[model] fix vlm utils (#8388)

Co-authored-by: viyer <vivek_iyer2@apple.com>

.dockerignore

@@ -3,11 +3,13 @@
 .github
 .venv
 cache
-data
+docker
+saves
 hf_cache
+ms_cache
+om_cache
+shared_data
 output
-examples
 .dockerignore
 .gitattributes
 .gitignore
-Dockerfile

.env.local

@@ -0,0 +1,42 @@
+# Note: actually we do not support .env, just for reference
+# api
+API_HOST=
+API_PORT=
+API_KEY=
+API_MODEL_NAME=
+API_VERBOSE=
+FASTAPI_ROOT_PATH=
+MAX_CONCURRENT=
+# general
+DISABLE_VERSION_CHECK=
+FORCE_CHECK_IMPORTS=
+ALLOW_EXTRA_ARGS=
+LLAMAFACTORY_VERBOSITY=
+USE_MODELSCOPE_HUB=
+USE_OPENMIND_HUB=
+USE_RAY=
+RECORD_VRAM=
+OPTIM_TORCH=
+NPU_JIT_COMPILE=
+# torchrun
+FORCE_TORCHRUN=
+MASTER_ADDR=
+MASTER_PORT=
+NNODES=
+NODE_RANK=
+NPROC_PER_NODE=
+# wandb
+WANDB_DISABLED=
+WANDB_PROJECT=
+WANDB_API_KEY=
+# gradio ui
+GRADIO_SHARE=
+GRADIO_SERVER_NAME=
+GRADIO_SERVER_PORT=
+GRADIO_ROOT_PATH=
+GRADIO_IPV6=
+# setup
+ENABLE_SHORT_CONSOLE=
+# reserved (do not use)
+LLAMABOARD_ENABLED=
+LLAMABOARD_WORKDIR=

.github/CONTRIBUTING.md

@@ -19,3 +19,49 @@ There are several ways you can contribute to LLaMA Factory:
 ### Style guide
 
 LLaMA Factory follows the [Google Python Style Guide](https://google.github.io/styleguide/pyguide.html), check it for details.
+
+### Create a Pull Request
+
+1. Fork the [repository](https://github.com/hiyouga/LLaMA-Factory) by clicking on the [Fork](https://github.com/hiyouga/LLaMA-Factory/fork) button on the repository's page. This creates a copy of the code under your GitHub user account.
+
+2. Clone your fork to your local disk, and add the base repository as a remote:
+
+```bash
+git clone git@github.com:[username]/LLaMA-Factory.git
+cd LLaMA-Factory
+git remote add upstream https://github.com/hiyouga/LLaMA-Factory.git
+```
+
+3. Create a new branch to hold your development changes:
+
+```bash
+git checkout -b dev_your_branch
+```
+
+4. Set up a development environment by running the following command in a virtual environment:
+
+```bash
+pip install -e ".[dev]"
+```
+
+If LLaMA Factory was already installed in the virtual environment, remove it with `pip uninstall llamafactory` before reinstalling it in editable mode with the -e flag.
+
+5. Check code before commit:
+
+```bash
+make commit
+make style && make quality
+make test
+```
+
+6. Submit changes:
+
+```bash
+git add .
+git commit -m "commit message"
+git fetch upstream
+git rebase upstream/main
+git push -u origin dev_your_branch
+```
+
+7. Create a merge request from your branch `dev_your_branch` at [origin repo](https://github.com/hiyouga/LLaMA-Factory).

.github/ISSUE_TEMPLATE/1-bug-report.yml

@@ -0,0 +1,61 @@
+name: "\U0001F41B Bug / help"
+description: Create a report to help us improve the LLaMA Factory
+labels: ["bug", "pending"]
+body:
+  - type: markdown
+    attributes:
+      value: |
+        Issues included in **[FAQs](https://github.com/hiyouga/LLaMA-Factory/issues/4614)** or those with **insufficient** information may be closed without a response.
+        已经包含在 **[常见问题](https://github.com/hiyouga/LLaMA-Factory/issues/4614)** 内或提供信息**不完整**的 issues 可能不会被回复。
+
+  - type: markdown
+    attributes:
+      value: |
+        Please do not create issues that are not related to framework bugs under this category, use **[Discussions](https://github.com/hiyouga/LLaMA-Factory/discussions/categories/q-a)** instead.
+        请勿在此分类下创建和框架 bug 无关的 issues，训练问题求助请使用 **[讨论区](https://github.com/hiyouga/LLaMA-Factory/discussions/categories/q-a)**。
+
+  - type: checkboxes
+    id: reminder
+    attributes:
+      label: Reminder
+      description: |
+        Please ensure you have read the above rules carefully and searched the existing issues (including FAQs).
+        请确保您已经认真阅读了上述规则并且搜索过现有的 issues（包括常见问题）。
+
+      options:
+        - label: I have read the above rules and searched the existing issues.
+          required: true
+
+  - type: textarea
+    id: system-info
+    validations:
+      required: true
+    attributes:
+      label: System Info
+      description: |
+        Please share your system info with us. You can run the command **llamafactory-cli env** and copy-paste its output below.
+        请提供您的系统信息。您可以在命令行运行 **llamafactory-cli env** 并将其输出复制到该文本框中。
+
+      placeholder: llamafactory version, platform, python version, ...
+
+  - type: textarea
+    id: reproduction
+    validations:
+      required: true
+    attributes:
+      label: Reproduction
+      description: |
+        Please provide entry arguments, error messages and stack traces that reproduces the problem.
+        请提供入口参数，错误日志以及异常堆栈以便于我们复现问题。
+
+      value: |
+        ```text
+        Put your message here.
+        ```
+
+  - type: textarea
+    id: others
+    validations:
+      required: false
+    attributes:
+      label: Others

.github/ISSUE_TEMPLATE/2-feature-request.yml

@@ -0,0 +1,41 @@
+name: "\U0001F680 Feature request"
+description: Submit a request for a new feature
+labels: ["enhancement", "pending"]
+body:
+  - type: markdown
+    attributes:
+      value: |
+        Please do not create issues that are not related to new features under this category.
+        请勿在此分类下创建和新特性无关的 issues。
+
+  - type: checkboxes
+    id: reminder
+    attributes:
+      label: Reminder
+      description: |
+        Please ensure you have read the above rules carefully and searched the existing issues.
+        请确保您已经认真阅读了上述规则并且搜索过现有的 issues。
+
+      options:
+        - label: I have read the above rules and searched the existing issues.
+          required: true
+
+  - type: textarea
+    id: description
+    validations:
+      required: true
+    attributes:
+      label: Description
+      description: |
+        A clear and concise description of the feature proposal.
+        请详细描述您希望加入的新功能特性。
+
+  - type: textarea
+    id: contribution
+    validations:
+      required: false
+    attributes:
+      label: Pull Request
+      description: |
+        Have you already created the relevant PR and submitted the code?
+        您是否已经创建了相关 PR 并提交了代码？

.github/ISSUE_TEMPLATE/bug-report.yml

@@ -1,60 +0,0 @@
-name: "\U0001F41B Bug / Help"
-description: Create a report to help us improve the LLaMA Factory
-body:
-  - type: checkboxes
-    id: reminder
-    attributes:
-      label: Reminder
-      description: |
-        Please ensure you have read the README carefully and searched the existing issues.
-        请确保您已经认真阅读了 README 并且搜索过现有的 Issue。
-
-      options:
-        - label: I have read the README and searched the existing issues.
-          required: true
-
-  - type: textarea
-    id: system-info
-    validations:
-      required: true
-    attributes:
-      label: System Info
-      description: |
-        Please share your system info with us. You can run the command **llamafactory-cli env** and copy-paste its output below.
-        请提供您的系统信息。您可以在命令行运行 **llamafactory-cli env** 并将其输出复制到该文本框中。
-
-      placeholder: llamafactory version, platform, python version, ...
-
-  - type: textarea
-    id: reproduction
-    validations:
-      required: true
-    attributes:
-      label: Reproduction
-      description: |
-        Please provide code snippets, error messages and stack traces that reproduces the problem.
-        请提供运行参数，错误信息以及异常堆栈以便于我们复现该问题。
-        Remember to use Markdown tags to correctly format your code.
-        请合理使用 Markdown 标签来格式化您的文本。
-
-      placeholder: |
-        ```bash
-        llamafactory-cli train ...
-        ```
-
-  - type: textarea
-    id: expected-behavior
-    validations:
-      required: false
-    attributes:
-      label: Expected behavior
-      description: |
-        Please provide a clear and concise description of what you would expect to happen.
-        请提供您原本的目的，即这段代码的期望行为。
-
-  - type: textarea
-    id: others
-    validations:
-      required: false
-    attributes:
-      label: Others

.github/ISSUE_TEMPLATE/config.yml

@@ -0,0 +1 @@
+blank_issues_enabled: false

.github/workflows/docker.yml

@@ -0,0 +1,66 @@
+name: docker
+
+on:
+  workflow_dispatch:
+  push:
+    branches:
+      - "main"
+    paths:
+      - "**/*.py"
+      - "requirements.txt"
+      - "docker/**"
+      - ".github/workflows/*.yml"
+  pull_request:
+    branches:
+      - "main"
+    paths:
+      - "**/*.py"
+      - "requirements.txt"
+      - "docker/**"
+      - ".github/workflows/*.yml"
+
+jobs:
+  build:
+    runs-on: ubuntu-latest
+
+    concurrency:
+      group: ${{ github.workflow }}-${{ github.ref }}
+      cancel-in-progress: ${{ github.ref != 'refs/heads/main' }}
+
+    environment:
+      name: docker
+      url: https://hub.docker.com/r/hiyouga/llamafactory
+
+    steps:
+      - name: Free up disk space
+        run: |
+          df -h
+          sudo rm -rf /usr/share/dotnet
+          sudo rm -rf /opt/ghc
+          sudo rm -rf /opt/hostedtoolcache
+          df -h
+
+      - name: Checkout
+        uses: actions/checkout@v4
+
+      - name: Set up Docker Buildx
+        uses: docker/setup-buildx-action@v3
+
+      - name: Login to Docker Hub
+        if: github.event_name != 'pull_request'
+        uses: docker/login-action@v3
+        with:
+          username: ${{ vars.DOCKERHUB_USERNAME }}
+          password: ${{ secrets.DOCKERHUB_TOKEN }}
+
+      - name: Build and push Docker image
+        uses: docker/build-push-action@v6
+        with:
+          context: .
+          file: ./docker/docker-cuda/Dockerfile
+          build-args: |
+            EXTRAS=metrics,deepspeed,liger-kernel
+          push: ${{ github.event_name != 'pull_request' }}
+          tags: docker.io/hiyouga/llamafactory:latest
+          cache-from: type=gha
+          cache-to: type=gha,mode=max

.github/workflows/label_issue.yml

@@ -9,9 +9,24 @@ jobs:
   label_issue:
     runs-on: ubuntu-latest
 
+    permissions:
+      issues: write
+
     steps:
       - env:
           GH_TOKEN: ${{ secrets.GITHUB_TOKEN }}
           ISSUE_URL: ${{ github.event.issue.html_url }}
+          ISSUE_TITLE: ${{ github.event.issue.title }}
         run: |
-          gh issue edit $ISSUE_URL --add-label "pending"
+          LABEL=""
+          NPU_KEYWORDS=(npu huawei ascend 华为 昇腾)
+          ISSUE_TITLE_LOWER=$(echo $ISSUE_TITLE | tr '[:upper:]' '[:lower:]')
+          for KEYWORD in ${NPU_KEYWORDS[@]}; do
+            if [[ $ISSUE_TITLE_LOWER == *$KEYWORD* ]] && [[ $ISSUE_TITLE_LOWER != *input* ]]; then
+              LABEL="npu"
+              break
+            fi
+          done
+          if [ -n "$LABEL" ]; then
+            gh issue edit $ISSUE_URL --add-label $LABEL
+          fi

.github/workflows/publish.yml

@@ -1,6 +1,7 @@
 name: publish
 
 on:
+  workflow_dispatch:
   release:
     types:
       - published
@@ -25,16 +26,11 @@ jobs:
       - name: Set up Python
         uses: actions/setup-python@v5
         with:
-          python-version: "3.8"
-      
-      - name: Install dependencies
-        run: |
-          python -m pip install --upgrade pip
-          python -m pip install build
+          python-version: "3.9"
 
       - name: Build package
         run: |
-          python -m build
+          make build
 
       - name: Publish package
         uses: pypa/gh-action-pypi-publish@release/v1

.github/workflows/tests.yml

@@ -1,16 +1,17 @@
 name: tests
 
 on:
+  workflow_dispatch:
   push:
     branches:
-      - main
+      - "main"
     paths:
       - "**.py"
       - "requirements.txt"
       - ".github/workflows/*.yml"
   pull_request:
     branches:
-      - main
+      - "main"
     paths:
       - "**.py"
       - "requirements.txt"
@@ -18,7 +19,40 @@ on:
 
 jobs:
   tests:
-    runs-on: ubuntu-latest
+    strategy:
+      fail-fast: false
+      matrix:
+        python:
+          - "3.9"
+          - "3.10"
+          - "3.11"
+          - "3.12"
+        os:
+          - "ubuntu-latest"
+          - "windows-latest"
+          - "macos-13"
+        transformers:
+          - null
+        include:  # test backward compatibility
+          - python: "3.9"
+            os: "ubuntu-latest"
+            transformers: "4.45.0"
+          - python: "3.9"
+            os: "ubuntu-latest"
+            transformers: "4.49.0"
+          - python: "3.9"
+            os: "ubuntu-latest"
+            transformers: "4.51.0"
+
+    runs-on: ${{ matrix.os }}
+
+    concurrency:
+      group: ${{ github.workflow }}-${{ github.ref }}-${{ matrix.os }}-${{ matrix.python }}-${{ matrix.transformers }}
+      cancel-in-progress: ${{ github.ref != 'refs/heads/main' }}
+
+    env:
+      HF_TOKEN: ${{ secrets.HF_TOKEN }}
+      OS_NAME: ${{ matrix.os }}
 
     steps:
       - name: Checkout
@@ -27,19 +61,42 @@ jobs:
       - name: Set up Python
         uses: actions/setup-python@v5
         with:
-          python-version: "3.8"
+          python-version: ${{ matrix.python }}
           cache: "pip"
-          cache-dependency-path: "setup.py"
+          cache-dependency-path: "**/requirements*.txt"
 
       - name: Install dependencies
         run: |
           python -m pip install --upgrade pip
-          python -m pip install .[torch,dev]
+          python -m pip install ".[torch,dev]"
+
+      - name: Install transformers
+        if: ${{ matrix.transformers }}
+        run: |
+          python -m pip install "transformers==${{ matrix.transformers }}"
+
+      - name: Cache files
+        id: hf-hub-cache
+        uses: actions/cache@v4
+        with:
+          path: ${{ runner.temp }}/huggingface
+          key: huggingface-${{ matrix.os }}-${{ matrix.python }}-${{ matrix.transformers }}-${{ hashFiles('tests/version.txt') }}
 
       - name: Check quality
         run: |
           make style && make quality
 
+      - name: Check license
+        run: |
+          make license
+
+      - name: Check build
+        run: |
+          make build
+
       - name: Test with pytest
         run: |
           make test
+        env:
+          HF_HOME: ${{ runner.temp }}/huggingface
+          HF_HUB_OFFLINE: "${{ steps.hf-hub-cache.outputs.cache-hit == 'true' && '1' || '0' }}"

.gitignore

@@ -159,7 +159,21 @@ cython_debug/
 #  option (not recommended) you can uncomment the following to ignore the entire idea folder.
 .idea/
 
+# vscode
+.vscode/
+
+# uv
+uv.lock
+
 # custom .gitignore
-user.config
-saves/
+hf_cache/
+ms_cache/
+om_cache/
 cache/
+config/
+saves/
+output/
+wandb/
+swanlog/
+generated_predictions.jsonl
+predictions_score.json

.pre-commit-config.yaml

@@ -0,0 +1,28 @@
+repos:
+-   repo: https://github.com/pre-commit/pre-commit-hooks
+    rev: v5.0.0
+    hooks:
+    -   id: check-ast
+    -   id: check-added-large-files
+        args: ['--maxkb=25000']
+    -   id: check-merge-conflict
+    -   id: check-yaml
+    -   id: debug-statements
+    -   id: end-of-file-fixer
+    -   id: trailing-whitespace
+        args: [--markdown-linebreak-ext=md]
+    -   id: no-commit-to-branch
+        args: ['--branch', 'main']
+
+-   repo: https://github.com/asottile/pyupgrade
+    rev: v3.17.0
+    hooks:
+    -   id: pyupgrade
+        args: [--py38-plus]
+
+-   repo: https://github.com/astral-sh/ruff-pre-commit
+    rev: v0.6.9
+    hooks:
+    -   id: ruff
+        args: [--fix]
+    -   id: ruff-format

CITATION.cff

@@ -12,12 +12,16 @@ authors:
   given-names: "Yanhan"
 - family-names: "Luo"
   given-names: "Zheyan"
+- family-names: "Feng"
+  given-names: "Zhangchi"
 - family-names: "Ma"
   given-names: "Yongqiang"
 title: "LlamaFactory: Unified Efficient Fine-Tuning of 100+ Language Models"
 url: "https://arxiv.org/abs/2403.13372"
 preferred-citation:
-  type: article
+  type: conference-paper
+  conference:
+    name: "Proceedings of the 62nd Annual Meeting of the Association for Computational Linguistics (Volume 3: System Demonstrations)"
   authors:
     - family-names: "Zheng"
       given-names: "Yaowei"
@@ -29,9 +33,12 @@ preferred-citation:
       given-names: "Yanhan"
     - family-names: "Luo"
       given-names: "Zheyan"
+    - family-names: "Feng"
+      given-names: "Zhangchi"
     - family-names: "Ma"
       given-names: "Yongqiang"
-  journal: "arXiv preprint arXiv:2403.13372"
   title: "LlamaFactory: Unified Efficient Fine-Tuning of 100+ Language Models"
   url: "https://arxiv.org/abs/2403.13372"
   year: 2024
+  publisher: "Association for Computational Linguistics"
+  address: "Bangkok, Thailand"

Dockerfile

@@ -1,47 +0,0 @@
-# Use the NVIDIA official image with PyTorch 2.3.0
-# https://docs.nvidia.com/deeplearning/frameworks/pytorch-release-notes/rel-24-02.html
-FROM nvcr.io/nvidia/pytorch:24.02-py3
-
-# Define installation arguments
-ARG INSTALL_BNB=false
-ARG INSTALL_VLLM=false
-ARG INSTALL_DEEPSPEED=false
-ARG PIP_INDEX=https://pypi.org/simple
-
-# Set the working directory
-WORKDIR /app
-
-# Install the requirements
-COPY requirements.txt /app/
-RUN pip config set global.index-url $PIP_INDEX
-RUN python -m pip install --upgrade pip
-RUN python -m pip install -r requirements.txt
-
-# Copy the rest of the application into the image
-COPY . /app/
-
-# Install the LLaMA Factory
-RUN EXTRA_PACKAGES="metrics"; \
-    if [ "$INSTALL_BNB" = "true" ]; then \
-        EXTRA_PACKAGES="${EXTRA_PACKAGES},bitsandbytes"; \
-    fi; \
-    if [ "$INSTALL_VLLM" = "true" ]; then \
-        EXTRA_PACKAGES="${EXTRA_PACKAGES},vllm"; \
-    fi; \
-    if [ "$INSTALL_DEEPSPEED" = "true" ]; then \
-        EXTRA_PACKAGES="${EXTRA_PACKAGES},deepspeed"; \
-    fi; \
-    pip install -e .[$EXTRA_PACKAGES] && \
-    pip uninstall -y transformer-engine flash-attn
-
-# Set up volumes
-VOLUME [ "/root/.cache/huggingface/", "/app/data", "/app/output" ]
-
-# Expose port 7860 for the LLaMA Board
-EXPOSE 7860
-
-# Expose port 8000 for the API service
-EXPOSE 8000
-
-# Launch LLaMA Board
-CMD [ "llamafactory-cli", "webui" ]

Makefile

@@ -1,6 +1,16 @@
-.PHONY: quality style test
+.PHONY: build commit license quality style test
 
-check_dirs := scripts src tests
+check_dirs := scripts src tests setup.py
+
+build:
+	pip3 install build && python3 -m build
+
+commit:
+	pre-commit install
+	pre-commit run --all-files
+
+license:
+	python3 tests/check_license.py $(check_dirs)
 
 quality:
 	ruff check $(check_dirs)
@@ -11,4 +21,4 @@ style:
 	ruff format $(check_dirs)
 
 test:
-	CUDA_VISIBLE_DEVICES= pytest tests/
+	CUDA_VISIBLE_DEVICES= WANDB_DISABLED=true pytest -vv tests/

README.md

@@ -1,44 +1,85 @@
 ![# LLaMA Factory](assets/logo.png)
 
 [![GitHub Repo stars](https://img.shields.io/github/stars/hiyouga/LLaMA-Factory?style=social)](https://github.com/hiyouga/LLaMA-Factory/stargazers)
-[![GitHub Code License](https://img.shields.io/github/license/hiyouga/LLaMA-Factory)](LICENSE)
 [![GitHub last commit](https://img.shields.io/github/last-commit/hiyouga/LLaMA-Factory)](https://github.com/hiyouga/LLaMA-Factory/commits/main)
+[![GitHub contributors](https://img.shields.io/github/contributors/hiyouga/LLaMA-Factory?color=orange)](https://github.com/hiyouga/LLaMA-Factory/graphs/contributors)
+[![GitHub workflow](https://github.com/hiyouga/LLaMA-Factory/actions/workflows/tests.yml/badge.svg)](https://github.com/hiyouga/LLaMA-Factory/actions/workflows/tests.yml)
 [![PyPI](https://img.shields.io/pypi/v/llamafactory)](https://pypi.org/project/llamafactory/)
-[![Citation](https://img.shields.io/badge/citation-44-green)](#projects-using-llama-factory)
-[![GitHub pull request](https://img.shields.io/badge/PRs-welcome-blue)](https://github.com/hiyouga/LLaMA-Factory/pulls)
-[![Discord](https://dcbadge.vercel.app/api/server/rKfvV9r9FK?compact=true&style=flat)](https://discord.gg/rKfvV9r9FK)
+[![Citation](https://img.shields.io/badge/citation-614-green)](https://scholar.google.com/scholar?cites=12620864006390196564)
+[![Docker Pulls](https://img.shields.io/docker/pulls/hiyouga/llamafactory)](https://hub.docker.com/r/hiyouga/llamafactory/tags)
+
 [![Twitter](https://img.shields.io/twitter/follow/llamafactory_ai)](https://twitter.com/llamafactory_ai)
+[![Discord](https://dcbadge.vercel.app/api/server/rKfvV9r9FK?compact=true&style=flat)](https://discord.gg/rKfvV9r9FK)
+[![GitCode](https://gitcode.com/zhengyaowei/LLaMA-Factory/star/badge.svg)](https://gitcode.com/zhengyaowei/LLaMA-Factory)
+
 [![Open in Colab](https://colab.research.google.com/assets/colab-badge.svg)](https://colab.research.google.com/drive/1eRTPn37ltBbYsISy9Aw2NuI2Aq5CQrD9?usp=sharing)
 [![Open in DSW](https://gallery.pai-ml.com/assets/open-in-dsw.svg)](https://gallery.pai-ml.com/#/preview/deepLearning/nlp/llama_factory)
-[![Spaces](https://img.shields.io/badge/🤗-Open%20in%20Spaces-blue)](https://huggingface.co/spaces/hiyouga/LLaMA-Board)
-[![Studios](https://img.shields.io/badge/ModelScope-Open%20in%20Studios-blue)](https://modelscope.cn/studios/hiyouga/LLaMA-Board)
+[![Open in Alaya](assets/alaya_new.svg)](https://docs.alayanew.com/docs/documents/newActivities/llamafactory/?utm_source=LLaMA-Factory)
+[![Open in Spaces](https://img.shields.io/badge/🤗-Open%20in%20Spaces-blue)](https://huggingface.co/spaces/hiyouga/LLaMA-Board)
+[![Open in Studios](https://img.shields.io/badge/ModelScope-Open%20in%20Studios-blue)](https://modelscope.cn/studios/hiyouga/LLaMA-Board)
+[![Open in Novita](https://img.shields.io/badge/Novita-Deploy%20Template-blue)](https://novita.ai/templates-library/105981?sharer=88115474-394e-4bda-968e-b88e123d0c47)
 
-[![GitHub Tread](https://trendshift.io/api/badge/repositories/4535)](https://trendshift.io/repositories/4535)
+### Used by [Amazon](https://aws.amazon.com/cn/blogs/machine-learning/how-apoidea-group-enhances-visual-information-extraction-from-banking-documents-with-multimodal-models-using-llama-factory-on-amazon-sagemaker-hyperpod/), [NVIDIA](https://developer.nvidia.com/rtx/ai-toolkit), [Aliyun](https://help.aliyun.com/zh/pai/use-cases/fine-tune-a-llama-3-model-with-llama-factory), etc.
 
-👋 Join our [WeChat](assets/wechat.jpg).
+<div align="center" markdown="1">
+
+### Supporters ❤️
+
+<a href="https://warp.dev/llama-factory">
+    <img alt="Warp sponsorship" width="400" src="https://github.com/user-attachments/assets/ab8dd143-b0fd-4904-bdc5-dd7ecac94eae">
+</a>
+
+#### [Warp, the agentic terminal for developers](https://warp.dev/llama-factory)
+
+[Available for MacOS, Linux, & Windows](https://warp.dev/llama-factory)
+
+----
+
+### Easily fine-tune 100+ large language models with zero-code [CLI](#quickstart) and [Web UI](#fine-tuning-with-llama-board-gui-powered-by-gradio)
+
+![GitHub Trend](https://trendshift.io/api/badge/repositories/4535)
+
+</div>
+
+👋 Join our [WeChat group](assets/wechat.jpg), [NPU user group](assets/wechat_npu.jpg) or [Alaya NeW user group](assets/wechat_alaya.png).
 
 \[ English | [中文](README_zh.md) \]
 
 **Fine-tuning a large language model can be easy as...**
 
-https://github.com/hiyouga/LLaMA-Factory/assets/16256802/9840a653-7e9c-41c8-ae89-7ace5698baf6
+https://github.com/user-attachments/assets/3991a3a8-4276-4d30-9cab-4cb0c4b9b99e
 
 Choose your path:
 
-- **Colab**: https://colab.research.google.com/drive/1eRTPn37ltBbYsISy9Aw2NuI2Aq5CQrD9?usp=sharing
-- **PAI-DSW**: https://gallery.pai-ml.com/#/preview/deepLearning/nlp/llama_factory
+- **Documentation**: https://llamafactory.readthedocs.io/en/latest/
+- **Colab (free)**: https://colab.research.google.com/drive/1eRTPn37ltBbYsISy9Aw2NuI2Aq5CQrD9?usp=sharing
 - **Local machine**: Please refer to [usage](#getting-started)
+- **PAI-DSW (free trial)**: https://gallery.pai-ml.com/#/preview/deepLearning/nlp/llama_factory
+- **Alaya NeW (cloud GPU deal)**: https://docs.alayanew.com/docs/documents/useGuide/LLaMAFactory/mutiple/?utm_source=LLaMA-Factory
+
+> [!NOTE]
+> Except for the above links, all other websites are unauthorized third-party websites. Please carefully use them.
 
 ## Table of Contents
 
 - [Features](#features)
-- [Benchmark](#benchmark)
+- [Blogs](#blogs)
 - [Changelog](#changelog)
 - [Supported Models](#supported-models)
 - [Supported Training Approaches](#supported-training-approaches)
 - [Provided Datasets](#provided-datasets)
 - [Requirement](#requirement)
 - [Getting Started](#getting-started)
+  - [Installation](#installation)
+  - [Data Preparation](#data-preparation)
+  - [Quickstart](#quickstart)
+  - [Fine-Tuning with LLaMA Board GUI](#fine-tuning-with-llama-board-gui-powered-by-gradio)
+  - [Build Docker](#build-docker)
+  - [Deploy with OpenAI-style API and vLLM](#deploy-with-openai-style-api-and-vllm)
+  - [Download from ModelScope Hub](#download-from-modelscope-hub)
+  - [Download from Modelers Hub](#download-from-modelers-hub)
+  - [Use W&B Logger](#use-wb-logger)
+  - [Use SwanLab Logger](#use-swanlab-logger)
 - [Projects using LLaMA Factory](#projects-using-llama-factory)
 - [License](#license)
 - [Citation](#citation)
@@ -46,40 +87,96 @@ Choose your path:
 
 ## Features
 
-- **Various models**: LLaMA, LLaVA, Mistral, Mixtral-MoE, Qwen, Yi, Gemma, Baichuan, ChatGLM, Phi, etc.
+- **Various models**: LLaMA, LLaVA, Mistral, Mixtral-MoE, Qwen, Qwen2-VL, DeepSeek, Yi, Gemma, ChatGLM, Phi, etc.
 - **Integrated methods**: (Continuous) pre-training, (multimodal) supervised fine-tuning, reward modeling, PPO, DPO, KTO, ORPO, etc.
-- **Scalable resources**: 32-bit full-tuning, 16-bit freeze-tuning, 16-bit LoRA and 2/4/8-bit QLoRA via AQLM/AWQ/GPTQ/LLM.int8.
-- **Advanced algorithms**: GaLore, BAdam, DoRA, LongLoRA, LLaMA Pro, Mixture-of-Depths, LoRA+, LoftQ, PiSSA and Agent tuning.
-- **Practical tricks**: FlashAttention-2, Unsloth, RoPE scaling, NEFTune and rsLoRA.
-- **Experiment monitors**: LlamaBoard, TensorBoard, Wandb, MLflow, etc.
-- **Faster inference**: OpenAI-style API, Gradio UI and CLI with vLLM worker.
+- **Scalable resources**: 16-bit full-tuning, freeze-tuning, LoRA and 2/3/4/5/6/8-bit QLoRA via AQLM/AWQ/GPTQ/LLM.int8/HQQ/EETQ.
+- **Advanced algorithms**: [GaLore](https://github.com/jiaweizzhao/GaLore), [BAdam](https://github.com/Ledzy/BAdam), [APOLLO](https://github.com/zhuhanqing/APOLLO), [Adam-mini](https://github.com/zyushun/Adam-mini), [Muon](https://github.com/KellerJordan/Muon), DoRA, LongLoRA, LLaMA Pro, Mixture-of-Depths, LoRA+, LoftQ and PiSSA.
+- **Practical tricks**: [FlashAttention-2](https://github.com/Dao-AILab/flash-attention), [Unsloth](https://github.com/unslothai/unsloth), [Liger Kernel](https://github.com/linkedin/Liger-Kernel), RoPE scaling, NEFTune and rsLoRA.
+- **Wide tasks**: Multi-turn dialogue, tool using, image understanding, visual grounding, video recognition, audio understanding, etc.
+- **Experiment monitors**: LlamaBoard, TensorBoard, Wandb, MLflow, [SwanLab](https://github.com/SwanHubX/SwanLab), etc.
+- **Faster inference**: OpenAI-style API, Gradio UI and CLI with [vLLM worker](https://github.com/vllm-project/vllm) or [SGLang worker](https://github.com/sgl-project/sglang).
 
-## Benchmark
+### Day-N Support for Fine-Tuning Cutting-Edge Models
 
-Compared to ChatGLM's [P-Tuning](https://github.com/THUDM/ChatGLM2-6B/tree/main/ptuning), LLaMA Factory's LoRA tuning offers up to **3.7 times faster** training speed with a better Rouge score on the advertising text generation task. By leveraging 4-bit quantization technique, LLaMA Factory's QLoRA further improves the efficiency regarding the GPU memory.
+| Support Date | Model Name                                                   |
+| ------------ | ------------------------------------------------------------ |
+| Day 0        | Qwen3 / Qwen2.5-VL / Gemma 3 / InternLM 3 / MiniCPM-o-2.6    |
+| Day 1        | Llama 3 / GLM-4 / Mistral Small / PaliGemma2 / Llama 4       |
 
-![benchmark](assets/benchmark.svg)
+## Blogs
 
-<details><summary>Definitions</summary>
+- [Fine-tune Qwen2.5-VL for Autonomous Driving using LLaMA-Factory](https://docs.alayanew.com/docs/documents/useGuide/LLaMAFactory/mutiple/?utm_source=LLaMA-Factory) (Chinese)
+- [How Apoidea Group enhances visual information extraction from banking documents with multimodal models using LLaMA-Factory on Amazon SageMaker HyperPod](https://aws.amazon.com/cn/blogs/machine-learning/how-apoidea-group-enhances-visual-information-extraction-from-banking-documents-with-multimodal-models-using-llama-factory-on-amazon-sagemaker-hyperpod/) (English)
+- [Easy Dataset × LLaMA Factory: Enabling LLMs to Efficiently Learn Domain Knowledge](https://buaa-act.feishu.cn/wiki/GVzlwYcRFiR8OLkHbL6cQpYin7g) (English)
 
-- **Training Speed**: the number of training samples processed per second during the training. (bs=4, cutoff_len=1024)
-- **Rouge Score**: Rouge-2 score on the development set of the [advertising text generation](https://aclanthology.org/D19-1321.pdf) task. (bs=4, cutoff_len=1024)
-- **GPU Memory**: Peak GPU memory usage in 4-bit quantized training. (bs=1, cutoff_len=1024)
-- We adopt `pre_seq_len=128` for ChatGLM's P-Tuning and `lora_rank=32` for LLaMA Factory's LoRA tuning.
+<details><summary>All Blogs</summary>
+
+- [LLaMA Factory: Fine-tuning the DeepSeek-R1-Distill-Qwen-7B Model for News Classifier](https://gallery.pai-ml.com/#/preview/deepLearning/nlp/llama_factory_deepseek_r1_distill_7b) (Chinese)
+- [A One-Stop Code-Free Model Fine-Tuning \& Deployment Platform based on SageMaker and LLaMA-Factory](https://aws.amazon.com/cn/blogs/china/a-one-stop-code-free-model-fine-tuning-deployment-platform-based-on-sagemaker-and-llama-factory/) (Chinese)
+- [LLaMA Factory Multi-Modal Fine-Tuning Practice: Fine-Tuning Qwen2-VL for Personal Tourist Guide](https://gallery.pai-ml.com/#/preview/deepLearning/nlp/llama_factory_qwen2vl) (Chinese)
+- [LLaMA Factory: Fine-tuning the LLaMA3 Model for Role-Playing](https://gallery.pai-ml.com/#/preview/deepLearning/nlp/llama_factory) (Chinese)
 
 </details>
 
 ## Changelog
 
-[24/06/16] We support **[PiSSA](https://arxiv.org/abs/2404.02948)** algorithm. See [examples](examples/README.md) for usage.
+[25/04/28] We supported fine-tuning the **[Qwen3](https://qwenlm.github.io/blog/qwen3/)** model family.
+
+[25/04/21] We supported the **[Muon](https://github.com/KellerJordan/Muon)** optimizer. See [examples](examples/README.md) for usage. Thank [@tianshijing](https://github.com/tianshijing)'s PR.
+
+[25/04/16] We supported fine-tuning the **[InternVL3](https://huggingface.co/OpenGVLab/InternVL3-8B)** model. See [PR #7258](https://github.com/hiyouga/LLaMA-Factory/pull/7258) to get started.
+
+[25/04/14] We supported fine-tuning the **[GLM-Z1](https://huggingface.co/THUDM/GLM-Z1-9B-0414)** and **[Kimi-VL](https://huggingface.co/moonshotai/Kimi-VL-A3B-Instruct)** models.
+
+[25/04/06] We supported fine-tuning the **[Llama 4](https://ai.meta.com/blog/llama-4-multimodal-intelligence/)** model. See [PR #7611](https://github.com/hiyouga/LLaMA-Factory/pull/7611) to get started.
+
+<details><summary>Full Changelog</summary>
+
+[25/03/31] We supported fine-tuning the **[Qwen2.5 Omni](https://qwenlm.github.io/blog/qwen2.5-omni/)** model. See [PR #7537](https://github.com/hiyouga/LLaMA-Factory/pull/7537) to get started.
+
+[25/03/15] We supported **[SGLang](https://github.com/sgl-project/sglang)** as inference backend. Try `infer_backend: sglang` to accelerate inference.
+
+[25/03/12] We supported fine-tuning the **[Gemma 3](https://huggingface.co/blog/gemma3)** model.
+
+[25/02/24] Announcing **[EasyR1](https://github.com/hiyouga/EasyR1)**, an efficient, scalable and multi-modality RL training framework for efficient GRPO training.
+
+[25/02/11] We supported saving the **[Ollama](https://github.com/ollama/ollama)** modelfile when exporting the model checkpoints. See [examples](examples/README.md) for usage.
+
+[25/02/05] We supported fine-tuning the **[Qwen2-Audio](Qwen/Qwen2-Audio-7B-Instruct)** and **[MiniCPM-o-2.6](https://huggingface.co/openbmb/MiniCPM-o-2_6)** on audio understanding tasks.
+
+[25/01/31] We supported fine-tuning the **[DeepSeek-R1](https://huggingface.co/deepseek-ai/DeepSeek-R1)** and **[Qwen2.5-VL](https://huggingface.co/Qwen/Qwen2.5-VL-7B-Instruct)** models.
+
+[25/01/15] We supported **[APOLLO](https://arxiv.org/abs/2412.05270)** optimizer. See [examples](examples/README.md) for usage.
+
+[25/01/14] We supported fine-tuning the **[MiniCPM-o-2.6](https://huggingface.co/openbmb/MiniCPM-o-2_6)** and **[MiniCPM-V-2.6](https://huggingface.co/openbmb/MiniCPM-V-2_6)** models. Thank [@BUAADreamer](https://github.com/BUAADreamer)'s PR.
+
+[25/01/14] We supported fine-tuning the **[InternLM 3](https://huggingface.co/collections/internlm/)** models. Thank [@hhaAndroid](https://github.com/hhaAndroid)'s PR.
+
+[25/01/10] We supported fine-tuning the **[Phi-4](https://huggingface.co/microsoft/phi-4)** model.
+
+[24/12/21] We supported using **[SwanLab](https://github.com/SwanHubX/SwanLab)** for experiment tracking and visualization. See [this section](#use-swanlab-logger) for details.
+
+[24/11/27] We supported fine-tuning the **[Skywork-o1](https://huggingface.co/Skywork/Skywork-o1-Open-Llama-3.1-8B)** model and the **[OpenO1](https://huggingface.co/datasets/O1-OPEN/OpenO1-SFT)** dataset.
+
+[24/10/09] We supported downloading pre-trained models and datasets from the **[Modelers Hub](https://modelers.cn/models)**. See [this tutorial](#download-from-modelers-hub) for usage.
+
+[24/09/19] We supported fine-tuning the **[Qwen2.5](https://qwenlm.github.io/blog/qwen2.5/)** models.
+
+[24/08/30] We supported fine-tuning the **[Qwen2-VL](https://qwenlm.github.io/blog/qwen2-vl/)** models. Thank [@simonJJJ](https://github.com/simonJJJ)'s PR.
+
+[24/08/27] We supported **[Liger Kernel](https://github.com/linkedin/Liger-Kernel)**. Try `enable_liger_kernel: true` for efficient training.
+
+[24/08/09] We supported **[Adam-mini](https://github.com/zyushun/Adam-mini)** optimizer. See [examples](examples/README.md) for usage. Thank [@relic-yuexi](https://github.com/relic-yuexi)'s PR.
+
+[24/07/04] We supported [contamination-free packed training](https://github.com/MeetKai/functionary/tree/main/functionary/train/packing). Use `neat_packing: true` to activate it. Thank [@chuan298](https://github.com/chuan298)'s PR.
+
+[24/06/16] We supported **[PiSSA](https://arxiv.org/abs/2404.02948)** algorithm. See [examples](examples/README.md) for usage.
 
 [24/06/07] We supported fine-tuning the **[Qwen2](https://qwenlm.github.io/blog/qwen2/)** and **[GLM-4](https://github.com/THUDM/GLM-4)** models.
 
 [24/05/26] We supported **[SimPO](https://arxiv.org/abs/2405.14734)** algorithm for preference learning. See [examples](examples/README.md) for usage.
 
-<details><summary>Full Changelog</summary>
-
-[24/05/20] We supported fine-tuning the **PaliGemma** series models. Note that the PaliGemma models are pre-trained models, you need to fine-tune them with `gemma` template for chat completion.
+[24/05/20] We supported fine-tuning the **PaliGemma** series models. Note that the PaliGemma models are pre-trained models, you need to fine-tune them with `paligemma` template for chat completion.
 
 [24/05/18] We supported **[KTO](https://arxiv.org/abs/2402.01306)** algorithm for preference learning. See [examples](examples/README.md) for usage.
 
@@ -91,7 +188,7 @@ Compared to ChatGLM's [P-Tuning](https://github.com/THUDM/ChatGLM2-6B/tree/main/
 
 [24/04/21] We supported **[Mixture-of-Depths](https://arxiv.org/abs/2404.02258)** according to [AstraMindAI's implementation](https://github.com/astramind-ai/Mixture-of-depths). See [examples](examples/README.md) for usage.
 
-[24/04/16] We supported **[BAdam](https://arxiv.org/abs/2404.02827)**. See [examples](examples/README.md) for usage.
+[24/04/16] We supported **[BAdam](https://arxiv.org/abs/2404.02827)** optimizer. See [examples](examples/README.md) for usage.
 
 [24/04/16] We supported **[unsloth](https://github.com/unslothai/unsloth)**'s long-sequence training (Llama-2-7B-56k within 24GB). It achieves **117%** speed and **50%** memory compared with FlashAttention-2, more benchmarks can be found in [this page](https://github.com/hiyouga/LLaMA-Factory/wiki/Performance-comparison).
 
@@ -103,7 +200,7 @@ Compared to ChatGLM's [P-Tuning](https://github.com/THUDM/ChatGLM2-6B/tree/main/
 
 [24/03/13] We supported **[LoRA+](https://arxiv.org/abs/2402.12354)**. See [examples](examples/README.md) for usage.
 
-[24/03/07] We supported gradient low-rank projection (**[GaLore](https://arxiv.org/abs/2403.03507)**) algorithm. See [examples](examples/README.md) for usage.
+[24/03/07] We supported **[GaLore](https://arxiv.org/abs/2403.03507)** optimizer. See [examples](examples/README.md) for usage.
 
 [24/03/07] We integrated **[vLLM](https://github.com/vllm-project/vllm)** for faster and concurrent inference. Try `infer_backend: vllm` to enjoy **270%** inference speed.
 
@@ -119,7 +216,7 @@ Compared to ChatGLM's [P-Tuning](https://github.com/THUDM/ChatGLM2-6B/tree/main/
 
 [23/12/12] We supported fine-tuning the latest MoE model **[Mixtral 8x7B](https://huggingface.co/mistralai/Mixtral-8x7B-v0.1)** in our framework. See hardware requirement [here](#hardware-requirement).
 
-[23/12/01] We supported downloading pre-trained models and datasets from the **[ModelScope Hub](https://modelscope.cn/models)** for Chinese mainland users. See [this tutorial](#download-from-modelscope-hub) for usage.
+[23/12/01] We supported downloading pre-trained models and datasets from the **[ModelScope Hub](https://modelscope.cn/models)**. See [this tutorial](#download-from-modelscope-hub) for usage.
 
 [23/10/21] We supported **[NEFTune](https://arxiv.org/abs/2310.05914)** trick for fine-tuning. Try `neftune_noise_alpha: 5` argument to activate NEFTune.
 
@@ -149,42 +246,74 @@ Compared to ChatGLM's [P-Tuning](https://github.com/THUDM/ChatGLM2-6B/tree/main/
 
 </details>
 
+> [!TIP]
+> If you cannot use the latest feature, please pull the latest code and install LLaMA-Factory again.
+
 ## Supported Models
 
 | Model                                                             | Model size                       | Template            |
-| --------------------------------------------------------- | -------------------------------- | --------- |
+| ----------------------------------------------------------------- | -------------------------------- | ------------------- |
 | [Baichuan 2](https://huggingface.co/baichuan-inc)                 | 7B/13B                           | baichuan2           |
-| [BLOOM](https://huggingface.co/bigscience)                | 560M/1.1B/1.7B/3B/7.1B/176B      | -         |
-| [BLOOMZ](https://huggingface.co/bigscience)               | 560M/1.1B/1.7B/3B/7.1B/176B      | -         |
+| [BLOOM/BLOOMZ](https://huggingface.co/bigscience)                 | 560M/1.1B/1.7B/3B/7.1B/176B      | -                   |
 | [ChatGLM3](https://huggingface.co/THUDM)                          | 6B                               | chatglm3            |
-| [Command-R](https://huggingface.co/CohereForAI)           | 35B/104B                         | cohere    |
+| [Command R](https://huggingface.co/CohereForAI)                   | 35B/104B                         | cohere              |
 | [DeepSeek (Code/MoE)](https://huggingface.co/deepseek-ai)         | 7B/16B/67B/236B                  | deepseek            |
+| [DeepSeek 2.5/3](https://huggingface.co/deepseek-ai)              | 236B/671B                        | deepseek3           |
+| [DeepSeek R1 (Distill)](https://huggingface.co/deepseek-ai)       | 1.5B/7B/8B/14B/32B/70B/671B      | deepseekr1          |
 | [Falcon](https://huggingface.co/tiiuae)                           | 7B/11B/40B/180B                  | falcon              |
-| [Gemma/CodeGemma](https://huggingface.co/google)          | 2B/7B                            | gemma     |
-| [GLM4](https://huggingface.co/THUDM)                      | 9B                               | glm4      |
-| [InternLM2](https://huggingface.co/internlm)              | 7B/20B                           | intern2   |
-| [LLaMA](https://github.com/facebookresearch/llama)        | 7B/13B/33B/65B                   | -         |
-| [LLaMA-2](https://huggingface.co/meta-llama)              | 7B/13B/70B                       | llama2    |
-| [LLaMA-3](https://huggingface.co/meta-llama)              | 8B/70B                           | llama3    |
-| [LLaVA-1.5](https://huggingface.co/llava-hf)              | 7B/13B                           | vicuna    |
+| [Gemma/Gemma 2/CodeGemma](https://huggingface.co/google)          | 2B/7B/9B/27B                     | gemma               |
+| [Gemma 3](https://huggingface.co/google)                          | 1B/4B/12B/27B                    | gemma3/gemma (1B)   |
+| [GLM-4/GLM-4-0414/GLM-Z1](https://huggingface.co/THUDM)           | 9B/32B                           | glm4/glmz1          |
+| [GPT-2](https://huggingface.co/openai-community)                  | 0.1B/0.4B/0.8B/1.5B              | -                   |
+| [Granite 3.0-3.3](https://huggingface.co/ibm-granite)             | 1B/2B/3B/8B                      | granite3            |
+| [Hunyuan](https://huggingface.co/tencent/)                        | 7B                               | hunyuan             |
+| [Index](https://huggingface.co/IndexTeam)                         | 1.9B                             | index               |
+| [InternLM 2-3](https://huggingface.co/internlm)                   | 7B/8B/20B                        | intern2             |
+| [InternVL 2.5-3](https://huggingface.co/OpenGVLab)                | 1B/2B/8B/14B/38B/78B             | intern_vl           |
+| [Kimi-VL](https://huggingface.co/moonshotai)                      | 16B                              | kimi_vl             |
+| [Llama](https://github.com/facebookresearch/llama)                | 7B/13B/33B/65B                   | -                   |
+| [Llama 2](https://huggingface.co/meta-llama)                      | 7B/13B/70B                       | llama2              |
+| [Llama 3-3.3](https://huggingface.co/meta-llama)                  | 1B/3B/8B/70B                     | llama3              |
+| [Llama 4](https://huggingface.co/meta-llama)                      | 109B/402B                        | llama4              |
+| [Llama 3.2 Vision](https://huggingface.co/meta-llama)             | 11B/90B                          | mllama              |
+| [LLaVA-1.5](https://huggingface.co/llava-hf)                      | 7B/13B                           | llava               |
+| [LLaVA-NeXT](https://huggingface.co/llava-hf)                     | 7B/8B/13B/34B/72B/110B           | llava_next          |
+| [LLaVA-NeXT-Video](https://huggingface.co/llava-hf)               | 7B/34B                           | llava_next_video    |
+| [MiMo](https://huggingface.co/XiaomiMiMo)                         | 7B                               | mimo                |
+| [MiniCPM](https://huggingface.co/openbmb)                         | 0.5B/1B/2B/4B/8B                 | cpm/cpm3/cpm4       |
+| [MiniCPM-o-2.6/MiniCPM-V-2.6](https://huggingface.co/openbmb)     | 8B                               | minicpm_o/minicpm_v |
+| [Ministral/Mistral-Nemo](https://huggingface.co/mistralai)        | 8B/12B                           | ministral           |
 | [Mistral/Mixtral](https://huggingface.co/mistralai)               | 7B/8x7B/8x22B                    | mistral             |
+| [Mistral Small](https://huggingface.co/mistralai)                 | 24B                              | mistral_small       |
 | [OLMo](https://huggingface.co/allenai)                            | 1B/7B                            | -                   |
-| [PaliGemma](https://huggingface.co/google)                | 3B                               | gemma     |
-| [Phi-1.5/2](https://huggingface.co/microsoft)             | 1.3B/2.7B                        | -         |
-| [Phi-3](https://huggingface.co/microsoft)                 | 4B/7B/14B                        | phi       |
-| [Qwen](https://huggingface.co/Qwen)                       | 1.8B/7B/14B/72B                  | qwen      |
-| [Qwen1.5 (Code/MoE)](https://huggingface.co/Qwen)         | 0.5B/1.8B/4B/7B/14B/32B/72B/110B | qwen      |
-| [Qwen2 (MoE)](https://huggingface.co/Qwen)                | 0.5B/1.5B/7B/57B/72B             | qwen      |
+| [PaliGemma/PaliGemma2](https://huggingface.co/google)             | 3B/10B/28B                       | paligemma           |
+| [Phi-1.5/Phi-2](https://huggingface.co/microsoft)                 | 1.3B/2.7B                        | -                   |
+| [Phi-3/Phi-3.5](https://huggingface.co/microsoft)                 | 4B/14B                           | phi                 |
+| [Phi-3-small](https://huggingface.co/microsoft)                   | 7B                               | phi_small           |
+| [Phi-4](https://huggingface.co/microsoft)                         | 14B                              | phi4                |
+| [Pixtral](https://huggingface.co/mistralai)                       | 12B                              | pixtral             |
+| [Qwen (1-2.5) (Code/Math/MoE/QwQ)](https://huggingface.co/Qwen)   | 0.5B/1.5B/3B/7B/14B/32B/72B/110B | qwen                |
+| [Qwen3 (MoE)](https://huggingface.co/Qwen)                        | 0.6B/1.7B/4B/8B/14B/32B/235B     | qwen3               |
+| [Qwen2-Audio](https://huggingface.co/Qwen)                        | 7B                               | qwen2_audio         |
+| [Qwen2.5-Omni](https://huggingface.co/Qwen)                       | 3B/7B                            | qwen2_omni          |
+| [Qwen2-VL/Qwen2.5-VL/QVQ](https://huggingface.co/Qwen)            | 2B/3B/7B/32B/72B                 | qwen2_vl            |
+| [Seed Coder](https://huggingface.co/ByteDance-Seed)               | 8B                               | seed_coder          |
+| [Skywork o1](https://huggingface.co/Skywork)                      | 8B                               | skywork_o1          |
 | [StarCoder 2](https://huggingface.co/bigcode)                     | 3B/7B/15B                        | -                   |
+| [TeleChat2](https://huggingface.co/Tele-AI)                       | 3B/7B/35B/115B                   | telechat2           |
 | [XVERSE](https://huggingface.co/xverse)                           | 7B/13B/65B                       | xverse              |
-| [Yi (1/1.5)](https://huggingface.co/01-ai)                | 6B/9B/34B                        | yi        |
+| [Yi/Yi-1.5 (Code)](https://huggingface.co/01-ai)                  | 1.5B/6B/9B/34B                   | yi                  |
 | [Yi-VL](https://huggingface.co/01-ai)                             | 6B/34B                           | yi_vl               |
-| [Yuan](https://huggingface.co/IEITYuan)                   | 2B/51B/102B                      | yuan      |
+| [Yuan 2](https://huggingface.co/IEITYuan)                         | 2B/51B/102B                      | yuan                |
 
 > [!NOTE]
 > For the "base" models, the `template` argument can be chosen from `default`, `alpaca`, `vicuna` etc. But make sure to use the **corresponding template** for the "instruct/chat" models.
 >
 > Remember to use the **SAME** template in training and inference.
+>
+> \*: You should install the `transformers` from main branch and use `DISABLE_VERSION_CHECK=1` to skip version check.
+>
+> \*\*: You need to install a specific version of `transformers` to use the corresponding model.
 
 Please refer to [constants.py](src/llamafactory/extras/constants.py) for a full list of models we supported.
 
@@ -203,6 +332,9 @@ You also can add a custom chat template to [template.py](src/llamafactory/data/t
 | ORPO Training          | :white_check_mark: | :white_check_mark: | :white_check_mark: | :white_check_mark: |
 | SimPO Training         | :white_check_mark: | :white_check_mark: | :white_check_mark: | :white_check_mark: |
 
+> [!TIP]
+> The implementation details of PPO can be found in [this blog](https://newfacade.github.io/notes-on-reinforcement-learning/17-ppo-trl.html).
+
 ## Provided Datasets
 
 <details><summary>Pre-training datasets</summary>
@@ -260,9 +392,15 @@ You also can add a custom chat template to [template.py](src/llamafactory/data/t
 - [STEM (zh)](https://huggingface.co/datasets/hfl/stem_zh_instruction)
 - [Ruozhiba (zh)](https://huggingface.co/datasets/hfl/ruozhiba_gpt4_turbo)
 - [Neo-sft (zh)](https://huggingface.co/datasets/m-a-p/neo_sft_phase2)
-- [WebInstructSub (en)](https://huggingface.co/datasets/TIGER-Lab/WebInstructSub)
 - [Magpie-Pro-300K-Filtered (en)](https://huggingface.co/datasets/Magpie-Align/Magpie-Pro-300K-Filtered)
+- [Magpie-ultra-v0.1 (en)](https://huggingface.co/datasets/argilla/magpie-ultra-v0.1)
+- [WebInstructSub (en)](https://huggingface.co/datasets/TIGER-Lab/WebInstructSub)
+- [OpenO1-SFT (en&zh)](https://huggingface.co/datasets/O1-OPEN/OpenO1-SFT)
+- [Open-Thoughts (en)](https://huggingface.co/datasets/open-thoughts/OpenThoughts-114k)
+- [Open-R1-Math (en)](https://huggingface.co/datasets/open-r1/OpenR1-Math-220k)
+- [Chinese-DeepSeek-R1-Distill (zh)](https://huggingface.co/datasets/Congliu/Chinese-DeepSeek-R1-Distill-data-110k-SFT)
 - [LLaVA mixed (en&zh)](https://huggingface.co/datasets/BUAADreamer/llava-en-zh-300k)
+- [Pokemon-gpt4o-captions (en&zh)](https://huggingface.co/datasets/jugg1024/pokemon-gpt4o-captions)
 - [Open Assistant (de)](https://huggingface.co/datasets/mayflowergmbh/oasst_de)
 - [Dolly 15k (de)](https://huggingface.co/datasets/mayflowergmbh/dolly-15k_de)
 - [Alpaca GPT4 (de)](https://huggingface.co/datasets/mayflowergmbh/alpaca-gpt4_de)
@@ -279,6 +417,10 @@ You also can add a custom chat template to [template.py](src/llamafactory/data/t
 
 - [DPO mixed (en&zh)](https://huggingface.co/datasets/hiyouga/DPO-En-Zh-20k)
 - [UltraFeedback (en)](https://huggingface.co/datasets/HuggingFaceH4/ultrafeedback_binarized)
+- [COIG-P (zh)](https://huggingface.co/datasets/m-a-p/COIG-P)
+- [RLHF-V (en)](https://huggingface.co/datasets/openbmb/RLHF-V-Dataset)
+- [VLFeedback (en)](https://huggingface.co/datasets/Zhihui/VLFeedback)
+- [RLAIF-V (en)](https://huggingface.co/datasets/openbmb/RLAIF-V-Dataset)
 - [Orca DPO Pairs (en)](https://huggingface.co/datasets/Intel/orca_dpo_pairs)
 - [HH-RLHF (en)](https://huggingface.co/datasets/Anthropic/hh-rlhf)
 - [Nectar (en)](https://huggingface.co/datasets/berkeley-nest/Nectar)
@@ -298,35 +440,35 @@ huggingface-cli login
 
 | Mandatory    | Minimum | Recommend |
 | ------------ | ------- | --------- |
-| python       | 3.8     | 3.11      |
-| torch        | 1.13.1  | 2.3.0     |
-| transformers | 4.41.2  | 4.41.2    |
-| datasets     | 2.16.0  | 2.19.2    |
-| accelerate   | 0.30.1  | 0.30.1    |
-| peft         | 0.11.1  | 0.11.1    |
-| trl          | 0.8.6   | 0.9.4     |
+| python       | 3.9     | 3.10      |
+| torch        | 2.0.0   | 2.6.0     |
+| torchvision  | 0.15.0  | 0.21.0    |
+| transformers | 4.45.0  | 4.50.0    |
+| datasets     | 2.16.0  | 3.2.0     |
+| accelerate   | 0.34.0  | 1.2.1     |
+| peft         | 0.14.0  | 0.15.1    |
+| trl          | 0.8.6   | 0.9.6     |
 
 | Optional     | Minimum | Recommend |
 | ------------ | ------- | --------- |
 | CUDA         | 11.6    | 12.2      |
-| deepspeed    | 0.10.0  | 0.14.0    |
+| deepspeed    | 0.10.0  | 0.16.4    |
 | bitsandbytes | 0.39.0  | 0.43.1    |
-| vllm         | 0.4.3   | 0.4.3     |
-| flash-attn   | 2.3.0   | 2.5.9     |
+| vllm         | 0.4.3   | 0.8.2     |
+| flash-attn   | 2.5.6   | 2.7.2     |
 
 ### Hardware Requirement
 
 \* *estimated*
 
-| Method            | Bits |   7B  |  13B  |  30B  |   70B  |  110B  |  8x7B |  8x22B |
-| ----------------- | ---- | ----- | ----- | ----- | ------ | ------ | ----- | ------ |
-| Full              | AMP  | 120GB | 240GB | 600GB | 1200GB | 2000GB | 900GB | 2400GB |
-| Full              |  16  |  60GB | 120GB | 300GB |  600GB |  900GB | 400GB | 1200GB |
-| Freeze            |  16  |  20GB |  40GB |  80GB |  200GB |  360GB | 160GB |  400GB |
-| LoRA/GaLore/BAdam |  16  |  16GB |  32GB |  64GB |  160GB |  240GB | 120GB |  320GB |
-| QLoRA             |   8  |  10GB |  20GB |  40GB |   80GB |  140GB |  60GB |  160GB |
-| QLoRA             |   4  |   6GB |  12GB |  24GB |   48GB |   72GB |  30GB |   96GB |
-| QLoRA             |   2  |   4GB |   8GB |  16GB |   24GB |   48GB |  18GB |   48GB |
+| Method                          | Bits |   7B  |  14B  |  30B  |   70B  |   `x`B  |
+| ------------------------------- | ---- | ----- | ----- | ----- | ------ | ------- |
+| Full (`bf16` or `fp16`)         |  32  | 120GB | 240GB | 600GB | 1200GB | `18x`GB |
+| Full (`pure_bf16`)              |  16  |  60GB | 120GB | 300GB |  600GB |  `8x`GB |
+| Freeze/LoRA/GaLore/APOLLO/BAdam |  16  |  16GB |  32GB |  64GB |  160GB |  `2x`GB |
+| QLoRA                           |   8  |  10GB |  20GB |  40GB |   80GB |   `x`GB |
+| QLoRA                           |   4  |   6GB |  12GB |  24GB |   48GB | `x/2`GB |
+| QLoRA                           |   2  |   4GB |   8GB |  16GB |   24GB | `x/4`GB |
 
 ## Getting Started
 
@@ -335,74 +477,151 @@ huggingface-cli login
 > [!IMPORTANT]
 > Installation is mandatory.
 
+#### Install from Source
+
 ```bash
 git clone --depth 1 https://github.com/hiyouga/LLaMA-Factory.git
 cd LLaMA-Factory
-pip install -e ".[torch,metrics]"
+pip install -e ".[torch,metrics]" --no-build-isolation
 ```
 
-Extra dependencies available: torch, torch_npu, metrics, deepspeed, bitsandbytes, vllm, galore, badam, gptq, awq, aqlm, qwen, modelscope, quality
+Extra dependencies available: torch, torch-npu, metrics, deepspeed, liger-kernel, bitsandbytes, hqq, eetq, gptq, aqlm, vllm, sglang, galore, apollo, badam, adam-mini, qwen, minicpm_v, modelscope, openmind, swanlab, dev
 
-> [!TIP]
-> Use `pip install --no-deps -e .` to resolve package conflicts.
+#### Install from Docker Image
+
+```bash
+docker run -it --rm --gpus=all --ipc=host hiyouga/llamafactory:latest
+```
+
+This image is built on Ubuntu 22.04 (x86\_64), CUDA 12.4, Python 3.11, PyTorch 2.6.0, and Flash-attn 2.7.4.
+
+Find the pre-built images: https://hub.docker.com/r/hiyouga/llamafactory/tags
+
+Please refer to [build docker](#build-docker) to build the image yourself.
+
+<details><summary>Setting up a virtual environment with <b>uv</b></summary>
+
+Create an isolated Python environment with [uv](https://github.com/astral-sh/uv):
+
+```bash
+uv sync --extra torch --extra metrics --prerelease=allow
+```
+
+Run LLaMA-Factory in the isolated environment:
+
+```bash
+uv run --prerelease=allow llamafactory-cli train examples/train_lora/llama3_lora_pretrain.yaml
+```
+
+</details>
 
 <details><summary>For Windows users</summary>
 
+#### Install PyTorch
+
+You need to manually install the GPU version of PyTorch on the Windows platform. Please refer to the [official website](https://pytorch.org/get-started/locally/) and the following command to install PyTorch with CUDA support:
+
+```bash
+pip uninstall torch torchvision torchaudio
+pip install torch torchvision torchaudio --index-url https://download.pytorch.org/whl/cu126
+python -c "import torch; print(torch.cuda.is_available())"
+```
+
+If you see `True` then you have successfully installed PyTorch with CUDA support.
+
+Try `dataloader_num_workers: 0` if you encounter `Can't pickle local object` error.
+
+#### Install BitsAndBytes
+
 If you want to enable the quantized LoRA (QLoRA) on the Windows platform, you need to install a pre-built version of `bitsandbytes` library, which supports CUDA 11.1 to 12.2, please select the appropriate [release version](https://github.com/jllllll/bitsandbytes-windows-webui/releases/tag/wheels) based on your CUDA version.
 
 ```bash
 pip install https://github.com/jllllll/bitsandbytes-windows-webui/releases/download/wheels/bitsandbytes-0.41.2.post2-py3-none-win_amd64.whl
 ```
 
-To enable FlashAttention-2 on the Windows platform, you need to install the precompiled `flash-attn` library, which supports CUDA 12.1 to 12.2. Please download the corresponding version from [flash-attention](https://github.com/bdashore3/flash-attention/releases) based on your requirements.
+#### Install Flash Attention-2
+
+To enable FlashAttention-2 on the Windows platform, please use the script from [flash-attention-windows-wheel](https://huggingface.co/lldacing/flash-attention-windows-wheel) to compile and install it by yourself.
 
 </details>
 
 <details><summary>For Ascend NPU users</summary>
 
-Join [NPU user group](assets/wechat_npu.jpg).
-
-To install LLaMA Factory on Ascend NPU devices, please specify extra dependencies: `pip install -e '.[torch-npu,metrics]'`. Additionally, you need to install the **[Ascend CANN Toolkit and Kernels](https://www.hiascend.com/developer/download/community/result?module=cann)**. Please follow the [installation tutorial](https://www.hiascend.com/document/detail/en/CANNCommunityEdition/600alphaX/softwareinstall/instg/atlasdeploy_03_0031.html) or use the following commands:
+To install LLaMA Factory on Ascend NPU devices, please upgrade Python to version 3.10 or higher and specify extra dependencies: `pip install -e ".[torch-npu,metrics]"`. Additionally, you need to install the **[Ascend CANN Toolkit and Kernels](https://www.hiascend.com/developer/download/community/result?module=cann)**. Please follow the [installation tutorial](https://www.hiascend.com/document/detail/en/CANNCommunityEdition/600alphaX/softwareinstall/instg/atlasdeploy_03_0031.html) or use the following commands:
 
 ```bash
 # replace the url according to your CANN version and devices
 # install CANN Toolkit
-wget https://ascend-repo.obs.cn-east-2.myhuaweicloud.com/Milan-ASL/Milan-ASL%20V100R001C17SPC701/Ascend-cann-toolkit_8.0.RC1.alpha001_linux-"$(uname -i)".run
-bash Ascend-cann-toolkit_8.0.RC1.alpha001_linux-"$(uname -i)".run --install
+wget https://ascend-repo.obs.cn-east-2.myhuaweicloud.com/Milan-ASL/Milan-ASL%20V100R001C20SPC702/Ascend-cann-toolkit_8.0.0.alpha002_linux-"$(uname -i)".run
+bash Ascend-cann-toolkit_8.0.0.alpha002_linux-"$(uname -i)".run --install
 
 # install CANN Kernels
-wget https://ascend-repo.obs.cn-east-2.myhuaweicloud.com/Milan-ASL/Milan-ASL%20V100R001C17SPC701/Ascend-cann-kernels-910b_8.0.RC1.alpha001_linux.run
-bash Ascend-cann-kernels-910b_8.0.RC1.alpha001_linux.run --install
+wget https://ascend-repo.obs.cn-east-2.myhuaweicloud.com/Milan-ASL/Milan-ASL%20V100R001C20SPC702/Ascend-cann-kernels-910b_8.0.0.alpha002_linux-"$(uname -i)".run
+bash Ascend-cann-kernels-910b_8.0.0.alpha002_linux-"$(uname -i)".run --install
 
 # set env variables
 source /usr/local/Ascend/ascend-toolkit/set_env.sh
 ```
 
 | Requirement  | Minimum | Recommend      |
-| ------------ | ------- | ----------- |
-| CANN         | 8.0.RC1 | 8.0.RC1     |
-| torch        | 2.1.0   | 2.1.0       |
-| torch-npu    | 2.1.0   | 2.1.0.post3 |
+| ------------ | ------- | -------------- |
+| CANN         | 8.0.RC1 | 8.0.0.alpha002 |
+| torch        | 2.1.0   | 2.4.0          |
+| torch-npu    | 2.1.0   | 2.4.0.post2    |
 | deepspeed    | 0.13.2  | 0.13.2         |
-
-Docker image:
-
-- 32GB: [Download page](http://mirrors.cn-central-221.ovaijisuan.com/detail/130.html)
-- 64GB: [Download page](http://mirrors.cn-central-221.ovaijisuan.com/detail/131.html)
+| vllm-ascend  | -       | 0.7.3          |
 
 Remember to use `ASCEND_RT_VISIBLE_DEVICES` instead of `CUDA_VISIBLE_DEVICES` to specify the device to use.
 
 If you cannot infer model on NPU devices, try setting `do_sample: false` in the configurations.
 
+Download the pre-built Docker images: [32GB](http://mirrors.cn-central-221.ovaijisuan.com/detail/130.html) | [64GB](http://mirrors.cn-central-221.ovaijisuan.com/detail/131.html)
+
+#### Install BitsAndBytes
+
+To use QLoRA based on bitsandbytes on Ascend NPU, please follow these 3 steps:
+
+1. Manually compile bitsandbytes: Refer to [the installation documentation](https://huggingface.co/docs/bitsandbytes/installation?backend=Ascend+NPU&platform=Ascend+NPU) for the NPU version of bitsandbytes to complete the compilation and installation. The compilation requires a cmake version of at least 3.22.1 and a g++ version of at least 12.x.
+
+```bash
+# Install bitsandbytes from source
+# Clone bitsandbytes repo, Ascend NPU backend is currently enabled on multi-backend-refactor branch
+git clone -b multi-backend-refactor https://github.com/bitsandbytes-foundation/bitsandbytes.git
+cd bitsandbytes/
+
+# Install dependencies
+pip install -r requirements-dev.txt
+
+# Install the dependencies for the compilation tools. Note that the commands for this step may vary depending on the operating system. The following are provided for reference
+apt-get install -y build-essential cmake
+
+# Compile & install  
+cmake -DCOMPUTE_BACKEND=npu -S .
+make
+pip install .
+```
+
+2. Install transformers from the main branch.
+
+```bash
+git clone -b main https://github.com/huggingface/transformers.git
+cd transformers
+pip install .
+```
+
+3. Set `double_quantization: false` in the configuration. You can refer to the [example](examples/train_qlora/llama3_lora_sft_bnb_npu.yaml).
+
 </details>
 
 ### Data Preparation
 
-Please refer to [data/README.md](data/README.md) for checking the details about the format of dataset files. You can either use datasets on HuggingFace / ModelScope hub or load the dataset in local disk.
+Please refer to [data/README.md](data/README.md) for checking the details about the format of dataset files. You can use datasets on HuggingFace / ModelScope / Modelers hub, load the dataset in local disk, or specify a path to s3/gcs cloud storage.
 
 > [!NOTE]
 > Please update `data/dataset_info.json` to use your custom dataset.
 
+You can also use **[Easy Dataset](https://github.com/ConardLi/easy-dataset)** or **[GraphGen](https://github.com/open-sciencelab/GraphGen)** to create synthetic data for fine-tuning.
+
 ### Quickstart
 
 Use the following 3 commands to run LoRA **fine-tuning**, **inference** and **merging** of the Llama3-8B-Instruct model, respectively.
@@ -417,6 +636,8 @@ See [examples/README.md](examples/README.md) for advanced usage (including distr
 
 > [!TIP]
 > Use `llamafactory-cli help` to show help information.
+>
+> Read [FAQs](https://github.com/hiyouga/LLaMA-Factory/issues/4614) first if you encounter any problems.
 
 ### Fine-Tuning with LLaMA Board GUI (powered by [Gradio](https://github.com/gradio-app/gradio))
 
@@ -426,50 +647,117 @@ llamafactory-cli webui
 
 ### Build Docker
 
-#### Use Docker
+For CUDA users:
 
 ```bash
-docker build -f ./Dockerfile \
-    --build-arg INSTALL_BNB=false \
-    --build-arg INSTALL_VLLM=false \
-    --build-arg INSTALL_DEEPSPEED=false \
+cd docker/docker-cuda/
+docker compose up -d
+docker compose exec llamafactory bash
+```
+
+For Ascend NPU users:
+
+```bash
+cd docker/docker-npu/
+docker compose up -d
+docker compose exec llamafactory bash
+```
+
+For AMD ROCm users:
+
+```bash
+cd docker/docker-rocm/
+docker compose up -d
+docker compose exec llamafactory bash
+```
+
+<details><summary>Build without Docker Compose</summary>
+
+For CUDA users:
+
+```bash
+docker build -f ./docker/docker-cuda/Dockerfile \
     --build-arg PIP_INDEX=https://pypi.org/simple \
+    --build-arg EXTRAS=metrics \
     -t llamafactory:latest .
 
-docker run -it --gpus=all \
-    -v ./hf_cache:/root/.cache/huggingface/ \
-    -v ./data:/app/data \
-    -v ./output:/app/output \
+docker run -dit --ipc=host --gpus=all \
     -p 7860:7860 \
     -p 8000:8000 \
-    --shm-size 16G \
     --name llamafactory \
     llamafactory:latest
+
+docker exec -it llamafactory bash
 ```
 
-#### Use Docker Compose
+For Ascend NPU users:
 
 ```bash
-docker-compose up -d
-docker-compose exec llamafactory bash
+docker build -f ./docker/docker-npu/Dockerfile \
+    --build-arg PIP_INDEX=https://pypi.org/simple \
+    --build-arg EXTRAS=torch-npu,metrics \
+    -t llamafactory:latest .
+
+docker run -dit --ipc=host \
+    -v /usr/local/dcmi:/usr/local/dcmi \
+    -v /usr/local/bin/npu-smi:/usr/local/bin/npu-smi \
+    -v /usr/local/Ascend/driver:/usr/local/Ascend/driver \
+    -v /etc/ascend_install.info:/etc/ascend_install.info \
+    -p 7860:7860 \
+    -p 8000:8000 \
+    --device /dev/davinci0 \
+    --device /dev/davinci_manager \
+    --device /dev/devmm_svm \
+    --device /dev/hisi_hdc \
+    --name llamafactory \
+    llamafactory:latest
+
+docker exec -it llamafactory bash
 ```
 
-<details><summary>Details about volume</summary>
+For AMD ROCm users:
 
-- hf_cache: Utilize Hugging Face cache on the host machine. Reassignable if a cache already exists in a different directory.
-- data: Place datasets on this dir of the host machine so that they can be selected on LLaMA Board GUI.
-- output: Set export dir to this location so that the merged result can be accessed directly on the host machine.
+```bash
+docker build -f ./docker/docker-rocm/Dockerfile \
+    --build-arg PIP_INDEX=https://pypi.org/simple \
+    --build-arg EXTRAS=metrics \
+    -t llamafactory:latest .
+
+docker run -dit --ipc=host \
+    -p 7860:7860 \
+    -p 8000:8000 \
+    --device /dev/kfd \
+    --device /dev/dri \
+    --name llamafactory \
+    llamafactory:latest
+
+docker exec -it llamafactory bash
+```
+
+</details>
+
+<details><summary>Use Docker volumes</summary>
+
+You can uncomment `VOLUME [ "/root/.cache/huggingface", "/app/shared_data", "/app/output" ]` in the Dockerfile to use data volumes.
+
+When building the Docker image, use `-v ./hf_cache:/root/.cache/huggingface` argument to mount the local directory to the container. The following data volumes are available.
+
+- `hf_cache`: Utilize Hugging Face cache on the host machine.
+- `shared_data`: The directionary to store datasets on the host machine.
+- `output`: Set export dir to this location so that the merged result can be accessed directly on the host machine.
 
 </details>
 
 ### Deploy with OpenAI-style API and vLLM
 
 ```bash
-API_PORT=8000 llamafactory-cli api examples/inference/llama3_vllm.yaml
+API_PORT=8000 llamafactory-cli api examples/inference/llama3.yaml infer_backend=vllm vllm_enforce_eager=true
 ```
 
 > [!TIP]
-> Visit https://platform.openai.com/docs/api-reference/chat/create for API document.
+> Visit [this page](https://platform.openai.com/docs/api-reference/chat/create) for API document.
+>
+> Examples: [Image understanding](scripts/api_example/test_image.py) | [Function calling](scripts/api_example/test_toolcall.py)
 
 ### Download from ModelScope Hub
 
@@ -481,6 +769,16 @@ export USE_MODELSCOPE_HUB=1 # `set USE_MODELSCOPE_HUB=1` for Windows
 
 Train the model by specifying a model ID of the ModelScope Hub as the `model_name_or_path`. You can find a full list of model IDs at [ModelScope Hub](https://modelscope.cn/models), e.g., `LLM-Research/Meta-Llama-3-8B-Instruct`.
 
+### Download from Modelers Hub
+
+You can also use Modelers Hub to download models and datasets.
+
+```bash
+export USE_OPENMIND_HUB=1 # `set USE_OPENMIND_HUB=1` for Windows
+```
+
+Train the model by specifying a model ID of the Modelers Hub as the `model_name_or_path`. You can find a full list of model IDs at [Modelers Hub](https://modelers.cn/models), e.g., `TeleAI/TeleChat-7B-pt`.
+
 ### Use W&B Logger
 
 To use [Weights & Biases](https://wandb.ai) for logging experimental results, you need to add the following arguments to yaml files.
@@ -492,6 +790,21 @@ run_name: test_run # optional
 
 Set `WANDB_API_KEY` to [your key](https://wandb.ai/authorize) when launching training tasks to log in with your W&B account.
 
+### Use SwanLab Logger
+
+To use [SwanLab](https://github.com/SwanHubX/SwanLab) for logging experimental results, you need to add the following arguments to yaml files.
+
+```yaml
+use_swanlab: true
+swanlab_run_name: test_run # optional
+```
+
+When launching training tasks, you can log in to SwanLab in three ways:
+
+1. Add `swanlab_api_key=<your_api_key>` to the yaml file, and set it to your [API key](https://swanlab.cn/settings).
+2. Set the environment variable `SWANLAB_API_KEY` to your [API key](https://swanlab.cn/settings).
+3. Use the `swanlab login` command to complete the login.
+
 ## Projects using LLaMA Factory
 
 If you have a project that should be incorporated, please contact via email or create a pull request.
@@ -503,63 +816,118 @@ If you have a project that should be incorporated, please contact via email or c
 1. Wang et al. UbiPhysio: Support Daily Functioning, Fitness, and Rehabilitation with Action Understanding and Feedback in Natural Language. 2023. [[arxiv]](https://arxiv.org/abs/2308.10526)
 1. Luceri et al. Leveraging Large Language Models to Detect Influence Campaigns in Social Media. 2023. [[arxiv]](https://arxiv.org/abs/2311.07816)
 1. Zhang et al. Alleviating Hallucinations of Large Language Models through Induced Hallucinations. 2023. [[arxiv]](https://arxiv.org/abs/2312.15710)
-1. Wang et al. Know Your Needs Better: Towards Structured Understanding of Marketer Demands with Analogical Reasoning Augmented LLMs. 2024. [[arxiv]](https://arxiv.org/abs/2401.04319)
-1. Wang et al. CANDLE: Iterative Conceptualization and Instantiation Distillation from Large Language Models for Commonsense Reasoning. 2024. [[arxiv]](https://arxiv.org/abs/2401.07286)
+1. Wang et al. Know Your Needs Better: Towards Structured Understanding of Marketer Demands with Analogical Reasoning Augmented LLMs. KDD 2024. [[arxiv]](https://arxiv.org/abs/2401.04319)
+1. Wang et al. CANDLE: Iterative Conceptualization and Instantiation Distillation from Large Language Models for Commonsense Reasoning. ACL 2024. [[arxiv]](https://arxiv.org/abs/2401.07286)
 1. Choi et al. FACT-GPT: Fact-Checking Augmentation via Claim Matching with LLMs. 2024. [[arxiv]](https://arxiv.org/abs/2402.05904)
 1. Zhang et al. AutoMathText: Autonomous Data Selection with Language Models for Mathematical Texts. 2024. [[arxiv]](https://arxiv.org/abs/2402.07625)
 1. Lyu et al. KnowTuning: Knowledge-aware Fine-tuning for Large Language Models. 2024. [[arxiv]](https://arxiv.org/abs/2402.11176)
 1. Yang et al. LaCo: Large Language Model Pruning via Layer Collaps. 2024. [[arxiv]](https://arxiv.org/abs/2402.11187)
 1. Bhardwaj et al. Language Models are Homer Simpson! Safety Re-Alignment of Fine-tuned Language Models through Task Arithmetic. 2024. [[arxiv]](https://arxiv.org/abs/2402.11746)
 1. Yang et al. Enhancing Empathetic Response Generation by Augmenting LLMs with Small-scale Empathetic Models. 2024. [[arxiv]](https://arxiv.org/abs/2402.11801)
-1. Yi et al. Generation Meets Verification: Accelerating Large Language Model Inference with Smart Parallel Auto-Correct Decoding. 2024. [[arxiv]](https://arxiv.org/abs/2402.11809)
+1. Yi et al. Generation Meets Verification: Accelerating Large Language Model Inference with Smart Parallel Auto-Correct Decoding. ACL 2024 Findings. [[arxiv]](https://arxiv.org/abs/2402.11809)
 1. Cao et al. Head-wise Shareable Attention for Large Language Models. 2024. [[arxiv]](https://arxiv.org/abs/2402.11819)
 1. Zhang et al. Enhancing Multilingual Capabilities of Large Language Models through Self-Distillation from Resource-Rich Languages. 2024. [[arxiv]](https://arxiv.org/abs/2402.12204)
 1. Kim et al. Efficient and Effective Vocabulary Expansion Towards Multilingual Large Language Models. 2024. [[arxiv]](https://arxiv.org/abs/2402.14714)
-1. Yu et al. KIEval: A Knowledge-grounded Interactive Evaluation Framework for Large Language Models. 2024. [[arxiv]](https://arxiv.org/abs/2402.15043)
+1. Yu et al. KIEval: A Knowledge-grounded Interactive Evaluation Framework for Large Language Models. ACL 2024. [[arxiv]](https://arxiv.org/abs/2402.15043)
 1. Huang et al. Key-Point-Driven Data Synthesis with its Enhancement on Mathematical Reasoning. 2024. [[arxiv]](https://arxiv.org/abs/2403.02333)
 1. Duan et al. Negating Negatives: Alignment without Human Positive Samples via Distributional Dispreference Optimization. 2024. [[arxiv]](https://arxiv.org/abs/2403.03419)
 1. Xie and Schwertfeger. Empowering Robotics with Large Language Models: osmAG Map Comprehension with LLMs. 2024. [[arxiv]](https://arxiv.org/abs/2403.08228)
 1. Wu et al. Large Language Models are Parallel Multilingual Learners. 2024. [[arxiv]](https://arxiv.org/abs/2403.09073)
 1. Zhang et al. EDT: Improving Large Language Models' Generation by Entropy-based Dynamic Temperature Sampling. 2024. [[arxiv]](https://arxiv.org/abs/2403.14541)
 1. Weller et al. FollowIR: Evaluating and Teaching Information Retrieval Models to Follow Instructions. 2024. [[arxiv]](https://arxiv.org/abs/2403.15246)
-1. Hongbin Na. CBT-LLM: A Chinese Large Language Model for Cognitive Behavioral Therapy-based Mental Health Question Answering. 2024. [[arxiv]](https://arxiv.org/abs/2403.16008)
+1. Hongbin Na. CBT-LLM: A Chinese Large Language Model for Cognitive Behavioral Therapy-based Mental Health Question Answering. COLING 2024. [[arxiv]](https://arxiv.org/abs/2403.16008)
 1. Zan et al. CodeS: Natural Language to Code Repository via Multi-Layer Sketch. 2024. [[arxiv]](https://arxiv.org/abs/2403.16443)
 1. Liu et al. Extensive Self-Contrast Enables Feedback-Free Language Model Alignment. 2024. [[arxiv]](https://arxiv.org/abs/2404.00604)
 1. Luo et al. BAdam: A Memory Efficient Full Parameter Training Method for Large Language Models. 2024. [[arxiv]](https://arxiv.org/abs/2404.02827)
 1. Du et al. Chinese Tiny LLM: Pretraining a Chinese-Centric Large Language Model. 2024. [[arxiv]](https://arxiv.org/abs/2404.04167)
-1. Ma et al. Parameter Efficient Quasi-Orthogonal Fine-Tuning via Givens Rotation. 2024. [[arxiv]](https://arxiv.org/abs/2404.04316)
+1. Ma et al. Parameter Efficient Quasi-Orthogonal Fine-Tuning via Givens Rotation. ICML 2024. [[arxiv]](https://arxiv.org/abs/2404.04316)
 1. Liu et al. Dynamic Generation of Personalities with Large Language Models. 2024. [[arxiv]](https://arxiv.org/abs/2404.07084)
 1. Shang et al. How Far Have We Gone in Stripped Binary Code Understanding Using Large Language Models. 2024. [[arxiv]](https://arxiv.org/abs/2404.09836)
 1. Huang et al. LLMTune: Accelerate Database Knob Tuning with Large Language Models. 2024. [[arxiv]](https://arxiv.org/abs/2404.11581)
 1. Deng et al. Text-Tuple-Table: Towards Information Integration in Text-to-Table Generation via Global Tuple Extraction. 2024. [[arxiv]](https://arxiv.org/abs/2404.14215)
 1. Acikgoz et al. Hippocrates: An Open-Source Framework for Advancing Large Language Models in Healthcare. 2024. [[arxiv]](https://arxiv.org/abs/2404.16621)
-1. Zhang et al. Small Language Models Need Strong Verifiers to Self-Correct Reasoning. 2024. [[arxiv]](https://arxiv.org/abs/2404.17140)
-1. Zhou et al. FREB-TQA: A Fine-Grained Robustness Evaluation Benchmark for Table Question Answering. 2024. [[arxiv]](https://arxiv.org/abs/2404.18585)
+1. Zhang et al. Small Language Models Need Strong Verifiers to Self-Correct Reasoning. ACL 2024 Findings. [[arxiv]](https://arxiv.org/abs/2404.17140)
+1. Zhou et al. FREB-TQA: A Fine-Grained Robustness Evaluation Benchmark for Table Question Answering. NAACL 2024. [[arxiv]](https://arxiv.org/abs/2404.18585)
+1. Xu et al. Large Language Models for Cyber Security: A Systematic Literature Review. 2024. [[arxiv]](https://arxiv.org/abs/2405.04760)
+1. Dammu et al. "They are uncultured": Unveiling Covert Harms and Social Threats in LLM Generated Conversations. 2024. [[arxiv]](https://arxiv.org/abs/2405.05378)
+1. Yi et al. A safety realignment framework via subspace-oriented model fusion for large language models. 2024. [[arxiv]](https://arxiv.org/abs/2405.09055)
+1. Lou et al. SPO: Multi-Dimensional Preference Sequential Alignment With Implicit Reward Modeling. 2024. [[arxiv]](https://arxiv.org/abs/2405.12739)
+1. Zhang et al. Getting More from Less: Large Language Models are Good Spontaneous Multilingual Learners. 2024. [[arxiv]](https://arxiv.org/abs/2405.13816)
+1. Zhang et al. TS-Align: A Teacher-Student Collaborative Framework for Scalable Iterative Finetuning of Large Language Models. 2024. [[arxiv]](https://arxiv.org/abs/2405.20215)
+1. Zihong Chen. Sentence Segmentation and Sentence Punctuation Based on XunziALLM. 2024. [[paper]](https://aclanthology.org/2024.lt4hala-1.30)
+1. Gao et al. The Best of Both Worlds: Toward an Honest and Helpful Large Language Model. 2024. [[arxiv]](https://arxiv.org/abs/2406.00380)
+1. Wang and Song. MARS: Benchmarking the Metaphysical Reasoning Abilities of Language Models with a Multi-task Evaluation Dataset. 2024. [[arxiv]](https://arxiv.org/abs/2406.02106)
+1. Hu et al. Computational Limits of Low-Rank Adaptation (LoRA) for Transformer-Based Models. 2024. [[arxiv]](https://arxiv.org/abs/2406.03136)
+1. Ge et al. Time Sensitive Knowledge Editing through Efficient Finetuning. ACL 2024. [[arxiv]](https://arxiv.org/abs/2406.04496)
+1. Tan et al. Peer Review as A Multi-Turn and Long-Context Dialogue with Role-Based Interactions. 2024. [[arxiv]](https://arxiv.org/abs/2406.05688)
+1. Song et al. Turbo Sparse: Achieving LLM SOTA Performance with Minimal Activated Parameters. 2024. [[arxiv]](https://arxiv.org/abs/2406.05955)
+1. Gu et al. RWKV-CLIP: A Robust Vision-Language Representation Learner. 2024. [[arxiv]](https://arxiv.org/abs/2406.06973)
+1. Chen et al. Advancing Tool-Augmented Large Language Models: Integrating Insights from Errors in Inference Trees. 2024. [[arxiv]](https://arxiv.org/abs/2406.07115)
+1. Zhu et al. Are Large Language Models Good Statisticians?. 2024. [[arxiv]](https://arxiv.org/abs/2406.07815)
+1. Li et al. Know the Unknown: An Uncertainty-Sensitive Method for LLM Instruction Tuning. 2024. [[arxiv]](https://arxiv.org/abs/2406.10099)
+1. Ding et al. IntentionQA: A Benchmark for Evaluating Purchase Intention Comprehension Abilities of Language Models in E-commerce. 2024. [[arxiv]](https://arxiv.org/abs/2406.10173)
+1. He et al. COMMUNITY-CROSS-INSTRUCT: Unsupervised Instruction Generation for Aligning Large Language Models to Online Communities. 2024. [[arxiv]](https://arxiv.org/abs/2406.12074)
+1. Lin et al. FVEL: Interactive Formal Verification Environment with Large Language Models via Theorem Proving. 2024. [[arxiv]](https://arxiv.org/abs/2406.14408)
+1. Treutlein et al. Connecting the Dots: LLMs can Infer and Verbalize Latent Structure from Disparate Training Data. 2024. [[arxiv]](https://arxiv.org/abs/2406.14546)
+1. Feng et al. SS-Bench: A Benchmark for Social Story Generation and Evaluation. 2024. [[arxiv]](https://arxiv.org/abs/2406.15695)
+1. Feng et al. Self-Constructed Context Decompilation with Fined-grained Alignment Enhancement. 2024. [[arxiv]](https://arxiv.org/abs/2406.17233)
+1. Liu et al. Large Language Models for Cuffless Blood Pressure Measurement From Wearable Biosignals. 2024. [[arxiv]](https://arxiv.org/abs/2406.18069)
+1. Iyer et al. Exploring Very Low-Resource Translation with LLMs: The University of Edinburgh's Submission to AmericasNLP 2024 Translation Task. AmericasNLP 2024. [[paper]](https://aclanthology.org/2024.americasnlp-1.25)
+1. Li et al. Calibrating LLMs with Preference Optimization on Thought Trees for Generating Rationale in Science Question Scoring. 2024. [[arxiv]](https://arxiv.org/abs/2406.19949)
+1. Yang et al. Financial Knowledge Large Language Model. 2024. [[arxiv]](https://arxiv.org/abs/2407.00365)
+1. Lin et al. DogeRM: Equipping Reward Models with Domain Knowledge through Model Merging. 2024. [[arxiv]](https://arxiv.org/abs/2407.01470)
+1. Bako et al. Evaluating the Semantic Profiling Abilities of LLMs for Natural Language Utterances in Data Visualization. 2024. [[arxiv]](https://arxiv.org/abs/2407.06129)
+1. Huang et al. RoLoRA: Fine-tuning Rotated Outlier-free LLMs for Effective Weight-Activation Quantization. 2024. [[arxiv]](https://arxiv.org/abs/2407.08044)
+1. Jiang et al. LLM-Collaboration on Automatic Science Journalism for the General Audience. 2024. [[arxiv]](https://arxiv.org/abs/2407.09756)
+1. Inouye et al. Applied Auto-tuning on LoRA Hyperparameters. 2024. [[paper]](https://scholarcommons.scu.edu/cseng_senior/272/)
+1. Qi et al. Research on Tibetan Tourism Viewpoints information generation system based on LLM. 2024. [[arxiv]](https://arxiv.org/abs/2407.13561)
+1. Xu et al. Course-Correction: Safety Alignment Using Synthetic Preferences. 2024. [[arxiv]](https://arxiv.org/abs/2407.16637)
+1. Sun et al. LAMBDA: A Large Model Based Data Agent. 2024. [[arxiv]](https://arxiv.org/abs/2407.17535)
+1. Zhu et al. CollectiveSFT: Scaling Large Language Models for Chinese Medical Benchmark with Collective Instructions in Healthcare. 2024. [[arxiv]](https://arxiv.org/abs/2407.19705)
+1. Yu et al. Correcting Negative Bias in Large Language Models through Negative Attention Score Alignment. 2024. [[arxiv]](https://arxiv.org/abs/2408.00137)
+1. Xie et al. The Power of Personalized Datasets: Advancing Chinese Composition Writing for Elementary School through Targeted Model Fine-Tuning. IALP 2024. [[paper]](https://www.asianlp.sg/conferences/ialp2024/proceedings/papers/IALP2024_P055.pdf)
+1. Liu et al. Instruct-Code-Llama: Improving Capabilities of Language Model in Competition Level Code Generation by Online Judge Feedback. ICIC 2024. [[paper]](https://link.springer.com/chapter/10.1007/978-981-97-5669-8_11)
+1. Wang et al. Cybernetic Sentinels: Unveiling the Impact of Safety Data Selection on Model Security in Supervised Fine-Tuning. ICIC 2024. [[paper]](https://link.springer.com/chapter/10.1007/978-981-97-5669-8_23)
+1. Xia et al. Understanding the Performance and Estimating the Cost of LLM Fine-Tuning. 2024. [[arxiv]](https://arxiv.org/abs/2408.04693)
+1. Zeng et al. Perceive, Reflect, and Plan: Designing LLM Agent for Goal-Directed City Navigation without Instructions. 2024. [[arxiv]](https://arxiv.org/abs/2408.04168)
+1. Xia et al. Using Pre-trained Language Model for Accurate ESG Prediction. FinNLP 2024. [[paper]](https://aclanthology.org/2024.finnlp-2.1/)
+1. Liang et al. I-SHEEP: Self-Alignment of LLM from Scratch through an Iterative Self-Enhancement Paradigm. 2024. [[arxiv]](https://arxiv.org/abs/2408.08072)
+1. Bai et al. Aligning Large Language Model with Direct Multi-Preference Optimization for Recommendation. CIKM 2024. [[paper]](https://dl.acm.org/doi/10.1145/3627673.3679611)
+1. Zhang et al. CPsyCoun: A Report-based Multi-turn Dialogue Reconstruction and Evaluation Framework for Chinese Psychological Counseling. ACL 2024. [[paper]](https://aclanthology.org/2024.findings-acl.830.pdf)
 1. **[StarWhisper](https://github.com/Yu-Yang-Li/StarWhisper)**: A large language model for Astronomy, based on ChatGLM2-6B and Qwen-14B.
 1. **[DISC-LawLLM](https://github.com/FudanDISC/DISC-LawLLM)**: A large language model specialized in Chinese legal domain, based on Baichuan-13B, is capable of retrieving and reasoning on legal knowledge.
 1. **[Sunsimiao](https://github.com/X-D-Lab/Sunsimiao)**: A large language model specialized in Chinese medical domain, based on Baichuan-7B and ChatGLM-6B.
 1. **[CareGPT](https://github.com/WangRongsheng/CareGPT)**: A series of large language models for Chinese medical domain, based on LLaMA2-7B and Baichuan-13B.
 1. **[MachineMindset](https://github.com/PKU-YuanGroup/Machine-Mindset/)**: A series of MBTI Personality large language models, capable of giving any LLM 16 different personality types based on different datasets and training methods.
-1. **[Luminia-13B-v3](https://huggingface.co/Nekochu/Luminia-13B-v3)**: A large language model specialized in generate metadata for stable diffusion. [[🤗Demo]](https://huggingface.co/spaces/Nekochu/Luminia-13B_SD_Prompt)
+1. **[Luminia-13B-v3](https://huggingface.co/Nekochu/Luminia-13B-v3)**: A large language model specialized in generate metadata for stable diffusion. [[demo]](https://huggingface.co/spaces/Nekochu/Luminia-13B_SD_Prompt)
 1. **[Chinese-LLaVA-Med](https://github.com/BUAADreamer/Chinese-LLaVA-Med)**: A multimodal large language model specialized in Chinese medical domain, based on LLaVA-1.5-7B.
-
+1. **[AutoRE](https://github.com/THUDM/AutoRE)**: A document-level relation extraction system based on large language models.
+1. **[NVIDIA RTX AI Toolkit](https://github.com/NVIDIA/RTX-AI-Toolkit)**: SDKs for fine-tuning LLMs on Windows PC for NVIDIA RTX.
+1. **[LazyLLM](https://github.com/LazyAGI/LazyLLM)**: An easy and lazy way for building multi-agent LLMs applications and supports model fine-tuning via LLaMA Factory.
+1. **[RAG-Retrieval](https://github.com/NLPJCL/RAG-Retrieval)**: A full pipeline for RAG retrieval model fine-tuning, inference, and distillation. [[blog]](https://zhuanlan.zhihu.com/p/987727357)
+1. **[360-LLaMA-Factory](https://github.com/Qihoo360/360-LLaMA-Factory)**: A modified library that supports long sequence SFT & DPO using ring attention.
+1. **[Sky-T1](https://novasky-ai.github.io/posts/sky-t1/)**: An o1-like model fine-tuned by NovaSky AI with very small cost.
+1. **[WeClone](https://github.com/xming521/WeClone)**: One-stop solution for creating your digital avatar from chat logs.
+1. **[EmoLLM](https://github.com/SmartFlowAI/EmoLLM)**: A project about large language models (LLMs) and mental health.
 </details>
 
 ## License
 
 This repository is licensed under the [Apache-2.0 License](LICENSE).
 
-Please follow the model licenses to use the corresponding model weights: [Baichuan2](https://huggingface.co/baichuan-inc/Baichuan2-7B-Base/blob/main/Community%20License%20for%20Baichuan%202%20Model.pdf) / [BLOOM](https://huggingface.co/spaces/bigscience/license) / [ChatGLM3](https://github.com/THUDM/ChatGLM3/blob/main/MODEL_LICENSE) / [Command-R](https://cohere.com/c4ai-cc-by-nc-license) / [DeepSeek](https://github.com/deepseek-ai/DeepSeek-LLM/blob/main/LICENSE-MODEL) / [Falcon](https://huggingface.co/tiiuae/falcon-180B/blob/main/LICENSE.txt) / [Gemma](https://ai.google.dev/gemma/terms) / [GLM4](https://huggingface.co/THUDM/glm-4-9b/blob/main/LICENSE) / [InternLM2](https://github.com/InternLM/InternLM#license) / [LLaMA](https://github.com/facebookresearch/llama/blob/main/MODEL_CARD.md) / [LLaMA-2 (LLaVA-1.5)](https://ai.meta.com/llama/license/) / [LLaMA-3](https://llama.meta.com/llama3/license/) / [Mistral](LICENSE) / [OLMo](LICENSE) / [Phi-1.5/2](https://huggingface.co/microsoft/phi-1_5/resolve/main/Research%20License.docx) / [Phi-3](https://huggingface.co/microsoft/Phi-3-mini-4k-instruct/blob/main/LICENSE) / [Qwen](https://github.com/QwenLM/Qwen/blob/main/Tongyi%20Qianwen%20LICENSE%20AGREEMENT) / [StarCoder2](https://huggingface.co/spaces/bigcode/bigcode-model-license-agreement) / [XVERSE](https://github.com/xverse-ai/XVERSE-13B/blob/main/MODEL_LICENSE.pdf) / [Yi](https://huggingface.co/01-ai/Yi-6B/blob/main/LICENSE) / [Yi-1.5](LICENSE) / [Yuan](https://github.com/IEIT-Yuan/Yuan-2.0/blob/main/LICENSE-Yuan)
+Please follow the model licenses to use the corresponding model weights: [Baichuan 2](https://huggingface.co/baichuan-inc/Baichuan2-7B-Base/blob/main/Community%20License%20for%20Baichuan%202%20Model.pdf) / [BLOOM](https://huggingface.co/spaces/bigscience/license) / [ChatGLM3](https://github.com/THUDM/ChatGLM3/blob/main/MODEL_LICENSE) / [Command R](https://cohere.com/c4ai-cc-by-nc-license) / [DeepSeek](https://github.com/deepseek-ai/DeepSeek-LLM/blob/main/LICENSE-MODEL) / [Falcon](https://huggingface.co/tiiuae/falcon-180B/blob/main/LICENSE.txt) / [Gemma](https://ai.google.dev/gemma/terms) / [GLM-4](https://huggingface.co/THUDM/glm-4-9b/blob/main/LICENSE) / [GPT-2](https://github.com/openai/gpt-2/blob/master/LICENSE) / [Granite](LICENSE) / [Index](https://huggingface.co/IndexTeam/Index-1.9B/blob/main/LICENSE) / [InternLM](https://github.com/InternLM/InternLM#license) / [Llama](https://github.com/facebookresearch/llama/blob/main/MODEL_CARD.md) / [Llama 2](https://ai.meta.com/llama/license/) / [Llama 3](https://llama.meta.com/llama3/license/) / [Llama 4](https://github.com/meta-llama/llama-models/blob/main/models/llama4/LICENSE) / [MiniCPM](https://github.com/OpenBMB/MiniCPM/blob/main/MiniCPM%20Model%20License.md) / [Mistral/Mixtral/Pixtral](LICENSE) / [OLMo](LICENSE) / [Phi-1.5/Phi-2](https://huggingface.co/microsoft/phi-1_5/resolve/main/Research%20License.docx) / [Phi-3/Phi-4](https://huggingface.co/microsoft/Phi-3-mini-4k-instruct/blob/main/LICENSE) / [Qwen](https://github.com/QwenLM/Qwen/blob/main/Tongyi%20Qianwen%20LICENSE%20AGREEMENT) / [Skywork](https://huggingface.co/Skywork/Skywork-13B-base/blob/main/Skywork%20Community%20License.pdf) / [StarCoder 2](https://huggingface.co/spaces/bigcode/bigcode-model-license-agreement) / [TeleChat2](https://huggingface.co/Tele-AI/telechat-7B/blob/main/TeleChat%E6%A8%A1%E5%9E%8B%E7%A4%BE%E5%8C%BA%E8%AE%B8%E5%8F%AF%E5%8D%8F%E8%AE%AE.pdf) / [XVERSE](https://github.com/xverse-ai/XVERSE-13B/blob/main/MODEL_LICENSE.pdf) / [Yi](https://huggingface.co/01-ai/Yi-6B/blob/main/LICENSE) / [Yi-1.5](LICENSE) / [Yuan 2](https://github.com/IEIT-Yuan/Yuan-2.0/blob/main/LICENSE-Yuan)
 
 ## Citation
 
 If this work is helpful, please kindly cite as:
 
 ```bibtex
-@article{zheng2024llamafactory,
+@inproceedings{zheng2024llamafactory,
   title={LlamaFactory: Unified Efficient Fine-Tuning of 100+ Language Models},
-  author={Yaowei Zheng and Richong Zhang and Junhao Zhang and Yanhan Ye and Zheyan Luo and Yongqiang Ma},
-  journal={arXiv preprint arXiv:2403.13372},
+  author={Yaowei Zheng and Richong Zhang and Junhao Zhang and Yanhan Ye and Zheyan Luo and Zhangchi Feng and Yongqiang Ma},
+  booktitle={Proceedings of the 62nd Annual Meeting of the Association for Computational Linguistics (Volume 3: System Demonstrations)},
+  address={Bangkok, Thailand},
+  publisher={Association for Computational Linguistics},
   year={2024},
   url={http://arxiv.org/abs/2403.13372}
 }

README_zh.md

@@ -1,44 +1,87 @@
 ![# LLaMA Factory](assets/logo.png)
 
 [![GitHub Repo stars](https://img.shields.io/github/stars/hiyouga/LLaMA-Factory?style=social)](https://github.com/hiyouga/LLaMA-Factory/stargazers)
-[![GitHub Code License](https://img.shields.io/github/license/hiyouga/LLaMA-Factory)](LICENSE)
 [![GitHub last commit](https://img.shields.io/github/last-commit/hiyouga/LLaMA-Factory)](https://github.com/hiyouga/LLaMA-Factory/commits/main)
+[![GitHub contributors](https://img.shields.io/github/contributors/hiyouga/LLaMA-Factory?color=orange)](https://github.com/hiyouga/LLaMA-Factory/graphs/contributors)
+[![GitHub workflow](https://github.com/hiyouga/LLaMA-Factory/actions/workflows/tests.yml/badge.svg)](https://github.com/hiyouga/LLaMA-Factory/actions/workflows/tests.yml)
 [![PyPI](https://img.shields.io/pypi/v/llamafactory)](https://pypi.org/project/llamafactory/)
-[![Citation](https://img.shields.io/badge/citation-44-green)](#使用了-llama-factory-的项目)
-[![GitHub pull request](https://img.shields.io/badge/PRs-welcome-blue)](https://github.com/hiyouga/LLaMA-Factory/pulls)
-[![Discord](https://dcbadge.vercel.app/api/server/rKfvV9r9FK?compact=true&style=flat)](https://discord.gg/rKfvV9r9FK)
+[![Citation](https://img.shields.io/badge/citation-614-green)](https://scholar.google.com/scholar?cites=12620864006390196564)
+[![Docker Pulls](https://img.shields.io/docker/pulls/hiyouga/llamafactory)](https://hub.docker.com/r/hiyouga/llamafactory/tags)
+
 [![Twitter](https://img.shields.io/twitter/follow/llamafactory_ai)](https://twitter.com/llamafactory_ai)
+[![Discord](https://dcbadge.vercel.app/api/server/rKfvV9r9FK?compact=true&style=flat)](https://discord.gg/rKfvV9r9FK)
+[![GitCode](https://gitcode.com/zhengyaowei/LLaMA-Factory/star/badge.svg)](https://gitcode.com/zhengyaowei/LLaMA-Factory)
+
 [![Open in Colab](https://colab.research.google.com/assets/colab-badge.svg)](https://colab.research.google.com/drive/1d5KQtbemerlSDSxZIfAaWXhKr30QypiK?usp=sharing)
 [![Open in DSW](https://gallery.pai-ml.com/assets/open-in-dsw.svg)](https://gallery.pai-ml.com/#/preview/deepLearning/nlp/llama_factory)
-[![Spaces](https://img.shields.io/badge/🤗-Open%20in%20Spaces-blue)](https://huggingface.co/spaces/hiyouga/LLaMA-Board)
-[![Studios](https://img.shields.io/badge/ModelScope-Open%20in%20Studios-blue)](https://modelscope.cn/studios/hiyouga/LLaMA-Board)
+[![Open in Alaya](assets/alaya_new.svg)](https://docs.alayanew.com/docs/documents/newActivities/llamafactory/?utm_source=LLaMA-Factory)
+[![Open in Spaces](https://img.shields.io/badge/🤗-Open%20in%20Spaces-blue)](https://huggingface.co/spaces/hiyouga/LLaMA-Board)
+[![Open in Studios](https://img.shields.io/badge/ModelScope-Open%20in%20Studios-blue)](https://modelscope.cn/studios/hiyouga/LLaMA-Board)
+[![Open in Novita](https://img.shields.io/badge/Novita-Deploy%20Template-blue)](https://novita.ai/templates-library/105981?sharer=88115474-394e-4bda-968e-b88e123d0c47)
 
-[![GitHub Tread](https://trendshift.io/api/badge/repositories/4535)](https://trendshift.io/repositories/4535)
+### 获得[亚马逊](https://aws.amazon.com/cn/blogs/china/a-one-stop-code-free-model-fine-tuning-deployment-platform-based-on-sagemaker-and-llama-factory/)、[英伟达](https://developer.nvidia.cn/rtx/ai-toolkit)、[阿里云](https://help.aliyun.com/zh/pai/use-cases/fine-tune-a-llama-3-model-with-llama-factory)等的应用。
 
-👋 加入我们的[微信群](assets/wechat.jpg)。
+<div align="center" markdown="1">
+
+### 赞助商 ❤️
+
+<a href="https://warp.dev/llama-factory">
+    <img alt="Warp sponsorship" width="400" src="https://github.com/user-attachments/assets/ab8dd143-b0fd-4904-bdc5-dd7ecac94eae">
+</a>
+
+#### [Warp，面向开发者的智能终端](https://warp.dev/llama-factory)
+
+[适用于 MacOS、Linux 和 Windows](https://warp.dev/llama-factory)
+
+----
+
+### 使用零代码[命令行](#快速开始)与 [Web UI](#llama-board-可视化微调由-gradio-驱动) 轻松微调百余种大模型
+
+![GitHub Trend](https://trendshift.io/api/badge/repositories/4535)
+
+</div>
+
+👋 加入我们的[微信群](assets/wechat.jpg)、[NPU 用户群](assets/wechat_npu.jpg)或 [九章智算云算力优惠群](assets/wechat_alaya.png)。
 
 \[ [English](README.md) | 中文 \]
 
 **微调大模型可以像这样轻松…**
 
-https://github.com/hiyouga/LLaMA-Factory/assets/16256802/ec36a9dd-37f4-4f72-81bd-d76c6d0a6594
+https://github.com/user-attachments/assets/43b700c6-a178-41db-b1f8-8190a5d3fcfc
 
 选择你的打开方式：
 
-- **Colab**：https://colab.research.google.com/drive/1d5KQtbemerlSDSxZIfAaWXhKr30QypiK?usp=sharing
-- **PAI-DSW**: https://gallery.pai-ml.com/#/preview/deepLearning/nlp/llama_factory
+- **入门教程**：https://zhuanlan.zhihu.com/p/695287607
+- **框架文档**：https://llamafactory.readthedocs.io/zh-cn/latest/
+- **框架文档（昇腾 NPU）**：https://ascend.github.io/docs/sources/llamafactory/
+- **Colab（免费）**：https://colab.research.google.com/drive/1d5KQtbemerlSDSxZIfAaWXhKr30QypiK?usp=sharing
 - **本地机器**：请见[如何使用](#如何使用)
+- **PAI-DSW（免费试用）**：https://gallery.pai-ml.com/#/preview/deepLearning/nlp/llama_factory
+- **九章智算云（算力优惠活动）**：https://docs.alayanew.com/docs/documents/useGuide/LLaMAFactory/mutiple/?utm_source=LLaMA-Factory
+
+> [!NOTE]
+> 除上述链接以外的其他网站均为未经许可的第三方网站，请小心甄别。
 
 ## 目录
 
 - [项目特色](#项目特色)
-- [性能指标](#性能指标)
+- [官方博客](#官方博客)
 - [更新日志](#更新日志)
 - [模型](#模型)
 - [训练方法](#训练方法)
 - [数据集](#数据集)
 - [软硬件依赖](#软硬件依赖)
 - [如何使用](#如何使用)
+  - [安装 LLaMA Factory](#安装-llama-factory)
+  - [数据准备](#数据准备)
+  - [快速开始](#快速开始)
+  - [LLaMA Board 可视化微调](#llama-board-可视化微调由-gradio-驱动)
+  - [构建 Docker](#构建-docker)
+  - [利用 vLLM 部署 OpenAI API](#利用-vllm-部署-openai-api)
+  - [从魔搭社区下载](#从魔搭社区下载)
+  - [从魔乐社区下载](#从魔乐社区下载)
+  - [使用 W&B 面板](#使用-wb-面板)
+  - [使用 SwanLab 面板](#使用-swanlab-面板)
 - [使用了 LLaMA Factory 的项目](#使用了-llama-factory-的项目)
 - [协议](#协议)
 - [引用](#引用)
@@ -46,40 +89,96 @@ https://github.com/hiyouga/LLaMA-Factory/assets/16256802/ec36a9dd-37f4-4f72-81bd
 
 ## 项目特色
 
-- **多种模型**：LLaMA、LLaVA、Mistral、Mixtral-MoE、Qwen、Yi、Gemma、Baichuan、ChatGLM、Phi 等等。
+- **多种模型**：LLaMA、LLaVA、Mistral、Mixtral-MoE、Qwen、Qwen2-VL、DeepSeek、Yi、Gemma、ChatGLM、Phi 等等。
 - **集成方法**：（增量）预训练、（多模态）指令监督微调、奖励模型训练、PPO 训练、DPO 训练、KTO 训练、ORPO 训练等等。
-- **多种精度**：32 比特全参数微调、16 比特冻结微调、16 比特 LoRA 微调和基于 AQLM/AWQ/GPTQ/LLM.int8 的 2/4/8 比特 QLoRA 微调。
-- **先进算法**：GaLore、BAdam、DoRA、LongLoRA、LLaMA Pro、Mixture-of-Depths、LoRA+、LoftQ、PiSSA 和 Agent 微调。
-- **实用技巧**：FlashAttention-2、Unsloth、RoPE scaling、NEFTune 和 rsLoRA。
-- **实验监控**：LlamaBoard、TensorBoard、Wandb、MLflow 等等。
-- **极速推理**：基于 vLLM 的 OpenAI 风格 API、浏览器界面和命令行接口。
+- **多种精度**：16 比特全参数微调、冻结微调、LoRA 微调和基于 AQLM/AWQ/GPTQ/LLM.int8/HQQ/EETQ 的 2/3/4/5/6/8 比特 QLoRA 微调。
+- **先进算法**：[GaLore](https://github.com/jiaweizzhao/GaLore)、[BAdam](https://github.com/Ledzy/BAdam)、[APOLLO](https://github.com/zhuhanqing/APOLLO)、[Adam-mini](https://github.com/zyushun/Adam-mini)、[Muon](https://github.com/KellerJordan/Muon)、DoRA、LongLoRA、LLaMA Pro、Mixture-of-Depths、LoRA+、LoftQ 和 PiSSA。
+- **实用技巧**：[FlashAttention-2](https://github.com/Dao-AILab/flash-attention)、[Unsloth](https://github.com/unslothai/unsloth)、[Liger Kernel](https://github.com/linkedin/Liger-Kernel)、RoPE scaling、NEFTune 和 rsLoRA。
+- **广泛任务**：多轮对话、工具调用、图像理解、视觉定位、视频识别和语音理解等等。
+- **实验监控**：LlamaBoard、TensorBoard、Wandb、MLflow、[SwanLab](https://github.com/SwanHubX/SwanLab) 等等。
+- **极速推理**：基于 [vLLM](https://github.com/vllm-project/vllm) 或 [SGLang](https://github.com/sgl-project/sglang) 的 OpenAI 风格 API、浏览器界面和命令行接口。
 
-## 性能指标
+### 最新模型的 Day-N 微调适配
 
-与 ChatGLM 官方的 [P-Tuning](https://github.com/THUDM/ChatGLM2-6B/tree/main/ptuning) 微调相比，LLaMA Factory 的 LoRA 微调提供了 **3.7 倍**的加速比，同时在广告文案生成任务上取得了更高的 Rouge 分数。结合 4 比特量化技术，LLaMA Factory 的 QLoRA 微调进一步降低了 GPU 显存消耗。
+| 适配时间      | 模型名称                                                       |
+| ------------ | ------------------------------------------------------------ |
+| Day 0        | Qwen3 / Qwen2.5-VL / Gemma 3 / InternLM 3 / MiniCPM-o-2.6    |
+| Day 1        | Llama 3 / GLM-4 / Mistral Small / PaliGemma2 / Llama 4       |
 
-![benchmark](assets/benchmark.svg)
+## 官方博客
 
-<details><summary>变量定义</summary>
+- [使用 LLaMA-Factory 微调 Qwen2.5-VL 实现自动驾驶场景微调](https://docs.alayanew.com/docs/documents/useGuide/LLaMAFactory/mutiple/?utm_source=LLaMA-Factory)（中文）
+- [通过亚马逊 SageMaker HyperPod 上的 LLaMA-Factory 增强多模态模型银行文档的视觉信息提取](https://aws.amazon.com/cn/blogs/machine-learning/how-apoidea-group-enhances-visual-information-extraction-from-banking-documents-with-multimodal-models-using-llama-factory-on-amazon-sagemaker-hyperpod/)（英文）
+- [Easy Dataset × LLaMA Factory: 让大模型高效学习领域知识](https://buaa-act.feishu.cn/wiki/KY9xwTGs1iqHrRkjXBwcZP9WnL9)（中文）
 
-- **Training Speed**: 训练阶段每秒处理的样本数量。（批处理大小=4，截断长度=1024）
-- **Rouge Score**: [广告文案生成](https://aclanthology.org/D19-1321.pdf)任务验证集上的 Rouge-2 分数。（批处理大小=4，截断长度=1024）
-- **GPU Memory**: 4 比特量化训练的 GPU 显存峰值。（批处理大小=1，截断长度=1024）
-- 我们在 ChatGLM 的 P-Tuning 中采用 `pre_seq_len=128`，在 LLaMA Factory 的 LoRA 微调中采用 `lora_rank=32`。
+<details><summary>全部博客</summary>
+
+- [LLaMA Factory：微调 DeepSeek-R1-Distill-Qwen-7B 模型实现新闻标题分类器](https://gallery.pai-ml.com/#/preview/deepLearning/nlp/llama_factory_deepseek_r1_distill_7b)（中文）
+- [基于 Amazon SageMaker 和 LLaMA-Factory 打造一站式无代码模型微调部署平台 Model Hub](https://aws.amazon.com/cn/blogs/china/a-one-stop-code-free-model-fine-tuning-deployment-platform-based-on-sagemaker-and-llama-factory/)（中文）
+- [LLaMA Factory 多模态微调实践：微调 Qwen2-VL 构建文旅大模型](https://gallery.pai-ml.com/#/preview/deepLearning/nlp/llama_factory_qwen2vl)（中文）
+- [LLaMA Factory：微调LLaMA3模型实现角色扮演](https://gallery.pai-ml.com/#/preview/deepLearning/nlp/llama_factory)（中文）
 
 </details>
 
 ## 更新日志
 
+[25/04/28] 我们支持了 **[Qwen3](https://qwenlm.github.io/blog/qwen3/)** 系列模型的微调。
+
+[25/04/21] 我们支持了 **[Muon](https://github.com/KellerJordan/Muon)** 优化器。详细用法请参照 [examples](examples/README_zh.md)。感谢 [@tianshijing](https://github.com/tianshijing) 的 PR。
+
+[25/04/16] 我们支持了 **[InternVL3](https://huggingface.co/OpenGVLab/InternVL3-8B)** 模型的微调。查看 [PR #7258](https://github.com/hiyouga/LLaMA-Factory/pull/7258) 以使用。
+
+[25/04/14] 我们支持了 **[GLM-Z1](https://huggingface.co/THUDM/GLM-Z1-9B-0414)** 和 **[Kimi-VL](https://huggingface.co/moonshotai/Kimi-VL-A3B-Instruct)** 模型的微调。
+
+[25/04/06] 我们支持了 **[Llama 4](https://ai.meta.com/blog/llama-4-multimodal-intelligence/)** 模型的微调。查看 [PR #7611](https://github.com/hiyouga/LLaMA-Factory/pull/7611) 以使用。
+
+<details><summary>展开日志</summary>
+
+[25/03/31] 我们支持了 **[Qwen2.5 Omni](https://qwenlm.github.io/blog/qwen2.5-omni/)** 模型的微调。查看 [PR #7537](https://github.com/hiyouga/LLaMA-Factory/pull/7537) 以使用。
+
+[25/03/15] 我们支持了 **[SGLang](https://github.com/sgl-project/sglang)** 推理后端，请使用 `infer_backend: sglang` 启用。
+
+[25/03/12] 我们支持了 **[Gemma 3](https://huggingface.co/blog/gemma3)** 模型的微调。
+
+[25/02/24] 我们宣布开源 **[EasyR1](https://github.com/hiyouga/EasyR1)**，一个高效可扩展的多模态强化学习框架，支持高效的 GRPO 训练。
+
+[25/02/11] 我们支持了在导出模型时保存 **[Ollama](https://github.com/ollama/ollama)** 配置文件。详细用法请参照 [examples](examples/README_zh.md)。
+
+[25/02/05] 我们支持了在语音理解任务上微调 **[Qwen2-Audio](Qwen/Qwen2-Audio-7B-Instruct)** 和 **[MiniCPM-o-2.6](https://huggingface.co/openbmb/MiniCPM-o-2_6)** 模型。
+
+[25/01/31] 我们支持了 **[DeepSeek-R1](https://huggingface.co/deepseek-ai/DeepSeek-R1)** 和 **[Qwen2.5-VL](https://huggingface.co/Qwen/Qwen2.5-VL-7B-Instruct)** 模型的微调。
+
+[25/01/15] 我们支持了 **[APOLLO](https://arxiv.org/abs/2412.05270)** 优化器。详细用法请参照 [examples](examples/README_zh.md)。
+
+[25/01/14] 我们支持了 **[MiniCPM-o-2.6](https://huggingface.co/openbmb/MiniCPM-o-2_6)** 和 **[MiniCPM-V-2.6](https://huggingface.co/openbmb/MiniCPM-V-2_6)** 模型的微调。 感谢 [@BUAADreamer](https://github.com/BUAADreamer) 的 PR.
+
+[25/01/14] 我们支持了 **[InternLM 3](https://huggingface.co/collections/internlm/)** 模型的微调。感谢 [@hhaAndroid](https://github.com/hhaAndroid) 的 PR。
+
+[25/01/10] 我们支持了 **[Phi-4](https://huggingface.co/microsoft/phi-4)** 模型的微调。
+
+[24/12/21] 我们支持了使用 **[SwanLab](https://github.com/SwanHubX/SwanLab)** 跟踪与可视化实验。详细用法请参考 [此部分](#使用-swanlab-面板)。
+
+[24/11/27] 我们支持了 **[Skywork-o1](https://huggingface.co/Skywork/Skywork-o1-Open-Llama-3.1-8B)** 模型的微调和 **[OpenO1](https://huggingface.co/datasets/O1-OPEN/OpenO1-SFT)** 数据集。
+
+[24/10/09] 我们支持了从 **[魔乐社区](https://modelers.cn/models)** 下载预训练模型和数据集。详细用法请参照 [此教程](#从魔乐社区下载)。
+
+[24/09/19] 我们支持了 **[Qwen2.5](https://qwenlm.github.io/blog/qwen2.5/)** 模型的微调。
+
+[24/08/30] 我们支持了 **[Qwen2-VL](https://qwenlm.github.io/blog/qwen2-vl/)** 模型的微调。感谢 [@simonJJJ](https://github.com/simonJJJ) 的 PR。
+
+[24/08/27] 我们支持了 **[Liger Kernel](https://github.com/linkedin/Liger-Kernel)**。请使用 `enable_liger_kernel: true` 来加速训练。
+
+[24/08/09] 我们支持了 **[Adam-mini](https://github.com/zyushun/Adam-mini)** 优化器。详细用法请参照 [examples](examples/README_zh.md)。感谢 [@relic-yuexi](https://github.com/relic-yuexi) 的 PR。
+
+[24/07/04] 我们支持了[无污染打包训练](https://github.com/MeetKai/functionary/tree/main/functionary/train/packing)。请使用 `neat_packing: true` 参数。感谢 [@chuan298](https://github.com/chuan298) 的 PR。
+
 [24/06/16] 我们支持了 **[PiSSA](https://arxiv.org/abs/2404.02948)** 算法。详细用法请参照 [examples](examples/README_zh.md)。
 
 [24/06/07] 我们支持了 **[Qwen2](https://qwenlm.github.io/blog/qwen2/)** 和 **[GLM-4](https://github.com/THUDM/GLM-4)** 模型的微调。
 
 [24/05/26] 我们支持了 **[SimPO](https://arxiv.org/abs/2405.14734)** 偏好对齐算法。详细用法请参照 [examples](examples/README_zh.md)。
 
-<details><summary>展开日志</summary>
-
-[24/05/20] 我们支持了 **PaliGemma** 系列模型的微调。注意 PaliGemma 是预训练模型，你需要使用 `gemma` 模板进行微调使其获得对话能力。
+[24/05/20] 我们支持了 **PaliGemma** 系列模型的微调。注意 PaliGemma 是预训练模型，你需要使用 `paligemma` 模板进行微调使其获得对话能力。
 
 [24/05/18] 我们支持了 **[KTO](https://arxiv.org/abs/2402.01306)** 偏好对齐算法。详细用法请参照 [examples](examples/README_zh.md)。
 
@@ -91,7 +190,7 @@ https://github.com/hiyouga/LLaMA-Factory/assets/16256802/ec36a9dd-37f4-4f72-81bd
 
 [24/04/21] 我们基于 [AstraMindAI 的仓库](https://github.com/astramind-ai/Mixture-of-depths)支持了 **[混合深度训练](https://arxiv.org/abs/2404.02258)**。详细用法请参照 [examples](examples/README_zh.md)。
 
-[24/04/16] 我们支持了 **[BAdam](https://arxiv.org/abs/2404.02827)**。详细用法请参照 [examples](examples/README_zh.md)。
+[24/04/16] 我们支持了 **[BAdam](https://arxiv.org/abs/2404.02827)** 优化器。详细用法请参照 [examples](examples/README_zh.md)。
 
 [24/04/16] 我们支持了 **[unsloth](https://github.com/unslothai/unsloth)** 的长序列训练（24GB 可训练 Llama-2-7B-56k）。该方法相比 FlashAttention-2 提供了 **117%** 的训练速度和 **50%** 的显存节约。更多数据请见[此页面](https://github.com/hiyouga/LLaMA-Factory/wiki/Performance-comparison)。
 
@@ -103,7 +202,7 @@ https://github.com/hiyouga/LLaMA-Factory/assets/16256802/ec36a9dd-37f4-4f72-81bd
 
 [24/03/13] 我们支持了 **[LoRA+](https://arxiv.org/abs/2402.12354)**。详细用法请参照 [examples](examples/README_zh.md)。
 
-[24/03/07] 我们支持了梯度低秩投影（**[GaLore](https://arxiv.org/abs/2403.03507)**）算法。详细用法请参照 [examples](examples/README_zh.md)。
+[24/03/07] 我们支持了 **[GaLore](https://arxiv.org/abs/2403.03507)** 优化器。详细用法请参照 [examples](examples/README_zh.md)。
 
 [24/03/07] 我们集成了 **[vLLM](https://github.com/vllm-project/vllm)** 以实现极速并发推理。请使用 `infer_backend: vllm` 来获得 **270%** 的推理速度。
 
@@ -149,42 +248,74 @@ https://github.com/hiyouga/LLaMA-Factory/assets/16256802/ec36a9dd-37f4-4f72-81bd
 
 </details>
 
+> [!TIP]
+> 如果您无法使用最新的功能，请尝试重新拉取代码并再次安装 LLaMA-Factory。
+
 ## 模型
 
-| 模型名                                                    | 模型大小                          | Template  |
-| --------------------------------------------------------- | -------------------------------- | --------- |
+| 模型名                                                             | 参数量                            | Template            |
+| ----------------------------------------------------------------- | -------------------------------- | ------------------- |
 | [Baichuan 2](https://huggingface.co/baichuan-inc)                 | 7B/13B                           | baichuan2           |
-| [BLOOM](https://huggingface.co/bigscience)                | 560M/1.1B/1.7B/3B/7.1B/176B      | -         |
-| [BLOOMZ](https://huggingface.co/bigscience)               | 560M/1.1B/1.7B/3B/7.1B/176B      | -         |
+| [BLOOM/BLOOMZ](https://huggingface.co/bigscience)                 | 560M/1.1B/1.7B/3B/7.1B/176B      | -                   |
 | [ChatGLM3](https://huggingface.co/THUDM)                          | 6B                               | chatglm3            |
-| [Command-R](https://huggingface.co/CohereForAI)           | 35B/104B                         | cohere    |
+| [Command R](https://huggingface.co/CohereForAI)                   | 35B/104B                         | cohere              |
 | [DeepSeek (Code/MoE)](https://huggingface.co/deepseek-ai)         | 7B/16B/67B/236B                  | deepseek            |
+| [DeepSeek 2.5/3](https://huggingface.co/deepseek-ai)              | 236B/671B                        | deepseek3           |
+| [DeepSeek R1 (Distill)](https://huggingface.co/deepseek-ai)       | 1.5B/7B/8B/14B/32B/70B/671B      | deepseekr1          |
 | [Falcon](https://huggingface.co/tiiuae)                           | 7B/11B/40B/180B                  | falcon              |
-| [Gemma/CodeGemma](https://huggingface.co/google)          | 2B/7B                            | gemma     |
-| [GLM4](https://huggingface.co/THUDM)                      | 9B                               | glm4      |
-| [InternLM2](https://huggingface.co/internlm)              | 7B/20B                           | intern2   |
-| [LLaMA](https://github.com/facebookresearch/llama)        | 7B/13B/33B/65B                   | -         |
-| [LLaMA-2](https://huggingface.co/meta-llama)              | 7B/13B/70B                       | llama2    |
-| [LLaMA-3](https://huggingface.co/meta-llama)              | 8B/70B                           | llama3    |
-| [LLaVA-1.5](https://huggingface.co/llava-hf)              | 7B/13B                           | vicuna    |
+| [Gemma/Gemma 2/CodeGemma](https://huggingface.co/google)          | 2B/7B/9B/27B                     | gemma               |
+| [Gemma 3](https://huggingface.co/google)                          | 1B/4B/12B/27B                    | gemma3/gemma (1B)   |
+| [GLM-4/GLM-4-0414/GLM-Z1](https://huggingface.co/THUDM)           | 9B/32B                           | glm4/glmz1          |
+| [GPT-2](https://huggingface.co/openai-community)                  | 0.1B/0.4B/0.8B/1.5B              | -                   |
+| [Granite 3.0-3.3](https://huggingface.co/ibm-granite)             | 1B/2B/3B/8B                      | granite3            |
+| [Hunyuan](https://huggingface.co/tencent/)                        | 7B                               | hunyuan             |
+| [Index](https://huggingface.co/IndexTeam)                         | 1.9B                             | index               |
+| [InternLM 2-3](https://huggingface.co/internlm)                   | 7B/8B/20B                        | intern2             |
+| [InternVL 2.5-3](https://huggingface.co/OpenGVLab)                | 1B/2B/8B/14B/38B/78B             | intern_vl           |
+| [Kimi-VL](https://huggingface.co/moonshotai)                      | 16B                              | kimi_vl             |
+| [Llama](https://github.com/facebookresearch/llama)                | 7B/13B/33B/65B                   | -                   |
+| [Llama 2](https://huggingface.co/meta-llama)                      | 7B/13B/70B                       | llama2              |
+| [Llama 3-3.3](https://huggingface.co/meta-llama)                  | 1B/3B/8B/70B                     | llama3              |
+| [Llama 4](https://huggingface.co/meta-llama)                      | 109B/402B                        | llama4              |
+| [Llama 3.2 Vision](https://huggingface.co/meta-llama)             | 11B/90B                          | mllama              |
+| [LLaVA-1.5](https://huggingface.co/llava-hf)                      | 7B/13B                           | llava               |
+| [LLaVA-NeXT](https://huggingface.co/llava-hf)                     | 7B/8B/13B/34B/72B/110B           | llava_next          |
+| [LLaVA-NeXT-Video](https://huggingface.co/llava-hf)               | 7B/34B                           | llava_next_video    |
+| [MiMo](https://huggingface.co/XiaomiMiMo)                         | 7B                               | mimo                |
+| [MiniCPM](https://huggingface.co/openbmb)                         | 0.5B/1B/2B/4B/8B                 | cpm/cpm3/cpm4       |
+| [MiniCPM-o-2.6/MiniCPM-V-2.6](https://huggingface.co/openbmb)     | 8B                               | minicpm_o/minicpm_v |
+| [Ministral/Mistral-Nemo](https://huggingface.co/mistralai)        | 8B/12B                           | ministral           |
 | [Mistral/Mixtral](https://huggingface.co/mistralai)               | 7B/8x7B/8x22B                    | mistral             |
+| [Mistral Small](https://huggingface.co/mistralai)                 | 24B                              | mistral_small       |
 | [OLMo](https://huggingface.co/allenai)                            | 1B/7B                            | -                   |
-| [PaliGemma](https://huggingface.co/google)                | 3B                               | gemma     |
-| [Phi-1.5/2](https://huggingface.co/microsoft)             | 1.3B/2.7B                        | -         |
-| [Phi-3](https://huggingface.co/microsoft)                 | 4B/7B/14B                        | phi       |
-| [Qwen](https://huggingface.co/Qwen)                       | 1.8B/7B/14B/72B                  | qwen      |
-| [Qwen1.5 (Code/MoE)](https://huggingface.co/Qwen)         | 0.5B/1.8B/4B/7B/14B/32B/72B/110B | qwen      |
-| [Qwen2 (MoE)](https://huggingface.co/Qwen)                | 0.5B/1.5B/7B/57B/72B             | qwen      |
+| [PaliGemma/PaliGemma2](https://huggingface.co/google)             | 3B/10B/28B                       | paligemma           |
+| [Phi-1.5/Phi-2](https://huggingface.co/microsoft)                 | 1.3B/2.7B                        | -                   |
+| [Phi-3/Phi-3.5](https://huggingface.co/microsoft)                 | 4B/14B                           | phi                 |
+| [Phi-3-small](https://huggingface.co/microsoft)                   | 7B                               | phi_small           |
+| [Phi-4](https://huggingface.co/microsoft)                         | 14B                              | phi4                |
+| [Pixtral](https://huggingface.co/mistralai)                       | 12B                              | pixtral             |
+| [Qwen (1-2.5) (Code/Math/MoE/QwQ)](https://huggingface.co/Qwen)   | 0.5B/1.5B/3B/7B/14B/32B/72B/110B | qwen                |
+| [Qwen3 (MoE)](https://huggingface.co/Qwen)                        | 0.6B/1.7B/4B/8B/14B/32B/235B     | qwen3               |
+| [Qwen2-Audio](https://huggingface.co/Qwen)                        | 7B                               | qwen2_audio         |
+| [Qwen2.5-Omni](https://huggingface.co/Qwen)                       | 3B/7B                            | qwen2_omni          |
+| [Qwen2-VL/Qwen2.5-VL/QVQ](https://huggingface.co/Qwen)            | 2B/3B/7B/32B/72B                 | qwen2_vl            |
+| [Seed Coder](https://huggingface.co/ByteDance-Seed)               | 8B                               | seed_coder          |
+| [Skywork o1](https://huggingface.co/Skywork)                      | 8B                               | skywork_o1          |
 | [StarCoder 2](https://huggingface.co/bigcode)                     | 3B/7B/15B                        | -                   |
+| [TeleChat2](https://huggingface.co/Tele-AI)                       | 3B/7B/35B/115B                   | telechat2           |
 | [XVERSE](https://huggingface.co/xverse)                           | 7B/13B/65B                       | xverse              |
-| [Yi (1/1.5)](https://huggingface.co/01-ai)                | 6B/9B/34B                        | yi        |
+| [Yi/Yi-1.5 (Code)](https://huggingface.co/01-ai)                  | 1.5B/6B/9B/34B                   | yi                  |
 | [Yi-VL](https://huggingface.co/01-ai)                             | 6B/34B                           | yi_vl               |
-| [Yuan](https://huggingface.co/IEITYuan)                   | 2B/51B/102B                      | yuan      |
+| [Yuan 2](https://huggingface.co/IEITYuan)                         | 2B/51B/102B                      | yuan                |
 
 > [!NOTE]
 > 对于所有“基座”（Base）模型，`template` 参数可以是 `default`, `alpaca`, `vicuna` 等任意值。但“对话”（Instruct/Chat）模型请务必使用**对应的模板**。
 >
 > 请务必在训练和推理时采用**完全一致**的模板。
+>
+> \*：您需要从 main 分支安装 `transformers` 并使用 `DISABLE_VERSION_CHECK=1` 来跳过版本检查。
+>
+> \*\*：您需要安装特定版本的 `transformers` 以使用该模型。
 
 项目所支持模型的完整列表请参阅 [constants.py](src/llamafactory/extras/constants.py)。
 
@@ -193,7 +324,7 @@ https://github.com/hiyouga/LLaMA-Factory/assets/16256802/ec36a9dd-37f4-4f72-81bd
 ## 训练方法
 
 | 方法                   |     全参数训练      |    部分参数训练     |       LoRA         |       QLoRA        |
-| ---------------------- | ------------------ | ------------------ | ------------------ | ------------------ |
+| --------------------- | ------------------ | ------------------ | ------------------ | ------------------ |
 | 预训练                 | :white_check_mark: | :white_check_mark: | :white_check_mark: | :white_check_mark: |
 | 指令监督微调            | :white_check_mark: | :white_check_mark: | :white_check_mark: | :white_check_mark: |
 | 奖励模型训练            | :white_check_mark: | :white_check_mark: | :white_check_mark: | :white_check_mark: |
@@ -203,6 +334,9 @@ https://github.com/hiyouga/LLaMA-Factory/assets/16256802/ec36a9dd-37f4-4f72-81bd
 | ORPO 训练              | :white_check_mark: | :white_check_mark: | :white_check_mark: | :white_check_mark: |
 | SimPO 训练             | :white_check_mark: | :white_check_mark: | :white_check_mark: | :white_check_mark: |
 
+> [!TIP]
+> 有关 PPO 的实现细节，请参考[此博客](https://newfacade.github.io/notes-on-reinforcement-learning/17-ppo-trl.html)。
+
 ## 数据集
 
 <details><summary>预训练数据集</summary>
@@ -260,9 +394,15 @@ https://github.com/hiyouga/LLaMA-Factory/assets/16256802/ec36a9dd-37f4-4f72-81bd
 - [STEM (zh)](https://huggingface.co/datasets/hfl/stem_zh_instruction)
 - [Ruozhiba (zh)](https://huggingface.co/datasets/hfl/ruozhiba_gpt4_turbo)
 - [Neo-sft (zh)](https://huggingface.co/datasets/m-a-p/neo_sft_phase2)
-- [WebInstructSub (en)](https://huggingface.co/datasets/TIGER-Lab/WebInstructSub)
 - [Magpie-Pro-300K-Filtered (en)](https://huggingface.co/datasets/Magpie-Align/Magpie-Pro-300K-Filtered)
+- [Magpie-ultra-v0.1 (en)](https://huggingface.co/datasets/argilla/magpie-ultra-v0.1)
+- [WebInstructSub (en)](https://huggingface.co/datasets/TIGER-Lab/WebInstructSub)
+- [OpenO1-SFT (en&zh)](https://huggingface.co/datasets/O1-OPEN/OpenO1-SFT)
+- [Open-Thoughts (en)](https://huggingface.co/datasets/open-thoughts/OpenThoughts-114k)
+- [Open-R1-Math (en)](https://huggingface.co/datasets/open-r1/OpenR1-Math-220k)
+- [Chinese-DeepSeek-R1-Distill (zh)](https://huggingface.co/datasets/Congliu/Chinese-DeepSeek-R1-Distill-data-110k-SFT)
 - [LLaVA mixed (en&zh)](https://huggingface.co/datasets/BUAADreamer/llava-en-zh-300k)
+- [Pokemon-gpt4o-captions (en&zh)](https://huggingface.co/datasets/jugg1024/pokemon-gpt4o-captions)
 - [Open Assistant (de)](https://huggingface.co/datasets/mayflowergmbh/oasst_de)
 - [Dolly 15k (de)](https://huggingface.co/datasets/mayflowergmbh/dolly-15k_de)
 - [Alpaca GPT4 (de)](https://huggingface.co/datasets/mayflowergmbh/alpaca-gpt4_de)
@@ -279,6 +419,10 @@ https://github.com/hiyouga/LLaMA-Factory/assets/16256802/ec36a9dd-37f4-4f72-81bd
 
 - [DPO mixed (en&zh)](https://huggingface.co/datasets/hiyouga/DPO-En-Zh-20k)
 - [UltraFeedback (en)](https://huggingface.co/datasets/HuggingFaceH4/ultrafeedback_binarized)
+- [COIG-P (zh)](https://huggingface.co/datasets/m-a-p/COIG-P)
+- [RLHF-V (en)](https://huggingface.co/datasets/openbmb/RLHF-V-Dataset)
+- [VLFeedback (en)](https://huggingface.co/datasets/Zhihui/VLFeedback)
+- [RLAIF-V (en)](https://huggingface.co/datasets/openbmb/RLAIF-V-Dataset)
 - [Orca DPO Pairs (en)](https://huggingface.co/datasets/Intel/orca_dpo_pairs)
 - [HH-RLHF (en)](https://huggingface.co/datasets/Anthropic/hh-rlhf)
 - [Nectar (en)](https://huggingface.co/datasets/berkeley-nest/Nectar)
@@ -298,35 +442,35 @@ huggingface-cli login
 
 | 必需项        | 至少     | 推荐      |
 | ------------ | ------- | --------- |
-| python       | 3.8     | 3.11      |
-| torch        | 1.13.1  | 2.3.0     |
-| transformers | 4.41.2  | 4.41.2    |
-| datasets     | 2.16.0  | 2.19.2    |
-| accelerate   | 0.30.1  | 0.30.1    |
-| peft         | 0.11.1  | 0.11.1    |
-| trl          | 0.8.6   | 0.9.4     |
+| python       | 3.9     | 3.10      |
+| torch        | 2.0.0   | 2.6.0     |
+| torchvision  | 0.15.0  | 0.21.0    |
+| transformers | 4.45.0  | 4.50.0    |
+| datasets     | 2.16.0  | 3.2.0     |
+| accelerate   | 0.34.0  | 1.2.1     |
+| peft         | 0.14.0  | 0.15.1    |
+| trl          | 0.8.6   | 0.9.6     |
 
 | 可选项        | 至少     | 推荐      |
 | ------------ | ------- | --------- |
 | CUDA         | 11.6    | 12.2      |
-| deepspeed    | 0.10.0  | 0.14.0    |
+| deepspeed    | 0.10.0  | 0.16.4    |
 | bitsandbytes | 0.39.0  | 0.43.1    |
-| vllm         | 0.4.3   | 0.4.3     |
-| flash-attn   | 2.3.0   | 2.5.9     |
+| vllm         | 0.4.3   | 0.8.2     |
+| flash-attn   | 2.5.6   | 2.7.2     |
 
 ### 硬件依赖
 
 \* *估算值*
 
-| 方法               | 精度 |   7B  |  13B  |  30B  |   70B  |  110B  |  8x7B |  8x22B |
-| ----------------- | ---- | ----- | ----- | ----- | ------ | ------ | ----- | ------ |
-| Full              | AMP  | 120GB | 240GB | 600GB | 1200GB | 2000GB | 900GB | 2400GB |
-| Full              |  16  |  60GB | 120GB | 300GB |  600GB |  900GB | 400GB | 1200GB |
-| Freeze            |  16  |  20GB |  40GB |  80GB |  200GB |  360GB | 160GB |  400GB |
-| LoRA/GaLore/BAdam |  16  |  16GB |  32GB |  64GB |  160GB |  240GB | 120GB |  320GB |
-| QLoRA             |   8  |  10GB |  20GB |  40GB |   80GB |  140GB |  60GB |  160GB |
-| QLoRA             |   4  |   6GB |  12GB |  24GB |   48GB |   72GB |  30GB |   96GB |
-| QLoRA             |   2  |   4GB |   8GB |  16GB |   24GB |   48GB |  18GB |   48GB |
+| 方法                             | 精度 |   7B  |  14B  |  30B  |   70B  |   `x`B  |
+| ------------------------------- | ---- | ----- | ----- | ----- | ------ | ------- |
+| Full (`bf16` or `fp16`)         |  32  | 120GB | 240GB | 600GB | 1200GB | `18x`GB |
+| Full (`pure_bf16`)              |  16  |  60GB | 120GB | 300GB |  600GB |  `8x`GB |
+| Freeze/LoRA/GaLore/APOLLO/BAdam |  16  |  16GB |  32GB |  64GB |  160GB |  `2x`GB |
+| QLoRA                           |   8  |  10GB |  20GB |  40GB |   80GB |   `x`GB |
+| QLoRA                           |   4  |   6GB |  12GB |  24GB |   48GB | `x/2`GB |
+| QLoRA                           |   2  |   4GB |   8GB |  16GB |   24GB | `x/4`GB |
 
 ## 如何使用
 
@@ -335,34 +479,77 @@ huggingface-cli login
 > [!IMPORTANT]
 > 此步骤为必需。
 
+#### 从源码安装
+
 ```bash
 git clone --depth 1 https://github.com/hiyouga/LLaMA-Factory.git
 cd LLaMA-Factory
-pip install -e ".[torch,metrics]"
+pip install -e ".[torch,metrics]" --no-build-isolation
 ```
 
-可选的额外依赖项：torch、torch_npu、metrics、deepspeed、bitsandbytes、vllm、galore、badam、gptq、awq、aqlm、qwen、modelscope、quality
+可选的额外依赖项：torch、torch-npu、metrics、deepspeed、liger-kernel、bitsandbytes、hqq、eetq、gptq、aqlm、vllm、sglang、galore、apollo、badam、adam-mini、qwen、minicpm_v、modelscope、openmind、swanlab、dev
 
-> [!TIP]
-> 遇到包冲突时，可使用 `pip install --no-deps -e .` 解决。
+#### 从镜像安装
+
+```bash
+docker run -it --rm --gpus=all --ipc=host hiyouga/llamafactory:latest
+```
+
+该镜像基于 Ubuntu 22.04（x86\_64）、CUDA 12.4、Python 3.11、PyTorch 2.6.0 和 Flash-attn 2.7.4 构建。
+
+查看全部镜像：https://hub.docker.com/r/hiyouga/llamafactory/tags
+
+请参阅[构建 Docker](#构建-docker) 来重新构建镜像。
+
+<details><summary>使用 <b>uv</b> 构建虚拟环境</summary>
+
+使用 [uv](https://github.com/astral-sh/uv) 创建隔离的 Python 环境：
+
+```bash
+uv sync --extra torch --extra metrics --prerelease=allow
+```
+
+在环境中运行 LLaMA-Factory：
+
+```bash
+uv run --prerelease=allow llamafactory-cli train examples/train_lora/llama3_lora_pretrain.yaml
+```
+
+</details>
 
 <details><summary>Windows 用户指南</summary>
 
+#### 安装 PyTorch
+
+Windows 平台需要额外手动安装 GPU 版本的 PyTorch 依赖包，您可以参考[官方网站](https://pytorch.org/get-started/locally/)和以下命令安装并测试 PyTorch 是否正确安装。
+
+```bash
+pip uninstall torch torchvision torchaudio
+pip install torch torchvision torchaudio --index-url https://download.pytorch.org/whl/cu126
+python -c "import torch; print(torch.cuda.is_available())"
+```
+
+如果看到 `True` 则说明安装成功。
+
+若遇到类似 `Can't pickle local object` 的报错，请设置 `dataloader_num_workers: 0`。
+
+#### 安装 BitsAndBytes
+
 如果要在 Windows 平台上开启量化 LoRA（QLoRA），需要安装预编译的 `bitsandbytes` 库, 支持 CUDA 11.1 到 12.2, 请根据您的 CUDA 版本情况选择适合的[发布版本](https://github.com/jllllll/bitsandbytes-windows-webui/releases/tag/wheels)。
 
 ```bash
 pip install https://github.com/jllllll/bitsandbytes-windows-webui/releases/download/wheels/bitsandbytes-0.41.2.post2-py3-none-win_amd64.whl
 ```
 
-如果要在 Windows 平台上开启 FlashAttention-2，需要安装预编译的 `flash-attn` 库，支持 CUDA 12.1 到 12.2，请根据需求到 [flash-attention](https://github.com/bdashore3/flash-attention/releases) 下载对应版本安装。
+#### 安装 Flash Attention-2
+
+如果要在 Windows 平台上开启 FlashAttention-2，请使用 [flash-attention-windows-wheel](https://huggingface.co/lldacing/flash-attention-windows-wheel) 中的脚本自行编译与安装。
 
 </details>
 
 <details><summary>昇腾 NPU 用户指南</summary>
 
-加入 [NPU 用户群](assets/wechat_npu.jpg)。
-
-在昇腾 NPU 设备上安装 LLaMA Factory 时，需要指定额外依赖项，使用 `pip install -e '.[torch-npu,metrics]'` 命令安装。此外，还需要安装 **[Ascend CANN Toolkit and Kernels](https://www.hiascend.com/developer/download/community/result?module=cann)**，安装方法请参考[安装教程](https://www.hiascend.com/document/detail/zh/CANNCommunityEdition/80RC2alpha002/quickstart/quickstart/quickstart_18_0004.html)或使用以下命令：
+在昇腾 NPU 设备上安装 LLaMA Factory 时，请升级 Python 到 3.10 及以上，并需要指定额外依赖项，使用 `pip install -e ".[torch-npu,metrics]"` 命令安装。此外，还需要安装 **[Ascend CANN Toolkit 与 Kernels](https://www.hiascend.com/developer/download/community/result?module=cann)**，安装方法请参考[安装教程](https://www.hiascend.com/document/detail/zh/CANNCommunityEdition/80RC2alpha002/quickstart/quickstart/quickstart_18_0004.html)或使用以下命令：
 
 ```bash
 # 请替换 URL 为 CANN 版本和设备型号对应的 URL
@@ -379,30 +566,64 @@ source /usr/local/Ascend/ascend-toolkit/set_env.sh
 ```
 
 | 依赖项        | 至少     | 推荐           |
-| ------------ | ------- | ----------- |
-| CANN         | 8.0.RC1 | 8.0.RC1     |
-| torch        | 2.1.0   | 2.1.0       |
-| torch-npu    | 2.1.0   | 2.1.0.post3 |
+| ------------ | ------- | -------------- |
+| CANN         | 8.0.RC1 | 8.0.0.alpha002 |
+| torch        | 2.1.0   | 2.4.0          |
+| torch-npu    | 2.1.0   | 2.4.0.post2    |
 | deepspeed    | 0.13.2  | 0.13.2         |
-
-Docker 镜像：
-
-- 32GB：[下载地址](http://mirrors.cn-central-221.ovaijisuan.com/detail/130.html)
-- 64GB：[下载地址](http://mirrors.cn-central-221.ovaijisuan.com/detail/131.html)
+| vllm-ascend  | -       | 0.7.3          |
 
 请使用 `ASCEND_RT_VISIBLE_DEVICES` 而非 `CUDA_VISIBLE_DEVICES` 来指定运算设备。
 
 如果遇到无法正常推理的情况，请尝试设置 `do_sample: false`。
 
+下载预构建 Docker 镜像：[32GB](http://mirrors.cn-central-221.ovaijisuan.com/detail/130.html) | [64GB](http://mirrors.cn-central-221.ovaijisuan.com/detail/131.html)
+
+#### 安装 BitsAndBytes
+
+如果要在 Ascend NPU 上进行基于 bitsandbytes 的 QLoRA 量化微调，请执行如下步骤：
+
+1. 手动编译 bitsandbytes：请参考[安装文档](https://huggingface.co/docs/bitsandbytes/installation?backend=Ascend+NPU&platform=Ascend+NPU)完成 NPU 版的 bitsandbytes 安装，编译要求环境 cmake 版本不低于 3.22.1，g++ 版本不低于 12.x。
+
+```bash
+# 从源码安装 bitsandbytes
+# 克隆 bitsandbytes 仓库, Ascend NPU 目前在 multi-backend-refactor 中支持
+git clone -b multi-backend-refactor https://github.com/bitsandbytes-foundation/bitsandbytes.git
+cd bitsandbytes/
+
+# 安装依赖
+pip install -r requirements-dev.txt
+
+# 安装编译工具依赖，该步骤在不同系统上命令有所不同，供参考
+apt-get install -y build-essential cmake
+
+# 编译 & 安装
+cmake -DCOMPUTE_BACKEND=npu -S .
+make
+pip install .
+```
+
+2. 安装 transformers 的 main 分支版本。
+
+```bash
+git clone -b main https://github.com/huggingface/transformers.git
+cd transformers
+pip install .
+```
+
+3. 在训练参数中设置 `double_quantization: false`，可参考[示例](examples/train_qlora/llama3_lora_sft_bnb_npu.yaml)。
+
 </details>
 
 ### 数据准备
 
-关于数据集文件的格式，请参考 [data/README_zh.md](data/README_zh.md) 的内容。你可以使用 HuggingFace / ModelScope 上的数据集或加载本地数据集。
+关于数据集文件的格式，请参考 [data/README_zh.md](data/README_zh.md) 的内容。你可以使用 HuggingFace / ModelScope / Modelers 上的数据集或加载本地数据集。
 
 > [!NOTE]
 > 使用自定义数据集时，请更新 `data/dataset_info.json` 文件。
 
+您也可以使用 **[Easy Dataset](https://github.com/ConardLi/easy-dataset)** 或 **[GraphGen](https://github.com/open-sciencelab/GraphGen)** 构建用于微调的合成数据。
+
 ### 快速开始
 
 下面三行命令分别对 Llama3-8B-Instruct 模型进行 LoRA **微调**、**推理**和**合并**。
@@ -417,6 +638,8 @@ llamafactory-cli export examples/merge_lora/llama3_lora_sft.yaml
 
 > [!TIP]
 > 使用 `llamafactory-cli help` 显示帮助信息。
+>
+> 遇到报错请先看[常见问题](https://github.com/hiyouga/LLaMA-Factory/issues/4614)。
 
 ### LLaMA Board 可视化微调（由 [Gradio](https://github.com/gradio-app/gradio) 驱动）
 
@@ -426,50 +649,117 @@ llamafactory-cli webui
 
 ### 构建 Docker
 
-#### 使用 Docker
+CUDA 用户：
 
 ```bash
-docker build -f ./Dockerfile \
-    --build-arg INSTALL_BNB=false \
-    --build-arg INSTALL_VLLM=false \
-    --build-arg INSTALL_DEEPSPEED=false \
+cd docker/docker-cuda/
+docker compose up -d
+docker compose exec llamafactory bash
+```
+
+昇腾 NPU 用户：
+
+```bash
+cd docker/docker-npu/
+docker compose up -d
+docker compose exec llamafactory bash
+```
+
+AMD ROCm 用户：
+
+```bash
+cd docker/docker-rocm/
+docker compose up -d
+docker compose exec llamafactory bash
+```
+
+<details><summary>不使用 Docker Compose 构建</summary>
+
+CUDA 用户：
+
+```bash
+docker build -f ./docker/docker-cuda/Dockerfile \
     --build-arg PIP_INDEX=https://pypi.org/simple \
+    --build-arg EXTRAS=metrics \
     -t llamafactory:latest .
 
-docker run -it --gpus=all \
-    -v ./hf_cache:/root/.cache/huggingface/ \
-    -v ./data:/app/data \
-    -v ./output:/app/output \
+docker run -dit --ipc=host --gpus=all \
     -p 7860:7860 \
     -p 8000:8000 \
-    --shm-size 16G \
     --name llamafactory \
     llamafactory:latest
+
+docker exec -it llamafactory bash
 ```
 
-#### 使用 Docker Compose
+昇腾 NPU 用户：
 
 ```bash
-docker-compose up -d
-docker-compose exec llamafactory bash
+docker build -f ./docker/docker-npu/Dockerfile \
+    --build-arg PIP_INDEX=https://pypi.org/simple \
+    --build-arg EXTRAS=torch-npu,metrics \
+    -t llamafactory:latest .
+
+docker run -dit --ipc=host \
+    -v /usr/local/dcmi:/usr/local/dcmi \
+    -v /usr/local/bin/npu-smi:/usr/local/bin/npu-smi \
+    -v /usr/local/Ascend/driver:/usr/local/Ascend/driver \
+    -v /etc/ascend_install.info:/etc/ascend_install.info \
+    -p 7860:7860 \
+    -p 8000:8000 \
+    --device /dev/davinci0 \
+    --device /dev/davinci_manager \
+    --device /dev/devmm_svm \
+    --device /dev/hisi_hdc \
+    --name llamafactory \
+    llamafactory:latest
+
+docker exec -it llamafactory bash
 ```
 
-<details><summary>数据卷详情</summary>
+AMD ROCm 用户：
 
-- hf_cache：使用宿主机的 Hugging Face 缓存文件夹，允许更改为新的目录。
-- data：宿主机中存放数据集的文件夹路径。
-- output：将导出目录设置为该路径后，即可在宿主机中访问导出后的模型。
+```bash
+docker build -f ./docker/docker-rocm/Dockerfile \
+    --build-arg PIP_INDEX=https://pypi.org/simple \
+    --build-arg EXTRAS=metrics \
+    -t llamafactory:latest .
+
+docker run -dit --ipc=host \
+    -p 7860:7860 \
+    -p 8000:8000 \
+    --device /dev/kfd \
+    --device /dev/dri \
+    --name llamafactory \
+    llamafactory:latest
+
+docker exec -it llamafactory bash
+```
+
+</details>
+
+<details><summary>使用数据卷</summary>
+
+您可以通过移除 Dockerfile 中 `VOLUME [ "/root/.cache/huggingface", "/app/shared_data", "/app/output" ]` 的注释来使用数据卷。
+
+在构建 Docker 时使用参数 `-v ./hf_cache:/root/.cache/huggingface` 来挂载数据卷。各个数据卷的含义表示如下。
+
+- `hf_cache`：使用宿主机的 Hugging Face 缓存文件夹。
+- `shared_data`：宿主机中存放数据集的文件夹路径。
+- `output`：将导出目录设置为该路径后，即可在宿主机中访问导出后的模型。
 
 </details>
 
 ### 利用 vLLM 部署 OpenAI API
 
 ```bash
-API_PORT=8000 llamafactory-cli api examples/inference/llama3_vllm.yaml
+API_PORT=8000 llamafactory-cli api examples/inference/llama3.yaml infer_backend=vllm vllm_enforce_eager=true
 ```
 
 > [!TIP]
-> API 文档请查阅 https://platform.openai.com/docs/api-reference/chat/create。
+> API 文档请查阅[这里](https://platform.openai.com/docs/api-reference/chat/create)。
+>
+> 示例：[图像理解](scripts/api_example/test_image.py) | [工具调用](scripts/api_example/test_toolcall.py)
 
 ### 从魔搭社区下载
 
@@ -481,6 +771,16 @@ export USE_MODELSCOPE_HUB=1 # Windows 使用 `set USE_MODELSCOPE_HUB=1`
 
 将 `model_name_or_path` 设置为模型 ID 来加载对应的模型。在[魔搭社区](https://modelscope.cn/models)查看所有可用的模型，例如 `LLM-Research/Meta-Llama-3-8B-Instruct`。
 
+### 从魔乐社区下载
+
+您也可以通过下述方法，使用魔乐社区下载数据集和模型。
+
+```bash
+export USE_OPENMIND_HUB=1 # Windows 使用 `set USE_OPENMIND_HUB=1`
+```
+
+将 `model_name_or_path` 设置为模型 ID 来加载对应的模型。在[魔乐社区](https://modelers.cn/models)查看所有可用的模型，例如 `TeleAI/TeleChat-7B-pt`。
+
 ### 使用 W&B 面板
 
 若要使用 [Weights & Biases](https://wandb.ai) 记录实验数据，请在 yaml 文件中添加下面的参数。
@@ -492,6 +792,21 @@ run_name: test_run # 可选
 
 在启动训练任务时，将 `WANDB_API_KEY` 设置为[密钥](https://wandb.ai/authorize)来登录 W&B 账户。
 
+### 使用 SwanLab 面板
+
+若要使用 [SwanLab](https://github.com/SwanHubX/SwanLab) 记录实验数据，请在 yaml 文件中添加下面的参数。
+
+```yaml
+use_swanlab: true
+swanlab_run_name: test_run # 可选
+```
+
+在启动训练任务时，登录SwanLab账户有以下三种方式：
+
+方式一：在 yaml 文件中添加 `swanlab_api_key=<your_api_key>` ，并设置为你的 [API 密钥](https://swanlab.cn/settings)。
+方式二：将环境变量 `SWANLAB_API_KEY` 设置为你的 [API 密钥](https://swanlab.cn/settings)。
+方式三：启动前使用 `swanlab login` 命令完成登录。
+
 ## 使用了 LLaMA Factory 的项目
 
 如果您有项目希望添加至下述列表，请通过邮件联系或者创建一个 PR。
@@ -503,45 +818,97 @@ run_name: test_run # 可选
 1. Wang et al. UbiPhysio: Support Daily Functioning, Fitness, and Rehabilitation with Action Understanding and Feedback in Natural Language. 2023. [[arxiv]](https://arxiv.org/abs/2308.10526)
 1. Luceri et al. Leveraging Large Language Models to Detect Influence Campaigns in Social Media. 2023. [[arxiv]](https://arxiv.org/abs/2311.07816)
 1. Zhang et al. Alleviating Hallucinations of Large Language Models through Induced Hallucinations. 2023. [[arxiv]](https://arxiv.org/abs/2312.15710)
-1. Wang et al. Know Your Needs Better: Towards Structured Understanding of Marketer Demands with Analogical Reasoning Augmented LLMs. 2024. [[arxiv]](https://arxiv.org/abs/2401.04319)
-1. Wang et al. CANDLE: Iterative Conceptualization and Instantiation Distillation from Large Language Models for Commonsense Reasoning. 2024. [[arxiv]](https://arxiv.org/abs/2401.07286)
+1. Wang et al. Know Your Needs Better: Towards Structured Understanding of Marketer Demands with Analogical Reasoning Augmented LLMs. KDD 2024. [[arxiv]](https://arxiv.org/abs/2401.04319)
+1. Wang et al. CANDLE: Iterative Conceptualization and Instantiation Distillation from Large Language Models for Commonsense Reasoning. ACL 2024. [[arxiv]](https://arxiv.org/abs/2401.07286)
 1. Choi et al. FACT-GPT: Fact-Checking Augmentation via Claim Matching with LLMs. 2024. [[arxiv]](https://arxiv.org/abs/2402.05904)
 1. Zhang et al. AutoMathText: Autonomous Data Selection with Language Models for Mathematical Texts. 2024. [[arxiv]](https://arxiv.org/abs/2402.07625)
 1. Lyu et al. KnowTuning: Knowledge-aware Fine-tuning for Large Language Models. 2024. [[arxiv]](https://arxiv.org/abs/2402.11176)
 1. Yang et al. LaCo: Large Language Model Pruning via Layer Collaps. 2024. [[arxiv]](https://arxiv.org/abs/2402.11187)
 1. Bhardwaj et al. Language Models are Homer Simpson! Safety Re-Alignment of Fine-tuned Language Models through Task Arithmetic. 2024. [[arxiv]](https://arxiv.org/abs/2402.11746)
 1. Yang et al. Enhancing Empathetic Response Generation by Augmenting LLMs with Small-scale Empathetic Models. 2024. [[arxiv]](https://arxiv.org/abs/2402.11801)
-1. Yi et al. Generation Meets Verification: Accelerating Large Language Model Inference with Smart Parallel Auto-Correct Decoding. 2024. [[arxiv]](https://arxiv.org/abs/2402.11809)
+1. Yi et al. Generation Meets Verification: Accelerating Large Language Model Inference with Smart Parallel Auto-Correct Decoding. ACL 2024 Findings. [[arxiv]](https://arxiv.org/abs/2402.11809)
 1. Cao et al. Head-wise Shareable Attention for Large Language Models. 2024. [[arxiv]](https://arxiv.org/abs/2402.11819)
 1. Zhang et al. Enhancing Multilingual Capabilities of Large Language Models through Self-Distillation from Resource-Rich Languages. 2024. [[arxiv]](https://arxiv.org/abs/2402.12204)
 1. Kim et al. Efficient and Effective Vocabulary Expansion Towards Multilingual Large Language Models. 2024. [[arxiv]](https://arxiv.org/abs/2402.14714)
-1. Yu et al. KIEval: A Knowledge-grounded Interactive Evaluation Framework for Large Language Models. 2024. [[arxiv]](https://arxiv.org/abs/2402.15043)
+1. Yu et al. KIEval: A Knowledge-grounded Interactive Evaluation Framework for Large Language Models. ACL 2024. [[arxiv]](https://arxiv.org/abs/2402.15043)
 1. Huang et al. Key-Point-Driven Data Synthesis with its Enhancement on Mathematical Reasoning. 2024. [[arxiv]](https://arxiv.org/abs/2403.02333)
 1. Duan et al. Negating Negatives: Alignment without Human Positive Samples via Distributional Dispreference Optimization. 2024. [[arxiv]](https://arxiv.org/abs/2403.03419)
 1. Xie and Schwertfeger. Empowering Robotics with Large Language Models: osmAG Map Comprehension with LLMs. 2024. [[arxiv]](https://arxiv.org/abs/2403.08228)
 1. Wu et al. Large Language Models are Parallel Multilingual Learners. 2024. [[arxiv]](https://arxiv.org/abs/2403.09073)
 1. Zhang et al. EDT: Improving Large Language Models' Generation by Entropy-based Dynamic Temperature Sampling. 2024. [[arxiv]](https://arxiv.org/abs/2403.14541)
 1. Weller et al. FollowIR: Evaluating and Teaching Information Retrieval Models to Follow Instructions. 2024. [[arxiv]](https://arxiv.org/abs/2403.15246)
-1. Hongbin Na. CBT-LLM: A Chinese Large Language Model for Cognitive Behavioral Therapy-based Mental Health Question Answering. 2024. [[arxiv]](https://arxiv.org/abs/2403.16008)
+1. Hongbin Na. CBT-LLM: A Chinese Large Language Model for Cognitive Behavioral Therapy-based Mental Health Question Answering. COLING 2024. [[arxiv]](https://arxiv.org/abs/2403.16008)
 1. Zan et al. CodeS: Natural Language to Code Repository via Multi-Layer Sketch. 2024. [[arxiv]](https://arxiv.org/abs/2403.16443)
 1. Liu et al. Extensive Self-Contrast Enables Feedback-Free Language Model Alignment. 2024. [[arxiv]](https://arxiv.org/abs/2404.00604)
 1. Luo et al. BAdam: A Memory Efficient Full Parameter Training Method for Large Language Models. 2024. [[arxiv]](https://arxiv.org/abs/2404.02827)
 1. Du et al. Chinese Tiny LLM: Pretraining a Chinese-Centric Large Language Model. 2024. [[arxiv]](https://arxiv.org/abs/2404.04167)
-1. Ma et al. Parameter Efficient Quasi-Orthogonal Fine-Tuning via Givens Rotation. 2024. [[arxiv]](https://arxiv.org/abs/2404.04316)
+1. Ma et al. Parameter Efficient Quasi-Orthogonal Fine-Tuning via Givens Rotation. ICML 2024. [[arxiv]](https://arxiv.org/abs/2404.04316)
 1. Liu et al. Dynamic Generation of Personalities with Large Language Models. 2024. [[arxiv]](https://arxiv.org/abs/2404.07084)
 1. Shang et al. How Far Have We Gone in Stripped Binary Code Understanding Using Large Language Models. 2024. [[arxiv]](https://arxiv.org/abs/2404.09836)
 1. Huang et al. LLMTune: Accelerate Database Knob Tuning with Large Language Models. 2024. [[arxiv]](https://arxiv.org/abs/2404.11581)
 1. Deng et al. Text-Tuple-Table: Towards Information Integration in Text-to-Table Generation via Global Tuple Extraction. 2024. [[arxiv]](https://arxiv.org/abs/2404.14215)
 1. Acikgoz et al. Hippocrates: An Open-Source Framework for Advancing Large Language Models in Healthcare. 2024. [[arxiv]](https://arxiv.org/abs/2404.16621)
-1. Zhang et al. Small Language Models Need Strong Verifiers to Self-Correct Reasoning. 2024. [[arxiv]](https://arxiv.org/abs/2404.17140)
-1. Zhou et al. FREB-TQA: A Fine-Grained Robustness Evaluation Benchmark for Table Question Answering. 2024. [[arxiv]](https://arxiv.org/abs/2404.18585)
+1. Zhang et al. Small Language Models Need Strong Verifiers to Self-Correct Reasoning. ACL 2024 Findings. [[arxiv]](https://arxiv.org/abs/2404.17140)
+1. Zhou et al. FREB-TQA: A Fine-Grained Robustness Evaluation Benchmark for Table Question Answering. NAACL 2024. [[arxiv]](https://arxiv.org/abs/2404.18585)
+1. Xu et al. Large Language Models for Cyber Security: A Systematic Literature Review. 2024. [[arxiv]](https://arxiv.org/abs/2405.04760)
+1. Dammu et al. "They are uncultured": Unveiling Covert Harms and Social Threats in LLM Generated Conversations. 2024. [[arxiv]](https://arxiv.org/abs/2405.05378)
+1. Yi et al. A safety realignment framework via subspace-oriented model fusion for large language models. 2024. [[arxiv]](https://arxiv.org/abs/2405.09055)
+1. Lou et al. SPO: Multi-Dimensional Preference Sequential Alignment With Implicit Reward Modeling. 2024. [[arxiv]](https://arxiv.org/abs/2405.12739)
+1. Zhang et al. Getting More from Less: Large Language Models are Good Spontaneous Multilingual Learners. 2024. [[arxiv]](https://arxiv.org/abs/2405.13816)
+1. Zhang et al. TS-Align: A Teacher-Student Collaborative Framework for Scalable Iterative Finetuning of Large Language Models. 2024. [[arxiv]](https://arxiv.org/abs/2405.20215)
+1. Zihong Chen. Sentence Segmentation and Sentence Punctuation Based on XunziALLM. 2024. [[paper]](https://aclanthology.org/2024.lt4hala-1.30)
+1. Gao et al. The Best of Both Worlds: Toward an Honest and Helpful Large Language Model. 2024. [[arxiv]](https://arxiv.org/abs/2406.00380)
+1. Wang and Song. MARS: Benchmarking the Metaphysical Reasoning Abilities of Language Models with a Multi-task Evaluation Dataset. 2024. [[arxiv]](https://arxiv.org/abs/2406.02106)
+1. Hu et al. Computational Limits of Low-Rank Adaptation (LoRA) for Transformer-Based Models. 2024. [[arxiv]](https://arxiv.org/abs/2406.03136)
+1. Ge et al. Time Sensitive Knowledge Editing through Efficient Finetuning. ACL 2024. [[arxiv]](https://arxiv.org/abs/2406.04496)
+1. Tan et al. Peer Review as A Multi-Turn and Long-Context Dialogue with Role-Based Interactions. 2024. [[arxiv]](https://arxiv.org/abs/2406.05688)
+1. Song et al. Turbo Sparse: Achieving LLM SOTA Performance with Minimal Activated Parameters. 2024. [[arxiv]](https://arxiv.org/abs/2406.05955)
+1. Gu et al. RWKV-CLIP: A Robust Vision-Language Representation Learner. 2024. [[arxiv]](https://arxiv.org/abs/2406.06973)
+1. Chen et al. Advancing Tool-Augmented Large Language Models: Integrating Insights from Errors in Inference Trees. 2024. [[arxiv]](https://arxiv.org/abs/2406.07115)
+1. Zhu et al. Are Large Language Models Good Statisticians?. 2024. [[arxiv]](https://arxiv.org/abs/2406.07815)
+1. Li et al. Know the Unknown: An Uncertainty-Sensitive Method for LLM Instruction Tuning. 2024. [[arxiv]](https://arxiv.org/abs/2406.10099)
+1. Ding et al. IntentionQA: A Benchmark for Evaluating Purchase Intention Comprehension Abilities of Language Models in E-commerce. 2024. [[arxiv]](https://arxiv.org/abs/2406.10173)
+1. He et al. COMMUNITY-CROSS-INSTRUCT: Unsupervised Instruction Generation for Aligning Large Language Models to Online Communities. 2024. [[arxiv]](https://arxiv.org/abs/2406.12074)
+1. Lin et al. FVEL: Interactive Formal Verification Environment with Large Language Models via Theorem Proving. 2024. [[arxiv]](https://arxiv.org/abs/2406.14408)
+1. Treutlein et al. Connecting the Dots: LLMs can Infer and Verbalize Latent Structure from Disparate Training Data. 2024. [[arxiv]](https://arxiv.org/abs/2406.14546)
+1. Feng et al. SS-Bench: A Benchmark for Social Story Generation and Evaluation. 2024. [[arxiv]](https://arxiv.org/abs/2406.15695)
+1. Feng et al. Self-Constructed Context Decompilation with Fined-grained Alignment Enhancement. 2024. [[arxiv]](https://arxiv.org/abs/2406.17233)
+1. Liu et al. Large Language Models for Cuffless Blood Pressure Measurement From Wearable Biosignals. 2024. [[arxiv]](https://arxiv.org/abs/2406.18069)
+1. Iyer et al. Exploring Very Low-Resource Translation with LLMs: The University of Edinburgh's Submission to AmericasNLP 2024 Translation Task. AmericasNLP 2024. [[paper]](https://aclanthology.org/2024.americasnlp-1.25)
+1. Li et al. Calibrating LLMs with Preference Optimization on Thought Trees for Generating Rationale in Science Question Scoring. 2024. [[arxiv]](https://arxiv.org/abs/2406.19949)
+1. Yang et al. Financial Knowledge Large Language Model. 2024. [[arxiv]](https://arxiv.org/abs/2407.00365)
+1. Lin et al. DogeRM: Equipping Reward Models with Domain Knowledge through Model Merging. 2024. [[arxiv]](https://arxiv.org/abs/2407.01470)
+1. Bako et al. Evaluating the Semantic Profiling Abilities of LLMs for Natural Language Utterances in Data Visualization. 2024. [[arxiv]](https://arxiv.org/abs/2407.06129)
+1. Huang et al. RoLoRA: Fine-tuning Rotated Outlier-free LLMs for Effective Weight-Activation Quantization. 2024. [[arxiv]](https://arxiv.org/abs/2407.08044)
+1. Jiang et al. LLM-Collaboration on Automatic Science Journalism for the General Audience. 2024. [[arxiv]](https://arxiv.org/abs/2407.09756)
+1. Inouye et al. Applied Auto-tuning on LoRA Hyperparameters. 2024. [[paper]](https://scholarcommons.scu.edu/cseng_senior/272/)
+1. Qi et al. Research on Tibetan Tourism Viewpoints information generation system based on LLM. 2024. [[arxiv]](https://arxiv.org/abs/2407.13561)
+1. Xu et al. Course-Correction: Safety Alignment Using Synthetic Preferences. 2024. [[arxiv]](https://arxiv.org/abs/2407.16637)
+1. Sun et al. LAMBDA: A Large Model Based Data Agent. 2024. [[arxiv]](https://arxiv.org/abs/2407.17535)
+1. Zhu et al. CollectiveSFT: Scaling Large Language Models for Chinese Medical Benchmark with Collective Instructions in Healthcare. 2024. [[arxiv]](https://arxiv.org/abs/2407.19705)
+1. Yu et al. Correcting Negative Bias in Large Language Models through Negative Attention Score Alignment. 2024. [[arxiv]](https://arxiv.org/abs/2408.00137)
+1. Xie et al. The Power of Personalized Datasets: Advancing Chinese Composition Writing for Elementary School through Targeted Model Fine-Tuning. IALP 2024. [[paper]](https://www.asianlp.sg/conferences/ialp2024/proceedings/papers/IALP2024_P055.pdf)
+1. Liu et al. Instruct-Code-Llama: Improving Capabilities of Language Model in Competition Level Code Generation by Online Judge Feedback. ICIC 2024. [[paper]](https://link.springer.com/chapter/10.1007/978-981-97-5669-8_11)
+1. Wang et al. Cybernetic Sentinels: Unveiling the Impact of Safety Data Selection on Model Security in Supervised Fine-Tuning. ICIC 2024. [[paper]](https://link.springer.com/chapter/10.1007/978-981-97-5669-8_23)
+1. Xia et al. Understanding the Performance and Estimating the Cost of LLM Fine-Tuning. 2024. [[arxiv]](https://arxiv.org/abs/2408.04693)
+1. Zeng et al. Perceive, Reflect, and Plan: Designing LLM Agent for Goal-Directed City Navigation without Instructions. 2024. [[arxiv]](https://arxiv.org/abs/2408.04168)
+1. Xia et al. Using Pre-trained Language Model for Accurate ESG Prediction. FinNLP 2024. [[paper]](https://aclanthology.org/2024.finnlp-2.1/)
+1. Liang et al. I-SHEEP: Self-Alignment of LLM from Scratch through an Iterative Self-Enhancement Paradigm. 2024. [[arxiv]](https://arxiv.org/abs/2408.08072)
+1. Bai et al. Aligning Large Language Model with Direct Multi-Preference Optimization for Recommendation. CIKM 2024. [[paper]](https://dl.acm.org/doi/10.1145/3627673.3679611)
 1. **[StarWhisper](https://github.com/Yu-Yang-Li/StarWhisper)**: 天文大模型 StarWhisper，基于 ChatGLM2-6B 和 Qwen-14B 在天文数据上微调而得。
 1. **[DISC-LawLLM](https://github.com/FudanDISC/DISC-LawLLM)**: 中文法律领域大模型 DISC-LawLLM，基于 Baichuan-13B 微调而得，具有法律推理和知识检索能力。
 1. **[Sunsimiao](https://github.com/X-D-Lab/Sunsimiao)**: 孙思邈中文医疗大模型 Sumsimiao，基于 Baichuan-7B 和 ChatGLM-6B 在中文医疗数据上微调而得。
 1. **[CareGPT](https://github.com/WangRongsheng/CareGPT)**: 医疗大模型项目 CareGPT，基于 LLaMA2-7B 和 Baichuan-13B 在中文医疗数据上微调而得。
 1. **[MachineMindset](https://github.com/PKU-YuanGroup/Machine-Mindset/)**：MBTI性格大模型项目，根据数据集与训练方式让任意 LLM 拥有 16 个不同的性格类型。
-1. **[Luminia-13B-v3](https://huggingface.co/Nekochu/Luminia-13B-v3)**：一个用于生成 Stable Diffusion 提示词的大型语言模型。[[🤗Demo]](https://huggingface.co/spaces/Nekochu/Luminia-13B_SD_Prompt)
+1. **[Luminia-13B-v3](https://huggingface.co/Nekochu/Luminia-13B-v3)**：一个用于生成 Stable Diffusion 提示词的大型语言模型。[[demo]](https://huggingface.co/spaces/Nekochu/Luminia-13B_SD_Prompt)
 1. **[Chinese-LLaVA-Med](https://github.com/BUAADreamer/Chinese-LLaVA-Med)**：中文多模态医学大模型，基于 LLaVA-1.5-7B 在中文多模态医疗数据上微调而得。
+1. **[AutoRE](https://github.com/THUDM/AutoRE)**：基于大语言模型的文档级关系抽取系统。
+1. **[NVIDIA RTX AI Toolkit](https://github.com/NVIDIA/RTX-AI-Toolkit)**：在 Windows 主机上利用英伟达 RTX 设备进行大型语言模型微调的开发包。
+1. **[LazyLLM](https://github.com/LazyAGI/LazyLLM)**：一个低代码构建多 Agent 大模型应用的开发工具，支持基于 LLaMA Factory 的模型微调.
+1. **[RAG-Retrieval](https://github.com/NLPJCL/RAG-Retrieval)**：一个全链路 RAG 检索模型微调、推理和蒸馏代码库。[[blog]](https://zhuanlan.zhihu.com/p/987727357)
+1. **[360-LLaMA-Factory](https://github.com/Qihoo360/360-LLaMA-Factory)**：一个魔改后的代码库，通过 Ring Attention 支持长序列的 SFT 和 DPO 训练。
+1. **[Sky-T1](https://novasky-ai.github.io/posts/sky-t1/)**：由 NovaSky AI 微调的低成本类 o1 长推理模型。
+1. **[WeClone](https://github.com/xming521/WeClone)**：从聊天记录创造数字分身的一站式解决方案。
 
 </details>
 
@@ -549,17 +916,19 @@ run_name: test_run # 可选
 
 本仓库的代码依照 [Apache-2.0](LICENSE) 协议开源。
 
-使用模型权重时，请遵循对应的模型协议：[Baichuan2](https://huggingface.co/baichuan-inc/Baichuan2-7B-Base/blob/main/Community%20License%20for%20Baichuan%202%20Model.pdf) / [BLOOM](https://huggingface.co/spaces/bigscience/license) / [ChatGLM3](https://github.com/THUDM/ChatGLM3/blob/main/MODEL_LICENSE) / [Command-R](https://cohere.com/c4ai-cc-by-nc-license) / [DeepSeek](https://github.com/deepseek-ai/DeepSeek-LLM/blob/main/LICENSE-MODEL) / [Falcon](https://huggingface.co/tiiuae/falcon-180B/blob/main/LICENSE.txt) / [Gemma](https://ai.google.dev/gemma/terms) / [GLM4](https://huggingface.co/THUDM/glm-4-9b/blob/main/LICENSE) / [InternLM2](https://github.com/InternLM/InternLM#license) / [LLaMA](https://github.com/facebookresearch/llama/blob/main/MODEL_CARD.md) / [LLaMA-2 (LLaVA-1.5)](https://ai.meta.com/llama/license/) / [LLaMA-3](https://llama.meta.com/llama3/license/) / [Mistral](LICENSE) / [OLMo](LICENSE) / [Phi-1.5/2](https://huggingface.co/microsoft/phi-1_5/resolve/main/Research%20License.docx) / [Phi-3](https://huggingface.co/microsoft/Phi-3-mini-4k-instruct/blob/main/LICENSE) / [Qwen](https://github.com/QwenLM/Qwen/blob/main/Tongyi%20Qianwen%20LICENSE%20AGREEMENT) / [StarCoder2](https://huggingface.co/spaces/bigcode/bigcode-model-license-agreement) / [XVERSE](https://github.com/xverse-ai/XVERSE-13B/blob/main/MODEL_LICENSE.pdf) / [Yi](https://huggingface.co/01-ai/Yi-6B/blob/main/LICENSE) / [Yi-1.5](LICENSE) / [Yuan](https://github.com/IEIT-Yuan/Yuan-2.0/blob/main/LICENSE-Yuan)
+使用模型权重时，请遵循对应的模型协议：[Baichuan 2](https://huggingface.co/baichuan-inc/Baichuan2-7B-Base/blob/main/Community%20License%20for%20Baichuan%202%20Model.pdf) / [BLOOM](https://huggingface.co/spaces/bigscience/license) / [ChatGLM3](https://github.com/THUDM/ChatGLM3/blob/main/MODEL_LICENSE) / [Command R](https://cohere.com/c4ai-cc-by-nc-license) / [DeepSeek](https://github.com/deepseek-ai/DeepSeek-LLM/blob/main/LICENSE-MODEL) / [Falcon](https://huggingface.co/tiiuae/falcon-180B/blob/main/LICENSE.txt) / [Gemma](https://ai.google.dev/gemma/terms) / [GLM-4](https://huggingface.co/THUDM/glm-4-9b/blob/main/LICENSE) / [GPT-2](https://github.com/openai/gpt-2/blob/master/LICENSE) / [Granite](LICENSE) / [Index](https://huggingface.co/IndexTeam/Index-1.9B/blob/main/LICENSE) / [InternLM](https://github.com/InternLM/InternLM#license) / [Llama](https://github.com/facebookresearch/llama/blob/main/MODEL_CARD.md) / [Llama 2](https://ai.meta.com/llama/license/) / [Llama 3](https://llama.meta.com/llama3/license/) / [Llama 4](https://github.com/meta-llama/llama-models/blob/main/models/llama4/LICENSE) / [MiniCPM](https://github.com/OpenBMB/MiniCPM/blob/main/MiniCPM%20Model%20License.md) / [Mistral/Mixtral/Pixtral](LICENSE) / [OLMo](LICENSE) / [Phi-1.5/Phi-2](https://huggingface.co/microsoft/phi-1_5/resolve/main/Research%20License.docx) / [Phi-3/Phi-4](https://huggingface.co/microsoft/Phi-3-mini-4k-instruct/blob/main/LICENSE) / [Qwen](https://github.com/QwenLM/Qwen/blob/main/Tongyi%20Qianwen%20LICENSE%20AGREEMENT) / [Skywork](https://huggingface.co/Skywork/Skywork-13B-base/blob/main/Skywork%20Community%20License.pdf) / [StarCoder 2](https://huggingface.co/spaces/bigcode/bigcode-model-license-agreement) / [TeleChat2](https://huggingface.co/Tele-AI/telechat-7B/blob/main/TeleChat%E6%A8%A1%E5%9E%8B%E7%A4%BE%E5%8C%BA%E8%AE%B8%E5%8F%AF%E5%8D%8F%E8%AE%AE.pdf) / [XVERSE](https://github.com/xverse-ai/XVERSE-13B/blob/main/MODEL_LICENSE.pdf) / [Yi](https://huggingface.co/01-ai/Yi-6B/blob/main/LICENSE) / [Yi-1.5](LICENSE) / [Yuan 2](https://github.com/IEIT-Yuan/Yuan-2.0/blob/main/LICENSE-Yuan)
 
 ## 引用
 
 如果您觉得此项目有帮助，请考虑以下列格式引用
 
 ```bibtex
-@article{zheng2024llamafactory,
+@inproceedings{zheng2024llamafactory,
   title={LlamaFactory: Unified Efficient Fine-Tuning of 100+ Language Models},
-  author={Yaowei Zheng and Richong Zhang and Junhao Zhang and Yanhan Ye and Zheyan Luo and Yongqiang Ma},
-  journal={arXiv preprint arXiv:2403.13372},
+  author={Yaowei Zheng and Richong Zhang and Junhao Zhang and Yanhan Ye and Zheyan Luo and Zhangchi Feng and Yongqiang Ma},
+  booktitle={Proceedings of the 62nd Annual Meeting of the Association for Computational Linguistics (Volume 3: System Demonstrations)},
+  address={Bangkok, Thailand},
+  publisher={Association for Computational Linguistics},
   year={2024},
   url={http://arxiv.org/abs/2403.13372}
 }

data/README.md

@@ -1,18 +1,22 @@
 The [dataset_info.json](dataset_info.json) contains all available datasets. If you are using a custom dataset, please **make sure** to add a *dataset description* in `dataset_info.json` and specify `dataset: dataset_name` before training to use it.
 
-Currently we support datasets in **alpaca** and **sharegpt** format.
+The `dataset_info.json` file should be put in the `dataset_dir` directory. You can change `dataset_dir` to use another directory. The default value is `./data`.
+
+Currently we support datasets in **alpaca** and **sharegpt** format. Allowed file types include json, jsonl, csv, parquet, arrow.
 
 ```json
 "dataset_name": {
-  "hf_hub_url": "the name of the dataset repository on the Hugging Face hub. (if specified, ignore script_url and file_name)",
-  "ms_hub_url": "the name of the dataset repository on the Model Scope hub. (if specified, ignore script_url and file_name)",
-  "script_url": "the name of the directory containing a dataset loading script. (if specified, ignore file_name)",
+  "hf_hub_url": "the name of the dataset repository on the Hugging Face hub. (if specified, ignore script_url, file_name and cloud_file_name)",
+  "ms_hub_url": "the name of the dataset repository on the Model Scope hub. (if specified, ignore script_url, file_name and cloud_file_name)",
+  "script_url": "the name of the directory containing a dataset loading script. (if specified, ignore file_name and cloud_file_name)",
+  "cloud_file_name": "the name of the dataset file in s3/gcs cloud storage. (if specified, ignore file_name)",
   "file_name": "the name of the dataset folder or dataset file in this directory. (required if above are not specified)",
   "formatting": "the format of the dataset. (optional, default: alpaca, can be chosen from {alpaca, sharegpt})",
   "ranking": "whether the dataset is a preference dataset or not. (default: False)",
   "subset": "the name of the subset. (optional, default: None)",
+  "split": "the name of dataset split to be used. (optional, default: train)",
   "folder": "the name of the folder of the dataset repository on the Hugging Face hub. (optional, default: None)",
-  "num_samples": "the number of samples in the dataset used for training. (optional, default: None)",
+  "num_samples": "the number of samples in the dataset to be used. (optional, default: None)",
   "columns (optional)": {
     "prompt": "the column name in the dataset containing the prompts. (default: instruction)",
     "query": "the column name in the dataset containing the queries. (default: input)",
@@ -22,6 +26,8 @@ Currently we support datasets in **alpaca** and **sharegpt** format.
     "system": "the column name in the dataset containing the system prompts. (default: None)",
     "tools": "the column name in the dataset containing the tool description. (default: None)",
     "images": "the column name in the dataset containing the image inputs. (default: None)",
+    "videos": "the column name in the dataset containing the videos inputs. (default: None)",
+    "audios": "the column name in the dataset containing the audios inputs. (default: None)",
     "chosen": "the column name in the dataset containing the chosen answers. (default: None)",
     "rejected": "the column name in the dataset containing the rejected answers. (default: None)",
     "kto_tag": "the column name in the dataset containing the kto tags. (default: None)"
@@ -44,7 +50,9 @@ Currently we support datasets in **alpaca** and **sharegpt** format.
 
 * [Example dataset](alpaca_en_demo.json)
 
-In supervised fine-tuning, the `instruction` column will be concatenated with the `input` column and used as the human prompt, then the human prompt would be `instruction\ninput`. The `output` column represents the model response.
+In supervised fine-tuning, the `instruction` column will be concatenated with the `input` column and used as the user prompt, then the user prompt would be `instruction\ninput`. The `output` column represents the model response.
+
+For reasoning models, if the dataset contains chain-of-thought (CoT), the CoT needs to be placed in the model responses, such as `<think>cot</think>output`.
 
 The `system` column will be used as the system prompt if specified.
 
@@ -53,13 +61,13 @@ The `history` column is a list consisting of string tuples representing prompt-r
 ```json
 [
   {
-    "instruction": "human instruction (required)",
-    "input": "human input (optional)",
+    "instruction": "user instruction (required)",
+    "input": "user input (optional)",
     "output": "model response (required)",
     "system": "system prompt (optional)",
     "history": [
-      ["human instruction in the first round (optional)", "model response in the first round (optional)"],
-      ["human instruction in the second round (optional)", "model response in the second round (optional)"]
+      ["user instruction in the first round (optional)", "model response in the first round (optional)"],
+      ["user instruction in the second round (optional)", "model response in the second round (optional)"]
     ]
   }
 ]
@@ -80,9 +88,14 @@ Regarding the above dataset, the *dataset description* in `dataset_info.json` sh
 }
 ```
 
+> [!TIP]  
+> If the model has reasoning capabilities (e.g. Qwen3) but the dataset does not contain chain-of-thought (CoT), LLaMA-Factory will automatically add empty CoT to the data. When `enable_thinking` is `True` (slow thinking, by default), the empty CoT will be added to the model responses and loss computation will be considered; otherwise (fast thinking), it will be added to the user prompts and loss computation will be ignored. Please keep the `enable_thinking` parameter consistent during training and inference.
+>
+> If you want to train data containing CoT with slow thinking and data without CoT with fast thinking, you can set `enable_thinking` to `None`. However, this feature is relatively complicated and should be used with caution.
+
 ### Pre-training Dataset
 
-- [Example dataset](c4_demo.json)
+- [Example dataset](c4_demo.jsonl)
 
 In pre-training, only the `text` column will be used for model learning.
 
@@ -106,15 +119,15 @@ Regarding the above dataset, the *dataset description* in `dataset_info.json` sh
 
 ### Preference Dataset
 
-Preference datasets are used for reward modeling, DPO training and ORPO training.
+Preference datasets are used for reward modeling, DPO training, ORPO and SimPO training.
 
 It requires a better response in `chosen` column and a worse response in `rejected` column.
 
 ```json
 [
   {
-    "instruction": "human instruction (required)",
-    "input": "human input (optional)",
+    "instruction": "user instruction (required)",
+    "input": "user input (optional)",
     "chosen": "chosen answer (required)",
     "rejected": "rejected answer (required)"
   }
@@ -138,67 +151,19 @@ Regarding the above dataset, the *dataset description* in `dataset_info.json` sh
 
 ### KTO Dataset
 
-- [Example dataset](kto_en_demo.json)
+An additional column `kto_tag` is required. Please refer to the [sharegpt](#sharegpt-format) format for details.
 
-KTO datasets require a extra `kto_tag` column containing the boolean human feedback.
+### Multimodal Image Dataset
 
-```json
-[
-  {
-    "instruction": "human instruction (required)",
-    "input": "human input (optional)",
-    "output": "model response (required)",
-    "kto_tag": "human feedback [true/false] (required)"
-  }
-]
-```
+An additional column `images` is required. Please refer to the [sharegpt](#sharegpt-format) format for details.
 
-Regarding the above dataset, the *dataset description* in `dataset_info.json` should be:
+### Multimodal Video Dataset
 
-```json
-"dataset_name": {
-  "file_name": "data.json",
-  "columns": {
-    "prompt": "instruction",
-    "query": "input",
-    "response": "output",
-    "kto_tag": "kto_tag"
-  }
-}
-```
+An additional column `videos` is required. Please refer to the [sharegpt](#sharegpt-format) format for details.
 
-### Multimodal Dataset
+### Multimodal Audio Dataset
 
-- [Example dataset](mllm_demo.json)
-
-Multimodal datasets require a `images` column containing the paths to the input images. Currently we only support one image.
-
-```json
-[
-  {
-    "instruction": "human instruction (required)",
-    "input": "human input (optional)",
-    "output": "model response (required)",
-    "images": [
-      "image path (required)"
-    ]
-  }
-]
-```
-
-Regarding the above dataset, the *dataset description* in `dataset_info.json` should be:
-
-```json
-"dataset_name": {
-  "file_name": "data.json",
-  "columns": {
-    "prompt": "instruction",
-    "query": "input",
-    "response": "output",
-    "images": "images"
-  }
-}
-```
+An additional column `audios` is required. Please refer to the [sharegpt](#sharegpt-format) format for details.
 
 ## Sharegpt Format
 
@@ -216,7 +181,7 @@ Note that the human and observation should appear in odd positions, while gpt an
     "conversations": [
       {
         "from": "human",
-        "value": "human instruction"
+        "value": "user instruction"
       },
       {
         "from": "function_call",
@@ -251,6 +216,10 @@ Regarding the above dataset, the *dataset description* in `dataset_info.json` sh
 }
 ```
 
+### Pre-training Dataset
+
+Not yet supported, please use the [alpaca](#alpaca-format) format.
+
 ### Preference Dataset
 
 - [Example dataset](dpo_en_demo.json)
@@ -263,7 +232,7 @@ Preference datasets in sharegpt format also require a better message in `chosen`
     "conversations": [
       {
         "from": "human",
-        "value": "human instruction"
+        "value": "user instruction"
       },
       {
         "from": "gpt",
@@ -271,7 +240,7 @@ Preference datasets in sharegpt format also require a better message in `chosen`
       },
       {
         "from": "human",
-        "value": "human instruction"
+        "value": "user instruction"
       }
     ],
     "chosen": {
@@ -301,6 +270,166 @@ Regarding the above dataset, the *dataset description* in `dataset_info.json` sh
 }
 ```
 
+### KTO Dataset
+
+- [Example dataset](kto_en_demo.json)
+
+KTO datasets require a extra `kto_tag` column containing the boolean human feedback.
+
+```json
+[
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "user instruction"
+      },
+      {
+        "from": "gpt",
+        "value": "model response"
+      }
+    ],
+    "kto_tag": "human feedback [true/false] (required)"
+  }
+]
+```
+
+Regarding the above dataset, the *dataset description* in `dataset_info.json` should be:
+
+```json
+"dataset_name": {
+  "file_name": "data.json",
+  "formatting": "sharegpt",
+  "columns": {
+    "messages": "conversations",
+    "kto_tag": "kto_tag"
+  }
+}
+```
+
+### Multimodal Image Dataset
+
+- [Example dataset](mllm_demo.json)
+
+Multimodal image datasets require an `images` column containing the paths to the input images.
+
+The number of images should be identical to the `<image>` tokens in the conversations.
+
+```json
+[
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<image>user instruction"
+      },
+      {
+        "from": "gpt",
+        "value": "model response"
+      }
+    ],
+    "images": [
+      "image path (required)"
+    ]
+  }
+]
+```
+
+Regarding the above dataset, the *dataset description* in `dataset_info.json` should be:
+
+```json
+"dataset_name": {
+  "file_name": "data.json",
+  "formatting": "sharegpt",
+  "columns": {
+    "messages": "conversations",
+    "images": "images"
+  }
+}
+```
+
+### Multimodal Video Dataset
+
+- [Example dataset](mllm_video_demo.json)
+
+Multimodal video datasets require a `videos` column containing the paths to the input videos.
+
+The number of videos should be identical to the `<video>` tokens in the conversations.
+
+```json
+[
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>user instruction"
+      },
+      {
+        "from": "gpt",
+        "value": "model response"
+      }
+    ],
+    "videos": [
+      "video path (required)"
+    ]
+  }
+]
+```
+
+Regarding the above dataset, the *dataset description* in `dataset_info.json` should be:
+
+```json
+"dataset_name": {
+  "file_name": "data.json",
+  "formatting": "sharegpt",
+  "columns": {
+    "messages": "conversations",
+    "videos": "videos"
+  }
+}
+```
+
+### Multimodal Audio Dataset
+
+- [Example dataset](mllm_audio_demo.json)
+
+Multimodal audio datasets require an `audios` column containing the paths to the input audios.
+
+The number of audios should be identical to the `<audio>` tokens in the conversations.
+
+```json
+[
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<audio>user instruction"
+      },
+      {
+        "from": "gpt",
+        "value": "model response"
+      }
+    ],
+    "audios": [
+      "audio path (required)"
+    ]
+  }
+]
+```
+
+Regarding the above dataset, the *dataset description* in `dataset_info.json` should be:
+
+```json
+"dataset_name": {
+  "file_name": "data.json",
+  "formatting": "sharegpt",
+  "columns": {
+    "messages": "conversations",
+    "audios": "audios"
+  }
+}
+```
+
 ### OpenAI Format
 
 The openai format is simply a special case of the sharegpt format, where the first message may be a system prompt.
@@ -315,7 +444,7 @@ The openai format is simply a special case of the sharegpt format, where the fir
       },
       {
         "role": "user",
-        "content": "human instruction"
+        "content": "user instruction"
       },
       {
         "role": "assistant",
@@ -344,7 +473,3 @@ Regarding the above dataset, the *dataset description* in `dataset_info.json` sh
   }
 }
 ```
-
-The KTO datasets and multimodal datasets in sharegpt format are similar to the alpaca format.
-
-Pre-training datasets are **incompatible** with the sharegpt format.

data/README_zh.md

@@ -1,6 +1,8 @@
 [dataset_info.json](dataset_info.json) 包含了所有可用的数据集。如果您希望使用自定义数据集，请**务必**在 `dataset_info.json` 文件中添加*数据集描述*，并通过修改 `dataset: 数据集名称` 配置来使用数据集。
 
-目前我们支持 **alpaca** 格式和 **sharegpt** 格式的数据集。
+其中 `dataset_info.json` 文件应放置在 `dataset_dir` 目录下。您可以通过修改 `dataset_dir` 参数来使用其他目录。默认值为 `./data`。
+
+目前我们支持 **alpaca** 格式和 **sharegpt** 格式的数据集。允许的文件类型包括 json、jsonl、csv、parquet 和 arrow。
 
 ```json
 "数据集名称": {
@@ -11,8 +13,9 @@
   "formatting": "数据集格式（可选，默认：alpaca，可以为 alpaca 或 sharegpt）",
   "ranking": "是否为偏好数据集（可选，默认：False）",
   "subset": "数据集子集的名称（可选，默认：None）",
+  "split": "所使用的数据集切分（可选，默认：train）",
   "folder": "Hugging Face 仓库的文件夹名称（可选，默认：None）",
-  "num_samples": "该数据集中用于训练的样本数量。（可选，默认：None）",
+  "num_samples": "该数据集所使用的样本数量。（可选，默认：None）",
   "columns（可选）": {
     "prompt": "数据集代表提示词的表头名称（默认：instruction）",
     "query": "数据集代表请求的表头名称（默认：input）",
@@ -22,6 +25,8 @@
     "system": "数据集代表系统提示的表头名称（默认：None）",
     "tools": "数据集代表工具描述的表头名称（默认：None）",
     "images": "数据集代表图像输入的表头名称（默认：None）",
+    "videos": "数据集代表视频输入的表头名称（默认：None）",
+    "audios": "数据集代表音频输入的表头名称（默认：None）",
     "chosen": "数据集代表更优回答的表头名称（默认：None）",
     "rejected": "数据集代表更差回答的表头名称（默认：None）",
     "kto_tag": "数据集代表 KTO 标签的表头名称（默认：None）"
@@ -44,7 +49,9 @@
 
 - [样例数据集](alpaca_zh_demo.json)
 
-在指令监督微调时，`instruction` 列对应的内容会与 `input` 列对应的内容拼接后作为人类指令，即人类指令为 `instruction\ninput`。而 `output` 列对应的内容为模型回答。
+在指令监督微调时，`instruction` 列对应的内容会与 `input` 列对应的内容拼接后作为提示词，即提示词为 `instruction\ninput`。而 `output` 列对应的内容为模型回答。
+
+对于推理类模型的微调，如果数据集包含思维链，则需要把思维链放在模型回答中，例如 `<think>cot</think>output`。
 
 如果指定，`system` 列对应的内容将被作为系统提示词。
 
@@ -53,8 +60,8 @@
 ```json
 [
   {
-    "instruction": "人类指令（必填）",
-    "input": "人类输入（选填）",
+    "instruction": "用户指令（必填）",
+    "input": "用户输入（选填）",
     "output": "模型回答（必填）",
     "system": "系统提示词（选填）",
     "history": [
@@ -80,9 +87,14 @@
 }
 ```
 
+> [!TIP]
+> 如果模型本身具备推理能力（如 Qwen3）而数据集不包含思维链，LLaMA-Factory 会自动为数据添加空思维链。当 `enable_thinking` 为 `True` 时（慢思考，默认），空思维链会添加到模型回答中并且计算损失，否则会添加到用户指令中并且不计算损失（快思考）。请在训练和推理时保持 `enable_thinking` 参数一致。
+>
+> 如果您希望训练包含思维链的数据时使用慢思考，训练不包含思维链的数据时使用快思考，可以设置 `enable_thinking` 为 `None`。但该功能较为复杂，请谨慎使用。
+
 ### 预训练数据集
 
-- [样例数据集](c4_demo.json)
+- [样例数据集](c4_demo.jsonl)
 
 在预训练时，只有 `text` 列中的内容会用于模型学习。
 
@@ -106,15 +118,15 @@
 
 ### 偏好数据集
 
-偏好数据集用于奖励模型训练、DPO 训练和 ORPO 训练。
+偏好数据集用于奖励模型训练、DPO 训练、ORPO 训练和 SimPO 训练。
 
 它需要在 `chosen` 列中提供更优的回答，并在 `rejected` 列中提供更差的回答。
 
 ```json
 [
   {
-    "instruction": "人类指令（必填）",
-    "input": "人类输入（选填）",
+    "instruction": "用户指令（必填）",
+    "input": "用户输入（选填）",
     "chosen": "优质回答（必填）",
     "rejected": "劣质回答（必填）"
   }
@@ -138,67 +150,19 @@
 
 ### KTO 数据集
 
-- [样例数据集](kto_en_demo.json)
+KTO 数据集需要提供额外的 `kto_tag` 列。详情请参阅 [sharegpt](#sharegpt-格式)。
 
-KTO 数据集需要额外添加一个 `kto_tag` 列，包含 bool 类型的人类反馈。
+### 多模态图像数据集
 
-```json
-[
-  {
-    "instruction": "人类指令（必填）",
-    "input": "人类输入（选填）",
-    "output": "模型回答（必填）",
-    "kto_tag": "人类反馈 [true/false]（必填）"
-  }
-]
-```
+多模态图像数据集需要提供额外的 `images` 列。详情请参阅 [sharegpt](#sharegpt-格式)。
 
-对于上述格式的数据，`dataset_info.json` 中的*数据集描述*应为：
+### 多模态视频数据集
 
-```json
-"数据集名称": {
-  "file_name": "data.json",
-  "columns": {
-    "prompt": "instruction",
-    "query": "input",
-    "response": "output",
-    "kto_tag": "kto_tag"
-  }
-}
-```
+多模态视频数据集需要提供额外的 `videos` 列。详情请参阅 [sharegpt](#sharegpt-格式)。
 
-### 多模态数据集
+### 多模态音频数据集
 
-- [样例数据集](mllm_demo.json)
-
-多模态数据集需要额外添加一个 `images` 列，包含输入图像的路径。目前我们仅支持单张图像输入。
-
-```json
-[
-  {
-    "instruction": "人类指令（必填）",
-    "input": "人类输入（选填）",
-    "output": "模型回答（必填）",
-    "images": [
-      "图像路径（必填）"
-    ]
-  }
-]
-```
-
-对于上述格式的数据，`dataset_info.json` 中的*数据集描述*应为：
-
-```json
-"数据集名称": {
-  "file_name": "data.json",
-  "columns": {
-    "prompt": "instruction",
-    "query": "input",
-    "response": "output",
-    "images": "images"
-  }
-}
-```
+多模态音频数据集需要提供额外的 `audios` 列。详情请参阅 [sharegpt](#sharegpt-格式)。
 
 ## Sharegpt 格式
 
@@ -216,7 +180,7 @@ KTO 数据集需要额外添加一个 `kto_tag` 列，包含 bool 类型的人
     "conversations": [
       {
         "from": "human",
-        "value": "人类指令"
+        "value": "用户指令"
       },
       {
         "from": "function_call",
@@ -251,6 +215,10 @@ KTO 数据集需要额外添加一个 `kto_tag` 列，包含 bool 类型的人
 }
 ```
 
+### 预训练数据集
+
+尚不支持，请使用 [alpaca](#alpaca-格式) 格式。
+
 ### 偏好数据集
 
 - [样例数据集](dpo_zh_demo.json)
@@ -263,7 +231,7 @@ Sharegpt 格式的偏好数据集同样需要在 `chosen` 列中提供更优的
     "conversations": [
       {
         "from": "human",
-        "value": "人类指令"
+        "value": "用户指令"
       },
       {
         "from": "gpt",
@@ -271,7 +239,7 @@ Sharegpt 格式的偏好数据集同样需要在 `chosen` 列中提供更优的
       },
       {
         "from": "human",
-        "value": "人类指令"
+        "value": "用户指令"
       }
     ],
     "chosen": {
@@ -301,6 +269,170 @@ Sharegpt 格式的偏好数据集同样需要在 `chosen` 列中提供更优的
 }
 ```
 
+### KTO 数据集
+
+- [样例数据集](kto_en_demo.json)
+
+KTO 数据集需要额外添加一个 `kto_tag` 列，包含 bool 类型的人类反馈。
+
+```json
+[
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "用户指令"
+      },
+      {
+        "from": "gpt",
+        "value": "模型回答"
+      }
+    ],
+    "kto_tag": "人类反馈 [true/false]（必填）"
+  }
+]
+```
+
+对于上述格式的数据，`dataset_info.json` 中的*数据集描述*应为：
+
+```json
+"数据集名称": {
+  "file_name": "data.json",
+  "formatting": "sharegpt",
+  "columns": {
+    "messages": "conversations",
+    "kto_tag": "kto_tag"
+  }
+}
+```
+
+### 多模态图像数据集
+
+- [样例数据集](mllm_demo.json)
+
+多模态图像数据集需要额外添加一个 `images` 列，包含输入图像的路径。
+
+注意图片的数量必须与文本中所有 `<image>` 标记的数量严格一致。
+
+```json
+[
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<image><image>用户指令"
+      },
+      {
+        "from": "gpt",
+        "value": "模型回答"
+      }
+    ],
+    "images": [
+      "图像路径（必填）",
+      "图像路径（必填）"
+    ]
+  }
+]
+```
+
+对于上述格式的数据，`dataset_info.json` 中的*数据集描述*应为：
+
+```json
+"数据集名称": {
+  "file_name": "data.json",
+  "formatting": "sharegpt",
+  "columns": {
+    "messages": "conversations",
+    "images": "images"
+  }
+}
+```
+
+### 多模态视频数据集
+
+- [样例数据集](mllm_video_demo.json)
+
+多模态视频数据集需要额外添加一个 `videos` 列，包含输入视频的路径。
+
+注意视频的数量必须与文本中所有 `<video>` 标记的数量严格一致。
+
+```json
+[
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video><video>用户指令"
+      },
+      {
+        "from": "gpt",
+        "value": "模型回答"
+      }
+    ],
+    "videos": [
+      "视频路径（必填）",
+      "视频路径（必填）"
+    ]
+  }
+]
+```
+
+对于上述格式的数据，`dataset_info.json` 中的*数据集描述*应为：
+
+```json
+"数据集名称": {
+  "file_name": "data.json",
+  "formatting": "sharegpt",
+  "columns": {
+    "messages": "conversations",
+    "videos": "videos"
+  }
+}
+```
+
+### 多模态音频数据集
+
+- [样例数据集](mllm_audio_demo.json)
+
+多模态音频数据集需要额外添加一个 `audios` 列，包含输入音频的路径。
+
+注意音频的数量必须与文本中所有 `<audio>` 标记的数量严格一致。
+
+```json
+[
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<audio><audio>用户指令"
+      },
+      {
+        "from": "gpt",
+        "value": "模型回答"
+      }
+    ],
+    "audios": [
+      "音频路径（必填）",
+      "音频路径（必填）"
+    ]
+  }
+]
+```
+
+对于上述格式的数据，`dataset_info.json` 中的*数据集描述*应为：
+
+```json
+"数据集名称": {
+  "file_name": "data.json",
+  "formatting": "sharegpt",
+  "columns": {
+    "messages": "conversations",
+    "audios": "audios"
+  }
+}
+```
+
+
 ### OpenAI 格式
 
 OpenAI 格式仅仅是 sharegpt 格式的一种特殊情况，其中第一条消息可能是系统提示词。
@@ -315,7 +447,7 @@ OpenAI 格式仅仅是 sharegpt 格式的一种特殊情况，其中第一条消
       },
       {
         "role": "user",
-        "content": "人类指令"
+        "content": "用户指令"
       },
       {
         "role": "assistant",
@@ -344,7 +476,3 @@ OpenAI 格式仅仅是 sharegpt 格式的一种特殊情况，其中第一条消
   }
 }
 ```
-
-Sharegpt 格式中的 KTO 数据集和多模态数据集与 alpaca 格式的类似。
-
-预训练数据集**不支持** sharegpt 格式。

data/belle_multiturn/belle_multiturn.py

@@ -1,3 +1,18 @@
+# Copyright 2025 the LlamaFactory team.
+# Copyright 2020 The HuggingFace Datasets Authors and the current dataset script contributor.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
 import json
 import os
 
@@ -10,16 +25,16 @@ _DESCRIPTION = "BELLE multiturn chat dataset."
 
 _CITATION = """\
 @article{belle2023exploring,
-  title={Exploring the Impact of Instruction Data Scaling on Large Language Models: An Empirical Study on Real-World Use Cases},
+  title={Exploring the Impact of Instruction Data Scaling on Large Language Models},
   author={Yunjie Ji, Yong Deng, Yan Gong, Yiping Peng, Qiang Niu, Lei Zhang, Baochang Ma, Xiangang Li},
   journal={arXiv preprint arXiv:2303.14742},
   year={2023}
 }
 """
 
-_HOMEPAGE = "{}/datasets/BelleGroup/multiturn_chat_0.8M".format(_HF_ENDPOINT)
+_HOMEPAGE = f"{_HF_ENDPOINT}/datasets/BelleGroup/multiturn_chat_0.8M"
 _LICENSE = "gpl-3.0"
-_URL = "{}/datasets/BelleGroup/multiturn_chat_0.8M/resolve/main/multiturn_chat_0.8M.json".format(_HF_ENDPOINT)
+_URL = f"{_HF_ENDPOINT}/datasets/BelleGroup/multiturn_chat_0.8M/resolve/main/multiturn_chat_0.8M.json"
 
 
 class BelleMultiturn(datasets.GeneratorBasedBuilder):
@@ -38,7 +53,7 @@ class BelleMultiturn(datasets.GeneratorBasedBuilder):
         return [datasets.SplitGenerator(name=datasets.Split.TRAIN, gen_kwargs={"filepath": file_path})]
 
     def _generate_examples(self, filepath: str):
-        with open(filepath, "r", encoding="utf-8") as f:
+        with open(filepath, encoding="utf-8") as f:
             for key, row in enumerate(f):
                 data = json.loads(row)
                 conversations = []

data/hh_rlhf_en/hh_rlhf_en.py

@@ -1,6 +1,20 @@
+# Copyright 2025 the LlamaFactory team.
+# Copyright 2020 The HuggingFace Datasets Authors and the current dataset script contributor.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
 import json
 import os
-from typing import List
 
 import datasets
 
@@ -8,9 +22,9 @@ import datasets
 _HF_ENDPOINT = os.getenv("HF_ENDPOINT", "https://huggingface.co")
 _DESCRIPTION = "Human preference data about helpfulness and harmlessness."
 _CITATION = ""
-_HOMEPAGE = "{}/datasets/Anthropic/hh-rlhf".format(_HF_ENDPOINT)
+_HOMEPAGE = f"{_HF_ENDPOINT}/datasets/Anthropic/hh-rlhf"
 _LICENSE = "mit"
-_URL = "{}/datasets/Anthropic/hh-rlhf/resolve/main/".format(_HF_ENDPOINT)
+_URL = f"{_HF_ENDPOINT}/datasets/Anthropic/hh-rlhf/resolve/main/"
 _URLS = {
     "train": [
         _URL + "harmless-base/train.jsonl.gz",
@@ -50,10 +64,10 @@ class HhRlhfEn(datasets.GeneratorBasedBuilder):
             datasets.SplitGenerator(name=datasets.Split.TEST, gen_kwargs={"filepaths": file_path["test"]}),
         ]
 
-    def _generate_examples(self, filepaths: List[str]):
+    def _generate_examples(self, filepaths: list[str]):
         key = 0
         for filepath in filepaths:
-            with open(filepath, "r", encoding="utf-8") as f:
+            with open(filepath, encoding="utf-8") as f:
                 for row in f:
                     data = json.loads(row)
                     chosen = data["chosen"]

data/ultra_chat/ultra_chat.py

@@ -1,6 +1,20 @@
+# Copyright 2025 the LlamaFactory team.
+# Copyright 2020 The HuggingFace Datasets Authors and the current dataset script contributor.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
 import json
 import os
-from typing import List
 
 import datasets
 
@@ -11,7 +25,7 @@ _DESCRIPTION = "UltraChat: Large-scale, Informative, and Diverse Multi-round Dia
 
 _CITATION = """\
 @misc{UltraChat,
-  author = {Ding, Ning and Chen, Yulin and Xu, Bokai and Hu, Shengding and Qin, Yujia and Liu, Zhiyuan and Sun, Maosong and Zhou, Bowen},
+  author = {Ding, Ning and Chen, Yulin and Xu, Bokai and Hu, Shengding and others},
   title = {UltraChat: A Large-scale Auto-generated Multi-round Dialogue Data},
   year = {2023},
   publisher = {GitHub},
@@ -20,9 +34,9 @@ _CITATION = """\
 }
 """
 
-_HOMEPAGE = "{}/datasets/stingning/ultrachat".format(_HF_ENDPOINT)
+_HOMEPAGE = f"{_HF_ENDPOINT}/datasets/stingning/ultrachat"
 _LICENSE = "cc-by-nc-4.0"
-_BASE_DATA_URL = "{}/datasets/stingning/ultrachat/resolve/main/train_{{idx}}.jsonl".format(_HF_ENDPOINT)
+_BASE_DATA_URL = f"{_HF_ENDPOINT}/datasets/stingning/ultrachat/resolve/main/train_{{idx}}.jsonl"
 
 
 class UltraChat(datasets.GeneratorBasedBuilder):
@@ -40,16 +54,16 @@ class UltraChat(datasets.GeneratorBasedBuilder):
         file_paths = [dl_manager.download(_BASE_DATA_URL.format(idx=idx)) for idx in range(10)]  # multiple shards
         return [datasets.SplitGenerator(name=datasets.Split.TRAIN, gen_kwargs={"filepaths": file_paths})]
 
-    def _generate_examples(self, filepaths: List[str]):
+    def _generate_examples(self, filepaths: list[str]):
         for filepath in filepaths:
-            with open(filepath, "r", encoding="utf-8") as f:
+            with open(filepath, encoding="utf-8") as f:
                 for row in f:
                     try:
                         data = json.loads(row)
                     except Exception:
                         continue
                     key: int = data["id"]
-                    content: List[str] = data["data"]
+                    content: list[str] = data["data"]
                     if len(content) % 2 == 1:
                         content.pop(-1)
                     if len(content) < 2:

docker/docker-cuda/Dockerfile

@@ -0,0 +1,66 @@
+# https://hub.docker.com/r/hiyouga/pytorch/tags
+ARG BASE_IMAGE=hiyouga/pytorch:th2.6.0-cu124-flashattn2.7.4-cxx11abi0-devel
+FROM ${BASE_IMAGE}
+
+# Installation arguments
+ARG PIP_INDEX=https://pypi.org/simple
+ARG EXTRAS=metrics
+ARG INSTALL_FLASHATTN=false
+ARG HTTP_PROXY=""
+
+# Define environments
+ENV MAX_JOBS=16
+ENV FLASH_ATTENTION_FORCE_BUILD=TRUE
+ENV VLLM_WORKER_MULTIPROC_METHOD=spawn
+ENV DEBIAN_FRONTEND=noninteractive
+ENV NODE_OPTIONS=""
+ENV PIP_ROOT_USER_ACTION=ignore
+ENV http_proxy="${HTTP_PROXY}"
+ENV https_proxy="${HTTP_PROXY}"
+
+# Use Bash instead of default /bin/sh
+SHELL ["/bin/bash", "-c"]
+
+# Set the working directory
+WORKDIR /app
+
+# Change pip source
+RUN pip config set global.index-url "${PIP_INDEX}" && \
+    pip config set global.extra-index-url "${PIP_INDEX}" && \
+    pip install --no-cache-dir --upgrade pip packaging wheel setuptools
+
+# Install the requirements
+COPY requirements.txt /app
+RUN pip install --no-cache-dir -r requirements.txt
+
+# Copy the rest of the application into the image
+COPY . /app
+
+# Install LLaMA Factory
+RUN pip install --no-cache-dir -e ".[${EXTRAS}]" --no-build-isolation
+
+# Rebuild flash attention
+RUN if [ "${INSTALL_FLASHATTN}" == "true" ]; then \
+        pip uninstall -y ninja && \
+        pip install --no-cache-dir ninja && \
+        pip install --no-cache-dir flash-attn --no-build-isolation; \
+    fi
+
+# Set up volumes
+# VOLUME [ "/root/.cache/huggingface", "/app/shared_data", "/app/output" ]
+
+# Expose port 7860 for LLaMA Board
+ENV GRADIO_SERVER_PORT=7860
+EXPOSE 7860
+
+# Expose port 8000 for API service
+ENV API_PORT=8000
+EXPOSE 8000
+
+# unset proxy
+ENV http_proxy=
+ENV https_proxy=
+
+# Reset pip config
+RUN pip config unset global.index-url && \
+    pip config unset global.extra-index-url

docker/docker-cuda/Dockerfile.base

@@ -0,0 +1,55 @@
+# Start from the pytorch official image (ubuntu-22.04 + cuda-12.4.1 + python-3.11)
+# https://hub.docker.com/r/pytorch/pytorch/tags
+FROM pytorch/pytorch:2.6.0-cuda12.4-cudnn9-devel
+
+# Define environments
+ENV MAX_JOBS=16
+ENV VLLM_WORKER_MULTIPROC_METHOD=spawn
+ENV DEBIAN_FRONTEND=noninteractive
+ENV NODE_OPTIONS=""
+ENV PIP_ROOT_USER_ACTION=ignore
+
+# Define installation arguments
+ARG APT_SOURCE=https://mirrors.tuna.tsinghua.edu.cn/ubuntu/
+ARG PIP_INDEX=https://mirrors.tuna.tsinghua.edu.cn/pypi/web/simple
+
+# Set apt source
+RUN cp /etc/apt/sources.list /etc/apt/sources.list.bak && \
+    { \
+    echo "deb ${APT_SOURCE} jammy main restricted universe multiverse"; \
+    echo "deb ${APT_SOURCE} jammy-updates main restricted universe multiverse"; \
+    echo "deb ${APT_SOURCE} jammy-backports main restricted universe multiverse"; \
+    echo "deb ${APT_SOURCE} jammy-security main restricted universe multiverse"; \
+    } > /etc/apt/sources.list
+
+# Install systemctl and wget
+RUN apt-get update && \
+    apt-get install -y -o Dpkg::Options::="--force-confdef" systemd wget && \
+    apt-get clean
+
+# Install git and vim
+RUN apt-get update && \
+    apt-get install -y git vim && \
+    apt-get clean
+
+# Install gcc and g++
+RUN apt-get update && \
+    apt-get install -y gcc g++ && \
+    apt-get clean
+
+# Change pip source
+RUN pip config set global.index-url "${PIP_INDEX}" && \
+    pip config set global.extra-index-url "${PIP_INDEX}" && \
+    pip install --no-cache-dir --upgrade pip packaging wheel setuptools
+
+# Install flash-attn-2.7.4.post1 (cxx11abi=False)
+RUN wget -nv https://github.com/Dao-AILab/flash-attention/releases/download/v2.7.4.post1/flash_attn-2.7.4.post1+cu12torch2.6cxx11abiFALSE-cp311-cp311-linux_x86_64.whl && \
+    pip install --no-cache-dir flash_attn-2.7.4.post1+cu12torch2.6cxx11abiFALSE-cp311-cp311-linux_x86_64.whl
+
+# Install flashinfer-0.2.2.post1+cu124 (cxx11abi=False)
+RUN wget -nv https://github.com/flashinfer-ai/flashinfer/releases/download/v0.2.2.post1/flashinfer_python-0.2.2.post1+cu124torch2.6-cp38-abi3-linux_x86_64.whl && \
+    pip install --no-cache-dir flashinfer_python-0.2.2.post1+cu124torch2.6-cp38-abi3-linux_x86_64.whl
+
+# Reset pip config
+RUN pip config unset global.index-url && \
+    pip config unset global.extra-index-url

docker/docker-cuda/docker-compose.yml

@@ -1,23 +1,18 @@
 services:
   llamafactory:
     build:
-      dockerfile: Dockerfile
-      context: .
+      dockerfile: ./docker/docker-cuda/Dockerfile
+      context: ../..
       args:
-        INSTALL_BNB: false
-        INSTALL_VLLM: false
-        INSTALL_DEEPSPEED: false
         PIP_INDEX: https://pypi.org/simple
+        EXTRAS: metrics
     container_name: llamafactory
-    volumes:
-      - ./hf_cache:/root/.cache/huggingface/
-      - ./data:/app/data
-      - ./output:/app/output
     ports:
       - "7860:7860"
       - "8000:8000"
     ipc: host
     tty: true
+    # shm_size: "16gb"  # ipc: host is set
     stdin_open: true
     command: bash
     deploy:

docker/docker-npu/Dockerfile

@@ -0,0 +1,58 @@
+# https://hub.docker.com/r/ascendai/cann/tags
+ARG BASE_IMAGE=ascendai/cann:8.0.0-910b-ubuntu22.04-py3.11
+FROM ${BASE_IMAGE}
+
+# Installation arguments
+ARG PIP_INDEX=https://pypi.org/simple
+ARG EXTRAS=torch-npu,metrics
+ARG HTTP_PROXY=""
+
+# Define environments
+ENV MAX_JOBS=16
+ENV FLASH_ATTENTION_FORCE_BUILD=TRUE
+ENV VLLM_WORKER_MULTIPROC_METHOD=spawn
+ENV DEBIAN_FRONTEND=noninteractive
+ENV NODE_OPTIONS=""
+ENV PIP_ROOT_USER_ACTION=ignore
+ENV http_proxy="${HTTP_PROXY}"
+ENV https_proxy="${HTTP_PROXY}"
+
+# Use Bash instead of default /bin/sh
+SHELL ["/bin/bash", "-c"]
+
+# Set the working directory
+WORKDIR /app
+
+# Change pip source
+RUN pip config set global.index-url "${PIP_INDEX}" && \
+    pip config set global.extra-index-url "${PIP_INDEX}" && \
+    pip install --no-cache-dir --upgrade pip packaging wheel setuptools
+
+# Install the requirements
+COPY requirements.txt /app
+RUN pip install --no-cache-dir -r requirements.txt
+
+# Copy the rest of the application into the image
+COPY . /app
+
+# Install LLaMA Factory
+RUN pip install --no-cache-dir -e ".[${EXTRAS}]" --no-build-isolation
+
+# Set up volumes
+# VOLUME [ "/root/.cache/huggingface", "/app/shared_data", "/app/output" ]
+
+# Expose port 7860 for LLaMA Board
+ENV GRADIO_SERVER_PORT=7860
+EXPOSE 7860
+
+# Expose port 8000 for API service
+ENV API_PORT=8000
+EXPOSE 8000
+
+# unset proxy
+ENV http_proxy=
+ENV https_proxy=
+
+# Reset pip config
+RUN pip config unset global.index-url && \
+    pip config unset global.extra-index-url

docker/docker-npu/docker-compose.yml

@@ -0,0 +1,28 @@
+services:
+  llamafactory:
+    build:
+      dockerfile: ./docker/docker-npu/Dockerfile
+      context: ../..
+      args:
+        PIP_INDEX: https://pypi.org/simple
+        EXTRAS: torch-npu,metrics
+    container_name: llamafactory
+    volumes:
+      - /usr/local/dcmi:/usr/local/dcmi
+      - /usr/local/bin/npu-smi:/usr/local/bin/npu-smi
+      - /usr/local/Ascend/driver:/usr/local/Ascend/driver
+      - /etc/ascend_install.info:/etc/ascend_install.info
+    ports:
+      - "7860:7860"
+      - "8000:8000"
+    ipc: host
+    tty: true
+    # shm_size: "16gb"  # ipc: host is set
+    stdin_open: true
+    command: bash
+    devices:
+      - /dev/davinci0
+      - /dev/davinci_manager
+      - /dev/devmm_svm
+      - /dev/hisi_hdc
+    restart: unless-stopped

docker/docker-rocm/Dockerfile

@@ -0,0 +1,71 @@
+# https://hub.docker.com/r/rocm/pytorch/tags
+ARG BASE_IMAGE=rocm/pytorch:rocm6.4.1_ubuntu22.04_py3.10_pytorch_release_2.6.0
+FROM ${BASE_IMAGE}
+
+# Installation arguments
+ARG PIP_INDEX=https://pypi.org/simple
+ARG EXTRAS=metrics
+ARG INSTALL_FLASHATTN=false
+ARG HTTP_PROXY=""
+ARG PYTORCH_INDEX=https://download.pytorch.org/whl/rocm6.3
+
+# Define environments
+ENV MAX_JOBS=16
+ENV FLASH_ATTENTION_FORCE_BUILD=TRUE
+ENV VLLM_WORKER_MULTIPROC_METHOD=spawn
+ENV DEBIAN_FRONTEND=noninteractive
+ENV NODE_OPTIONS=""
+ENV PIP_ROOT_USER_ACTION=ignore
+ENV http_proxy="${HTTP_PROXY}"
+ENV https_proxy="${HTTP_PROXY}"
+
+# Use Bash instead of default /bin/sh
+SHELL ["/bin/bash", "-c"]
+
+# Set the working directory
+WORKDIR /app
+
+# Change pip source
+RUN pip config set global.index-url "${PIP_INDEX}" && \
+    pip config set global.extra-index-url "${PIP_INDEX}" && \
+    pip install --no-cache-dir --upgrade pip packaging wheel setuptools
+
+# Reinstall pytorch rocm
+RUN pip uninstall -y torch torchvision torchaudio && \
+    pip install --no-cache-dir --pre torch torchvision torchaudio --index-url "${PYTORCH_INDEX}"
+
+# Install the requirements
+COPY requirements.txt /app
+RUN pip install --no-cache-dir -r requirements.txt
+
+# Copy the rest of the application into the image
+COPY . /app
+
+# Install LLaMA Factory
+RUN pip install --no-cache-dir -e ".[${EXTRAS}]" --no-build-isolation
+
+# Rebuild flash attention
+RUN if [ "${INSTALL_FLASHATTN}" == "true" ]; then \
+        pip uninstall -y ninja && \
+        pip install --no-cache-dir ninja && \
+        pip install --no-cache-dir flash-attn --no-build-isolation; \
+    fi
+
+# Set up volumes
+# VOLUME [ "/root/.cache/huggingface", "/app/shared_data", "/app/output" ]
+
+# Expose port 7860 for LLaMA Board
+ENV GRADIO_SERVER_PORT=7860
+EXPOSE 7860
+
+# Expose port 8000 for API service
+ENV API_PORT=8000
+EXPOSE 8000
+
+# unset proxy
+ENV http_proxy=
+ENV https_proxy=
+
+# Reset pip config
+RUN pip config unset global.index-url && \
+    pip config unset global.extra-index-url

docker/docker-rocm/docker-compose.yml

@@ -0,0 +1,21 @@
+services:
+  llamafactory:
+    build:
+      dockerfile: ./docker/docker-rocm/Dockerfile
+      context: ../..
+      args:
+        PIP_INDEX: https://pypi.org/simple
+        EXTRAS: metrics
+    container_name: llamafactory
+    ports:
+      - "7860:7860"
+      - "8000:8000"
+    ipc: host
+    tty: true
+    # shm_size: "16gb"  # ipc: host is set
+    stdin_open: true
+    command: bash
+    devices:
+      - /dev/kfd:/dev/kfd
+      - /dev/dri:/dev/dri
+    restart: unless-stopped

evaluation/ceval/ceval.py

@@ -1,3 +1,4 @@
+# Copyright 2025 the LlamaFactory team.
 # Copyright 2020 The HuggingFace Datasets Authors and the current dataset script contributor.
 #
 # Licensed under the Apache License, Version 2.0 (the "License");
@@ -21,14 +22,15 @@ import pandas as pd
 _CITATION = """\
 @article{huang2023ceval,
   title={C-Eval: A Multi-Level Multi-Discipline Chinese Evaluation Suite for Foundation Models},
-  author={Huang, Yuzhen and Bai, Yuzhuo and Zhu, Zhihao and Zhang, Junlei and Zhang, Jinghan and Su, Tangjun and Liu, Junteng and Lv, Chuancheng and Zhang, Yikai and Lei, Jiayi and Fu, Yao and Sun, Maosong and He, Junxian},
+  author={Huang, Yuzhen and Bai, Yuzhuo and Zhu, Zhihao and others},
   journal={arXiv preprint arXiv:2305.08322},
   year={2023}
 }
 """
 
 _DESCRIPTION = """\
-C-Eval is a comprehensive Chinese evaluation suite for foundation models. It consists of 13948 multi-choice questions spanning 52 diverse disciplines and four difficulty levels.
+C-Eval is a comprehensive Chinese evaluation suite for foundation models.
+It consists of 13948 multi-choice questions spanning 52 diverse disciplines and four difficulty levels.
 """
 
 _HOMEPAGE = "https://cevalbenchmark.com"

evaluation/cmmlu/cmmlu.py

@@ -1,3 +1,4 @@
+# Copyright 2025 the LlamaFactory team.
 # Copyright 2020 The HuggingFace Datasets Authors and the current dataset script contributor.
 #
 # Licensed under the Apache License, Version 2.0 (the "License");
@@ -21,14 +22,15 @@ import pandas as pd
 _CITATION = """\
 @article{li2023cmmlu,
   title={CMMLU: Measuring massive multitask language understanding in Chinese},
-  author={Haonan Li and Yixuan Zhang and Fajri Koto and Yifei Yang and Hai Zhao and Yeyun Gong and Nan Duan and Timothy Baldwin},
+  author={Haonan Li and Yixuan Zhang and Fajri Koto and Yifei Yang and others,
   journal={arXiv preprint arXiv:2306.09212},
   year={2023}
 }
 """
 
 _DESCRIPTION = """\
-CMMLU is a comprehensive Chinese assessment suite specifically designed to evaluate the advanced knowledge and reasoning abilities of LLMs within the Chinese language and cultural context.
+CMMLU is a comprehensive Chinese assessment suite specifically designed to evaluate the advanced knowledge
+and reasoning abilities of LLMs within the Chinese language and cultural context.
 """
 
 _HOMEPAGE = "https://github.com/haonan-li/CMMLU"

evaluation/mmlu/mmlu.py

@@ -1,3 +1,4 @@
+# Copyright 2025 the LlamaFactory team.
 # Copyright 2020 The HuggingFace Datasets Authors and the current dataset script contributor.
 #
 # Licensed under the Apache License, Version 2.0 (the "License");
@@ -21,14 +22,15 @@ import pandas as pd
 _CITATION = """\
 @article{hendryckstest2021,
   title={Measuring Massive Multitask Language Understanding},
-  author={Dan Hendrycks and Collin Burns and Steven Basart and Andy Zou and Mantas Mazeika and Dawn Song and Jacob Steinhardt},
+  author={Dan Hendrycks and Collin Burns and others},
   journal={Proceedings of the International Conference on Learning Representations (ICLR)},
   year={2021}
 }
 """
 
 _DESCRIPTION = """\
-Measuring Massive Multitask Language Understanding by Dan Hendrycks, Collin Burns, Steven Basart, Andy Zou, Mantas Mazeika, Dawn Song, and Jacob Steinhardt (ICLR 2021).
+Measuring Massive Multitask Language Understanding by Dan Hendrycks, Collin Burns, Steven Basart,
+Andy Zou, Mantas Mazeika, Dawn Song, and Jacob Steinhardt (ICLR 2021).
 """
 
 _HOMEPAGE = "https://github.com/hendrycks/test"
@@ -158,5 +160,4 @@ class MMLU(datasets.GeneratorBasedBuilder):
         df = pd.read_csv(filepath, header=None)
         df.columns = ["question", "A", "B", "C", "D", "answer"]
 
-        for i, instance in enumerate(df.to_dict(orient="records")):
-            yield i, instance
+        yield from enumerate(df.to_dict(orient="records"))

examples/README.md

@@ -13,6 +13,26 @@ Make sure to execute these commands in the `LLaMA-Factory` directory.
 
 Use `CUDA_VISIBLE_DEVICES` (GPU) or `ASCEND_RT_VISIBLE_DEVICES` (NPU) to choose computing devices.
 
+By default, LLaMA-Factory uses all visible computing devices.
+
+Basic usage:
+
+```bash
+llamafactory-cli train examples/train_lora/llama3_lora_sft.yaml
+```
+
+Advanced usage:
+
+```bash
+CUDA_VISIBLE_DEVICES=0,1 llamafactory-cli train examples/train_lora/llama3_lora_sft.yaml \
+    learning_rate=1e-5 \
+    logging_steps=1
+```
+
+```bash
+bash examples/train_lora/llama3_lora_sft.sh
+```
+
 ## Examples
 
 ### LoRA Fine-Tuning
@@ -32,7 +52,19 @@ llamafactory-cli train examples/train_lora/llama3_lora_sft.yaml
 #### Multimodal Supervised Fine-Tuning
 
 ```bash
-llamafactory-cli train examples/train_lora/llava1_5_lora_sft.yaml
+llamafactory-cli train examples/train_lora/qwen2_5vl_lora_sft.yaml
+```
+
+#### DPO/ORPO/SimPO Training
+
+```bash
+llamafactory-cli train examples/train_lora/llama3_lora_dpo.yaml
+```
+
+#### Multimodal DPO/ORPO/SimPO Training
+
+```bash
+llamafactory-cli train examples/train_lora/qwen2_5vl_lora_dpo.yaml
 ```
 
 #### Reward Modeling
@@ -47,12 +79,6 @@ llamafactory-cli train examples/train_lora/llama3_lora_reward.yaml
 llamafactory-cli train examples/train_lora/llama3_lora_ppo.yaml
 ```
 
-#### DPO/ORPO/SimPO Training
-
-```bash
-llamafactory-cli train examples/train_lora/llama3_lora_dpo.yaml
-```
-
 #### KTO Training
 
 ```bash
@@ -73,17 +99,11 @@ llamafactory-cli train examples/train_lora/llama3_preprocess.yaml
 llamafactory-cli eval examples/train_lora/llama3_lora_eval.yaml
 ```
 
-#### Batch Predicting and Computing BLEU and ROUGE Scores
-
-```bash
-llamafactory-cli train examples/train_lora/llama3_lora_predict.yaml
-```
-
 #### Supervised Fine-Tuning on Multiple Nodes
 
 ```bash
-FORCE_TORCHRUN=1 NNODES=2 RANK=0 MASTER_ADDR=192.168.0.1 MASTER_PORT=29500 llamafactory-cli train examples/train_lora/llama3_lora_sft.yaml
-FORCE_TORCHRUN=1 NNODES=2 RANK=1 MASTER_ADDR=192.168.0.1 MASTER_PORT=29500 llamafactory-cli train examples/train_lora/llama3_lora_sft.yaml
+FORCE_TORCHRUN=1 NNODES=2 NODE_RANK=0 MASTER_ADDR=192.168.0.1 MASTER_PORT=29500 llamafactory-cli train examples/train_lora/llama3_lora_sft.yaml
+FORCE_TORCHRUN=1 NNODES=2 NODE_RANK=1 MASTER_ADDR=192.168.0.1 MASTER_PORT=29500 llamafactory-cli train examples/train_lora/llama3_lora_sft.yaml
 ```
 
 #### Supervised Fine-Tuning with DeepSpeed ZeRO-3 (Weight Sharding)
@@ -92,12 +112,24 @@ FORCE_TORCHRUN=1 NNODES=2 RANK=1 MASTER_ADDR=192.168.0.1 MASTER_PORT=29500 llama
 FORCE_TORCHRUN=1 llamafactory-cli train examples/train_lora/llama3_lora_sft_ds3.yaml
 ```
 
-### QLoRA Fine-Tuning
-
-#### Supervised Fine-Tuning with 4/8-bit Bitsandbytes Quantization (Recommended)
+#### Supervised Fine-Tuning with Ray on 4 GPUs
 
 ```bash
-llamafactory-cli train examples/train_qlora/llama3_lora_sft_bitsandbytes.yaml
+USE_RAY=1 llamafactory-cli train examples/train_lora/llama3_lora_sft_ray.yaml
+```
+
+### QLoRA Fine-Tuning
+
+#### Supervised Fine-Tuning with 4/8-bit Bitsandbytes/HQQ/EETQ Quantization (Recommended)
+
+```bash
+llamafactory-cli train examples/train_qlora/llama3_lora_sft_otfq.yaml
+```
+
+#### Supervised Fine-Tuning with 4-bit Bitsandbytes Quantization on Ascend NPU
+
+```bash
+llamafactory-cli train examples/train_qlora/llama3_lora_sft_bnb_npu.yaml
 ```
 
 #### Supervised Fine-Tuning with 4/8-bit GPTQ Quantization
@@ -123,20 +155,28 @@ llamafactory-cli train examples/train_qlora/llama3_lora_sft_aqlm.yaml
 #### Supervised Fine-Tuning on Single Node
 
 ```bash
-FORCE_TORCHRUN=1 llamafactory-cli train examples/train_full/llama3_full_sft_ds3.yaml
+FORCE_TORCHRUN=1 llamafactory-cli train examples/train_full/llama3_full_sft.yaml
 ```
 
 #### Supervised Fine-Tuning on Multiple Nodes
 
 ```bash
-FORCE_TORCHRUN=1 NNODES=2 RANK=0 MASTER_ADDR=192.168.0.1 MASTER_PORT=29500 llamafactory-cli train examples/train_full/llama3_full_sft_ds3.yaml
-FORCE_TORCHRUN=1 NNODES=2 RANK=1 MASTER_ADDR=192.168.0.1 MASTER_PORT=29500 llamafactory-cli train examples/train_full/llama3_full_sft_ds3.yaml
+FORCE_TORCHRUN=1 NNODES=2 NODE_RANK=0 MASTER_ADDR=192.168.0.1 MASTER_PORT=29500 llamafactory-cli train examples/train_full/llama3_full_sft.yaml
+FORCE_TORCHRUN=1 NNODES=2 NODE_RANK=1 MASTER_ADDR=192.168.0.1 MASTER_PORT=29500 llamafactory-cli train examples/train_full/llama3_full_sft.yaml
 ```
 
-#### Batch Predicting and Computing BLEU and ROUGE Scores
+### Elastic and Fault-Tolerant Supervised Fine-Tuning on Multiple Nodes
+
+To launch an elastic job with `MAX_RESTARTS` failures retries, run the following on at least `MIN_NNODES` nodes and at most `MAX_NNODES` nodes. `RDZV_ID` should be set as a unique job id (shared by all nodes participating in the job). See also [torchrun](https://docs.pytorch.org/docs/stable/elastic/run.html).
 
 ```bash
-llamafactory-cli train examples/train_full/llama3_full_predict.yaml
+FORCE_TORCHRUN=1 MIN_NNODES=1 MAX_NNODES=3 MAX_RESTARTS=3 RDZV_ID=llamafactory MASTER_ADDR=192.168.0.1 MASTER_PORT=29500 llamafactory-cli train examples/train_full/llama3_full_sft.yaml
+```
+
+#### Multimodal Supervised Fine-Tuning
+
+```bash
+FORCE_TORCHRUN=1 llamafactory-cli train examples/train_full/qwen2_5vl_full_sft.yaml
 ```
 
 ### Merging LoRA Adapters and Quantization
@@ -155,15 +195,28 @@ llamafactory-cli export examples/merge_lora/llama3_lora_sft.yaml
 llamafactory-cli export examples/merge_lora/llama3_gptq.yaml
 ```
 
+### Save Ollama modelfile
+
+```bash
+llamafactory-cli export examples/merge_lora/llama3_full_sft.yaml
+```
+
 ### Inferring LoRA Fine-Tuned Models
 
-#### Use CLI
+#### Evaluation using vLLM's Multi-GPU Inference
+
+```
+python scripts/vllm_infer.py --model_name_or_path meta-llama/Meta-Llama-3-8B-Instruct --template llama3 --dataset alpaca_en_demo
+python scripts/eval_bleu_rouge.py generated_predictions.jsonl
+```
+
+#### Use CLI ChatBox
 
 ```bash
 llamafactory-cli chat examples/inference/llama3_lora_sft.yaml
 ```
 
-#### Use Web UI
+#### Use Web UI ChatBox
 
 ```bash
 llamafactory-cli webchat examples/inference/llama3_lora_sft.yaml
@@ -183,12 +236,30 @@ llamafactory-cli api examples/inference/llama3_lora_sft.yaml
 llamafactory-cli train examples/extras/galore/llama3_full_sft.yaml
 ```
 
+#### Full-Parameter Fine-Tuning using APOLLO
+
+```bash
+llamafactory-cli train examples/extras/apollo/llama3_full_sft.yaml
+```
+
 #### Full-Parameter Fine-Tuning using BAdam
 
 ```bash
 llamafactory-cli train examples/extras/badam/llama3_full_sft.yaml
 ```
 
+#### Full-Parameter Fine-Tuning using Adam-mini
+
+```bash
+llamafactory-cli train examples/extras/adam_mini/qwen2_full_sft.yaml
+```
+
+#### Full-Parameter Fine-Tuning using Muon
+
+```bash
+llamafactory-cli train examples/extras/muon/qwen2_full_sft.yaml
+```
+
 #### LoRA+ Fine-Tuning
 
 ```bash

examples/README_zh.md

@@ -13,6 +13,26 @@
 
 使用 `CUDA_VISIBLE_DEVICES`（GPU）或 `ASCEND_RT_VISIBLE_DEVICES`（NPU）选择计算设备。
 
+LLaMA-Factory 默认使用所有可见的计算设备。
+
+基础用法：
+
+```bash
+llamafactory-cli train examples/train_lora/llama3_lora_sft.yaml
+```
+
+高级用法：
+
+```bash
+CUDA_VISIBLE_DEVICES=0,1 llamafactory-cli train examples/train_lora/llama3_lora_sft.yaml \
+    learning_rate=1e-5 \
+    logging_steps=1
+```
+
+```bash
+bash examples/train_lora/llama3_lora_sft.sh
+```
+
 ## 示例
 
 ### LoRA 微调
@@ -32,7 +52,19 @@ llamafactory-cli train examples/train_lora/llama3_lora_sft.yaml
 #### 多模态指令监督微调
 
 ```bash
-llamafactory-cli train examples/train_lora/llava1_5_lora_sft.yaml
+llamafactory-cli train examples/train_lora/qwen2_5vl_lora_sft.yaml
+```
+
+#### DPO/ORPO/SimPO 训练
+
+```bash
+llamafactory-cli train examples/train_lora/llama3_lora_dpo.yaml
+```
+
+#### 多模态 DPO/ORPO/SimPO 训练
+
+```bash
+llamafactory-cli train examples/train_lora/qwen2_5vl_lora_dpo.yaml
 ```
 
 #### 奖励模型训练
@@ -47,12 +79,6 @@ llamafactory-cli train examples/train_lora/llama3_lora_reward.yaml
 llamafactory-cli train examples/train_lora/llama3_lora_ppo.yaml
 ```
 
-#### DPO/ORPO/SimPO 训练
-
-```bash
-llamafactory-cli train examples/train_lora/llama3_lora_dpo.yaml
-```
-
 #### KTO 训练
 
 ```bash
@@ -73,17 +99,19 @@ llamafactory-cli train examples/train_lora/llama3_preprocess.yaml
 llamafactory-cli eval examples/train_lora/llama3_lora_eval.yaml
 ```
 
-#### 批量预测并计算 BLEU 和 ROUGE 分数
-
-```bash
-llamafactory-cli train examples/train_lora/llama3_lora_predict.yaml
-```
-
 #### 多机指令监督微调
 
 ```bash
-FORCE_TORCHRUN=1 NNODES=2 RANK=0 MASTER_ADDR=192.168.0.1 MASTER_PORT=29500 llamafactory-cli train examples/train_lora/llama3_lora_sft.yaml
-FORCE_TORCHRUN=1 NNODES=2 RANK=1 MASTER_ADDR=192.168.0.1 MASTER_PORT=29500 llamafactory-cli train examples/train_lora/llama3_lora_sft.yaml
+FORCE_TORCHRUN=1 NNODES=2 NODE_RANK=0 MASTER_ADDR=192.168.0.1 MASTER_PORT=29500 llamafactory-cli train examples/train_lora/llama3_lora_sft.yaml
+FORCE_TORCHRUN=1 NNODES=2 NODE_RANK=1 MASTER_ADDR=192.168.0.1 MASTER_PORT=29500 llamafactory-cli train examples/train_lora/llama3_lora_sft.yaml
+```
+
+### 支持弹性和容错的多机指令监督微调
+
+要启动一个支持弹性节点和容错的多机指令微调，在每个节点上执行以下命令。弹性节点数量范围为 `MIN_NNODES:MAX_NNODES`，每个节点最多允许因为错误重启 `MAX_RESTARTS` 次。`RDZV_ID` 应设置为一个唯一的作业 ID（由参与该作业的所有节点共享）。更多新可以参考官方文档 [torchrun](https://docs.pytorch.org/docs/stable/elastic/run.html)。
+
+```bash
+FORCE_TORCHRUN=1 MIN_NNODES=1 MAX_NNODES=3 MAX_RESTARTS=3 RDZV_ID=llamafactory MASTER_ADDR=192.168.0.1 MASTER_PORT=29500 llamafactory-cli train examples/train_full/llama3_full_sft.yaml
 ```
 
 #### 使用 DeepSpeed ZeRO-3 平均分配显存
@@ -92,12 +120,24 @@ FORCE_TORCHRUN=1 NNODES=2 RANK=1 MASTER_ADDR=192.168.0.1 MASTER_PORT=29500 llama
 FORCE_TORCHRUN=1 llamafactory-cli train examples/train_lora/llama3_lora_sft_ds3.yaml
 ```
 
-### QLoRA 微调
-
-#### 基于 4/8 比特 Bitsandbytes 量化进行指令监督微调（推荐）
+#### 使用 Ray 在 4 张 GPU 上微调
 
 ```bash
-llamafactory-cli train examples/train_qlora/llama3_lora_sft_bitsandbytes.yaml
+USE_RAY=1 llamafactory-cli train examples/train_lora/llama3_lora_sft_ray.yaml
+```
+
+### QLoRA 微调
+
+#### 基于 4/8 比特 Bitsandbytes/HQQ/EETQ 量化进行指令监督微调（推荐）
+
+```bash
+llamafactory-cli train examples/train_qlora/llama3_lora_sft_otfq.yaml
+```
+
+#### 在 NPU 上基于 4 比特 Bitsandbytes 量化进行指令监督微调
+
+```bash
+llamafactory-cli train examples/train_qlora/llama3_lora_sft_bnb_npu.yaml
 ```
 
 #### 基于 4/8 比特 GPTQ 量化进行指令监督微调
@@ -123,20 +163,20 @@ llamafactory-cli train examples/train_qlora/llama3_lora_sft_aqlm.yaml
 #### 在单机上进行指令监督微调
 
 ```bash
-FORCE_TORCHRUN=1 llamafactory-cli train examples/train_full/llama3_full_sft_ds3.yaml
+FORCE_TORCHRUN=1 llamafactory-cli train examples/train_full/llama3_full_sft.yaml
 ```
 
 #### 在多机上进行指令监督微调
 
 ```bash
-FORCE_TORCHRUN=1 NNODES=2 RANK=0 MASTER_ADDR=192.168.0.1 MASTER_PORT=29500 llamafactory-cli train examples/train_full/llama3_full_sft_ds3.yaml
-FORCE_TORCHRUN=1 NNODES=2 RANK=1 MASTER_ADDR=192.168.0.1 MASTER_PORT=29500 llamafactory-cli train examples/train_full/llama3_full_sft_ds3.yaml
+FORCE_TORCHRUN=1 NNODES=2 NODE_RANK=0 MASTER_ADDR=192.168.0.1 MASTER_PORT=29500 llamafactory-cli train examples/train_full/llama3_full_sft.yaml
+FORCE_TORCHRUN=1 NNODES=2 NODE_RANK=1 MASTER_ADDR=192.168.0.1 MASTER_PORT=29500 llamafactory-cli train examples/train_full/llama3_full_sft.yaml
 ```
 
-#### 批量预测并计算 BLEU 和 ROUGE 分数
+#### 多模态指令监督微调
 
 ```bash
-llamafactory-cli train examples/train_full/llama3_full_predict.yaml
+FORCE_TORCHRUN=1 llamafactory-cli train examples/train_full/qwen2_5vl_full_sft.yaml
 ```
 
 ### 合并 LoRA 适配器与模型量化
@@ -155,15 +195,28 @@ llamafactory-cli export examples/merge_lora/llama3_lora_sft.yaml
 llamafactory-cli export examples/merge_lora/llama3_gptq.yaml
 ```
 
+### 保存 Ollama 配置文件
+
+```bash
+llamafactory-cli export examples/merge_lora/llama3_full_sft.yaml
+```
+
 ### 推理 LoRA 模型
 
-#### 使用命令行接口
+#### 使用 vLLM 多卡推理评估
+
+```
+python scripts/vllm_infer.py --model_name_or_path meta-llama/Meta-Llama-3-8B-Instruct --template llama3 --dataset alpaca_en_demo
+python scripts/eval_bleu_rouge.py generated_predictions.jsonl
+```
+
+#### 使用命令行对话框
 
 ```bash
 llamafactory-cli chat examples/inference/llama3_lora_sft.yaml
 ```
 
-#### 使用浏览器界面
+#### 使用浏览器对话框
 
 ```bash
 llamafactory-cli webchat examples/inference/llama3_lora_sft.yaml
@@ -183,12 +236,30 @@ llamafactory-cli api examples/inference/llama3_lora_sft.yaml
 llamafactory-cli train examples/extras/galore/llama3_full_sft.yaml
 ```
 
+#### 使用 APOLLO 进行全参数训练
+
+```bash
+llamafactory-cli train examples/extras/apollo/llama3_full_sft.yaml
+```
+
 #### 使用 BAdam 进行全参数训练
 
 ```bash
 llamafactory-cli train examples/extras/badam/llama3_full_sft.yaml
 ```
 
+#### 使用 Adam-mini 进行全参数训练
+
+```bash
+llamafactory-cli train examples/extras/adam_mini/qwen2_full_sft.yaml
+```
+
+#### 使用 Muon 进行全参数训练
+
+```bash
+llamafactory-cli train examples/extras/muon/qwen2_full_sft.yaml
+```
+
 #### LoRA+ 微调
 
 ```bash

examples/accelerate/fsdp_config.yaml

@@ -7,14 +7,14 @@ fsdp_config:
   fsdp_backward_prefetch: BACKWARD_PRE
   fsdp_forward_prefetch: false
   fsdp_cpu_ram_efficient_loading: true
-  fsdp_offload_params: true # offload may affect training speed
+  fsdp_offload_params: false
   fsdp_sharding_strategy: FULL_SHARD
   fsdp_state_dict_type: FULL_STATE_DICT
   fsdp_sync_module_states: true
   fsdp_use_orig_params: true
 machine_rank: 0
 main_training_function: main
-mixed_precision: fp16 # or bf16
+mixed_precision: bf16  # or fp16
 num_machines: 1  # the number of nodes
 num_processes: 2  # the number of GPUs in all nodes
 rdzv_backend: static

examples/accelerate/fsdp_config_offload.yaml

@@ -0,0 +1,25 @@
+compute_environment: LOCAL_MACHINE
+debug: false
+distributed_type: FSDP
+downcast_bf16: 'no'
+fsdp_config:
+  fsdp_auto_wrap_policy: TRANSFORMER_BASED_WRAP
+  fsdp_backward_prefetch: BACKWARD_PRE
+  fsdp_forward_prefetch: false
+  fsdp_cpu_ram_efficient_loading: true
+  fsdp_offload_params: true  # offload may affect training speed
+  fsdp_sharding_strategy: FULL_SHARD
+  fsdp_state_dict_type: FULL_STATE_DICT
+  fsdp_sync_module_states: true
+  fsdp_use_orig_params: true
+machine_rank: 0
+main_training_function: main
+mixed_precision: bf16  # or fp16
+num_machines: 1  # the number of nodes
+num_processes: 2  # the number of GPUs in all nodes
+rdzv_backend: static
+same_network: true
+tpu_env: []
+tpu_use_cluster: false
+tpu_use_sudo: false
+use_cpu: false

examples/extras/adam_mini/qwen2_full_sft.yaml

@@ -0,0 +1,43 @@
+### model
+model_name_or_path: Qwen/Qwen2-1.5B-Instruct
+trust_remote_code: true
+
+### method
+stage: sft
+do_train: true
+finetuning_type: full
+use_adam_mini: true
+
+### dataset
+dataset: identity,alpaca_en_demo
+template: qwen
+cutoff_len: 2048
+max_samples: 1000
+overwrite_cache: true
+preprocessing_num_workers: 16
+dataloader_num_workers: 4
+
+### output
+output_dir: saves/qwen2-1_5b/full/sft
+logging_steps: 10
+save_steps: 500
+plot_loss: true
+overwrite_output_dir: true
+save_only_model: false
+report_to: none  # choices: [none, wandb, tensorboard, swanlab, mlflow]
+
+### train
+per_device_train_batch_size: 1
+gradient_accumulation_steps: 8
+learning_rate: 1.0e-5
+num_train_epochs: 3.0
+lr_scheduler_type: cosine
+warmup_ratio: 0.1
+bf16: true
+ddp_timeout: 180000000
+
+### eval
+# val_size: 0.1
+# per_device_eval_batch_size: 1
+# eval_strategy: steps
+# eval_steps: 500

examples/extras/apollo/llama3_full_sft.yaml

@@ -0,0 +1,48 @@
+### model
+model_name_or_path: meta-llama/Meta-Llama-3-8B-Instruct
+trust_remote_code: true
+
+### method
+stage: sft
+do_train: true
+finetuning_type: full
+use_apollo: true
+apollo_layerwise: true  # choices: [true, false], use false for DDP training
+apollo_target: all
+apollo_rank: 128
+apollo_scale: 32.0
+apollo_scale_type: channel
+
+### dataset
+dataset: identity,alpaca_en_demo
+template: llama3
+cutoff_len: 2048
+max_samples: 1000
+overwrite_cache: true
+preprocessing_num_workers: 16
+dataloader_num_workers: 4
+
+### output
+output_dir: saves/llama3-8b/full/sft
+logging_steps: 10
+save_steps: 500
+plot_loss: true
+overwrite_output_dir: true
+save_only_model: false
+report_to: none  # choices: [none, wandb, tensorboard, swanlab, mlflow]
+
+### train
+per_device_train_batch_size: 1
+gradient_accumulation_steps: 1  # use 1 for layerwise apollo
+learning_rate: 1.0e-5
+num_train_epochs: 3.0
+lr_scheduler_type: cosine
+warmup_ratio: 0.1
+pure_bf16: true
+ddp_timeout: 180000000
+
+### eval
+# val_size: 0.1
+# per_device_eval_batch_size: 1
+# eval_strategy: steps
+# eval_steps: 500

examples/extras/badam/llama3_full_sft.yaml

@@ -1,22 +1,26 @@
 ### model
 model_name_or_path: meta-llama/Meta-Llama-3-8B-Instruct
+trust_remote_code: true
 
 ### method
 stage: sft
 do_train: true
 finetuning_type: full
 use_badam: true
+badam_mode: layer
 badam_switch_mode: ascending
 badam_switch_interval: 50
 badam_verbose: 2
+# deepspeed: examples/deepspeed/ds_z3_config.json
 
 ### dataset
 dataset: identity,alpaca_en_demo
 template: llama3
-cutoff_len: 1024
+cutoff_len: 2048
 max_samples: 1000
 overwrite_cache: true
 preprocessing_num_workers: 16
+dataloader_num_workers: 4
 
 ### output
 output_dir: saves/llama3-8b/full/sft
@@ -24,18 +28,19 @@ logging_steps: 10
 save_steps: 500
 plot_loss: true
 overwrite_output_dir: true
+save_only_model: false
+report_to: none  # choices: [none, wandb, tensorboard, swanlab, mlflow]
 
 ### train
 per_device_train_batch_size: 1
 gradient_accumulation_steps: 8
-learning_rate: 1.0e-4
+learning_rate: 1.0e-5
 num_train_epochs: 3.0
 lr_scheduler_type: cosine
 warmup_ratio: 0.1
-pure_bf16: true
 
 ### eval
-val_size: 0.1
-per_device_eval_batch_size: 1
-eval_strategy: steps
-eval_steps: 500
+# val_size: 0.1
+# per_device_eval_batch_size: 1
+# eval_strategy: steps
+# eval_steps: 500

examples/extras/fsdp_qlora/llama3_lora_sft.yaml

@@ -1,20 +1,23 @@
 ### model
 model_name_or_path: meta-llama/Meta-Llama-3-8B-Instruct
 quantization_bit: 4
+trust_remote_code: true
 
 ### method
 stage: sft
 do_train: true
 finetuning_type: lora
+lora_rank: 8
 lora_target: all
 
 ### dataset
 dataset: identity,alpaca_en_demo
 template: llama3
-cutoff_len: 1024
+cutoff_len: 2048
 max_samples: 1000
 overwrite_cache: true
 preprocessing_num_workers: 16
+dataloader_num_workers: 4
 
 ### output
 output_dir: saves/llama3-8b/lora/sft
@@ -22,6 +25,8 @@ logging_steps: 10
 save_steps: 500
 plot_loss: true
 overwrite_output_dir: true
+save_only_model: false
+report_to: none  # choices: [none, wandb, tensorboard, swanlab, mlflow]
 
 ### train
 per_device_train_batch_size: 1
@@ -30,11 +35,11 @@ learning_rate: 1.0e-4
 num_train_epochs: 3.0
 lr_scheduler_type: cosine
 warmup_ratio: 0.1
-fp16: true
+bf16: true
 ddp_timeout: 180000000
 
 ### eval
-val_size: 0.1
-per_device_eval_batch_size: 1
-eval_strategy: steps
-eval_steps: 500
+# val_size: 0.1
+# per_device_eval_batch_size: 1
+# eval_strategy: steps
+# eval_steps: 500

examples/extras/galore/llama3_full_sft.yaml

@@ -1,23 +1,25 @@
 ### model
 model_name_or_path: meta-llama/Meta-Llama-3-8B-Instruct
+trust_remote_code: true
 
 ### method
 stage: sft
 do_train: true
 finetuning_type: full
 use_galore: true
-galore_layerwise: true
-galore_target: mlp,self_attn
+galore_layerwise: true  # choices: [true, false], use false for DDP training
+galore_target: all
 galore_rank: 128
 galore_scale: 2.0
 
 ### dataset
 dataset: identity,alpaca_en_demo
 template: llama3
-cutoff_len: 1024
+cutoff_len: 2048
 max_samples: 1000
 overwrite_cache: true
 preprocessing_num_workers: 16
+dataloader_num_workers: 4
 
 ### output
 output_dir: saves/llama3-8b/full/sft
@@ -25,18 +27,21 @@ logging_steps: 10
 save_steps: 500
 plot_loss: true
 overwrite_output_dir: true
+save_only_model: false
+report_to: none  # choices: [none, wandb, tensorboard, swanlab, mlflow]
 
 ### train
 per_device_train_batch_size: 1
-gradient_accumulation_steps: 1
-learning_rate: 1.0e-4
+gradient_accumulation_steps: 1  # use 1 for layerwise galore
+learning_rate: 1.0e-5
 num_train_epochs: 3.0
 lr_scheduler_type: cosine
 warmup_ratio: 0.1
 pure_bf16: true
+ddp_timeout: 180000000
 
 ### eval
-val_size: 0.1
-per_device_eval_batch_size: 1
-eval_strategy: steps
-eval_steps: 500
+# val_size: 0.1
+# per_device_eval_batch_size: 1
+# eval_strategy: steps
+# eval_steps: 500

examples/extras/llama_pro/expand.sh

@@ -2,5 +2,5 @@
 
 python scripts/llama_pro.py \
     --model_name_or_path meta-llama/Meta-Llama-3-8B-Instruct \
-    --output_dir models/llama3-8b-instruct-pro \
+    --output_dir models/llama3-8b-pro \
     --num_expand 8

examples/extras/llama_pro/llama3_freeze_sft.yaml

@@ -1,5 +1,6 @@
 ### model
-model_name_or_path: models/llama3-8b-instruct-pro
+model_name_or_path: models/llama3-8b-pro
+trust_remote_code: true
 
 ### method
 stage: sft
@@ -12,17 +13,20 @@ use_llama_pro: true
 ### dataset
 dataset: identity,alpaca_en_demo
 template: llama3
-cutoff_len: 1024
+cutoff_len: 2048
 max_samples: 1000
 overwrite_cache: true
 preprocessing_num_workers: 16
+dataloader_num_workers: 4
 
 ### output
-output_dir: saves/llama3-8b-instruct-pro/freeze/sft
+output_dir: saves/llama3-8b-pro/freeze/sft
 logging_steps: 10
 save_steps: 500
 plot_loss: true
 overwrite_output_dir: true
+save_only_model: false
+report_to: none  # choices: [none, wandb, tensorboard, swanlab, mlflow]
 
 ### train
 per_device_train_batch_size: 1
@@ -31,11 +35,11 @@ learning_rate: 1.0e-4
 num_train_epochs: 3.0
 lr_scheduler_type: cosine
 warmup_ratio: 0.1
-fp16: true
+bf16: true
 ddp_timeout: 180000000
 
 ### eval
-val_size: 0.1
-per_device_eval_batch_size: 1
-eval_strategy: steps
-eval_steps: 500
+# val_size: 0.1
+# per_device_eval_batch_size: 1
+# eval_strategy: steps
+# eval_steps: 500

examples/extras/loraplus/llama3_lora_sft.yaml

@@ -1,20 +1,23 @@
 ### model
 model_name_or_path: meta-llama/Meta-Llama-3-8B-Instruct
+trust_remote_code: true
 
 ### method
 stage: sft
 do_train: true
 finetuning_type: lora
+lora_rank: 8
 lora_target: all
 loraplus_lr_ratio: 16.0
 
 ### dataset
 dataset: identity,alpaca_en_demo
 template: llama3
-cutoff_len: 1024
+cutoff_len: 2048
 max_samples: 1000
 overwrite_cache: true
 preprocessing_num_workers: 16
+dataloader_num_workers: 4
 
 ### output
 output_dir: saves/llama3-8b/lora/sft
@@ -22,6 +25,8 @@ logging_steps: 10
 save_steps: 500
 plot_loss: true
 overwrite_output_dir: true
+save_only_model: false
+report_to: none  # choices: [none, wandb, tensorboard, swanlab, mlflow]
 
 ### train
 per_device_train_batch_size: 1
@@ -30,11 +35,11 @@ learning_rate: 1.0e-4
 num_train_epochs: 3.0
 lr_scheduler_type: cosine
 warmup_ratio: 0.1
-fp16: true
+bf16: true
 ddp_timeout: 180000000
 
 ### eval
-val_size: 0.1
-per_device_eval_batch_size: 1
-eval_strategy: steps
-eval_steps: 500
+# val_size: 0.1
+# per_device_eval_batch_size: 1
+# eval_strategy: steps
+# eval_steps: 500

examples/extras/mod/llama3_full_sft.yaml

@@ -1,5 +1,6 @@
 ### model
 model_name_or_path: meta-llama/Meta-Llama-3-8B-Instruct
+trust_remote_code: true
 
 ### method
 stage: sft
@@ -10,10 +11,11 @@ mixture_of_depths: convert
 ### dataset
 dataset: identity,alpaca_en_demo
 template: llama3
-cutoff_len: 1024
+cutoff_len: 2048
 max_samples: 1000
 overwrite_cache: true
 preprocessing_num_workers: 16
+dataloader_num_workers: 4
 
 ### output
 output_dir: saves/llama3-8b-mod/full/sft
@@ -21,12 +23,14 @@ logging_steps: 10
 save_steps: 500
 plot_loss: true
 overwrite_output_dir: true
+save_only_model: false
+report_to: none  # choices: [none, wandb, tensorboard, swanlab, mlflow]
 
 ### train
 per_device_train_batch_size: 1
 gradient_accumulation_steps: 8
 optim: paged_adamw_8bit
-learning_rate: 1.0e-4
+learning_rate: 1.0e-5
 num_train_epochs: 3.0
 lr_scheduler_type: cosine
 warmup_ratio: 0.1
@@ -34,7 +38,7 @@ pure_bf16: true
 ddp_timeout: 180000000
 
 ### eval
-val_size: 0.1
-per_device_eval_batch_size: 1
-eval_strategy: steps
-eval_steps: 500
+# val_size: 0.1
+# per_device_eval_batch_size: 1
+# eval_strategy: steps
+# eval_steps: 500

examples/extras/muon/qwen2_full_sft.yaml

@@ -0,0 +1,43 @@
+### model
+model_name_or_path: Qwen/Qwen2-1.5B-Instruct
+trust_remote_code: true
+
+### method
+stage: sft
+do_train: true
+finetuning_type: full
+use_muon: true
+
+### dataset
+dataset: identity,alpaca_en_demo
+template: qwen
+cutoff_len: 2048
+max_samples: 1000
+overwrite_cache: true
+preprocessing_num_workers: 16
+dataloader_num_workers: 4
+
+### output
+output_dir: saves/qwen2-1_5b/full/sft
+logging_steps: 10
+save_steps: 500
+plot_loss: true
+overwrite_output_dir: true
+save_only_model: false
+report_to: none  # choices: [none, wandb, tensorboard, swanlab, mlflow]
+
+### train
+per_device_train_batch_size: 1
+gradient_accumulation_steps: 8
+learning_rate: 1.0e-5
+num_train_epochs: 3.0
+lr_scheduler_type: cosine
+warmup_ratio: 0.1
+bf16: true
+ddp_timeout: 180000000
+
+### eval
+# val_size: 0.1
+# per_device_eval_batch_size: 1
+# eval_strategy: steps
+# eval_steps: 500

examples/extras/nlg_eval/llama3_lora_predict.yaml

@@ -1,6 +1,10 @@
+# The batch generation can be SLOW using this config.
+# For faster inference, we recommend to use `scripts/vllm_infer.py`.
+
 ### model
 model_name_or_path: meta-llama/Meta-Llama-3-8B-Instruct
 adapter_name_or_path: saves/llama3-8b/lora/sft
+trust_remote_code: true
 
 ### method
 stage: sft
@@ -8,16 +12,18 @@ do_predict: true
 finetuning_type: lora
 
 ### dataset
-dataset: identity,alpaca_en_demo
+eval_dataset: identity,alpaca_en_demo
 template: llama3
-cutoff_len: 1024
+cutoff_len: 2048
 max_samples: 50
 overwrite_cache: true
 preprocessing_num_workers: 16
+dataloader_num_workers: 4
 
 ### output
 output_dir: saves/llama3-8b/lora/predict
 overwrite_output_dir: true
+report_to: none  # choices: [none, wandb, tensorboard, swanlab, mlflow]
 
 ### eval
 per_device_eval_batch_size: 1

examples/extras/pissa/init.sh

@@ -0,0 +1,5 @@
+#!/bin/bash
+
+python scripts/pissa_init.py \
+    --model_name_or_path meta-llama/Meta-Llama-3-8B-Instruct \
+    --output_dir models/llama3-8b-pissa

examples/extras/pissa/llama3_lora_sft.yaml

@@ -1,22 +1,25 @@
 ### model
 model_name_or_path: meta-llama/Meta-Llama-3-8B-Instruct
+trust_remote_code: true
 
 ### method
 stage: sft
 do_train: true
 finetuning_type: lora
+lora_rank: 8
 lora_target: all
 pissa_init: true
-pissa_iter: 4
+pissa_iter: 16
 pissa_convert: true
 
 ### dataset
 dataset: identity,alpaca_en_demo
 template: llama3
-cutoff_len: 1024
+cutoff_len: 2048
 max_samples: 1000
 overwrite_cache: true
 preprocessing_num_workers: 16
+dataloader_num_workers: 4
 
 ### output
 output_dir: saves/llama3-8b/lora/sft
@@ -24,6 +27,8 @@ logging_steps: 10
 save_steps: 500
 plot_loss: true
 overwrite_output_dir: true
+save_only_model: false
+report_to: none  # choices: [none, wandb, tensorboard, swanlab, mlflow]
 
 ### train
 per_device_train_batch_size: 1
@@ -32,11 +37,11 @@ learning_rate: 1.0e-4
 num_train_epochs: 3.0
 lr_scheduler_type: cosine
 warmup_ratio: 0.1
-fp16: true
+bf16: true
 ddp_timeout: 180000000
 
 ### eval
-val_size: 0.1
-per_device_eval_batch_size: 1
-eval_strategy: steps
-eval_steps: 500
+# val_size: 0.1
+# per_device_eval_batch_size: 1
+# eval_strategy: steps
+# eval_steps: 500

examples/inference/llama3.yaml

@@ -1,2 +1,4 @@
 model_name_or_path: meta-llama/Meta-Llama-3-8B-Instruct
 template: llama3
+infer_backend: huggingface  # choices: [huggingface, vllm, sglang]
+trust_remote_code: true

examples/inference/llama3_full_sft.yaml

@@ -0,0 +1,4 @@
+model_name_or_path: saves/llama3-8b/full/sft
+template: llama3
+infer_backend: huggingface  # choices: [huggingface, vllm, sglang]
+trust_remote_code: true

examples/inference/llama3_lora_sft.yaml

@@ -1,4 +1,5 @@
 model_name_or_path: meta-llama/Meta-Llama-3-8B-Instruct
 adapter_name_or_path: saves/llama3-8b/lora/sft
 template: llama3
-finetuning_type: lora
+infer_backend: huggingface  # choices: [huggingface, vllm, sglang]
+trust_remote_code: true

examples/inference/llama3_vllm.yaml

@@ -1,4 +0,0 @@
-model_name_or_path: meta-llama/Meta-Llama-3-8B-Instruct
-template: llama3
-infer_backend: vllm
-vllm_enforce_eager: true

examples/inference/qwen2_5vl.yaml

@@ -0,0 +1,4 @@
+model_name_or_path: Qwen/Qwen2.5-VL-7B-Instruct
+template: qwen2_vl
+infer_backend: huggingface  # choices: [huggingface, vllm, sglang]
+trust_remote_code: true

examples/merge_lora/llama3_full_sft.yaml

@@ -0,0 +1,10 @@
+### model
+model_name_or_path: saves/llama3-8b/full/sft
+template: llama3
+trust_remote_code: true
+
+### export
+export_dir: output/llama3_full_sft
+export_size: 5
+export_device: cpu  # choices: [cpu, auto]
+export_legacy_format: false

examples/merge_lora/llama3_gptq.yaml

@@ -1,11 +1,12 @@
 ### model
 model_name_or_path: meta-llama/Meta-Llama-3-8B-Instruct
 template: llama3
+trust_remote_code: true
 
 ### export
-export_dir: models/llama3_gptq
+export_dir: output/llama3_gptq
 export_quantization_bit: 4
-export_quantization_dataset: data/c4_demo.json
-export_size: 2
-export_device: cpu
+export_quantization_dataset: data/c4_demo.jsonl
+export_size: 5
+export_device: cpu  # choices: [cpu, auto]
 export_legacy_format: false

examples/merge_lora/llama3_lora_sft.yaml

@@ -4,10 +4,10 @@
 model_name_or_path: meta-llama/Meta-Llama-3-8B-Instruct
 adapter_name_or_path: saves/llama3-8b/lora/sft
 template: llama3
-finetuning_type: lora
+trust_remote_code: true
 
 ### export
-export_dir: models/llama3_lora_sft
-export_size: 2
-export_device: cpu
+export_dir: output/llama3_lora_sft
+export_size: 5
+export_device: cpu  # choices: [cpu, auto]
 export_legacy_format: false

examples/merge_lora/qwen2_5vl_lora_sft.yaml

@@ -0,0 +1,13 @@
+### Note: DO NOT use quantized model or quantization_bit when merging lora adapters
+
+### model
+model_name_or_path: Qwen/Qwen2.5-VL-7B-Instruct
+adapter_name_or_path: saves/qwen2_5vl-7b/lora/sft
+template: qwen2_vl
+trust_remote_code: true
+
+### export
+export_dir: output/qwen2_5vl_lora_sft
+export_size: 5
+export_device: cpu  # choices: [cpu, auto]
+export_legacy_format: false

examples/train_full/llama3_full_predict.yaml

@@ -1,23 +0,0 @@
-### model
-model_name_or_path: saves/llama3-8b/full/sft
-
-### method
-stage: sft
-do_predict: true
-finetuning_type: full
-
-### dataset
-dataset: identity,alpaca_en_demo
-template: llama3
-cutoff_len: 1024
-max_samples: 50
-overwrite_cache: true
-preprocessing_num_workers: 16
-
-### output
-output_dir: saves/llama3-8b/full/predict
-overwrite_output_dir: true
-
-### eval
-per_device_eval_batch_size: 1
-predict_with_generate: true

examples/train_full/llama3_full_sft.yaml

@@ -1,19 +1,21 @@
 ### model
 model_name_or_path: meta-llama/Meta-Llama-3-8B-Instruct
+trust_remote_code: true
 
 ### method
 stage: sft
 do_train: true
 finetuning_type: full
-deepspeed: examples/deepspeed/ds_z3_config.json
+deepspeed: examples/deepspeed/ds_z3_config.json  # choices: [ds_z0_config.json, ds_z2_config.json, ds_z3_config.json]
 
 ### dataset
 dataset: identity,alpaca_en_demo
 template: llama3
-cutoff_len: 1024
+cutoff_len: 2048
 max_samples: 1000
 overwrite_cache: true
 preprocessing_num_workers: 16
+dataloader_num_workers: 4
 
 ### output
 output_dir: saves/llama3-8b/full/sft
@@ -21,19 +23,23 @@ logging_steps: 10
 save_steps: 500
 plot_loss: true
 overwrite_output_dir: true
+save_only_model: false
+report_to: none  # choices: [none, wandb, tensorboard, swanlab, mlflow]
 
 ### train
 per_device_train_batch_size: 1
 gradient_accumulation_steps: 2
-learning_rate: 1.0e-4
+learning_rate: 1.0e-5
 num_train_epochs: 3.0
 lr_scheduler_type: cosine
 warmup_ratio: 0.1
-fp16: true
+bf16: true
 ddp_timeout: 180000000
+resume_from_checkpoint: null
 
 ### eval
-val_size: 0.1
-per_device_eval_batch_size: 1
-eval_strategy: steps
-eval_steps: 500
+# eval_dataset: alpaca_en_demo
+# val_size: 0.1
+# per_device_eval_batch_size: 1
+# eval_strategy: steps
+# eval_steps: 500

examples/train_full/qwen2_5vl_full_sft.yaml

@@ -0,0 +1,49 @@
+### model
+model_name_or_path: Qwen/Qwen2.5-VL-7B-Instruct
+image_max_pixels: 262144
+video_max_pixels: 16384
+trust_remote_code: true
+
+### method
+stage: sft
+do_train: true
+finetuning_type: full
+freeze_vision_tower: true
+freeze_multi_modal_projector: true
+freeze_language_model: false
+deepspeed: examples/deepspeed/ds_z3_config.json
+
+### dataset
+dataset: mllm_demo,identity,alpaca_en_demo
+template: qwen2_vl
+cutoff_len: 2048
+max_samples: 1000
+overwrite_cache: true
+preprocessing_num_workers: 16
+dataloader_num_workers: 4
+
+### output
+output_dir: saves/qwen2_5vl-7b/full/sft
+logging_steps: 10
+save_steps: 500
+plot_loss: true
+overwrite_output_dir: true
+save_only_model: false
+report_to: none  # choices: [none, wandb, tensorboard, swanlab, mlflow]
+
+### train
+per_device_train_batch_size: 1
+gradient_accumulation_steps: 2
+learning_rate: 1.0e-5
+num_train_epochs: 3.0
+lr_scheduler_type: cosine
+warmup_ratio: 0.1
+bf16: true
+ddp_timeout: 180000000
+resume_from_checkpoint: null
+
+### eval
+# val_size: 0.1
+# per_device_eval_batch_size: 1
+# eval_strategy: steps
+# eval_steps: 500

examples/train_lora/llama3_lora_dpo.yaml

@@ -1,21 +1,24 @@
 ### model
 model_name_or_path: meta-llama/Meta-Llama-3-8B-Instruct
+trust_remote_code: true
 
 ### method
 stage: dpo
 do_train: true
 finetuning_type: lora
+lora_rank: 8
 lora_target: all
 pref_beta: 0.1
-pref_loss: sigmoid  # [sigmoid (dpo), orpo, simpo]
+pref_loss: sigmoid  # choices: [sigmoid (dpo), orpo, simpo]
 
 ### dataset
 dataset: dpo_en_demo
 template: llama3
-cutoff_len: 1024
+cutoff_len: 2048
 max_samples: 1000
 overwrite_cache: true
 preprocessing_num_workers: 16
+dataloader_num_workers: 4
 
 ### output
 output_dir: saves/llama3-8b/lora/dpo
@@ -23,6 +26,8 @@ logging_steps: 10
 save_steps: 500
 plot_loss: true
 overwrite_output_dir: true
+save_only_model: false
+report_to: none  # choices: [none, wandb, tensorboard, swanlab, mlflow]
 
 ### train
 per_device_train_batch_size: 1
@@ -31,11 +36,13 @@ learning_rate: 5.0e-6
 num_train_epochs: 3.0
 lr_scheduler_type: cosine
 warmup_ratio: 0.1
-fp16: true
+bf16: true
 ddp_timeout: 180000000
+resume_from_checkpoint: null
 
 ### eval
-val_size: 0.1
-per_device_eval_batch_size: 1
-eval_strategy: steps
-eval_steps: 500
+# eval_dataset: dpo_en_demo
+# val_size: 0.1
+# per_device_eval_batch_size: 1
+# eval_strategy: steps
+# eval_steps: 500

examples/train_lora/llama3_lora_eval.yaml

@@ -1,13 +1,13 @@
 ### model
 model_name_or_path: meta-llama/Meta-Llama-3-8B-Instruct
 adapter_name_or_path: saves/llama3-8b/lora/sft
+trust_remote_code: true
 
 ### method
 finetuning_type: lora
 
 ### dataset
-task: mmlu
-split: test
+task: mmlu_test  # choices: [mmlu_test, ceval_validation, cmmlu_test]
 template: fewshot
 lang: en
 n_shot: 5

examples/train_lora/llama3_lora_kto.yaml

@@ -1,20 +1,23 @@
 ### model
 model_name_or_path: meta-llama/Meta-Llama-3-8B-Instruct
+trust_remote_code: true
 
 ### method
 stage: kto
 do_train: true
 finetuning_type: lora
+lora_rank: 8
 lora_target: all
 pref_beta: 0.1
 
 ### dataset
 dataset: kto_en_demo
 template: llama3
-cutoff_len: 1024
+cutoff_len: 2048
 max_samples: 1000
 overwrite_cache: true
 preprocessing_num_workers: 16
+dataloader_num_workers: 4
 
 ### output
 output_dir: saves/llama3-8b/lora/kto
@@ -22,6 +25,7 @@ logging_steps: 10
 save_steps: 500
 plot_loss: true
 overwrite_output_dir: true
+report_to: none  # choices: [none, wandb, tensorboard, swanlab, mlflow]
 
 ### train
 per_device_train_batch_size: 1
@@ -30,11 +34,11 @@ learning_rate: 5.0e-6
 num_train_epochs: 3.0
 lr_scheduler_type: cosine
 warmup_ratio: 0.1
-fp16: true
+bf16: true
 ddp_timeout: 180000000
 
 ### eval
-val_size: 0.1
-per_device_eval_batch_size: 1
-eval_strategy: steps
-eval_steps: 500
+# val_size: 0.1
+# per_device_eval_batch_size: 1
+# eval_strategy: steps
+# eval_steps: 500

examples/train_lora/llama3_lora_ppo.yaml

@@ -1,20 +1,23 @@
 ### model
 model_name_or_path: meta-llama/Meta-Llama-3-8B-Instruct
 reward_model: saves/llama3-8b/lora/reward
+trust_remote_code: true
 
 ### method
 stage: ppo
 do_train: true
 finetuning_type: lora
+lora_rank: 8
 lora_target: all
 
 ### dataset
 dataset: identity,alpaca_en_demo
 template: llama3
-cutoff_len: 1024
+cutoff_len: 2048
 max_samples: 1000
 overwrite_cache: true
 preprocessing_num_workers: 16
+dataloader_num_workers: 4
 
 ### output
 output_dir: saves/llama3-8b/lora/ppo
@@ -22,6 +25,7 @@ logging_steps: 10
 save_steps: 500
 plot_loss: true
 overwrite_output_dir: true
+report_to: none  # choices: [none, wandb, tensorboard, swanlab, mlflow]
 
 ### train
 per_device_train_batch_size: 1
@@ -30,7 +34,7 @@ learning_rate: 1.0e-5
 num_train_epochs: 3.0
 lr_scheduler_type: cosine
 warmup_ratio: 0.1
-fp16: true
+bf16: true
 ddp_timeout: 180000000
 
 ### generate

examples/train_lora/llama3_lora_pretrain.yaml

@@ -1,25 +1,30 @@
 ### model
 model_name_or_path: meta-llama/Meta-Llama-3-8B-Instruct
+trust_remote_code: true
 
 ### method
 stage: pt
 do_train: true
 finetuning_type: lora
+lora_rank: 8
 lora_target: all
 
 ### dataset
 dataset: c4_demo
-cutoff_len: 1024
+cutoff_len: 2048
 max_samples: 1000
 overwrite_cache: true
 preprocessing_num_workers: 16
+dataloader_num_workers: 4
 
 ### output
-output_dir: saves/llama3-8b/lora/sft
+output_dir: saves/llama3-8b/lora/pretrain
 logging_steps: 10
 save_steps: 500
 plot_loss: true
 overwrite_output_dir: true
+save_only_model: false
+report_to: none  # choices: [none, wandb, tensorboard, swanlab, mlflow]
 
 ### train
 per_device_train_batch_size: 1
@@ -28,11 +33,13 @@ learning_rate: 1.0e-4
 num_train_epochs: 3.0
 lr_scheduler_type: cosine
 warmup_ratio: 0.1
-fp16: true
+bf16: true
 ddp_timeout: 180000000
+resume_from_checkpoint: null
 
 ### eval
-val_size: 0.1
-per_device_eval_batch_size: 1
-eval_strategy: steps
-eval_steps: 500
+# eval_dataset: c4_demo
+# val_size: 0.1
+# per_device_eval_batch_size: 1
+# eval_strategy: steps
+# eval_steps: 500

examples/train_lora/llama3_lora_reward.yaml

@@ -1,19 +1,22 @@
 ### model
 model_name_or_path: meta-llama/Meta-Llama-3-8B-Instruct
+trust_remote_code: true
 
 ### method
 stage: rm
 do_train: true
 finetuning_type: lora
+lora_rank: 8
 lora_target: all
 
 ### dataset
 dataset: dpo_en_demo
 template: llama3
-cutoff_len: 1024
+cutoff_len: 2048
 max_samples: 1000
 overwrite_cache: true
 preprocessing_num_workers: 16
+dataloader_num_workers: 4
 
 ### output
 output_dir: saves/llama3-8b/lora/reward
@@ -21,19 +24,23 @@ logging_steps: 10
 save_steps: 500
 plot_loss: true
 overwrite_output_dir: true
+save_only_model: false
+report_to: none  # choices: [none, wandb, tensorboard, swanlab, mlflow]
 
 ### train
 per_device_train_batch_size: 1
 gradient_accumulation_steps: 8
-learning_rate: 1.0e-5
+learning_rate: 1.0e-4
 num_train_epochs: 3.0
 lr_scheduler_type: cosine
 warmup_ratio: 0.1
-fp16: true
+bf16: true
 ddp_timeout: 180000000
+resume_from_checkpoint: null
 
 ### eval
-val_size: 0.1
-per_device_eval_batch_size: 1
-eval_strategy: steps
-eval_steps: 500
+# eval_dataset: dpo_en_demo
+# val_size: 0.1
+# per_device_eval_batch_size: 1
+# eval_strategy: steps
+# eval_steps: 500

examples/train_lora/llama3_lora_sft.sh

@@ -0,0 +1,36 @@
+#!/bin/bash
+
+set -x
+
+MODEL_PATH=meta-llama/Meta-Llama-3-8B-Instruct
+
+llamafactory-cli train \
+    --model_name_or_path ${MODEL_PATH} \
+    --trust_remote_code \
+    --stage sft \
+    --do_train \
+    --finetuning_type lora \
+    --lora_rank 8 \
+    --lora_target all \
+    --dataset identity,alpaca_en_demo \
+    --template llama3 \
+    --cutoff_len 2048 \
+    --max_samples 1000 \
+    --overwrite_cache \
+    --preprocessing_num_workers 16 \
+    --dataloader_num_workers 4 \
+    --output_dir saves/llama3-8b/lora/sft \
+    --logging_steps 10 \
+    --save_steps 500 \
+    --plot_loss \
+    --overwrite_output_dir \
+    --save_only_model false \
+    --report_to none \
+    --per_device_train_batch_size 1 \
+    --gradient_accumulation_steps 8 \
+    --learning_rate 1e-4 \
+    --num_train_epochs 3.0 \
+    --lr_scheduler_type cosine \
+    --warmup_ratio 0.1 \
+    --bf16 \
+    --ddp_timeout 180000000

examples/train_lora/llama3_lora_sft.yaml

@@ -1,19 +1,22 @@
 ### model
 model_name_or_path: meta-llama/Meta-Llama-3-8B-Instruct
+trust_remote_code: true
 
 ### method
 stage: sft
 do_train: true
 finetuning_type: lora
+lora_rank: 8
 lora_target: all
 
 ### dataset
 dataset: identity,alpaca_en_demo
 template: llama3
-cutoff_len: 1024
+cutoff_len: 2048
 max_samples: 1000
 overwrite_cache: true
 preprocessing_num_workers: 16
+dataloader_num_workers: 4
 
 ### output
 output_dir: saves/llama3-8b/lora/sft
@@ -21,6 +24,8 @@ logging_steps: 10
 save_steps: 500
 plot_loss: true
 overwrite_output_dir: true
+save_only_model: false
+report_to: none  # choices: [none, wandb, tensorboard, swanlab, mlflow]
 
 ### train
 per_device_train_batch_size: 1
@@ -29,11 +34,13 @@ learning_rate: 1.0e-4
 num_train_epochs: 3.0
 lr_scheduler_type: cosine
 warmup_ratio: 0.1
-fp16: true
+bf16: true
 ddp_timeout: 180000000
+resume_from_checkpoint: null
 
 ### eval
-val_size: 0.1
-per_device_eval_batch_size: 1
-eval_strategy: steps
-eval_steps: 500
+# eval_dataset: alpaca_en_demo
+# val_size: 0.1
+# per_device_eval_batch_size: 1
+# eval_strategy: steps
+# eval_steps: 500

examples/train_lora/llama3_lora_sft_ds3.yaml

@@ -1,20 +1,23 @@
 ### model
 model_name_or_path: meta-llama/Meta-Llama-3-8B-Instruct
+trust_remote_code: true
 
 ### method
 stage: sft
 do_train: true
 finetuning_type: lora
+lora_rank: 8
 lora_target: all
-deepspeed: examples/deepspeed/ds_z3_config.json
+deepspeed: examples/deepspeed/ds_z3_config.json  # choices: [ds_z0_config.json, ds_z2_config.json, ds_z3_config.json]
 
 ### dataset
 dataset: identity,alpaca_en_demo
 template: llama3
-cutoff_len: 1024
+cutoff_len: 2048
 max_samples: 1000
 overwrite_cache: true
 preprocessing_num_workers: 16
+dataloader_num_workers: 4
 
 ### output
 output_dir: saves/llama3-8b/lora/sft
@@ -22,6 +25,8 @@ logging_steps: 10
 save_steps: 500
 plot_loss: true
 overwrite_output_dir: true
+save_only_model: false
+report_to: none  # choices: [none, wandb, tensorboard, swanlab, mlflow]
 
 ### train
 per_device_train_batch_size: 1
@@ -30,11 +35,13 @@ learning_rate: 1.0e-4
 num_train_epochs: 3.0
 lr_scheduler_type: cosine
 warmup_ratio: 0.1
-fp16: true
+bf16: true
 ddp_timeout: 180000000
+resume_from_checkpoint: null
 
 ### eval
-val_size: 0.1
-per_device_eval_batch_size: 1
-eval_strategy: steps
-eval_steps: 500
+# eval_dataset: alpaca_en_demo
+# val_size: 0.1
+# per_device_eval_batch_size: 1
+# eval_strategy: steps
+# eval_steps: 500

examples/train_lora/llama3_lora_sft_ray.yaml

@@ -0,0 +1,61 @@
+### model
+model_name_or_path: meta-llama/Meta-Llama-3-8B-Instruct  # or use local absolute path
+trust_remote_code: true
+
+### method
+stage: sft
+do_train: true
+finetuning_type: lora
+lora_rank: 8
+lora_target: all
+
+### dataset
+dataset: identity,alpaca_en_demo
+dataset_dir: REMOTE:llamafactory/demo_data  # or use local absolute path
+template: llama3
+cutoff_len: 2048
+max_samples: 1000
+overwrite_cache: true
+preprocessing_num_workers: 16
+dataloader_num_workers: 4
+
+### output
+output_dir: tmp_dir
+logging_steps: 10
+save_steps: 500
+plot_loss: true
+overwrite_output_dir: true
+save_only_model: false
+report_to: none  # choices: [none, wandb, tensorboard, swanlab, mlflow]
+
+### ray
+ray_run_name: llama3_8b_sft_lora
+ray_storage_path: ./saves
+ray_num_workers: 4  # Number of GPUs to use.
+placement_strategy: PACK
+resources_per_worker:
+  GPU: 1
+# ray_init_kwargs:
+#   runtime_env:
+#     env_vars:
+#       <YOUR-ENV-VAR-HERE>: "<YOUR-ENV-VAR-HERE>"
+#     pip:
+#       - emoji
+
+### train
+per_device_train_batch_size: 1
+gradient_accumulation_steps: 8
+learning_rate: 1.0e-4
+num_train_epochs: 3.0
+lr_scheduler_type: cosine
+warmup_ratio: 0.1
+bf16: true
+ddp_timeout: 180000000
+resume_from_checkpoint: null
+
+### eval
+# eval_dataset: alpaca_en_demo
+# val_size: 0.1
+# per_device_eval_batch_size: 1
+# eval_strategy: steps
+# eval_steps: 500

examples/train_lora/llama3_preprocess.yaml

@@ -1,16 +1,18 @@
 ### model
 model_name_or_path: meta-llama/Meta-Llama-3-8B-Instruct
+trust_remote_code: true
 
 ### method
 stage: sft
 do_train: true
 finetuning_type: lora
+lora_rank: 8
 lora_target: all
 
 ### dataset
 dataset: identity,alpaca_en_demo
 template: llama3
-cutoff_len: 1024
+cutoff_len: 2048
 max_samples: 1000
 overwrite_cache: true
 preprocessing_num_workers: 16

examples/train_lora/llama4_lora_sft_ds3.yaml

@@ -0,0 +1,49 @@
+# pip install git+https://github.com/hiyouga/transformers.git@llama4_train
+
+### model
+model_name_or_path: meta-llama/Llama-4-Scout-17B-16E-Instruct
+trust_remote_code: true
+
+### method
+stage: sft
+do_train: true
+finetuning_type: lora
+lora_rank: 8
+lora_target: all
+deepspeed: examples/deepspeed/ds_z3_config.json  # choices: [ds_z0_config.json, ds_z2_config.json, ds_z3_config.json]
+
+### dataset
+dataset: mllm_demo,identity,alpaca_en_demo
+template: llama4
+cutoff_len: 2048
+max_samples: 1000
+overwrite_cache: true
+preprocessing_num_workers: 16
+dataloader_num_workers: 4
+
+### output
+output_dir: saves/llama4-8b/lora/sft
+logging_steps: 10
+save_steps: 500
+plot_loss: true
+overwrite_output_dir: true
+save_only_model: false
+report_to: none  # choices: [none, wandb, tensorboard, swanlab, mlflow]
+
+### train
+per_device_train_batch_size: 1
+gradient_accumulation_steps: 2
+learning_rate: 1.0e-4
+num_train_epochs: 3.0
+lr_scheduler_type: cosine
+warmup_ratio: 0.1
+bf16: true
+ddp_timeout: 180000000
+resume_from_checkpoint: null
+
+### eval
+# eval_dataset: alpaca_en_demo
+# val_size: 0.1
+# per_device_eval_batch_size: 1
+# eval_strategy: steps
+# eval_steps: 500

examples/train_lora/llava1_5_lora_sft.yaml

@@ -1,40 +0,0 @@
-### model
-model_name_or_path: llava-hf/llava-1.5-7b-hf
-visual_inputs: true
-
-### method
-stage: sft
-do_train: true
-finetuning_type: lora
-lora_target: all
-
-### dataset
-dataset: mllm_demo
-template: vicuna
-cutoff_len: 1024
-max_samples: 1000
-overwrite_cache: true
-preprocessing_num_workers: 16
-
-### output
-output_dir: saves/llava1_5-7b/lora/sft
-logging_steps: 10
-save_steps: 500
-plot_loss: true
-overwrite_output_dir: true
-
-### train
-per_device_train_batch_size: 1
-gradient_accumulation_steps: 8
-learning_rate: 1.0e-4
-num_train_epochs: 3.0
-lr_scheduler_type: cosine
-warmup_ratio: 0.1
-fp16: true
-ddp_timeout: 180000000
-
-### eval
-val_size: 0.1
-per_device_eval_batch_size: 1
-eval_strategy: steps
-eval_steps: 500

examples/train_lora/qwen2_5vl_lora_dpo.yaml

@@ -0,0 +1,49 @@
+### model
+model_name_or_path: Qwen/Qwen2.5-VL-7B-Instruct
+image_max_pixels: 262144
+video_max_pixels: 16384
+trust_remote_code: true
+
+### method
+stage: dpo
+do_train: true
+finetuning_type: lora
+lora_rank: 8
+lora_target: all
+pref_beta: 0.1
+pref_loss: sigmoid  # choices: [sigmoid (dpo), orpo, simpo]
+
+### dataset
+dataset: rlhf_v
+template: qwen2_vl
+cutoff_len: 2048
+max_samples: 1000
+overwrite_cache: true
+preprocessing_num_workers: 16
+dataloader_num_workers: 4
+
+### output
+output_dir: saves/qwen2_5vl-7b/lora/dpo
+logging_steps: 10
+save_steps: 500
+plot_loss: true
+overwrite_output_dir: true
+save_only_model: false
+report_to: none  # choices: [none, wandb, tensorboard, swanlab, mlflow]
+
+### train
+per_device_train_batch_size: 1
+gradient_accumulation_steps: 8
+learning_rate: 5.0e-6
+num_train_epochs: 3.0
+lr_scheduler_type: cosine
+warmup_ratio: 0.1
+bf16: true
+ddp_timeout: 180000000
+resume_from_checkpoint: null
+
+### eval
+# val_size: 0.1
+# per_device_eval_batch_size: 1
+# eval_strategy: steps
+# eval_steps: 500

examples/train_lora/qwen2_5vl_lora_sft.yaml

@@ -0,0 +1,47 @@
+### model
+model_name_or_path: Qwen/Qwen2.5-VL-7B-Instruct
+image_max_pixels: 262144
+video_max_pixels: 16384
+trust_remote_code: true
+
+### method
+stage: sft
+do_train: true
+finetuning_type: lora
+lora_rank: 8
+lora_target: all
+
+### dataset
+dataset: mllm_demo,identity,alpaca_en_demo  # video: mllm_video_demo
+template: qwen2_vl
+cutoff_len: 2048
+max_samples: 1000
+overwrite_cache: true
+preprocessing_num_workers: 16
+dataloader_num_workers: 4
+
+### output
+output_dir: saves/qwen2_5vl-7b/lora/sft
+logging_steps: 10
+save_steps: 500
+plot_loss: true
+overwrite_output_dir: true
+save_only_model: false
+report_to: none  # choices: [none, wandb, tensorboard, swanlab, mlflow]
+
+### train
+per_device_train_batch_size: 1
+gradient_accumulation_steps: 8
+learning_rate: 1.0e-4
+num_train_epochs: 3.0
+lr_scheduler_type: cosine
+warmup_ratio: 0.1
+bf16: true
+ddp_timeout: 180000000
+resume_from_checkpoint: null
+
+### eval
+# val_size: 0.1
+# per_device_eval_batch_size: 1
+# eval_strategy: steps
+# eval_steps: 500

examples/train_qlora/llama3_lora_sft_aqlm.yaml

@@ -1,19 +1,22 @@
 ### model
 model_name_or_path: ISTA-DASLab/Meta-Llama-3-8B-Instruct-AQLM-2Bit-1x16
+trust_remote_code: true
 
 ### method
 stage: sft
 do_train: true
 finetuning_type: lora
+lora_rank: 8
 lora_target: all
 
 ### dataset
 dataset: identity,alpaca_en_demo
 template: llama3
-cutoff_len: 1024
+cutoff_len: 2048
 max_samples: 1000
 overwrite_cache: true
 preprocessing_num_workers: 16
+dataloader_num_workers: 4
 
 ### output
 output_dir: saves/llama3-8b/lora/sft
@@ -21,6 +24,8 @@ logging_steps: 10
 save_steps: 500
 plot_loss: true
 overwrite_output_dir: true
+save_only_model: false
+report_to: none  # choices: [none, wandb, tensorboard, swanlab, mlflow]
 
 ### train
 per_device_train_batch_size: 1
@@ -29,11 +34,11 @@ learning_rate: 1.0e-4
 num_train_epochs: 3.0
 lr_scheduler_type: cosine
 warmup_ratio: 0.1
-fp16: true
+bf16: true
 ddp_timeout: 180000000
 
 ### eval
-val_size: 0.1
-per_device_eval_batch_size: 1
-eval_strategy: steps
-eval_steps: 500
+# val_size: 0.1
+# per_device_eval_batch_size: 1
+# eval_strategy: steps
+# eval_steps: 500

examples/train_qlora/llama3_lora_sft_awq.yaml

@@ -1,19 +1,22 @@
 ### model
 model_name_or_path: TechxGenus/Meta-Llama-3-8B-Instruct-AWQ
+trust_remote_code: true
 
 ### method
 stage: sft
 do_train: true
 finetuning_type: lora
+lora_rank: 8
 lora_target: all
 
 ### dataset
 dataset: identity,alpaca_en_demo
 template: llama3
-cutoff_len: 1024
+cutoff_len: 2048
 max_samples: 1000
 overwrite_cache: true
 preprocessing_num_workers: 16
+dataloader_num_workers: 4
 
 ### output
 output_dir: saves/llama3-8b/lora/sft
@@ -21,6 +24,8 @@ logging_steps: 10
 save_steps: 500
 plot_loss: true
 overwrite_output_dir: true
+save_only_model: false
+report_to: none  # choices: [none, wandb, tensorboard, swanlab, mlflow]
 
 ### train
 per_device_train_batch_size: 1
@@ -29,11 +34,11 @@ learning_rate: 1.0e-4
 num_train_epochs: 3.0
 lr_scheduler_type: cosine
 warmup_ratio: 0.1
-fp16: true
+bf16: true
 ddp_timeout: 180000000
 
 ### eval
-val_size: 0.1
-per_device_eval_batch_size: 1
-eval_strategy: steps
-eval_steps: 500
+# val_size: 0.1
+# per_device_eval_batch_size: 1
+# eval_strategy: steps
+# eval_steps: 500

examples/train_qlora/llama3_lora_sft_bnb_npu.yaml

@@ -1,20 +1,25 @@
 ### model
 model_name_or_path: meta-llama/Meta-Llama-3-8B-Instruct
 quantization_bit: 4
+quantization_method: bnb
+double_quantization: false
+trust_remote_code: true
 
 ### method
 stage: sft
 do_train: true
 finetuning_type: lora
+lora_rank: 8
 lora_target: all
 
 ### dataset
 dataset: identity,alpaca_en_demo
 template: llama3
-cutoff_len: 1024
+cutoff_len: 2048
 max_samples: 1000
 overwrite_cache: true
 preprocessing_num_workers: 16
+dataloader_num_workers: 4
 
 ### output
 output_dir: saves/llama3-8b/lora/sft
@@ -22,6 +27,8 @@ logging_steps: 10
 save_steps: 500
 plot_loss: true
 overwrite_output_dir: true
+save_only_model: false
+report_to: none  # choices: [none, wandb, tensorboard, swanlab, mlflow]
 
 ### train
 per_device_train_batch_size: 1
@@ -30,11 +37,11 @@ learning_rate: 1.0e-4
 num_train_epochs: 3.0
 lr_scheduler_type: cosine
 warmup_ratio: 0.1
-fp16: true
+bf16: true
 ddp_timeout: 180000000
 
 ### eval
-val_size: 0.1
-per_device_eval_batch_size: 1
-eval_strategy: steps
-eval_steps: 500
+# val_size: 0.1
+# per_device_eval_batch_size: 1
+# eval_strategy: steps
+# eval_steps: 500

examples/train_qlora/llama3_lora_sft_gptq.yaml

@@ -1,19 +1,22 @@
 ### model
 model_name_or_path: TechxGenus/Meta-Llama-3-8B-Instruct-GPTQ
+trust_remote_code: true
 
 ### method
 stage: sft
 do_train: true
 finetuning_type: lora
+lora_rank: 8
 lora_target: all
 
 ### dataset
 dataset: identity,alpaca_en_demo
 template: llama3
-cutoff_len: 1024
+cutoff_len: 2048
 max_samples: 1000
 overwrite_cache: true
 preprocessing_num_workers: 16
+dataloader_num_workers: 4
 
 ### output
 output_dir: saves/llama3-8b/lora/sft
@@ -21,6 +24,8 @@ logging_steps: 10
 save_steps: 500
 plot_loss: true
 overwrite_output_dir: true
+save_only_model: false
+report_to: none  # choices: [none, wandb, tensorboard, swanlab, mlflow]
 
 ### train
 per_device_train_batch_size: 1
@@ -29,11 +34,11 @@ learning_rate: 1.0e-4
 num_train_epochs: 3.0
 lr_scheduler_type: cosine
 warmup_ratio: 0.1
-fp16: true
+bf16: true
 ddp_timeout: 180000000
 
 ### eval
-val_size: 0.1
-per_device_eval_batch_size: 1
-eval_strategy: steps
-eval_steps: 500
+# val_size: 0.1
+# per_device_eval_batch_size: 1
+# eval_strategy: steps
+# eval_steps: 500

examples/train_qlora/llama3_lora_sft_otfq.yaml

@@ -1,20 +1,24 @@
 ### model
 model_name_or_path: meta-llama/Meta-Llama-3-8B-Instruct
+quantization_bit: 4  # choices: [8 (bnb/hqq/eetq), 4 (bnb/hqq), 3 (hqq), 2 (hqq)]
+quantization_method: bnb  # choices: [bnb, hqq, eetq]
+trust_remote_code: true
 
 ### method
 stage: sft
 do_train: true
 finetuning_type: lora
+lora_rank: 8
 lora_target: all
-deepspeed: examples/deepspeed/ds_z0_config.json
 
 ### dataset
 dataset: identity,alpaca_en_demo
 template: llama3
-cutoff_len: 1024
+cutoff_len: 2048
 max_samples: 1000
 overwrite_cache: true
 preprocessing_num_workers: 16
+dataloader_num_workers: 4
 
 ### output
 output_dir: saves/llama3-8b/lora/sft
@@ -22,19 +26,21 @@ logging_steps: 10
 save_steps: 500
 plot_loss: true
 overwrite_output_dir: true
+save_only_model: false
+report_to: none  # choices: [none, wandb, tensorboard, swanlab, mlflow]
 
 ### train
 per_device_train_batch_size: 1
-gradient_accumulation_steps: 2
+gradient_accumulation_steps: 8
 learning_rate: 1.0e-4
 num_train_epochs: 3.0
 lr_scheduler_type: cosine
 warmup_ratio: 0.1
-fp16: true
+bf16: true
 ddp_timeout: 180000000
 
 ### eval
-val_size: 0.1
-per_device_eval_batch_size: 1
-eval_strategy: steps
-eval_steps: 500
+# val_size: 0.1
+# per_device_eval_batch_size: 1
+# eval_strategy: steps
+# eval_steps: 500

pyproject.toml

@@ -2,14 +2,53 @@
 requires = ["setuptools>=61.0"]
 build-backend = "setuptools.build_meta"
 
+[project]
+name = "llamafactory"
+dynamic = [
+    "version",
+    "dependencies",
+    "optional-dependencies",
+    "requires-python",
+    "scripts",
+    "authors",
+    "description",
+    "readme",
+    "license",
+    "keywords",
+    "classifiers"
+]
+
 [tool.ruff]
-target-version = "py38"
+target-version = "py39"
 line-length = 119
 indent-width = 4
 
 [tool.ruff.lint]
-ignore = ["C408", "C901", "E501", "E731", "E741", "W605"]
-select = ["C", "E", "F", "I", "W"]
+ignore = [
+    "C408", # collection
+    "C901", # complex
+    "E501", # line too long
+    "E731", # lambda function
+    "E741", # ambiguous var name
+    "D100", # no doc public module
+    "D101", # no doc public class
+    "D102", # no doc public method
+    "D103", # no doc public function
+    "D104", # no doc public package
+    "D105", # no doc magic method
+    "D107", # no doc __init__
+]
+extend-select = [
+    "C",      # complexity
+    "E",      # error
+    "F",      # pyflakes
+    "I",      # isort
+    "W",      # warning
+    "UP",     # pyupgrade
+    "D",      # pydocstyle
+    "PT009",  # pytest assert
+    "RUF022", # sort __all__
+]
 
 [tool.ruff.lint.isort]
 lines-after-imports = 2
@@ -22,12 +61,35 @@ known-third-party = [
     "peft",
     "torch",
     "transformers",
-    "trl"
+    "trl",
 ]
 
+[tool.ruff.lint.pydocstyle]
+convention = "google"
+
 [tool.ruff.format]
 quote-style = "double"
 indent-style = "space"
 docstring-code-format = true
 skip-magic-trailing-comma = false
 line-ending = "auto"
+
+[tool.uv]
+conflicts = [
+    [
+        { extra = "torch-npu" },
+        { extra = "aqlm" },
+    ],
+    [
+        { extra = "torch-npu" },
+        { extra = "vllm" },
+    ],
+    [
+        { extra = "torch-npu" },
+        { extra = "sglang" },
+    ],
+    [
+        { extra = "vllm" },
+        { extra = "sglang" },
+    ],
+]

requirements.txt

@@ -1,20 +1,27 @@
-transformers>=4.41.2
-datasets>=2.16.0
-accelerate>=0.30.1
-peft>=0.11.1
-trl>=0.8.6
-gradio>=4.0.0
-pandas>=2.0.0
+transformers>=4.45.0,<=4.52.4,!=4.46.*,!=4.47.*,!=4.48.0,!=4.52.0; sys_platform != 'darwin'
+transformers>=4.45.0,<=4.51.3,!=4.46.*,!=4.47.*,!=4.48.0,!=4.52.0; sys_platform == 'darwin'
+datasets>=2.16.0,<=3.6.0
+accelerate>=0.34.0,<=1.7.0
+peft>=0.14.0,<=0.15.2
+trl>=0.8.6,<=0.9.6
+tokenizers>=0.19.0,<=0.21.1
+gradio>=4.38.0,<=5.31.0
 scipy
 einops
 sentencepiece
 tiktoken
 protobuf
 uvicorn
-pydantic
 fastapi
 sse-starlette
 matplotlib>=3.7.0
 fire
+omegaconf
 packaging
 pyyaml
+numpy<2.0.0
+pydantic<=2.10.6
+pandas>=2.0.0
+av
+librosa
+tyro<0.9.0

scripts/api_example/test_image.py

@@ -0,0 +1,65 @@
+# Copyright 2025 the LlamaFactory team.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+import os
+
+from openai import OpenAI
+from transformers.utils.versions import require_version
+
+
+require_version("openai>=1.5.0", "To fix: pip install openai>=1.5.0")
+
+
+def main():
+    client = OpenAI(
+        api_key="{}".format(os.getenv("API_KEY", "0")),
+        base_url="http://localhost:{}/v1".format(os.getenv("API_PORT", 8000)),
+    )
+    messages = []
+    messages.append(
+        {
+            "role": "user",
+            "content": [
+                {"type": "text", "text": "Output the color and number of each box."},
+                {
+                    "type": "image_url",
+                    "image_url": {"url": "https://qianwen-res.oss-cn-beijing.aliyuncs.com/Qwen2-VL/boxes.png"},
+                },
+            ],
+        }
+    )
+    result = client.chat.completions.create(messages=messages, model="test")
+    messages.append(result.choices[0].message)
+    print("Round 1:", result.choices[0].message.content)
+    # The image shows a pyramid of colored blocks with numbers on them. Here are the colors and numbers of ...
+    messages.append(
+        {
+            "role": "user",
+            "content": [
+                {"type": "text", "text": "What kind of flower is this?"},
+                {
+                    "type": "image_url",
+                    "image_url": {"url": "https://qianwen-res.oss-cn-beijing.aliyuncs.com/Qwen2-VL/flowers.jpg"},
+                },
+            ],
+        }
+    )
+    result = client.chat.completions.create(messages=messages, model="test")
+    messages.append(result.choices[0].message)
+    print("Round 2:", result.choices[0].message.content)
+    # The image shows a cluster of forget-me-not flowers. Forget-me-nots are small ...
+
+
+if __name__ == "__main__":
+    main()

scripts/api_example/test_toolcall.py

@@ -1,5 +1,4 @@
-# coding=utf-8
-# Copyright 2024 the LlamaFactory team.
+# Copyright 2025 the LlamaFactory team.
 #
 # Licensed under the Apache License, Version 2.0 (the "License");
 # you may not use this file except in compliance with the License.
@@ -15,7 +14,6 @@
 
 import json
 import os
-from typing import Sequence
 
 from openai import OpenAI
 from transformers.utils.versions import require_version
@@ -24,7 +22,7 @@ from transformers.utils.versions import require_version
 require_version("openai>=1.5.0", "To fix: pip install openai>=1.5.0")
 
 
-def calculate_gpa(grades: Sequence[str], hours: Sequence[int]) -> float:
+def calculate_gpa(grades: list[str], hours: list[int]) -> float:
     grade_to_score = {"A": 4, "B": 3, "C": 2}
     total_score, total_hour = 0, 0
     for grade, hour in zip(grades, hours):
@@ -35,8 +33,8 @@ def calculate_gpa(grades: Sequence[str], hours: Sequence[int]) -> float:
 
 def main():
     client = OpenAI(
-        api_key="{}".format(os.environ.get("API_KEY", "0")),
-        base_url="http://localhost:{}/v1".format(os.environ.get("API_PORT", 8000)),
+        api_key="{}".format(os.getenv("API_KEY", "0")),
+        base_url="http://localhost:{}/v1".format(os.getenv("API_PORT", 8000)),
     )
     tools = [
         {

scripts/convert_ckpt/llamafy_baichuan2.py

@@ -1,5 +1,4 @@
-# coding=utf-8
-# Copyright 2024 the LlamaFactory team.
+# Copyright 2025 the LlamaFactory team.
 #
 # Licensed under the Apache License, Version 2.0 (the "License");
 # you may not use this file except in compliance with the License.
@@ -16,64 +15,67 @@
 import json
 import os
 from collections import OrderedDict
-from typing import Any, Dict, Optional
+from typing import Any
 
 import fire
 import torch
+from huggingface_hub import split_torch_state_dict_into_shards
 from safetensors.torch import save_file
 from tqdm import tqdm
-from transformers.modeling_utils import (
-    SAFE_WEIGHTS_INDEX_NAME,
-    SAFE_WEIGHTS_NAME,
-    WEIGHTS_INDEX_NAME,
-    WEIGHTS_NAME,
-    shard_checkpoint,
-)
+from transformers.modeling_utils import SAFE_WEIGHTS_INDEX_NAME, SAFE_WEIGHTS_NAME, WEIGHTS_INDEX_NAME, WEIGHTS_NAME
 
 
 CONFIG_NAME = "config.json"
 
 
 def save_weight(input_dir: str, output_dir: str, shard_size: str, save_safetensors: bool):
-    baichuan2_state_dict: Dict[str, torch.Tensor] = OrderedDict()
+    baichuan2_state_dict: dict[str, torch.Tensor] = OrderedDict()
     for filepath in tqdm(os.listdir(input_dir), desc="Load weights"):
         if os.path.isfile(os.path.join(input_dir, filepath)) and filepath.endswith(".bin"):
-            shard_weight = torch.load(os.path.join(input_dir, filepath), map_location="cpu")
+            shard_weight = torch.load(os.path.join(input_dir, filepath), map_location="cpu", weights_only=True)
             baichuan2_state_dict.update(shard_weight)
 
-    llama2_state_dict: Dict[str, torch.Tensor] = OrderedDict()
+    llama_state_dict: dict[str, torch.Tensor] = OrderedDict()
     for key, value in tqdm(baichuan2_state_dict.items(), desc="Convert format"):
         if "W_pack" in key:
             proj_size = value.size(0) // 3
-            llama2_state_dict[key.replace("W_pack", "q_proj")] = value[:proj_size, :]
-            llama2_state_dict[key.replace("W_pack", "k_proj")] = value[proj_size : 2 * proj_size, :]
-            llama2_state_dict[key.replace("W_pack", "v_proj")] = value[2 * proj_size :, :]
+            llama_state_dict[key.replace("W_pack", "q_proj")] = value[:proj_size, :]
+            llama_state_dict[key.replace("W_pack", "k_proj")] = value[proj_size : 2 * proj_size, :]
+            llama_state_dict[key.replace("W_pack", "v_proj")] = value[2 * proj_size :, :]
         elif "lm_head" in key:
-            llama2_state_dict[key] = torch.nn.functional.normalize(value)
+            llama_state_dict[key] = torch.nn.functional.normalize(value)
         else:
-            llama2_state_dict[key] = value
+            llama_state_dict[key] = value
 
     weights_name = SAFE_WEIGHTS_NAME if save_safetensors else WEIGHTS_NAME
-    shards, index = shard_checkpoint(llama2_state_dict, max_shard_size=shard_size, weights_name=weights_name)
-
-    for shard_file, shard in tqdm(shards.items(), desc="Save weights"):
+    filename_pattern = weights_name.replace(".bin", "{suffix}.bin").replace(".safetensors", "{suffix}.safetensors")
+    state_dict_split = split_torch_state_dict_into_shards(
+        llama_state_dict, filename_pattern=filename_pattern, max_shard_size=shard_size
+    )
+    for shard_file, tensors in tqdm(state_dict_split.filename_to_tensors.items(), desc="Save weights"):
+        shard = {tensor: llama_state_dict[tensor].contiguous() for tensor in tensors}
         if save_safetensors:
             save_file(shard, os.path.join(output_dir, shard_file), metadata={"format": "pt"})
         else:
             torch.save(shard, os.path.join(output_dir, shard_file))
 
-    if index is None:
-        print("Model weights saved in {}".format(os.path.join(output_dir, WEIGHTS_NAME)))
+    if not state_dict_split.is_sharded:
+        print(f"Model weights saved in {os.path.join(output_dir, weights_name)}.")
     else:
+        index = {
+            "metadata": state_dict_split.metadata,
+            "weight_map": state_dict_split.tensor_to_filename,
+        }
         index_name = SAFE_WEIGHTS_INDEX_NAME if save_safetensors else WEIGHTS_INDEX_NAME
         with open(os.path.join(output_dir, index_name), "w", encoding="utf-8") as f:
             json.dump(index, f, indent=2, sort_keys=True)
-        print("Model weights saved in {}".format(output_dir))
+
+        print(f"Model weights saved in {output_dir}.")
 
 
 def save_config(input_dir: str, output_dir: str):
-    with open(os.path.join(input_dir, CONFIG_NAME), "r", encoding="utf-8") as f:
-        llama2_config_dict: Dict[str, Any] = json.load(f)
+    with open(os.path.join(input_dir, CONFIG_NAME), encoding="utf-8") as f:
+        llama2_config_dict: dict[str, Any] = json.load(f)
 
     llama2_config_dict["architectures"] = ["LlamaForCausalLM"]
     llama2_config_dict.pop("auto_map", None)
@@ -82,14 +84,18 @@ def save_config(input_dir: str, output_dir: str):
 
     with open(os.path.join(output_dir, CONFIG_NAME), "w", encoding="utf-8") as f:
         json.dump(llama2_config_dict, f, indent=2)
-    print("Model config saved in {}".format(os.path.join(output_dir, CONFIG_NAME)))
+
+    print(f"Model config saved in {os.path.join(output_dir, CONFIG_NAME)}")
 
 
 def llamafy_baichuan2(
-    input_dir: str, output_dir: str, shard_size: Optional[str] = "2GB", save_safetensors: Optional[bool] = False
+    input_dir: str,
+    output_dir: str,
+    shard_size: str = "2GB",
+    save_safetensors: bool = True,
 ):
-    r"""
-    Converts the Baichuan2-7B model in the same format as LLaMA2-7B.
+    r"""Convert the Baichuan2-7B model in the same format as LLaMA2-7B.
+
     Usage: python llamafy_baichuan2.py --input_dir input --output_dir output
     Converted model: https://huggingface.co/hiyouga/Baichuan2-7B-Base-LLaMAfied
     """

scripts/convert_ckpt/llamafy_qwen.py

@@ -1,5 +1,4 @@
-# coding=utf-8
-# Copyright 2024 the LlamaFactory team.
+# Copyright 2025 the LlamaFactory team.
 #
 # Licensed under the Apache License, Version 2.0 (the "License");
 # you may not use this file except in compliance with the License.
@@ -16,20 +15,15 @@
 import json
 import os
 from collections import OrderedDict
-from typing import Any, Dict, Optional
+from typing import Any
 
 import fire
 import torch
+from huggingface_hub import split_torch_state_dict_into_shards
 from safetensors import safe_open
 from safetensors.torch import save_file
 from tqdm import tqdm
-from transformers.modeling_utils import (
-    SAFE_WEIGHTS_INDEX_NAME,
-    SAFE_WEIGHTS_NAME,
-    WEIGHTS_INDEX_NAME,
-    WEIGHTS_NAME,
-    shard_checkpoint,
-)
+from transformers.modeling_utils import SAFE_WEIGHTS_INDEX_NAME, SAFE_WEIGHTS_NAME, WEIGHTS_INDEX_NAME, WEIGHTS_NAME
 from transformers.utils import check_min_version
 
 
@@ -43,76 +37,84 @@ CONFIG_NAME = "config.json"
 
 
 def save_weight(input_dir: str, output_dir: str, shard_size: str, save_safetensors: bool) -> str:
-    qwen_state_dict: Dict[str, torch.Tensor] = OrderedDict()
+    qwen_state_dict: dict[str, torch.Tensor] = OrderedDict()
     for filepath in tqdm(os.listdir(input_dir), desc="Load weights"):
         if os.path.isfile(os.path.join(input_dir, filepath)) and filepath.endswith(".safetensors"):
             with safe_open(os.path.join(input_dir, filepath), framework="pt", device="cpu") as f:
                 for key in f.keys():
                     qwen_state_dict[key] = f.get_tensor(key)
 
-    llama2_state_dict: Dict[str, torch.Tensor] = OrderedDict()
+    llama_state_dict: dict[str, torch.Tensor] = OrderedDict()
     torch_dtype = None
     for key, value in tqdm(qwen_state_dict.items(), desc="Convert format"):
         if torch_dtype is None:
             torch_dtype = value.dtype
         if "wte" in key:
-            llama2_state_dict["model.embed_tokens.weight"] = value
+            llama_state_dict["model.embed_tokens.weight"] = value
         elif "ln_f" in key:
-            llama2_state_dict["model.norm.weight"] = value
+            llama_state_dict["model.norm.weight"] = value
         else:
             key = key.replace("transformer.h", "model.layers")
             if "attn.c_attn" in key:
                 proj_size = value.size(0) // 3
-                llama2_state_dict[key.replace("attn.c_attn", "self_attn.q_proj")] = value[:proj_size, ...]
-                llama2_state_dict[key.replace("attn.c_attn", "self_attn.k_proj")] = value[
+                llama_state_dict[key.replace("attn.c_attn", "self_attn.q_proj")] = value[:proj_size, ...]
+                llama_state_dict[key.replace("attn.c_attn", "self_attn.k_proj")] = value[
                     proj_size : 2 * proj_size, ...
                 ]
-                llama2_state_dict[key.replace("attn.c_attn", "self_attn.v_proj")] = value[2 * proj_size :, ...]
+                llama_state_dict[key.replace("attn.c_attn", "self_attn.v_proj")] = value[2 * proj_size :, ...]
             elif "attn.c_proj" in key:
-                llama2_state_dict[key.replace("attn.c_proj", "self_attn.o_proj")] = value
-                llama2_state_dict[key.replace("attn.c_proj.weight", "self_attn.o_proj.bias")] = torch.zeros_like(
+                llama_state_dict[key.replace("attn.c_proj", "self_attn.o_proj")] = value
+                llama_state_dict[key.replace("attn.c_proj.weight", "self_attn.o_proj.bias")] = torch.zeros_like(
                     value[:, 0]
                 ).squeeze()
             elif "ln_1" in key:
-                llama2_state_dict[key.replace("ln_1", "input_layernorm")] = value
+                llama_state_dict[key.replace("ln_1", "input_layernorm")] = value
             elif "ln_2" in key:
-                llama2_state_dict[key.replace("ln_2", "post_attention_layernorm")] = value
+                llama_state_dict[key.replace("ln_2", "post_attention_layernorm")] = value
             elif "mlp.w1" in key:
-                llama2_state_dict[key.replace("mlp.w1", "mlp.up_proj")] = value
+                llama_state_dict[key.replace("mlp.w1", "mlp.up_proj")] = value
             elif "mlp.w2" in key:
-                llama2_state_dict[key.replace("mlp.w2", "mlp.gate_proj")] = value
+                llama_state_dict[key.replace("mlp.w2", "mlp.gate_proj")] = value
             elif "mlp.c_proj" in key:
-                llama2_state_dict[key.replace("mlp.c_proj", "mlp.down_proj")] = value
+                llama_state_dict[key.replace("mlp.c_proj", "mlp.down_proj")] = value
             elif "lm_head" in key:
-                llama2_state_dict[key] = value
+                llama_state_dict[key] = value
             else:
-                raise KeyError("Unable to process key {}".format(key))
+                raise KeyError(f"Unable to process key {key}")
 
     weights_name = SAFE_WEIGHTS_NAME if save_safetensors else WEIGHTS_NAME
-    shards, index = shard_checkpoint(llama2_state_dict, max_shard_size=shard_size, weights_name=weights_name)
-
-    for shard_file, shard in tqdm(shards.items(), desc="Save weights"):
+    filename_pattern = weights_name.replace(".bin", "{suffix}.bin").replace(".safetensors", "{suffix}.safetensors")
+    state_dict_split = split_torch_state_dict_into_shards(
+        llama_state_dict, filename_pattern=filename_pattern, max_shard_size=shard_size
+    )
+    for shard_file, tensors in tqdm(state_dict_split.filename_to_tensors.items(), desc="Save weights"):
+        shard = {tensor: llama_state_dict[tensor].contiguous() for tensor in tensors}
         if save_safetensors:
             save_file(shard, os.path.join(output_dir, shard_file), metadata={"format": "pt"})
         else:
             torch.save(shard, os.path.join(output_dir, shard_file))
 
-    if index is None:
-        print("Model weights saved in {}".format(os.path.join(output_dir, weights_name)))
+    if not state_dict_split.is_sharded:
+        print(f"Model weights saved in {os.path.join(output_dir, weights_name)}.")
     else:
+        index = {
+            "metadata": state_dict_split.metadata,
+            "weight_map": state_dict_split.tensor_to_filename,
+        }
         index_name = SAFE_WEIGHTS_INDEX_NAME if save_safetensors else WEIGHTS_INDEX_NAME
         with open(os.path.join(output_dir, index_name), "w", encoding="utf-8") as f:
             json.dump(index, f, indent=2, sort_keys=True)
-        print("Model weights saved in {}".format(output_dir))
+
+        print(f"Model weights saved in {output_dir}.")
 
     return str(torch_dtype).replace("torch.", "")
 
 
 def save_config(input_dir: str, output_dir: str, torch_dtype: str):
-    with open(os.path.join(input_dir, CONFIG_NAME), "r", encoding="utf-8") as f:
-        qwen_config_dict: Dict[str, Any] = json.load(f)
+    with open(os.path.join(input_dir, CONFIG_NAME), encoding="utf-8") as f:
+        qwen_config_dict: dict[str, Any] = json.load(f)
 
-    llama2_config_dict: Dict[str, Any] = OrderedDict()
+    llama2_config_dict: dict[str, Any] = OrderedDict()
     llama2_config_dict["architectures"] = ["LlamaForCausalLM"]
     llama2_config_dict["hidden_act"] = "silu"
     llama2_config_dict["hidden_size"] = qwen_config_dict["hidden_size"]
@@ -135,14 +137,18 @@ def save_config(input_dir: str, output_dir: str, torch_dtype: str):
 
     with open(os.path.join(output_dir, CONFIG_NAME), "w", encoding="utf-8") as f:
         json.dump(llama2_config_dict, f, indent=2)
-    print("Model config saved in {}".format(os.path.join(output_dir, CONFIG_NAME)))
+
+    print(f"Model config saved in {os.path.join(output_dir, CONFIG_NAME)}")
 
 
 def llamafy_qwen(
-    input_dir: str, output_dir: str, shard_size: Optional[str] = "2GB", save_safetensors: Optional[bool] = False
+    input_dir: str,
+    output_dir: str,
+    shard_size: str = "2GB",
+    save_safetensors: bool = False,
 ):
-    r"""
-    Converts the Qwen models in the same format as LLaMA2.
+    r"""Convert the Qwen models in the same format as LLaMA2.
+
     Usage: python llamafy_qwen.py --input_dir input --output_dir output
     Converted model: https://huggingface.co/hiyouga/Qwen-14B-Chat-LLaMAfied
     """

scripts/convert_ckpt/tiny_llama4.py

@@ -0,0 +1,39 @@
+# Copyright 2025 the LlamaFactory team.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+from transformers import Llama4Config, Llama4ForConditionalGeneration, Llama4TextConfig, Llama4VisionConfig
+
+
+if __name__ == "__main__":
+    vision_config = Llama4VisionConfig(
+        hidden_size=1408,
+        image_size=336,
+        intermediate_size=5632,
+        num_attention_heads=16,
+        num_hidden_layers=4,
+        vision_output_dim=4096,
+    )
+    text_config = Llama4TextConfig(
+        hidden_size=512,
+        intermediate_size=1024,
+        intermediate_size_mlp=1024,
+        num_hidden_layers=4,
+        num_attention_heads=8,
+        num_key_value_heads=2,
+        head_dim=512 // 8,
+        num_local_experts=2,
+    )
+    config = Llama4Config(vision_config=vision_config, text_config=text_config)
+    model = Llama4ForConditionalGeneration._from_config(config)
+    model.save_pretrained("tiny-llama4")