Upload folder using huggingface_hub

README.md

@@ -1,12 +1,6 @@
 ---
-title: Src
-emoji: 👀
-colorFrom: yellow
-colorTo: blue
+title: src
+app_file: webui.py
 sdk: gradio
 sdk_version: 4.44.1
-app_file: app.py
-pinned: false
 ---
-
-Check out the configuration reference at https://huggingface.co/docs/hub/spaces-config-reference

api.py

@@ -0,0 +1,33 @@
+# Copyright 2024 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
+
+import uvicorn
+
+from llamafactory.api.app import create_app
+from llamafactory.chat import ChatModel
+
+
+def main():
+    chat_model = ChatModel()
+    app = create_app(chat_model)
+    api_host = os.environ.get("API_HOST", "0.0.0.0")
+    api_port = int(os.environ.get("API_PORT", "8000"))
+    print("Visit http://localhost:{}/docs for API document.".format(api_port))
+    uvicorn.run(app, host=api_host, port=api_port)
+
+
+if __name__ == "__main__":
+    main()

llamafactory.egg-info/PKG-INFO

@@ -0,0 +1,815 @@
+Metadata-Version: 2.1
+Name: llamafactory
+Version: 0.9.1.dev0
+Summary: Easy-to-use LLM fine-tuning framework
+Home-page: https://github.com/hiyouga/LLaMA-Factory
+Author: hiyouga
+Author-email: hiyouga@buaa.edu.cn
+License: Apache 2.0 License
+Keywords: LLaMA,BLOOM,Falcon,LLM,ChatGPT,transformer,pytorch,deep learning
+Classifier: Development Status :: 4 - Beta
+Classifier: Intended Audience :: Developers
+Classifier: Intended Audience :: Education
+Classifier: Intended Audience :: Science/Research
+Classifier: License :: OSI Approved :: Apache Software License
+Classifier: Operating System :: OS Independent
+Classifier: Programming Language :: Python :: 3
+Classifier: Programming Language :: Python :: 3.8
+Classifier: Programming Language :: Python :: 3.9
+Classifier: Programming Language :: Python :: 3.10
+Classifier: Programming Language :: Python :: 3.11
+Classifier: Topic :: Scientific/Engineering :: Artificial Intelligence
+Requires-Python: >=3.8.0
+Description-Content-Type: text/markdown
+License-File: LICENSE
+Requires-Dist: transformers<=4.45.0,>=4.41.2
+Requires-Dist: datasets<=2.21.0,>=2.16.0
+Requires-Dist: accelerate<=0.34.2,>=0.30.1
+Requires-Dist: peft<=0.12.0,>=0.11.1
+Requires-Dist: trl<=0.9.6,>=0.8.6
+Requires-Dist: gradio>=4.0.0
+Requires-Dist: pandas>=2.0.0
+Requires-Dist: scipy
+Requires-Dist: einops
+Requires-Dist: sentencepiece
+Requires-Dist: tiktoken
+Requires-Dist: protobuf
+Requires-Dist: uvicorn
+Requires-Dist: pydantic
+Requires-Dist: fastapi
+Requires-Dist: sse-starlette
+Requires-Dist: matplotlib>=3.7.0
+Requires-Dist: fire
+Requires-Dist: packaging
+Requires-Dist: pyyaml
+Requires-Dist: numpy<2.0.0
+Requires-Dist: av
+Provides-Extra: torch
+Requires-Dist: torch>=1.13.1; extra == "torch"
+Provides-Extra: torch-npu
+Requires-Dist: torch==2.1.0; extra == "torch-npu"
+Requires-Dist: torch-npu==2.1.0.post3; extra == "torch-npu"
+Requires-Dist: decorator; extra == "torch-npu"
+Provides-Extra: metrics
+Requires-Dist: nltk; extra == "metrics"
+Requires-Dist: jieba; extra == "metrics"
+Requires-Dist: rouge-chinese; extra == "metrics"
+Provides-Extra: deepspeed
+Requires-Dist: deepspeed<=0.14.4,>=0.10.0; extra == "deepspeed"
+Provides-Extra: liger-kernel
+Requires-Dist: liger-kernel; extra == "liger-kernel"
+Provides-Extra: bitsandbytes
+Requires-Dist: bitsandbytes>=0.39.0; extra == "bitsandbytes"
+Provides-Extra: hqq
+Requires-Dist: hqq; extra == "hqq"
+Provides-Extra: eetq
+Requires-Dist: eetq; extra == "eetq"
+Provides-Extra: gptq
+Requires-Dist: optimum>=1.17.0; extra == "gptq"
+Requires-Dist: auto-gptq>=0.5.0; extra == "gptq"
+Provides-Extra: awq
+Requires-Dist: autoawq; extra == "awq"
+Provides-Extra: aqlm
+Requires-Dist: aqlm[gpu]>=1.1.0; extra == "aqlm"
+Provides-Extra: vllm
+Requires-Dist: vllm<=0.6.2,>=0.4.3; extra == "vllm"
+Provides-Extra: galore
+Requires-Dist: galore-torch; extra == "galore"
+Provides-Extra: badam
+Requires-Dist: badam>=1.2.1; extra == "badam"
+Provides-Extra: adam-mini
+Requires-Dist: adam-mini; extra == "adam-mini"
+Provides-Extra: qwen
+Requires-Dist: transformers_stream_generator; extra == "qwen"
+Provides-Extra: modelscope
+Requires-Dist: modelscope; extra == "modelscope"
+Provides-Extra: dev
+Requires-Dist: ruff; extra == "dev"
+Requires-Dist: pytest; extra == "dev"
+
+![# 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)
+[![PyPI](https://img.shields.io/pypi/v/llamafactory)](https://pypi.org/project/llamafactory/)
+[![Citation](https://img.shields.io/badge/citation-91-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)
+[![Twitter](https://img.shields.io/twitter/follow/llamafactory_ai)](https://twitter.com/llamafactory_ai)
+[![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)
+
+[![GitHub Tread](https://trendshift.io/api/badge/repositories/4535)](https://trendshift.io/repositories/4535)
+
+👋 Join our [WeChat](assets/wechat.jpg) or [NPU user group](assets/wechat_npu.jpg).
+
+\[ English | [中文](README_zh.md) \]
+
+**Fine-tuning a large language model can be easy as...**
+
+https://github.com/user-attachments/assets/7c96b465-9df7-45f4-8053-bf03e58386d3
+
+Choose your path:
+
+- **Colab**: https://colab.research.google.com/drive/1eRTPn37ltBbYsISy9Aw2NuI2Aq5CQrD9?usp=sharing
+- **PAI-DSW**: [Llama3 Example](https://gallery.pai-ml.com/#/preview/deepLearning/nlp/llama_factory) | [Qwen2-VL Example](https://gallery.pai-ml.com/#/preview/deepLearning/nlp/llama_factory_qwen2vl)
+- **Local machine**: Please refer to [usage](#getting-started)
+- **Documentation (WIP)**: https://llamafactory.readthedocs.io/zh-cn/latest/
+
+> [!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)
+- [Changelog](#changelog)
+- [Supported Models](#supported-models)
+- [Supported Training Approaches](#supported-training-approaches)
+- [Provided Datasets](#provided-datasets)
+- [Requirement](#requirement)
+- [Getting Started](#getting-started)
+- [Projects using LLaMA Factory](#projects-using-llama-factory)
+- [License](#license)
+- [Citation](#citation)
+- [Acknowledgement](#acknowledgement)
+
+## Features
+
+- **Various models**: LLaMA, LLaVA, Mistral, Mixtral-MoE, Qwen, Qwen2-VL, Yi, Gemma, Baichuan, ChatGLM, Phi, etc.
+- **Integrated methods**: (Continuous) pre-training, (multimodal) supervised fine-tuning, reward modeling, PPO, DPO, KTO, ORPO, etc.
+- **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), [Adam-mini](https://github.com/zyushun/Adam-mini), DoRA, LongLoRA, LLaMA Pro, Mixture-of-Depths, LoRA+, LoftQ, PiSSA and Agent tuning.
+- **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.
+- **Experiment monitors**: LlamaBoard, TensorBoard, Wandb, MLflow, etc.
+- **Faster inference**: OpenAI-style API, Gradio UI and CLI with vLLM worker.
+
+## Benchmark
+
+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.
+
+![benchmark](assets/benchmark.svg)
+
+<details><summary>Definitions</summary>
+
+- **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>
+
+## Changelog
+
+[24/09/19] We support fine-tuning the **[Qwen2.5](https://qwenlm.github.io/blog/qwen2.5/)** models.
+
+[24/08/30] We support 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 support **[Liger Kernel](https://github.com/linkedin/Liger-Kernel)**. Try `enable_liger_kernel: true` for efficient training.
+
+[24/08/09] We support **[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.
+
+<details><summary>Full Changelog</summary>
+
+[24/07/04] We support [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 support **[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.
+
+[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.
+
+[24/05/14] We supported training and inference on the Ascend NPU devices. Check [installation](#installation) section for details.
+
+[24/04/26] We supported fine-tuning the **LLaVA-1.5** multimodal LLMs. See [examples](examples/README.md) for usage.
+
+[24/04/22] We provided a **[Colab notebook](https://colab.research.google.com/drive/1eRTPn37ltBbYsISy9Aw2NuI2Aq5CQrD9?usp=sharing)** for fine-tuning the Llama-3 model on a free T4 GPU. Two Llama-3-derived models fine-tuned using LLaMA Factory are available at Hugging Face, check [Llama3-8B-Chinese-Chat](https://huggingface.co/shenzhi-wang/Llama3-8B-Chinese-Chat) and [Llama3-Chinese](https://huggingface.co/zhichen/Llama3-Chinese) for details.
+
+[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)** 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).
+
+[24/03/31] We supported **[ORPO](https://arxiv.org/abs/2403.07691)**. See [examples](examples/README.md) for usage.
+
+[24/03/21] Our paper "[LlamaFactory: Unified Efficient Fine-Tuning of 100+ Language Models](https://arxiv.org/abs/2403.13372)" is available at arXiv!
+
+[24/03/20] We supported **FSDP+QLoRA** that fine-tunes a 70B model on 2x24GB GPUs. See [examples](examples/README.md) for usage.
+
+[24/03/13] We supported **[LoRA+](https://arxiv.org/abs/2402.12354)**. 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.
+
+[24/02/28] We supported weight-decomposed LoRA (**[DoRA](https://arxiv.org/abs/2402.09353)**). Try `use_dora: true` to activate DoRA training.
+
+[24/02/15] We supported **block expansion** proposed by [LLaMA Pro](https://github.com/TencentARC/LLaMA-Pro). See [examples](examples/README.md) for usage.
+
+[24/02/05] Qwen1.5 (Qwen2 beta version) series models are supported in LLaMA-Factory. Check this [blog post](https://qwenlm.github.io/blog/qwen1.5/) for details.
+
+[24/01/18] We supported **agent tuning** for most models, equipping model with tool using abilities by fine-tuning with `dataset: glaive_toolcall_en`.
+
+[23/12/23] We supported **[unsloth](https://github.com/unslothai/unsloth)**'s implementation to boost LoRA tuning for the LLaMA, Mistral and Yi models. Try `use_unsloth: true` argument to activate unsloth patch. It achieves **170%** speed in our benchmark, check [this page](https://github.com/hiyouga/LLaMA-Factory/wiki/Performance-comparison) for details.
+
+[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.
+
+[23/10/21] We supported **[NEFTune](https://arxiv.org/abs/2310.05914)** trick for fine-tuning. Try `neftune_noise_alpha: 5` argument to activate NEFTune.
+
+[23/09/27] We supported **$S^2$-Attn** proposed by [LongLoRA](https://github.com/dvlab-research/LongLoRA) for the LLaMA models. Try `shift_attn: true` argument to enable shift short attention.
+
+[23/09/23] We integrated MMLU, C-Eval and CMMLU benchmarks in this repo. See [examples](examples/README.md) for usage.
+
+[23/09/10] We supported **[FlashAttention-2](https://github.com/Dao-AILab/flash-attention)**. Try `flash_attn: fa2` argument to enable FlashAttention-2 if you are using RTX4090, A100 or H100 GPUs.
+
+[23/08/12] We supported **RoPE scaling** to extend the context length of the LLaMA models. Try `rope_scaling: linear` argument in training and `rope_scaling: dynamic` argument at inference to extrapolate the position embeddings.
+
+[23/08/11] We supported **[DPO training](https://arxiv.org/abs/2305.18290)** for instruction-tuned models. See [examples](examples/README.md) for usage.
+
+[23/07/31] We supported **dataset streaming**. Try `streaming: true` and `max_steps: 10000` arguments to load your dataset in streaming mode.
+
+[23/07/29] We released two instruction-tuned 13B models at Hugging Face. See these Hugging Face Repos ([LLaMA-2](https://huggingface.co/hiyouga/Llama-2-Chinese-13b-chat) / [Baichuan](https://huggingface.co/hiyouga/Baichuan-13B-sft)) for details.
+
+[23/07/18] We developed an **all-in-one Web UI** for training, evaluation and inference. Try `train_web.py` to fine-tune models in your Web browser. Thank [@KanadeSiina](https://github.com/KanadeSiina) and [@codemayq](https://github.com/codemayq) for their efforts in the development.
+
+[23/07/09] We released **[FastEdit](https://github.com/hiyouga/FastEdit)** ⚡🩹, an easy-to-use package for editing the factual knowledge of large language models efficiently. Please follow [FastEdit](https://github.com/hiyouga/FastEdit) if you are interested.
+
+[23/06/29] We provided a **reproducible example** of training a chat model using instruction-following datasets, see [Baichuan-7B-sft](https://huggingface.co/hiyouga/Baichuan-7B-sft) for details.
+
+[23/06/22] We aligned the [demo API](src/api_demo.py) with the [OpenAI's](https://platform.openai.com/docs/api-reference/chat) format where you can insert the fine-tuned model in **arbitrary ChatGPT-based applications**.
+
+[23/06/03] We supported quantized training and inference (aka **[QLoRA](https://github.com/artidoro/qlora)**). See [examples](examples/README.md) for usage.
+
+</details>
+
+## Supported Models
+
+| Model                                                             | Model size                       | Template         |
+| ----------------------------------------------------------------- | -------------------------------- | ---------------- |
+| [Baichuan 2](https://huggingface.co/baichuan-inc)                 | 7B/13B                           | baichuan2        |
+| [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           |
+| [DeepSeek (Code/MoE)](https://huggingface.co/deepseek-ai)         | 7B/16B/67B/236B                  | deepseek         |
+| [Falcon](https://huggingface.co/tiiuae)                           | 7B/11B/40B/180B                  | falcon           |
+| [Gemma/Gemma 2/CodeGemma](https://huggingface.co/google)          | 2B/7B/9B/27B                     | gemma            |
+| [GLM-4](https://huggingface.co/THUDM)                             | 9B                               | glm4             |
+| [InternLM2/InternLM2.5](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 3-3.2](https://huggingface.co/meta-llama)                  | 1B/3B/8B/70B                     | llama3           |
+| [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 |
+| [MiniCPM](https://huggingface.co/openbmb)                         | 1B/2B/4B                         | cpm/cpm3         |
+| [Mistral/Mixtral](https://huggingface.co/mistralai)               | 7B/8x7B/8x22B                    | mistral          |
+| [OLMo](https://huggingface.co/allenai)                            | 1B/7B                            | -                |
+| [PaliGemma](https://huggingface.co/google)                        | 3B                               | paligemma        |
+| [Phi-1.5/Phi-2](https://huggingface.co/microsoft)                 | 1.3B/2.7B                        | -                |
+| [Phi-3](https://huggingface.co/microsoft)                         | 4B/7B/14B                        | phi              |
+| [Qwen (1-2.5) (Code/Math/MoE)](https://huggingface.co/Qwen)       | 0.5B/1.5B/3B/7B/14B/32B/72B/110B | qwen             |
+| [Qwen2-VL](https://huggingface.co/Qwen)                           | 2B/7B/72B                        | qwen2_vl         |
+| [StarCoder 2](https://huggingface.co/bigcode)                     | 3B/7B/15B                        | -                |
+| [XVERSE](https://huggingface.co/xverse)                           | 7B/13B/65B                       | xverse           |
+| [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 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.
+
+Please refer to [constants.py](src/llamafactory/extras/constants.py) for a full list of models we supported.
+
+You also can add a custom chat template to [template.py](src/llamafactory/data/template.py).
+
+## Supported Training Approaches
+
+| Approach               |     Full-tuning    |    Freeze-tuning   |       LoRA         |       QLoRA        |
+| ---------------------- | ------------------ | ------------------ | ------------------ | ------------------ |
+| Pre-Training           | :white_check_mark: | :white_check_mark: | :white_check_mark: | :white_check_mark: |
+| Supervised Fine-Tuning | :white_check_mark: | :white_check_mark: | :white_check_mark: | :white_check_mark: |
+| Reward Modeling        | :white_check_mark: | :white_check_mark: | :white_check_mark: | :white_check_mark: |
+| PPO Training           | :white_check_mark: | :white_check_mark: | :white_check_mark: | :white_check_mark: |
+| DPO Training           | :white_check_mark: | :white_check_mark: | :white_check_mark: | :white_check_mark: |
+| KTO Training           | :white_check_mark: | :white_check_mark: | :white_check_mark: | :white_check_mark: |
+| 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>
+
+- [Wiki Demo (en)](data/wiki_demo.txt)
+- [RefinedWeb (en)](https://huggingface.co/datasets/tiiuae/falcon-refinedweb)
+- [RedPajama V2 (en)](https://huggingface.co/datasets/togethercomputer/RedPajama-Data-V2)
+- [Wikipedia (en)](https://huggingface.co/datasets/olm/olm-wikipedia-20221220)
+- [Wikipedia (zh)](https://huggingface.co/datasets/pleisto/wikipedia-cn-20230720-filtered)
+- [Pile (en)](https://huggingface.co/datasets/EleutherAI/pile)
+- [SkyPile (zh)](https://huggingface.co/datasets/Skywork/SkyPile-150B)
+- [FineWeb (en)](https://huggingface.co/datasets/HuggingFaceFW/fineweb)
+- [FineWeb-Edu (en)](https://huggingface.co/datasets/HuggingFaceFW/fineweb-edu)
+- [The Stack (en)](https://huggingface.co/datasets/bigcode/the-stack)
+- [StarCoder (en)](https://huggingface.co/datasets/bigcode/starcoderdata)
+
+</details>
+
+<details><summary>Supervised fine-tuning datasets</summary>
+
+- [Identity (en&zh)](data/identity.json)
+- [Stanford Alpaca (en)](https://github.com/tatsu-lab/stanford_alpaca)
+- [Stanford Alpaca (zh)](https://github.com/ymcui/Chinese-LLaMA-Alpaca-3)
+- [Alpaca GPT4 (en&zh)](https://github.com/Instruction-Tuning-with-GPT-4/GPT-4-LLM)
+- [Glaive Function Calling V2 (en&zh)](https://huggingface.co/datasets/glaiveai/glaive-function-calling-v2)
+- [LIMA (en)](https://huggingface.co/datasets/GAIR/lima)
+- [Guanaco Dataset (multilingual)](https://huggingface.co/datasets/JosephusCheung/GuanacoDataset)
+- [BELLE 2M (zh)](https://huggingface.co/datasets/BelleGroup/train_2M_CN)
+- [BELLE 1M (zh)](https://huggingface.co/datasets/BelleGroup/train_1M_CN)
+- [BELLE 0.5M (zh)](https://huggingface.co/datasets/BelleGroup/train_0.5M_CN)
+- [BELLE Dialogue 0.4M (zh)](https://huggingface.co/datasets/BelleGroup/generated_chat_0.4M)
+- [BELLE School Math 0.25M (zh)](https://huggingface.co/datasets/BelleGroup/school_math_0.25M)
+- [BELLE Multiturn Chat 0.8M (zh)](https://huggingface.co/datasets/BelleGroup/multiturn_chat_0.8M)
+- [UltraChat (en)](https://github.com/thunlp/UltraChat)
+- [OpenPlatypus (en)](https://huggingface.co/datasets/garage-bAInd/Open-Platypus)
+- [CodeAlpaca 20k (en)](https://huggingface.co/datasets/sahil2801/CodeAlpaca-20k)
+- [Alpaca CoT (multilingual)](https://huggingface.co/datasets/QingyiSi/Alpaca-CoT)
+- [OpenOrca (en)](https://huggingface.co/datasets/Open-Orca/OpenOrca)
+- [SlimOrca (en)](https://huggingface.co/datasets/Open-Orca/SlimOrca)
+- [MathInstruct (en)](https://huggingface.co/datasets/TIGER-Lab/MathInstruct)
+- [Firefly 1.1M (zh)](https://huggingface.co/datasets/YeungNLP/firefly-train-1.1M)
+- [Wiki QA (en)](https://huggingface.co/datasets/wiki_qa)
+- [Web QA (zh)](https://huggingface.co/datasets/suolyer/webqa)
+- [WebNovel (zh)](https://huggingface.co/datasets/zxbsmk/webnovel_cn)
+- [Nectar (en)](https://huggingface.co/datasets/berkeley-nest/Nectar)
+- [deepctrl (en&zh)](https://www.modelscope.cn/datasets/deepctrl/deepctrl-sft-data)
+- [Advertise Generating (zh)](https://huggingface.co/datasets/HasturOfficial/adgen)
+- [ShareGPT Hyperfiltered (en)](https://huggingface.co/datasets/totally-not-an-llm/sharegpt-hyperfiltered-3k)
+- [ShareGPT4 (en&zh)](https://huggingface.co/datasets/shibing624/sharegpt_gpt4)
+- [UltraChat 200k (en)](https://huggingface.co/datasets/HuggingFaceH4/ultrachat_200k)
+- [AgentInstruct (en)](https://huggingface.co/datasets/THUDM/AgentInstruct)
+- [LMSYS Chat 1M (en)](https://huggingface.co/datasets/lmsys/lmsys-chat-1m)
+- [Evol Instruct V2 (en)](https://huggingface.co/datasets/WizardLM/WizardLM_evol_instruct_V2_196k)
+- [Cosmopedia (en)](https://huggingface.co/datasets/HuggingFaceTB/cosmopedia)
+- [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)
+- [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)
+- [OpenSchnabeltier (de)](https://huggingface.co/datasets/mayflowergmbh/openschnabeltier_de)
+- [Evol Instruct (de)](https://huggingface.co/datasets/mayflowergmbh/evol-instruct_de)
+- [Dolphin (de)](https://huggingface.co/datasets/mayflowergmbh/dolphin_de)
+- [Booksum (de)](https://huggingface.co/datasets/mayflowergmbh/booksum_de)
+- [Airoboros (de)](https://huggingface.co/datasets/mayflowergmbh/airoboros-3.0_de)
+- [Ultrachat (de)](https://huggingface.co/datasets/mayflowergmbh/ultra-chat_de)
+
+</details>
+
+<details><summary>Preference datasets</summary>
+
+- [DPO mixed (en&zh)](https://huggingface.co/datasets/hiyouga/DPO-En-Zh-20k)
+- [UltraFeedback (en)](https://huggingface.co/datasets/HuggingFaceH4/ultrafeedback_binarized)
+- [RLHF-V (en)](https://huggingface.co/datasets/openbmb/RLHF-V-Dataset)
+- [VLFeedback (en)](https://huggingface.co/datasets/Zhihui/VLFeedback)
+- [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)
+- [Orca DPO (de)](https://huggingface.co/datasets/mayflowergmbh/intel_orca_dpo_pairs_de)
+- [KTO mixed (en)](https://huggingface.co/datasets/argilla/kto-mix-15k)
+
+</details>
+
+Some datasets require confirmation before using them, so we recommend logging in with your Hugging Face account using these commands.
+
+```bash
+pip install --upgrade huggingface_hub
+huggingface-cli login
+```
+
+## Requirement
+
+| Mandatory    | Minimum | Recommend |
+| ------------ | ------- | --------- |
+| python       | 3.8     | 3.11      |
+| torch        | 1.13.1  | 2.4.0     |
+| transformers | 4.41.2  | 4.43.4    |
+| datasets     | 2.16.0  | 2.20.0    |
+| accelerate   | 0.30.1  | 0.32.0    |
+| peft         | 0.11.1  | 0.12.0    |
+| trl          | 0.8.6   | 0.9.6     |
+
+| Optional     | Minimum | Recommend |
+| ------------ | ------- | --------- |
+| CUDA         | 11.6    | 12.2      |
+| deepspeed    | 0.10.0  | 0.14.0    |
+| bitsandbytes | 0.39.0  | 0.43.1    |
+| vllm         | 0.4.3   | 0.5.0     |
+| flash-attn   | 2.3.0   | 2.6.3     |
+
+### 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 |
+
+## Getting Started
+
+### Installation
+
+> [!IMPORTANT]
+> Installation is mandatory.
+
+```bash
+git clone --depth 1 https://github.com/hiyouga/LLaMA-Factory.git
+cd LLaMA-Factory
+pip install -e ".[torch,metrics]"
+```
+
+Extra dependencies available: torch, torch-npu, metrics, deepspeed, liger-kernel, bitsandbytes, hqq, eetq, gptq, awq, aqlm, vllm, galore, badam, adam-mini, qwen, modelscope, quality
+
+> [!TIP]
+> Use `pip install --no-deps -e .` to resolve package conflicts.
+
+<details><summary>For Windows users</summary>
+
+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.
+
+</details>
+
+<details><summary>For Ascend NPU users</summary>
+
+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:
+
+```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
+
+# 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
+
+# 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 |
+| deepspeed    | 0.13.2  | 0.13.2      |
+
+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)
+
+</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.
+
+> [!NOTE]
+> Please update `data/dataset_info.json` to use your custom dataset.
+
+### Quickstart
+
+Use the following 3 commands to run LoRA **fine-tuning**, **inference** and **merging** of the Llama3-8B-Instruct model, respectively.
+
+```bash
+llamafactory-cli train examples/train_lora/llama3_lora_sft.yaml
+llamafactory-cli chat examples/inference/llama3_lora_sft.yaml
+llamafactory-cli export examples/merge_lora/llama3_lora_sft.yaml
+```
+
+See [examples/README.md](examples/README.md) for advanced usage (including distributed training).
+
+> [!TIP]
+> Use `llamafactory-cli help` to show help information.
+
+### Fine-Tuning with LLaMA Board GUI (powered by [Gradio](https://github.com/gradio-app/gradio))
+
+```bash
+llamafactory-cli webui
+```
+
+### Build Docker
+
+For CUDA users:
+
+```bash
+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 INSTALL_BNB=false \
+    --build-arg INSTALL_VLLM=false \
+    --build-arg INSTALL_DEEPSPEED=false \
+    --build-arg INSTALL_FLASHATTN=false \
+    --build-arg PIP_INDEX=https://pypi.org/simple \
+    -t llamafactory:latest .
+
+docker run -dit --gpus=all \
+    -v ./hf_cache:/root/.cache/huggingface \
+    -v ./ms_cache:/root/.cache/modelscope \
+    -v ./data:/app/data \
+    -v ./output:/app/output \
+    -p 7860:7860 \
+    -p 8000:8000 \
+    --shm-size 16G \
+    --name llamafactory \
+    llamafactory:latest
+
+docker exec -it llamafactory bash
+```
+
+For Ascend NPU users:
+
+```bash
+# Choose docker image upon your environment
+docker build -f ./docker/docker-npu/Dockerfile \
+    --build-arg INSTALL_DEEPSPEED=false \
+    --build-arg PIP_INDEX=https://pypi.org/simple \
+    -t llamafactory:latest .
+
+# Change `device` upon your resources
+docker run -dit \
+    -v ./hf_cache:/root/.cache/huggingface \
+    -v ./ms_cache:/root/.cache/modelscope \
+    -v ./data:/app/data \
+    -v ./output:/app/output \
+    -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 \
+    --shm-size 16G \
+    --name llamafactory \
+    llamafactory:latest
+
+docker exec -it llamafactory bash
+```
+
+For AMD ROCm users:
+
+```bash
+docker build -f ./docker/docker-rocm/Dockerfile \
+    --build-arg INSTALL_BNB=false \
+    --build-arg INSTALL_VLLM=false \
+    --build-arg INSTALL_DEEPSPEED=false \
+    --build-arg INSTALL_FLASHATTN=false \
+    --build-arg PIP_INDEX=https://pypi.org/simple \
+    -t llamafactory:latest .
+
+docker run -dit \
+    -v ./hf_cache:/root/.cache/huggingface \
+    -v ./ms_cache:/root/.cache/modelscope \
+    -v ./data:/app/data \
+    -v ./output:/app/output \
+    -v ./saves:/app/saves \
+    -p 7860:7860 \
+    -p 8000:8000 \
+    --device /dev/kfd \
+    --device /dev/dri \
+    --shm-size 16G \
+    --name llamafactory \
+    llamafactory:latest
+
+docker exec -it llamafactory bash
+```
+
+</details>
+
+<details><summary>Details about volume</summary>
+
+- `hf_cache`: Utilize Hugging Face cache on the host machine. Reassignable if a cache already exists in a different directory.
+- `ms_cache`: Similar to Hugging Face cache but for ModelScope users.
+- `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.
+
+</details>
+
+### Deploy with OpenAI-style API and vLLM
+
+```bash
+API_PORT=8000 llamafactory-cli api examples/inference/llama3_vllm.yaml
+```
+
+> [!TIP]
+> Visit [this page](https://platform.openai.com/docs/api-reference/chat/create) for API document.
+
+### Download from ModelScope Hub
+
+If you have trouble with downloading models and datasets from Hugging Face, you can use ModelScope.
+
+```bash
+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`.
+
+### 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.
+
+```yaml
+report_to: wandb
+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.
+
+## Projects using LLaMA Factory
+
+If you have a project that should be incorporated, please contact via email or create a pull request.
+
+<details><summary>Click to show</summary>
+
+1. Wang et al. ESRL: Efficient Sampling-based Reinforcement Learning for Sequence Generation. 2023. [[arxiv]](https://arxiv.org/abs/2308.02223)
+1. Yu et al. Open, Closed, or Small Language Models for Text Classification? 2023. [[arxiv]](https://arxiv.org/abs/2308.10092)
+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. 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. 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. 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. 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. 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. 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. **[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. **[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.
+
+</details>
+
+## License
+
+This repository is licensed under the [Apache-2.0 License](LICENSE).
+
+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) / [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/) / [MiniCPM](https://github.com/OpenBMB/MiniCPM/blob/main/MiniCPM%20Model%20License.md) / [Mistral](LICENSE) / [OLMo](LICENSE) / [Phi-1.5/Phi-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) / [StarCoder 2](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 2](https://github.com/IEIT-Yuan/Yuan-2.0/blob/main/LICENSE-Yuan)
+
+## Citation
+
+If this work is helpful, please kindly cite as:
+
+```bibtex
+@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 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}
+}
+```
+
+## Acknowledgement
+
+This repo benefits from [PEFT](https://github.com/huggingface/peft), [TRL](https://github.com/huggingface/trl), [QLoRA](https://github.com/artidoro/qlora) and [FastChat](https://github.com/lm-sys/FastChat). Thanks for their wonderful works.
+
+## Star History
+
+![Star History Chart](https://api.star-history.com/svg?repos=hiyouga/LLaMA-Factory&type=Date)

llamafactory.egg-info/SOURCES.txt

@@ -0,0 +1,123 @@
+LICENSE
+MANIFEST.in
+README.md
+pyproject.toml
+requirements.txt
+setup.py
+src/llamafactory/__init__.py
+src/llamafactory/cli.py
+src/llamafactory/launcher.py
+src/llamafactory.egg-info/PKG-INFO
+src/llamafactory.egg-info/SOURCES.txt
+src/llamafactory.egg-info/dependency_links.txt
+src/llamafactory.egg-info/entry_points.txt
+src/llamafactory.egg-info/requires.txt
+src/llamafactory.egg-info/top_level.txt
+src/llamafactory/api/__init__.py
+src/llamafactory/api/app.py
+src/llamafactory/api/chat.py
+src/llamafactory/api/common.py
+src/llamafactory/api/protocol.py
+src/llamafactory/chat/__init__.py
+src/llamafactory/chat/base_engine.py
+src/llamafactory/chat/chat_model.py
+src/llamafactory/chat/hf_engine.py
+src/llamafactory/chat/vllm_engine.py
+src/llamafactory/data/__init__.py
+src/llamafactory/data/aligner.py
+src/llamafactory/data/collator.py
+src/llamafactory/data/data_utils.py
+src/llamafactory/data/formatter.py
+src/llamafactory/data/loader.py
+src/llamafactory/data/mm_plugin.py
+src/llamafactory/data/parser.py
+src/llamafactory/data/preprocess.py
+src/llamafactory/data/template.py
+src/llamafactory/data/tool_utils.py
+src/llamafactory/data/processors/__init__.py
+src/llamafactory/data/processors/feedback.py
+src/llamafactory/data/processors/pairwise.py
+src/llamafactory/data/processors/pretrain.py
+src/llamafactory/data/processors/processor_utils.py
+src/llamafactory/data/processors/supervised.py
+src/llamafactory/data/processors/unsupervised.py
+src/llamafactory/eval/__init__.py
+src/llamafactory/eval/evaluator.py
+src/llamafactory/eval/template.py
+src/llamafactory/extras/__init__.py
+src/llamafactory/extras/constants.py
+src/llamafactory/extras/env.py
+src/llamafactory/extras/logging.py
+src/llamafactory/extras/misc.py
+src/llamafactory/extras/packages.py
+src/llamafactory/extras/ploting.py
+src/llamafactory/hparams/__init__.py
+src/llamafactory/hparams/data_args.py
+src/llamafactory/hparams/evaluation_args.py
+src/llamafactory/hparams/finetuning_args.py
+src/llamafactory/hparams/generating_args.py
+src/llamafactory/hparams/model_args.py
+src/llamafactory/hparams/parser.py
+src/llamafactory/model/__init__.py
+src/llamafactory/model/adapter.py
+src/llamafactory/model/loader.py
+src/llamafactory/model/patcher.py
+src/llamafactory/model/model_utils/__init__.py
+src/llamafactory/model/model_utils/attention.py
+src/llamafactory/model/model_utils/checkpointing.py
+src/llamafactory/model/model_utils/embedding.py
+src/llamafactory/model/model_utils/liger_kernel.py
+src/llamafactory/model/model_utils/longlora.py
+src/llamafactory/model/model_utils/misc.py
+src/llamafactory/model/model_utils/mod.py
+src/llamafactory/model/model_utils/moe.py
+src/llamafactory/model/model_utils/packing.py
+src/llamafactory/model/model_utils/quantization.py
+src/llamafactory/model/model_utils/rope.py
+src/llamafactory/model/model_utils/unsloth.py
+src/llamafactory/model/model_utils/valuehead.py
+src/llamafactory/model/model_utils/visual.py
+src/llamafactory/train/__init__.py
+src/llamafactory/train/callbacks.py
+src/llamafactory/train/test_utils.py
+src/llamafactory/train/trainer_utils.py
+src/llamafactory/train/tuner.py
+src/llamafactory/train/dpo/__init__.py
+src/llamafactory/train/dpo/trainer.py
+src/llamafactory/train/dpo/workflow.py
+src/llamafactory/train/kto/__init__.py
+src/llamafactory/train/kto/trainer.py
+src/llamafactory/train/kto/workflow.py
+src/llamafactory/train/ppo/__init__.py
+src/llamafactory/train/ppo/ppo_utils.py
+src/llamafactory/train/ppo/trainer.py
+src/llamafactory/train/ppo/workflow.py
+src/llamafactory/train/pt/__init__.py
+src/llamafactory/train/pt/trainer.py
+src/llamafactory/train/pt/workflow.py
+src/llamafactory/train/rm/__init__.py
+src/llamafactory/train/rm/metric.py
+src/llamafactory/train/rm/trainer.py
+src/llamafactory/train/rm/workflow.py
+src/llamafactory/train/sft/__init__.py
+src/llamafactory/train/sft/metric.py
+src/llamafactory/train/sft/trainer.py
+src/llamafactory/train/sft/workflow.py
+src/llamafactory/webui/__init__.py
+src/llamafactory/webui/chatter.py
+src/llamafactory/webui/common.py
+src/llamafactory/webui/css.py
+src/llamafactory/webui/engine.py
+src/llamafactory/webui/interface.py
+src/llamafactory/webui/locales.py
+src/llamafactory/webui/manager.py
+src/llamafactory/webui/runner.py
+src/llamafactory/webui/utils.py
+src/llamafactory/webui/components/__init__.py
+src/llamafactory/webui/components/chatbot.py
+src/llamafactory/webui/components/data.py
+src/llamafactory/webui/components/eval.py
+src/llamafactory/webui/components/export.py
+src/llamafactory/webui/components/infer.py
+src/llamafactory/webui/components/top.py
+src/llamafactory/webui/components/train.py

llamafactory.egg-info/dependency_links.txt

@@ -0,0 +1 @@
+

llamafactory.egg-info/entry_points.txt

@@ -0,0 +1,3 @@
+[console_scripts]
+llamafactory-cli = llamafactory.cli:main
+lmf = llamafactory.cli:main

llamafactory.egg-info/requires.txt

@@ -0,0 +1,82 @@
+transformers<=4.45.0,>=4.41.2
+datasets<=2.21.0,>=2.16.0
+accelerate<=0.34.2,>=0.30.1
+peft<=0.12.0,>=0.11.1
+trl<=0.9.6,>=0.8.6
+gradio>=4.0.0
+pandas>=2.0.0
+scipy
+einops
+sentencepiece
+tiktoken
+protobuf
+uvicorn
+pydantic
+fastapi
+sse-starlette
+matplotlib>=3.7.0
+fire
+packaging
+pyyaml
+numpy<2.0.0
+av
+
+[adam-mini]
+adam-mini
+
+[aqlm]
+aqlm[gpu]>=1.1.0
+
+[awq]
+autoawq
+
+[badam]
+badam>=1.2.1
+
+[bitsandbytes]
+bitsandbytes>=0.39.0
+
+[deepspeed]
+deepspeed<=0.14.4,>=0.10.0
+
+[dev]
+ruff
+pytest
+
+[eetq]
+eetq
+
+[galore]
+galore-torch
+
+[gptq]
+optimum>=1.17.0
+auto-gptq>=0.5.0
+
+[hqq]
+hqq
+
+[liger-kernel]
+liger-kernel
+
+[metrics]
+nltk
+jieba
+rouge-chinese
+
+[modelscope]
+modelscope
+
+[qwen]
+transformers_stream_generator
+
+[torch]
+torch>=1.13.1
+
+[torch-npu]
+torch==2.1.0
+torch-npu==2.1.0.post3
+decorator
+
+[vllm]
+vllm<=0.6.2,>=0.4.3

llamafactory.egg-info/top_level.txt

@@ -0,0 +1 @@
+llamafactory

llamafactory/__init__.py

@@ -0,0 +1,46 @@
+# Copyright 2024 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.
+
+r"""
+Efficient fine-tuning of large language models.
+
+Level:
+  api, webui > chat, eval, train > data, model > hparams > extras
+
+Dependency graph:
+  main:
+    transformers>=4.41.2,<=4.45.0
+    datasets>=2.16.0,<=2.21.0
+    accelerate>=0.30.1,<=0.34.2
+    peft>=0.11.1,<=0.12.0
+    trl>=0.8.6,<=0.9.6
+  attention:
+    transformers>=4.42.4 (gemma+fa2)
+  longlora:
+    transformers>=4.41.2,<=4.45.0
+  packing:
+    transformers>=4.41.2,<=4.45.0
+
+Disable version checking: DISABLE_VERSION_CHECK=1
+Enable VRAM recording: RECORD_VRAM=1
+Force check imports: FORCE_CHECK_IMPORTS=1
+Force using torchrun: FORCE_TORCHRUN=1
+Set logging verbosity: LLAMAFACTORY_VERBOSITY=WARN
+Use modelscope: USE_MODELSCOPE_HUB=1
+"""
+
+from .extras.env import VERSION
+
+
+__version__ = VERSION

llamafactory/api/app.py

@@ -0,0 +1,134 @@
+# Copyright 2024 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 asyncio
+import os
+from contextlib import asynccontextmanager
+from functools import partial
+from typing import Optional
+
+from typing_extensions import Annotated
+
+from ..chat import ChatModel
+from ..extras.misc import torch_gc
+from ..extras.packages import is_fastapi_available, is_starlette_available, is_uvicorn_available
+from .chat import (
+    create_chat_completion_response,
+    create_score_evaluation_response,
+    create_stream_chat_completion_response,
+)
+from .protocol import (
+    ChatCompletionRequest,
+    ChatCompletionResponse,
+    ModelCard,
+    ModelList,
+    ScoreEvaluationRequest,
+    ScoreEvaluationResponse,
+)
+
+
+if is_fastapi_available():
+    from fastapi import Depends, FastAPI, HTTPException, status
+    from fastapi.middleware.cors import CORSMiddleware
+    from fastapi.security.http import HTTPAuthorizationCredentials, HTTPBearer
+
+
+if is_starlette_available():
+    from sse_starlette import EventSourceResponse
+
+
+if is_uvicorn_available():
+    import uvicorn
+
+
+async def sweeper() -> None:
+    while True:
+        torch_gc()
+        await asyncio.sleep(300)
+
+
+@asynccontextmanager
+async def lifespan(app: "FastAPI", chat_model: "ChatModel"):  # collects GPU memory
+    if chat_model.engine_type == "huggingface":
+        asyncio.create_task(sweeper())
+
+    yield
+    torch_gc()
+
+
+def create_app(chat_model: "ChatModel") -> "FastAPI":
+    root_path = os.environ.get("FASTAPI_ROOT_PATH", "")
+    app = FastAPI(lifespan=partial(lifespan, chat_model=chat_model), root_path=root_path)
+    app.add_middleware(
+        CORSMiddleware,
+        allow_origins=["*"],
+        allow_credentials=True,
+        allow_methods=["*"],
+        allow_headers=["*"],
+    )
+    api_key = os.environ.get("API_KEY", None)
+    security = HTTPBearer(auto_error=False)
+
+    async def verify_api_key(auth: Annotated[Optional[HTTPAuthorizationCredentials], Depends(security)]):
+        if api_key and (auth is None or auth.credentials != api_key):
+            raise HTTPException(status_code=status.HTTP_401_UNAUTHORIZED, detail="Invalid API key.")
+
+    @app.get(
+        "/v1/models",
+        response_model=ModelList,
+        status_code=status.HTTP_200_OK,
+        dependencies=[Depends(verify_api_key)],
+    )
+    async def list_models():
+        model_card = ModelCard(id=os.environ.get("API_MODEL_NAME", "gpt-3.5-turbo"))
+        return ModelList(data=[model_card])
+
+    @app.post(
+        "/v1/chat/completions",
+        response_model=ChatCompletionResponse,
+        status_code=status.HTTP_200_OK,
+        dependencies=[Depends(verify_api_key)],
+    )
+    async def create_chat_completion(request: ChatCompletionRequest):
+        if not chat_model.engine.can_generate:
+            raise HTTPException(status_code=status.HTTP_405_METHOD_NOT_ALLOWED, detail="Not allowed")
+
+        if request.stream:
+            generate = create_stream_chat_completion_response(request, chat_model)
+            return EventSourceResponse(generate, media_type="text/event-stream")
+        else:
+            return await create_chat_completion_response(request, chat_model)
+
+    @app.post(
+        "/v1/score/evaluation",
+        response_model=ScoreEvaluationResponse,
+        status_code=status.HTTP_200_OK,
+        dependencies=[Depends(verify_api_key)],
+    )
+    async def create_score_evaluation(request: ScoreEvaluationRequest):
+        if chat_model.engine.can_generate:
+            raise HTTPException(status_code=status.HTTP_405_METHOD_NOT_ALLOWED, detail="Not allowed")
+
+        return await create_score_evaluation_response(request, chat_model)
+
+    return app
+
+
+def run_api() -> None:
+    chat_model = ChatModel()
+    app = create_app(chat_model)
+    api_host = os.environ.get("API_HOST", "0.0.0.0")
+    api_port = int(os.environ.get("API_PORT", "8000"))
+    print("Visit http://localhost:{}/docs for API document.".format(api_port))
+    uvicorn.run(app, host=api_host, port=api_port)

llamafactory/api/chat.py

@@ -0,0 +1,237 @@
+# Copyright 2024 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 base64
+import io
+import json
+import os
+import re
+import uuid
+from typing import TYPE_CHECKING, AsyncGenerator, Dict, List, Optional, Tuple
+
+from ..data import Role as DataRole
+from ..extras.logging import get_logger
+from ..extras.packages import is_fastapi_available, is_pillow_available, is_requests_available
+from .common import dictify, jsonify
+from .protocol import (
+    ChatCompletionMessage,
+    ChatCompletionResponse,
+    ChatCompletionResponseChoice,
+    ChatCompletionResponseUsage,
+    ChatCompletionStreamResponse,
+    ChatCompletionStreamResponseChoice,
+    Finish,
+    Function,
+    FunctionCall,
+    Role,
+    ScoreEvaluationResponse,
+)
+
+
+if is_fastapi_available():
+    from fastapi import HTTPException, status
+
+
+if is_pillow_available():
+    from PIL import Image
+
+
+if is_requests_available():
+    import requests
+
+
+if TYPE_CHECKING:
+    from ..chat import ChatModel
+    from ..data.mm_plugin import ImageInput
+    from .protocol import ChatCompletionRequest, ScoreEvaluationRequest
+
+
+logger = get_logger(__name__)
+ROLE_MAPPING = {
+    Role.USER: DataRole.USER.value,
+    Role.ASSISTANT: DataRole.ASSISTANT.value,
+    Role.SYSTEM: DataRole.SYSTEM.value,
+    Role.FUNCTION: DataRole.FUNCTION.value,
+    Role.TOOL: DataRole.OBSERVATION.value,
+}
+
+
+def _process_request(
+    request: "ChatCompletionRequest",
+) -> Tuple[List[Dict[str, str]], Optional[str], Optional[str], Optional["ImageInput"]]:
+    logger.info("==== request ====\n{}".format(json.dumps(dictify(request), indent=2, ensure_ascii=False)))
+
+    if len(request.messages) == 0:
+        raise HTTPException(status_code=status.HTTP_400_BAD_REQUEST, detail="Invalid length")
+
+    if request.messages[0].role == Role.SYSTEM:
+        system = request.messages.pop(0).content
+    else:
+        system = None
+
+    if len(request.messages) % 2 == 0:
+        raise HTTPException(status_code=status.HTTP_400_BAD_REQUEST, detail="Only supports u/a/u/a/u...")
+
+    input_messages = []
+    image = None
+    for i, message in enumerate(request.messages):
+        if i % 2 == 0 and message.role not in [Role.USER, Role.TOOL]:
+            raise HTTPException(status_code=status.HTTP_400_BAD_REQUEST, detail="Invalid role")
+        elif i % 2 == 1 and message.role not in [Role.ASSISTANT, Role.FUNCTION]:
+            raise HTTPException(status_code=status.HTTP_400_BAD_REQUEST, detail="Invalid role")
+
+        if message.role == Role.ASSISTANT and isinstance(message.tool_calls, list) and len(message.tool_calls):
+            tool_calls = [
+                {"name": tool_call.function.name, "arguments": tool_call.function.arguments}
+                for tool_call in message.tool_calls
+            ]
+            content = json.dumps(tool_calls, ensure_ascii=False)
+            input_messages.append({"role": ROLE_MAPPING[Role.FUNCTION], "content": content})
+        elif isinstance(message.content, list):
+            for input_item in message.content:
+                if input_item.type == "text":
+                    input_messages.append({"role": ROLE_MAPPING[message.role], "content": input_item.text})
+                else:
+                    image_url = input_item.image_url.url
+                    if re.match(r"^data:image\/(png|jpg|jpeg|gif|bmp);base64,(.+)$", image_url):  # base64 image
+                        image_stream = io.BytesIO(base64.b64decode(image_url.split(",", maxsplit=1)[1]))
+                    elif os.path.isfile(image_url):  # local file
+                        image_stream = open(image_url, "rb")
+                    else:  # web uri
+                        image_stream = requests.get(image_url, stream=True).raw
+
+                    image = Image.open(image_stream).convert("RGB")
+        else:
+            input_messages.append({"role": ROLE_MAPPING[message.role], "content": message.content})
+
+    tool_list = request.tools
+    if isinstance(tool_list, list) and len(tool_list):
+        try:
+            tools = json.dumps([dictify(tool.function) for tool in tool_list], ensure_ascii=False)
+        except json.JSONDecodeError:
+            raise HTTPException(status_code=status.HTTP_400_BAD_REQUEST, detail="Invalid tools")
+    else:
+        tools = None
+
+    return input_messages, system, tools, image
+
+
+def _create_stream_chat_completion_chunk(
+    completion_id: str,
+    model: str,
+    delta: "ChatCompletionMessage",
+    index: Optional[int] = 0,
+    finish_reason: Optional["Finish"] = None,
+) -> str:
+    choice_data = ChatCompletionStreamResponseChoice(index=index, delta=delta, finish_reason=finish_reason)
+    chunk = ChatCompletionStreamResponse(id=completion_id, model=model, choices=[choice_data])
+    return jsonify(chunk)
+
+
+async def create_chat_completion_response(
+    request: "ChatCompletionRequest", chat_model: "ChatModel"
+) -> "ChatCompletionResponse":
+    completion_id = "chatcmpl-{}".format(uuid.uuid4().hex)
+    input_messages, system, tools, image = _process_request(request)
+    responses = await chat_model.achat(
+        input_messages,
+        system,
+        tools,
+        image,
+        do_sample=request.do_sample,
+        temperature=request.temperature,
+        top_p=request.top_p,
+        max_new_tokens=request.max_tokens,
+        num_return_sequences=request.n,
+        stop=request.stop,
+    )
+
+    prompt_length, response_length = 0, 0
+    choices = []
+    for i, response in enumerate(responses):
+        if tools:
+            result = chat_model.engine.template.extract_tool(response.response_text)
+        else:
+            result = response.response_text
+
+        if isinstance(result, list):
+            tool_calls = []
+            for tool in result:
+                function = Function(name=tool[0], arguments=tool[1])
+                tool_calls.append(FunctionCall(id="call_{}".format(uuid.uuid4().hex), function=function))
+
+            response_message = ChatCompletionMessage(role=Role.ASSISTANT, tool_calls=tool_calls)
+            finish_reason = Finish.TOOL
+        else:
+            response_message = ChatCompletionMessage(role=Role.ASSISTANT, content=result)
+            finish_reason = Finish.STOP if response.finish_reason == "stop" else Finish.LENGTH
+
+        choices.append(ChatCompletionResponseChoice(index=i, message=response_message, finish_reason=finish_reason))
+        prompt_length = response.prompt_length
+        response_length += response.response_length
+
+    usage = ChatCompletionResponseUsage(
+        prompt_tokens=prompt_length,
+        completion_tokens=response_length,
+        total_tokens=prompt_length + response_length,
+    )
+
+    return ChatCompletionResponse(id=completion_id, model=request.model, choices=choices, usage=usage)
+
+
+async def create_stream_chat_completion_response(
+    request: "ChatCompletionRequest", chat_model: "ChatModel"
+) -> AsyncGenerator[str, None]:
+    completion_id = "chatcmpl-{}".format(uuid.uuid4().hex)
+    input_messages, system, tools, image = _process_request(request)
+    if tools:
+        raise HTTPException(status_code=status.HTTP_400_BAD_REQUEST, detail="Cannot stream function calls.")
+
+    if request.n > 1:
+        raise HTTPException(status_code=status.HTTP_400_BAD_REQUEST, detail="Cannot stream multiple responses.")
+
+    yield _create_stream_chat_completion_chunk(
+        completion_id=completion_id, model=request.model, delta=ChatCompletionMessage(role=Role.ASSISTANT, content="")
+    )
+    async for new_token in chat_model.astream_chat(
+        input_messages,
+        system,
+        tools,
+        image,
+        do_sample=request.do_sample,
+        temperature=request.temperature,
+        top_p=request.top_p,
+        max_new_tokens=request.max_tokens,
+        stop=request.stop,
+    ):
+        if len(new_token) != 0:
+            yield _create_stream_chat_completion_chunk(
+                completion_id=completion_id, model=request.model, delta=ChatCompletionMessage(content=new_token)
+            )
+
+    yield _create_stream_chat_completion_chunk(
+        completion_id=completion_id, model=request.model, delta=ChatCompletionMessage(), finish_reason=Finish.STOP
+    )
+    yield "[DONE]"
+
+
+async def create_score_evaluation_response(
+    request: "ScoreEvaluationRequest", chat_model: "ChatModel"
+) -> "ScoreEvaluationResponse":
+    score_id = "scoreval-{}".format(uuid.uuid4().hex)
+    if len(request.messages) == 0:
+        raise HTTPException(status_code=status.HTTP_400_BAD_REQUEST, detail="Invalid request")
+
+    scores = await chat_model.aget_scores(request.messages, max_length=request.max_length)
+    return ScoreEvaluationResponse(id=score_id, model=request.model, scores=scores)

llamafactory/api/common.py

@@ -0,0 +1,34 @@
+# Copyright 2024 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 json
+from typing import TYPE_CHECKING, Any, Dict
+
+
+if TYPE_CHECKING:
+    from pydantic import BaseModel
+
+
+def dictify(data: "BaseModel") -> Dict[str, Any]:
+    try:  # pydantic v2
+        return data.model_dump(exclude_unset=True)
+    except AttributeError:  # pydantic v1
+        return data.dict(exclude_unset=True)
+
+
+def jsonify(data: "BaseModel") -> str:
+    try:  # pydantic v2
+        return json.dumps(data.model_dump(exclude_unset=True), ensure_ascii=False)
+    except AttributeError:  # pydantic v1
+        return data.json(exclude_unset=True, ensure_ascii=False)

llamafactory/api/protocol.py

@@ -0,0 +1,153 @@
+# Copyright 2024 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 time
+from enum import Enum, unique
+from typing import Any, Dict, List, Optional, Union
+
+from pydantic import BaseModel, Field
+from typing_extensions import Literal
+
+
+@unique
+class Role(str, Enum):
+    USER = "user"
+    ASSISTANT = "assistant"
+    SYSTEM = "system"
+    FUNCTION = "function"
+    TOOL = "tool"
+
+
+@unique
+class Finish(str, Enum):
+    STOP = "stop"
+    LENGTH = "length"
+    TOOL = "tool_calls"
+
+
+class ModelCard(BaseModel):
+    id: str
+    object: Literal["model"] = "model"
+    created: int = Field(default_factory=lambda: int(time.time()))
+    owned_by: Literal["owner"] = "owner"
+
+
+class ModelList(BaseModel):
+    object: Literal["list"] = "list"
+    data: List[ModelCard] = []
+
+
+class Function(BaseModel):
+    name: str
+    arguments: str
+
+
+class FunctionDefinition(BaseModel):
+    name: str
+    description: str
+    parameters: Dict[str, Any]
+
+
+class FunctionAvailable(BaseModel):
+    type: Literal["function", "code_interpreter"] = "function"
+    function: Optional[FunctionDefinition] = None
+
+
+class FunctionCall(BaseModel):
+    id: str
+    type: Literal["function"] = "function"
+    function: Function
+
+
+class ImageURL(BaseModel):
+    url: str
+
+
+class MultimodalInputItem(BaseModel):
+    type: Literal["text", "image_url"]
+    text: Optional[str] = None
+    image_url: Optional[ImageURL] = None
+
+
+class ChatMessage(BaseModel):
+    role: Role
+    content: Optional[Union[str, List[MultimodalInputItem]]] = None
+    tool_calls: Optional[List[FunctionCall]] = None
+
+
+class ChatCompletionMessage(BaseModel):
+    role: Optional[Role] = None
+    content: Optional[str] = None
+    tool_calls: Optional[List[FunctionCall]] = None
+
+
+class ChatCompletionRequest(BaseModel):
+    model: str
+    messages: List[ChatMessage]
+    tools: Optional[List[FunctionAvailable]] = None
+    do_sample: Optional[bool] = None
+    temperature: Optional[float] = None
+    top_p: Optional[float] = None
+    n: int = 1
+    max_tokens: Optional[int] = None
+    stop: Optional[Union[str, List[str]]] = None
+    stream: bool = False
+
+
+class ChatCompletionResponseChoice(BaseModel):
+    index: int
+    message: ChatCompletionMessage
+    finish_reason: Finish
+
+
+class ChatCompletionStreamResponseChoice(BaseModel):
+    index: int
+    delta: ChatCompletionMessage
+    finish_reason: Optional[Finish] = None
+
+
+class ChatCompletionResponseUsage(BaseModel):
+    prompt_tokens: int
+    completion_tokens: int
+    total_tokens: int
+
+
+class ChatCompletionResponse(BaseModel):
+    id: str
+    object: Literal["chat.completion"] = "chat.completion"
+    created: int = Field(default_factory=lambda: int(time.time()))
+    model: str
+    choices: List[ChatCompletionResponseChoice]
+    usage: ChatCompletionResponseUsage
+
+
+class ChatCompletionStreamResponse(BaseModel):
+    id: str
+    object: Literal["chat.completion.chunk"] = "chat.completion.chunk"
+    created: int = Field(default_factory=lambda: int(time.time()))
+    model: str
+    choices: List[ChatCompletionStreamResponseChoice]
+
+
+class ScoreEvaluationRequest(BaseModel):
+    model: str
+    messages: List[str]
+    max_length: Optional[int] = None
+
+
+class ScoreEvaluationResponse(BaseModel):
+    id: str
+    object: Literal["score.evaluation"] = "score.evaluation"
+    model: str
+    scores: List[float]

llamafactory/chat/__init__.py

@@ -0,0 +1,19 @@
+# Copyright 2024 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 .base_engine import BaseEngine
+from .chat_model import ChatModel
+
+
+__all__ = ["BaseEngine", "ChatModel"]

llamafactory/chat/base_engine.py

@@ -0,0 +1,102 @@
+# Copyright 2024 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 abc import ABC, abstractmethod
+from dataclasses import dataclass
+from typing import TYPE_CHECKING, Any, AsyncGenerator, Dict, List, Literal, Optional, Sequence, Union
+
+
+if TYPE_CHECKING:
+    from transformers import PreTrainedModel, PreTrainedTokenizer
+    from vllm import AsyncLLMEngine
+
+    from ..data import Template
+    from ..data.mm_plugin import ImageInput, VideoInput
+    from ..hparams import DataArguments, FinetuningArguments, GeneratingArguments, ModelArguments
+
+
+@dataclass
+class Response:
+    response_text: str
+    response_length: int
+    prompt_length: int
+    finish_reason: Literal["stop", "length"]
+
+
+class BaseEngine(ABC):
+    r"""
+    Base class for inference engine of chat models.
+
+    Must implements async methods: chat(), stream_chat() and get_scores().
+    """
+
+    model: Union["PreTrainedModel", "AsyncLLMEngine"]
+    tokenizer: "PreTrainedTokenizer"
+    can_generate: bool
+    template: "Template"
+    generating_args: Dict[str, Any]
+
+    @abstractmethod
+    def __init__(
+        self,
+        model_args: "ModelArguments",
+        data_args: "DataArguments",
+        finetuning_args: "FinetuningArguments",
+        generating_args: "GeneratingArguments",
+    ) -> None:
+        r"""
+        Initializes an inference engine.
+        """
+        ...
+
+    @abstractmethod
+    async def chat(
+        self,
+        messages: Sequence[Dict[str, str]],
+        system: Optional[str] = None,
+        tools: Optional[str] = None,
+        image: Optional["ImageInput"] = None,
+        video: Optional["VideoInput"] = None,
+        **input_kwargs,
+    ) -> List["Response"]:
+        r"""
+        Gets a list of responses of the chat model.
+        """
+        ...
+
+    @abstractmethod
+    async def stream_chat(
+        self,
+        messages: Sequence[Dict[str, str]],
+        system: Optional[str] = None,
+        tools: Optional[str] = None,
+        image: Optional["ImageInput"] = None,
+        video: Optional["VideoInput"] = None,
+        **input_kwargs,
+    ) -> AsyncGenerator[str, None]:
+        r"""
+        Gets the response token-by-token of the chat model.
+        """
+        ...
+
+    @abstractmethod
+    async def get_scores(
+        self,
+        batch_input: List[str],
+        **input_kwargs,
+    ) -> List[float]:
+        r"""
+        Gets a list of scores of the reward model.
+        """
+        ...

llamafactory/chat/chat_model.py

@@ -0,0 +1,187 @@
+# Copyright 2024 THUDM and the LlamaFactory team.
+#
+# This code is inspired by the THUDM's ChatGLM implementation.
+# https://github.com/THUDM/ChatGLM-6B/blob/main/cli_demo.py
+#
+# 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 asyncio
+import os
+from threading import Thread
+from typing import TYPE_CHECKING, Any, AsyncGenerator, Dict, Generator, List, Optional, Sequence
+
+from ..extras.misc import torch_gc
+from ..hparams import get_infer_args
+from .hf_engine import HuggingfaceEngine
+from .vllm_engine import VllmEngine
+
+
+if TYPE_CHECKING:
+    from ..data.mm_plugin import ImageInput, VideoInput
+    from .base_engine import BaseEngine, Response
+
+
+def _start_background_loop(loop: "asyncio.AbstractEventLoop") -> None:
+    asyncio.set_event_loop(loop)
+    loop.run_forever()
+
+
+class ChatModel:
+    r"""
+    General class for chat models. Backed by huggingface or vllm engines.
+
+    Supports both sync and async methods.
+    Sync methods: chat(), stream_chat() and get_scores().
+    Async methods: achat(), astream_chat() and aget_scores().
+    """
+
+    def __init__(self, args: Optional[Dict[str, Any]] = None) -> None:
+        model_args, data_args, finetuning_args, generating_args = get_infer_args(args)
+        self.engine_type = model_args.infer_backend
+        if model_args.infer_backend == "huggingface":
+            self.engine: "BaseEngine" = HuggingfaceEngine(model_args, data_args, finetuning_args, generating_args)
+        elif model_args.infer_backend == "vllm":
+            self.engine: "BaseEngine" = VllmEngine(model_args, data_args, finetuning_args, generating_args)
+        else:
+            raise NotImplementedError("Unknown backend: {}".format(model_args.infer_backend))
+
+        self._loop = asyncio.new_event_loop()
+        self._thread = Thread(target=_start_background_loop, args=(self._loop,), daemon=True)
+        self._thread.start()
+
+    def chat(
+        self,
+        messages: Sequence[Dict[str, str]],
+        system: Optional[str] = None,
+        tools: Optional[str] = None,
+        image: Optional["ImageInput"] = None,
+        video: Optional["VideoInput"] = None,
+        **input_kwargs,
+    ) -> List["Response"]:
+        r"""
+        Gets a list of responses of the chat model.
+        """
+        task = asyncio.run_coroutine_threadsafe(
+            self.achat(messages, system, tools, image, video, **input_kwargs), self._loop
+        )
+        return task.result()
+
+    async def achat(
+        self,
+        messages: Sequence[Dict[str, str]],
+        system: Optional[str] = None,
+        tools: Optional[str] = None,
+        image: Optional["ImageInput"] = None,
+        video: Optional["VideoInput"] = None,
+        **input_kwargs,
+    ) -> List["Response"]:
+        r"""
+        Asynchronously gets a list of responses of the chat model.
+        """
+        return await self.engine.chat(messages, system, tools, image, video, **input_kwargs)
+
+    def stream_chat(
+        self,
+        messages: Sequence[Dict[str, str]],
+        system: Optional[str] = None,
+        tools: Optional[str] = None,
+        image: Optional["ImageInput"] = None,
+        video: Optional["VideoInput"] = None,
+        **input_kwargs,
+    ) -> Generator[str, None, None]:
+        r"""
+        Gets the response token-by-token of the chat model.
+        """
+        generator = self.astream_chat(messages, system, tools, image, video, **input_kwargs)
+        while True:
+            try:
+                task = asyncio.run_coroutine_threadsafe(generator.__anext__(), self._loop)
+                yield task.result()
+            except StopAsyncIteration:
+                break
+
+    async def astream_chat(
+        self,
+        messages: Sequence[Dict[str, str]],
+        system: Optional[str] = None,
+        tools: Optional[str] = None,
+        image: Optional["ImageInput"] = None,
+        video: Optional["VideoInput"] = None,
+        **input_kwargs,
+    ) -> AsyncGenerator[str, None]:
+        r"""
+        Asynchronously gets the response token-by-token of the chat model.
+        """
+        async for new_token in self.engine.stream_chat(messages, system, tools, image, video, **input_kwargs):
+            yield new_token
+
+    def get_scores(
+        self,
+        batch_input: List[str],
+        **input_kwargs,
+    ) -> List[float]:
+        r"""
+        Gets a list of scores of the reward model.
+        """
+        task = asyncio.run_coroutine_threadsafe(self.aget_scores(batch_input, **input_kwargs), self._loop)
+        return task.result()
+
+    async def aget_scores(
+        self,
+        batch_input: List[str],
+        **input_kwargs,
+    ) -> List[float]:
+        r"""
+        Asynchronously gets a list of scores of the reward model.
+        """
+        return await self.engine.get_scores(batch_input, **input_kwargs)
+
+
+def run_chat() -> None:
+    if os.name != "nt":
+        try:
+            import readline  # noqa: F401
+        except ImportError:
+            print("Install `readline` for a better experience.")
+
+    chat_model = ChatModel()
+    messages = []
+    print("Welcome to the CLI application, use `clear` to remove the history, use `exit` to exit the application.")
+
+    while True:
+        try:
+            query = input("\nUser: ")
+        except UnicodeDecodeError:
+            print("Detected decoding error at the inputs, please set the terminal encoding to utf-8.")
+            continue
+        except Exception:
+            raise
+
+        if query.strip() == "exit":
+            break
+
+        if query.strip() == "clear":
+            messages = []
+            torch_gc()
+            print("History has been removed.")
+            continue
+
+        messages.append({"role": "user", "content": query})
+        print("Assistant: ", end="", flush=True)
+
+        response = ""
+        for new_text in chat_model.stream_chat(messages):
+            print(new_text, end="", flush=True)
+            response += new_text
+        print()
+        messages.append({"role": "assistant", "content": response})

llamafactory/chat/hf_engine.py

@@ -0,0 +1,343 @@
+# Copyright 2024 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 asyncio
+import concurrent.futures
+import os
+from threading import Thread
+from typing import TYPE_CHECKING, Any, AsyncGenerator, Callable, Dict, List, Optional, Sequence, Tuple, Union
+
+import torch
+from transformers import GenerationConfig, TextIteratorStreamer
+from typing_extensions import override
+
+from ..data import get_template_and_fix_tokenizer
+from ..extras.constants import IMAGE_PLACEHOLDER, VIDEO_PLACEHOLDER
+from ..extras.logging import get_logger
+from ..extras.misc import get_logits_processor
+from ..model import load_model, load_tokenizer
+from .base_engine import BaseEngine, Response
+
+
+if TYPE_CHECKING:
+    from transformers import PreTrainedModel, PreTrainedTokenizer, ProcessorMixin
+    from trl import PreTrainedModelWrapper
+
+    from ..data import Template
+    from ..data.mm_plugin import ImageInput, VideoInput
+    from ..hparams import DataArguments, FinetuningArguments, GeneratingArguments, ModelArguments
+
+
+logger = get_logger(__name__)
+
+
+class HuggingfaceEngine(BaseEngine):
+    def __init__(
+        self,
+        model_args: "ModelArguments",
+        data_args: "DataArguments",
+        finetuning_args: "FinetuningArguments",
+        generating_args: "GeneratingArguments",
+    ) -> None:
+        self.can_generate = finetuning_args.stage == "sft"
+        tokenizer_module = load_tokenizer(model_args)
+        self.tokenizer = tokenizer_module["tokenizer"]
+        self.processor = tokenizer_module["processor"]
+        self.tokenizer.padding_side = "left" if self.can_generate else "right"
+        self.template = get_template_and_fix_tokenizer(self.tokenizer, data_args)
+        self.model = load_model(
+            self.tokenizer, model_args, finetuning_args, is_trainable=False, add_valuehead=(not self.can_generate)
+        )  # must after fixing tokenizer to resize vocab
+        self.generating_args = generating_args.to_dict()
+        try:
+            asyncio.get_event_loop()
+        except RuntimeError:
+            logger.warning("There is no current event loop, creating a new one.")
+            loop = asyncio.new_event_loop()
+            asyncio.set_event_loop(loop)
+
+        self.semaphore = asyncio.Semaphore(int(os.environ.get("MAX_CONCURRENT", "1")))
+
+    @staticmethod
+    def _process_args(
+        model: "PreTrainedModel",
+        tokenizer: "PreTrainedTokenizer",
+        processor: Optional["ProcessorMixin"],
+        template: "Template",
+        generating_args: Dict[str, Any],
+        messages: Sequence[Dict[str, str]],
+        system: Optional[str] = None,
+        tools: Optional[str] = None,
+        image: Optional["ImageInput"] = None,
+        video: Optional["VideoInput"] = None,
+        input_kwargs: Optional[Dict[str, Any]] = {},
+    ) -> Tuple[Dict[str, Any], int]:
+        mm_input_dict = {"images": [], "videos": [], "imglens": [0], "vidlens": [0]}
+        if image is not None:
+            mm_input_dict.update({"images": [image], "imglens": [1]})
+            if IMAGE_PLACEHOLDER not in messages[0]["content"]:
+                messages[0]["content"] = IMAGE_PLACEHOLDER + messages[0]["content"]
+
+        if video is not None:
+            mm_input_dict.update({"videos": [video], "vidlens": [1]})
+            if VIDEO_PLACEHOLDER not in messages[0]["content"]:
+                messages[0]["content"] = VIDEO_PLACEHOLDER + messages[0]["content"]
+
+        messages = template.mm_plugin.process_messages(
+            messages, mm_input_dict["images"], mm_input_dict["videos"], processor
+        )
+        paired_messages = messages + [{"role": "assistant", "content": ""}]
+        system = system or generating_args["default_system"]
+        prompt_ids, _ = template.encode_oneturn(tokenizer, paired_messages, system, tools)
+        prompt_ids, _ = template.mm_plugin.process_token_ids(
+            prompt_ids, None, mm_input_dict["images"], mm_input_dict["videos"], tokenizer, processor
+        )
+        prompt_length = len(prompt_ids)
+        inputs = torch.tensor([prompt_ids], device=model.device)
+        attention_mask = torch.ones_like(inputs, dtype=torch.bool)
+
+        do_sample: Optional[bool] = input_kwargs.pop("do_sample", None)
+        temperature: Optional[float] = input_kwargs.pop("temperature", None)
+        top_p: Optional[float] = input_kwargs.pop("top_p", None)
+        top_k: Optional[float] = input_kwargs.pop("top_k", None)
+        num_return_sequences: int = input_kwargs.pop("num_return_sequences", 1)
+        repetition_penalty: Optional[float] = input_kwargs.pop("repetition_penalty", None)
+        length_penalty: Optional[float] = input_kwargs.pop("length_penalty", None)
+        max_length: Optional[int] = input_kwargs.pop("max_length", None)
+        max_new_tokens: Optional[int] = input_kwargs.pop("max_new_tokens", None)
+        stop: Optional[Union[str, List[str]]] = input_kwargs.pop("stop", None)
+
+        if stop is not None:
+            logger.warning("Stop parameter is not supported by the huggingface engine yet.")
+
+        generating_args = generating_args.copy()
+        generating_args.update(
+            dict(
+                do_sample=do_sample if do_sample is not None else generating_args["do_sample"],
+                temperature=temperature if temperature is not None else generating_args["temperature"],
+                top_p=top_p if top_p is not None else generating_args["top_p"],
+                top_k=top_k if top_k is not None else generating_args["top_k"],
+                num_return_sequences=num_return_sequences,
+                repetition_penalty=repetition_penalty
+                if repetition_penalty is not None
+                else generating_args["repetition_penalty"],
+                length_penalty=length_penalty if length_penalty is not None else generating_args["length_penalty"],
+                eos_token_id=[tokenizer.eos_token_id] + tokenizer.additional_special_tokens_ids,
+                pad_token_id=tokenizer.pad_token_id,
+            )
+        )
+
+        if isinstance(num_return_sequences, int) and num_return_sequences > 1:  # do_sample needs temperature > 0
+            generating_args["do_sample"] = True
+            generating_args["temperature"] = generating_args["temperature"] or 1.0
+
+        if not generating_args["temperature"]:
+            generating_args["do_sample"] = False
+
+        if not generating_args["do_sample"]:
+            generating_args.pop("temperature", None)
+            generating_args.pop("top_p", None)
+
+        if max_length:
+            generating_args.pop("max_new_tokens", None)
+            generating_args["max_length"] = max_length
+
+        if max_new_tokens:
+            generating_args.pop("max_length", None)
+            generating_args["max_new_tokens"] = max_new_tokens
+
+        gen_kwargs = dict(
+            inputs=inputs,
+            attention_mask=attention_mask,
+            generation_config=GenerationConfig(**generating_args),
+            logits_processor=get_logits_processor(),
+        )
+
+        mm_inputs = template.mm_plugin.get_mm_inputs(**mm_input_dict, seqlens=[prompt_length], processor=processor)
+        for key, value in mm_inputs.items():
+            value = value if isinstance(value, torch.Tensor) else torch.tensor(value)
+            gen_kwargs[key] = value.to(model.device)
+
+        return gen_kwargs, prompt_length
+
+    @staticmethod
+    @torch.inference_mode()
+    def _chat(
+        model: "PreTrainedModel",
+        tokenizer: "PreTrainedTokenizer",
+        processor: Optional["ProcessorMixin"],
+        template: "Template",
+        generating_args: Dict[str, Any],
+        messages: Sequence[Dict[str, str]],
+        system: Optional[str] = None,
+        tools: Optional[str] = None,
+        image: Optional["ImageInput"] = None,
+        video: Optional["VideoInput"] = None,
+        input_kwargs: Optional[Dict[str, Any]] = {},
+    ) -> List["Response"]:
+        gen_kwargs, prompt_length = HuggingfaceEngine._process_args(
+            model, tokenizer, processor, template, generating_args, messages, system, tools, image, video, input_kwargs
+        )
+        generate_output = model.generate(**gen_kwargs)
+        response_ids = generate_output[:, prompt_length:]
+        response = tokenizer.batch_decode(response_ids, skip_special_tokens=True, clean_up_tokenization_spaces=True)
+        results = []
+        for i in range(len(response)):
+            eos_index = (response_ids[i] == tokenizer.eos_token_id).nonzero()
+            response_length = (eos_index[0].item() + 1) if len(eos_index) else len(response_ids[i])
+            results.append(
+                Response(
+                    response_text=response[i],
+                    response_length=response_length,
+                    prompt_length=prompt_length,
+                    finish_reason="stop" if len(eos_index) else "length",
+                )
+            )
+
+        return results
+
+    @staticmethod
+    @torch.inference_mode()
+    def _stream_chat(
+        model: "PreTrainedModel",
+        tokenizer: "PreTrainedTokenizer",
+        processor: Optional["ProcessorMixin"],
+        template: "Template",
+        generating_args: Dict[str, Any],
+        messages: Sequence[Dict[str, str]],
+        system: Optional[str] = None,
+        tools: Optional[str] = None,
+        image: Optional["ImageInput"] = None,
+        video: Optional["VideoInput"] = None,
+        input_kwargs: Optional[Dict[str, Any]] = {},
+    ) -> Callable[[], str]:
+        gen_kwargs, _ = HuggingfaceEngine._process_args(
+            model, tokenizer, processor, template, generating_args, messages, system, tools, image, video, input_kwargs
+        )
+        streamer = TextIteratorStreamer(tokenizer, skip_prompt=True, skip_special_tokens=True)
+        gen_kwargs["streamer"] = streamer
+        thread = Thread(target=model.generate, kwargs=gen_kwargs, daemon=True)
+        thread.start()
+
+        def stream():
+            try:
+                return streamer.__next__()
+            except StopIteration:
+                raise StopAsyncIteration()
+
+        return stream
+
+    @staticmethod
+    @torch.inference_mode()
+    def _get_scores(
+        model: "PreTrainedModelWrapper",
+        tokenizer: "PreTrainedTokenizer",
+        batch_input: List[str],
+        input_kwargs: Optional[Dict[str, Any]] = {},
+    ) -> List[float]:
+        max_length: Optional[int] = input_kwargs.pop("max_length", None)
+        device = getattr(model.pretrained_model, "device", "cuda")
+        inputs: Dict[str, "torch.Tensor"] = tokenizer(
+            batch_input,
+            padding=True,
+            truncation=True,
+            max_length=max_length or getattr(model.config, "max_position_embeddings", 1024),
+            return_tensors="pt",
+            add_special_tokens=False,
+        ).to(device)
+        values: "torch.Tensor" = model(**inputs, return_dict=True, use_cache=False)[-1]
+        scores = values.gather(dim=-1, index=(inputs["attention_mask"].sum(dim=-1, keepdim=True) - 1))
+        return scores
+
+    @override
+    async def chat(
+        self,
+        messages: Sequence[Dict[str, str]],
+        system: Optional[str] = None,
+        tools: Optional[str] = None,
+        image: Optional["ImageInput"] = None,
+        video: Optional["VideoInput"] = None,
+        **input_kwargs,
+    ) -> List["Response"]:
+        if not self.can_generate:
+            raise ValueError("The current model does not support `chat`.")
+
+        loop = asyncio.get_running_loop()
+        input_args = (
+            self.model,
+            self.tokenizer,
+            self.processor,
+            self.template,
+            self.generating_args,
+            messages,
+            system,
+            tools,
+            image,
+            video,
+            input_kwargs,
+        )
+        async with self.semaphore:
+            with concurrent.futures.ThreadPoolExecutor() as pool:
+                return await loop.run_in_executor(pool, self._chat, *input_args)
+
+    @override
+    async def stream_chat(
+        self,
+        messages: Sequence[Dict[str, str]],
+        system: Optional[str] = None,
+        tools: Optional[str] = None,
+        image: Optional["ImageInput"] = None,
+        video: Optional["VideoInput"] = None,
+        **input_kwargs,
+    ) -> AsyncGenerator[str, None]:
+        if not self.can_generate:
+            raise ValueError("The current model does not support `stream_chat`.")
+
+        loop = asyncio.get_running_loop()
+        input_args = (
+            self.model,
+            self.tokenizer,
+            self.processor,
+            self.template,
+            self.generating_args,
+            messages,
+            system,
+            tools,
+            image,
+            video,
+            input_kwargs,
+        )
+        async with self.semaphore:
+            with concurrent.futures.ThreadPoolExecutor() as pool:
+                stream = self._stream_chat(*input_args)
+                while True:
+                    try:
+                        yield await loop.run_in_executor(pool, stream)
+                    except StopAsyncIteration:
+                        break
+
+    @override
+    async def get_scores(
+        self,
+        batch_input: List[str],
+        **input_kwargs,
+    ) -> List[float]:
+        if self.can_generate:
+            raise ValueError("Cannot get scores using an auto-regressive model.")
+
+        loop = asyncio.get_running_loop()
+        input_args = (self.model, self.tokenizer, batch_input, input_kwargs)
+        async with self.semaphore:
+            with concurrent.futures.ThreadPoolExecutor() as pool:
+                return await loop.run_in_executor(pool, self._get_scores, *input_args)

llamafactory/chat/vllm_engine.py

@@ -0,0 +1,230 @@
+# Copyright 2024 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 uuid
+from typing import TYPE_CHECKING, Any, AsyncGenerator, AsyncIterator, Dict, List, Optional, Sequence, Union
+
+from typing_extensions import override
+
+from ..data import get_template_and_fix_tokenizer
+from ..extras.constants import IMAGE_PLACEHOLDER
+from ..extras.logging import get_logger
+from ..extras.misc import get_device_count
+from ..extras.packages import is_pillow_available, is_vllm_available
+from ..model import load_config, load_tokenizer
+from ..model.model_utils.quantization import QuantizationMethod
+from ..model.model_utils.visual import LlavaMultiModalProjectorForYiVLForVLLM
+from .base_engine import BaseEngine, Response
+
+
+if is_pillow_available():
+    from PIL import Image
+    from PIL.Image import Image as ImageObject
+
+
+if is_vllm_available():
+    from vllm import AsyncEngineArgs, AsyncLLMEngine, RequestOutput, SamplingParams
+    from vllm.lora.request import LoRARequest
+
+
+if TYPE_CHECKING:
+    from ..data.mm_plugin import ImageInput, VideoInput
+    from ..hparams import DataArguments, FinetuningArguments, GeneratingArguments, ModelArguments
+
+
+logger = get_logger(__name__)
+
+
+class VllmEngine(BaseEngine):
+    def __init__(
+        self,
+        model_args: "ModelArguments",
+        data_args: "DataArguments",
+        finetuning_args: "FinetuningArguments",
+        generating_args: "GeneratingArguments",
+    ) -> None:
+        config = load_config(model_args)  # may download model from ms hub
+        if getattr(config, "quantization_config", None):  # gptq models should use float16
+            quantization_config: Dict[str, Any] = getattr(config, "quantization_config", None)
+            quant_method = quantization_config.get("quant_method", "")
+            if quant_method == QuantizationMethod.GPTQ and model_args.infer_dtype == "auto":
+                model_args.infer_dtype = "float16"
+
+        self.can_generate = finetuning_args.stage == "sft"
+        tokenizer_module = load_tokenizer(model_args)
+        self.tokenizer = tokenizer_module["tokenizer"]
+        self.processor = tokenizer_module["processor"]
+        self.tokenizer.padding_side = "left"
+        self.template = get_template_and_fix_tokenizer(self.tokenizer, data_args)
+        self.generating_args = generating_args.to_dict()
+
+        engine_args = {
+            "model": model_args.model_name_or_path,
+            "trust_remote_code": True,
+            "download_dir": model_args.cache_dir,
+            "dtype": model_args.infer_dtype,
+            "max_model_len": model_args.vllm_maxlen,
+            "tensor_parallel_size": get_device_count() or 1,
+            "gpu_memory_utilization": model_args.vllm_gpu_util,
+            "disable_log_stats": True,
+            "disable_log_requests": True,
+            "enforce_eager": model_args.vllm_enforce_eager,
+            "enable_lora": model_args.adapter_name_or_path is not None,
+            "max_lora_rank": model_args.vllm_max_lora_rank,
+        }
+
+        if getattr(config, "is_yi_vl_derived_model", None):
+            import vllm.model_executor.models.llava
+
+            logger.info("Detected Yi-VL model, applying projector patch.")
+            vllm.model_executor.models.llava.LlavaMultiModalProjector = LlavaMultiModalProjectorForYiVLForVLLM
+
+        self.model = AsyncLLMEngine.from_engine_args(AsyncEngineArgs(**engine_args))
+        if model_args.adapter_name_or_path is not None:
+            self.lora_request = LoRARequest("default", 1, model_args.adapter_name_or_path[0])
+        else:
+            self.lora_request = None
+
+    async def _generate(
+        self,
+        messages: Sequence[Dict[str, str]],
+        system: Optional[str] = None,
+        tools: Optional[str] = None,
+        image: Optional["ImageInput"] = None,
+        video: Optional["VideoInput"] = None,
+        **input_kwargs,
+    ) -> AsyncIterator["RequestOutput"]:
+        request_id = "chatcmpl-{}".format(uuid.uuid4().hex)
+        if image is not None:
+            if IMAGE_PLACEHOLDER not in messages[0]["content"]:
+                messages[0]["content"] = IMAGE_PLACEHOLDER + messages[0]["content"]
+
+        paired_messages = messages + [{"role": "assistant", "content": ""}]
+        system = system or self.generating_args["default_system"]
+        prompt_ids, _ = self.template.encode_oneturn(self.tokenizer, paired_messages, system, tools)
+        prompt_length = len(prompt_ids)
+
+        use_beam_search: bool = self.generating_args["num_beams"] > 1
+        temperature: Optional[float] = input_kwargs.pop("temperature", None)
+        top_p: Optional[float] = input_kwargs.pop("top_p", None)
+        top_k: Optional[float] = input_kwargs.pop("top_k", None)
+        num_return_sequences: int = input_kwargs.pop("num_return_sequences", 1)
+        repetition_penalty: Optional[float] = input_kwargs.pop("repetition_penalty", None)
+        length_penalty: Optional[float] = input_kwargs.pop("length_penalty", None)
+        max_length: Optional[int] = input_kwargs.pop("max_length", None)
+        max_new_tokens: Optional[int] = input_kwargs.pop("max_new_tokens", None)
+        stop: Optional[Union[str, List[str]]] = input_kwargs.pop("stop", None)
+
+        if "max_new_tokens" in self.generating_args:
+            max_tokens = self.generating_args["max_new_tokens"]
+        elif "max_length" in self.generating_args:
+            if self.generating_args["max_length"] > prompt_length:
+                max_tokens = self.generating_args["max_length"] - prompt_length
+            else:
+                max_tokens = 1
+
+        if max_length:
+            max_tokens = max_length - prompt_length if max_length > prompt_length else 1
+
+        if max_new_tokens:
+            max_tokens = max_new_tokens
+
+        sampling_params = SamplingParams(
+            n=num_return_sequences,
+            repetition_penalty=(
+                repetition_penalty if repetition_penalty is not None else self.generating_args["repetition_penalty"]
+            )
+            or 1.0,  # repetition_penalty must > 0
+            temperature=temperature if temperature is not None else self.generating_args["temperature"],
+            top_p=(top_p if top_p is not None else self.generating_args["top_p"]) or 1.0,  # top_p must > 0
+            top_k=top_k if top_k is not None else self.generating_args["top_k"],
+            use_beam_search=use_beam_search,
+            length_penalty=length_penalty if length_penalty is not None else self.generating_args["length_penalty"],
+            stop=stop,
+            stop_token_ids=[self.tokenizer.eos_token_id] + self.tokenizer.additional_special_tokens_ids,
+            max_tokens=max_tokens,
+            skip_special_tokens=True,
+        )
+
+        if image is not None:  # add image features
+            if not isinstance(image, (str, ImageObject)):
+                raise ValueError("Expected image input is a path or PIL.Image, but got {}.".format(type(image)))
+
+            if isinstance(image, str):
+                image = Image.open(image).convert("RGB")
+
+            multi_modal_data = {"image": image}
+        else:
+            multi_modal_data = None
+
+        result_generator = self.model.generate(
+            inputs={"prompt_token_ids": prompt_ids, "multi_modal_data": multi_modal_data},
+            sampling_params=sampling_params,
+            request_id=request_id,
+            lora_request=self.lora_request,
+        )
+        return result_generator
+
+    @override
+    async def chat(
+        self,
+        messages: Sequence[Dict[str, str]],
+        system: Optional[str] = None,
+        tools: Optional[str] = None,
+        image: Optional["ImageInput"] = None,
+        video: Optional["VideoInput"] = None,
+        **input_kwargs,
+    ) -> List["Response"]:
+        final_output = None
+        generator = await self._generate(messages, system, tools, image, video, **input_kwargs)
+        async for request_output in generator:
+            final_output = request_output
+
+        results = []
+        for output in final_output.outputs:
+            results.append(
+                Response(
+                    response_text=output.text,
+                    response_length=len(output.token_ids),
+                    prompt_length=len(final_output.prompt_token_ids),
+                    finish_reason=output.finish_reason,
+                )
+            )
+
+        return results
+
+    @override
+    async def stream_chat(
+        self,
+        messages: Sequence[Dict[str, str]],
+        system: Optional[str] = None,
+        tools: Optional[str] = None,
+        image: Optional["ImageInput"] = None,
+        video: Optional["VideoInput"] = None,
+        **input_kwargs,
+    ) -> AsyncGenerator[str, None]:
+        generated_text = ""
+        generator = await self._generate(messages, system, tools, image, video, **input_kwargs)
+        async for result in generator:
+            delta_text = result.outputs[0].text[len(generated_text) :]
+            generated_text = result.outputs[0].text
+            yield delta_text
+
+    @override
+    async def get_scores(
+        self,
+        batch_input: List[str],
+        **input_kwargs,
+    ) -> List[float]:
+        raise NotImplementedError("vLLM engine does not support get_scores.")

llamafactory/cli.py

@@ -0,0 +1,121 @@
+# Copyright 2024 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
+import random
+import subprocess
+import sys
+from enum import Enum, unique
+
+from . import launcher
+from .api.app import run_api
+from .chat.chat_model import run_chat
+from .eval.evaluator import run_eval
+from .extras.env import VERSION, print_env
+from .extras.logging import get_logger
+from .extras.misc import get_device_count
+from .train.tuner import export_model, run_exp
+from .webui.interface import run_web_demo, run_web_ui
+
+
+USAGE = (
+    "-" * 70
+    + "\n"
+    + "| Usage:                                                             |\n"
+    + "|   llamafactory-cli api -h: launch an OpenAI-style API server       |\n"
+    + "|   llamafactory-cli chat -h: launch a chat interface in CLI         |\n"
+    + "|   llamafactory-cli eval -h: evaluate models                        |\n"
+    + "|   llamafactory-cli export -h: merge LoRA adapters and export model |\n"
+    + "|   llamafactory-cli train -h: train models                          |\n"
+    + "|   llamafactory-cli webchat -h: launch a chat interface in Web UI   |\n"
+    + "|   llamafactory-cli webui: launch LlamaBoard                        |\n"
+    + "|   llamafactory-cli version: show version info                      |\n"
+    + "-" * 70
+)
+
+WELCOME = (
+    "-" * 58
+    + "\n"
+    + "| Welcome to LLaMA Factory, version {}".format(VERSION)
+    + " " * (21 - len(VERSION))
+    + "|\n|"
+    + " " * 56
+    + "|\n"
+    + "| Project page: https://github.com/hiyouga/LLaMA-Factory |\n"
+    + "-" * 58
+)
+
+logger = get_logger(__name__)
+
+
+@unique
+class Command(str, Enum):
+    API = "api"
+    CHAT = "chat"
+    ENV = "env"
+    EVAL = "eval"
+    EXPORT = "export"
+    TRAIN = "train"
+    WEBDEMO = "webchat"
+    WEBUI = "webui"
+    VER = "version"
+    HELP = "help"
+
+
+def main():
+    command = sys.argv.pop(1) if len(sys.argv) != 1 else Command.HELP
+    if command == Command.API:
+        run_api()
+    elif command == Command.CHAT:
+        run_chat()
+    elif command == Command.ENV:
+        print_env()
+    elif command == Command.EVAL:
+        run_eval()
+    elif command == Command.EXPORT:
+        export_model()
+    elif command == Command.TRAIN:
+        force_torchrun = os.environ.get("FORCE_TORCHRUN", "0").lower() in ["true", "1"]
+        if force_torchrun or get_device_count() > 1:
+            master_addr = os.environ.get("MASTER_ADDR", "127.0.0.1")
+            master_port = os.environ.get("MASTER_PORT", str(random.randint(20001, 29999)))
+            logger.info("Initializing distributed tasks at: {}:{}".format(master_addr, master_port))
+            process = subprocess.run(
+                (
+                    "torchrun --nnodes {nnodes} --node_rank {node_rank} --nproc_per_node {nproc_per_node} "
+                    "--master_addr {master_addr} --master_port {master_port} {file_name} {args}"
+                ).format(
+                    nnodes=os.environ.get("NNODES", "1"),
+                    node_rank=os.environ.get("RANK", "0"),
+                    nproc_per_node=os.environ.get("NPROC_PER_NODE", str(get_device_count())),
+                    master_addr=master_addr,
+                    master_port=master_port,
+                    file_name=launcher.__file__,
+                    args=" ".join(sys.argv[1:]),
+                ),
+                shell=True,
+            )
+            sys.exit(process.returncode)
+        else:
+            run_exp()
+    elif command == Command.WEBDEMO:
+        run_web_demo()
+    elif command == Command.WEBUI:
+        run_web_ui()
+    elif command == Command.VER:
+        print(WELCOME)
+    elif command == Command.HELP:
+        print(USAGE)
+    else:
+        raise NotImplementedError("Unknown command: {}.".format(command))

llamafactory/data/__init__.py

@@ -0,0 +1,37 @@
+# Copyright 2024 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 .collator import (
+    KTODataCollatorWithPadding,
+    MultiModalDataCollatorForSeq2Seq,
+    PairwiseDataCollatorWithPadding,
+    SFTDataCollatorWith4DAttentionMask,
+)
+from .data_utils import Role, split_dataset
+from .loader import get_dataset
+from .template import TEMPLATES, Template, get_template_and_fix_tokenizer
+
+
+__all__ = [
+    "KTODataCollatorWithPadding",
+    "MultiModalDataCollatorForSeq2Seq",
+    "PairwiseDataCollatorWithPadding",
+    "SFTDataCollatorWith4DAttentionMask",
+    "Role",
+    "split_dataset",
+    "get_dataset",
+    "TEMPLATES",
+    "Template",
+    "get_template_and_fix_tokenizer",
+]

llamafactory/data/aligner.py

@@ -0,0 +1,258 @@
+# Copyright 2024 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 functools import partial
+from typing import TYPE_CHECKING, Any, Dict, List, Optional, Sequence, Union
+
+from ..extras.logging import get_logger
+from .data_utils import Role
+
+
+if TYPE_CHECKING:
+    from datasets import Dataset, IterableDataset
+    from transformers import Seq2SeqTrainingArguments
+
+    from ..hparams import DataArguments
+    from .mm_plugin import ImageInput, VideoInput
+    from .parser import DatasetAttr
+
+
+logger = get_logger(__name__)
+
+
+def _convert_images(
+    images: Sequence["ImageInput"],
+    dataset_attr: "DatasetAttr",
+    data_args: "DataArguments",
+) -> Optional[List["ImageInput"]]:
+    r"""
+    Optionally concatenates image path to dataset dir when loading from local disk.
+    """
+    if len(images) == 0:
+        return None
+
+    images = images[:]
+    if dataset_attr.load_from in ["script", "file"]:
+        for i in range(len(images)):
+            if isinstance(images[i], str) and os.path.isfile(os.path.join(data_args.dataset_dir, images[i])):
+                images[i] = os.path.join(data_args.dataset_dir, images[i])
+
+    return images
+
+
+def _convert_videos(
+    videos: Sequence["VideoInput"],
+    dataset_attr: "DatasetAttr",
+    data_args: "DataArguments",
+) -> Optional[List["VideoInput"]]:
+    r"""
+    Optionally concatenates video path to dataset dir when loading from local disk.
+    """
+    if len(videos) == 0:
+        return None
+
+    videos = videos[:]
+    if dataset_attr.load_from in ["script", "file"]:
+        for i in range(len(videos)):
+            if isinstance(videos[i], str) and os.path.isfile(os.path.join(data_args.dataset_dir, videos[i])):
+                videos[i] = os.path.join(data_args.dataset_dir, videos[i])
+
+    return videos
+
+
+def convert_alpaca(
+    example: Dict[str, Any],
+    dataset_attr: "DatasetAttr",
+    data_args: "DataArguments",
+) -> Dict[str, Any]:
+    r"""
+    Converts alpaca format dataset to the standard format.
+    """
+    prompt = []
+    if dataset_attr.history and isinstance(example[dataset_attr.history], list):
+        for old_prompt, old_response in example[dataset_attr.history]:
+            prompt.append({"role": Role.USER.value, "content": old_prompt})
+            prompt.append({"role": Role.ASSISTANT.value, "content": old_response})
+
+    query = []
+    if dataset_attr.prompt and example[dataset_attr.prompt]:
+        query.append(example[dataset_attr.prompt])
+
+    if dataset_attr.query and example[dataset_attr.query]:
+        query.append(example[dataset_attr.query])
+
+    prompt.append({"role": Role.USER.value, "content": "\n".join(query)})  # "prompt\nquery"
+
+    if dataset_attr.kto_tag and isinstance(example[dataset_attr.kto_tag], bool):  # kto example
+        response = [{"role": Role.ASSISTANT.value, "content": example[dataset_attr.response]}]
+        if example[dataset_attr.kto_tag]:
+            response = response + [{"role": Role.ASSISTANT.value, "content": ""}]
+        else:
+            response = [{"role": Role.ASSISTANT.value, "content": ""}] + response
+    elif (
+        dataset_attr.ranking
+        and isinstance(example[dataset_attr.chosen], str)
+        and isinstance(example[dataset_attr.rejected], str)
+    ):  # pairwise example
+        response = [
+            {"role": Role.ASSISTANT.value, "content": example[dataset_attr.chosen]},
+            {"role": Role.ASSISTANT.value, "content": example[dataset_attr.rejected]},
+        ]
+    elif dataset_attr.response and isinstance(example[dataset_attr.response], str):  # normal example
+        response = [{"role": Role.ASSISTANT.value, "content": example[dataset_attr.response]}]
+    else:  # unsupervised
+        response = []
+
+    convert_images = partial(_convert_images, dataset_attr=dataset_attr, data_args=data_args)
+    convert_videos = partial(_convert_videos, dataset_attr=dataset_attr, data_args=data_args)
+    output = {
+        "_prompt": prompt,
+        "_response": response,
+        "_system": example[dataset_attr.system] if dataset_attr.system else "",
+        "_tools": example[dataset_attr.tools] if dataset_attr.tools else "",
+        "_images": convert_images(example[dataset_attr.images]) if dataset_attr.images else None,
+        "_videos": convert_videos(example[dataset_attr.videos]) if dataset_attr.videos else None,
+    }
+    return output
+
+
+def convert_sharegpt(
+    example: Dict[str, Any],
+    dataset_attr: "DatasetAttr",
+    data_args: "DataArguments",
+) -> Dict[str, Any]:
+    r"""
+    Converts sharegpt format dataset to the standard format.
+    """
+    tag_mapping = {
+        dataset_attr.user_tag: Role.USER.value,
+        dataset_attr.assistant_tag: Role.ASSISTANT.value,
+        dataset_attr.observation_tag: Role.OBSERVATION.value,
+        dataset_attr.function_tag: Role.FUNCTION.value,
+        dataset_attr.system_tag: Role.SYSTEM.value,
+    }
+    odd_tags = (dataset_attr.user_tag, dataset_attr.observation_tag)
+    even_tags = (dataset_attr.assistant_tag, dataset_attr.function_tag)
+    accept_tags = (odd_tags, even_tags)
+    messages = example[dataset_attr.messages]
+    if (
+        dataset_attr.system_tag
+        and len(messages) != 0
+        and messages[0][dataset_attr.role_tag] == dataset_attr.system_tag
+    ):
+        system = messages[0][dataset_attr.content_tag]
+        messages = messages[1:]
+    else:
+        system = example[dataset_attr.system] if dataset_attr.system else ""
+
+    aligned_messages = []
+    broken_data = False
+    for turn_idx, message in enumerate(messages):
+        if message[dataset_attr.role_tag] not in accept_tags[turn_idx % 2]:
+            logger.warning("Invalid role tag in {}.".format(messages))
+            broken_data = True
+
+        aligned_messages.append(
+            {"role": tag_mapping[message[dataset_attr.role_tag]], "content": message[dataset_attr.content_tag]}
+        )
+
+    if (not dataset_attr.ranking and len(aligned_messages) % 2 != 0) or (
+        dataset_attr.ranking and len(aligned_messages) % 2 == 0
+    ):
+        logger.warning("Invalid message count in {}.".format(messages))
+        broken_data = True
+
+    if dataset_attr.kto_tag and isinstance(example[dataset_attr.kto_tag], bool):  # kto example
+        prompt = aligned_messages[:-1]
+        response = aligned_messages[-1:]
+        if example[dataset_attr.kto_tag]:
+            response = response + [{"role": Role.ASSISTANT.value, "content": ""}]
+        else:
+            response = [{"role": Role.ASSISTANT.value, "content": ""}] + response
+    elif (
+        dataset_attr.ranking
+        and isinstance(example[dataset_attr.chosen], dict)
+        and isinstance(example[dataset_attr.rejected], dict)
+    ):  # pairwise example
+        chosen = example[dataset_attr.chosen]
+        rejected = example[dataset_attr.rejected]
+        if (
+            chosen[dataset_attr.role_tag] not in accept_tags[-1]
+            or rejected[dataset_attr.role_tag] not in accept_tags[-1]
+        ):
+            logger.warning("Invalid role tag in {}.".format([chosen, rejected]))
+            broken_data = True
+
+        prompt = aligned_messages
+        response = [
+            {"role": tag_mapping[chosen[dataset_attr.role_tag]], "content": chosen[dataset_attr.content_tag]},
+            {"role": tag_mapping[rejected[dataset_attr.role_tag]], "content": rejected[dataset_attr.content_tag]},
+        ]
+    else:  # normal example
+        prompt = aligned_messages[:-1]
+        response = aligned_messages[-1:]
+
+    if broken_data:
+        logger.warning("Skipping this abnormal example.")
+        prompt, response = [], []
+
+    convert_images = partial(_convert_images, dataset_attr=dataset_attr, data_args=data_args)
+    convert_videos = partial(_convert_videos, dataset_attr=dataset_attr, data_args=data_args)
+    output = {
+        "_prompt": prompt,
+        "_response": response,
+        "_system": system,
+        "_tools": example[dataset_attr.tools] if dataset_attr.tools else "",
+        "_images": convert_images(example[dataset_attr.images]) if dataset_attr.images else None,
+        "_videos": convert_videos(example[dataset_attr.videos]) if dataset_attr.videos else None,
+    }
+    return output
+
+
+def align_dataset(
+    dataset: Union["Dataset", "IterableDataset"],
+    dataset_attr: "DatasetAttr",
+    data_args: "DataArguments",
+    training_args: "Seq2SeqTrainingArguments",
+) -> Union["Dataset", "IterableDataset"]:
+    r"""
+    Aligned dataset:
+        _prompt: [{"role": "user", "content": "..."}] * (2T - 1)
+        _response: [{"role": "assistant", "content": "..."}] * N (N > 1 for ranking dataset)
+        _system: "..."
+        _tools: "...",
+        _images: [],
+        _videos: [],
+    """
+    if dataset_attr.formatting == "alpaca":
+        convert_func = partial(convert_alpaca, dataset_attr=dataset_attr, data_args=data_args)
+    else:
+        convert_func = partial(convert_sharegpt, dataset_attr=dataset_attr, data_args=data_args)
+
+    column_names = list(next(iter(dataset)).keys())
+    kwargs = {}
+    if not data_args.streaming:
+        kwargs = dict(
+            num_proc=data_args.preprocessing_num_workers,
+            load_from_cache_file=(not data_args.overwrite_cache) or (training_args.local_process_index != 0),
+            desc="Converting format of dataset",
+        )
+
+    return dataset.map(
+        convert_func,
+        batched=False,
+        remove_columns=column_names,
+        **kwargs,
+    )

llamafactory/data/collator.py

@@ -0,0 +1,189 @@
+# Copyright 2024 OpenAccess AI Collective and the LlamaFactory team.
+#
+# This code is inspired by the OpenAccess AI Collective's axolotl library.
+# https://github.com/OpenAccess-AI-Collective/axolotl/blob/main/src/axolotl/monkeypatch/utils.py
+#
+# 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 dataclasses import dataclass
+from typing import TYPE_CHECKING, Any, Dict, Literal, Optional, Sequence
+
+import torch
+from transformers import DataCollatorForSeq2Seq
+
+
+if TYPE_CHECKING:
+    from transformers import ProcessorMixin
+
+    from .template import Template
+
+
+def prepare_4d_attention_mask(attention_mask_with_indices: "torch.Tensor", dtype: "torch.dtype") -> "torch.Tensor":
+    r"""
+    Expands the attention mask with indices from (batch_size, seq_len) to (batch_size, 1, seq_len, seq_len),
+    while handles packed sequences and transforms the mask to lower triangular form to prevent future peeking.
+
+    e.g.
+    ```python
+    # input
+    [[1, 1, 2, 2, 2, 0]]
+    # output
+    [
+        [
+            [
+                [o, x, x, x, x, x],
+                [o, o, x, x, x, x],
+                [x, x, o, x, x, x],
+                [x, x, o, o, x, x],
+                [x, x, o, o, o, x],
+                [x, x, x, x, x, x],
+            ]
+        ]
+    ]
+    ```
+    where `o` equals to `0.0`, `x` equals to `min_dtype`.
+    """
+    bsz, seq_len = attention_mask_with_indices.size()
+    min_dtype = torch.finfo(dtype).min
+    expanded_mask = attention_mask_with_indices[:, None, None, :].expand(bsz, 1, seq_len, seq_len)
+    # Create a binary mask from the original mask where zeros remain zeros and all other values are set to one
+    padding_mask = torch.where(expanded_mask != 0, 1, 0)
+    # Create a block-diagonal mask.
+    attention_mask_4d = torch.eq(expanded_mask, expanded_mask.transpose(-1, -2)).int() * padding_mask
+    # Use the lower triangular mask to zero out the upper triangular part
+    attention_mask_4d *= torch.tril(torch.ones((seq_len, seq_len), dtype=torch.long))
+    # Invert the attention mask.
+    attention_mask_4d = torch.where(attention_mask_4d != 0, torch.tensor(0, dtype=dtype), min_dtype)
+    return attention_mask_4d
+
+
+@dataclass
+class MultiModalDataCollatorForSeq2Seq(DataCollatorForSeq2Seq):
+    r"""
+    Data collator that supports VLMs.
+
+    Features should contain input_ids, attention_mask, labels and images.
+    """
+
+    template: Optional["Template"] = None
+    processor: Optional["ProcessorMixin"] = None
+
+    def __call__(self, features: Sequence[Dict[str, Any]]) -> Dict[str, "torch.Tensor"]:
+        batch_images, batch_videos, batch_imglens, batch_vidlens, batch_seqlens = [], [], [], [], []
+        for feature in features:
+            images = feature.pop("images", None) or []
+            videos = feature.pop("videos", None) or []
+            batch_images.extend(images)
+            batch_videos.extend(videos)
+            batch_imglens.append(len(images))
+            batch_vidlens.append(len(videos))
+            batch_seqlens.append(len(feature["input_ids"]))
+
+        mm_inputs = self.template.mm_plugin.get_mm_inputs(
+            batch_images, batch_videos, batch_imglens, batch_vidlens, batch_seqlens, self.processor
+        )
+        if "token_type_ids" in mm_inputs:
+            token_type_ids = mm_inputs.pop("token_type_ids")
+            for i, feature in enumerate(features):
+                feature["token_type_ids"] = token_type_ids[i]
+
+        features: Dict[str, "torch.Tensor"] = super().__call__(features)
+        features.update(mm_inputs)
+        return features
+
+
+@dataclass
+class SFTDataCollatorWith4DAttentionMask(MultiModalDataCollatorForSeq2Seq):
+    r"""
+    Data collator for 4d attention mask.
+    """
+
+    block_diag_attn: bool = False
+    attn_implementation: Literal["eager", "sdpa", "flash_attention_2"] = "eager"
+    compute_dtype: "torch.dtype" = torch.float32
+
+    def __call__(self, features: Sequence[Dict[str, Any]]) -> Dict[str, "torch.Tensor"]:
+        features = super().__call__(features)
+        if self.block_diag_attn and self.attn_implementation != "flash_attention_2":
+            features["attention_mask"] = prepare_4d_attention_mask(features["attention_mask"], self.compute_dtype)
+
+        return features
+
+
+@dataclass
+class PairwiseDataCollatorWithPadding(MultiModalDataCollatorForSeq2Seq):
+    r"""
+    Data collator for pairwise data.
+    """
+
+    def __call__(self, features: Sequence[Dict[str, Any]]) -> Dict[str, "torch.Tensor"]:
+        r"""
+        Pads batched data to the longest sequence in the batch.
+
+        We generate 2 * n examples where the first n examples represent chosen examples and
+        the last n examples represent rejected examples.
+        """
+        concatenated_features = []
+        for key in ("chosen", "rejected"):
+            for feature in features:
+                target_feature = {
+                    "input_ids": feature["{}_input_ids".format(key)],
+                    "attention_mask": feature["{}_attention_mask".format(key)],
+                    "labels": feature["{}_labels".format(key)],
+                    "images": feature["images"],
+                    "videos": feature["videos"],
+                }
+                concatenated_features.append(target_feature)
+
+        return super().__call__(concatenated_features)
+
+
+@dataclass
+class KTODataCollatorWithPadding(MultiModalDataCollatorForSeq2Seq):
+    r"""
+    Data collator for KTO data.
+    """
+
+    def __call__(self, features: Sequence[Dict[str, Any]]) -> Dict[str, "torch.Tensor"]:
+        target_features = []
+        kl_features = []
+        kto_tags = []
+        for feature in features:
+            target_feature = {
+                "input_ids": feature["input_ids"],
+                "attention_mask": feature["attention_mask"],
+                "labels": feature["labels"],
+                "images": feature["images"],
+                "videos": feature["videos"],
+            }
+            kl_feature = {
+                "input_ids": feature["kl_input_ids"],
+                "attention_mask": feature["kl_attention_mask"],
+                "labels": feature["kl_labels"],
+                "images": feature["images"],
+                "videos": feature["videos"],
+            }
+            target_features.append(target_feature)
+            kl_features.append(kl_feature)
+            kto_tags.append(feature["kto_tags"])
+
+        batch = super().__call__(target_features)
+        kl_batch = super().__call__(kl_features)
+        batch["kl_input_ids"] = kl_batch["input_ids"]
+        batch["kl_attention_mask"] = kl_batch["attention_mask"]
+        batch["kl_labels"] = kl_batch["labels"]
+        if "token_type_ids" in kl_batch:
+            batch["kl_token_type_ids"] = kl_batch["token_type_ids"]
+
+        batch["kto_tags"] = torch.tensor(kto_tags)
+        return batch

llamafactory/data/data_utils.py

@@ -0,0 +1,92 @@
+# Copyright 2024 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 enum import Enum, unique
+from typing import TYPE_CHECKING, Dict, List, Optional, Sequence, Set, TypedDict, Union
+
+from datasets import DatasetDict, concatenate_datasets, interleave_datasets
+
+from ..extras.logging import get_logger
+
+
+if TYPE_CHECKING:
+    from datasets import Dataset, IterableDataset
+
+    from ..hparams import DataArguments
+
+
+logger = get_logger(__name__)
+
+
+SLOTS = Sequence[Union[str, Set[str], Dict[str, str]]]
+
+
+@unique
+class Role(str, Enum):
+    USER = "user"
+    ASSISTANT = "assistant"
+    SYSTEM = "system"
+    FUNCTION = "function"
+    OBSERVATION = "observation"
+
+
+class DatasetModule(TypedDict):
+    train_dataset: Optional[Union["Dataset", "IterableDataset"]]
+    eval_dataset: Optional[Union["Dataset", "IterableDataset"]]
+
+
+def merge_dataset(
+    all_datasets: List[Union["Dataset", "IterableDataset"]], data_args: "DataArguments", seed: int
+) -> Union["Dataset", "IterableDataset"]:
+    r"""
+    Merges multiple datasets to a unified dataset.
+    """
+    if len(all_datasets) == 1:
+        return all_datasets[0]
+    elif data_args.mix_strategy == "concat":
+        if data_args.streaming:
+            logger.warning("The samples between different datasets will not be mixed in streaming mode.")
+
+        return concatenate_datasets(all_datasets)
+    elif data_args.mix_strategy.startswith("interleave"):
+        if not data_args.streaming:
+            logger.warning("We recommend using `mix_strategy=concat` in non-streaming mode.")
+
+        return interleave_datasets(
+            datasets=all_datasets,
+            probabilities=data_args.interleave_probs,
+            seed=seed,
+            stopping_strategy="first_exhausted" if data_args.mix_strategy.endswith("under") else "all_exhausted",
+        )
+    else:
+        raise ValueError("Unknown mixing strategy: {}.".format(data_args.mix_strategy))
+
+
+def split_dataset(
+    dataset: Union["Dataset", "IterableDataset"], data_args: "DataArguments", seed: int
+) -> "DatasetDict":
+    r"""
+    Splits the dataset and returns a dataset dict containing train set and validation set.
+
+    Supports both map dataset and iterable dataset.
+    """
+    if data_args.streaming:
+        dataset = dataset.shuffle(buffer_size=data_args.buffer_size, seed=seed)
+        val_set = dataset.take(int(data_args.val_size))
+        train_set = dataset.skip(int(data_args.val_size))
+        return DatasetDict({"train": train_set, "validation": val_set})
+    else:
+        val_size = int(data_args.val_size) if data_args.val_size > 1 else data_args.val_size
+        dataset = dataset.train_test_split(test_size=val_size, seed=seed)
+        return DatasetDict({"train": dataset["train"], "validation": dataset["test"]})

llamafactory/data/formatter.py

@@ -0,0 +1,148 @@
+# Copyright 2024 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 json
+import re
+from abc import ABC, abstractmethod
+from dataclasses import dataclass, field
+from typing import TYPE_CHECKING, List, Optional, Tuple, Union
+
+from typing_extensions import override
+
+from .data_utils import SLOTS
+from .tool_utils import get_tool_utils
+
+
+if TYPE_CHECKING:
+    from .tool_utils import FunctionCall
+
+
+@dataclass
+class Formatter(ABC):
+    slots: SLOTS = field(default_factory=list)
+    tool_format: Optional[str] = None
+
+    @abstractmethod
+    def apply(self, **kwargs) -> SLOTS:
+        r"""
+        Forms a list of slots according to the inputs to encode.
+        """
+        ...
+
+    def extract(self, content: str) -> Union[str, List["FunctionCall"]]:
+        r"""
+        Extract a list of tuples from the response message if using tools.
+
+        Each tuple consists of function name and function arguments.
+        """
+        raise NotImplementedError
+
+
+@dataclass
+class EmptyFormatter(Formatter):
+    def __post_init__(self):
+        has_placeholder = False
+        for slot in filter(lambda s: isinstance(s, str), self.slots):
+            if re.search(r"\{\{[a-zA-Z_][a-zA-Z0-9_]*\}\}", slot):
+                has_placeholder = True
+
+        if has_placeholder:
+            raise ValueError("Empty formatter should not contain any placeholder.")
+
+    @override
+    def apply(self, **kwargs) -> SLOTS:
+        return self.slots
+
+
+@dataclass
+class StringFormatter(Formatter):
+    def __post_init__(self):
+        has_placeholder = False
+        for slot in filter(lambda s: isinstance(s, str), self.slots):
+            if re.search(r"\{\{[a-zA-Z_][a-zA-Z0-9_]*\}\}", slot):
+                has_placeholder = True
+
+        if not has_placeholder:
+            raise ValueError("A placeholder is required in the string formatter.")
+
+    @override
+    def apply(self, **kwargs) -> SLOTS:
+        elements = []
+        for slot in self.slots:
+            if isinstance(slot, str):
+                for name, value in kwargs.items():
+                    if not isinstance(value, str):
+                        raise RuntimeError("Expected a string, got {}".format(value))
+
+                    slot = slot.replace("{{" + name + "}}", value, 1)
+                elements.append(slot)
+            elif isinstance(slot, (dict, set)):
+                elements.append(slot)
+            else:
+                raise RuntimeError("Input must be string, set[str] or dict[str, str], got {}".format(type(slot)))
+
+        return elements
+
+
+@dataclass
+class FunctionFormatter(Formatter):
+    def __post_init__(self):
+        self.slots = get_tool_utils(self.tool_format).get_function_slots() + self.slots
+
+    @override
+    def apply(self, **kwargs) -> SLOTS:
+        content = kwargs.pop("content")
+        functions: List[Tuple[str, str]] = []
+        try:
+            tool_calls = json.loads(content)
+            if not isinstance(tool_calls, list):  # parallel function call
+                tool_calls = [tool_calls]
+
+            for tool_call in tool_calls:
+                functions.append((tool_call["name"], json.dumps(tool_call["arguments"], ensure_ascii=False)))
+
+        except json.JSONDecodeError:
+            raise RuntimeError("Invalid JSON format in function message: {}".format(str([content])))  # flat string
+
+        elements = []
+        for name, arguments in functions:
+            for slot in self.slots:
+                if isinstance(slot, str):
+                    slot = slot.replace("{{name}}", name).replace("{{arguments}}", arguments)
+                    elements.append(slot)
+                elif isinstance(slot, (dict, set)):
+                    elements.append(slot)
+                else:
+                    raise RuntimeError("Input must be string, set[str] or dict[str, str], got {}".format(type(slot)))
+
+        return elements
+
+
+@dataclass
+class ToolFormatter(Formatter):
+    def __post_init__(self):
+        self.tool_utils = get_tool_utils(self.tool_format)
+
+    @override
+    def apply(self, **kwargs) -> SLOTS:
+        content = kwargs.pop("content")
+        try:
+            tools = json.loads(content)
+            return [self.tool_utils.tool_formatter(tools) if len(tools) != 0 else ""]
+        except json.JSONDecodeError:
+            raise RuntimeError("Invalid JSON format in tool description: {}".format(str([content])))  # flat string
+
+    @override
+    def extract(self, content: str) -> Union[str, List["FunctionCall"]]:
+        return self.tool_utils.tool_extractor(content)

llamafactory/data/loader.py

@@ -0,0 +1,292 @@
+# Copyright 2024 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
+import sys
+from typing import TYPE_CHECKING, Dict, Literal, Optional, Sequence, Union
+
+import numpy as np
+from datasets import DatasetDict, load_dataset, load_from_disk
+from transformers.utils.versions import require_version
+
+from ..extras.constants import FILEEXT2TYPE
+from ..extras.logging import get_logger
+from ..extras.misc import has_tokenized_data
+from .aligner import align_dataset
+from .data_utils import merge_dataset, split_dataset
+from .parser import get_dataset_list
+from .preprocess import get_preprocess_and_print_func
+
+
+if TYPE_CHECKING:
+    from datasets import Dataset, IterableDataset
+    from transformers import PreTrainedTokenizer, ProcessorMixin, Seq2SeqTrainingArguments
+
+    from ..hparams import DataArguments, ModelArguments
+    from .data_utils import DatasetModule
+    from .parser import DatasetAttr
+    from .template import Template
+
+
+logger = get_logger(__name__)
+
+
+def _load_single_dataset(
+    dataset_attr: "DatasetAttr",
+    model_args: "ModelArguments",
+    data_args: "DataArguments",
+    training_args: "Seq2SeqTrainingArguments",
+) -> Union["Dataset", "IterableDataset"]:
+    r"""
+    Loads a single dataset and aligns it to the standard format.
+    """
+    logger.info("Loading dataset {}...".format(dataset_attr))
+    data_path, data_name, data_dir, data_files = None, None, None, None
+    if dataset_attr.load_from in ["hf_hub", "ms_hub"]:
+        data_path = dataset_attr.dataset_name
+        data_name = dataset_attr.subset
+        data_dir = dataset_attr.folder
+
+    elif dataset_attr.load_from == "script":
+        data_path = os.path.join(data_args.dataset_dir, dataset_attr.dataset_name)
+        data_name = dataset_attr.subset
+        data_dir = dataset_attr.folder
+
+    elif dataset_attr.load_from == "file":
+        data_files = []
+        local_path = os.path.join(data_args.dataset_dir, dataset_attr.dataset_name)
+        if os.path.isdir(local_path):  # is directory
+            for file_name in os.listdir(local_path):
+                data_files.append(os.path.join(local_path, file_name))
+                if data_path is None:
+                    data_path = FILEEXT2TYPE.get(file_name.split(".")[-1], None)
+                elif data_path != FILEEXT2TYPE.get(file_name.split(".")[-1], None):
+                    raise ValueError("File types should be identical.")
+        elif os.path.isfile(local_path):  # is file
+            data_files.append(local_path)
+            data_path = FILEEXT2TYPE.get(local_path.split(".")[-1], None)
+        else:
+            raise ValueError("File {} not found.".format(local_path))
+
+        if data_path is None:
+            raise ValueError("Allowed file types: {}.".format(",".join(FILEEXT2TYPE.keys())))
+    else:
+        raise NotImplementedError("Unknown load type: {}.".format(dataset_attr.load_from))
+
+    if dataset_attr.load_from == "ms_hub":
+        require_version("modelscope>=1.11.0", "To fix: pip install modelscope>=1.11.0")
+        from modelscope import MsDataset
+        from modelscope.utils.config_ds import MS_DATASETS_CACHE
+
+        cache_dir = model_args.cache_dir or MS_DATASETS_CACHE
+        dataset = MsDataset.load(
+            dataset_name=data_path,
+            subset_name=data_name,
+            data_dir=data_dir,
+            data_files=data_files,
+            split=dataset_attr.split,
+            cache_dir=cache_dir,
+            token=model_args.ms_hub_token,
+            use_streaming=(data_args.streaming and (dataset_attr.load_from != "file")),
+        )
+        if isinstance(dataset, MsDataset):
+            dataset = dataset.to_hf_dataset()
+    else:
+        dataset = load_dataset(
+            path=data_path,
+            name=data_name,
+            data_dir=data_dir,
+            data_files=data_files,
+            split=dataset_attr.split,
+            cache_dir=model_args.cache_dir,
+            token=model_args.hf_hub_token,
+            streaming=(data_args.streaming and (dataset_attr.load_from != "file")),
+            trust_remote_code=True,
+        )
+
+    if data_args.streaming and (dataset_attr.load_from == "file"):  # faster than specifying streaming=True
+        dataset = dataset.to_iterable_dataset()  # TODO: add num shards parameter
+
+    if dataset_attr.num_samples is not None and not data_args.streaming:
+        target_num = dataset_attr.num_samples
+        indexes = np.random.permutation(len(dataset))[:target_num]  # all samples should be included
+        target_num -= len(indexes)
+        if target_num > 0:
+            expand_indexes = np.random.choice(len(dataset), target_num)
+            indexes = np.concatenate((indexes, expand_indexes), axis=0)
+
+        assert len(indexes) == dataset_attr.num_samples, "Sample num mismatched."
+        dataset = dataset.select(indexes)
+        logger.info("Sampled {} examples from dataset {}.".format(dataset_attr.num_samples, dataset_attr))
+
+    if data_args.max_samples is not None:  # truncate dataset
+        max_samples = min(data_args.max_samples, len(dataset))
+        dataset = dataset.select(range(max_samples))
+
+    return align_dataset(dataset, dataset_attr, data_args, training_args)
+
+
+def _get_merged_dataset(
+    dataset_names: Optional[Sequence[str]],
+    model_args: "ModelArguments",
+    data_args: "DataArguments",
+    training_args: "Seq2SeqTrainingArguments",
+    stage: Literal["pt", "sft", "rm", "ppo", "kto"],
+) -> Optional[Union["Dataset", "IterableDataset"]]:
+    r"""
+    Gets the merged datasets in the standard format.
+    """
+    if dataset_names is None:
+        return None
+
+    datasets = []
+    for dataset_attr in get_dataset_list(dataset_names, data_args.dataset_dir):
+        if (stage == "rm" and dataset_attr.ranking is False) or (stage != "rm" and dataset_attr.ranking is True):
+            raise ValueError("The dataset is not applicable in the current training stage.")
+
+        datasets.append(_load_single_dataset(dataset_attr, model_args, data_args, training_args))
+
+    return merge_dataset(datasets, data_args, seed=training_args.seed)
+
+
+def _get_preprocessed_dataset(
+    dataset: Optional[Union["Dataset", "IterableDataset"]],
+    data_args: "DataArguments",
+    training_args: "Seq2SeqTrainingArguments",
+    stage: Literal["pt", "sft", "rm", "ppo", "kto"],
+    template: "Template",
+    tokenizer: "PreTrainedTokenizer",
+    processor: Optional["ProcessorMixin"] = None,
+    is_eval: bool = False,
+) -> Optional[Union["Dataset", "IterableDataset"]]:
+    r"""
+    Preprocesses the dataset, including format checking and tokenization.
+    """
+    if dataset is None:
+        return None
+
+    preprocess_func, print_function = get_preprocess_and_print_func(
+        data_args, stage, template, tokenizer, processor, do_generate=(training_args.predict_with_generate and is_eval)
+    )
+    column_names = list(next(iter(dataset)).keys())
+    kwargs = {}
+    if not data_args.streaming:
+        kwargs = dict(
+            num_proc=data_args.preprocessing_num_workers,
+            load_from_cache_file=(not data_args.overwrite_cache) or (training_args.local_process_index != 0),
+            desc="Running tokenizer on dataset",
+        )
+
+    dataset = dataset.map(
+        preprocess_func,
+        batched=True,
+        batch_size=data_args.preprocessing_batch_size,
+        remove_columns=column_names,
+        **kwargs,
+    )
+
+    if training_args.should_log:
+        try:
+            print("eval example:" if is_eval else "training example:")
+            print_function(next(iter(dataset)))
+        except StopIteration:
+            if stage == "pt":
+                raise RuntimeError("Cannot find sufficient samples, consider increasing dataset size.")
+            else:
+                raise RuntimeError("Cannot find valid samples, check `data/README.md` for the data format.")
+
+    return dataset
+
+
+def get_dataset(
+    template: "Template",
+    model_args: "ModelArguments",
+    data_args: "DataArguments",
+    training_args: "Seq2SeqTrainingArguments",
+    stage: Literal["pt", "sft", "rm", "ppo", "kto"],
+    tokenizer: "PreTrainedTokenizer",
+    processor: Optional["ProcessorMixin"] = None,
+) -> "DatasetModule":
+    r"""
+    Gets the train dataset and optionally gets the evaluation dataset.
+    """
+    # Load tokenized dataset
+    if data_args.tokenized_path is not None:
+        if has_tokenized_data(data_args.tokenized_path):
+            logger.warning("Loading dataset from disk will ignore other data arguments.")
+            dataset_dict: "DatasetDict" = load_from_disk(data_args.tokenized_path)
+            logger.info("Loaded tokenized dataset from {}.".format(data_args.tokenized_path))
+
+            dataset_module: Dict[str, "Dataset"] = {}
+            if "train" in dataset_dict:
+                dataset_module["train_dataset"] = dataset_dict["train"]
+
+            if "validation" in dataset_dict:
+                dataset_module["eval_dataset"] = dataset_dict["validation"]
+
+            if data_args.streaming:
+                dataset_module = {k: v.to_iterable_dataset() for k, v in dataset_module.items()}
+
+            return dataset_module
+
+        if data_args.streaming:
+            raise ValueError("Turn off `streaming` when saving dataset to disk.")
+
+    # Load and preprocess dataset
+    with training_args.main_process_first(desc="load dataset"):
+        dataset = _get_merged_dataset(data_args.dataset, model_args, data_args, training_args, stage)
+        eval_dataset = _get_merged_dataset(data_args.eval_dataset, model_args, data_args, training_args, stage)
+
+    with training_args.main_process_first(desc="pre-process dataset"):
+        dataset = _get_preprocessed_dataset(
+            dataset, data_args, training_args, stage, template, tokenizer, processor, is_eval=False
+        )
+        eval_dataset = _get_preprocessed_dataset(
+            eval_dataset, data_args, training_args, stage, template, tokenizer, processor, is_eval=True
+        )
+
+        if data_args.val_size > 1e-6:
+            dataset_dict = split_dataset(dataset, data_args, seed=training_args.seed)
+        else:
+            dataset_dict = {}
+            if dataset is not None:
+                if data_args.streaming:
+                    dataset = dataset.shuffle(buffer_size=data_args.buffer_size, seed=training_args.seed)
+
+                dataset_dict["train"] = dataset
+
+            if eval_dataset is not None:
+                if data_args.streaming:
+                    eval_dataset = eval_dataset.shuffle(buffer_size=data_args.buffer_size, seed=training_args.seed)
+
+                dataset_dict["validation"] = eval_dataset
+
+            dataset_dict = DatasetDict(dataset_dict)
+
+        if data_args.tokenized_path is not None:
+            if training_args.should_save:
+                dataset_dict.save_to_disk(data_args.tokenized_path)
+                logger.info("Tokenized dataset saved at {}.".format(data_args.tokenized_path))
+                logger.info("Please restart the training with `tokenized_path: {}`.".format(data_args.tokenized_path))
+
+            sys.exit(0)
+
+        dataset_module = {}
+        if "train" in dataset_dict:
+            dataset_module["train_dataset"] = dataset_dict["train"]
+
+        if "validation" in dataset_dict:
+            dataset_module["eval_dataset"] = dataset_dict["validation"]
+
+        return dataset_module

llamafactory/data/mm_plugin.py

@@ -0,0 +1,627 @@
+import math
+from copy import deepcopy
+from io import BytesIO
+from typing import TYPE_CHECKING, Dict, List, Optional, Sequence, Tuple, TypedDict, Union
+
+import numpy as np
+from transformers.image_utils import get_image_size, to_numpy_array
+from typing_extensions import override
+
+from ..extras.constants import IGNORE_INDEX, IMAGE_PLACEHOLDER, VIDEO_PLACEHOLDER
+from ..extras.packages import is_pillow_available, is_pyav_available
+
+
+if is_pillow_available():
+    from PIL import Image
+    from PIL.Image import Image as ImageObject
+
+
+if is_pyav_available():
+    import av
+
+
+if TYPE_CHECKING:
+    import torch
+    from av.stream import Stream
+    from transformers import PreTrainedTokenizer, ProcessorMixin
+    from transformers.image_processing_utils import BaseImageProcessor
+
+    class EncodedImage(TypedDict):
+        path: Optional[str]
+        bytes: Optional[bytes]
+
+    ImageInput = Union[str, EncodedImage, ImageObject]
+    VideoInput = str
+
+
+def _get_paligemma_token_type_ids(
+    imglens: Sequence[int], seqlens: Sequence[int], processor: "ProcessorMixin"
+) -> List[List[int]]:
+    r"""
+    Gets paligemma token type ids for computing loss.
+
+    Returns:
+        batch_token_type_ids: shape (batch_size, sequence_length)
+    """
+    batch_token_type_ids = []
+    for imglen, seqlen in zip(imglens, seqlens):
+        image_seqlen = imglen * getattr(processor, "image_seqlen")
+        batch_token_type_ids.append([0] * image_seqlen + [1] * (seqlen - image_seqlen))
+
+    return batch_token_type_ids
+
+
+class BasePlugin:
+    def __init__(self, image_token: Optional[str], video_token: Optional[str]) -> None:
+        self.image_token = image_token
+        self.video_token = video_token
+
+    def _validate_input(
+        self,
+        images: Sequence["ImageInput"],
+        videos: Sequence["VideoInput"],
+    ) -> None:
+        r"""
+        Validates if this model accepts the input modalities.
+        """
+        if len(images) != 0 and self.image_token is None:
+            raise ValueError("This model does not support image input.")
+
+        if len(videos) != 0 and self.video_token is None:
+            raise ValueError("This model does not support video input.")
+
+    def _preprocess_image(self, image: "ImageObject", **kwargs) -> "ImageObject":
+        r"""
+        Pre-processes a single image.
+        """
+        image_resolution: int = kwargs.get("image_resolution")
+        if max(image.width, image.height) > image_resolution:
+            resize_factor = image_resolution / max(image.width, image.height)
+            width, height = int(image.width * resize_factor), int(image.height * resize_factor)
+            image = image.resize((width, height), resample=Image.NEAREST)
+
+        if image.mode != "RGB":
+            image = image.convert("RGB")
+
+        return image
+
+    def _get_video_sample_frames(self, video_stream: "Stream", **kwargs) -> int:
+        r"""
+        Computes video sample frames according to fps.
+        """
+        video_fps: float = kwargs.get("video_fps")
+        video_maxlen: int = kwargs.get("video_maxlen")
+        total_frames = video_stream.frames
+        sample_frames = float(video_stream.duration * video_stream.time_base) * video_fps
+        sample_frames = min(total_frames, video_maxlen, sample_frames)
+        return math.floor(sample_frames)
+
+    def _regularize_images(self, images: Sequence["ImageInput"], **kwargs) -> List["ImageObject"]:
+        r"""
+        Regularizes images to avoid error. Including reading and pre-processing.
+        """
+        results = []
+        for image in images:
+            if isinstance(image, str):
+                image = Image.open(image)
+            elif isinstance(image, dict):
+                if image["bytes"] is not None:
+                    image = Image.open(BytesIO(image["bytes"]))
+                else:
+                    image = Image.open(image["path"])
+
+            if not isinstance(image, ImageObject):
+                raise ValueError("Expect input is a list of Images, but got {}.".format(type(image)))
+
+            results.append(self._preprocess_image(image, **kwargs))
+
+        return results
+
+    def _regularize_videos(self, videos: Sequence["VideoInput"], **kwargs) -> List[List["ImageObject"]]:
+        r"""
+        Regularizes videos to avoid error. Including reading, resizing and converting.
+        """
+        results = []
+        for video in videos:
+            container = av.open(video, "r")
+            video_stream = next(stream for stream in container.streams if stream.type == "video")
+            total_frames = video_stream.frames
+            sample_frames = self._get_video_sample_frames(video_stream, **kwargs)
+            sample_indices = np.linspace(0, total_frames - 1, sample_frames).astype(np.int32)
+            frames: List["ImageObject"] = []
+            container.seek(0)
+            for frame_idx, frame in enumerate(container.decode(video_stream)):
+                if frame_idx in sample_indices:
+                    frames.append(frame.to_image())
+
+            frames = self._regularize_images(frames, **kwargs)
+            results.append(frames)
+
+        return results
+
+    def _get_mm_inputs(
+        self,
+        images: Sequence["ImageInput"],
+        videos: Sequence["VideoInput"],
+        processor: "ProcessorMixin",
+    ) -> Dict[str, "torch.Tensor"]:
+        r"""
+        Processes visual inputs.
+
+        Returns: (llava and paligemma)
+            pixel_values: tensor with shape (B, C, H, W)
+
+        Returns: (qwen2-vl)
+            pixel_values: tensor with shape (num_patches, patch_dim)
+            image_grid_thw: tensor with shape (num_images, 3), where the three numbers are time, width, height
+
+        It holds num_patches == torch.prod(image_grid_thw)
+        """
+        image_processor: "BaseImageProcessor" = getattr(processor, "image_processor")
+        video_processor: "BaseImageProcessor" = getattr(processor, "video_processor", image_processor)
+        input_dict = {"images": None}  # default key
+        if len(images) != 0:
+            images = self._regularize_images(
+                images,
+                image_resolution=getattr(processor, "image_resolution", 512),
+            )
+            input_dict["images"] = images
+
+        if len(videos) != 0:
+            videos = self._regularize_videos(
+                videos,
+                image_resolution=getattr(processor, "video_resolution", 128),
+                video_fps=getattr(processor, "video_fps", 1.0),
+                video_maxlen=getattr(processor, "video_maxlen", 64),
+            )
+            input_dict["videos"] = videos
+
+        mm_inputs = {}
+        if image_processor != video_processor:
+            if input_dict.get("images") is not None:
+                mm_inputs.update(image_processor(input_dict["images"], return_tensors="pt"))
+            if input_dict.get("videos") is not None:
+                mm_inputs.update(video_processor(input_dict["videos"], return_tensors="pt"))
+        elif input_dict.get("images") is not None or input_dict.get("videos") is not None:  # same processor (qwen2-vl)
+            mm_inputs.update(image_processor(**input_dict, return_tensors="pt"))
+
+        return mm_inputs
+
+    def process_messages(
+        self,
+        messages: Sequence[Dict[str, str]],
+        images: Sequence["ImageInput"],
+        videos: Sequence["VideoInput"],
+        processor: Optional["ProcessorMixin"],
+    ) -> List[Dict[str, str]]:
+        r"""
+        Pre-processes input messages before tokenization for VLMs.
+        """
+        self._validate_input(images, videos)
+        return messages
+
+    def process_token_ids(
+        self,
+        input_ids: List[int],
+        labels: Optional[List[int]],
+        images: Sequence["ImageInput"],
+        videos: Sequence["VideoInput"],
+        tokenizer: "PreTrainedTokenizer",
+        processor: Optional["ProcessorMixin"],
+    ) -> Tuple[List[int], Optional[List[int]]]:
+        r"""
+        Pre-processes token ids after tokenization for VLMs.
+        """
+        self._validate_input(images, videos)
+        return input_ids, labels
+
+    def get_mm_inputs(
+        self,
+        images: Sequence["ImageInput"],
+        videos: Sequence["VideoInput"],
+        imglens: Sequence[int],
+        vidlens: Sequence[int],
+        seqlens: Sequence[int],
+        processor: Optional["ProcessorMixin"],
+    ) -> Dict[str, Union[List[int], "torch.Tensor"]]:
+        r"""
+        Builds batched multimodal inputs for VLMs.
+        """
+        self._validate_input(images, videos)
+        return {}
+
+
+class LlavaPlugin(BasePlugin):
+    @override
+    def process_messages(
+        self,
+        messages: Sequence[Dict[str, str]],
+        images: Sequence["ImageInput"],
+        videos: Sequence["VideoInput"],
+        processor: Optional["ProcessorMixin"],
+    ) -> List[Dict[str, str]]:
+        self._validate_input(images, videos)
+        num_image_tokens = 0
+        image_seqlen = getattr(processor, "image_seqlen")
+        messages = deepcopy(messages)
+        for message in messages:
+            content = message["content"]
+            while IMAGE_PLACEHOLDER in content:
+                num_image_tokens += 1
+                content = content.replace(IMAGE_PLACEHOLDER, "{{image}}", 1)
+
+            message["content"] = content.replace("{{image}}", self.image_token * image_seqlen)
+
+        if len(images) != num_image_tokens:
+            raise ValueError("The number of images does not match the number of {} tokens".format(IMAGE_PLACEHOLDER))
+
+        return messages
+
+    @override
+    def get_mm_inputs(
+        self,
+        images: Sequence["ImageInput"],
+        videos: Sequence["VideoInput"],
+        imglens: Sequence[int],
+        vidlens: Sequence[int],
+        seqlens: Sequence[int],
+        processor: Optional["ProcessorMixin"],
+    ) -> Dict[str, Union[List[int], "torch.Tensor"]]:
+        self._validate_input(images, videos)
+        return self._get_mm_inputs(images, videos, processor)
+
+
+class LlavaNextPlugin(BasePlugin):
+    @override
+    def process_messages(
+        self,
+        messages: Sequence[Dict[str, str]],
+        images: Sequence["ImageInput"],
+        videos: Sequence["VideoInput"],
+        processor: Optional["ProcessorMixin"],
+    ) -> List[Dict[str, str]]:
+        self._validate_input(images, videos)
+        num_image_tokens = 0
+        messages = deepcopy(messages)
+        mm_inputs = self._get_mm_inputs(images, videos, processor)
+        if "image_sizes" in mm_inputs:
+            image_sizes = iter(mm_inputs["image_sizes"])
+        if "pixel_values" in mm_inputs:
+            height, width = get_image_size(to_numpy_array(mm_inputs["pixel_values"][0][0]))
+        for message in messages:
+            content = message["content"]
+            while self.image_token in content:
+                image_size = next(image_sizes)
+                orig_height, orig_width = image_size
+                image_seqlen = processor._get_number_of_features(orig_height, orig_width, height, width)
+                if processor.vision_feature_select_strategy == "default":
+                    image_seqlen -= 1
+                num_image_tokens += 1
+                content = content.replace(self.image_token, "{{image}}" * image_seqlen, 1)
+
+            message["content"] = content.replace("{{image}}", self.image_token)
+
+        if len(images) != num_image_tokens:
+            raise ValueError("The number of images does not match the number of {} tokens".format(IMAGE_PLACEHOLDER))
+        return messages
+
+    @override
+    def get_mm_inputs(
+        self,
+        images: Sequence["ImageInput"],
+        videos: Sequence["VideoInput"],
+        imglens: Sequence[int],
+        vidlens: Sequence[int],
+        seqlens: Sequence[int],
+        processor: Optional["ProcessorMixin"],
+    ) -> Dict[str, Union[List[int], "torch.Tensor"]]:
+        self._validate_input(images, videos)
+        res = self._get_mm_inputs(images, videos, processor)
+        return res
+
+
+class LlavaNextVideoPlugin(BasePlugin):
+    @override
+    def process_messages(
+        self,
+        messages: Sequence[Dict[str, str]],
+        images: Sequence["ImageInput"],
+        videos: Sequence["VideoInput"],
+        processor: Optional["ProcessorMixin"],
+    ) -> List[Dict[str, str]]:
+        self._validate_input(images, videos)
+        num_image_tokens = 0
+        num_video_tokens = 0
+        messages = deepcopy(messages)
+        mm_inputs = self._get_mm_inputs(images, videos, processor)
+        if "pixel_values" in mm_inputs:
+            image_sizes = iter(mm_inputs["image_sizes"])
+            height, width = get_image_size(to_numpy_array(mm_inputs["pixel_values"][0][0]))
+            for message in messages:
+                content = message["content"]
+
+                while self.image_token in content:
+                    image_size = next(image_sizes)
+                    orig_height, orig_width = image_size
+                    image_seqlen = processor._get_number_of_features(orig_height, orig_width, height, width)
+                    if processor.vision_feature_select_strategy == "default":
+                        image_seqlen -= 1
+                    num_image_tokens += 1
+                    content = content.replace(self.image_token, "{{image}}" * image_seqlen, 1)
+
+                message["content"] = content.replace("{{image}}", self.image_token)
+
+        if "pixel_values_videos" in mm_inputs:
+            pixel_values_video = to_numpy_array(mm_inputs.get("pixel_values_videos")[0])
+            height, width = get_image_size(pixel_values_video[0])
+            num_frames = pixel_values_video.shape[0]  # frame dim is always after batch dim
+            image_seqlen = (height // processor.patch_size) * (width // processor.patch_size)
+            video_seqlen = image_seqlen // 4 * num_frames  # divide by 4 needed for avg pooling layer
+
+            for message in messages:
+                content = message["content"]
+                while self.video_token in content:
+                    num_video_tokens += 1
+                    content = content.replace(self.video_token, "{{video}}", 1)
+                message["content"] = content.replace("{{video}}", self.video_token * video_seqlen)
+
+        if len(images) != num_image_tokens:
+            raise ValueError("The number of images does not match the number of {} tokens".format(IMAGE_PLACEHOLDER))
+
+        if len(videos) != num_video_tokens:
+            raise ValueError("The number of videos does not match the number of {} tokens".format(IMAGE_PLACEHOLDER))
+
+        return messages
+
+    @override
+    def get_mm_inputs(
+        self,
+        images: Sequence["ImageInput"],
+        videos: Sequence["VideoInput"],
+        imglens: Sequence[int],
+        vidlens: Sequence[int],
+        seqlens: Sequence[int],
+        processor: Optional["ProcessorMixin"],
+    ) -> Dict[str, Union[List[int], "torch.Tensor"]]:
+        self._validate_input(images, videos)
+        return self._get_mm_inputs(images, videos, processor)
+
+
+class PaliGemmaPlugin(BasePlugin):
+    @override
+    def process_messages(
+        self,
+        messages: Sequence[Dict[str, str]],
+        images: Sequence["ImageInput"],
+        videos: Sequence["VideoInput"],
+        processor: Optional["ProcessorMixin"],
+    ) -> List[Dict[str, str]]:
+        self._validate_input(images, videos)
+        num_image_tokens = 0
+        messages = deepcopy(messages)
+        for message in messages:
+            content = message["content"]
+            while IMAGE_PLACEHOLDER in content:
+                num_image_tokens += 1
+                content = content.replace(IMAGE_PLACEHOLDER, "{{image}}", 1)
+
+            message["content"] = content.replace("{{image}}", "")
+
+        if len(images) != num_image_tokens:
+            raise ValueError("The number of images does not match the number of {} tokens".format(IMAGE_PLACEHOLDER))
+
+        return messages
+
+    @override
+    def process_token_ids(
+        self,
+        input_ids: List[int],
+        labels: Optional[List[int]],
+        images: Sequence["ImageInput"],
+        videos: Sequence["VideoInput"],
+        tokenizer: "PreTrainedTokenizer",
+        processor: Optional["ProcessorMixin"],
+    ) -> Tuple[List[int], Optional[List[int]]]:
+        self._validate_input(images, videos)
+        num_images = len(images)
+        image_seqlen = num_images * getattr(processor, "image_seqlen")
+        image_token_id = tokenizer.convert_tokens_to_ids(self.image_token)
+        input_ids = [image_token_id] * image_seqlen + input_ids
+        if labels is not None:
+            labels = [IGNORE_INDEX] * image_seqlen + labels
+
+        return input_ids, labels
+
+    @override
+    def get_mm_inputs(
+        self,
+        images: Sequence["ImageInput"],
+        videos: Sequence["VideoInput"],
+        imglens: Sequence[int],
+        vidlens: Sequence[int],
+        seqlens: Sequence[int],
+        processor: Optional["ProcessorMixin"],
+    ) -> Dict[str, Union[List[int], "torch.Tensor"]]:
+        self._validate_input(images, videos)
+        mm_inputs = self._get_mm_inputs(images, videos, processor)
+        mm_inputs["token_type_ids"] = _get_paligemma_token_type_ids(imglens, seqlens, processor)
+        return mm_inputs
+
+
+class Qwen2vlPlugin(BasePlugin):
+    @override
+    def _preprocess_image(self, image: "ImageObject", **kwargs) -> "ImageObject":
+        image = super()._preprocess_image(image, **kwargs)
+        if min(image.width, image.height) < 28:
+            width, height = max(image.width, 28), max(image.height, 28)
+            image = image.resize((width, height), resample=Image.NEAREST)
+
+        if image.width / image.height > 200:
+            width, height = image.height * 180, image.height
+            image = image.resize((width, height), resample=Image.NEAREST)
+
+        if image.height / image.width > 200:
+            width, height = image.width, image.width * 180
+            image = image.resize((width, height), resample=Image.NEAREST)
+
+        return image
+
+    @override
+    def _get_video_sample_frames(self, video_stream: "Stream", **kwargs) -> int:
+        sample_frames = super()._get_video_sample_frames(video_stream, **kwargs)
+        sample_frames = sample_frames // 2 * 2
+        return sample_frames
+
+    @override
+    def process_messages(
+        self,
+        messages: Sequence[Dict[str, str]],
+        images: Sequence["ImageInput"],
+        videos: Sequence["VideoInput"],
+        processor: Optional["ProcessorMixin"],
+    ) -> List[Dict[str, str]]:
+        self._validate_input(images, videos)
+        image_processor: "BaseImageProcessor" = getattr(processor, "image_processor")
+        merge_length: int = getattr(image_processor, "merge_size") ** 2
+        mm_inputs = self._get_mm_inputs(images, videos, processor)
+        image_grid_thw = mm_inputs.get("image_grid_thw", [])
+        video_grid_thw = mm_inputs.get("video_grid_thw", [])
+
+        num_image_tokens, num_video_tokens = 0, 0
+        messages = deepcopy(messages)
+        for message in messages:
+            content = message["content"]
+            while IMAGE_PLACEHOLDER in content:
+                if num_image_tokens >= len(image_grid_thw):
+                    raise ValueError("`len(images)` is less than the number of {} tokens.".format(IMAGE_PLACEHOLDER))
+
+                content = content.replace(
+                    IMAGE_PLACEHOLDER,
+                    "<|vision_start|>{}<|vision_end|>".format(
+                        self.image_token * (image_grid_thw[num_image_tokens].prod() // merge_length)
+                    ),
+                    1,
+                )
+                num_image_tokens += 1
+
+            while VIDEO_PLACEHOLDER in content:
+                if num_video_tokens >= len(video_grid_thw):
+                    raise ValueError("`len(videos)` is less than the number of {} tokens.".format(VIDEO_PLACEHOLDER))
+
+                content = content.replace(
+                    VIDEO_PLACEHOLDER,
+                    "<|vision_start|>{}<|vision_end|>".format(
+                        self.video_token * (video_grid_thw[num_video_tokens].prod() // merge_length)
+                    ),
+                    1,
+                )
+                num_video_tokens += 1
+
+            message["content"] = content
+
+        if len(images) != num_image_tokens:
+            raise ValueError("The number of images does not match the number of {} tokens".format(IMAGE_PLACEHOLDER))
+
+        if len(videos) != num_video_tokens:
+            raise ValueError("The number of videos does not match the number of {} tokens".format(VIDEO_PLACEHOLDER))
+
+        return messages
+
+    @override
+    def get_mm_inputs(
+        self,
+        images: Sequence["ImageInput"],
+        videos: Sequence["VideoInput"],
+        imglens: Sequence[int],
+        vidlens: Sequence[int],
+        seqlens: Sequence[int],
+        processor: Optional["ProcessorMixin"],
+    ) -> Dict[str, Union[List[int], "torch.Tensor"]]:
+        self._validate_input(images, videos)
+        return self._get_mm_inputs(images, videos, processor)
+
+
+class VideoLlavaPlugin(BasePlugin):
+    @override
+    def process_messages(
+        self,
+        messages: Sequence[Dict[str, str]],
+        images: Sequence["ImageInput"],
+        videos: Sequence["VideoInput"],
+        processor: Optional["ProcessorMixin"],
+    ) -> List[Dict[str, str]]:
+        self._validate_input(images, videos)
+        num_image_tokens = 0
+        num_video_tokens = 0
+        messages = deepcopy(messages)
+        mm_inputs = self._get_mm_inputs(images, videos, processor)
+        num_frames = 0
+        exist_images = "pixel_values_images" in mm_inputs
+        exist_videos = "pixel_values_videos" in mm_inputs
+        if exist_videos or exist_images:
+            if exist_images:
+                height, width = get_image_size(to_numpy_array(mm_inputs.get("pixel_values_images")[0]))
+                num_frames = 1
+            if exist_videos:
+                pixel_values_video = to_numpy_array(mm_inputs.get("pixel_values_videos")[0])
+                height, width = get_image_size(pixel_values_video[0])
+                num_frames = pixel_values_video.shape[0]  # frame dim is always after batch dim
+            image_seqlen = (height // processor.patch_size) * (width // processor.patch_size) + 1
+            video_seqlen = image_seqlen * num_frames
+            if processor.vision_feature_select_strategy == "default":
+                image_seqlen -= 1
+            for message in messages:
+                content = message["content"]
+                while self.image_token in content:
+                    num_image_tokens += 1
+                    content = content.replace(self.image_token, "{{image}}", 1)
+                while self.video_token in content:
+                    num_video_tokens += 1
+                    content = content.replace(self.video_token, "{{video}}", 1)
+
+                content = content.replace("{{image}}", self.image_token * image_seqlen)
+                message["content"] = content.replace("{{video}}", self.video_token * video_seqlen)
+
+        if len(images) != num_image_tokens:
+            raise ValueError("The number of images does not match the number of {} tokens".format(self.image_token))
+
+        if len(videos) != num_video_tokens:
+            raise ValueError("The number of videos does not match the number of {} tokens".format(self.video_token))
+
+        return messages
+
+    @override
+    def get_mm_inputs(
+        self,
+        images: Sequence["ImageInput"],
+        videos: Sequence["VideoInput"],
+        imglens: Sequence[int],
+        vidlens: Sequence[int],
+        seqlens: Sequence[int],
+        processor: Optional["ProcessorMixin"],
+    ) -> Dict[str, Union[List[int], "torch.Tensor"]]:
+        self._validate_input(images, videos)
+        return self._get_mm_inputs(images, videos, processor)
+
+
+PLUGINS = {
+    "base": BasePlugin,
+    "llava": LlavaPlugin,
+    "llava_next": LlavaNextPlugin,
+    "llava_next_video": LlavaNextVideoPlugin,
+    "paligemma": PaliGemmaPlugin,
+    "qwen2_vl": Qwen2vlPlugin,
+    "video_llava": VideoLlavaPlugin,
+}
+
+
+def get_mm_plugin(
+    name: str,
+    image_token: Optional[str] = None,
+    video_token: Optional[str] = None,
+) -> "BasePlugin":
+    plugin_class = PLUGINS.get(name, None)
+    if plugin_class is None:
+        raise ValueError("Multimodal plugin `{}` not found.".format(name))
+
+    return plugin_class(image_token, video_token)

llamafactory/data/parser.py

@@ -0,0 +1,154 @@
+# Copyright 2024 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 json
+import os
+from dataclasses import dataclass
+from typing import Any, Dict, List, Literal, Optional, Sequence
+
+from transformers.utils import cached_file
+
+from ..extras.constants import DATA_CONFIG
+from ..extras.misc import use_modelscope
+
+
+@dataclass
+class DatasetAttr:
+    r"""
+    Dataset attributes.
+    """
+
+    # basic configs
+    load_from: Literal["hf_hub", "ms_hub", "script", "file"]
+    dataset_name: str
+    formatting: Literal["alpaca", "sharegpt"] = "alpaca"
+    ranking: bool = False
+    # extra configs
+    subset: Optional[str] = None
+    split: str = "train"
+    folder: Optional[str] = None
+    num_samples: Optional[int] = None
+    # common columns
+    system: Optional[str] = None
+    tools: Optional[str] = None
+    images: Optional[str] = None
+    videos: Optional[str] = None
+    # rlhf columns
+    chosen: Optional[str] = None
+    rejected: Optional[str] = None
+    kto_tag: Optional[str] = None
+    # alpaca columns
+    prompt: Optional[str] = "instruction"
+    query: Optional[str] = "input"
+    response: Optional[str] = "output"
+    history: Optional[str] = None
+    # sharegpt columns
+    messages: Optional[str] = "conversations"
+    # sharegpt tags
+    role_tag: Optional[str] = "from"
+    content_tag: Optional[str] = "value"
+    user_tag: Optional[str] = "human"
+    assistant_tag: Optional[str] = "gpt"
+    observation_tag: Optional[str] = "observation"
+    function_tag: Optional[str] = "function_call"
+    system_tag: Optional[str] = "system"
+
+    def __repr__(self) -> str:
+        return self.dataset_name
+
+    def set_attr(self, key: str, obj: Dict[str, Any], default: Optional[Any] = None) -> None:
+        setattr(self, key, obj.get(key, default))
+
+
+def get_dataset_list(dataset_names: Optional[Sequence[str]], dataset_dir: str) -> List["DatasetAttr"]:
+    r"""
+    Gets the attributes of the datasets.
+    """
+    if dataset_names is None:
+        dataset_names = []
+
+    if dataset_dir == "ONLINE":
+        dataset_info = None
+    else:
+        if dataset_dir.startswith("REMOTE:"):
+            config_path = cached_file(path_or_repo_id=dataset_dir[7:], filename=DATA_CONFIG, repo_type="dataset")
+        else:
+            config_path = os.path.join(dataset_dir, DATA_CONFIG)
+
+        try:
+            with open(config_path, "r") as f:
+                dataset_info = json.load(f)
+        except Exception as err:
+            if len(dataset_names) != 0:
+                raise ValueError("Cannot open {} due to {}.".format(config_path, str(err)))
+
+            dataset_info = None
+
+    dataset_list: List["DatasetAttr"] = []
+    for name in dataset_names:
+        if dataset_info is None:  # dataset_dir is ONLINE
+            load_from = "ms_hub" if use_modelscope() else "hf_hub"
+            dataset_attr = DatasetAttr(load_from, dataset_name=name)
+            dataset_list.append(dataset_attr)
+            continue
+
+        if name not in dataset_info:
+            raise ValueError("Undefined dataset {} in {}.".format(name, DATA_CONFIG))
+
+        has_hf_url = "hf_hub_url" in dataset_info[name]
+        has_ms_url = "ms_hub_url" in dataset_info[name]
+
+        if has_hf_url or has_ms_url:
+            if (use_modelscope() and has_ms_url) or (not has_hf_url):
+                dataset_attr = DatasetAttr("ms_hub", dataset_name=dataset_info[name]["ms_hub_url"])
+            else:
+                dataset_attr = DatasetAttr("hf_hub", dataset_name=dataset_info[name]["hf_hub_url"])
+        elif "script_url" in dataset_info[name]:
+            dataset_attr = DatasetAttr("script", dataset_name=dataset_info[name]["script_url"])
+        else:
+            dataset_attr = DatasetAttr("file", dataset_name=dataset_info[name]["file_name"])
+
+        dataset_attr.set_attr("formatting", dataset_info[name], default="alpaca")
+        dataset_attr.set_attr("ranking", dataset_info[name], default=False)
+        dataset_attr.set_attr("subset", dataset_info[name])
+        dataset_attr.set_attr("split", dataset_info[name], default="train")
+        dataset_attr.set_attr("folder", dataset_info[name])
+        dataset_attr.set_attr("num_samples", dataset_info[name])
+
+        if "columns" in dataset_info[name]:
+            column_names = ["system", "tools", "images", "videos", "chosen", "rejected", "kto_tag"]
+            if dataset_attr.formatting == "alpaca":
+                column_names.extend(["prompt", "query", "response", "history"])
+            else:
+                column_names.extend(["messages"])
+
+            for column_name in column_names:
+                dataset_attr.set_attr(column_name, dataset_info[name]["columns"])
+
+        if dataset_attr.formatting == "sharegpt" and "tags" in dataset_info[name]:
+            tag_names = (
+                "role_tag",
+                "content_tag",
+                "user_tag",
+                "assistant_tag",
+                "observation_tag",
+                "function_tag",
+                "system_tag",
+            )
+            for tag in tag_names:
+                dataset_attr.set_attr(tag, dataset_info[name]["tags"])
+
+        dataset_list.append(dataset_attr)
+
+    return dataset_list

llamafactory/data/preprocess.py

@@ -0,0 +1,111 @@
+# Copyright 2024 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 functools import partial
+from typing import TYPE_CHECKING, Callable, Literal, Optional, Tuple
+
+from .processors.feedback import preprocess_feedback_dataset
+from .processors.pairwise import preprocess_pairwise_dataset, print_pairwise_dataset_example
+from .processors.pretrain import preprocess_pretrain_dataset
+from .processors.supervised import (
+    preprocess_packed_supervised_dataset,
+    preprocess_supervised_dataset,
+    print_supervised_dataset_example,
+)
+from .processors.unsupervised import preprocess_unsupervised_dataset, print_unsupervised_dataset_example
+
+
+if TYPE_CHECKING:
+    from transformers import PreTrainedTokenizer, ProcessorMixin
+
+    from ..hparams import DataArguments
+    from .template import Template
+
+
+def get_preprocess_and_print_func(
+    data_args: "DataArguments",
+    stage: Literal["pt", "sft", "rm", "ppo", "kto"],
+    template: "Template",
+    tokenizer: "PreTrainedTokenizer",
+    processor: Optional["ProcessorMixin"],
+    do_generate: bool = False,
+) -> Tuple[Callable, Callable]:
+    if stage == "pt":
+        preprocess_func = partial(
+            preprocess_pretrain_dataset,
+            tokenizer=tokenizer,
+            data_args=data_args,
+        )
+        print_function = partial(print_unsupervised_dataset_example, tokenizer=tokenizer)
+    elif stage == "sft" and not do_generate:
+        if data_args.packing:
+            if data_args.neat_packing:  # hack datasets to have int32 attention mask
+                from datasets.arrow_writer import OptimizedTypedSequence, TypedSequence
+
+                def __init__(self, data, **kwargs):
+                    return TypedSequence.__init__(
+                        self,
+                        data,
+                        type=kwargs.pop("type", None),
+                        try_type=kwargs.pop("try_type", None),
+                        optimized_int_type=kwargs.pop("optimized_int_type", None),
+                    )
+
+                OptimizedTypedSequence.__init__ = __init__
+            preprocess_func = partial(
+                preprocess_packed_supervised_dataset,
+                template=template,
+                tokenizer=tokenizer,
+                processor=processor,
+                data_args=data_args,
+            )
+        else:
+            preprocess_func = partial(
+                preprocess_supervised_dataset,
+                template=template,
+                tokenizer=tokenizer,
+                processor=processor,
+                data_args=data_args,
+            )
+
+        print_function = partial(print_supervised_dataset_example, tokenizer=tokenizer)
+    elif stage == "rm":
+        preprocess_func = partial(
+            preprocess_pairwise_dataset,
+            template=template,
+            tokenizer=tokenizer,
+            processor=processor,
+            data_args=data_args,
+        )
+        print_function = partial(print_pairwise_dataset_example, tokenizer=tokenizer)
+    elif stage == "kto":
+        preprocess_func = partial(
+            preprocess_feedback_dataset,
+            template=template,
+            tokenizer=tokenizer,
+            processor=processor,
+            data_args=data_args,
+        )
+        print_function = partial(print_supervised_dataset_example, tokenizer=tokenizer)
+    else:
+        preprocess_func = partial(
+            preprocess_unsupervised_dataset,
+            template=template,
+            tokenizer=tokenizer,
+            processor=processor,
+            data_args=data_args,
+        )
+        print_function = partial(print_unsupervised_dataset_example, tokenizer=tokenizer)
+
+    return preprocess_func, print_function