.svg" width="400" height="104" style="max-width: 100%;">
     </picture>
 </p>
-  <p align="center">
-      <strong>A Free and Open Source LLM Fine-tuning Framework</strong><br>
-  </p>
 
 <p align="center">
     <img src="https://img.shields.io/github/license/axolotl-ai-cloud/axolotl.svg?color=blue" alt="GitHub License">
@@ -20,7 +17,6 @@
     <br/>
     <a href="https://discord.com/invite/HhrNrHJPRb"><img src="https://img.shields.io/badge/discord-7289da.svg?style=flat-square&logo=discord" alt="discord" style="height: 20px;"></a>
     <a href="https://twitter.com/axolotl_ai"><img src="https://img.shields.io/twitter/follow/axolotl_ai?style=social" alt="twitter" style="height: 20px;"></a>
-    <a href="https://colab.research.google.com/github/axolotl-ai-cloud/axolotl/blob/main/examples/colab-notebooks/colab-axolotl-example.ipynb"><img src="https://colab.research.google.com/assets/colab-badge.svg" alt="google-colab" style="height: 20px;"></a>
     <br/>
     <img src="https://github.com/axolotl-ai-cloud/axolotl/actions/workflows/tests-nightly.yml/badge.svg" alt="tests-nightly">
     <img src="https://github.com/axolotl-ai-cloud/axolotl/actions/workflows/multi-gpu-e2e.yml/badge.svg" alt="multigpu-semi-weekly tests">
@@ -53,21 +49,20 @@
 
 ## ✨ Overview
 
-Axolotl is a free and open-source tool designed to streamline post-training and fine-tuning for the latest large language models (LLMs).
+Axolotl is a tool designed to streamline post-training for various AI models.
 
 Features:
 
-- **Multiple Model Support**: Train various models like GPT-OSS, LLaMA, Mistral, Mixtral, Pythia, and many more models available on the Hugging Face Hub.
+- **Multiple Model Support**: Train various models like LLaMA, Mistral, Mixtral, Pythia, and more. We are compatible with HuggingFace transformers causal language models.
-- **Multimodal Training**: Fine-tune vision-language models (VLMs) including LLaMA-Vision, Qwen2-VL, Pixtral, LLaVA, SmolVLM2, and audio models like Voxtral with image, video, and audio support.
+- **Training Methods**: Full fine-tuning, LoRA, QLoRA, GPTQ, QAT, Preference Tuning (DPO, IPO, KTO, ORPO), RL (GRPO), Multimodal, and Reward Modelling (RM) / Process Reward Modelling (PRM).
-- **Training Methods**: Full fine-tuning, LoRA, QLoRA, GPTQ, QAT, Preference Tuning (DPO, IPO, KTO, ORPO), RL (GRPO), and Reward Modelling (RM) / Process Reward Modelling (PRM).
+- **Easy Configuration**: Re-use a single YAML file between dataset preprocess, training, evaluation, quantization, and inference.
-- **Easy Configuration**: Re-use a single YAML configuration file across the full fine-tuning pipeline: dataset preprocessing, training, evaluation, quantization, and inference.
 - **Performance Optimizations**: [Multipacking](https://docs.axolotl.ai/docs/multipack.html), [Flash Attention](https://github.com/Dao-AILab/flash-attention), [Xformers](https://github.com/facebookresearch/xformers), [Flex Attention](https://pytorch.org/blog/flexattention/), [Liger Kernel](https://github.com/linkedin/Liger-Kernel), [Cut Cross Entropy](https://github.com/apple/ml-cross-entropy/tree/main), [Sequence Parallelism (SP)](https://docs.axolotl.ai/docs/sequence_parallelism.html), [LoRA optimizations](https://docs.axolotl.ai/docs/lora_optims.html), [Multi-GPU training (FSDP1, FSDP2, DeepSpeed)](https://docs.axolotl.ai/docs/multi-gpu.html), [Multi-node training (Torchrun, Ray)](https://docs.axolotl.ai/docs/multi-node.html), and many more!
 - **Flexible Dataset Handling**: Load from local, HuggingFace, and cloud (S3, Azure, GCP, OCI) datasets.
 - **Cloud Ready**: We ship [Docker images](https://hub.docker.com/u/axolotlai) and also [PyPI packages](https://pypi.org/project/axolotl/) for use on cloud platforms and local hardware.
 
 
 
-## 🚀 Quick Start - LLM Fine-tuning in Minutes
+## 🚀 Quick Start
 
 **Requirements**:
 
@@ -75,10 +70,6 @@ Features:
 - Python 3.11
 - PyTorch ≥2.6.0
 
-### Google Colab
-
-[![Open In Colab](https://colab.research.google.com/assets/colab-badge.svg)](https://colab.research.google.com/github/axolotl-ai-cloud/axolotl/blob/main/examples/colab-notebooks/colab-axolotl-example.ipynb#scrollTo=msOCO4NRmRLa)
-
 ### Installation
 
 #### Using pip
@@ -164,7 +155,7 @@ If you use Axolotl in your research or projects, please cite it as follows:
 
 ```bibtex
 @software{axolotl,
-  title = {Axolotl: Open Source LLM Post-Training},
+  title = {Axolotl: Post-Training for AI Models},
   author = {{Axolotl maintainers and contributors}},
   url = {https://github.com/axolotl-ai-cloud/axolotl},
   license = {Apache-2.0},

_quarto.yml

@@ -153,7 +153,7 @@ quartodoc:
         - utils.distributed
         - utils.dict
         - utils.optimizers.adopt
-        - utils.data.streaming
+        - utils.data.pretraining
         - utils.data.sft
         - utils.quantization
     - title: Schemas
@@ -272,7 +272,6 @@ website:
           contents:
             - docs/batch_vs_grad.qmd
             - docs/dataset_preprocessing.qmd
-            - docs/streaming.qmd
             - docs/multipack.qmd
             - docs/mixed_precision.qmd
             - docs/optimizers.qmd

cicd/multigpu.py

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

cicd/single_gpu.py

@@ -1,5 +1,7 @@
 """Modal app to run axolotl GPU tests"""
 
+# pylint: disable=duplicate-code
+
 import os
 import pathlib
 import tempfile
@@ -57,8 +59,7 @@ VOLUME_CONFIG = {
 }
 
 N_GPUS = int(os.environ.get("N_GPUS", 1))
-GPU_TYPE = os.environ.get("GPU_TYPE", "L40S")
+GPU_CONFIG = f"L40S:{N_GPUS}"
-GPU_CONFIG = f"{GPU_TYPE}:{N_GPUS}"
 
 
 def run_cmd(cmd: str, run_folder: str):
@@ -69,4 +70,4 @@ def run_cmd(cmd: str, run_folder: str):
 
     # Propagate errors from subprocess.
     if exit_code := subprocess.call(cmd.split(), cwd=run_folder, env=sp_env):  # nosec
-        exit(exit_code)
+        exit(exit_code)  # pylint: disable=consider-using-sys-exit

codecov.yml

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

docker/Dockerfile-base

@@ -37,7 +37,7 @@ WORKDIR /workspace
 
 RUN python3 -m pip install --upgrade pip && pip3 install -U packaging==23.2 setuptools==75.8.0 wheel && \
     python3 -m pip install --no-cache-dir -U torch==${PYTORCH_VERSION}+cu${CUDA} torchvision --extra-index-url https://download.pytorch.org/whl/cu$CUDA && \
-    CAUSAL_CONV1D_FORCE_CXX11_ABI=TRUE CAUSAL_CONV1D_FORCE_BUILD=TRUE python3 -m pip install --no-cache-dir causal_conv1d==1.5.2 && \
+    python3 -m pip install --no-cache-dir "causal_conv1d @ git+https://github.com/Dao-AILab/causal-conv1d.git@main" && \
     python3 -m pip install --no-cache-dir "mamba_ssm @ git+https://github.com/state-spaces/mamba.git@main" && \
     python3 -m pip cache purge
 

docs/installation.qmd

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

docs/multi-gpu.qmd

@@ -63,6 +63,15 @@ Start from Stage 1 -> Stage 2 -> Stage 3.
 
 :::
 
+::: {.callout-tip}
+
+Using ZeRO Stage 3 with Single-GPU training
+
+ZeRO Stage 3 can be used for training on a single GPU by manually setting the environment variables:
+`WORLD_SIZE=1 LOCAL_RANK=0 MASTER_ADDR=0.0.0.0 MASTER_PORT=29500`
+
+:::
+
 ## Fully Sharded Data Parallel (FSDP) {#sec-fsdp}
 
 ::: {.callout-note}

docs/multimodal.qmd

@@ -13,13 +13,10 @@ format:
 - [Pixtral](#sec-pixtral)
 - [Llava-1.5](#sec-llava-15)
 - [Mistral-Small-3.1](#sec-mistral-small-31)
-- [Voxtral](#sec-voxtral)
 - [Gemma-3](#sec-gemma-3)
 - [Gemma-3n](#sec-gemma-3n)
 - [Qwen2-VL](#sec-qwen2-vl)
 - [Qwen2.5-VL](#sec-qwen25-vl)
-- [SmolVLM2](#sec-smolvlm2)
-- [LFM2-VL](#sec-lfm2-vl)
 
 ## Usage
 
@@ -34,7 +31,7 @@ skip_prepare_dataset: true
 remove_unused_columns: false  # leave columns in place as they are needed to handle image embeddings during training
 sample_packing: false  # not yet supported with multimodal
 
-chat_template:  # see in next section if specified
+chat_template:  # see in next section
 
 # example dataset
 datasets:
@@ -100,16 +97,6 @@ base_model: mistralai/Mistral-Small-3.1-24B-Instruct-2503
 chat_template: mistral_v7_tekken
 ```
 
-### Voxtral {#sec-voxtral}
-
-::: {.callout-tip}
-Please make sure to install audio lib via `pip3 install librosa==0.11.0 'mistral_common[audio]==1.8.3'`
-:::
-
-```yaml
-base_model: mistralai/Voxtral-Mini-3B-2507
-```
-
 ### Gemma-3 {#sec-gemma-3}
 
 ::: {.callout-tip}
@@ -156,26 +143,6 @@ base_model: Qwen/Qwen2.5-VL-7B-Instruct
 chat_template: qwen2_vl  # same as qwen2-vl
 ```
 
-### SmolVLM2 {#sec-smolvlm2}
-
-::: {.callout-tip}
-Please make sure to install `num2words` via `pip3 install num2words==0.5.14`
-:::
-
-```yaml
-base_model: HuggingFaceTB/SmolVLM2-500M-Video-Instruct
-```
-
-### LFM2-VL {#sec-lfm2-vl}
-
-::: {.callout-warning}
-Please uninstall `causal-conv1d` via `pip3 uninstall -y causal-conv1d`
-:::
-
-```yaml
-base_model: LiquidAI/LFM2-VL-450M
-```
-
 ## Dataset Format
 
 For multi-modal datasets, we adopt an extended `chat_template` format similar to OpenAI's Message format.
@@ -214,20 +181,6 @@ You may need to install `librosa` via `pip3 install librosa==0.11.0`.
 
 :::
 
-### Video
-
-::: {.callout-warning}
-
-This is not well tested at the moment. We welcome contributors!
-
-:::
-
-For video loading, you can use the following keys within `content` alongside `"type": "video"`:
-
-- `"path": "/path/to/video.mp4"`
-- `"url": "https://example.com/video.mp4"`
-- `"video": np.ndarray | list[PIL.Image.Image] | torch.Tensor` (or list of the aforementioned)
-
 ### Example
 
 Here is an example of a multi-modal dataset:

docs/quantize.qmd

@@ -51,11 +51,3 @@ axolotl quantize qat.yml
 ```
 
 This ensures that an identical quantization configuration is used to quantize the model as was used to train it.
-
-
-::: {.callout-note}
-
-If you have configured pushing to hub with `hub_model_id`, your model hub name will have the quantization schema appended to it,
-e.g. `axolotl-ai-cloud/qat-nvfp4-llama3B` will become `axolotl-ai-cloud/qat-nvfp4-llama3B-nvfp4w`
-
-:::

docs/reward_modelling.qmd

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

docs/scripts/generate_config_docs.py

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

docs/streaming.qmd

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

examples/LiquidAI/README.md

@@ -1,58 +0,0 @@
-# Finetune Liquid Foundation Models 2 (LFM2) with Axolotl
-
-[Liquid Foundation Models 2 (LFM2)](https://huggingface.co/collections/LiquidAI/lfm2-686d721927015b2ad73eaa38) are a family of small, open-weight models from [Liquid AI](https://www.liquid.ai/) focused on quality, speed, and memory efficiency. Liquid AI released text-only [LFM2](https://huggingface.co/collections/LiquidAI/lfm2-686d721927015b2ad73eaa38) and text+vision [LFM2-VL](https://huggingface.co/collections/LiquidAI/lfm2-vl-68963bbc84a610f7638d5ffa) models.
-
-LFM2 features a new hybrid Liquid architecture with multiplicative gates, short-range convolutions, and grouped query attention, enabling fast training and inference.
-
-This guide shows how to fine-tune both the LFM2 and LFM2-VL models with Axolotl.
-
-## Getting Started
-
-1.  Install Axolotl following the [installation guide](https://docs.axolotl.ai/docs/installation.html).
-
-    Here is an example of how to install from pip:
-    ```bash
-    # Ensure you have a compatible version of Pytorch installed
-    pip3 install packaging setuptools wheel ninja
-    pip3 install --no-build-isolation 'axolotl[flash-attn]>=0.12.0'
-    ```
-
-2.  Run one of the finetuning examples below.
-
-    **LFM2**
-    ```bash
-    # FFT SFT (1x48GB @ 25GiB)
-    axolotl train examples/LiquidAI/lfm2-350m-fft.yaml
-    ```
-
-    **LFM2-VL**
-    ```bash
-    # LoRA SFT (1x48GB @ 2.7GiB)
-    axolotl train examples/LiquidAI/lfm2-vl-lora.yaml
-    ```
-
-### TIPS
-
-- **Installation Error**: If you encounter `ImportError: ... undefined symbol ...` or `ModuleNotFoundError: No module named 'causal_conv1d_cuda'`, the `causal-conv1d` package may have been installed incorrectly. Try uninstalling it:
-  ```bash
-  pip uninstall -y causal-conv1d
-  ```
-
-- **Dataset Loading**: Read more on how to load your own dataset in our [documentation](https://docs.axolotl.ai/docs/dataset_loading.html).
-- **Dataset Formats**:
-  - For LFM2 models, the dataset format follows the OpenAI Messages format as seen [here](https://docs.axolotl.ai/docs/dataset-formats/conversation.html#chat_template).
-  - For LFM2-VL models, Axolotl follows the multi-content Messages format. See our [Multimodal docs](https://docs.axolotl.ai/docs/multimodal.html#dataset-format) for details.
-
-## Optimization Guides
-
-- [Multi-GPU Training](https://docs.axolotl.ai/docs/multi-gpu.html)
-- [LoRA Optimizations](https://docs.axolotl.ai/docs/lora_optims.html)
-- [Multi-Node Training](https://docs.axolotl.ai/docs/multi-node.html)
-
-## Related Resources
-
-- [LFM2 Blog](https://www.liquid.ai/blog/liquid-foundation-models-v2-our-second-series-of-generative-ai-models)
-- [LFM2-VL Blog](https://www.liquid.ai/blog/lfm2-vl-efficient-vision-language-models)
-- [Axolotl Docs](https://docs.axolotl.ai)
-- [Axolotl GitHub](https://github.com/axolotl-ai-cloud/axolotl)
-- [Axolotl Discord](https://discord.gg/7m9sfhzaf3)

examples/LiquidAI/lfm2-vl-lora.yaml

@@ -1,58 +0,0 @@
-base_model: LiquidAI/LFM2-VL-450M
-trust_remote_code: true
-model_type: AutoModelForImageTextToText
-processor_type: AutoProcessor
-
-# these 3 lines are needed for now to handle vision chat templates w images
-skip_prepare_dataset: true
-remove_unused_columns: false
-sample_packing: false
-
-datasets:
-  - path: HuggingFaceH4/llava-instruct-mix-vsft
-    type: chat_template
-    split: train[:1%]
-
-dataset_prepared_path: last_run_prepared
-val_set_size: 0.0
-output_dir: ./outputs/out
-
-adapter: lora
-lora_model_dir:
-
-sequence_len: 8192
-pad_to_sequence_len: false
-
-lora_r: 32
-lora_alpha: 16
-lora_dropout: 0.05
-lora_target_modules: 'model.language_model.layers.[\d]+.(mlp|cross_attn|self_attn).(up|down|gate|q|k|v|o)_proj'
-
-wandb_project:
-wandb_entity:
-wandb_watch:
-wandb_name:
-wandb_log_model:
-
-gradient_accumulation_steps: 4
-micro_batch_size: 1
-num_epochs: 1
-optimizer: adamw_bnb_8bit
-lr_scheduler: cosine
-learning_rate: 0.0002
-
-bf16: true
-fp16:
-tf32: true
-
-gradient_checkpointing: true
-logging_steps: 1
-flash_attention: true
-eager_attention:
-
-warmup_ratio: 0.1
-evals_per_epoch: 1
-saves_per_epoch: 1
-weight_decay: 0.0
-
-# save_first_step: true  # uncomment this to validate checkpoint saving works with your config

examples/cloud/baseten.yaml

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

examples/devstral/README.md

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

examples/gemma3/270m-qlora.yml

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

examples/gpt-oss/README.md

@@ -33,64 +33,13 @@ Note: Memory usage taken from `device_mem_reserved(gib)` from logs.
 
 ### Training 120B
 
-On 8xH100s, make sure you have ~3TB of free disk space. With each checkpoint clocking in at ~720GB, along with the base
+On 8xH100s
-model, and final model output, you may need at least 3TB of free disk space to keep at least 2 checkpoints.
 
 ```bash
 # FFT SFT with offloading (8x80GB @ ~49GiB/GPU)
 axolotl train examples/gpt-oss/gpt-oss-120b-fft-fsdp2-offload.yaml
 ```
 
-To simplify fine-tuning across 2 nodes × 8x H100 (80GB) GPUs, we've partnered with [Baseten](https://baseten.co) to showcase multi-node
-training of the 120B model using Baseten Truss. You can read more about this recipe on
-[Baseten's blog](https://www.baseten.co/blog/how-to-fine-tune-gpt-oss-120b-with-baseten-and-axolotl/). The recipe can
-be found on their
-[GitHub](https://github.com/basetenlabs/ml-cookbook/tree/main/examples/oss-gpt-120b-axolotl/training).
-
-ERRATA: Transformers saves the model Architecture prefixed with `FSDP` which needs to be manually renamed in `config.json`.
-See https://github.com/huggingface/transformers/pull/40207 for the status of this issue.
-
-```bash
-sed -i 's/FSDPGptOssForCausalLM/GptOssForCausalLM/g' ./outputs/gpt-oss-out/config.json
-```
-
-When using SHARDED_STATE_DICT with FSDP, the final checkpoint should automatically merge the sharded weights to your
-configured `output_dir`. However, if that step fails due to a disk space error, you can take an additional step to
-merge the sharded weights.  This step will automatically determine the last checkpoint directory and merge the sharded
-weights to `{output_dir}/merged`.
-
-```bash
-axolotl merge-sharded-fsdp-weights examples/gpt-oss/gpt-oss-120b-fft-fsdp2-offload.yaml
-mv ./outputs/gpt-oss-out/merged/* ./outputs/gpt-oss-out/
-```
-
-
-### Inferencing your fine-tuned model
-
-#### vLLM
-
-GPT-OSS support in vLLM does not exist in a stable release yet. See https://x.com/MaziyarPanahi/status/1955741905515323425
-for more information about using a special vllm-openai docker image for inferencing with vLLM.
-
-Optionally, vLLM can be installed from nightly:
-
-```bash
-pip install --no-build-isolation --pre -U vllm --extra-index-url https://wheels.vllm.ai/nightly
-```
-and the vLLM server can be started with the following command (modify `--tensor-parallel-size 8` to match your environment):
-```bash
-vllm serve ./outputs/gpt-oss-out/ --served-model-name axolotl/gpt-oss-20b --host 0.0.0.0 --port 8888  --tensor-parallel-size 8
-```
-
-#### SGLang
-
-SGLang has 0-day support in main, see https://github.com/sgl-project/sglang/issues/8833 for infomation on installing
-SGLang from source. Once you've installed SGLang, run the following command to launch a SGLang server:
-
-```bash
-python3 -m sglang.launch_server --model ./outputs/gpt-oss-out/ --served-model-name axolotl/gpt-oss-120b --host 0.0.0.0 --port 8888 --tp 8
-```
-
 ### Tool use
 
 GPT-OSS has a comprehensive tool understanding. Axolotl supports tool calling datasets for Supervised Fine-tuning.
@@ -106,16 +55,6 @@ See [Nanobit/text-tools-2k-test](https://huggingface.co/datasets/Nanobit/text-to
 
 Refer to [our docs](https://docs.axolotl.ai/docs/dataset-formats/conversation.html#using-tool-use) for more info.
 
-### Thinking and chat_template masking conflict
-
-OpenAI’s Harmony template hides `thinking` in all non-final turns, which conflicts with Axolotl’s `chat_template` masking.
-
-If your dataset has `thinking` content mid-turn, there are two paths we recommend:
-
-- Train only on the last turn. This can be accomplished via chat_template's [train on last doc](https://docs.axolotl.ai/docs/dataset-formats/conversation.html#training-on-last-message).
-
-- Adjust your dataset to only have `thinking` content in the last turn.
-
 ### TIPS
 
 - Read more on how to load your own dataset at [docs](https://docs.axolotl.ai/docs/dataset_loading.html).

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

@@ -20,7 +20,6 @@ datasets:
 dataset_prepared_path: last_run_prepared
 val_set_size: 0
 output_dir: ./outputs/gpt-oss-out/
-save_total_limit: 2  # the 120B model can use up to 720GB of disk space per checkpoint, so let's only keep the last 2
 
 sequence_len: 4096
 sample_packing: true
@@ -44,7 +43,7 @@ bf16: true
 tf32: true
 
 flash_attention: true
-attn_implementation: kernels-community/vllm-flash-attn3  # this is not needed if using flash_attn >= 2.8.3
+attn_implementation: kernels-community/vllm-flash-attn3
 
 gradient_checkpointing: true
 activation_offloading: true

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

@@ -40,7 +40,7 @@ bf16: true
 tf32: true
 
 flash_attention: true
-attn_implementation: kernels-community/vllm-flash-attn3  # this is not needed if using flash_attn >= 2.8.3
+attn_implementation: kernels-community/vllm-flash-attn3
 
 gradient_checkpointing: true
 activation_offloading: true

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

@@ -15,7 +15,7 @@ datasets:
     field_thinking: thinking
     template_thinking_key: thinking
 
-dataset_prepared_path: ./outputs/last_run_prepared
+dataset_prepared_path: last_run_prepared
 val_set_size: 0
 output_dir: ./outputs/gpt-oss-out/
 
@@ -41,7 +41,7 @@ bf16: true
 tf32: true
 
 flash_attention: true
-attn_implementation: kernels-community/vllm-flash-attn3  # this is not needed if using flash_attn >= 2.8.3
+attn_implementation: kernels-community/vllm-flash-attn3
 
 gradient_checkpointing: true
 activation_offloading: true

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

@@ -15,7 +15,7 @@ datasets:
     field_thinking: thinking
     template_thinking_key: thinking
 
-dataset_prepared_path: ./outputs/last_run_prepared
+dataset_prepared_path: last_run_prepared
 val_set_size: 0
 output_dir: ./outputs/gpt-oss-out/
 
@@ -40,7 +40,7 @@ bf16: true
 tf32: true
 
 flash_attention: true
-attn_implementation: kernels-community/vllm-flash-attn3  # this is not needed if using flash_attn >= 2.8.3
+attn_implementation: kernels-community/vllm-flash-attn3
 
 gradient_checkpointing: true
 activation_offloading: true

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

@@ -53,7 +53,7 @@ bf16: true
 tf32: true
 
 flash_attention: true
-attn_implementation: kernels-community/vllm-flash-attn3  # this is not needed if using flash_attn >= 2.8.3
+attn_implementation: kernels-community/vllm-flash-attn3
 
 gradient_checkpointing: true
 activation_offloading: true

examples/hunyuan/README.md

@@ -1,85 +0,0 @@
-# Finetune HunYuan with Axolotl
-
-Tencent released a family of opensource models called HunYuan with varying parameter scales of 0.5B, 1.8B, 4B, and 7B scale for both Pre-trained and Instruct variants. The models can be found at [HuggingFace](https://huggingface.co/collections/tencent/hunyuan-dense-model-6890632cda26b19119c9c5e7). This guide shows how to fine-tune it with Axolotl with multi-turn conversations and proper masking.
-
-## Getting started
-
-1. Install Axolotl following the [installation guide](https://docs.axolotl.ai/docs/installation.html). You need to install from main as HunYuan is only on nightly or use our latest [Docker images](https://docs.axolotl.ai/docs/docker.html).
-
-    Here is an example of how to install from main for pip:
-
-```bash
-# Ensure you have Pytorch installed (Pytorch 2.6.0 min)
-git clone https://github.com/axolotl-ai-cloud/axolotl.git
-cd axolotl
-
-pip3 install packaging==23.2 setuptools==75.8.0 wheel ninja
-pip3 install --no-build-isolation -e '.[flash-attn]'
-
-# Install CCE https://docs.axolotl.ai/docs/custom_integrations.html#cut-cross-entropy
-python scripts/cutcrossentropy_install.py | sh
-```
-
-2. Run the finetuning example:
-
-```bash
-axolotl train examples/hunyuan/hunyuan-v1-dense-qlora.yaml
-```
-
-This config uses about 4.7 GB VRAM.
-
-Let us know how it goes. Happy finetuning! 🚀
-
-### Dataset
-
-HunYuan Instruct models can choose to enter a slow think or fast think pattern. For best performance on fine-tuning their Instruct models, your dataset should be adjusted to match their pattern.
-
-```python
-# fast think pattern
-messages = [
-    {"role": "system", "content": "You are a helpful assistant."},
-    {"role": "user", "content": "/no_think What color is the sun?" },
-    {"role": "assistant", "content": "<think>\n\n</think>\n<answer>\nThe sun is yellow.\n</answer>"}
-]
-
-# slow think pattern
-messages = [
-    {"role": "system", "content": "You are a helpful assistant."},
-    {"role": "user", "content": "/no_think What color is the sun?" },
-    {"role": "assistant", "content": "<think>\nThe user is asking about the color of the sun. I need to ...\n</think>\n<answer>\nThe sun is yellow.\n</answer>"}
-]
-```
-
-### TIPS
-
-- For inference, the official Tencent team recommends
-
-```json
-
-{
-  "do_sample": true,
-  "top_k": 20,
-  "top_p": 0.8,
-  "repetition_penalty": 1.05,
-  "temperature": 0.7
-}
-
-```
-
-- You can run a full finetuning by removing the `adapter: qlora` and `load_in_4bit: true` from the config.
-- Read more on how to load your own dataset at [docs](https://docs.axolotl.ai/docs/dataset_loading.html).
-- The dataset format follows the OpenAI Messages format as seen [here](https://docs.axolotl.ai/docs/dataset-formats/conversation.html#chat_template).
-
-## Optimization Guides
-
-- [Multi-GPU Training](https://docs.axolotl.ai/docs/multi-gpu.html)
-- [Multi-Node Training](https://docs.axolotl.ai/docs/multi-node.html)
-- [LoRA Optimizations](https://docs.axolotl.ai/docs/lora_optims.html)
-
-## Related Resources
-
-- [Tencent HunYuan Blog](https://hunyuan.tencent.com/)
-- [Axolotl Docs](https://docs.axolotl.ai)
-- [Axolotl Website](https://axolotl.ai)
-- [Axolotl GitHub](https://github.com/axolotl-ai-cloud/axolotl)
-- [Axolotl Discord](https://discord.gg/7m9sfhzaf3)

examples/hunyuan/hunyuan-v1-dense-qlora.yaml

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

examples/lfm2/README.md

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

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

@@ -2,6 +2,7 @@ base_model: LiquidAI/LFM2-350M
 
 chunked_cross_entropy: true
 
+chat_template: tokenizer_default
 eot_tokens:
   - "<|im_end|>"
 datasets:

examples/llama-3/3b-qat-fsdp2-nvfp4.yaml

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

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

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

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

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

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

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

examples/magistral/README.md

@@ -18,13 +18,7 @@ pip3 install packaging==23.2 setuptools==75.8.0 wheel ninja
 pip3 install --no-build-isolation 'axolotl[flash-attn]>=0.12.0'
 ```
 
-2. Install [Cut Cross Entropy](https://docs.axolotl.ai/docs/custom_integrations.html#cut-cross-entropy) to reduce training VRAM usage
+2. Run the finetuning example:
-
-```bash
-python scripts/cutcrossentropy_install.py | sh
-```
-
-3. Run the finetuning example:
 
 ```bash
 axolotl train examples/magistral/magistral-small-qlora.yaml

examples/qwen3/reward-model.yaml

@@ -1,44 +0,0 @@
-base_model: Skywork/Skywork-Reward-V2-Qwen3-8B
-model_type: AutoModelForSequenceClassification
-num_labels: 1
-
-reward_model: true
-center_rewards_coefficient: 0.01  # Incentivize mean-zero rewards for improved stability
-chat_template: qwen3
-datasets:
-  - path: argilla/distilabel-intel-orca-dpo-pairs
-    type: bradley_terry.chat_template
-
-val_set_size: 0.0
-output_dir: ./outputs/out
-
-sequence_len: 8192
-sample_packing: false
-eval_sample_packing: false
-pad_to_sequence_len: true
-
-deepspeed: deepspeed_configs/zero1.json
-
-wandb_project:
-wandb_entity:
-wandb_watch:
-wandb_name:
-wandb_log_model:
-
-gradient_accumulation_steps: 4
-micro_batch_size: 1
-eval_batch_size: 1
-num_epochs: 3
-optimizer: adamw_bnb_8bit
-lr_scheduler: linear
-learning_rate: 0.00002
-
-bf16: true
-tf32: true
-
-gradient_checkpointing: true
-gradient_checkpointing_kwargs:
-  use_reentrant: false
-warmup_ratio: 0.1
-logging_steps: 1
-weight_decay: 0.01

examples/seed-oss/README.md

@@ -1,54 +0,0 @@
-# Finetune ByteDance's Seed-OSS with Axolotl
-
-[Seed-OSS](https://huggingface.co/collections/ByteDance-Seed/seed-oss-68a609f4201e788db05b5dcd) are a series of 36B parameter open source models trained by ByteDance's Seed Team.
-
-This guide shows how to fine-tune it with Axolotl with multi-turn conversations and proper masking.
-
-## Getting started
-
-1. Install Axolotl following the [installation guide](https://docs.axolotl.ai/docs/installation.html). You need to install from main as Seed-OSS is only on nightly or use our latest [Docker images](https://docs.axolotl.ai/docs/docker.html).
-
-    Here is an example of how to install from main for pip:
-
-```bash
-# Ensure you have Pytorch installed (Pytorch 2.6.0 min)
-git clone https://github.com/axolotl-ai-cloud/axolotl.git
-cd axolotl
-
-pip3 install packaging==23.2 setuptools==75.8.0 wheel ninja
-pip3 install --no-build-isolation -e '.[flash-attn]'
-
-# Install Cut Cross Entropy
-python scripts/cutcrossentropy_install.py | sh
-```
-
-2. Run the finetuning example:
-
-```bash
-axolotl train examples/seed-oss/seed-oss-36b-qlora.yaml
-```
-
-This config uses about 27.7 GiB VRAM.
-
-Let us know how it goes. Happy finetuning! 🚀
-
-### TIPS
-
-- For inference, the official Seed Team recommends `top_p=0.95` and `temperature=1.1`.
-- You can run a full finetuning by removing the `adapter: qlora` and `load_in_4bit: true` from the config.
-- Read more on how to load your own dataset at [docs](https://docs.axolotl.ai/docs/dataset_loading.html).
-- The dataset format follows the OpenAI Messages format as seen [here](https://docs.axolotl.ai/docs/dataset-formats/conversation.html#chat_template).
-
-## Optimization Guides
-
-- [Multi-GPU Training](https://docs.axolotl.ai/docs/multi-gpu.html)
-- [Multi-Node Training](https://docs.axolotl.ai/docs/multi-node.html)
-- [LoRA Optimizations](https://docs.axolotl.ai/docs/lora_optims.html)
-
-## Related Resources
-
-- [ByteDance Seed Website](https://seed.bytedance.com/)
-- [Axolotl Docs](https://docs.axolotl.ai)
-- [Axolotl Website](https://axolotl.ai)
-- [Axolotl GitHub](https://github.com/axolotl-ai-cloud/axolotl)
-- [Axolotl Discord](https://discord.gg/7m9sfhzaf3)

examples/seed-oss/seed-oss-36b-qlora.yaml

@@ -1,56 +0,0 @@
-base_model: ByteDance-Seed/Seed-OSS-36B-Instruct
-
-# Automatically upload checkpoint and final model to HF
-# hub_model_id: username/custom_model_name
-
-plugins:
-  - axolotl.integrations.cut_cross_entropy.CutCrossEntropyPlugin
-
-load_in_8bit: false
-load_in_4bit: true
-
-datasets:
-  - path: fozziethebeat/alpaca_messages_2k_test
-    type: chat_template
-
-dataset_prepared_path: last_run_prepared
-val_set_size: 0.1
-output_dir: ./outputs/lora-out
-
-adapter: qlora
-lora_model_dir:
-
-sequence_len: 2048
-sample_packing: true
-
-lora_r: 32
-lora_alpha: 16
-lora_dropout: 0.05
-lora_target_linear: true
-
-wandb_project:
-wandb_entity:
-wandb_watch:
-wandb_name:
-wandb_log_model:
-
-gradient_accumulation_steps: 4
-micro_batch_size: 2
-num_epochs: 1
-optimizer: adamw_bnb_8bit
-lr_scheduler: cosine
-learning_rate: 0.0002
-
-bf16: auto
-tf32: false
-
-gradient_checkpointing: true
-resume_from_checkpoint:
-logging_steps: 1
-flash_attention: true
-
-warmup_ratio: 0.1
-evals_per_epoch: 1
-saves_per_epoch: 1
-
-# save_first_step: true  # uncomment this to validate checkpoint saving works with your config

examples/smolvlm2/README.md

@@ -1,49 +0,0 @@
-# Finetune SmolVLM2 with Axolotl
-
-[SmolVLM2](https://huggingface.co/collections/HuggingFaceTB/smolvlm2-smallest-video-lm-ever-67ab6b5e84bf8aaa60cb17c7) are a family of lightweight, open-source multimodal models from HuggingFace designed to analyze and understand video, image, and text content.
-
-These models are built for efficiency, making them well-suited for on-device applications where computational resources are limited. Models are available in multiple sizes, including 2.2B, 500M, and 256M.
-
-This guide shows how to fine-tune SmolVLM2 models with Axolotl.
-
-## Getting Started
-
-1.  Install Axolotl following the [installation guide](https://docs.axolotl.ai/docs/installation.html).
-
-    Here is an example of how to install from pip:
-    ```bash
-    # Ensure you have a compatible version of Pytorch installed
-    pip3 install packaging setuptools wheel ninja
-    pip3 install --no-build-isolation 'axolotl[flash-attn]>=0.12.0'
-    ```
-
-2. Install an extra dependency:
-
-    ```bash
-    pip3 install num2words==0.5.14
-    ```
-
-3.  Run the finetuning example:
-
-    ```bash
-    # LoRA SFT (1x48GB @ 6.8GiB)
-    axolotl train examples/smolvlm2/smolvlm2-2B-lora.yaml
-    ```
-
-## TIPS
-
-- **Dataset Format**: For video finetuning, your dataset must be compatible with the multi-content Messages format. For more details, see our documentation on [Multimodal Formats](https://docs.axolotl.ai/docs/multimodal.html#dataset-format).
-- **Dataset Loading**: Read more on how to prepare and load your own datasets in our [documentation](https://docs.axolotl.ai/docs/dataset_loading.html).
-
-## Optimization Guides
-
-- [Multi-GPU Training](https://docs.axolotl.ai/docs/multi-gpu.html)
-- [LoRA Optimizations](https://docs.axolotl.ai/docs/lora_optims.html)
-- [Multi-Node Training](https://docs.axolotl.ai/docs/multi-node.html)
-
-## Related Resources
-
-- [SmolVLM2 Blog](https://huggingface.co/blog/smolvlm2)
-- [Axolotl Docs](https://docs.axolotl.ai)
-- [Axolotl GitHub](https://github.com/axolotl-ai-cloud/axolotl)
-- [Axolotl Discord](https://discord.gg/7m9sfhzaf3)

examples/smolvlm2/smolvlm2-2B-lora.yaml

@@ -1,56 +0,0 @@
-base_model: HuggingFaceTB/SmolVLM2-2.2B-Instruct
-trust_remote_code: true
-processor_type: AutoProcessor
-
-# these 3 lines are needed for now to handle vision chat templates w images
-skip_prepare_dataset: true
-remove_unused_columns: false
-sample_packing: false
-
-datasets:
-  - path: HuggingFaceH4/llava-instruct-mix-vsft
-    type: chat_template
-    split: train[:1%]
-dataset_prepared_path: last_run_prepared
-val_set_size: 0.0
-output_dir: ./outputs/out
-
-adapter: lora
-lora_model_dir:
-
-sequence_len: 8192
-pad_to_sequence_len: false
-
-lora_r: 32
-lora_alpha: 16
-lora_dropout: 0.05
-lora_target_modules: 'model.text_model.layers.[\d]+.(mlp|cross_attn|self_attn).(up|down|gate|q|k|v|o)_proj'
-
-wandb_project:
-wandb_entity:
-wandb_watch:
-wandb_name:
-wandb_log_model:
-
-gradient_accumulation_steps: 4
-micro_batch_size: 1
-num_epochs: 1
-optimizer: adamw_bnb_8bit
-lr_scheduler: cosine
-learning_rate: 0.0002
-
-bf16: true
-fp16:
-tf32: true
-
-gradient_checkpointing: true
-logging_steps: 1
-flash_attention: true
-eager_attention:
-
-warmup_ratio: 0.1
-evals_per_epoch: 1
-saves_per_epoch: 1
-weight_decay: 0.0
-
-# save_first_step: true  # uncomment this to validate checkpoint saving works with your config

examples/streaming/README.md

@@ -1,50 +0,0 @@
-# Streaming Dataset Examples
-
-This directory contains example configurations for using Axolotl's streaming dataset
-functionality, which enables memory-efficient training with large datasets.
-
-## Examples
-
-Run the following examples with e.g. `axolotl train examples/streaming/sft.yaml`; no
-`axolotl preprocess` required!
-
-### Pretraining (`pretrain.yaml`)
-
-Demonstrates streaming configuration for pretraining tasks using the fineweb-edu dataset
-with SmolLM2-135M.
-
-- Uses `pretraining_dataset` configuration for automatic streaming
-- Multipack attention control to prevent cross-attention between packed sequences
-- Buffer size configuration for memory management
-
-### SFT (`sft.yaml`)
-
-Shows how to use streaming for supervised fine-tuning with the Alpaca dataset.
-
-- Explicit `streaming: true` flag for SFT datasets
-- Memory-efficient training on instruction datasets
-- Evaluation datasets are currently not streamed
-
-## Key Configuration Options
-
-### `streaming`
-- Enables streaming mode for standard datasets
-- Automatically enabled for `pretraining_dataset`
-
-### `streaming_multipack_buffer_size`
-- Controls buffer size for sample packing (default: 10,000)
-- Larger values improve packing efficiency but use more memory
-- Adjust based on available memory
-
-### `shuffle_merged_datasets`
-- Enables shuffling of streaming datasets
-- Requires additional memory for shuffle buffer
-
-### `sample_packing`
-- Packs multiple samples into single sequences
-- Minimize per-step padding tokens
-
-## Performance Tips
-
-- Download small / frequently-used datasets locally for better performance
-- Larger buffer sizes improve packing efficiency

examples/streaming/pretrain.yaml

@@ -1,57 +0,0 @@
-base_model: HuggingFaceTB/SmolLM2-135M
-
-# Streaming pretraining configuration
-pretraining_dataset:
-  - path: HuggingFaceFW/fineweb-edu
-    name: sample-10BT
-    type: pretrain
-    text_column: text
-    split: train
-
-# Streaming-specific settings
-streaming_multipack_buffer_size: 10000
-shuffle_merged_datasets: true
-
-# Training configuration
-max_steps: 1000
-output_dir: ./outputs/smollm2-135m-pretrain-streaming
-
-# Sequence and packing settings
-sequence_len: 1024
-sample_packing: true
-pretrain_multipack_attn: true  # Prevent cross-attention between packed sequences
-flash_attention: true
-
-# Batch size settings
-gradient_accumulation_steps: 8
-micro_batch_size: 1
-
-# Optimizer and scheduler
-optimizer: adamw_torch
-lr_scheduler: cosine
-learning_rate: 5e-4
-warmup_ratio: 0.1
-weight_decay: 0.01
-
-# Precision and performance
-bf16: auto
-tf32: true
-
-# Logging and checkpointing
-logging_steps: 10
-save_strategy: steps
-save_steps: 250
-save_total_limit: 3
-
-# Weights & Biases (optional)
-wandb_project:
-wandb_entity:
-wandb_watch:
-wandb_name:
-wandb_log_model:
-
-# Special tokens
-special_tokens:
-  pad_token: "<|endoftext|>"
-
-# save_first_step: true  # uncomment this to validate checkpoint saving works with your config

examples/streaming/sft.yaml

@@ -1,55 +0,0 @@
-base_model: HuggingFaceTB/SmolLM2-135M
-
-# Dataset configuration
-datasets:
-  - path: tatsu-lab/alpaca
-    type: alpaca
-    split: train
-
-# Streaming-specific settings
-streaming: true
-streaming_multipack_buffer_size: 10000
-shuffle_merged_datasets: true
-
-# Training configuration
-max_steps: 1000
-output_dir: ./outputs/smollm2-135m-sft-streaming
-
-# Sequence and packing settings
-sequence_len: 1024
-sample_packing: true
-flash_attention: true
-
-# Batch size settings
-gradient_accumulation_steps: 4
-micro_batch_size: 1
-
-# Optimizer and scheduler
-optimizer: adamw_torch
-lr_scheduler: cosine
-learning_rate: 2e-4
-warmup_ratio: 0.1
-weight_decay: 0.0
-
-# Precision and performance
-bf16: auto
-tf32: true
-
-# Logging and checkpointing
-logging_steps: 10
-save_strategy: steps
-save_steps: 100
-save_total_limit: 3
-
-# Weights & Biases (optional)
-wandb_project:
-wandb_entity:
-wandb_watch:
-wandb_name:
-wandb_log_model:
-
-# Special tokens
-special_tokens:
-  pad_token: "<|endoftext|>"
-
-# save_first_step: true  # uncomment this to validate checkpoint saving works with your config

examples/voxtral/README.md

@@ -22,9 +22,6 @@ pip3 install --no-build-isolation 'axolotl[flash-attn]>=0.12.0'
 # audio
 pip3 install librosa==0.11.0
 pip3 install 'mistral_common[audio]==1.8.3'
-
-# Install CCE https://docs.axolotl.ai/docs/custom_integrations.html#cut-cross-entropy
-python scripts/cutcrossentropy_install.py | sh
 ```
 
 3. Run the finetuning example:

pyproject.toml

@@ -26,34 +26,3 @@ include-package-data = true
 
 [tool.setuptools.cmdclass]
 build_py = "setuptools_axolotl_dynamic_dependencies.BuildPyCommand"
-
-[tool.ruff]
-line-length = 88
-target-version = "py310"
-
-[tool.ruff.lint]
-select = ["E", "F", "W", "C90", "B"]
-ignore = [
-    "E203",  # Whitespace before ':'
-    "E501",  # Line too long
-    "C901",  # Too complex
-    "B019",  # Use of functools.cache on methods
-    "E722",  # Bare except
-    "F821",  # Undefined name (for dynamic exec)
-]
-
-[tool.ruff.lint.isort]
-known-third-party = ["wandb", "comet_ml"]
-known-local-folder = ["src", "tests"]
-# Black-compatible isort settings
-force-single-line = false
-combine-as-imports = true
-split-on-trailing-comma = true
-
-[tool.ruff.format]
-# Use black's formatting style exactly
-quote-style = "double"
-indent-style = "space"
-skip-magic-trailing-comma = false
-line-ending = "auto"
-docstring-code-format = false

requirements.txt

@@ -1,8 +1,9 @@
 --extra-index-url https://huggingface.github.io/autogptq-index/whl/cu118/
 
 # START section of dependencies that don't install on Darwin/MacOS
-bitsandbytes==0.47.0
+bitsandbytes==0.46.1
-triton>=3.0.0
+# triton 3.4.0 is not compatible with CCE
+triton>=3.0.0,<3.4.0
 mamba-ssm==1.2.0.post1
 xformers>=0.0.23.post1
 autoawq==0.2.7.post3
@@ -12,8 +13,8 @@ liger-kernel==0.6.1
 packaging==23.2
 
 huggingface_hub>=0.33.0
-peft>=0.17.0
+peft==0.17.0
-transformers==4.56.1
+transformers==4.55.0
 tokenizers>=0.21.1
 accelerate==1.10.0
 datasets==4.0.0
@@ -64,7 +65,7 @@ langdetect==1.0.9
 immutabledict==4.2.0
 antlr4-python3-runtime==4.13.2
 
-torchao==0.13.0
+torchao==0.12.0
 schedulefree==1.4.1
 
 axolotl-contribs-lgpl==0.0.6

scripts/chat_datasets.py

@@ -27,7 +27,7 @@ def parse_dataset(dataset=None, split="train"):
             break
     if not field_messages:
         raise ValueError(
-            f"No conversation field found in dataset: {', '.join(feature_keys)}"
+            f'No conversation field found in dataset: {", ".join(feature_keys)}'
         )
     ds_cfg["field_messages"] = field_messages
 
@@ -40,7 +40,7 @@ def parse_dataset(dataset=None, split="train"):
             break
     if not message_property_mappings["role"]:
         raise ValueError(
-            f"No role field found in messages: {', '.join(message_fields)}"
+            f'No role field found in messages: {", ".join(message_fields)}'
         )
 
     for key in ["content", "text", "value"]:
@@ -49,7 +49,7 @@ def parse_dataset(dataset=None, split="train"):
             break
     if not message_property_mappings["content"]:
         raise ValueError(
-            f"No content field found in messages: {', '.join(message_fields)}"
+            f'No content field found in messages: {", ".join(message_fields)}'
         )
     ds_cfg["message_property_mappings"] = message_property_mappings
 

scripts/cutcrossentropy_install.py

@@ -29,5 +29,5 @@ UV_PREFIX = "uv " if USE_UV else ""
 
 print(
     UNINSTALL_PREFIX
-    + f'{UV_PREFIX}pip install "cut-cross-entropy[transformers] @ git+https://github.com/axolotl-ai-cloud/ml-cross-entropy.git@c6a32c5"'
+    + f'{UV_PREFIX}pip install "cut-cross-entropy[transformers] @ git+https://github.com/axolotl-ai-cloud/ml-cross-entropy.git@0ee9ee8"'
 )

scripts/unsloth_install.py

@@ -1,10 +1,11 @@
 # noqa
+# pylint: skip-file
 import sys
 
 try:
     import torch
-except ImportError as error:
+except ImportError:
-    raise ImportError("Install torch via `pip install torch`") from error
+    raise ImportError("Install torch via `pip install torch`")
 from packaging.version import Version as V
 
 use_uv = "--uv" in sys.argv[1:]

setup.py

@@ -64,9 +64,7 @@ def parse_requirements(extras_require_map):
             else:
                 raise ValueError("Invalid version format")
 
-            if (major, minor) >= (2, 8):
+            if (major, minor) >= (2, 7):
-                pass
-            elif (major, minor) >= (2, 7):
                 _install_requires.pop(_install_requires.index(xformers_version))
                 if patch == 0:
                     _install_requires.append("xformers==0.0.30")
@@ -120,14 +118,14 @@ def get_package_version():
 
 
 extras_require = {
-    "flash-attn": ["flash-attn==2.8.3"],
+    "flash-attn": ["flash-attn==2.8.2"],
     "ring-flash-attn": [
-        "flash-attn==2.8.3",
+        "flash-attn==2.8.2",
         "ring-flash-attn>=0.1.7",
         "yunchang==0.6.0",
     ],
     "deepspeed": [
-        "deepspeed==0.17.5",
+        "deepspeed==0.17.2",
         "deepspeed-kernels",
     ],
     "mamba-ssm": [
@@ -162,7 +160,6 @@ extras_require = {
     "llmcompressor": [
         "llmcompressor==0.5.1",
     ],
-    "fbgemm-gpu": ["fbgemm-gpu-genai>=1.2.0"],
 }
 install_requires, dependency_links, extras_require_build = parse_requirements(
     extras_require

src/axolotl/cli/args.py

@@ -14,13 +14,9 @@ class PreprocessCliArgs:
     prompter: Optional[str] = field(default=None)
     download: Optional[bool] = field(default=True)
     iterable: Optional[bool] = field(
-        default=False,
+        default=None,
         metadata={
-            "help": (
+            "help": "Use IterableDataset for streaming processing of large datasets"
-                "Deprecated in v0.13.0, will be removed in v0.14.0. For streaming "
-                "datasets, use 'axolotl train' and set 'streaming: true' in your YAML "
-                "config, or pass --streaming instead in the CLI."
-            )
         },
     )
 
@@ -44,12 +40,6 @@ class VllmServeCliArgs:
         default=None,
         metadata={"help": "Number of tensor parallel workers to use."},
     )
-    data_parallel_size: Optional[int] = field(
-        default=None,
-        metadata={
-            "help": "Number of data parallel workers to use for vLLM serving. This controls how many model replicas are used for parallel inference."
-        },
-    )
     host: Optional[str] = field(
         default=None,  # nosec B104
         metadata={"help": "Host address to run the server on."},
@@ -115,7 +105,6 @@ class QuantizeCliArgs:
     quantize_embedding: Optional[bool] = field(default=None)
     group_size: Optional[int] = field(default=None)
     output_dir: Optional[str] = field(default=None)
-    hub_model_id: Optional[str] = field(default=None)
 
 
 @dataclass

src/axolotl/cli/art.py

@@ -22,7 +22,7 @@ HAS_PRINTED_LOGO = False
 def print_axolotl_text_art():
     """Prints axolotl ASCII art."""
 
-    global HAS_PRINTED_LOGO
+    global HAS_PRINTED_LOGO  # pylint: disable=global-statement
     if HAS_PRINTED_LOGO:
         return
     if is_main_process():

src/axolotl/cli/cloud/__init__.py

@@ -7,8 +7,6 @@ from typing import Literal
 
 import yaml
 
-from axolotl.cli.cloud.base import Cloud
-from axolotl.cli.cloud.baseten import BasetenCloud
 from axolotl.cli.cloud.modal_ import ModalCloud
 from axolotl.utils.dict import DictDefault
 
@@ -40,15 +38,8 @@ def do_cli_train(
     cwd=None,
     **kwargs,
 ) -> None:
-    cloud_cfg: DictDefault = load_cloud_cfg(cloud_config)
+    cloud_cfg = load_cloud_cfg(cloud_config)
-    provider = cloud_cfg.provider or "modal"
+    cloud = ModalCloud(cloud_cfg)
-    cloud: Cloud | None
-    if provider == "modal":
-        cloud = ModalCloud(cloud_cfg)
-    elif provider == "baseten":
-        cloud = BasetenCloud(cloud_cfg.to_dict())
-    else:
-        raise ValueError(f"Unsupported cloud provider: {provider}")
     with open(config, "r", encoding="utf-8") as file:
         config_yaml = file.read()
     local_dirs = {}

src/axolotl/cli/cloud/baseten/__init__.py

@@ -1,48 +0,0 @@
-"""Baseten Cloud CLI"""
-
-import shutil
-import subprocess  # nosec B404
-import tempfile
-from os.path import dirname
-from typing import Literal
-
-import yaml
-
-from axolotl.cli.cloud.base import Cloud
-
-
-class BasetenCloud(Cloud):
-    """Baseten Cloud Axolotl CLI"""
-
-    def __init__(self, config: dict):
-        self.config = config
-
-    def preprocess(self, config_yaml: str, *args, **kwargs) -> None:
-        raise NotImplementedError(
-            "Separate preprocess function for Baseten is not "
-            "implemented and will happen during hte train step."
-        )
-
-    def train(
-        self,
-        config_yaml: str,
-        launcher: Literal["accelerate", "torchrun", "python"] = "accelerate",
-        launcher_args: list[str] | None = None,
-        local_dirs: dict[str, str] | None = None,  # pylint: disable=unused-argument
-        **kwargs,
-    ):
-        with tempfile.TemporaryDirectory() as tmp_dir:
-            config = self.config.copy()
-            config["launcher"] = launcher
-            config["launcher_args"] = launcher_args
-            with open(tmp_dir + "/cloud.yaml", "w", encoding="utf-8") as cloud_fout:
-                yaml.dump(config, cloud_fout)
-            with open(tmp_dir + "/train.yaml", "w", encoding="utf-8") as config_fout:
-                config_fout.write(config_yaml)
-            shutil.copyfile(dirname(__file__) + "/template/run.sh", tmp_dir + "/run.sh")
-            shutil.copyfile(
-                dirname(__file__) + "/template/train_sft.py", tmp_dir + "/train_sft.py"
-            )
-            subprocess.run(  # nosec B603 B607
-                ["truss", "train", "push", "train_sft.py"], cwd=tmp_dir, check=False
-            )

src/axolotl/cli/cloud/baseten/template/run.sh

@@ -1,9 +0,0 @@
-#!/bin/bash
-set -eux
-
-export NCCL_SOCKET_IFNAME="^docker0,lo"
-export NCCL_IB_DISABLE=0
-export NCCL_TIMEOUT=1800000
-
-axolotl preprocess train.yaml
-axolotl train train.yaml --launcher ${AXOLOTL_LAUNCHER} ${AXOLOTL_LAUNCHER_ARGS}

src/axolotl/cli/cloud/baseten/template/train_sft.py

@@ -1,71 +0,0 @@
-"""
-Baseten Training Script for Axolotl
-"""
-
-# pylint: skip-file
-import yaml
-from truss.base import truss_config
-
-# Import necessary classes from the Baseten Training SDK
-from truss_train import definitions
-
-cloud_config = yaml.safe_load(open("cloud.yaml", "r"))
-gpu = cloud_config.get("gpu", "h100")
-gpu_count = int(cloud_config.get("gpu_count", 1))
-node_count = int(cloud_config.get("node_count", 1))
-project_name = cloud_config.get("project_name", "axolotl-project") or "axolotl-project"
-secrets = cloud_config.get("secrets", [])
-launcher = cloud_config.get("launcher", "accelerate")
-launcher_args = cloud_config.get("launcher_args", [])
-script_name = "run.sh"
-
-launcher_args_str = ""
-if launcher_args:
-    launcher_args_str = "-- " + " ".join(launcher_args)
-
-# 1. Define a base image for your training job
-# must use torch 2.7.0 for vllm
-BASE_IMAGE = "axolotlai/axolotl:main-py3.11-cu126-2.7.1"
-
-# 2. Define the Runtime Environment for the Training Job
-# This includes start commands and environment variables.a
-# Secrets from the baseten workspace like API keys are referenced using
-# `SecretReference`.
-
-env_vars = {
-    "AXOLOTL_LAUNCHER": launcher,
-    "AXOLOTL_LAUNCHER_ARGS": launcher_args_str,
-}
-for secret_name in secrets:
-    env_vars[secret_name] = definitions.SecretReference(name=secret_name)
-
-training_runtime = definitions.Runtime(
-    start_commands=[  # Example: list of commands to run your training script
-        f"/bin/sh -c 'chmod +x ./{script_name} && ./{script_name}'"
-    ],
-    environment_variables=env_vars,
-)
-
-# 3. Define the Compute Resources for the Training Job
-training_compute = definitions.Compute(
-    node_count=node_count,
-    accelerator=truss_config.AcceleratorSpec(
-        accelerator=truss_config.Accelerator.H100,
-        count=gpu_count,
-    ),
-)
-
-# 4. Define the Training Job
-# This brings together the image, compute, and runtime configurations.
-my_training_job = definitions.TrainingJob(
-    image=definitions.Image(base_image=BASE_IMAGE),
-    compute=training_compute,
-    runtime=training_runtime,
-)
-
-
-# This config will be pushed using the Truss CLI.
-# The association of the job to the project happens at the time of push.
-first_project_with_job = definitions.TrainingProject(
-    name=project_name, job=my_training_job
-)

src/axolotl/cli/cloud/modal_.py

@@ -41,7 +41,7 @@ def run_cmd(cmd: str, run_folder: str, volumes=None):
     if exit_code := subprocess.call(  # nosec B603
         cmd.split(), cwd=run_folder, env=new_env
     ):
-        exit(exit_code)
+        exit(exit_code)  # pylint: disable=consider-using-sys-exit
 
     # Commit writes to volume.
     if volumes:
@@ -82,7 +82,7 @@ class ModalCloud(Cloud):
         return res
 
     def get_image(self):
-        docker_tag = "main-py3.11-cu126-2.7.1"
+        docker_tag = "main-py3.11-cu124-2.6.0"
         if self.config.docker_tag:
             docker_tag = self.config.docker_tag
         docker_image = f"axolotlai/axolotl:{docker_tag}"
@@ -130,6 +130,7 @@ class ModalCloud(Cloud):
         res = []
         if self.config.secrets:
             for key in self.config.get("secrets", []):
+                # pylint: disable=duplicate-code
                 if isinstance(key, str):
                     if val := os.environ.get(key, ""):
                         res.append(modal.Secret.from_dict({key: val}))
@@ -176,8 +177,8 @@ class ModalCloud(Cloud):
             with self.app.run(detach=True):
                 modal_fn.remote(
                     config_yaml,
-                    *args,
                     volumes={k: v[0] for k, v in self.volumes.items()},
+                    *args,
                     **kwargs,
                 )
 
@@ -186,7 +187,7 @@ class ModalCloud(Cloud):
             return int(self.config.timeout)
         return 60 * 60 * 24  # 24 hours
 
-    def get_train_gpu(self):
+    def get_train_gpu(self):  # pylint: disable=too-many-return-statements
         count = self.config.gpu_count or 1
         family = self.config.gpu.lower() or "l40s"
 
@@ -199,7 +200,7 @@ class ModalCloud(Cloud):
         if family in ["a10", "a10g"]:
             return modal.gpu.A10G(count=count)
         if family == "h100":
-            return f"H100:{count}"
+            return modal.gpu.H100(count=count)
         if family == "t4":
             return modal.gpu.T4(count=count)
         if family == "l4":
@@ -276,7 +277,7 @@ def _train(
     launcher: Literal["accelerate", "torchrun", "python"] = "accelerate",
     launcher_args: list[str] | None = None,
     volumes=None,
-    **kwargs,
+    **kwargs,  # pylint: disable=unused-argument
 ):
     Path("/workspace/mounts").mkdir(parents=True, exist_ok=True)
     with open("/workspace/mounts/config.yaml", "w", encoding="utf-8") as f_out:

src/axolotl/cli/config.py

@@ -210,7 +210,7 @@ def load_cfg(
     try:
         device_props = torch.cuda.get_device_properties("cuda")
         gpu_version = "sm_" + str(device_props.major) + str(device_props.minor)
-    except:
+    except:  # pylint: disable=bare-except # noqa: E722
         gpu_version = None
 
     prepare_plugins(cfg)

src/axolotl/cli/evaluate.py

@@ -28,7 +28,7 @@ def do_evaluate(cfg: DictDefault, cli_args: TrainerCliArgs) -> None:
         cfg: Dictionary mapping `axolotl` config keys to values.
         cli_args: CLI arguments.
     """
-
+    # pylint: disable=duplicate-code
     check_accelerate_default_config()
     if int(os.getenv("LOCAL_RANK", "0")) == 0:
         check_user_token()
@@ -49,7 +49,7 @@ def do_cli(config: Union[Path, str] = Path("examples/"), **kwargs) -> None:
         config: Path to `axolotl` config YAML file.
         kwargs: Additional keyword arguments to override config file values.
     """
-
+    # pylint: disable=duplicate-code
     parsed_cfg = load_cfg(config, **kwargs)
     parser = HfArgumentParser(TrainerCliArgs)
     parsed_cli_args, _ = parser.parse_args_into_dataclasses(

src/axolotl/cli/inference.py

@@ -14,14 +14,10 @@ from transformers import GenerationConfig, TextIteratorStreamer, TextStreamer
 from axolotl.cli.args import InferenceCliArgs
 from axolotl.cli.config import load_cfg
 from axolotl.cli.utils import load_model_and_tokenizer
-from axolotl.cli.utils.diffusion import (
+from axolotl.utils.chat_templates import (
-    diffusion_inference,
+    get_chat_template,
-    launch_diffusion_gradio_ui,
+    get_chat_template_from_config,
-    render_html,
-    run_diffusion,
 )
-from axolotl.integrations.base import PluginManager
-from axolotl.utils.chat_templates import get_chat_template_from_config
 from axolotl.utils.dict import DictDefault
 from axolotl.utils.logging import get_logger
 
@@ -36,11 +32,10 @@ def get_multi_line_input() -> str:
         Possibly multi-line, possibly empty stdin input as a string.
     """
     print("Give me an instruction (Ctrl + D to submit): ")
-    print("=" * 80)
 
     instruction = ""
     for line in sys.stdin:
-        instruction += line
+        instruction += line  # pylint: disable=consider-using-join
 
     return instruction
 
@@ -51,9 +46,9 @@ def do_inference(
     cli_args: InferenceCliArgs,
 ):
     """
-    Runs inference on the command line in a loop. User input is accepted, a chat
+    Runs inference on the command line in a loop. User input is accepted, a chat template
-    template is (optionally) applied, and the model specified in the `axolotl` config is
+    is (optionally) applied, and the model specified in the `axolotl` config is used to
-    used to generate completions according to a default generation config.
+    generate completions according to a default generation config.
 
     Args:
         cfg: Dictionary mapping `axolotl` config keys to values.
@@ -69,31 +64,17 @@ def do_inference(
             importlib.import_module("axolotl.prompters"), prompter
         )
     elif cfg.chat_template:
-        chat_template_str = get_chat_template_from_config(
+        chat_template_str = get_chat_template(cfg.chat_template)
-            cfg, ds_cfg=None, tokenizer=tokenizer
+    elif cfg.datasets[0].type == "chat_template":
-        )
-    elif cfg.datasets and cfg.datasets[0].type == "chat_template":
         chat_template_str = get_chat_template_from_config(
             cfg=cfg, ds_cfg=cfg.datasets[0], tokenizer=tokenizer
         )
 
     model = model.to(cfg.device, dtype=cfg.torch_dtype)
 
-    # Detect diffusion mode
-    plugin_manager = PluginManager.get_instance()
-    is_diffusion = any(
-        plugin.__class__.__name__ == "DiffusionPlugin"
-        for plugin in plugin_manager.plugins.values()
-    )
-
-    if is_diffusion:
-        print("=" * 80)
-        print("Commands:")
-        print(":complete N -> completion mode with N tokens (default 64)")
-        print(":mask R     -> random masking with ratio R (0.0–1.0)")
-
     while True:
         print("=" * 80)
+        # support for multiline inputs
         instruction = get_multi_line_input()
         if not instruction:
             return
@@ -123,19 +104,9 @@ def do_inference(
         else:
             batch = tokenizer(prompt, return_tensors="pt", add_special_tokens=True)
 
-        print("=" * 80)
+        print("=" * 40)
         model.eval()
         with torch.no_grad():
-            if is_diffusion:
-                diffusion_inference(
-                    model=model,
-                    tokenizer=tokenizer,
-                    cfg=cfg,
-                    prompt=prompt,
-                    chat_template_str=chat_template_str,
-                )
-                continue
-
             generation_config = GenerationConfig(
                 repetition_penalty=1.1,
                 max_new_tokens=1024,
@@ -158,7 +129,7 @@ def do_inference(
                 generation_config=generation_config,
                 streamer=streamer,
             )
-        print("=" * 80)
+        print("=" * 40)
         print(tokenizer.decode(generated["sequences"].cpu().tolist()[0]))
 
 
@@ -188,37 +159,15 @@ def do_inference_gradio(
             importlib.import_module("axolotl.prompters"), prompter
         )
     elif cfg.chat_template:
-        chat_template_str = get_chat_template_from_config(
+        chat_template_str = get_chat_template(cfg.chat_template, tokenizer=tokenizer)
-            cfg, ds_cfg=None, tokenizer=tokenizer
-        )
-    elif cfg.datasets and cfg.datasets[0].type == "chat_template":
-        chat_template_str = get_chat_template_from_config(
-            cfg=cfg, ds_cfg=cfg.datasets[0], tokenizer=tokenizer
-        )
 
     model = model.to(cfg.device, dtype=cfg.torch_dtype)
 
-    # Detect diffusion mode
-    plugin_manager = PluginManager.get_instance()
-    is_diffusion = any(
-        plugin.__class__.__name__ == "DiffusionPlugin"
-        for plugin in plugin_manager.plugins.values()
-    )
-
-    if is_diffusion:
-        launch_diffusion_gradio_ui(
-            model=model,
-            tokenizer=tokenizer,
-            cfg=cfg,
-            prompter_module=prompter_module,
-            chat_template_str=chat_template_str,
-        )
-        return
-
     def generate(instruction):
         if not instruction:
             return
         if prompter_module:
+            # pylint: disable=stop-iteration-return
             prompt: str = next(
                 prompter_module().build_prompt(instruction=instruction.strip("\n"))
             )
@@ -303,7 +252,7 @@ def do_cli(
         config: Path to `axolotl` config YAML file.
         kwargs: Additional keyword arguments to override config file values.
     """
-
+    # pylint: disable=duplicate-code
     parsed_cfg = load_cfg(config, inference=True, rl=None, **kwargs)
     parsed_cfg.sample_packing = False
     parser = transformers.HfArgumentParser(InferenceCliArgs)

src/axolotl/cli/main.py

@@ -1,5 +1,7 @@
 """Click CLI definitions for various axolotl commands."""
 
+# pylint: disable=redefined-outer-name
+
 import os
 import subprocess  # nosec B404
 from typing import Literal, Optional

src/axolotl/cli/merge_lora.py

@@ -43,10 +43,7 @@ def do_merge_lora(*, cfg: DictDefault) -> None:
             safe_serialization=safe_serialization,
             progressbar=True,
         )
-        tokenizer.save_pretrained(
+        tokenizer.save_pretrained(str(Path(cfg.output_dir) / "merged"))
-            str(Path(cfg.output_dir) / "merged"),
-            save_jinja_files=cfg.tokenizer_save_jinja_files,
-        )
 
         if processor:
             processor.save_pretrained(str(Path(cfg.output_dir) / "merged"))

src/axolotl/cli/merge_sharded_fsdp_weights.py

@@ -10,7 +10,6 @@ import fire
 import torch
 import torch.distributed.checkpoint as dist_cp
 import torch.distributed.checkpoint.format_utils as dist_cp_format_utils
-from accelerate import PartialState
 from accelerate.utils import (
     SAFE_WEIGHTS_INDEX_NAME,
     SAFE_WEIGHTS_NAME,
@@ -24,7 +23,6 @@ from torch.distributed.checkpoint.format_utils import _EmptyStateDictLoadPlanner
 
 from axolotl.cli.config import load_cfg
 from axolotl.utils.logging import get_logger
-from axolotl.utils.train import determine_last_checkpoint
 
 LOG = get_logger(__name__)
 
@@ -32,7 +30,7 @@ LOG = get_logger(__name__)
 class BFloat16CastPlanner(_EmptyStateDictLoadPlanner):
     """A custom planner to cast tensors to bfloat16 on the fly during loading."""
 
-    def commit_tensor(self, read_item, tensor):
+    def commit_tensor(self, read_item, tensor):  # pylint: disable=unused-argument
         tensor.copy_(tensor.to(torch.bfloat16))
 
 
@@ -59,10 +57,10 @@ def _distributed_checkpoint_to_merged_weights(
     state_dict: Dict = {}
     save_path_ = Path(save_path)
     save_path_.mkdir(exist_ok=True)
-    dist_cp_format_utils._load_state_dict(
+    dist_cp_format_utils._load_state_dict(  # pylint: disable=protected-access
         state_dict,
         storage_reader=dist_cp.FileSystemReader(checkpoint_dir),
-        planner=BFloat16CastPlanner(),
+        planner=BFloat16CastPlanner(),  # pylint: disable=protected-access
         no_dist=True,
     )
 
@@ -145,6 +143,7 @@ def merge_fsdp_weights(
         ValueError: If torch version < 2.3.0, or if `checkpoint_dir` does not exist.
     """
     checkpoint_dir_ = Path(checkpoint_dir)
+    from accelerate.state import PartialState
 
     if not is_torch_version(">=", "2.3.0"):
         raise ValueError("`merge_fsdp_weights` requires PyTorch >= 2.3.0`")
@@ -181,6 +180,7 @@ def merge_fsdp_weights(
         if remove_checkpoint_dir:
             LOG.info(f"Removing old checkpoint directory {checkpoint_dir_}")
             shutil.rmtree(checkpoint_dir_)
+    state.wait_for_everyone()
 
 
 def do_cli(config: Union[Path, str] = Path("examples/"), **kwargs):
@@ -191,36 +191,15 @@ def do_cli(config: Union[Path, str] = Path("examples/"), **kwargs):
         config: Path to `axolotl` config YAML file.
         kwargs: Additional keyword arguments to override config file values.
     """
-
+    # pylint: disable=duplicate-code
     parsed_cfg = load_cfg(config, **kwargs)
 
     fsdp_dir = Path(parsed_cfg.output_dir) / "pytorch_model_fsdp_0"
-    if not fsdp_dir.exists():
-        checkpoint_dir = determine_last_checkpoint(parsed_cfg, update=False)
-        if checkpoint_dir:
-            fsdp_dir = Path(checkpoint_dir) / "pytorch_model_fsdp_0"
-        if not fsdp_dir.exists():
-            raise ValueError(
-                f"Could not find FSDP checkpoint `pytorch_model_fsdp_0` in {checkpoint_dir}"
-            )
-
-    output_path = str(Path(parsed_cfg.output_dir) / "merged")
     merge_fsdp_weights(
         checkpoint_dir=str(fsdp_dir),
-        output_path=output_path,
+        output_path=str(Path(parsed_cfg.output_dir) / "merged"),
         safe_serialization=True,
     )
-    state = PartialState()
-    state.wait_for_everyone()
-    LOG.info(
-        f"FSDP SHARDED_STATE_DICT weights successfully merged to: {output_path}",
-        main_process_only=True,
-    )
-    LOG.info(
-        "Merged weights are only the safetensors and doesn't include the model configuration "
-        f"or tokenizer which may be found in {parsed_cfg.output_dir}.",
-        main_process_only=True,
-    )
 
 
 if __name__ == "__main__":

src/axolotl/cli/preprocess.py

@@ -35,20 +35,10 @@ def do_preprocess(cfg: DictDefault, cli_args: PreprocessCliArgs) -> None:
     check_accelerate_default_config()
     check_user_token()
 
-    if cli_args.iterable:
-        LOG.error(
-            "The --iterable CLI argument for 'axolotl preprocess' is no longer "
-            "supported. For training, set 'streaming: true' in your YAML config or "
-            "pass '--streaming' in your 'axolotl train' command for on-the-fly "
-            "preprocessing."
-        )
-        return
-
     for key in ["skip_prepare_dataset", "pretraining_dataset"]:
         if cfg.get(key):
             LOG.error(
-                f"You have set `{key}:`. `preprocess` is not needed. Run the 'axolotl "
+                f"You have set `{key}:`. `preprocess` is not needed. Run the `axolotl train` CLI directly instead."
-                "train' CLI directly instead."
             )
             return
 
@@ -83,7 +73,7 @@ def do_preprocess(cfg: DictDefault, cli_args: PreprocessCliArgs) -> None:
                     AutoModelForCausalLM.from_pretrained(
                         model_name, trust_remote_code=True
                     )
-                except Exception:  # nosec B110
+                except Exception as exc:  # pylint: disable=broad-exception-caught,unused-variable  # nosec B110  # noqa F841
                     pass
                 # fmt: on
 
@@ -105,10 +95,9 @@ def do_cli(
         config: Path to `axolotl` config YAML file.
         kwargs: Additional keyword arguments to override config file values.
     """
-
+    # pylint: disable=duplicate-code
     os.environ["AXOLOTL_IS_PREPROCESS"] = "1"
-    is_preprocess = kwargs.pop("is_preprocess", True)
+    parsed_cfg = load_cfg(config, **kwargs)
-    parsed_cfg = load_cfg(config, is_preprocess=is_preprocess, **kwargs)
     parsed_cfg.is_preprocess = True
     parser = transformers.HfArgumentParser(PreprocessCliArgs)
     parsed_cli_args, _ = parser.parse_args_into_dataclasses(

src/axolotl/cli/quantize.py

@@ -5,17 +5,12 @@ CLI to post-training quantize a model using torchao
 from pathlib import Path
 from typing import Union
 
-from transformers import AutoConfig, AutoModelForCausalLM, TorchAoConfig
+from transformers import AutoModelForCausalLM
 
 from axolotl.cli.config import load_cfg
 from axolotl.loaders import load_tokenizer
 from axolotl.utils.logging import get_logger
-from axolotl.utils.quantization import (
+from axolotl.utils.quantization import TorchIntDType, quantize_model_for_ptq
-    TorchAOQuantDType,
-    get_quantization_config,
-    quantization_config_to_str,
-    quantize_model,
-)
 
 LOG = get_logger(__name__)
 
@@ -48,13 +43,13 @@ def do_quantize(
             "No quantization configuration found. Please specify either qat or quantization in your config file."
         )
 
-    model_path = cli_args.get("base_model") or cfg.output_dir
+    model_path = cli_args.get("model_path") or cfg.output_dir
     if weight_dtype := cli_args.get("weight_dtype"):
-        weight_dtype = TorchAOQuantDType.from_string(weight_dtype)
+        weight_dtype = TorchIntDType[weight_dtype]
     else:
         weight_dtype = quantize_cfg.weight_dtype
     if activation_dtype := cli_args.get("activation_dtype"):
-        activation_dtype = TorchAOQuantDType.from_string(activation_dtype)
+        activation_dtype = TorchIntDType[activation_dtype]
     else:
         activation_dtype = quantize_cfg.activation_dtype
     group_size = cli_args.get("group_size") or quantize_cfg.group_size
@@ -62,15 +57,10 @@ def do_quantize(
         cli_args.get("quantize_embedding") or quantize_cfg.quantize_embedding
     )
     output_dir = cli_args.get("output_dir") or cfg.output_dir
-    hub_model_id = cli_args.get("hub_model_id") or cfg.hub_model_id
 
-    LOG.info(f"Loading model from {model_path}.")
+    LOG.info(f"Loading model from {model_path}...")
     tokenizer = load_tokenizer(cfg)
-    config = AutoConfig.from_pretrained(model_path)
+    model = AutoModelForCausalLM.from_pretrained(model_path, device_map="auto")
-    torch_dtype = config.torch_dtype if hasattr(config, "torch_dtype") else None
-    model = AutoModelForCausalLM.from_pretrained(
-        model_path, device_map="auto", torch_dtype=torch_dtype
-    )
 
     LOG.info(
         f"Quantizing model with configuration: \n"
@@ -80,21 +70,11 @@ def do_quantize(
         f"\tquantize_embedding: {quantize_embedding}"
     )
 
-    quantize_model(
+    quantize_model_for_ptq(
         model, weight_dtype, group_size, activation_dtype, quantize_embedding
     )
 
-    quantization_config = get_quantization_config(
+    LOG.info(f"Saving quantized model to: {str(Path(output_dir) / 'quantized')}...")
-        weight_dtype, activation_dtype, group_size
-    )
-
-    ao_config = TorchAoConfig(
-        quant_type=quantization_config,
-        include_input_output_embeddings=quantize_embedding,
-    )
-    model.config.quantization_config = ao_config
-
-    LOG.info(f"Saving quantized model to: {str(Path(output_dir) / 'quantized')}.")
     model.save_pretrained(
         str(Path(output_dir) / "quantized"),
         safe_serialization=False,
@@ -104,16 +84,5 @@ def do_quantize(
         str(Path(output_dir) / "quantized"),
         safe_serialization=False,
         progressbar=True,
-        save_jinja_files=cfg.tokenizer_save_jinja_files,
     )
-
+    LOG.info(f"Quantized model saved to: {str(Path(output_dir) / 'quantized')}...")
-    if hub_model_id:
-        hub_model_id = (
-            hub_model_id.rstrip("-")
-            + f"-{quantization_config_to_str[type(quantization_config)]}"
-        )
-        model.push_to_hub(hub_model_id, safe_serialization=False)
-        tokenizer.push_to_hub(hub_model_id)
-        LOG.info(f"Quantized model pushed to: {hub_model_id}.")
-
-    LOG.info(f"Quantized model saved to: {str(Path(output_dir) / 'quantized')}.")

src/axolotl/cli/train.py

@@ -59,7 +59,7 @@ def do_cli(config: Union[Path, str] = Path("examples/"), **kwargs):
         config: Path to `axolotl` config YAML file.
         kwargs: Additional keyword arguments to override config file values.
     """
-
+    # pylint: disable=duplicate-code
     parsed_cfg = load_cfg(config, **kwargs)
     parser = HfArgumentParser(TrainerCliArgs)
     parsed_cli_args, _ = parser.parse_args_into_dataclasses(

src/axolotl/cli/utils/args.py

@@ -65,7 +65,7 @@ def add_options_from_dataclass(config_class: Type[Any]) -> Callable:
         for field in reversed(dataclasses.fields(config_class)):
             field_type = _strip_optional_type(field.type)
 
-            if field_type is bool:
+            if field_type == bool:
                 field_name = field.name.replace("_", "-")
                 option_name = f"--{field_name}/--no-{field_name}"
                 function = click.option(
@@ -103,7 +103,7 @@ def add_options_from_config(config_class: Type[BaseModel]) -> Callable:
         for name, field in reversed(config_class.model_fields.items()):
             field_type = _strip_optional_type(field.annotation)
 
-            if field_type is bool:
+            if field_type == bool:
                 field_name = name.replace("_", "-")
                 option_name = f"--{field_name}/--no-{field_name}"
                 function = click.option(

src/axolotl/cli/utils/diffusion.py

@@ -1,375 +0,0 @@
-"""Helpers for diffusion-mode inference in CLI and Gradio."""
-
-from __future__ import annotations
-
-import gradio as gr
-import torch
-from colorama import Fore, Style
-
-from axolotl.integrations.diffusion import generate, resolve_mask_token_id
-from axolotl.utils.dict import DictDefault
-
-
-def diffusion_inference(
-    model,
-    tokenizer,
-    cfg,
-    prompt: str,
-    chat_template_str: str | None = None,
-):
-    """Diffusion inference helper method."""
-    mode = "random"
-    completion_tokens = 0
-    target_mask_ratio = None
-    mode, completion_tokens, target_mask_ratio, cleaned = _parse_commands(prompt)
-
-    if cleaned:
-        prompt = cleaned
-
-    info = run_diffusion(
-        model=model,
-        tokenizer=tokenizer,
-        cfg=cfg,
-        prompt=prompt,
-        chat_template_str=chat_template_str,
-        mode=mode,
-        target_mask_ratio=target_mask_ratio,
-        completion_tokens=completion_tokens,
-    )
-    masked_text = info["masked_text"]
-    mask_ratio = info["mask_ratio"]
-    generated_ids = info["generated_ids"]
-    masked_positions = info["masked_positions"]
-    orig_ids = info["orig_ids"]
-
-    # Display with masked preview and colored diff
-    if masked_text is not None and mask_ratio is not None:
-        print(f"Masked ({mask_ratio:.1%}):\n{masked_text}\n")
-    if generated_ids is not None:
-        # Compute per-token style
-        styles: list[str] = []
-        for i, tid in enumerate(generated_ids):
-            if i in masked_positions:
-                if i < len(orig_ids) and tid == orig_ids[i]:
-                    styles.append("green")  # correct fill
-                elif i < len(orig_ids):
-                    styles.append("red")  # incorrect fill
-                else:
-                    styles.append("normal")  # appended
-            else:
-                same = i < len(orig_ids) and tid == orig_ids[i]
-                styles.append("dim" if same else "normal")
-
-        # Group contiguous spans by style
-        styled_spans: list[tuple[str, int, int]] = []
-        if generated_ids:
-            current_style = styles[0]
-            start = 0
-            for i in range(1, len(generated_ids)):
-                s = styles[i]
-                if s != current_style:
-                    styled_spans.append((current_style, start, i))
-                    current_style, start = s, i
-            styled_spans.append((current_style, start, len(generated_ids)))
-
-        out_parts = []
-        for style_name, a, b in styled_spans:
-            chunk_text = tokenizer.decode(generated_ids[a:b], skip_special_tokens=False)
-            if style_name == "green":
-                out_parts.append(Fore.GREEN + chunk_text + Style.RESET_ALL)
-            elif style_name == "red":
-                out_parts.append(Fore.RED + chunk_text + Style.RESET_ALL)
-            else:
-                if style_name == "dim":
-                    out_parts.append(Style.DIM + chunk_text + Style.RESET_ALL)
-                else:
-                    out_parts.append(chunk_text)
-        print("Generated:\n" + "".join(out_parts))
-    else:
-        print("Generated:\n(no output)")
-
-
-def _parse_commands(text: str):
-    """
-    Parse leading diffusion commands.
-
-    Supported at start of input (can be chained):
-      :complete N  -> completion mode with N tokens (default 64)
-      :mask R      -> random masking with ratio R in [0, 1]
-    """
-    tokens = text.strip().split()
-    i = 0
-    mode = "random"
-    completion_tokens = 0
-    target_mask_ratio = None
-    consumed = 0
-    while i < len(tokens) and tokens[i].startswith(":"):
-        cmd = tokens[i]
-        i += 1
-        consumed = i
-        if cmd == ":complete":
-            mode = "completion"
-            if i < len(tokens):
-                try:
-                    completion_tokens = int(tokens[i])
-                    i += 1
-                    consumed = i
-                except Exception:
-                    completion_tokens = 64
-            else:
-                completion_tokens = 64
-        elif cmd == ":mask":
-            mode = "random"
-            if i < len(tokens):
-                try:
-                    target_mask_ratio = float(tokens[i])
-                    i += 1
-                    consumed = i
-                except Exception:
-                    target_mask_ratio = None
-        else:
-            i -= 1
-            consumed = i
-            break
-
-    cleaned = " ".join(tokens[consumed:])
-
-    return mode, completion_tokens, target_mask_ratio, cleaned
-
-
-def run_diffusion(
-    *,
-    model,
-    tokenizer,
-    cfg: DictDefault,
-    prompt: str,
-    chat_template_str: str | None,
-    mode: str = "random",
-    target_mask_ratio: float | None = None,
-    completion_tokens: int = 0,
-):
-    """Run a single diffusion generation and return a structured result dict."""
-    if chat_template_str:
-        batch = tokenizer.apply_chat_template(
-            [{"role": "user", "content": prompt}],
-            return_tensors="pt",
-            add_special_tokens=True,
-            add_generation_prompt=True,
-            chat_template=chat_template_str,
-            tokenize=True,
-            return_dict=True,
-        )
-    else:
-        batch = tokenizer(prompt, return_tensors="pt", add_special_tokens=True)
-
-    mask_token_id = resolve_mask_token_id(tokenizer, cfg, allow_add=False)
-
-    seq = batch["input_ids"].to(cfg.device)
-    gen_mode = "completion" if mode == "completion" else "random"
-    comp_tokens = int(completion_tokens) if gen_mode == "completion" else 0
-
-    result = generate(
-        model,
-        tokenizer,
-        original_sequence=seq[:1],
-        num_diffusion_steps=cfg.diffusion.num_diffusion_steps,
-        temperature=cfg.diffusion.generation_temperature,
-        mask_token_id=int(mask_token_id),
-        mode=gen_mode,  # type: ignore[arg-type]
-        completion_tokens=comp_tokens,
-        target_mask_ratio=target_mask_ratio,
-    )
-
-    masked_text = result.get("masked") if isinstance(result, dict) else None
-    mask_ratio = result.get("mask_ratio") if isinstance(result, dict) else None
-    generated_ids = result.get("generated_ids") if isinstance(result, dict) else None
-    masked_positions = (
-        set(result.get("masked_positions") or []) if isinstance(result, dict) else set()
-    )
-    orig_ids = seq[0].detach().cpu().tolist()
-
-    return {
-        "masked_text": masked_text,
-        "mask_ratio": mask_ratio,
-        "generated_ids": generated_ids,
-        "masked_positions": masked_positions,
-        "orig_ids": orig_ids,
-    }
-
-
-def render_html(
-    *,
-    generated_ids: list[int] | None,
-    orig_ids: list[int],
-    masked_positions: set[int],
-    tokenizer,
-) -> str:
-    """Render HTML visualizing diffusion outputs."""
-    if not generated_ids:
-        return "<pre>Generated:\n(no output)</pre>"
-
-    def _style_for(i: int, tid: int) -> str:
-        if i in masked_positions:
-            if i < len(orig_ids) and tid == orig_ids[i]:
-                return "green"
-            if i < len(orig_ids):
-                return "red"
-            return "normal"
-        same = i < len(orig_ids) and tid == orig_ids[i]
-        return "dim" if same else "normal"
-
-    # Group contiguous spans by style to reduce HTML size
-    spans: list[tuple[str, int, int]] = []
-    if generated_ids:
-        cur = _style_for(0, generated_ids[0])
-        start = 0
-        for i in range(1, len(generated_ids)):
-            s = _style_for(i, generated_ids[i])
-            if s != cur:
-                spans.append((cur, start, i))
-                cur, start = s, i
-        spans.append((cur, start, len(generated_ids)))
-
-    html_parts = []
-    for style_name, a, b in spans:
-        txt = tokenizer.decode(generated_ids[a:b], skip_special_tokens=False)
-        if style_name == "green":
-            html_parts.append(f'<span style="color:#2e7d32">{txt}</span>')
-        elif style_name == "red":
-            html_parts.append(f'<span style="color:#c62828">{txt}</span>')
-        elif style_name == "dim":
-            html_parts.append(f'<span style="opacity:0.6">{txt}</span>')
-        else:
-            html_parts.append(txt)
-
-    legend = (
-        '<div style="font-size:0.9em;margin-bottom:4px">'
-        '<span style="color:#2e7d32">correct</span>, '
-        '<span style="color:#c62828">incorrect</span>, '
-        '<span style="opacity:0.6">unchanged</span>'
-        "</div>"
-    )
-
-    return (
-        legend
-        + '<pre style="white-space:pre-wrap">Generated:\n'
-        + "".join(html_parts)
-        + "</pre>"
-    )
-
-
-def launch_diffusion_gradio_ui(
-    *,
-    model,
-    tokenizer,
-    cfg: DictDefault,
-    prompter_module=None,
-    chat_template_str: str | None = None,
-):
-    """Build and launch a simple Gradio UI for diffusion inference."""
-    with gr.Blocks(
-        title=cfg.get("gradio_title", "Axolotl Diffusion Interface")
-    ) as demo:
-        gr.Markdown(
-            """
-            ## Axolotl Diffusion Inference
-            - Mode "Random" masks tokens at a target ratio and fills them.
-            - Mode "Completion" appends N masked tokens at the end and fills them.
-            """
-        )
-
-        with gr.Row():
-            mode = gr.Radio(
-                choices=["random", "completion"],
-                value="random",
-                label="Mode",
-            )
-            mask_ratio = gr.Slider(
-                minimum=0.0,
-                maximum=1.0,
-                step=0.05,
-                value=0.4,
-                label="Mask ratio (random mode)",
-                interactive=True,
-            )
-            completion_tokens = gr.Number(
-                value=64,
-                precision=0,
-                label="Completion tokens (completion mode)",
-                interactive=True,
-                visible=False,
-            )
-
-        instruction = gr.Textbox(label="Instruction", lines=6)
-        run_btn = gr.Button("Generate")
-
-        masked_preview = gr.Textbox(label="Masked preview", lines=6)
-        html_out = gr.HTML(label="Generated")
-
-        def _toggle_controls(selected_mode: str):
-            return (
-                gr.update(visible=(selected_mode == "random")),
-                gr.update(visible=(selected_mode == "completion")),
-            )
-
-        mode.change(
-            _toggle_controls,
-            inputs=[mode],
-            outputs=[mask_ratio, completion_tokens],
-        )
-
-        def _gen(instruction_text: str, selected_mode: str, mratio: float, ctoks: int):
-            if not instruction_text:
-                return "", "<pre>Generated:\n(no output)</pre>"
-
-            if prompter_module:
-                prompt: str = next(
-                    prompter_module().build_prompt(
-                        instruction=instruction_text.strip("\n")
-                    )
-                )
-            else:
-                prompt = instruction_text.strip()
-
-            info = run_diffusion(
-                model=model,
-                tokenizer=tokenizer,
-                cfg=cfg,
-                prompt=prompt,
-                chat_template_str=chat_template_str,
-                mode=selected_mode,
-                target_mask_ratio=mratio if selected_mode == "random" else None,
-                completion_tokens=int(ctoks) if selected_mode == "completion" else 0,
-            )
-
-            masked_text = info.get("masked_text")
-            mask_ratio_val = info.get("mask_ratio")
-            generated_ids = info.get("generated_ids")
-            masked_positions = info.get("masked_positions") or set()
-            orig_ids = info.get("orig_ids") or []
-
-            preview = (
-                f"Masked ({mask_ratio_val:.1%}):\n{masked_text}"
-                if masked_text is not None and mask_ratio_val is not None
-                else ""
-            )
-            html = render_html(
-                generated_ids=generated_ids,
-                orig_ids=orig_ids,
-                masked_positions=masked_positions,
-                tokenizer=tokenizer,
-            )
-            return preview, html
-
-        run_btn.click(
-            _gen,
-            inputs=[instruction, mode, mask_ratio, completion_tokens],
-            outputs=[masked_preview, html_out],
-        )
-
-        demo.queue().launch(
-            show_api=False,
-            share=cfg.get("gradio_share", True),
-            server_name=cfg.get("gradio_server_name", "127.0.0.1"),
-            server_port=cfg.get("gradio_server_port", None),
-        )

src/axolotl/cli/utils/sweeps.py

@@ -3,12 +3,11 @@
 import random
 from copy import deepcopy
 from itertools import product
-from typing import Any
 
 
 def generate_sweep_configs(
     base_config: dict[str, list], sweeps_config: dict[str, list]
-) -> list[dict[str, Any]]:
+) -> list[dict[str, list]]:
     """
     Recursively generates all possible configurations by applying sweeps to the base config.
 
@@ -49,10 +48,7 @@ def generate_sweep_configs(
                 new_config = {}
                 # new_config = deepcopy(base_config)
                 # Combine regular parameters with paired parameters
-                full_combo = {
+                full_combo = {**dict(zip(param_names, reg_combo)), **paired_set}
-                    **dict(zip(param_names, reg_combo, strict=False)),
-                    **paired_set,
-                }
                 for param_name, param_value in full_combo.items():
                     new_config[param_name] = param_value
                 print(new_config)
@@ -61,7 +57,7 @@ def generate_sweep_configs(
             # If no paired values, just use regular combinations
             # new_config = deepcopy(base_config)
             new_config = {}
-            for param_name, param_value in zip(param_names, reg_combo, strict=False):
+            for param_name, param_value in zip(param_names, reg_combo):
                 new_config[param_name] = param_value
             print(new_config)
             all_combinations.append(new_config)

src/axolotl/cli/utils/train.py

@@ -4,7 +4,6 @@ import os
 import subprocess  # nosec
 import sys
 import tempfile
-from pathlib import Path
 from typing import Any, Iterator, Literal
 
 import yaml
@@ -89,12 +88,8 @@ def generate_config_files(config: str, sweep: str | None) -> Iterator[tuple[str,
     # Generate all possible configurations
     permutations = generate_sweep_configs(base_config, sweep_config)
     is_group = len(permutations) > 1
-    base_output_dir = base_config.get("output_dir", "./model-out")
+    for permutation in permutations:
-    for idx, permutation in enumerate(permutations, start=1):
+        # pylint: disable=consider-using-with
-        permutation_dir = Path(permutation.get("output_dir", base_output_dir))
-        permutation_id = f"sweep{idx:04d}"
-        permutation["output_dir"] = str(permutation_dir / permutation_id)
-
         temp_file = tempfile.NamedTemporaryFile(
             mode="w",
             suffix=".yaml",

src/axolotl/cli/vllm_serve.py

@@ -39,7 +39,7 @@ def do_vllm_serve(
     model = cfg.base_model
 
     serve_module = cli_args.get("serve_module", "trl.scripts.vllm_serve")
-    vllm_serve_main = __import__(serve_module, fromlist=["main"]).main
+    vllm_serve_main = getattr(__import__(serve_module, fromlist=["main"]), "main")
     tensor_parallel_size = 1
     data_parallel_size = 1
 
@@ -68,6 +68,7 @@ def do_vllm_serve(
         cli_args.get("enable_reasoning") or cfg.vllm.enable_reasoning or False
     )
 
+    # pylint: disable=unexpected-keyword-arg
     vllm_script_args = AxolotlScriptArguments(
         model=model,
         tensor_parallel_size=tensor_parallel_size,

src/axolotl/common/datasets.py

@@ -6,7 +6,7 @@ from dataclasses import dataclass
 
 from datasets import Dataset
 
-import axolotl.monkeypatch.data.batch_dataset_fetcher  # noqa: F401
+import axolotl.monkeypatch.data.batch_dataset_fetcher  # pylint: disable=unused-import  # noqa: F401
 from axolotl.cli.args import PreprocessCliArgs, TrainerCliArgs
 from axolotl.loaders import load_processor, load_tokenizer
 from axolotl.utils.data import prepare_datasets, prepare_preference_datasets
@@ -55,11 +55,13 @@ def load_datasets(
     """
     tokenizer = load_tokenizer(cfg)
     processor = load_processor(cfg, tokenizer=tokenizer) if cfg.processor_type else None
+    preprocess_iterable = getattr(cli_args, "iterable", False)
 
     train_dataset, eval_dataset, total_num_steps, prompters = prepare_datasets(
         cfg,
         tokenizer,
         processor=processor,
+        preprocess_iterable=preprocess_iterable,
     )
 
     if (

src/axolotl/convert.py

@@ -67,7 +67,9 @@ class JsonToJsonlConverter:
         self.json_parser = json_parser
         self.jsonl_serializer = jsonl_serializer
 
-    def convert(self, input_file_path, output_file_path):
+    def convert(
+        self, input_file_path, output_file_path
+    ):  # pylint: disable=unused-argument
         content = self.file_reader.read(input_file_path)
         data = self.json_parser.parse(content)
         # data = [r for r in data if r["conversations"]]  # vicuna cleaned has rows with empty conversations

src/axolotl/core/attention/flex_block_mask.py

@@ -84,7 +84,9 @@ def create_causal_mask(
     batch_size, dtype = input_embeds.shape[0], input_embeds.dtype
     if attention_mask is not None:
 
-        def causal_doc_mask_mod(batch_idx, head_idx, q_idx, kv_idx):
+        def causal_doc_mask_mod(
+            batch_idx, head_idx, q_idx, kv_idx
+        ):  # pylint: disable=unused-argument
             """
             Defines the logic of a block causal mask by combining both a standard causal mask
             and a block diagonal document mask.
@@ -101,7 +103,9 @@ def create_causal_mask(
         mask_factory_function = causal_doc_mask_mod
     else:
         mask_factory_function = causal_mask_function
-    mask_interface = ALL_MASK_ATTENTION_FUNCTIONS[config._attn_implementation]
+    mask_interface = ALL_MASK_ATTENTION_FUNCTIONS[
+        config._attn_implementation  # pylint: disable=protected-access
+    ]
 
     # Do not allow skip if we are compiling (this is to match BC)
     allow_is_causal_skip = (

src/axolotl/core/builders/base.py

@@ -24,7 +24,9 @@ from pathlib import Path
 from typing import Any
 
 import torch
-from transformers import TrainerCallback
+from transformers import (
+    TrainerCallback,
+)
 from transformers.trainer_pt_utils import AcceleratorConfig
 
 from axolotl.integrations.base import PluginManager
@@ -42,7 +44,7 @@ from axolotl.utils.schemas.enums import CustomSupportedOptimizers
 LOG = logging.getLogger(__name__)
 
 with suppress(ImportError):
-    import torch._dynamo
+    import torch._dynamo  # pylint: disable=ungrouped-imports
 
 
 class TrainerBuilderBase(abc.ABC):
@@ -258,14 +260,14 @@ class TrainerBuilderBase(abc.ABC):
                 adam_kwargs["eps"] = training_args_kwargs.get("adam_epsilon")
 
             if self.cfg.optimizer == "muon":
-                from axolotl.contribs.mit.muon import (
+                from axolotl.contribs.mit.muon import (  # pylint: disable=no-name-in-module
                     MuonOptimizerFactory,
                 )
 
                 optimizer_cls = MuonOptimizerFactory
                 optimizer_kwargs.update(adam_kwargs)
             elif self.cfg.optimizer == "dion":
-                from axolotl.contribs.mit.dion import (
+                from axolotl.contribs.mit.dion import (  # pylint: disable=no-name-in-module
                     DionOptimizerFactory,
                 )
 
@@ -412,8 +414,12 @@ class TrainerBuilderBase(abc.ABC):
 
     def _configure_torch_compile(self, training_args_kwargs: dict):
         if self.cfg.torch_compile and getattr(torch, "_dynamo", None):
-            torch._dynamo.config.suppress_errors = True
+            torch._dynamo.config.suppress_errors = (  # pylint: disable=protected-access
-            torch._dynamo.config.accumulated_cache_size_limit = 256
+                True
+            )
+            torch._dynamo.config.accumulated_cache_size_limit = (  # pylint: disable=protected-access
+                256
+            )
             training_args_kwargs["torch_compile"] = self.cfg.torch_compile
             if self.cfg.torch_compile_backend:
                 training_args_kwargs["torch_compile_backend"] = (
@@ -510,7 +516,6 @@ class TrainerBuilderBase(abc.ABC):
                 self.cfg.eval_batch_size
             )
 
-        training_args_kwargs["include_tkps"] = self.cfg.include_tkps
         training_args_kwargs["max_steps"] = self.cfg.max_steps or total_num_steps or -1
         training_args_kwargs["num_train_epochs"] = self.cfg.num_epochs
 

src/axolotl/core/builders/causal.py

@@ -10,7 +10,6 @@ import transformers
 from transformers import (
     DataCollatorWithFlattening,
     EarlyStoppingCallback,
-    Trainer,
 )
 from trl.trainer.utils import RewardDataCollatorWithPadding
 
@@ -36,7 +35,6 @@ from axolotl.utils.callbacks import (
 )
 from axolotl.utils.callbacks.lisa import lisa_callback_factory
 from axolotl.utils.callbacks.qat import QATCallback
-from axolotl.utils.callbacks.tokens_per_second import TokensPerSecondCallback
 from axolotl.utils.chat_templates import get_chat_template_from_config
 from axolotl.utils.collators import (
     BatchSamplerDataCollatorForSeq2Seq,
@@ -76,12 +74,6 @@ class HFCausalTrainerBuilder(TrainerBuilderBase):
         if self.cfg.qat:
             callbacks.append(QATCallback(self.cfg.qat))
 
-        if self.cfg.include_tkps:
-            callbacks.append(
-                TokensPerSecondCallback(
-                    self.cfg.tensor_parallel_size, self.cfg.context_parallel_size
-                )
-            )
         return callbacks
 
     def get_post_trainer_create_callbacks(self, trainer):
@@ -348,22 +340,20 @@ class HFCausalTrainerBuilder(TrainerBuilderBase):
 
         if self.cfg.reward_model:
             training_args_cls = AxolotlRewardConfig
-            if self.cfg.center_rewards_coefficient is not None:
-                training_arguments_kwargs["center_rewards_coefficient"] = (
-                    self.cfg.center_rewards_coefficient
-                )
         elif self.cfg.process_reward_model:
             training_args_cls = AxolotlPRMConfig
         else:
             training_args_cls = AxolotlTrainingArguments
-        training_args = training_args_cls(
+        training_args = training_args_cls(  # pylint: disable=unexpected-keyword-arg
             **training_arguments_kwargs,
         )
         training_args = self.hook_post_create_training_args(training_args)
 
         # unset run_name so wandb sets up experiment names
         if self.cfg.use_wandb and training_args.run_name == training_args.output_dir:
-            training_args.run_name = None
+            training_args.run_name = (  # pylint: disable=attribute-defined-outside-init
+                None
+            )
 
         data_collator_kwargs = {
             "padding": True,  # True/"longest" is the default
@@ -395,11 +385,10 @@ class HFCausalTrainerBuilder(TrainerBuilderBase):
                 **data_collator_kwargs,
             )
         sig = inspect.signature(trainer_cls)
-        if "processing_class" in sig.parameters or issubclass(trainer_cls, Trainer):
+        if "processing_class" in sig.parameters:
             trainer_kwargs["processing_class"] = self.tokenizer
         elif "tokenizer" in sig.parameters:
             trainer_kwargs["tokenizer"] = self.tokenizer
-
         if (
             trainer_cls not in [AxolotlRewardTrainer, AxolotlPRMTrainer]
             and self.cfg.datasets is not None
@@ -417,9 +406,6 @@ class HFCausalTrainerBuilder(TrainerBuilderBase):
             **trainer_kwargs,
         )
         trainer = self.hook_post_create_trainer(trainer)
-        # if the trainer has the `axolotl_cfg` property, set it
-        if hasattr(trainer, "axolotl_cfg"):
-            trainer.axolotl_cfg = self.cfg
         for callback in self.get_post_trainer_create_callbacks(trainer):
             trainer.add_callback(callback)
 

src/axolotl/core/builders/rl.py

@@ -168,14 +168,16 @@ class HFRLTrainerBuilder(TrainerBuilderBase):
             if plugin_training_args:
                 training_args_kwargs.update(plugin_training_args)
 
-        training_args = training_args_cls(
+        training_args = training_args_cls(  # pylint: disable=unexpected-keyword-arg
             logging_first_step=True,
             **training_args_kwargs,
         )
 
         # unset run_name so wandb sets up experiment names
         if self.cfg.use_wandb and training_args.run_name == training_args.output_dir:
-            training_args.run_name = None
+            training_args.run_name = (  # pylint: disable=attribute-defined-outside-init
+                None
+            )
 
         return training_args, trainer_kwargs
 

src/axolotl/core/chat/format/chatml.py

@@ -10,7 +10,7 @@ from .shared import wrap_tools
 
 def format_message(
     message: Messages,
-    message_index: Optional[int] = None,
+    message_index: Optional[int] = None,  # pylint: disable=unused-argument
 ) -> Messages:
     if message.is_chat_formatted:
         return message

src/axolotl/core/chat/messages.py

@@ -15,11 +15,11 @@ class MessageRoles(str, Enum):
     Message roles for the system, user, assistant, and tools
     """
 
-    system = "system"
+    system = "system"  # pylint: disable=invalid-name
-    user = "user"
+    user = "user"  # pylint: disable=invalid-name
-    assistant = "assistant"
+    assistant = "assistant"  # pylint: disable=invalid-name
-    tool = "tool"
+    tool = "tool"  # pylint: disable=invalid-name
-    ipython = (
+    ipython = (  # pylint: disable=invalid-name
         # for responses from builtin tools
         "ipython"
     )
@@ -30,12 +30,12 @@ class MessageContentTypes(str, Enum):
     Message content types for text, image, audio, tool calls, and tool responses
     """
 
-    special_token = "special_token"  # nosec B105
+    special_token = "special_token"  # pylint: disable=invalid-name  # nosec B105
-    text = "text"
+    text = "text"  # pylint: disable=invalid-name
-    image = "image"
+    image = "image"  # pylint: disable=invalid-name
-    audio = "audio"
+    audio = "audio"  # pylint: disable=invalid-name
-    tool_call = "tool_call"
+    tool_call = "tool_call"  # pylint: disable=invalid-name  # to differentiate regular responses from tool calls from the assistant
-    tool_response = "tool_response"
+    tool_response = "tool_response"  # pylint: disable=invalid-name
 
 
 class SpecialToken(str, Enum):
@@ -43,8 +43,8 @@ class SpecialToken(str, Enum):
     Special tokens for beginning of string and end of string
     """
 
-    bos_token = "bos_token"  # nosec B105
+    bos_token = "bos_token"  # pylint: disable=invalid-name  # nosec B105
-    eos_token = "eos_token"  # nosec B105
+    eos_token = "eos_token"  # pylint: disable=invalid-name  # nosec B105
 
 
 class ToolCallFunction(BaseModel):
@@ -73,7 +73,7 @@ class ToolCallContents(BaseModel):
 
     name: str
     arguments: dict[str, Union[str, int]]
-    id: Optional[str] = None
+    id: Optional[str] = None  # pylint: disable=invalid-name
 
     def __str__(self) -> str:
         data = {"name": self.name, "arguments": self.arguments}
@@ -89,7 +89,7 @@ class ToolResponseContents(BaseModel):
 
     name: str
     content: Union[str, dict[str, Union[str, int, float]]]
-    id: Optional[str] = None
+    id: Optional[str] = None  # pylint: disable=invalid-name
 
     def __str__(self) -> str:
         data = {"name": self.name, "content": self.content}

src/axolotl/core/datasets/transforms/chat_builder.py

@@ -1,17 +1,23 @@
 """
-This module contains a function that builds a transform that takes a row from the
+This module contains a function that builds a transform that takes a row from the dataset and converts it to a Chat.
-dataset and converts it to a Chat.
 """
 
-from typing import Any, Mapping
+from typing import Any, Mapping, Union
 
 
-def chat_message_transform_builder(
+def chat_message_transform_builder(  # pylint: disable=dangerous-default-value
     train_on_inputs=False,
     conversations_field: str = "conversations",
-    message_field_role: str | list[str] | None = None,  # commonly "role"
+    message_field_role: Union[str, list[str]] = ["role", "from"],  # commonly "role"
-    message_field_content: str | list[str] | None = None,  # commonly "content"
+    message_field_content: Union[str, list[str]] = [
-    message_field_training: str | list[str] | None = None,  # commonly "weight"
+        "value",
+        "text",
+        "content",
+    ],  # commonly "content"
+    message_field_training: Union[str, list[str]] = [
+        "train",
+        "weight",
+    ],  # commonly "weight"
 ):
     """Builds a transform that takes a row from the dataset and converts it to a Chat
 
@@ -33,12 +39,6 @@ def chat_message_transform_builder(
             A function that takes a list of conversations and returns a list of messages.
     """
 
-    if message_field_training is None:
-        message_field_training = ["train", "weight"]
-    if message_field_content is None:
-        message_field_content = ["value", "text", "content"]
-    if message_field_role is None:
-        message_field_role = ["role", "from"]
     message_field_role = (
         [message_field_role]
         if isinstance(message_field_role, str)

src/axolotl/core/trainers/__init__.py

@@ -1,9 +1,11 @@
 """Init for axolotl.core.trainers"""
 
+# pylint: disable=unused-import
 # flake8: noqa
 
 from .base import AxolotlTrainer
 from .dpo.trainer import AxolotlDPOTrainer
+from .grpo.trainer import AxolotlGRPOSequenceParallelTrainer, AxolotlGRPOTrainer
 from .mamba import AxolotlMambaTrainer
 from .trl import (
     AxolotlCPOTrainer,

src/axolotl/core/trainers/base.py

@@ -1,5 +1,7 @@
 """Module for customized trainers"""
 
+# pylint: disable=too-many-lines
+
 from __future__ import annotations
 
 import os
@@ -42,20 +44,12 @@ from axolotl.core.trainers.utils import (
 )
 from axolotl.utils import get_not_null
 from axolotl.utils.bench import get_gpu_memory_usage
-from axolotl.utils.dict import DictDefault
 from axolotl.utils.distributed import is_main_process
 from axolotl.utils.logging import get_logger
 from axolotl.utils.samplers import MultipackBatchSampler, get_dataset_lengths
 
 LOG = get_logger(__name__)
 
-REDUCTION_FNS = {
-    "mean": torch.mean,
-    "min": torch.min,
-    "max": torch.max,
-    "sum": torch.sum,
-}
-
 
 class AxolotlTrainer(
     PackingMixin,
@@ -71,15 +65,6 @@ class AxolotlTrainer(
 
     args = None  # type: "AxolotlTrainingArguments"  # type: ignore[name-defined]
     tag_names = ["axolotl"]
-    _axolotl_cfg: DictDefault | None = None
-
-    @property
-    def axolotl_cfg(self):
-        return self._axolotl_cfg
-
-    @axolotl_cfg.setter
-    def axolotl_cfg(self, cfg):
-        self._axolotl_cfg = cfg
 
     def __init__(
         self,
@@ -95,10 +80,9 @@ class AxolotlTrainer(
         self._signature_columns = None  # workaround for pylint
 
         super().__init__(*_args, **kwargs)
+
         self.train_data_collator = self.data_collator
-        self._stored_metrics = defaultdict(
+        self._stored_metrics = defaultdict(lambda: defaultdict(list))
-            lambda: defaultdict(lambda: {"values": [], "reduction": "mean"})
-        )
         if self.args.orpo_alpha:
             self.loss_fct = torch.nn.CrossEntropyLoss(reduction="none")
 
@@ -301,9 +285,9 @@ class AxolotlTrainer(
         # fmt: off
         if dataloader_key is not None and self.args.dataloader_persistent_workers:
             if hasattr(self, "_eval_dataloaders"):
-                self._eval_dataloaders[dataloader_key] = dataloader  # type: ignore
+                self._eval_dataloaders[dataloader_key] = dataloader  # type: ignore  # pylint: disable=access-member-before-definition
             else:
-                self._eval_dataloaders = {dataloader_key: dataloader}
+                self._eval_dataloaders = {dataloader_key: dataloader}  # pylint: disable=attribute-defined-outside-init
         # fmt: on
 
         return self.accelerator.prepare(dataloader)
@@ -345,17 +329,6 @@ class AxolotlTrainer(
         #     outputs = model(**inputs)
         #     loss = trainer_weighted_loss(outputs, labels, shift_labels=True)
         #     return (loss, outputs) if return_outputs else loss
-
-        # track number of tokens for tokens per second calculation
-        if self.args.include_tkps:
-            inputs_key = "labels" if "labels" in inputs else "input_ids"
-            if hasattr(self.state, "num_tokens"):
-                self.state.num_tokens = (
-                    self.state.num_tokens + (inputs[inputs_key] != -100).sum().cpu()
-                )
-            else:
-                self.state.num_tokens = (inputs[inputs_key] != -100).sum().cpu()
-
         if self.args.orpo_alpha:
             return self.orpo_compute_loss(
                 model,
@@ -371,11 +344,6 @@ class AxolotlTrainer(
             num_items_in_batch=num_items_in_batch,
         )
 
-    @override
-    def evaluate(self, *args, **kwargs):
-        LOG.info("Running evaluation step...")
-        return super().evaluate(*args, **kwargs)
-
     @staticmethod
     def orpo_concatenate_inputs(inputs, label_pad_token=-100, pad_token=0, device=None):
         concatenated_batch = {}
@@ -475,7 +443,7 @@ class AxolotlTrainer(
         model,
         inputs,
         return_outputs=False,
-        num_items_in_batch=None,
+        num_items_in_batch=None,  # pylint: disable=unused-argument
     ):
         concat_inputs = AxolotlTrainer.orpo_concatenate_inputs(
             inputs,
@@ -556,10 +524,15 @@ class AxolotlTrainer(
         accelerator_config = self.args.accelerator_config.to_dict()
         use_configured_state = accelerator_config.get("use_configured_state", False)
         if not use_configured_state:
-            AcceleratorState._reset_state(reset_partial_state=True)
+            AcceleratorState._reset_state(  # pylint: disable=protected-access
+                reset_partial_state=True
+            )
 
         super().create_accelerator_and_postprocess()
 
+        # now we need to put parallelism_config back on the PartialState since we rely on that info in other places
+        # PartialState().parallelism_config = self.accelerator.state.parallelism_config
+
         if self.is_fsdp_enabled:
             if (
                 "limit_all_gathers" in self.args.fsdp_config
@@ -567,6 +540,7 @@ class AxolotlTrainer(
             ):
                 self.accelerator.state.fsdp_plugin.limit_all_gathers = True
 
+    # pylint: disable=unused-argument
     def additional_accelerator_args(
         self, fp8: bool = False, enable_fsdp_float8_all_gather: bool = False, **kwargs
     ) -> dict[str, Any]:
@@ -599,61 +573,29 @@ class AxolotlTrainer(
         """
         # logs either has 'loss' or 'eval_loss'
         train_eval = "train" if "loss" in logs else "eval"
-
+        # Add averaged stored metrics to logs
-        for key, metric_data in self._stored_metrics[train_eval].items():
+        for key, metrics in self._stored_metrics[train_eval].items():
-            values = torch.tensor(metric_data["values"])  # type: ignore[arg-type]
+            logs[key] = torch.tensor(metrics).mean().item()
-            reduction_type = metric_data["reduction"]
-
-            fn = REDUCTION_FNS.get(reduction_type)
-            if fn is None:
-                raise NotImplementedError(
-                    "Metric reduction must be one of [mean, min, max, sum]"
-                )
-            logs[key] = round(fn(values).item(), 4)
 
         if is_main_process():
             # Add memory usage
             try:
                 active, allocated, reserved = get_gpu_memory_usage()
-                logs["memory/max_active (GiB)"] = round(active, 2)
+                logs["memory/max_mem_active(gib)"] = round(active, 2)
-                logs["memory/max_allocated (GiB)"] = round(allocated, 2)
+                logs["memory/max_mem_allocated(gib)"] = round(allocated, 2)
-                logs["memory/device_reserved (GiB)"] = round(reserved, 2)
+                logs["memory/device_mem_reserved(gib)"] = round(reserved, 2)
             except (ValueError, TypeError, FileNotFoundError):
                 pass
 
-        if self.args.include_tkps and train_eval == "train":
-            # each rank will log its own tokens per second
-            # for logging_steps > 1 we obtain a moving average of this metric
-            logs["tokens_per_second_per_gpu"] = round(
-                self.state.last_tokens_per_second.item() / self.args.logging_steps, 2
-            )
-
         del self._stored_metrics[train_eval]
 
         return super().log(logs, start_time)
 
     def store_metrics(
-        self,
+        self, metrics: dict[str, float], train_eval: Literal["train", "eval"] = "train"
-        metrics: dict[str, float] | dict[str, tuple[int | float, str]],
-        train_eval: Literal["train", "eval"] = "train",
-        reduction: Literal["mean", "min", "max", "sum"] = "mean",
     ) -> None:
-        """
-        Store metrics with specified reduction type.
-
-        Args:
-            metrics: Dictionary of metric names to values, or metric names to (value,
-                reduction_type) tuples.
-            train_eval: Whether this is for training or evaluation.
-        """
         for key, value in metrics.items():
-            if isinstance(value, tuple):
+            self._stored_metrics[train_eval][key].append(value)
-                value, _reduction = value  # type: ignore[assignment]
-            else:
-                value, _reduction = value, reduction
-
-            self._stored_metrics[train_eval][key]["values"].append(value)
-            self._stored_metrics[train_eval][key]["reduction"] = _reduction
 
     def _save_checkpoint(self, model, trial, **kwargs):
         # make sure the checkpoint dir exists, since trainer is flakey
@@ -720,11 +662,6 @@ class AxolotlTrainer(
                 LOG.info(
                     "Saving Trainer.data_collator.tokenizer by default as Trainer.processing_class is `None`"
                 )
-                save_jinja_files = True
+                self.data_collator.tokenizer.save_pretrained(output_dir)
-                if self.axolotl_cfg:
-                    save_jinja_files = self.axolotl_cfg.tokenizer_save_jinja_files
-                self.data_collator.tokenizer.save_pretrained(
-                    output_dir, save_jinja_files=save_jinja_files
-                )
             # Good practice: save your training arguments together with the trained model
             torch.save(self.args, os.path.join(output_dir, TRAINING_ARGS_NAME))

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

@@ -101,11 +101,11 @@ class AxolotlDPOTrainer(
     ) -> dict[str, torch.Tensor]:
         if self.args.dpo_norm_loss:
             # fmt: off
-            loss_type: str = self.loss_type  # type: ignore[has-type]
+            loss_type: str = self.loss_type  # type: ignore[has-type]  # pylint: disable=access-member-before-definition
             # fmt: on
             # concatenated_forward handles avg token logprob for ipo case already
-            self.loss_type = "ipo"
+            self.loss_type = "ipo"  # pylint: disable=attribute-defined-outside-init
             res = super().concatenated_forward(model, batch, is_ref_model=is_ref_model)
-            self.loss_type = loss_type
+            self.loss_type = loss_type  # pylint: disable=attribute-defined-outside-init
             return res
         return super().concatenated_forward(model, batch, is_ref_model=is_ref_model)

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

@@ -128,7 +128,9 @@ class GRPOStrategy:
         return grpo_args_kwargs
 
     @classmethod
-    def set_trainer_args(cls, cfg: DictDefault) -> list[Any]:
+    def set_trainer_args(
+        cls, cfg: DictDefault
+    ) -> list[Any]:  # pylint: disable=unused-argument
         trainer_args = []
         if cfg.trl and cfg.trl.reward_funcs:
             reward_funcs = []
@@ -149,7 +151,7 @@ class GRPOStrategy:
         return trainer_kwargs
 
     @classmethod
-    def get_collator(cls, *args, **kwargs):
+    def get_collator(cls, *args, **kwargs):  # pylint: disable=unused-argument
         # No data collation is needed in GRPO, handled by trl's trainer __init__
         return None
 

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

@@ -1,5 +1,7 @@
 """Axolotl GRPO trainers (with and without sequence parallelism handling)"""
 
+# pylint: disable=too-many-lines,duplicate-code,protected-access,no-member
+
 import warnings
 from functools import partial
 from typing import Any
@@ -50,6 +52,7 @@ from axolotl.core.trainers.mixins.optimizer import OptimizerInitMixin, Optimizer
 from axolotl.monkeypatch.ring_attn import get_ring_attn_group
 
 if is_peft_available():
+    # pylint: disable=unused-import
     from peft import PeftConfig
 
 
@@ -250,7 +253,7 @@ class AxolotlGRPOSequenceParallelTrainer(AxolotlGRPOTrainer):
     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
@@ -263,7 +266,7 @@ class AxolotlGRPOSequenceParallelTrainer(AxolotlGRPOTrainer):
                     train_dataset, description="training"
                 )
         else:
-            self.data_collator = self._get_collator_with_removed_columns(
+            self.data_collator = self._get_collator_with_removed_columns(  # pylint: disable=attribute-defined-outside-init
                 data_collator,
                 description="training",
             )
@@ -305,10 +308,10 @@ class AxolotlGRPOSequenceParallelTrainer(AxolotlGRPOTrainer):
         # 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
@@ -330,9 +333,8 @@ class AxolotlGRPOSequenceParallelTrainer(AxolotlGRPOTrainer):
                         # 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
-                                group_leader_rank + 1
+                            * len(prompts_text) : (group_leader_rank + 1)
-                            )
                             * len(prompts_text) : self.num_generations
                         ]
 
@@ -483,7 +485,7 @@ class AxolotlGRPOSequenceParallelTrainer(AxolotlGRPOTrainer):
         )
         if is_conversational(inputs[0]):
             completions = []
-            for prompt, completion in zip(prompts, completions_text, strict=False):
+            for prompt, completion in zip(prompts, completions_text):
                 bootstrap = (
                     prompt.pop()["content"] if prompt[-1]["role"] == "assistant" else ""
                 )
@@ -501,7 +503,6 @@ class AxolotlGRPOSequenceParallelTrainer(AxolotlGRPOTrainer):
                 self.reward_funcs,
                 self.reward_processing_classes,
                 self.reward_func_names,
-                strict=False,
             )
         ):
             with profiling_context(self, reward_func_name):
@@ -510,17 +511,14 @@ class AxolotlGRPOSequenceParallelTrainer(AxolotlGRPOTrainer):
                 ):  # Module instead of PretrainedModel for compat with compiled models
                     if is_conversational(inputs[0]):
                         messages = [
-                            {"messages": p + c}
+                            {"messages": p + c} for p, c in zip(prompts, completions)
-                            for p, c in zip(prompts, completions, strict=False)
                         ]
                         texts = [
                             apply_chat_template(x, reward_processing_class)["text"]
                             for x in messages
                         ]
                     else:
-                        texts = [
+                        texts = [p + c for p, c in zip(prompts, completions)]
-                            p + c for p, c in zip(prompts, completions, strict=False)
-                        ]
                     reward_inputs = reward_processing_class(
                         text=texts,
                         return_tensors="pt",
@@ -566,8 +564,7 @@ class AxolotlGRPOSequenceParallelTrainer(AxolotlGRPOTrainer):
             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.",
+                "Please ensure that at least one reward function returns a valid reward."
-                stacklevel=2,
             )
 
         # Gather the reward per function: this part is crucial, because the rewards are normalized per group and the

src/axolotl/core/trainers/mamba.py

@@ -5,6 +5,7 @@ import torch
 from axolotl.core.trainers.base import AxolotlTrainer
 
 
+# pylint: disable=too-many-ancestors
 class AxolotlMambaTrainer(AxolotlTrainer):
     """Mamba specific trainer to handle loss calculation"""
 
@@ -14,8 +15,8 @@ class AxolotlMambaTrainer(AxolotlTrainer):
         self,
         model,
         inputs,
-        return_outputs=False,
+        return_outputs=False,  # pylint: disable=unused-argument
-        num_items_in_batch=None,
+        num_items_in_batch=None,  # pylint: disable=unused-argument
     ):
         input_ids = inputs.pop("input_ids")
         lm_logits = model(input_ids).logits

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

@@ -1,5 +1,6 @@
 """Init for axolotl.core.trainers.mixins"""
 
+# pylint: disable=unused-import
 # flake8: noqa
 
 from .activation_checkpointing import ActivationOffloadingMixin

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

@@ -92,7 +92,7 @@ def get_lora_act_offloading_ctx_manager(
         `contextlib.ContextDecorator`:
             Activation offloading context manager for the model.
     """
-
+    # pylint: disable=unnecessary-dunder-call
     activations_handling_ctx = OffloadActivations(
         use_pin_memory=use_pin_memory,
         use_streams=use_streams,

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

@@ -26,6 +26,7 @@ class DistributedParallelMixin(Trainer):
             self.accelerator.distributed_type == "FSDP"
             and self.accelerator.state.fsdp_plugin is None
         ):
+            # pylint: disable=protected-access
             # handle Context Parallelism without FSDP
             self.accelerator.state.distributed_type = "MULTI_GPU"
             self.accelerator.state._shared_state["distributed_type"] = "MULTI_GPU"

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

@@ -70,11 +70,11 @@ class OptimizerMixin(Trainer):
                 }
             )
         if params["embeddings"]:
-            lr = optimizer_kwargs["lr"]
+            lr = optimizer_kwargs["lr"]  # pylint: disable=invalid-name
             if self.args.embedding_lr_scale:
-                lr *= self.args.embedding_lr_scale
+                lr *= self.args.embedding_lr_scale  # pylint: disable=invalid-name
             elif self.args.embedding_lr:
-                lr = self.args.embedding_lr
+                lr = self.args.embedding_lr  # pylint: disable=invalid-name
             optimizer_grouped_parameters.append(
                 {
                     "params": list(params["embeddings"].values()),
@@ -143,7 +143,7 @@ class OptimizerMixin(Trainer):
                 loraplus_lr_embedding = getattr(
                     self.args, "loraplus_lr_embedding", 1e-6
                 )
-                self.optimizer = create_loraplus_optimizer(
+                self.optimizer = create_loraplus_optimizer(  # pylint: disable=attribute-defined-outside-init
                     opt_model,
                     optimizer_cls,
                     loraplus_lr_ratio=loraplus_lr_ratio,
@@ -185,15 +185,17 @@ class OptimizerMixin(Trainer):
                                 p.data_ptr(): p.numel() for p in module.parameters()
                             }.values()
                         )
-                        LOG.info(f"skipped {module}: {skipped / 2**20}M params")
+                        LOG.info(f"skipped {module}: {skipped/2**20}M params")
                         manager.register_module_override(
                             module, "weight", {"optim_bits": 32}
                         )
                         LOG.debug(f"bitsandbytes: will optimize {module} in fp32")
-                LOG.info(f"skipped: {skipped / 2**20}M params")
+                LOG.info(f"skipped: {skipped/2**20}M params")
 
         if is_sagemaker_mp_enabled():
-            self.optimizer = smp.DistributedOptimizer(self.optimizer)
+            self.optimizer = smp.DistributedOptimizer(  # pylint: disable=attribute-defined-outside-init
+                self.optimizer
+            )
 
         return self.optimizer