release v0.7.1

Former-commit-id: a4f8adb021b6218d624303b51cd5e93ffa3111a1

.github/SECURITY.md

@@ -1,6 +1,6 @@
 # Reporting Security Issues
 
-To report a security issue, please use the GitHub Security Advisory ["Report a Vulnerability"](https://github.com/electron/electron/security/advisories/new) tab.
+To report a security issue, please use the GitHub Security Advisory ["Report a Vulnerability"](https://github.com/hiyouga/LLaMA-Factory/security/advisories/new) tab.
 
 We will send a response indicating the next steps in handling your report. After the initial reply to your report, the security team will keep you informed of the progress towards a fix and full announcement, and may ask for additional information or guidance.
 

Dockerfile

@@ -11,4 +11,4 @@ RUN pip install -e .[deepspeed,metrics,bitsandbytes,qwen]
 VOLUME [ "/root/.cache/huggingface/", "/app/data", "/app/output" ]
 EXPOSE 7860
 
-CMD [ "python", "src/train_web.py" ]
+CMD [ "llamafactory-cli", "webui" ]

README.md

@@ -5,7 +5,7 @@
 [![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/llmtuner)](https://pypi.org/project/llmtuner/)
 [![Downloads](https://static.pepy.tech/badge/llmtuner)](https://pypi.org/project/llmtuner/)
-[![Citation](https://img.shields.io/badge/citation-27-green)](#projects-using-llama-factory)
+[![Citation](https://img.shields.io/badge/citation-44-green)](#projects-using-llama-factory)
 [![GitHub pull request](https://img.shields.io/badge/PRs-welcome-blue)](https://github.com/hiyouga/LLaMA-Factory/pulls)
 [![Discord](https://dcbadge.vercel.app/api/server/rKfvV9r9FK?compact=true&style=flat)](https://discord.gg/rKfvV9r9FK)
 [![Twitter](https://img.shields.io/twitter/follow/llamafactory_ai)](https://twitter.com/llamafactory_ai)
@@ -13,6 +13,8 @@
 [![Studios](https://img.shields.io/badge/ModelScope-Open%20in%20Studios-blue)](https://modelscope.cn/studios/hiyouga/LLaMA-Board)
 [![Open in Colab](https://colab.research.google.com/assets/colab-badge.svg)](https://colab.research.google.com/drive/1eRTPn37ltBbYsISy9Aw2NuI2Aq5CQrD9?usp=sharing)
 
+[![GitHub Tread](https://trendshift.io/api/badge/repositories/4535)](https://trendshift.io/repositories/4535)
+
 👋 Join our [WeChat](assets/wechat.jpg).
 
 \[ English | [中文](README_zh.md) \]
@@ -43,17 +45,17 @@ Choose your path:
 
 ## Features
 
-- **Various models**: LLaMA, Mistral, Mixtral-MoE, Qwen, Yi, Gemma, Baichuan, ChatGLM, Phi, etc.
+- **Various models**: LLaMA, LLaVA, Mistral, Mixtral-MoE, Qwen, Yi, Gemma, Baichuan, ChatGLM, Phi, etc.
-- **Integrated methods**: (Continuous) pre-training, supervised fine-tuning, reward modeling, PPO and DPO.
+- **Integrated methods**: (Continuous) pre-training, (multimodal) supervised fine-tuning, reward modeling, PPO, DPO and ORPO.
 - **Scalable resources**: 32-bit full-tuning, 16-bit freeze-tuning, 16-bit LoRA and 2/4/8-bit QLoRA via AQLM/AWQ/GPTQ/LLM.int8.
-- **Advanced algorithms**: GaLore, DoRA, LongLoRA, LLaMA Pro, LoRA+, LoftQ and Agent tuning.
+- **Advanced algorithms**: GaLore, BAdam, DoRA, LongLoRA, LLaMA Pro, Mixture-of-Depths, LoRA+, LoftQ and Agent tuning.
 - **Practical tricks**: FlashAttention-2, Unsloth, RoPE scaling, NEFTune and rsLoRA.
 - **Experiment monitors**: LlamaBoard, TensorBoard, Wandb, MLflow, etc.
 - **Faster inference**: OpenAI-style API, Gradio UI and CLI with vLLM worker.
 
 ## 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.
+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)
 
@@ -62,51 +64,67 @@ Compared to ChatGLM's [P-Tuning](https://github.com/THUDM/ChatGLM2-6B/tree/main/
 - **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.
+- We adopt `pre_seq_len=128` for ChatGLM's P-Tuning and `lora_rank=32` for LLaMA Factory's LoRA tuning.
 
 </details>
 
 ## Changelog
 
-[24/03/21] Our paper "[LlamaFactory: Unified Efficient Fine-Tuning of 100+ Language Models](https://arxiv.org/abs/2403.13372)" is available at arXiv!
+[24/05/14] We supported training and inference on the Ascend NPU devices. Check [installation](#installation) section for details.
 
-[24/03/20] We supported **FSDP+QLoRA** that fine-tunes a 70B model on 2x24GB GPUs. See `examples/fsdp_qlora` for usage.
+[24/05/13] We supported fine-tuning the **Yi-1.5** series models.
 
-[24/03/13] We supported **[LoRA+](https://arxiv.org/abs/2402.12354)**. See `examples/extras/loraplus` for usage.
+[24/04/26] We supported fine-tuning the **LLaVA-1.5** multimodal LLMs. See [examples](examples/README.md) for usage.
-
-[24/03/07] We supported gradient low-rank projection (**[GaLore](https://arxiv.org/abs/2403.03507)**) algorithm. See `examples/extras/galore` for usage.
 
 <details><summary>Full Changelog</summary>
 
-[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. (LoRA is not yet supported, merge it first.)
+[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.
 
-[24/02/28] We supported weight-decomposed LoRA (**[DoRA](https://arxiv.org/abs/2402.09353)**). Try `--use_dora` to activate DoRA training.
+[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.
 
-[24/02/15] We supported **block expansion** proposed by [LLaMA Pro](https://github.com/TencentARC/LLaMA-Pro). See `examples/extras/llama_pro` for usage.
+[24/04/16] We supported **[BAdam](https://arxiv.org/abs/2404.02827)**. See [examples](examples/README.md) for usage.
+
+[24/04/16] We supported **[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 gradient low-rank projection (**[GaLore](https://arxiv.org/abs/2403.03507)**) algorithm. 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`.
+[24/01/18] We supported **agent tuning** for most models, equipping model with tool using abilities by fine-tuning with `dataset: glaive_toolcall`.
 
-[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` 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.
+[23/12/23] 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](#use-modelscope-hub-optional) for usage.
+[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` argument to activate NEFTune, e.g., `--neftune_noise_alpha 5`.
+[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` argument to enable shift short attention.
+[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 [this example](#evaluation) to evaluate your models.
+[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` argument to enable FlashAttention-2 if you are using RTX4090, A100 or H100 GPUs.
+[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.
+[23/08/12] 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 [this example](#dpo-training) to train your models.
+[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` and `--max_steps 10000` arguments to load your dataset in streaming mode.
+[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.
 
@@ -118,39 +136,45 @@ Compared to ChatGLM's [P-Tuning](https://github.com/THUDM/ChatGLM2-6B/tree/main/
 
 [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)**). Try `--quantization_bit 4/8` argument to work with quantized models.
+[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                  | Default module    | Template  |
+| Model                                                    | Model size                       | Default module    | Template  |
-| -------------------------------------------------------- | --------------------------- | ----------------- | --------- |
+| -------------------------------------------------------- | -------------------------------- | ----------------- | --------- |
-| [Baichuan2](https://huggingface.co/baichuan-inc)         | 7B/13B                      | W_pack            | baichuan2 |
+| [Baichuan2](https://huggingface.co/baichuan-inc)         | 7B/13B                           | W_pack            | baichuan2 |
-| [BLOOM](https://huggingface.co/bigscience/bloom)         | 560M/1.1B/1.7B/3B/7.1B/176B | query_key_value   | -         |
+| [BLOOM](https://huggingface.co/bigscience)               | 560M/1.1B/1.7B/3B/7.1B/176B      | query_key_value   | -         |
-| [BLOOMZ](https://huggingface.co/bigscience/bloomz)       | 560M/1.1B/1.7B/3B/7.1B/176B | query_key_value   | -         |
+| [BLOOMZ](https://huggingface.co/bigscience)              | 560M/1.1B/1.7B/3B/7.1B/176B      | query_key_value   | -         |
-| [ChatGLM3](https://huggingface.co/THUDM/chatglm3-6b)     | 6B                          | query_key_value   | chatglm3  |
+| [ChatGLM3](https://huggingface.co/THUDM)                 | 6B                               | query_key_value   | chatglm3  |
-| [DeepSeek (MoE)](https://huggingface.co/deepseek-ai)     | 7B/16B/67B                  | q_proj,v_proj     | deepseek  |
+| [Command-R](https://huggingface.co/CohereForAI)          | 35B/104B                         | q_proj,v_proj     | cohere    |
-| [Falcon](https://huggingface.co/tiiuae)                  | 7B/40B/180B                 | query_key_value   | falcon    |
+| [DeepSeek (MoE)](https://huggingface.co/deepseek-ai)     | 7B/16B/67B/236B                  | q_proj,v_proj     | deepseek  |
-| [Gemma](https://huggingface.co/google)                   | 2B/7B                       | q_proj,v_proj     | gemma     |
+| [Falcon](https://huggingface.co/tiiuae)                  | 7B/40B/180B                      | query_key_value   | falcon    |
-| [InternLM2](https://huggingface.co/internlm)             | 7B/20B                      | wqkv              | intern2   |
+| [Gemma/CodeGemma](https://huggingface.co/google)         | 2B/7B                            | q_proj,v_proj     | gemma     |
-| [LLaMA](https://github.com/facebookresearch/llama)       | 7B/13B/33B/65B              | q_proj,v_proj     | -         |
+| [InternLM2](https://huggingface.co/internlm)             | 7B/20B                           | wqkv              | intern2   |
-| [LLaMA-2](https://huggingface.co/meta-llama)             | 7B/13B/70B                  | q_proj,v_proj     | llama2    |
+| [LLaMA](https://github.com/facebookresearch/llama)       | 7B/13B/33B/65B                   | q_proj,v_proj     | -         |
-| [Mistral](https://huggingface.co/mistralai)              | 7B                          | q_proj,v_proj     | mistral   |
+| [LLaMA-2](https://huggingface.co/meta-llama)             | 7B/13B/70B                       | q_proj,v_proj     | llama2    |
-| [Mixtral](https://huggingface.co/mistralai)              | 8x7B                        | q_proj,v_proj     | mistral   |
+| [LLaMA-3](https://huggingface.co/meta-llama)             | 8B/70B                           | q_proj,v_proj     | llama3    |
-| [OLMo](https://huggingface.co/allenai)                   | 1B/7B                       | att_proj          | olmo      |
+| [LLaVA-1.5](https://huggingface.co/llava-hf)             | 7B/13B                           | q_proj,v_proj     | vicuna    |
-| [Phi-1.5/2](https://huggingface.co/microsoft)            | 1.3B/2.7B                   | q_proj,v_proj     | -         |
+| [Mistral/Mixtral](https://huggingface.co/mistralai)      | 7B/8x7B/8x22B                    | q_proj,v_proj     | mistral   |
-| [Qwen](https://huggingface.co/Qwen)                      | 1.8B/7B/14B/72B             | c_attn            | qwen      |
+| [OLMo](https://huggingface.co/allenai)                   | 1B/7B                            | q_proj,v_proj     | -         |
-| [Qwen1.5](https://huggingface.co/Qwen)                   | 0.5B/1.8B/4B/7B/14B/72B     | q_proj,v_proj     | qwen      |
+| [Phi-1.5/2](https://huggingface.co/microsoft)            | 1.3B/2.7B                        | q_proj,v_proj     | -         |
-| [StarCoder2](https://huggingface.co/bigcode)             | 3B/7B/15B                   | q_proj,v_proj     | -         |
+| [Phi-3](https://huggingface.co/microsoft)                | 3.8B                             | qkv_proj          | phi       |
-| [XVERSE](https://huggingface.co/xverse)                  | 7B/13B/65B                  | q_proj,v_proj     | xverse    |
+| [Qwen](https://huggingface.co/Qwen)                      | 1.8B/7B/14B/72B                  | c_attn            | qwen      |
-| [Yi](https://huggingface.co/01-ai)                       | 6B/9B/34B                   | q_proj,v_proj     | yi        |
+| [Qwen1.5 (Code/MoE)](https://huggingface.co/Qwen)        | 0.5B/1.8B/4B/7B/14B/32B/72B/110B | q_proj,v_proj     | qwen      |
-| [Yuan](https://huggingface.co/IEITYuan)                  | 2B/51B/102B                 | q_proj,v_proj     | yuan      |
+| [StarCoder2](https://huggingface.co/bigcode)             | 3B/7B/15B                        | q_proj,v_proj     | -         |
+| [XVERSE](https://huggingface.co/xverse)                  | 7B/13B/65B                       | q_proj,v_proj     | xverse    |
+| [Yi (1/1.5)](https://huggingface.co/01-ai)               | 6B/9B/34B                        | q_proj,v_proj     | yi        |
+| [Yi-VL](https://huggingface.co/01-ai)                    | 6B/34B                           | q_proj,v_proj     | yi_vl     |
+| [Yuan](https://huggingface.co/IEITYuan)                  | 2B/51B/102B                      | q_proj,v_proj     | yuan      |
 
 > [!NOTE]
-> **Default module** is used for the `--lora_target` argument, you can use `--lora_target all` to specify all the available modules.
+> **Default module** is used for the `--lora_target` argument, you can use `--lora_target all` to specify all the available modules for better convergence.
 >
-> 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 "chat" models.
+> 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/llmtuner/extras/constants.py) for a full list of models we supported.
 
@@ -165,9 +189,7 @@ You also can add a custom chat template to [template.py](src/llmtuner/data/templ
 | 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: |
-
+| ORPO Training          | :white_check_mark: | :white_check_mark: | :white_check_mark: | :white_check_mark: |
-> [!NOTE]
-> Use `--quantization_bit 4` argument to enable QLoRA.
 
 ## Provided Datasets
 
@@ -190,8 +212,8 @@ You also can add a custom chat template to [template.py](src/llmtuner/data/templ
 - [Stanford Alpaca (en)](https://github.com/tatsu-lab/stanford_alpaca)
 - [Stanford Alpaca (zh)](https://github.com/ymcui/Chinese-LLaMA-Alpaca)
 - [Alpaca GPT4 (en&zh)](https://github.com/Instruction-Tuning-with-GPT-4/GPT-4-LLM)
-- [Self Cognition (zh)](data/self_cognition.json)
+- [Identity (en&zh)](data/identity.json)
-- [Open Assistant (multilingual)](https://huggingface.co/datasets/OpenAssistant/oasst1)
+- [Open Assistant (zh)](https://huggingface.co/datasets/OpenAssistant/oasst1)
 - [ShareGPT (zh)](https://huggingface.co/datasets/QingyiSi/Alpaca-CoT/tree/main/Chinese-instruction-collection)
 - [Guanaco Dataset (multilingual)](https://huggingface.co/datasets/JosephusCheung/GuanacoDataset)
 - [BELLE 2M (zh)](https://huggingface.co/datasets/BelleGroup/train_2M_CN)
@@ -223,6 +245,7 @@ You also can add a custom chat template to [template.py](src/llmtuner/data/templ
 - [Evol Instruct V2 (en)](https://huggingface.co/datasets/WizardLM/WizardLM_evol_instruct_V2_196k)
 - [Glaive Function Calling V2 (en)](https://huggingface.co/datasets/glaiveai/glaive-function-calling-v2)
 - [Cosmopedia (en)](https://huggingface.co/datasets/HuggingFaceTB/cosmopedia)
+- [LLaVA mixed (en&zh)](https://huggingface.co/datasets/BUAADreamer/llava-en-zh-300k)
 - [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)
@@ -238,16 +261,15 @@ You also can add a custom chat template to [template.py](src/llmtuner/data/templ
 <details><summary>Preference datasets</summary>
 
 - [HH-RLHF (en)](https://huggingface.co/datasets/Anthropic/hh-rlhf)
-- [Open Assistant (multilingual)](https://huggingface.co/datasets/OpenAssistant/oasst1)
 - [GPT-4 Generated Data (en&zh)](https://github.com/Instruction-Tuning-with-GPT-4/GPT-4-LLM)
 - [Orca DPO (en)](https://huggingface.co/datasets/Intel/orca_dpo_pairs)
 - [Nectar (en)](https://huggingface.co/datasets/berkeley-nest/Nectar)
+- [DPO mixed (en&zh)](https://huggingface.co/datasets/hiyouga/DPO-En-Zh-20k)
+- [Open Assistant (zh)](https://huggingface.co/datasets/OpenAssistant/oasst1)
 - [Orca DPO (de)](https://huggingface.co/datasets/mayflowergmbh/intel_orca_dpo_pairs_de)
 
 </details>
 
-Please refer to [data/README.md](data/README.md) for details.
-
 Some datasets require confirmation before using them, so we recommend logging in with your Hugging Face account using these commands.
 
 ```bash
@@ -261,54 +283,55 @@ huggingface-cli login
 | ------------ | ------- | --------- |
 | python       | 3.8     | 3.10      |
 | torch        | 1.13.1  | 2.2.0     |
-| transformers | 4.37.2  | 4.39.1    |
+| transformers | 4.37.2  | 4.40.1    |
-| datasets     | 2.14.3  | 2.17.1    |
+| datasets     | 2.14.3  | 2.19.1    |
-| accelerate   | 0.27.2  | 0.28.0    |
+| accelerate   | 0.27.2  | 0.30.0    |
 | peft         | 0.9.0   | 0.10.0    |
-| trl          | 0.8.1   | 0.8.1     |
+| trl          | 0.8.1   | 0.8.6     |
 
 | Optional     | Minimum | Recommend |
 | ------------ | ------- | --------- |
 | CUDA         | 11.6    | 12.2      |
 | deepspeed    | 0.10.0  | 0.14.0    |
-| bitsandbytes | 0.39.0  | 0.43.0    |
+| bitsandbytes | 0.39.0  | 0.43.1    |
-| flash-attn   | 2.3.0   | 2.5.6     |
+| vllm         | 0.4.0   | 0.4.2     |
+| flash-attn   | 2.3.0   | 2.5.8     |
 
 ### Hardware Requirement
 
 \* *estimated*
 
-| Method | Bits |   7B  |  13B  |  30B  |   70B  |   8x7B |
+| Method            | Bits |   7B  |  13B  |  30B  |   70B  |  110B  |  8x7B |  8x22B |
-| ------ | ---- | ----- | ----- | ----- | ------ | ------ |
+| ----------------- | ---- | ----- | ----- | ----- | ------ | ------ | ----- | ------ |
-| Full   | AMP  | 120GB | 240GB | 600GB | 1200GB |  900GB |
+| Full              | AMP  | 120GB | 240GB | 600GB | 1200GB | 2000GB | 900GB | 2400GB |
-| Full   |  16  |  60GB | 120GB | 300GB |  600GB |  400GB |
+| Full              |  16  |  60GB | 120GB | 300GB |  600GB |  900GB | 400GB | 1200GB |
-| GaLore |  16  |  16GB |  32GB |  64GB |  160GB |  120GB |
+| Freeze            |  16  |  20GB |  40GB |  80GB |  200GB |  360GB | 160GB |  400GB |
-| Freeze |  16  |  20GB |  40GB |  80GB |  200GB |  160GB |
+| LoRA/GaLore/BAdam |  16  |  16GB |  32GB |  64GB |  160GB |  240GB | 120GB |  320GB |
-| LoRA   |  16  |  16GB |  32GB |  64GB |  160GB |  120GB |
+| QLoRA             |   8  |  10GB |  20GB |  40GB |   80GB |  140GB |  60GB |  160GB |
-| QLoRA  |   8  |  10GB |  20GB |  40GB |   80GB |   60GB |
+| QLoRA             |   4  |   6GB |  12GB |  24GB |   48GB |   72GB |  30GB |   96GB |
-| QLoRA  |   4  |   6GB |  12GB |  24GB |   48GB |   30GB |
+| QLoRA             |   2  |   4GB |   8GB |  16GB |   24GB |   48GB |  18GB |   48GB |
-| QLoRA  |   2  |   4GB |   8GB |  16GB |   24GB |   18GB |
 
 ## Getting Started
 
-### Data Preparation (optional)
+### Installation
 
-Please refer to [data/README.md](data/README.md) for checking the details about the format of dataset files. You can either use a single `.json` file or a [dataset loading script](https://huggingface.co/docs/datasets/dataset_script) with multiple files to create a custom dataset.
+> [!IMPORTANT]
-
+> Installation is mandatory.
-> [!NOTE]
-> Please update `data/dataset_info.json` to use your custom dataset. About the format of this file, please refer to `data/README.md`.
-
-### Dependence Installation (optional)
 
 ```bash
 git clone https://github.com/hiyouga/LLaMA-Factory.git
-conda create -n llama_factory python=3.10
-conda activate llama_factory
 cd LLaMA-Factory
-pip install -r requirements.txt
+pip install -e .[torch,metrics]
 ```
 
-If you want to enable the quantized LoRA (QLoRA) on the Windows platform, you will be required 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.
+Extra dependencies available: torch, metrics, deepspeed, bitsandbytes, vllm, galore, badam, gptq, awq, aqlm, 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
@@ -316,357 +339,83 @@ pip install https://github.com/jllllll/bitsandbytes-windows-webui/releases/downl
 
 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.
 
-### Use ModelScope Hub (optional)
+</details>
 
-If you have trouble with downloading models and datasets from Hugging Face, you can use LLaMA-Factory together with ModelScope in the following manner.
+<details><summary>For Ascend NPU users</summary>
+
+To utilize Ascend NPU devices for (distributed) training and inference, you need to install the **[torch-npu](https://gitee.com/ascend/pytorch)** library and the **[Ascend CANN Kernels](https://www.hiascend.com/developer/download/community/result?module=cann)**.
+
+| Requirement  | Minimum | Recommend |
+| ------------ | ------- | --------- |
+| CANN         | 8.0.RC1 | 8.0.RC1   |
+| torch        | 2.2.0   | 2.2.0     |
+| torch-npu    | 2.2.0   | 2.2.0     |
+| deepspeed    | 0.13.2  | 0.13.2    |
+
+Docker image:
+
+- 32GB: [Download page](http://mirrors.cn-central-221.ovaijisuan.com/detail/130.html)
+- 64GB: Coming soon
+
+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.
+
+</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
-export USE_MODELSCOPE_HUB=1 # `set USE_MODELSCOPE_HUB=1` for Windows
+CUDA_VISIBLE_DEVICES=0 llamafactory-cli train examples/lora_single_gpu/llama3_lora_sft.yaml
+CUDA_VISIBLE_DEVICES=0 llamafactory-cli chat examples/inference/llama3_lora_sft.yaml
+CUDA_VISIBLE_DEVICES=0 llamafactory-cli export examples/merge_lora/llama3_lora_sft.yaml
 ```
 
-Then you can train the corresponding model by specifying a model ID of the ModelScope Hub. (find a full list of model IDs at [ModelScope Hub](https://modelscope.cn/models))
+See [examples/README.md](examples/README.md) for advanced usage (including distributed training).
 
-```bash
+> [!TIP]
-CUDA_VISIBLE_DEVICES=0 python src/train_bash.py \
+> Use `llamafactory-cli help` to show help information.
-    --model_name_or_path modelscope/Llama-2-7b-ms \
-    ... # arguments (same as below)
-```
 
-LLaMA Board also supports using the models and datasets on the ModelScope Hub.
+### Fine-Tuning with LLaMA Board GUI (powered by [Gradio](https://github.com/gradio-app/gradio))
-
-```bash
-CUDA_VISIBLE_DEVICES=0 USE_MODELSCOPE_HUB=1 python src/train_web.py
-```
-
-### Train on a single GPU
 
 > [!IMPORTANT]
-> If you want to train models on multiple GPUs, please refer to [Distributed Training](#distributed-training).
+> LLaMA Board GUI only supports training on a single GPU.
 
-
+#### Use local environment
-#### LLaMA Board GUI
 
 ```bash
-CUDA_VISIBLE_DEVICES=0 python src/train_web.py
+CUDA_VISIBLE_DEVICES=0 GRADIO_SHARE=1 llamafactory-cli webui
 ```
 
-#### Pre-Training
+<details><summary>For Alibaba Cloud PAI or AutoDL users</summary>
+
+If you encountered display problems in LLaMA Board on Alibaba Cloud PAI, try using the following command to set environment variables before starting LLaMA Board:
 
 ```bash
-CUDA_VISIBLE_DEVICES=0 python src/train_bash.py \
+export GRADIO_SERVER_PORT=7860 GRADIO_ROOT_PATH=/${JUPYTER_NAME}/proxy/7860/
-    --stage pt \
-    --do_train \
-    --model_name_or_path path_to_llama_model \
-    --dataset wiki_demo \
-    --finetuning_type lora \
-    --lora_target q_proj,v_proj \
-    --output_dir path_to_pt_checkpoint \
-    --overwrite_cache \
-    --per_device_train_batch_size 4 \
-    --gradient_accumulation_steps 4 \
-    --lr_scheduler_type cosine \
-    --logging_steps 10 \
-    --save_steps 1000 \
-    --learning_rate 5e-5 \
-    --num_train_epochs 3.0 \
-    --plot_loss \
-    --fp16
 ```
 
-#### Supervised Fine-Tuning
+If you are using AutoDL, please install a specific version of Gradio:
 
 ```bash
-CUDA_VISIBLE_DEVICES=0 python src/train_bash.py \
+pip install gradio==4.10.0
-    --stage sft \
-    --do_train \
-    --model_name_or_path path_to_llama_model \
-    --dataset alpaca_gpt4_en \
-    --template default \
-    --finetuning_type lora \
-    --lora_target q_proj,v_proj \
-    --output_dir path_to_sft_checkpoint \
-    --overwrite_cache \
-    --per_device_train_batch_size 4 \
-    --gradient_accumulation_steps 4 \
-    --lr_scheduler_type cosine \
-    --logging_steps 10 \
-    --save_steps 1000 \
-    --learning_rate 5e-5 \
-    --num_train_epochs 3.0 \
-    --plot_loss \
-    --fp16
-```
-
-#### Reward Modeling
-
-```bash
-CUDA_VISIBLE_DEVICES=0 python src/train_bash.py \
-    --stage rm \
-    --do_train \
-    --model_name_or_path path_to_llama_model \
-    --adapter_name_or_path path_to_sft_checkpoint \
-    --create_new_adapter \
-    --dataset comparison_gpt4_en \
-    --template default \
-    --finetuning_type lora \
-    --lora_target q_proj,v_proj \
-    --output_dir path_to_rm_checkpoint \
-    --per_device_train_batch_size 2 \
-    --gradient_accumulation_steps 4 \
-    --lr_scheduler_type cosine \
-    --logging_steps 10 \
-    --save_steps 1000 \
-    --learning_rate 1e-5 \
-    --num_train_epochs 1.0 \
-    --plot_loss \
-    --fp16
-```
-
-#### PPO Training
-
-```bash
-CUDA_VISIBLE_DEVICES=0 python src/train_bash.py \
-    --stage ppo \
-    --do_train \
-    --model_name_or_path path_to_llama_model \
-    --adapter_name_or_path path_to_sft_checkpoint \
-    --create_new_adapter \
-    --dataset alpaca_gpt4_en \
-    --template default \
-    --finetuning_type lora \
-    --lora_target q_proj,v_proj \
-    --reward_model path_to_rm_checkpoint \
-    --output_dir path_to_ppo_checkpoint \
-    --per_device_train_batch_size 2 \
-    --gradient_accumulation_steps 4 \
-    --lr_scheduler_type cosine \
-    --top_k 0 \
-    --top_p 0.9 \
-    --logging_steps 10 \
-    --save_steps 1000 \
-    --learning_rate 1e-5 \
-    --num_train_epochs 1.0 \
-    --plot_loss \
-    --fp16
-```
-
-> [!TIP]
-> Use `--adapter_name_or_path path_to_sft_checkpoint,path_to_ppo_checkpoint` to infer the fine-tuned model if `--create_new_adapter` was enabled.
-
-> [!WARNING]
-> Use `--per_device_train_batch_size=1` for LLaMA-2 models in fp16 PPO training.
-
-#### DPO Training
-
-```bash
-CUDA_VISIBLE_DEVICES=0 python src/train_bash.py \
-    --stage dpo \
-    --do_train \
-    --model_name_or_path path_to_llama_model \
-    --adapter_name_or_path path_to_sft_checkpoint \
-    --create_new_adapter \
-    --dataset comparison_gpt4_en \
-    --template default \
-    --finetuning_type lora \
-    --lora_target q_proj,v_proj \
-    --output_dir path_to_dpo_checkpoint \
-    --per_device_train_batch_size 2 \
-    --gradient_accumulation_steps 4 \
-    --lr_scheduler_type cosine \
-    --logging_steps 10 \
-    --save_steps 1000 \
-    --learning_rate 1e-5 \
-    --num_train_epochs 1.0 \
-    --plot_loss \
-    --fp16
-```
-
-> [!TIP]
-> Use `--adapter_name_or_path path_to_sft_checkpoint,path_to_dpo_checkpoint` to infer the fine-tuned model if `--create_new_adapter` was enabled.
-
-### Distributed Training
-
-#### Use Huggingface Accelerate
-
-```bash
-accelerate launch --config_file config.yaml src/train_bash.py \
-    --ddp_timeout 180000000 \
-    ... # arguments (same as above)
-```
-
-<details><summary>Example config.yaml for LoRA training</summary>
-
-```yaml
-compute_environment: LOCAL_MACHINE
-debug: false
-distributed_type: MULTI_GPU
-downcast_bf16: 'no'
-gpu_ids: all
-machine_rank: 0
-main_training_function: main
-mixed_precision: fp16
-num_machines: 1
-num_processes: 4
-rdzv_backend: static
-same_network: true
-tpu_env: []
-tpu_use_cluster: false
-tpu_use_sudo: false
-use_cpu: false
 ```
 
 </details>
 
-> [!TIP]
-> We commend using Accelerate for LoRA tuning.
-
-#### Use DeepSpeed
-
-```bash
-deepspeed --num_gpus 8 src/train_bash.py \
-    --deepspeed ds_config.json \
-    --ddp_timeout 180000000 \
-    ... # arguments (same as above)
-```
-
-<details><summary>Example ds_config.json for full-parameter training with DeepSpeed ZeRO-2</summary>
-
-```json
-{
-  "train_batch_size": "auto",
-  "train_micro_batch_size_per_gpu": "auto",
-  "gradient_accumulation_steps": "auto",
-  "gradient_clipping": "auto",
-  "zero_allow_untested_optimizer": true,
-  "fp16": {
-    "enabled": "auto",
-    "loss_scale": 0,
-    "loss_scale_window": 1000,
-    "initial_scale_power": 16,
-    "hysteresis": 2,
-    "min_loss_scale": 1
-  },
-  "bf16": {
-    "enabled": "auto"
-  },
-  "zero_optimization": {
-    "stage": 2,
-    "allgather_partitions": true,
-    "allgather_bucket_size": 5e8,
-    "overlap_comm": true,
-    "reduce_scatter": true,
-    "reduce_bucket_size": 5e8,
-    "contiguous_gradients": true,
-    "round_robin_gradients": true
-  }
-}
-```
-
-</details>
-
-> [!TIP]
-> Refer to [examples](examples) for more training scripts.
-
-### Merge LoRA weights and export model
-
-```bash
-CUDA_VISIBLE_DEVICES= python src/export_model.py \
-    --model_name_or_path path_to_llama_model \
-    --adapter_name_or_path path_to_checkpoint \
-    --template default \
-    --finetuning_type lora \
-    --export_dir path_to_export \
-    --export_size 2 \
-    --export_legacy_format False
-```
-
-> [!WARNING]
-> Merging LoRA weights into a quantized model is not supported.
-
-> [!TIP]
-> Use `--model_name_or_path path_to_export` solely to use the exported model.
-> 
-> Use `CUDA_VISIBLE_DEVICES=0`, `--export_quantization_bit 4` and `--export_quantization_dataset data/c4_demo.json` to quantize the model with AutoGPTQ after merging the LoRA weights.
-
-### Inference with OpenAI-style API
-
-```bash
-CUDA_VISIBLE_DEVICES=0 API_PORT=8000 python src/api_demo.py \
-    --model_name_or_path path_to_llama_model \
-    --adapter_name_or_path path_to_checkpoint \
-    --template default \
-    --finetuning_type lora
-```
-
-> [!TIP]
-> Visit `http://localhost:8000/docs` for API documentation.
-
-### Inference with command line
-
-```bash
-CUDA_VISIBLE_DEVICES=0 python src/cli_demo.py \
-    --model_name_or_path path_to_llama_model \
-    --adapter_name_or_path path_to_checkpoint \
-    --template default \
-    --finetuning_type lora
-```
-
-### Inference with web browser
-
-```bash
-CUDA_VISIBLE_DEVICES=0 python src/web_demo.py \
-    --model_name_or_path path_to_llama_model \
-    --adapter_name_or_path path_to_checkpoint \
-    --template default \
-    --finetuning_type lora
-```
-
-### Evaluation
-
-```bash
-CUDA_VISIBLE_DEVICES=0 python src/evaluate.py \
-    --model_name_or_path path_to_llama_model \
-    --adapter_name_or_path path_to_checkpoint \
-    --template vanilla \
-    --finetuning_type lora \
-    --task mmlu \
-    --split test \
-    --lang en \
-    --n_shot 5 \
-    --batch_size 4
-```
-
-### Predict
-
-```bash
-CUDA_VISIBLE_DEVICES=0 python src/train_bash.py \
-    --stage sft \
-    --do_predict \
-    --model_name_or_path path_to_llama_model \
-    --adapter_name_or_path path_to_checkpoint \
-    --dataset alpaca_gpt4_en \
-    --template default \
-    --finetuning_type lora \
-    --output_dir path_to_predict_result \
-    --per_device_eval_batch_size 1 \
-    --max_samples 100 \
-    --predict_with_generate \
-    --fp16
-```
-
-> [!WARNING]
-> Use `--per_device_train_batch_size=1` for LLaMA-2 models in fp16 predict.
-
-> [!TIP]
-> We recommend using `--per_device_eval_batch_size=1` and `--max_target_length 128` at 4/8-bit predict.
-
-### Dockerize Training
-
 #### Use Docker
 
 ```bash
 docker build -f ./Dockerfile -t llama-factory:latest .
-
 docker run --gpus=all \
     -v ./hf_cache:/root/.cache/huggingface/ \
     -v ./data:/app/data \
@@ -684,14 +433,36 @@ docker run --gpus=all \
 docker compose -f ./docker-compose.yml up -d
 ```
 
-> [!TIP]
+<details><summary>Details about volume</summary>
-> Details about volume:
+
-> * hf_cache: Utilize Hugging Face cache on the host machine. Reassignable if a cache already exists in a different directory.
+- hf_cache: Utilize Hugging Face cache on the host machine. Reassignable if a cache already exists in a different directory.
-> * data: Place datasets on this dir of the host machine so that they can be selected on LLaMA Board GUI.
+- 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.
+- 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
+CUDA_VISIBLE_DEVICES=0,1 API_PORT=8000 llamafactory-cli api examples/inference/llama3_vllm.yaml
+```
+
+### 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`.
 
 ## 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)
@@ -713,21 +484,37 @@ docker compose -f ./docker-compose.yml up -d
 1. Huang et al. Key-Point-Driven Data Synthesis with its Enhancement on Mathematical Reasoning. 2024. [[arxiv]](https://arxiv.org/abs/2403.02333)
 1. Duan et al. Negating Negatives: Alignment without Human Positive Samples via Distributional Dispreference Optimization. 2024. [[arxiv]](https://arxiv.org/abs/2403.03419)
 1. Xie and Schwertfeger. Empowering Robotics with Large Language Models: osmAG Map Comprehension with LLMs. 2024. [[arxiv]](https://arxiv.org/abs/2403.08228)
+1. Wu et al. Large Language Models are Parallel Multilingual Learners. 2024. [[arxiv]](https://arxiv.org/abs/2403.09073)
+1. Zhang et al. EDT: Improving Large Language Models' Generation by Entropy-based Dynamic Temperature Sampling. 2024. [[arxiv]](https://arxiv.org/abs/2403.14541)
+1. Weller et al. FollowIR: Evaluating and Teaching Information Retrieval Models to Follow Instructions. 2024. [[arxiv]](https://arxiv.org/abs/2403.15246)
 1. Hongbin Na. CBT-LLM: A Chinese Large Language Model for Cognitive Behavioral Therapy-based Mental Health Question Answering. 2024. [[arxiv]](https://arxiv.org/abs/2403.16008)
+1. Zan et al. CodeS: Natural Language to Code Repository via Multi-Layer Sketch. 2024. [[arxiv]](https://arxiv.org/abs/2403.16443)
+1. Liu et al. Extensive Self-Contrast Enables Feedback-Free Language Model Alignment. 2024. [[arxiv]](https://arxiv.org/abs/2404.00604)
+1. Luo et al. BAdam: A Memory Efficient Full Parameter Training Method for Large Language Models. 2024. [[arxiv]](https://arxiv.org/abs/2404.02827)
+1. Du et al. Chinese Tiny LLM: Pretraining a Chinese-Centric Large Language Model. 2024. [[arxiv]](https://arxiv.org/abs/2404.04167)
+1. Ma et al. Parameter Efficient Quasi-Orthogonal Fine-Tuning via Givens Rotation. 2024. [[arxiv]](https://arxiv.org/abs/2404.04316)
+1. Liu et al. Dynamic Generation of Personalities with Large Language Models. 2024. [[arxiv]](https://arxiv.org/abs/2404.07084)
+1. Shang et al. How Far Have We Gone in Stripped Binary Code Understanding Using Large Language Models. 2024. [[arxiv]](https://arxiv.org/abs/2404.09836)
+1. Huang et al. LLMTune: Accelerate Database Knob Tuning with Large Language Models. 2024. [[arxiv]](https://arxiv.org/abs/2404.11581)
+1. Deng et al. Text-Tuple-Table: Towards Information Integration in Text-to-Table Generation via Global Tuple Extraction. 2024. [[arxiv]](https://arxiv.org/abs/2404.14215)
+1. Acikgoz et al. Hippocrates: An Open-Source Framework for Advancing Large Language Models in Healthcare. 2024. [[arxiv]](https://arxiv.org/abs/2404.16621)
+1. Zhang et al. Small Language Models Need Strong Verifiers to Self-Correct Reasoning. 2024. [[arxiv]](https://arxiv.org/abs/2404.17140)
+1. Zhou et al. FREB-TQA: A Fine-Grained Robustness Evaluation Benchmark for Table Question Answering. 2024. [[arxiv]](https://arxiv.org/abs/2404.18585)
 1. **[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/thomas-yanxin/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.
 
-> [!TIP]
+</details>
-> If you have a project that should be incorporated, please contact via email or create a pull request.
 
 ## License
 
 This repository is licensed under the [Apache-2.0 License](LICENSE).
 
-Please follow the model licenses to use the corresponding model weights: [Baichuan2](https://huggingface.co/baichuan-inc/Baichuan2-7B-Base/blob/main/Community%20License%20for%20Baichuan%202%20Model.pdf) / [BLOOM](https://huggingface.co/spaces/bigscience/license) / [ChatGLM3](https://github.com/THUDM/ChatGLM3/blob/main/MODEL_LICENSE) / [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) / [InternLM2](https://github.com/InternLM/InternLM#license) / [LLaMA](https://github.com/facebookresearch/llama/blob/main/MODEL_CARD.md) / [LLaMA-2](https://ai.meta.com/llama/license/) / [Mistral](LICENSE) / [OLMo](LICENSE) / [Phi-1.5/2](https://huggingface.co/microsoft/phi-1_5/resolve/main/Research%20License.docx) / [Qwen](https://github.com/QwenLM/Qwen/blob/main/Tongyi%20Qianwen%20LICENSE%20AGREEMENT) / [StarCoder2](https://huggingface.co/spaces/bigcode/bigcode-model-license-agreement) / [XVERSE](https://github.com/xverse-ai/XVERSE-13B/blob/main/MODEL_LICENSE.pdf) / [Yi](https://huggingface.co/01-ai/Yi-6B/blob/main/LICENSE) / [Yuan](https://github.com/IEIT-Yuan/Yuan-2.0/blob/main/LICENSE-Yuan)
+Please follow the model licenses to use the corresponding model weights: [Baichuan2](https://huggingface.co/baichuan-inc/Baichuan2-7B-Base/blob/main/Community%20License%20for%20Baichuan%202%20Model.pdf) / [BLOOM](https://huggingface.co/spaces/bigscience/license) / [ChatGLM3](https://github.com/THUDM/ChatGLM3/blob/main/MODEL_LICENSE) / [Command-R](https://cohere.com/c4ai-cc-by-nc-license) / [DeepSeek](https://github.com/deepseek-ai/DeepSeek-LLM/blob/main/LICENSE-MODEL) / [Falcon](https://huggingface.co/tiiuae/falcon-180B/blob/main/LICENSE.txt) / [Gemma](https://ai.google.dev/gemma/terms) / [InternLM2](https://github.com/InternLM/InternLM#license) / [LLaMA](https://github.com/facebookresearch/llama/blob/main/MODEL_CARD.md) / [LLaMA-2 (LLaVA-1.5)](https://ai.meta.com/llama/license/) / [LLaMA-3](https://llama.meta.com/llama3/license/) / [Mistral](LICENSE) / [OLMo](LICENSE) / [Phi-1.5/2](https://huggingface.co/microsoft/phi-1_5/resolve/main/Research%20License.docx) / [Phi-3](https://huggingface.co/microsoft/Phi-3-mini-4k-instruct/blob/main/LICENSE) / [Qwen](https://github.com/QwenLM/Qwen/blob/main/Tongyi%20Qianwen%20LICENSE%20AGREEMENT) / [StarCoder2](https://huggingface.co/spaces/bigcode/bigcode-model-license-agreement) / [XVERSE](https://github.com/xverse-ai/XVERSE-13B/blob/main/MODEL_LICENSE.pdf) / [Yi](https://huggingface.co/01-ai/Yi-6B/blob/main/LICENSE) / [Yi-1.5](LICENSE) / [Yuan](https://github.com/IEIT-Yuan/Yuan-2.0/blob/main/LICENSE-Yuan)
 
 ## Citation
 
@@ -745,7 +532,7 @@ If this work is helpful, please kindly cite as:
 
 ## Acknowledgement
 
-This repo benefits from [PEFT](https://github.com/huggingface/peft), [QLoRA](https://github.com/artidoro/qlora) and [FastChat](https://github.com/lm-sys/FastChat). Thanks for their wonderful works.
+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
 

README_zh.md

@@ -5,13 +5,15 @@
 [![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/llmtuner)](https://pypi.org/project/llmtuner/)
 [![Downloads](https://static.pepy.tech/badge/llmtuner)](https://pypi.org/project/llmtuner/)
-[![Citation](https://img.shields.io/badge/citation-27-green)](#使用了-llama-factory-的项目)
+[![Citation](https://img.shields.io/badge/citation-44-green)](#使用了-llama-factory-的项目)
 [![GitHub pull request](https://img.shields.io/badge/PRs-welcome-blue)](https://github.com/hiyouga/LLaMA-Factory/pulls)
 [![Discord](https://dcbadge.vercel.app/api/server/rKfvV9r9FK?compact=true&style=flat)](https://discord.gg/rKfvV9r9FK)
 [![Twitter](https://img.shields.io/twitter/follow/llamafactory_ai)](https://twitter.com/llamafactory_ai)
 [![Spaces](https://img.shields.io/badge/🤗-Open%20in%20Spaces-blue)](https://huggingface.co/spaces/hiyouga/LLaMA-Board)
 [![Studios](https://img.shields.io/badge/ModelScope-Open%20in%20Studios-blue)](https://modelscope.cn/studios/hiyouga/LLaMA-Board)
-[![Open in Colab](https://colab.research.google.com/assets/colab-badge.svg)](https://colab.research.google.com/drive/1eRTPn37ltBbYsISy9Aw2NuI2Aq5CQrD9?usp=sharing)
+[![Open in Colab](https://colab.research.google.com/assets/colab-badge.svg)](https://colab.research.google.com/drive/1d5KQtbemerlSDSxZIfAaWXhKr30QypiK?usp=sharing)
+
+[![GitHub Tread](https://trendshift.io/api/badge/repositories/4535)](https://trendshift.io/repositories/4535)
 
 👋 加入我们的[微信群](assets/wechat.jpg)。
 
@@ -23,7 +25,7 @@ https://github.com/hiyouga/LLaMA-Factory/assets/16256802/ec36a9dd-37f4-4f72-81bd
 
 选择你的打开方式：
 
-- **Colab**：https://colab.research.google.com/drive/1eRTPn37ltBbYsISy9Aw2NuI2Aq5CQrD9?usp=sharing
+- **Colab**：https://colab.research.google.com/drive/1d5KQtbemerlSDSxZIfAaWXhKr30QypiK?usp=sharing
 - **本地机器**：请见[如何使用](#如何使用)
 
 ## 目录
@@ -43,17 +45,17 @@ https://github.com/hiyouga/LLaMA-Factory/assets/16256802/ec36a9dd-37f4-4f72-81bd
 
 ## 项目特色
 
-- **多种模型**：LLaMA、Mistral、Mixtral-MoE、Qwen、Yi、Gemma、Baichuan、ChatGLM、Phi 等等。
+- **多种模型**：LLaMA、LLaVA、Mistral、Mixtral-MoE、Qwen、Yi、Gemma、Baichuan、ChatGLM、Phi 等等。
-- **集成方法**：（增量）预训练、指令监督微调、奖励模型训练、PPO 训练和 DPO 训练。
+- **集成方法**：（增量）预训练、（多模态）指令监督微调、奖励模型训练、PPO 训练、DPO 训练和 ORPO 训练。
 - **多种精度**：32 比特全参数微调、16 比特冻结微调、16 比特 LoRA 微调和基于 AQLM/AWQ/GPTQ/LLM.int8 的 2/4/8 比特 QLoRA 微调。
-- **先进算法**：GaLore、DoRA、LongLoRA、LLaMA Pro、LoRA+、LoftQ 和 Agent 微调。
+- **先进算法**：GaLore、BAdam、DoRA、LongLoRA、LLaMA Pro、Mixture-of-Depths、LoRA+、LoftQ 和 Agent 微调。
 - **实用技巧**：FlashAttention-2、Unsloth、RoPE scaling、NEFTune 和 rsLoRA。
 - **实验监控**：LlamaBoard、TensorBoard、Wandb、MLflow 等等。
 - **极速推理**：基于 vLLM 的 OpenAI 风格 API、浏览器界面和命令行接口。
 
 ## 性能指标
 
-与 ChatGLM 官方的 [P-Tuning](https://github.com/THUDM/ChatGLM2-6B/tree/main/ptuning) 微调相比，LLaMA-Factory 的 LoRA 微调提供了 **3.7 倍**的加速比，同时在广告文案生成任务上取得了更高的 Rouge 分数。结合 4 比特量化技术，LLaMA-Factory 的 QLoRA 微调进一步降低了 GPU 显存消耗。
+与 ChatGLM 官方的 [P-Tuning](https://github.com/THUDM/ChatGLM2-6B/tree/main/ptuning) 微调相比，LLaMA Factory 的 LoRA 微调提供了 **3.7 倍**的加速比，同时在广告文案生成任务上取得了更高的 Rouge 分数。结合 4 比特量化技术，LLaMA Factory 的 QLoRA 微调进一步降低了 GPU 显存消耗。
 
 ![benchmark](assets/benchmark.svg)
 
@@ -62,51 +64,67 @@ https://github.com/hiyouga/LLaMA-Factory/assets/16256802/ec36a9dd-37f4-4f72-81bd
 - **Training Speed**: 训练阶段每秒处理的样本数量。（批处理大小=4，截断长度=1024）
 - **Rouge Score**: [广告文案生成](https://aclanthology.org/D19-1321.pdf)任务验证集上的 Rouge-2 分数。（批处理大小=4，截断长度=1024）
 - **GPU Memory**: 4 比特量化训练的 GPU 显存峰值。（批处理大小=1，截断长度=1024）
-- 我们在 ChatGLM 的 P-Tuning 中采用 `pre_seq_len=128`，在 LLaMA-Factory 的 LoRA 微调中采用 `lora_rank=32`。
+- 我们在 ChatGLM 的 P-Tuning 中采用 `pre_seq_len=128`，在 LLaMA Factory 的 LoRA 微调中采用 `lora_rank=32`。
 
 </details>
 
 ## 更新日志
 
-[24/03/21] 我们的论文 "[LlamaFactory: Unified Efficient Fine-Tuning of 100+ Language Models](https://arxiv.org/abs/2403.13372)" 可在 arXiv 上查看！
+[24/05/14] 我们支持了昇腾 NPU 设备的训练和推理。详情请查阅[安装](#安装-llama-factory)部分。
 
-[24/03/20] 我们支持了能在 2x24GB GPU 上微调 70B 模型的 **FSDP+QLoRA**。详细用法请参照 `examples/fsdp_qlora`。
+[24/05/13] 我们支持了 Yi-1.5 系列模型的微调。
 
-[24/03/13] 我们支持了 **[LoRA+](https://arxiv.org/abs/2402.12354)**。详细用法请参照 `examples/extras/loraplus`。
+[24/04/26] 我们支持了多模态模型 **LLaVA-1.5** 的微调。详细用法请参照 [examples](examples/README_zh.md)。
-
-[24/03/07] 我们支持了梯度低秩投影（**[GaLore](https://arxiv.org/abs/2403.03507)**）算法。详细用法请参照 `examples/extras/galore`。
 
 <details><summary>展开日志</summary>
 
-[24/03/07] 我们集成了 **[vLLM](https://github.com/vllm-project/vllm)** 以实现极速并发推理。请使用 `--infer_backend vllm` 来获得 **270%** 的推理速度。（尚不支持 LoRA，请先合并权重。）
+[24/04/22] 我们提供了在免费 T4 GPU 上微调 Llama-3 模型的 **[Colab 笔记本](https://colab.research.google.com/drive/1d5KQtbemerlSDSxZIfAaWXhKr30QypiK?usp=sharing)**。Hugging Face 社区公开了两个利用 LLaMA Factory 微调的 Llama-3 模型，详情请见 [Llama3-8B-Chinese-Chat](https://huggingface.co/shenzhi-wang/Llama3-8B-Chinese-Chat) 和 [Llama3-Chinese](https://huggingface.co/zhichen/Llama3-Chinese)。
 
-[24/02/28] 我们支持了 **[DoRA](https://arxiv.org/abs/2402.09353)** 微调。请使用 `--use_dora` 参数进行 DoRA 微调。
+[24/04/21] 我们基于 [AstraMindAI 的仓库](https://github.com/astramind-ai/Mixture-of-depths)支持了 **[混合深度训练](https://arxiv.org/abs/2404.02258)**。详细用法请参照 [examples](examples/README_zh.md)。
 
-[24/02/15] 我们支持了 [LLaMA Pro](https://github.com/TencentARC/LLaMA-Pro) 提出的**块扩展**方法。详细用法请参照 `examples/extras/llama_pro`。
+[24/04/16] 我们支持了 **[BAdam](https://arxiv.org/abs/2404.02827)**。详细用法请参照 [examples](examples/README_zh.md)。
+
+[24/04/16] 我们支持了 **[unsloth](https://github.com/unslothai/unsloth)** 的长序列训练（24GB 可训练 Llama-2-7B-56k）。该方法相比 FlashAttention-2 提供了 **117%** 的训练速度和 **50%** 的显存节约。更多数据请见[此页面](https://github.com/hiyouga/LLaMA-Factory/wiki/Performance-comparison)。
+
+[24/03/31] 我们支持了 **[ORPO](https://arxiv.org/abs/2403.07691)**。详细用法请参照 [examples](examples/README_zh.md)。
+
+[24/03/21] 我们的论文 "[LlamaFactory: Unified Efficient Fine-Tuning of 100+ Language Models](https://arxiv.org/abs/2403.13372)" 可在 arXiv 上查看！
+
+[24/03/20] 我们支持了能在 2x24GB GPU 上微调 70B 模型的 **FSDP+QLoRA**。详细用法请参照 [examples](examples/README_zh.md)。
+
+[24/03/13] 我们支持了 **[LoRA+](https://arxiv.org/abs/2402.12354)**。详细用法请参照 [examples](examples/README_zh.md)。
+
+[24/03/07] 我们支持了梯度低秩投影（**[GaLore](https://arxiv.org/abs/2403.03507)**）算法。详细用法请参照 [examples](examples/README_zh.md)。
+
+[24/03/07] 我们集成了 **[vLLM](https://github.com/vllm-project/vllm)** 以实现极速并发推理。请使用 `infer_backend: vllm` 来获得 **270%** 的推理速度。
+
+[24/02/28] 我们支持了 **[DoRA](https://arxiv.org/abs/2402.09353)** 微调。请使用 `use_dora: true` 参数进行 DoRA 微调。
+
+[24/02/15] 我们支持了 [LLaMA Pro](https://github.com/TencentARC/LLaMA-Pro) 提出的**块扩展**方法。详细用法请参照 [examples](examples/README_zh.md)。
 
 [24/02/05] Qwen1.5（Qwen2 测试版）系列模型已在 LLaMA-Factory 中实现微调支持。详情请查阅该[博客页面](https://qwenlm.github.io/zh/blog/qwen1.5/)。
 
-[24/01/18] 我们针对绝大多数模型实现了 **Agent 微调**，微调时指定 `--dataset glaive_toolcall` 即可使模型获得工具调用能力。
+[24/01/18] 我们针对绝大多数模型实现了 **Agent 微调**，微调时指定 `dataset: glaive_toolcall` 即可使模型获得工具调用能力。
 
-[23/12/23] 我们针对 LLaMA, Mistral 和 Yi 模型支持了 **[unsloth](https://github.com/unslothai/unsloth)** 的 LoRA 训练加速。请使用 `--use_unsloth` 参数启用 unsloth 优化。该方法可提供 **170%** 的训练速度，详情请查阅[此页面](https://github.com/hiyouga/LLaMA-Factory/wiki/Performance-comparison)。
+[23/12/23] 我们针对 LLaMA, Mistral 和 Yi 模型支持了 **[unsloth](https://github.com/unslothai/unsloth)** 的 LoRA 训练加速。请使用 `use_unsloth: true` 参数启用 unsloth 优化。该方法可提供 **170%** 的训练速度，详情请查阅[此页面](https://github.com/hiyouga/LLaMA-Factory/wiki/Performance-comparison)。
 
 [23/12/12] 我们支持了微调最新的混合专家模型 **[Mixtral 8x7B](https://huggingface.co/mistralai/Mixtral-8x7B-v0.1)**。硬件需求请查阅[此处](#硬件依赖)。
 
-[23/12/01] 我们支持了从 **[魔搭社区](https://modelscope.cn/models)** 下载预训练模型和数据集。详细用法请参照 [此教程](#使用魔搭社区可跳过)。
+[23/12/01] 我们支持了从 **[魔搭社区](https://modelscope.cn/models)** 下载预训练模型和数据集。详细用法请参照 [此教程](#从魔搭社区下载)。
 
-[23/10/21] 我们支持了 **[NEFTune](https://arxiv.org/abs/2310.05914)** 训练技巧。请使用 `--neftune_noise_alpha` 参数启用 NEFTune，例如 `--neftune_noise_alpha 5`。
+[23/10/21] 我们支持了 **[NEFTune](https://arxiv.org/abs/2310.05914)** 训练技巧。请使用 `neftune_noise_alpha: 5` 参数启用 NEFTune。
 
-[23/09/27] 我们针对 LLaMA 模型支持了 [LongLoRA](https://github.com/dvlab-research/LongLoRA) 提出的 **$S^2$-Attn**。请使用 `--shift_attn` 参数以启用该功能。
+[23/09/27] 我们针对 LLaMA 模型支持了 [LongLoRA](https://github.com/dvlab-research/LongLoRA) 提出的 **$S^2$-Attn**。请使用 `shift_attn: true` 参数以启用该功能。
 
-[23/09/23] 我们在项目中集成了 MMLU、C-Eval 和 CMMLU 评估集。使用方法请参阅[此示例](#模型评估)。
+[23/09/23] 我们在项目中集成了 MMLU、C-Eval 和 CMMLU 评估集。详细用法请参照 [examples](examples/README_zh.md)。
 
-[23/09/10] 我们支持了 **[FlashAttention-2](https://github.com/Dao-AILab/flash-attention)**。如果您使用的是 RTX4090、A100 或 H100 GPU，请使用 `--flash_attn` 参数以启用 FlashAttention-2。
+[23/09/10] 我们支持了 **[FlashAttention-2](https://github.com/Dao-AILab/flash-attention)**。如果您使用的是 RTX4090、A100 或 H100 GPU，请使用 `flash_attn: fa2` 参数以启用 FlashAttention-2。
 
-[23/08/12] 我们支持了 **RoPE 插值**来扩展 LLaMA 模型的上下文长度。请使用 `--rope_scaling linear` 参数训练模型或使用 `--rope_scaling dynamic` 参数评估模型。
+[23/08/12] 我们支持了 **RoPE 插值**来扩展 LLaMA 模型的上下文长度。请使用 `rope_scaling: linear` 参数训练模型或使用 `rope_scaling: dynamic` 参数评估模型。
 
-[23/08/11] 我们支持了指令模型的 **[DPO 训练](https://arxiv.org/abs/2305.18290)**。使用方法请参阅[此示例](#dpo-训练)。
+[23/08/11] 我们支持了指令模型的 **[DPO 训练](https://arxiv.org/abs/2305.18290)**。详细用法请参照 [examples](examples/README_zh.md)。
 
-[23/07/31] 我们支持了**数据流式加载**。请使用 `--streaming` 和 `--max_steps 10000` 参数来流式加载数据集。
+[23/07/31] 我们支持了**数据流式加载**。请使用 `streaming: true` 和 `max_steps: 10000` 参数来流式加载数据集。
 
 [23/07/29] 我们在 Hugging Face 发布了两个 13B 指令微调模型。详细内容请查阅我们的 Hugging Face 项目（[LLaMA-2](https://huggingface.co/hiyouga/Llama-2-Chinese-13b-chat) / [Baichuan](https://huggingface.co/hiyouga/Baichuan-13B-sft)）。
 
@@ -118,39 +136,45 @@ https://github.com/hiyouga/LLaMA-Factory/assets/16256802/ec36a9dd-37f4-4f72-81bd
 
 [23/06/22] 我们对齐了[示例 API](src/api_demo.py) 与 [OpenAI API](https://platform.openai.com/docs/api-reference/chat) 的格式，您可以将微调模型接入**任意基于 ChatGPT 的应用**中。
 
-[23/06/03] 我们实现了 4 比特的 LoRA 训练（也称 **[QLoRA](https://github.com/artidoro/qlora)**）。请使用 `--quantization_bit 4` 参数进行 4 比特量化微调。
+[23/06/03] 我们实现了 4 比特的 LoRA 训练（也称 **[QLoRA](https://github.com/artidoro/qlora)**）。详细用法请参照 [examples](examples/README_zh.md)。
 
 </details>
 
 ## 模型
 
-| 模型名                                                   | 模型大小                     | 默认模块           | Template  |
+| 模型名                                                   | 模型大小                          | 默认模块           | Template  |
-| -------------------------------------------------------- | --------------------------- | ----------------- | --------- |
+| -------------------------------------------------------- | -------------------------------- | ----------------- | --------- |
-| [Baichuan2](https://huggingface.co/baichuan-inc)         | 7B/13B                      | W_pack            | baichuan2 |
+| [Baichuan2](https://huggingface.co/baichuan-inc)         | 7B/13B                           | W_pack            | baichuan2 |
-| [BLOOM](https://huggingface.co/bigscience/bloom)         | 560M/1.1B/1.7B/3B/7.1B/176B | query_key_value   | -         |
+| [BLOOM](https://huggingface.co/bigscience)               | 560M/1.1B/1.7B/3B/7.1B/176B      | query_key_value   | -         |
-| [BLOOMZ](https://huggingface.co/bigscience/bloomz)       | 560M/1.1B/1.7B/3B/7.1B/176B | query_key_value   | -         |
+| [BLOOMZ](https://huggingface.co/bigscience)              | 560M/1.1B/1.7B/3B/7.1B/176B      | query_key_value   | -         |
-| [ChatGLM3](https://huggingface.co/THUDM/chatglm3-6b)     | 6B                          | query_key_value   | chatglm3  |
+| [ChatGLM3](https://huggingface.co/THUDM)                 | 6B                               | query_key_value   | chatglm3  |
-| [DeepSeek (MoE)](https://huggingface.co/deepseek-ai)     | 7B/16B/67B                  | q_proj,v_proj     | deepseek  |
+| [Command-R](https://huggingface.co/CohereForAI)          | 35B/104B                         | q_proj,v_proj     | cohere    |
-| [Falcon](https://huggingface.co/tiiuae)                  | 7B/40B/180B                 | query_key_value   | falcon    |
+| [DeepSeek (MoE)](https://huggingface.co/deepseek-ai)     | 7B/16B/67B/236B                  | q_proj,v_proj     | deepseek  |
-| [Gemma](https://huggingface.co/google)                   | 2B/7B                       | q_proj,v_proj     | gemma     |
+| [Falcon](https://huggingface.co/tiiuae)                  | 7B/40B/180B                      | query_key_value   | falcon    |
-| [InternLM2](https://huggingface.co/internlm)             | 7B/20B                      | wqkv              | intern2   |
+| [Gemma/CodeGemma](https://huggingface.co/google)         | 2B/7B                            | q_proj,v_proj     | gemma     |
-| [LLaMA](https://github.com/facebookresearch/llama)       | 7B/13B/33B/65B              | q_proj,v_proj     | -         |
+| [InternLM2](https://huggingface.co/internlm)             | 7B/20B                           | wqkv              | intern2   |
-| [LLaMA-2](https://huggingface.co/meta-llama)             | 7B/13B/70B                  | q_proj,v_proj     | llama2    |
+| [LLaMA](https://github.com/facebookresearch/llama)       | 7B/13B/33B/65B                   | q_proj,v_proj     | -         |
-| [Mistral](https://huggingface.co/mistralai)              | 7B                          | q_proj,v_proj     | mistral   |
+| [LLaMA-2](https://huggingface.co/meta-llama)             | 7B/13B/70B                       | q_proj,v_proj     | llama2    |
-| [Mixtral](https://huggingface.co/mistralai)              | 8x7B                        | q_proj,v_proj     | mistral   |
+| [LLaMA-3](https://huggingface.co/meta-llama)             | 8B/70B                           | q_proj,v_proj     | llama3    |
-| [OLMo](https://huggingface.co/allenai)                   | 1B/7B                       | att_proj          | olmo      |
+| [LLaVA-1.5](https://huggingface.co/llava-hf)             | 7B/13B                           | q_proj,v_proj     | vicuna    |
-| [Phi-1.5/2](https://huggingface.co/microsoft)            | 1.3B/2.7B                   | q_proj,v_proj     | -         |
+| [Mistral/Mixtral](https://huggingface.co/mistralai)      | 7B/8x7B/8x22B                    | q_proj,v_proj     | mistral   |
-| [Qwen](https://huggingface.co/Qwen)                      | 1.8B/7B/14B/72B             | c_attn            | qwen      |
+| [OLMo](https://huggingface.co/allenai)                   | 1B/7B                            | q_proj,v_proj     | -         |
-| [Qwen1.5](https://huggingface.co/Qwen)                   | 0.5B/1.8B/4B/7B/14B/72B     | q_proj,v_proj     | qwen      |
+| [Phi-1.5/2](https://huggingface.co/microsoft)            | 1.3B/2.7B                        | q_proj,v_proj     | -         |
-| [StarCoder2](https://huggingface.co/bigcode)             | 3B/7B/15B                   | q_proj,v_proj     | -         |
+| [Phi-3](https://huggingface.co/microsoft)                | 3.8B                             | qkv_proj          | phi       |
-| [XVERSE](https://huggingface.co/xverse)                  | 7B/13B/65B                  | q_proj,v_proj     | xverse    |
+| [Qwen](https://huggingface.co/Qwen)                      | 1.8B/7B/14B/72B                  | c_attn            | qwen      |
-| [Yi](https://huggingface.co/01-ai)                       | 6B/9B/34B                   | q_proj,v_proj     | yi        |
+| [Qwen1.5 (Code/MoE)](https://huggingface.co/Qwen)        | 0.5B/1.8B/4B/7B/14B/32B/72B/110B | q_proj,v_proj     | qwen      |
-| [Yuan](https://huggingface.co/IEITYuan)                  | 2B/51B/102B                 | q_proj,v_proj     | yuan      |
+| [StarCoder2](https://huggingface.co/bigcode)             | 3B/7B/15B                        | q_proj,v_proj     | -         |
+| [XVERSE](https://huggingface.co/xverse)                  | 7B/13B/65B                       | q_proj,v_proj     | xverse    |
+| [Yi (1/1.5)](https://huggingface.co/01-ai)               | 6B/9B/34B                        | q_proj,v_proj     | yi        |
+| [Yi-VL](https://huggingface.co/01-ai)                    | 6B/34B                           | q_proj,v_proj     | yi_vl     |
+| [Yuan](https://huggingface.co/IEITYuan)                  | 2B/51B/102B                      | q_proj,v_proj     | yuan      |
 
 > [!NOTE]
-> **默认模块**应作为 `--lora_target` 参数的默认值，可使用 `--lora_target all` 参数指定全部模块。
+> **默认模块**应作为 `--lora_target` 参数的默认值，可使用 `--lora_target all` 参数指定全部模块以取得更好的效果。
 >
-> 对于所有“基座”（Base）模型，`--template` 参数可以是 `default`, `alpaca`, `vicuna` 等任意值。但“对话”（Chat）模型请务必使用**对应的模板**。
+> 对于所有“基座”（Base）模型，`--template` 参数可以是 `default`, `alpaca`, `vicuna` 等任意值。但“对话”（Instruct/Chat）模型请务必使用**对应的模板**。
+>
+> 请务必在训练和推理时使用**完全一致**的模板。
 
 项目所支持模型的完整列表请参阅 [constants.py](src/llmtuner/extras/constants.py)。
 
@@ -165,9 +189,7 @@ https://github.com/hiyouga/LLaMA-Factory/assets/16256802/ec36a9dd-37f4-4f72-81bd
 | 奖励模型训练            | :white_check_mark: | :white_check_mark: | :white_check_mark: | :white_check_mark: |
 | PPO 训练               | :white_check_mark: | :white_check_mark: | :white_check_mark: | :white_check_mark: |
 | DPO 训练               | :white_check_mark: | :white_check_mark: | :white_check_mark: | :white_check_mark: |
-
+| ORPO 训练              | :white_check_mark: | :white_check_mark: | :white_check_mark: | :white_check_mark: |
-> [!NOTE]
-> 请使用 `--quantization_bit 4` 参数来启用 QLoRA 训练。
 
 ## 数据集
 
@@ -190,8 +212,8 @@ https://github.com/hiyouga/LLaMA-Factory/assets/16256802/ec36a9dd-37f4-4f72-81bd
 - [Stanford Alpaca (en)](https://github.com/tatsu-lab/stanford_alpaca)
 - [Stanford Alpaca (zh)](https://github.com/ymcui/Chinese-LLaMA-Alpaca)
 - [Alpaca GPT4 (en&zh)](https://github.com/Instruction-Tuning-with-GPT-4/GPT-4-LLM)
-- [Self Cognition (zh)](data/self_cognition.json)
+- [Identity (en&zh)](data/identity.json)
-- [Open Assistant (multilingual)](https://huggingface.co/datasets/OpenAssistant/oasst1)
+- [Open Assistant (zh)](https://huggingface.co/datasets/OpenAssistant/oasst1)
 - [ShareGPT (zh)](https://huggingface.co/datasets/QingyiSi/Alpaca-CoT/tree/main/Chinese-instruction-collection)
 - [Guanaco Dataset (multilingual)](https://huggingface.co/datasets/JosephusCheung/GuanacoDataset)
 - [BELLE 2M (zh)](https://huggingface.co/datasets/BelleGroup/train_2M_CN)
@@ -223,6 +245,7 @@ https://github.com/hiyouga/LLaMA-Factory/assets/16256802/ec36a9dd-37f4-4f72-81bd
 - [Evol Instruct V2 (en)](https://huggingface.co/datasets/WizardLM/WizardLM_evol_instruct_V2_196k)
 - [Glaive Function Calling V2 (en)](https://huggingface.co/datasets/glaiveai/glaive-function-calling-v2)
 - [Cosmopedia (en)](https://huggingface.co/datasets/HuggingFaceTB/cosmopedia)
+- [LLaVA mixed (en&zh)](https://huggingface.co/datasets/BUAADreamer/llava-en-zh-300k)
 - [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)
@@ -238,16 +261,15 @@ https://github.com/hiyouga/LLaMA-Factory/assets/16256802/ec36a9dd-37f4-4f72-81bd
 <details><summary>偏好数据集</summary>
 
 - [HH-RLHF (en)](https://huggingface.co/datasets/Anthropic/hh-rlhf)
-- [Open Assistant (multilingual)](https://huggingface.co/datasets/OpenAssistant/oasst1)
 - [GPT-4 Generated Data (en&zh)](https://github.com/Instruction-Tuning-with-GPT-4/GPT-4-LLM)
 - [Orca DPO (en)](https://huggingface.co/datasets/Intel/orca_dpo_pairs)
 - [Nectar (en)](https://huggingface.co/datasets/berkeley-nest/Nectar)
+- [DPO mixed (en&zh)](https://huggingface.co/datasets/hiyouga/DPO-En-Zh-20k)
+- [Open Assistant (zh)](https://huggingface.co/datasets/OpenAssistant/oasst1)
 - [Orca DPO (de)](https://huggingface.co/datasets/mayflowergmbh/intel_orca_dpo_pairs_de)
 
 </details>
 
-使用方法请参考 [data/README_zh.md](data/README_zh.md) 文件。
-
 部分数据集的使用需要确认，我们推荐使用下述命令登录您的 Hugging Face 账户。
 
 ```bash
@@ -261,53 +283,54 @@ huggingface-cli login
 | ------------ | ------- | --------- |
 | python       | 3.8     | 3.10      |
 | torch        | 1.13.1  | 2.2.0     |
-| transformers | 4.37.2  | 4.39.1    |
+| transformers | 4.37.2  | 4.40.1    |
-| datasets     | 2.14.3  | 2.17.1    |
+| datasets     | 2.14.3  | 2.19.1    |
-| accelerate   | 0.27.2  | 0.28.0    |
+| accelerate   | 0.27.2  | 0.30.0    |
 | peft         | 0.9.0   | 0.10.0    |
-| trl          | 0.8.1   | 0.8.1     |
+| trl          | 0.8.1   | 0.8.6     |
 
 | 可选项       | 至少     | 推荐      |
 | ------------ | ------- | --------- |
 | CUDA         | 11.6    | 12.2      |
 | deepspeed    | 0.10.0  | 0.14.0    |
-| bitsandbytes | 0.39.0  | 0.43.0    |
+| bitsandbytes | 0.39.0  | 0.43.1    |
-| flash-attn   | 2.3.0   | 2.5.6     |
+| vllm         | 0.4.0   | 0.4.2     |
+| flash-attn   | 2.3.0   | 2.5.8     |
 
 ### 硬件依赖
 
 \* *估算值*
 
-| 训练方法 | 精度 |   7B  |  13B  |  30B  |   70B  |   8x7B |
+| 方法               | 精度 |   7B  |  13B  |  30B  |   70B  |  110B  |  8x7B |  8x22B |
-| ------- | ---- | ----- | ----- | ----- | ------ | ------ |
+| ----------------- | ---- | ----- | ----- | ----- | ------ | ------ | ----- | ------ |
-| 全参数   | AMP  | 120GB | 240GB | 600GB | 1200GB |  900GB |
+| Full              | AMP  | 120GB | 240GB | 600GB | 1200GB | 2000GB | 900GB | 2400GB |
-| 全参数   |  16  |  60GB | 120GB | 300GB |  600GB |  400GB |
+| Full              |  16  |  60GB | 120GB | 300GB |  600GB |  900GB | 400GB | 1200GB |
-| GaLore  |  16  |  16GB |  32GB |  64GB |  160GB |  120GB |
+| Freeze            |  16  |  20GB |  40GB |  80GB |  200GB |  360GB | 160GB |  400GB |
-| 部分参数 |  16  |  20GB |  40GB |  80GB |  200GB |  160GB |
+| LoRA/GaLore/BAdam |  16  |  16GB |  32GB |  64GB |  160GB |  240GB | 120GB |  320GB |
-| LoRA    |  16  |  16GB |  32GB |  64GB |  160GB |  120GB |
+| QLoRA             |   8  |  10GB |  20GB |  40GB |   80GB |  140GB |  60GB |  160GB |
-| QLoRA   |   8  |  10GB |  20GB |  40GB |   80GB |   60GB |
+| QLoRA             |   4  |   6GB |  12GB |  24GB |   48GB |   72GB |  30GB |   96GB |
-| QLoRA   |   4  |   6GB |  12GB |  24GB |   48GB |   30GB |
+| QLoRA             |   2  |   4GB |   8GB |  16GB |   24GB |   48GB |  18GB |   48GB |
-| QLoRA   |   2  |   4GB |   8GB |  16GB |   24GB |   18GB |
 
 ## 如何使用
 
-### 数据准备（可跳过）
+### 安装 LLaMA Factory
 
-关于数据集文件的格式，请参考 [data/README_zh.md](data/README_zh.md) 的内容。构建自定义数据集时，既可以使用单个 `.json` 文件，也可以使用一个[数据加载脚本](https://huggingface.co/docs/datasets/dataset_script)和多个文件。
+> [!IMPORTANT]
-
+> 此步骤为必需。
-> [!NOTE]
-> 使用自定义数据集时，请更新 `data/dataset_info.json` 文件，该文件的格式请参考 `data/README_zh.md`。
-
-### 环境搭建（可跳过）
 
 ```bash
 git clone https://github.com/hiyouga/LLaMA-Factory.git
-conda create -n llama_factory python=3.10
-conda activate llama_factory
 cd LLaMA-Factory
-pip install -r requirements.txt
+pip install -e .[torch,metrics]
 ```
 
+可选的额外依赖项：torch、metrics、deepspeed、bitsandbytes、vllm、galore、badam、gptq、awq、aqlm、qwen、modelscope、quality
+
+> [!TIP]
+> 遇到包冲突时，可使用 `pip install --no-deps -e .` 解决。
+
+<details><summary>Windows 用户指南</summary>
+
 如果要在 Windows 平台上开启量化 LoRA（QLoRA），需要安装预编译的 `bitsandbytes` 库, 支持 CUDA 11.1 到 12.2, 请根据您的 CUDA 版本情况选择适合的[发布版本](https://github.com/jllllll/bitsandbytes-windows-webui/releases/tag/wheels)。
 
 ```bash
@@ -316,356 +339,83 @@ pip install https://github.com/jllllll/bitsandbytes-windows-webui/releases/downl
 
 如果要在 Windows 平台上开启 FlashAttention-2，需要安装预编译的 `flash-attn` 库，支持 CUDA 12.1 到 12.2，请根据需求到 [flash-attention](https://github.com/bdashore3/flash-attention/releases) 下载对应版本安装。
 
-### 使用魔搭社区（可跳过）
+</details>
 
-如果您在 Hugging Face 模型和数据集的下载中遇到了问题，可以通过下述方法使用魔搭社区。
+<details><summary>昇腾 NPU 用户指南</summary>
+
+如果使用昇腾 NPU 设备进行（分布式）训练或推理，需要安装 **[torch-npu](https://gitee.com/ascend/pytorch)** 库和 **[Ascend CANN Kernels](https://www.hiascend.com/developer/download/community/result?module=cann)**。
+
+| 依赖项       | 至少     | 推荐      |
+| ------------ | ------- | --------- |
+| CANN         | 8.0.RC1 | 8.0.RC1   |
+| torch        | 2.2.0   | 2.2.0     |
+| torch-npu    | 2.2.0   | 2.2.0     |
+| deepspeed    | 0.13.2  | 0.13.2    |
+
+Docker 镜像：
+
+- 32GB：[下载地址](http://mirrors.cn-central-221.ovaijisuan.com/detail/130.html)
+- 64GB：敬请期待
+
+请记得使用 `ASCEND_RT_VISIBLE_DEVICES` 而非 `CUDA_VISIBLE_DEVICES` 来指定您使用的设备。
+
+如果遇到无法正常推理的情况，请尝试设置 `do_sample: false`。
+
+</details>
+
+### 数据准备
+
+关于数据集文件的格式，请参考 [data/README_zh.md](data/README_zh.md) 的内容。你可以使用 HuggingFace / ModelScope 上的数据集或加载本地数据集。
+
+> [!NOTE]
+> 使用自定义数据集时，请更新 `data/dataset_info.json` 文件。
+
+### 快速开始
+
+下面三行命令分别对 Llama3-8B-Instruct 模型进行 LoRA **微调**、**推理**和**合并**。
 
 ```bash
-export USE_MODELSCOPE_HUB=1 # Windows 使用 `set USE_MODELSCOPE_HUB=1`
+CUDA_VISIBLE_DEVICES=0 llamafactory-cli train examples/lora_single_gpu/llama3_lora_sft.yaml
+CUDA_VISIBLE_DEVICES=0 llamafactory-cli chat examples/inference/llama3_lora_sft.yaml
+CUDA_VISIBLE_DEVICES=0 llamafactory-cli export examples/merge_lora/llama3_lora_sft.yaml
 ```
 
-接着即可通过指定模型名称来训练对应的模型。（在[魔搭社区](https://modelscope.cn/models)查看所有可用的模型）
+高级用法请参考 [examples/README_zh.md](examples/README_zh.md)（包括多 GPU 微调）。
 
-```bash
+> [!TIP]
-CUDA_VISIBLE_DEVICES=0 python src/train_bash.py \
+> 使用 `llamafactory-cli help` 显示帮助信息。
-    --model_name_or_path modelscope/Llama-2-7b-ms \
-    ... # 参数同下
-```
 
-LLaMA Board 同样支持魔搭社区的模型和数据集下载。
+### LLaMA Board 可视化微调（由 [Gradio](https://github.com/gradio-app/gradio) 驱动）
-
-```bash
-CUDA_VISIBLE_DEVICES=0 USE_MODELSCOPE_HUB=1 python src/train_web.py
-```
-
-### 单 GPU 训练
 
 > [!IMPORTANT]
-> 如果您使用多张 GPU 训练模型，请移步[多 GPU 分布式训练](#多-gpu-分布式训练)部分。
+> LLaMA Board 可视化界面目前仅支持单 GPU 训练。
 
-#### LLaMA Board GUI
+#### 使用本地环境
 
 ```bash
-CUDA_VISIBLE_DEVICES=0 python src/train_web.py
+CUDA_VISIBLE_DEVICES=0 GRADIO_SHARE=1 llamafactory-cli webui
 ```
 
-#### 预训练
+<details><summary>阿里云 PAI 和 AutoDL 用户指南</summary>
+
+如果您在阿里云 PAI 上使用 LLaMA Board 时遇到显示问题，请尝试在启动前使用以下命令设置环境变量：
 
 ```bash
-CUDA_VISIBLE_DEVICES=0 python src/train_bash.py \
+export GRADIO_SERVER_PORT=7860 GRADIO_ROOT_PATH=/${JUPYTER_NAME}/proxy/7860/
-    --stage pt \
-    --do_train \
-    --model_name_or_path path_to_llama_model \
-    --dataset wiki_demo \
-    --finetuning_type lora \
-    --lora_target q_proj,v_proj \
-    --output_dir path_to_pt_checkpoint \
-    --overwrite_cache \
-    --per_device_train_batch_size 4 \
-    --gradient_accumulation_steps 4 \
-    --lr_scheduler_type cosine \
-    --logging_steps 10 \
-    --save_steps 1000 \
-    --learning_rate 5e-5 \
-    --num_train_epochs 3.0 \
-    --plot_loss \
-    --fp16
 ```
 
-#### 指令监督微调
+如果您正在使用 AutoDL，请安装下述 Gradio 版本：
 
 ```bash
-CUDA_VISIBLE_DEVICES=0 python src/train_bash.py \
+pip install gradio==4.10.0
-    --stage sft \
-    --do_train \
-    --model_name_or_path path_to_llama_model \
-    --dataset alpaca_gpt4_zh \
-    --template default \
-    --finetuning_type lora \
-    --lora_target q_proj,v_proj \
-    --output_dir path_to_sft_checkpoint \
-    --overwrite_cache \
-    --per_device_train_batch_size 4 \
-    --gradient_accumulation_steps 4 \
-    --lr_scheduler_type cosine \
-    --logging_steps 10 \
-    --save_steps 1000 \
-    --learning_rate 5e-5 \
-    --num_train_epochs 3.0 \
-    --plot_loss \
-    --fp16
-```
-
-#### 奖励模型训练
-
-```bash
-CUDA_VISIBLE_DEVICES=0 python src/train_bash.py \
-    --stage rm \
-    --do_train \
-    --model_name_or_path path_to_llama_model \
-    --adapter_name_or_path path_to_sft_checkpoint \
-    --create_new_adapter \
-    --dataset comparison_gpt4_zh \
-    --template default \
-    --finetuning_type lora \
-    --lora_target q_proj,v_proj \
-    --output_dir path_to_rm_checkpoint \
-    --per_device_train_batch_size 2 \
-    --gradient_accumulation_steps 4 \
-    --lr_scheduler_type cosine \
-    --logging_steps 10 \
-    --save_steps 1000 \
-    --learning_rate 1e-5 \
-    --num_train_epochs 1.0 \
-    --plot_loss \
-    --fp16
-```
-
-#### PPO 训练
-
-```bash
-CUDA_VISIBLE_DEVICES=0 python src/train_bash.py \
-    --stage ppo \
-    --do_train \
-    --model_name_or_path path_to_llama_model \
-    --adapter_name_or_path path_to_sft_checkpoint \
-    --create_new_adapter \
-    --dataset alpaca_gpt4_zh \
-    --template default \
-    --finetuning_type lora \
-    --lora_target q_proj,v_proj \
-    --reward_model path_to_rm_checkpoint \
-    --output_dir path_to_ppo_checkpoint \
-    --per_device_train_batch_size 2 \
-    --gradient_accumulation_steps 4 \
-    --lr_scheduler_type cosine \
-    --top_k 0 \
-    --top_p 0.9 \
-    --logging_steps 10 \
-    --save_steps 1000 \
-    --learning_rate 1e-5 \
-    --num_train_epochs 1.0 \
-    --plot_loss \
-    --fp16
-```
-
-> [!TIP]
-> 如果开启了 `--create_new_adapter`，则使用 `--adapter_name_or_path path_to_sft_checkpoint,path_to_ppo_checkpoint` 来进行微调模型的推理。
-
-> [!WARNING]
-> 如果使用 fp16 精度进行 LLaMA-2 模型的 PPO 训练，请使用 `--per_device_train_batch_size=1`。
-
-#### DPO 训练
-
-```bash
-CUDA_VISIBLE_DEVICES=0 python src/train_bash.py \
-    --stage dpo \
-    --do_train \
-    --model_name_or_path path_to_llama_model \
-    --adapter_name_or_path path_to_sft_checkpoint \
-    --create_new_adapter \
-    --dataset comparison_gpt4_zh \
-    --template default \
-    --finetuning_type lora \
-    --lora_target q_proj,v_proj \
-    --output_dir path_to_dpo_checkpoint \
-    --per_device_train_batch_size 2 \
-    --gradient_accumulation_steps 4 \
-    --lr_scheduler_type cosine \
-    --logging_steps 10 \
-    --save_steps 1000 \
-    --learning_rate 1e-5 \
-    --num_train_epochs 1.0 \
-    --plot_loss \
-    --fp16
-```
-
-> [!TIP]
-> 如果开启了 `--create_new_adapter`，则使用 `--adapter_name_or_path path_to_sft_checkpoint,path_to_dpo_checkpoint` 来进行微调模型的推理。
-
-### 多 GPU 分布式训练
-
-#### 使用 Huggingface Accelerate
-
-```bash
-accelerate launch --config_file config.yaml src/train_bash.py \
-    --ddp_timeout 180000000 \
-    ... # 参数同上
-```
-
-<details><summary>使用 Accelerate 进行 LoRA 训练的 config.yaml 示例</summary>
-
-```yaml
-compute_environment: LOCAL_MACHINE
-debug: false
-distributed_type: MULTI_GPU
-downcast_bf16: 'no'
-gpu_ids: all
-machine_rank: 0
-main_training_function: main
-mixed_precision: fp16
-num_machines: 1
-num_processes: 4
-rdzv_backend: static
-same_network: true
-tpu_env: []
-tpu_use_cluster: false
-tpu_use_sudo: false
-use_cpu: false
 ```
 
 </details>
 
-> [!TIP]
-> 我们推荐使用 Accelerate 进行 LoRA 训练。
-
-#### 使用 DeepSpeed
-
-```bash
-deepspeed --num_gpus 8 src/train_bash.py \
-    --deepspeed ds_config.json \
-    --ddp_timeout 180000000 \
-    ... # 参数同上
-```
-
-<details><summary>使用 DeepSpeed ZeRO-2 进行全参数训练的 ds_config.json 示例</summary>
-
-```json
-{
-  "train_batch_size": "auto",
-  "train_micro_batch_size_per_gpu": "auto",
-  "gradient_accumulation_steps": "auto",
-  "gradient_clipping": "auto",
-  "zero_allow_untested_optimizer": true,
-  "fp16": {
-    "enabled": "auto",
-    "loss_scale": 0,
-    "loss_scale_window": 1000,
-    "initial_scale_power": 16,
-    "hysteresis": 2,
-    "min_loss_scale": 1
-  },
-  "bf16": {
-    "enabled": "auto"
-  },
-  "zero_optimization": {
-    "stage": 2,
-    "allgather_partitions": true,
-    "allgather_bucket_size": 5e8,
-    "overlap_comm": true,
-    "reduce_scatter": true,
-    "reduce_bucket_size": 5e8,
-    "contiguous_gradients": true,
-    "round_robin_gradients": true
-  }
-}
-```
-
-</details>
-
-> [!TIP]
-> 更多训练脚本请查看 [examples](examples)。
-
-### 合并 LoRA 权重并导出模型
-
-```bash
-CUDA_VISIBLE_DEVICES= python src/export_model.py \
-    --model_name_or_path path_to_llama_model \
-    --adapter_name_or_path path_to_checkpoint \
-    --template default \
-    --finetuning_type lora \
-    --export_dir path_to_export \
-    --export_size 2 \
-    --export_legacy_format False
-```
-
-> [!WARNING]
-> 尚不支持量化模型的 LoRA 权重合并及导出。
-
-> [!TIP]
-> 仅使用 `--model_name_or_path path_to_export` 来加载导出后的模型。
-> 
-> 合并 LoRA 权重之后可再次使用 `CUDA_VISIBLE_DEVICES=0`、`--export_quantization_bit 4` 和 `--export_quantization_dataset data/c4_demo.json` 基于 AutoGPTQ 量化模型。
-
-### 使用 OpenAI 风格 API 推理
-
-```bash
-CUDA_VISIBLE_DEVICES=0 API_PORT=8000 python src/api_demo.py \
-    --model_name_or_path path_to_llama_model \
-    --adapter_name_or_path path_to_checkpoint \
-    --template default \
-    --finetuning_type lora
-```
-
-> [!TIP]
-> 关于 API 文档请见 `http://localhost:8000/docs`。
-
-### 使用命令行推理
-
-```bash
-CUDA_VISIBLE_DEVICES=0 python src/cli_demo.py \
-    --model_name_or_path path_to_llama_model \
-    --adapter_name_or_path path_to_checkpoint \
-    --template default \
-    --finetuning_type lora
-```
-
-### 使用浏览器推理
-
-```bash
-CUDA_VISIBLE_DEVICES=0 python src/web_demo.py \
-    --model_name_or_path path_to_llama_model \
-    --adapter_name_or_path path_to_checkpoint \
-    --template default \
-    --finetuning_type lora
-```
-
-### 模型评估
-
-```bash
-CUDA_VISIBLE_DEVICES=0 python src/evaluate.py \
-    --model_name_or_path path_to_llama_model \
-    --adapter_name_or_path path_to_checkpoint \
-    --template vanilla \
-    --finetuning_type lora \
-    --task ceval \
-    --split validation \
-    --lang zh \
-    --n_shot 5 \
-    --batch_size 4
-```
-
-### 模型预测
-
-```bash
-CUDA_VISIBLE_DEVICES=0 python src/train_bash.py \
-    --stage sft \
-    --do_predict \
-    --model_name_or_path path_to_llama_model \
-    --adapter_name_or_path path_to_checkpoint \
-    --dataset alpaca_gpt4_zh \
-    --template default \
-    --finetuning_type lora \
-    --output_dir path_to_predict_result \
-    --per_device_eval_batch_size 1 \
-    --max_samples 100 \
-    --predict_with_generate \
-    --fp16
-```
-
-> [!WARNING]
-> 如果使用 fp16 精度进行 LLaMA-2 模型的预测，请使用 `--per_device_eval_batch_size=1`。
-
-> [!TIP]
-> 我们建议在量化模型的预测中使用 `--per_device_eval_batch_size=1` 和 `--max_target_length 128`。
-
-### 使用容器
-
 #### 使用 Docker
 
 ```bash
 docker build -f ./Dockerfile -t llama-factory:latest .
-
 docker run --gpus=all \
     -v ./hf_cache:/root/.cache/huggingface/ \
     -v ./data:/app/data \
@@ -683,14 +433,36 @@ docker run --gpus=all \
 docker compose -f ./docker-compose.yml up -d
 ```
 
-> [!TIP]
+<details><summary>数据卷详情</summary>
-> 数据卷详情：
+
-> * hf_cache：使用宿主机的 Hugging Face 缓存文件夹，允许更改为新的目录。
+- hf_cache：使用宿主机的 Hugging Face 缓存文件夹，允许更改为新的目录。
-> * data：宿主机中存放数据集的文件夹路径。
+- data：宿主机中存放数据集的文件夹路径。
-> * output：将导出目录设置为该路径后，即可在宿主机中访问导出后的模型。
+- output：将导出目录设置为该路径后，即可在宿主机中访问导出后的模型。
+
+</details>
+
+### 利用 vLLM 部署 OpenAI API
+
+```bash
+CUDA_VISIBLE_DEVICES=0,1 API_PORT=8000 llamafactory-cli api examples/inference/llama3_vllm.yaml
+```
+
+### 从魔搭社区下载
+
+如果您在 Hugging Face 模型和数据集的下载中遇到了问题，可以通过下述方法使用魔搭社区。
+
+```bash
+export USE_MODELSCOPE_HUB=1 # Windows 使用 `set USE_MODELSCOPE_HUB=1`
+```
+
+将 `--model_name_or_path` 设置为模型 ID 来加载对应的模型。在[魔搭社区](https://modelscope.cn/models)查看所有可用的模型，例如 `LLM-Research/Meta-Llama-3-8B-Instruct`。
 
 ## 使用了 LLaMA Factory 的项目
 
+如果您有项目希望添加至下述列表，请通过邮件联系或者创建一个 PR。
+
+<details><summary>点击显示</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)
@@ -712,21 +484,37 @@ docker compose -f ./docker-compose.yml up -d
 1. Huang et al. Key-Point-Driven Data Synthesis with its Enhancement on Mathematical Reasoning. 2024. [[arxiv]](https://arxiv.org/abs/2403.02333)
 1. Duan et al. Negating Negatives: Alignment without Human Positive Samples via Distributional Dispreference Optimization. 2024. [[arxiv]](https://arxiv.org/abs/2403.03419)
 1. Xie and Schwertfeger. Empowering Robotics with Large Language Models: osmAG Map Comprehension with LLMs. 2024. [[arxiv]](https://arxiv.org/abs/2403.08228)
+1. Wu et al. Large Language Models are Parallel Multilingual Learners. 2024. [[arxiv]](https://arxiv.org/abs/2403.09073)
+1. Zhang et al. EDT: Improving Large Language Models' Generation by Entropy-based Dynamic Temperature Sampling. 2024. [[arxiv]](https://arxiv.org/abs/2403.14541)
+1. Weller et al. FollowIR: Evaluating and Teaching Information Retrieval Models to Follow Instructions. 2024. [[arxiv]](https://arxiv.org/abs/2403.15246)
 1. Hongbin Na. CBT-LLM: A Chinese Large Language Model for Cognitive Behavioral Therapy-based Mental Health Question Answering. 2024. [[arxiv]](https://arxiv.org/abs/2403.16008)
+1. Zan et al. CodeS: Natural Language to Code Repository via Multi-Layer Sketch. 2024. [[arxiv]](https://arxiv.org/abs/2403.16443)
+1. Liu et al. Extensive Self-Contrast Enables Feedback-Free Language Model Alignment. 2024. [[arxiv]](https://arxiv.org/abs/2404.00604)
+1. Luo et al. BAdam: A Memory Efficient Full Parameter Training Method for Large Language Models. 2024. [[arxiv]](https://arxiv.org/abs/2404.02827)
+1. Du et al. Chinese Tiny LLM: Pretraining a Chinese-Centric Large Language Model. 2024. [[arxiv]](https://arxiv.org/abs/2404.04167)
+1. Ma et al. Parameter Efficient Quasi-Orthogonal Fine-Tuning via Givens Rotation. 2024. [[arxiv]](https://arxiv.org/abs/2404.04316)
+1. Liu et al. Dynamic Generation of Personalities with Large Language Models. 2024. [[arxiv]](https://arxiv.org/abs/2404.07084)
+1. Shang et al. How Far Have We Gone in Stripped Binary Code Understanding Using Large Language Models. 2024. [[arxiv]](https://arxiv.org/abs/2404.09836)
+1. Huang et al. LLMTune: Accelerate Database Knob Tuning with Large Language Models. 2024. [[arxiv]](https://arxiv.org/abs/2404.11581)
+1. Deng et al. Text-Tuple-Table: Towards Information Integration in Text-to-Table Generation via Global Tuple Extraction. 2024. [[arxiv]](https://arxiv.org/abs/2404.14215)
+1. Acikgoz et al. Hippocrates: An Open-Source Framework for Advancing Large Language Models in Healthcare. 2024. [[arxiv]](https://arxiv.org/abs/2404.16621)
+1. Zhang et al. Small Language Models Need Strong Verifiers to Self-Correct Reasoning. 2024. [[arxiv]](https://arxiv.org/abs/2404.17140)
+1. Zhou et al. FREB-TQA: A Fine-Grained Robustness Evaluation Benchmark for Table Question Answering. 2024. [[arxiv]](https://arxiv.org/abs/2404.18585)
 1. **[StarWhisper](https://github.com/Yu-Yang-Li/StarWhisper)**: 天文大模型 StarWhisper，基于 ChatGLM2-6B 和 Qwen-14B 在天文数据上微调而得。
 1. **[DISC-LawLLM](https://github.com/FudanDISC/DISC-LawLLM)**: 中文法律领域大模型 DISC-LawLLM，基于 Baichuan-13B 微调而得，具有法律推理和知识检索能力。
 1. **[Sunsimiao](https://github.com/thomas-yanxin/Sunsimiao)**: 孙思邈中文医疗大模型 Sumsimiao，基于 Baichuan-7B 和 ChatGLM-6B 在中文医疗数据上微调而得。
 1. **[CareGPT](https://github.com/WangRongsheng/CareGPT)**: 医疗大模型项目 CareGPT，基于 LLaMA2-7B 和 Baichuan-13B 在中文医疗数据上微调而得。
 1. **[MachineMindset](https://github.com/PKU-YuanGroup/Machine-Mindset/)**：MBTI性格大模型项目，根据数据集与训练方式让任意 LLM 拥有 16 个不同的性格类型。
+1. **[Luminia-13B-v3](https://huggingface.co/Nekochu/Luminia-13B-v3)**：一个用于生成 Stable Diffusion 提示词的大型语言模型。[[🤗Demo]](https://huggingface.co/spaces/Nekochu/Luminia-13B_SD_Prompt)
+1. **[Chinese-LLaVA-Med](https://github.com/BUAADreamer/Chinese-LLaVA-Med)**：中文多模态医学大模型，基于 LLaVA-1.5-7B 在中文多模态医疗数据上微调而得。
 
-> [!TIP]
+</details>
-> 如果您有项目希望添加至上述列表，请通过邮件联系或者创建一个 PR。
 
 ## 协议
 
 本仓库的代码依照 [Apache-2.0](LICENSE) 协议开源。
 
-使用模型权重时，请遵循对应的模型协议：[Baichuan2](https://huggingface.co/baichuan-inc/Baichuan2-7B-Base/blob/main/Community%20License%20for%20Baichuan%202%20Model.pdf) / [BLOOM](https://huggingface.co/spaces/bigscience/license) / [ChatGLM3](https://github.com/THUDM/ChatGLM3/blob/main/MODEL_LICENSE) / [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) / [InternLM2](https://github.com/InternLM/InternLM#license) / [LLaMA](https://github.com/facebookresearch/llama/blob/main/MODEL_CARD.md) / [LLaMA-2](https://ai.meta.com/llama/license/) / [Mistral](LICENSE) / [OLMo](LICENSE) / [Phi-1.5/2](https://huggingface.co/microsoft/phi-1_5/resolve/main/Research%20License.docx) / [Qwen](https://github.com/QwenLM/Qwen/blob/main/Tongyi%20Qianwen%20LICENSE%20AGREEMENT) / [StarCoder2](https://huggingface.co/spaces/bigcode/bigcode-model-license-agreement) / [XVERSE](https://github.com/xverse-ai/XVERSE-13B/blob/main/MODEL_LICENSE.pdf) / [Yi](https://huggingface.co/01-ai/Yi-6B/blob/main/LICENSE) / [Yuan](https://github.com/IEIT-Yuan/Yuan-2.0/blob/main/LICENSE-Yuan)
+使用模型权重时，请遵循对应的模型协议：[Baichuan2](https://huggingface.co/baichuan-inc/Baichuan2-7B-Base/blob/main/Community%20License%20for%20Baichuan%202%20Model.pdf) / [BLOOM](https://huggingface.co/spaces/bigscience/license) / [ChatGLM3](https://github.com/THUDM/ChatGLM3/blob/main/MODEL_LICENSE) / [Command-R](https://cohere.com/c4ai-cc-by-nc-license) / [DeepSeek](https://github.com/deepseek-ai/DeepSeek-LLM/blob/main/LICENSE-MODEL) / [Falcon](https://huggingface.co/tiiuae/falcon-180B/blob/main/LICENSE.txt) / [Gemma](https://ai.google.dev/gemma/terms) / [InternLM2](https://github.com/InternLM/InternLM#license) / [LLaMA](https://github.com/facebookresearch/llama/blob/main/MODEL_CARD.md) / [LLaMA-2 (LLaVA-1.5)](https://ai.meta.com/llama/license/) / [LLaMA-3](https://llama.meta.com/llama3/license/) / [Mistral](LICENSE) / [OLMo](LICENSE) / [Phi-1.5/2](https://huggingface.co/microsoft/phi-1_5/resolve/main/Research%20License.docx) / [Phi-3](https://huggingface.co/microsoft/Phi-3-mini-4k-instruct/blob/main/LICENSE) / [Qwen](https://github.com/QwenLM/Qwen/blob/main/Tongyi%20Qianwen%20LICENSE%20AGREEMENT) / [StarCoder2](https://huggingface.co/spaces/bigcode/bigcode-model-license-agreement) / [XVERSE](https://github.com/xverse-ai/XVERSE-13B/blob/main/MODEL_LICENSE.pdf) / [Yi](https://huggingface.co/01-ai/Yi-6B/blob/main/LICENSE) / [Yi-1.5](LICENSE) / [Yuan](https://github.com/IEIT-Yuan/Yuan-2.0/blob/main/LICENSE-Yuan)
 
 ## 引用
 
@@ -744,7 +532,7 @@ docker compose -f ./docker-compose.yml up -d
 
 ## 致谢
 
-本项目受益于 [PEFT](https://github.com/huggingface/peft)、[QLoRA](https://github.com/artidoro/qlora) 和 [FastChat](https://github.com/lm-sys/FastChat)，感谢以上诸位作者的付出。
+本项目受益于 [PEFT](https://github.com/huggingface/peft)、[TRL](https://github.com/huggingface/trl)、[QLoRA](https://github.com/artidoro/qlora) 和 [FastChat](https://github.com/lm-sys/FastChat)，感谢以上诸位作者的付出。
 
 ## Star History
 

data/README.md

@@ -1,4 +1,4 @@
-If you are using a custom dataset, please provide your dataset definition in the following format in `dataset_info.json`.
+If you are using a custom dataset, please add your **dataset description** to `dataset_info.json` according to the following format. We also provide several examples in the next section.
 
 ```json
 "dataset_name": {
@@ -18,7 +18,8 @@ If you are using a custom dataset, please provide your dataset definition in the
     "history": "the column name in the dataset containing the histories. (default: None)",
     "messages": "the column name in the dataset containing the messages. (default: conversations)",
     "system": "the column name in the dataset containing the system prompts. (default: None)",
-    "tools": "the column name in the dataset containing the tool description. (default: None)"
+    "tools": "the column name in the dataset containing the tool description. (default: None)",
+    "images": "the column name in the dataset containing the image inputs. (default: None)"
   },
   "tags (optional, used for the sharegpt format)": {
     "role_tag": "the key in the message represents the identity. (default: from)",
@@ -32,7 +33,9 @@ If you are using a custom dataset, please provide your dataset definition in the
 }
 ```
 
-Given above, you can use the custom dataset via specifying `--dataset dataset_name`.
+After that, you can load the custom dataset by specifying `--dataset dataset_name`.
+
+----
 
 Currently we support dataset in **alpaca** or **sharegpt** format, the dataset in alpaca format should follow the below format:
 
@@ -51,10 +54,11 @@ Currently we support dataset in **alpaca** or **sharegpt** format, the dataset i
 ]
 ```
 
-Regarding the above dataset, the `columns` in `dataset_info.json` should be:
+Regarding the above dataset, the description in `dataset_info.json` should be:
 
 ```json
 "dataset_name": {
+  "file_name": "data.json",
   "columns": {
     "prompt": "instruction",
     "query": "input",
@@ -67,24 +71,60 @@ Regarding the above dataset, the `columns` in `dataset_info.json` should be:
 
 The `query` column will be concatenated with the `prompt` column and used as the user prompt, then the user prompt would be `prompt\nquery`. The `response` column represents the model response.
 
-The `system` column will be used as the system prompt. The `history` column is a list consisting string tuples representing prompt-response pairs in the history. Note that the responses in the history **will also be used for training**.
+The `system` column will be used as the system prompt. The `history` column is a list consisting string tuples representing prompt-response pairs in the history. Note that the responses in the history **will also be used for training** in supervised fine-tuning.
 
-For the pre-training datasets, only the `prompt` column will be used for training.
+For the **pre-training datasets**, only the `prompt` column will be used for training, for example:
-
-For the preference datasets, the `response` column should be a string list whose length is 2, with the preferred answers appearing first, for example:
 
 ```json
-{
+[
-  "instruction": "user instruction",
+  {"text": "document"},
-  "input": "user input",
+  {"text": "document"}
-  "output": [
+]
-    "chosen answer",
+```
-    "rejected answer"
+
-  ]
+Regarding the above dataset, the description in `dataset_info.json` should be:
+
+```json
+"dataset_name": {
+  "file_name": "data.json",
+  "columns": {
+    "prompt": "text"
+  }
 }
 ```
 
-The dataset in sharegpt format should follow the below format:
+For the **preference datasets**, the `response` column should be a string list whose length is 2, with the preferred answers appearing first, for example:
+
+```json
+[
+  {
+    "instruction": "user instruction",
+    "input": "user input",
+    "output": [
+      "chosen answer",
+      "rejected answer"
+    ]
+  }
+]
+```
+
+Regarding the above dataset, the description in `dataset_info.json` should be:
+
+```json
+"dataset_name": {
+  "file_name": "data.json",
+  "ranking": true,
+  "columns": {
+    "prompt": "instruction",
+    "query": "input",
+    "response": "output",
+  }
+}
+```
+
+----
+
+The dataset in **sharegpt** format should follow the below format:
 
 ```json
 [
@@ -105,10 +145,12 @@ The dataset in sharegpt format should follow the below format:
 ]
 ```
 
-Regarding the above dataset, the `columns` in `dataset_info.json` should be:
+Regarding the above dataset, the description in `dataset_info.json` should be:
 
 ```json
 "dataset_name": {
+  "file_name": "data.json",
+  "formatting": "sharegpt",
   "columns": {
     "messages": "conversations",
     "system": "system",
@@ -125,4 +167,46 @@ Regarding the above dataset, the `columns` in `dataset_info.json` should be:
 
 where the `messages` column should be a list following the `u/a/u/a/u/a` order.
 
-Pre-training datasets and preference datasets are incompatible with the sharegpt format yet.
+We also supports the dataset in the **openai** format:
+
+```json
+[
+  {
+    "messages": [
+      {
+        "role": "system",
+        "content": "system prompt (optional)"
+      },
+      {
+        "role": "user",
+        "content": "user instruction"
+      },
+      {
+        "role": "assistant",
+        "content": "model response"
+      }
+    ]
+  }
+]
+```
+
+Regarding the above dataset, the description in `dataset_info.json` should be:
+
+```json
+"dataset_name": {
+  "file_name": "data.json",
+  "formatting": "sharegpt",
+  "columns": {
+    "messages": "messages"
+  },
+  "tags": {
+    "role_tag": "role",
+    "content_tag": "content",
+    "user_tag": "user",
+    "assistant_tag": "assistant",
+    "system_tag": "system"
+  }
+}
+```
+
+Pre-training datasets and preference datasets are **incompatible** with the sharegpt format yet.

data/README_zh.md

@@ -1,4 +1,4 @@
-如果您使用自定义数据集，请务必在 `dataset_info.json` 文件中按照以下格式提供数据集定义。
+如果您使用自定义数据集，请务必按照以下格式在 `dataset_info.json` 文件中添加**数据集描述**。我们在下面也提供了一些例子。
 
 ```json
 "数据集名称": {
@@ -18,7 +18,8 @@
     "history": "数据集代表历史对话的表头名称（默认：None）",
     "messages": "数据集代表消息列表的表头名称（默认：conversations）",
     "system": "数据集代表系统提示的表头名称（默认：None）",
-    "tools": "数据集代表工具描述的表头名称（默认：None）"
+    "tools": "数据集代表工具描述的表头名称（默认：None）",
+    "images": "数据集代表图像输入的表头名称（默认：None）"
   },
   "tags（可选，用于 sharegpt 格式）": {
     "role_tag": "消息中代表发送者身份的键名（默认：from）",
@@ -32,7 +33,9 @@
 }
 ```
 
-添加后可通过指定 `--dataset 数据集名称` 参数使用自定义数据集。
+然后，可通过使用 `--dataset 数据集名称` 参数加载自定义数据集。
+
+----
 
 该项目目前支持两种格式的数据集：**alpaca** 和 **sharegpt**，其中 alpaca 格式的数据集按照以下方式组织：
 
@@ -51,10 +54,11 @@
 ]
 ```
 
-对于上述格式的数据，`dataset_info.json` 中的 `columns` 应为：
+对于上述格式的数据，`dataset_info.json` 中的描述应为：
 
 ```json
 "数据集名称": {
+  "file_name": "data.json",
   "columns": {
     "prompt": "instruction",
     "query": "input",
@@ -67,24 +71,60 @@
 
 其中 `query` 列对应的内容会与 `prompt` 列对应的内容拼接后作为用户指令，即用户指令为 `prompt\nquery`。`response` 列对应的内容为模型回答。
 
-`system` 列对应的内容将被作为系统提示词。`history` 列是由多个字符串二元组构成的列表，分别代表历史消息中每轮的指令和回答。注意历史消息中的回答**也会被用于训练**。
+`system` 列对应的内容将被作为系统提示词。`history` 列是由多个字符串二元组构成的列表，分别代表历史消息中每轮的指令和回答。注意在指令监督学习时，历史消息中的回答**也会被用于训练**。
 
-对于预训练数据集，仅 `prompt` 列中的内容会用于模型训练。
+对于**预训练数据集**，仅 `prompt` 列中的内容会用于模型训练，例如：
-
-对于偏好数据集，`response` 列应当是一个长度为 2 的字符串列表，排在前面的代表更优的回答，例如：
 
 ```json
-{
+[
-  "instruction": "用户指令",
+  {"text": "document"},
-  "input": "用户输入",
+  {"text": "document"}
-  "output": [
+]
-    "优质回答",
+```
-    "劣质回答"
+
-  ]
+对于上述格式的数据，`dataset_info.json` 中的描述应为：
+
+```json
+"数据集名称": {
+  "file_name": "data.json",
+  "columns": {
+    "prompt": "text"
+  }
 }
 ```
 
-而 sharegpt 格式的数据集按照以下方式组织：
+对于**偏好数据集**，`response` 列应当是一个长度为 2 的字符串列表，排在前面的代表更优的回答，例如：
+
+```json
+[
+  {
+    "instruction": "用户指令",
+    "input": "用户输入",
+    "output": [
+      "优质回答",
+      "劣质回答"
+    ]
+  }
+]
+```
+
+对于上述格式的数据，`dataset_info.json` 中的描述应为：
+
+```json
+"数据集名称": {
+  "file_name": "data.json",
+  "ranking": true,
+  "columns": {
+    "prompt": "instruction",
+    "query": "input",
+    "response": "output",
+  }
+}
+```
+
+----
+
+而 **sharegpt** 格式的数据集按照以下方式组织：
 
 ```json
 [
@@ -105,10 +145,12 @@
 ]
 ```
 
-对于上述格式的数据，`dataset_info.json` 中的 `columns` 应为：
+对于上述格式的数据，`dataset_info.json` 中的描述应为：
 
 ```json
 "数据集名称": {
+  "file_name": "data.json",
+  "formatting": "sharegpt",
   "columns": {
     "messages": "conversations",
     "system": "system",
@@ -125,4 +167,46 @@
 
 其中 `messages` 列应当是一个列表，且符合 `用户/模型/用户/模型/用户/模型` 的顺序。
 
-预训练数据集和偏好数据集尚不支持 sharegpt 格式。
+我们同样支持 **openai** 格式的数据集：
+
+```json
+[
+  {
+    "messages": [
+      {
+        "role": "system",
+        "content": "系统提示词（选填）"
+      },
+      {
+        "role": "user",
+        "content": "用户指令"
+      },
+      {
+        "role": "assistant",
+        "content": "模型回答"
+      }
+    ]
+  }
+]
+```
+
+对于上述格式的数据，`dataset_info.json` 中的描述应为：
+
+```json
+"数据集名称": {
+  "file_name": "data.json",
+  "formatting": "sharegpt",
+  "columns": {
+    "messages": "messages"
+  },
+  "tags": {
+    "role_tag": "role",
+    "content_tag": "content",
+    "user_tag": "user",
+    "assistant_tag": "assistant",
+    "system_tag": "system"
+  }
+}
+```
+
+预训练数据集和偏好数据集**尚不支持** sharegpt 格式。

data/alpaca_data_zh_51k.json.REMOVED.git-id

@@ -1 +1 @@
-34c723573fbc2d7601f6d9c882ccf5aa4f9bcc4b
+a97cf9475291591843976554878568e046d8a46d

data/belle_multiturn/belle_multiturn.py

@@ -1,5 +1,6 @@
-import os
 import json
+import os
+
 import datasets
 
 
@@ -22,31 +23,19 @@ _URL = "{}/datasets/BelleGroup/multiturn_chat_0.8M/resolve/main/multiturn_chat_0
 
 
 class BelleMultiturn(datasets.GeneratorBasedBuilder):
-
     VERSION = datasets.Version("0.0.0")
 
     def _info(self):
-        features = datasets.Features({
+        features = datasets.Features(
-            "conversations": [{"from": datasets.Value("string"), "value": datasets.Value("string")}]
+            {"conversations": [{"from": datasets.Value("string"), "value": datasets.Value("string")}]}
-        })
+        )
         return datasets.DatasetInfo(
-            description=_DESCRIPTION,
+            description=_DESCRIPTION, features=features, homepage=_HOMEPAGE, license=_LICENSE, citation=_CITATION
-            features=features,
-            homepage=_HOMEPAGE,
-            license=_LICENSE,
-            citation=_CITATION
         )
 
     def _split_generators(self, dl_manager: datasets.DownloadManager):
         file_path = dl_manager.download(_URL)
-        return [
+        return [datasets.SplitGenerator(name=datasets.Split.TRAIN, gen_kwargs={"filepath": file_path})]
-            datasets.SplitGenerator(
-                name=datasets.Split.TRAIN,
-                gen_kwargs={
-                    "filepath": file_path
-                }
-            )
-        ]
 
     def _generate_examples(self, filepath: str):
         with open(filepath, "r", encoding="utf-8") as f:
@@ -58,7 +47,7 @@ class BelleMultiturn(datasets.GeneratorBasedBuilder):
 
                 assist_idx = prompt.rfind("Assistant:")
                 human_idx = prompt.rfind("Human:")
-                query = prompt[human_idx+6:assist_idx].strip()
+                query = prompt[human_idx + 6 : assist_idx].strip()
                 prompt = prompt[:human_idx].strip()
                 conversations.insert(0, {"from": "gpt", "value": response})
                 conversations.insert(0, {"from": "human", "value": query})
@@ -67,8 +56,8 @@ class BelleMultiturn(datasets.GeneratorBasedBuilder):
                     assist_idx = prompt.rfind("Assistant:")
                     human_idx = prompt.rfind("Human:")
                     if human_idx != -1:
-                        old_query = prompt[human_idx+6:assist_idx].strip()
+                        old_query = prompt[human_idx + 6 : assist_idx].strip()
-                        old_resp = prompt[assist_idx+10:].strip()
+                        old_resp = prompt[assist_idx + 10 :].strip()
                         conversations.insert(0, {"from": "gpt", "value": old_resp})
                         conversations.insert(0, {"from": "human", "value": old_query})
                     else:

data/example_dataset/example_dataset.py

@@ -1,7 +1,8 @@
 import json
-import datasets
 from typing import Any, Dict, Generator, List, Tuple
 
+import datasets
+
 
 _DESCRIPTION = "An example of dataset."
 _CITATION = ""
@@ -11,34 +12,24 @@ _URL = "examples.json"
 
 
 class ExampleDataset(datasets.GeneratorBasedBuilder):
-
     VERSION = datasets.Version("0.0.0")
 
     def _info(self) -> datasets.DatasetInfo:
-        features = datasets.Features({
+        features = datasets.Features(
-            "instruction": datasets.Value("string"),
+            {
-            "input": datasets.Value("string"),
+                "instruction": datasets.Value("string"),
-            "output": datasets.Value("string"),
+                "input": datasets.Value("string"),
-            "history": datasets.Sequence(datasets.Sequence(datasets.Value("string")))
+                "output": datasets.Value("string"),
-        })
+                "history": datasets.Sequence(datasets.Sequence(datasets.Value("string"))),
+            }
+        )
         return datasets.DatasetInfo(
-            description=_DESCRIPTION,
+            description=_DESCRIPTION, features=features, homepage=_HOMEPAGE, license=_LICENSE, citation=_CITATION
-            features=features,
-            homepage=_HOMEPAGE,
-            license=_LICENSE,
-            citation=_CITATION
         )
 
     def _split_generators(self, dl_manager: datasets.DownloadManager) -> List[datasets.SplitGenerator]:
         file_path = dl_manager.download(_URL)
-        return [
+        return [datasets.SplitGenerator(name=datasets.Split.TRAIN, gen_kwargs={"filepath": file_path})]
-            datasets.SplitGenerator(
-                name=datasets.Split.TRAIN,
-                gen_kwargs={
-                    "filepath": file_path
-                }
-            )
-        ]
 
     def _generate_examples(self, filepath: str) -> Generator[Tuple[int, Dict[str, Any]], None, None]:
         example_dataset = json.load(open(filepath, "r", encoding="utf-8"))

data/hh_rlhf_en/hh_rlhf_en.py

@@ -1,8 +1,10 @@
-import os
 import json
-import datasets
+import os
 from typing import List
 
+import datasets
+
+
 _HF_ENDPOINT = os.getenv("HF_ENDPOINT", "https://huggingface.co")
 _DESCRIPTION = "Human preference data about helpfulness and harmlessness."
 _CITATION = ""
@@ -14,50 +16,37 @@ _URLS = {
         _URL + "harmless-base/train.jsonl.gz",
         _URL + "helpful-base/train.jsonl.gz",
         _URL + "helpful-online/train.jsonl.gz",
-        _URL + "helpful-rejection-sampled/train.jsonl.gz"
+        _URL + "helpful-rejection-sampled/train.jsonl.gz",
     ],
     "test": [
         _URL + "harmless-base/test.jsonl.gz",
         _URL + "helpful-base/test.jsonl.gz",
         _URL + "helpful-online/test.jsonl.gz",
-        _URL + "helpful-rejection-sampled/test.jsonl.gz"
+        _URL + "helpful-rejection-sampled/test.jsonl.gz",
-    ]
+    ],
 }
 
 
 class HhRlhfEn(datasets.GeneratorBasedBuilder):
-
     VERSION = datasets.Version("0.0.0")
 
     def _info(self) -> datasets.DatasetInfo:
-        features = datasets.Features({
+        features = datasets.Features(
-            "instruction": datasets.Value("string"),
+            {
-            "output": datasets.Sequence(datasets.Value("string")),
+                "instruction": datasets.Value("string"),
-            "history": datasets.Sequence(datasets.Sequence(datasets.Value("string")))
+                "output": datasets.Sequence(datasets.Value("string")),
-        })
+                "history": datasets.Sequence(datasets.Sequence(datasets.Value("string"))),
+            }
+        )
         return datasets.DatasetInfo(
-            description=_DESCRIPTION,
+            description=_DESCRIPTION, features=features, homepage=_HOMEPAGE, license=_LICENSE, citation=_CITATION
-            features=features,
-            homepage=_HOMEPAGE,
-            license=_LICENSE,
-            citation=_CITATION
         )
 
     def _split_generators(self, dl_manager: datasets.DownloadManager):
         file_path = dl_manager.download_and_extract(_URLS)
         return [
-            datasets.SplitGenerator(
+            datasets.SplitGenerator(name=datasets.Split.TRAIN, gen_kwargs={"filepaths": file_path["train"]}),
-                name=datasets.Split.TRAIN,
+            datasets.SplitGenerator(name=datasets.Split.TEST, gen_kwargs={"filepaths": file_path["test"]}),
-                gen_kwargs={
-                    "filepaths": file_path["train"]
-                }
-            ),
-            datasets.SplitGenerator(
-                name=datasets.Split.TEST,
-                gen_kwargs={
-                    "filepaths": file_path["test"]
-                }
-            )
         ]
 
     def _generate_examples(self, filepaths: List[str]):
@@ -70,12 +59,12 @@ class HhRlhfEn(datasets.GeneratorBasedBuilder):
                     rejected = data["rejected"]
 
                     assist_idx = rejected.rfind("\n\nAssistant: ")
-                    r_reject = rejected[assist_idx+13:].strip()
+                    r_reject = rejected[assist_idx + 13 :].strip()
                     assist_idx = chosen.rfind("\n\nAssistant: ")
-                    r_accept = chosen[assist_idx+13:].strip()
+                    r_accept = chosen[assist_idx + 13 :].strip()
 
                     human_idx = chosen.rfind("\n\nHuman: ")
-                    query = chosen[human_idx+9:assist_idx].strip()
+                    query = chosen[human_idx + 9 : assist_idx].strip()
                     prompt = chosen[:human_idx]
                     history = []
 
@@ -83,16 +72,12 @@ class HhRlhfEn(datasets.GeneratorBasedBuilder):
                         assist_idx = prompt.rfind("\n\nAssistant: ")
                         human_idx = prompt.rfind("\n\nHuman: ")
                         if human_idx != -1:
-                            old_query = prompt[human_idx+9:assist_idx].strip()
+                            old_query = prompt[human_idx + 9 : assist_idx].strip()
-                            old_resp = prompt[assist_idx+13:].strip()
+                            old_resp = prompt[assist_idx + 13 :].strip()
                             history.insert(0, (old_query, old_resp))
                         else:
                             break
                         prompt = prompt[:human_idx]
 
-                    yield key, {
+                    yield key, {"instruction": query, "output": [r_accept, r_reject], "history": history}
-                        "instruction": query,
-                        "output": [r_accept, r_reject],
-                        "history": history
-                    }
                     key += 1

data/oaast_rm.json.REMOVED.git-id

@@ -1 +0,0 @@
-274079ea921762be356de85b18f13fa60b7ba8cb

data/oaast_sft.json.REMOVED.git-id

@@ -1 +0,0 @@
-57fd080be5bffe4153fe3ee26a175e3d56da30f3

data/ultra_chat/ultra_chat.py

@@ -1,8 +1,10 @@
-import os
 import json
-import datasets
+import os
 from typing import List
 
+import datasets
+
+
 _HF_ENDPOINT = os.getenv("HF_ENDPOINT", "https://huggingface.co")
 
 _DESCRIPTION = "UltraChat: Large-scale, Informative, and Diverse Multi-round Dialogue Data."
@@ -24,31 +26,19 @@ _BASE_DATA_URL = "{}/datasets/stingning/ultrachat/resolve/main/train_{{idx}}.jso
 
 
 class UltraChat(datasets.GeneratorBasedBuilder):
-
     VERSION = datasets.Version("0.0.0")
 
     def _info(self):
-        features = datasets.Features({
+        features = datasets.Features(
-            "conversations": [{"from": datasets.Value("string"), "value": datasets.Value("string")}]
+            {"conversations": [{"from": datasets.Value("string"), "value": datasets.Value("string")}]}
-        })
+        )
         return datasets.DatasetInfo(
-            description=_DESCRIPTION,
+            description=_DESCRIPTION, features=features, homepage=_HOMEPAGE, license=_LICENSE, citation=_CITATION
-            features=features,
-            homepage=_HOMEPAGE,
-            license=_LICENSE,
-            citation=_CITATION
         )
 
     def _split_generators(self, dl_manager: datasets.DownloadManager):
-        file_paths = [dl_manager.download(_BASE_DATA_URL.format(idx=idx)) for idx in range(10)] # multiple shards
+        file_paths = [dl_manager.download(_BASE_DATA_URL.format(idx=idx)) for idx in range(10)]  # multiple shards
-        return [
+        return [datasets.SplitGenerator(name=datasets.Split.TRAIN, gen_kwargs={"filepaths": file_paths})]
-            datasets.SplitGenerator(
-                name=datasets.Split.TRAIN,
-                gen_kwargs={
-                    "filepaths": file_paths
-                }
-            )
-        ]
 
     def _generate_examples(self, filepaths: List[str]):
         for filepath in filepaths:
@@ -56,7 +46,7 @@ class UltraChat(datasets.GeneratorBasedBuilder):
                 for row in f:
                     try:
                         data = json.loads(row)
-                    except:
+                    except Exception:
                         continue
                     key: int = data["id"]
                     content: List[str] = data["data"]
@@ -64,8 +54,7 @@ class UltraChat(datasets.GeneratorBasedBuilder):
                         content.pop(-1)
                     if len(content) < 2:
                         continue
-                    conversations = [{
+                    conversations = [
-                        "from": "human" if i % 2 == 0 else "gpt",
+                        {"from": "human" if i % 2 == 0 else "gpt", "value": content[i]} for i in range(len(content))
-                        "value": content[i]
+                    ]
-                    } for i in range(len(content))]
                     yield key, {"conversations": conversations}

evaluation/ceval/ceval.py

@@ -133,25 +133,19 @@ class Ceval(datasets.GeneratorBasedBuilder):
             datasets.SplitGenerator(
                 name=datasets.Split.TEST,
                 gen_kwargs={
-                    "filepath": os.path.join(
+                    "filepath": os.path.join(data_dir, "test", f"{task_name}_test.csv"),
-                        data_dir, "test", f"{task_name}_test.csv"
-                    ),
                 },
             ),
             datasets.SplitGenerator(
                 name=datasets.Split.VALIDATION,
                 gen_kwargs={
-                    "filepath": os.path.join(
+                    "filepath": os.path.join(data_dir, "val", f"{task_name}_val.csv"),
-                        data_dir, "val", f"{task_name}_val.csv"
-                    ),
                 },
             ),
             datasets.SplitGenerator(
                 name=datasets.Split.TRAIN,
                 gen_kwargs={
-                    "filepath": os.path.join(
+                    "filepath": os.path.join(data_dir, "dev", f"{task_name}_dev.csv"),
-                        data_dir, "dev", f"{task_name}_dev.csv"
-                    ),
                 },
             ),
         ]

evaluation/cmmlu/cmmlu.py

@@ -37,73 +37,73 @@ _LICENSE = "Creative Commons Attribution-NonCommercial-ShareAlike 4.0 Internatio
 _URL = "cmmlu.zip"
 
 task_list = [
-     'agronomy',
+    "agronomy",
-     'anatomy',
+    "anatomy",
-     'ancient_chinese',
+    "ancient_chinese",
-     'arts',
+    "arts",
-     'astronomy',
+    "astronomy",
-     'business_ethics',
+    "business_ethics",
-     'chinese_civil_service_exam',
+    "chinese_civil_service_exam",
-     'chinese_driving_rule',
+    "chinese_driving_rule",
-     'chinese_food_culture',
+    "chinese_food_culture",
-     'chinese_foreign_policy',
+    "chinese_foreign_policy",
-     'chinese_history',
+    "chinese_history",
-     'chinese_literature',
+    "chinese_literature",
-     'chinese_teacher_qualification',
+    "chinese_teacher_qualification",
-     'clinical_knowledge',
+    "clinical_knowledge",
-     'college_actuarial_science',
+    "college_actuarial_science",
-     'college_education',
+    "college_education",
-     'college_engineering_hydrology',
+    "college_engineering_hydrology",
-     'college_law',
+    "college_law",
-     'college_mathematics',
+    "college_mathematics",
-     'college_medical_statistics',
+    "college_medical_statistics",
-     'college_medicine',
+    "college_medicine",
-     'computer_science',
+    "computer_science",
-     'computer_security',
+    "computer_security",
-     'conceptual_physics',
+    "conceptual_physics",
-     'construction_project_management',
+    "construction_project_management",
-     'economics',
+    "economics",
-     'education',
+    "education",
-     'electrical_engineering',
+    "electrical_engineering",
-     'elementary_chinese',
+    "elementary_chinese",
-     'elementary_commonsense',
+    "elementary_commonsense",
-     'elementary_information_and_technology',
+    "elementary_information_and_technology",
-     'elementary_mathematics',
+    "elementary_mathematics",
-     'ethnology',
+    "ethnology",
-     'food_science',
+    "food_science",
-     'genetics',
+    "genetics",
-     'global_facts',
+    "global_facts",
-     'high_school_biology',
+    "high_school_biology",
-     'high_school_chemistry',
+    "high_school_chemistry",
-     'high_school_geography',
+    "high_school_geography",
-     'high_school_mathematics',
+    "high_school_mathematics",
-     'high_school_physics',
+    "high_school_physics",
-     'high_school_politics',
+    "high_school_politics",
-     'human_sexuality',
+    "human_sexuality",
-     'international_law',
+    "international_law",
-     'journalism',
+    "journalism",
-     'jurisprudence',
+    "jurisprudence",
-     'legal_and_moral_basis',
+    "legal_and_moral_basis",
-     'logical',
+    "logical",
-     'machine_learning',
+    "machine_learning",
-     'management',
+    "management",
-     'marketing',
+    "marketing",
-     'marxist_theory',
+    "marxist_theory",
-     'modern_chinese',
+    "modern_chinese",
-     'nutrition',
+    "nutrition",
-     'philosophy',
+    "philosophy",
-     'professional_accounting',
+    "professional_accounting",
-     'professional_law',
+    "professional_law",
-     'professional_medicine',
+    "professional_medicine",
-     'professional_psychology',
+    "professional_psychology",
-     'public_relations',
+    "public_relations",
-     'security_study',
+    "security_study",
-     'sociology',
+    "sociology",
-     'sports_science',
+    "sports_science",
-     'traditional_chinese_medicine',
+    "traditional_chinese_medicine",
-     'virology',
+    "virology",
-     'world_history',
+    "world_history",
-     'world_religions',
+    "world_religions",
 ]
 
 

evaluation/mmlu/mmlu.py

@@ -136,25 +136,19 @@ class MMLU(datasets.GeneratorBasedBuilder):
             datasets.SplitGenerator(
                 name=datasets.Split.TEST,
                 gen_kwargs={
-                    "filepath": os.path.join(
+                    "filepath": os.path.join(data_dir, "data", "test", f"{task_name}_test.csv"),
-                        data_dir, "data", "test", f"{task_name}_test.csv"
-                    ),
                 },
             ),
             datasets.SplitGenerator(
                 name=datasets.Split.VALIDATION,
                 gen_kwargs={
-                    "filepath": os.path.join(
+                    "filepath": os.path.join(data_dir, "data", "val", f"{task_name}_val.csv"),
-                        data_dir, "data", "val", f"{task_name}_val.csv"
-                    ),
                 },
             ),
             datasets.SplitGenerator(
                 name=datasets.Split.TRAIN,
                 gen_kwargs={
-                    "filepath": os.path.join(
+                    "filepath": os.path.join(data_dir, "data", "dev", f"{task_name}_dev.csv"),
-                        data_dir, "data", "dev", f"{task_name}_dev.csv"
-                    ),
                 },
             ),
         ]

examples/README.md

@@ -0,0 +1,229 @@
+We provide diverse examples about fine-tuning LLMs.
+
+Make sure to execute these commands in the `LLaMA-Factory` directory.
+
+## Table of Contents
+
+- [LoRA Fine-Tuning on A Single GPU](#lora-fine-tuning-on-a-single-gpu)
+- [QLoRA Fine-Tuning on a Single GPU](#qlora-fine-tuning-on-a-single-gpu)
+- [LoRA Fine-Tuning on Multiple GPUs](#lora-fine-tuning-on-multiple-gpus)
+- [LoRA Fine-Tuning on Multiple NPUs](#lora-fine-tuning-on-multiple-npus)
+- [Full-Parameter Fine-Tuning on Multiple GPUs](#full-parameter-fine-tuning-on-multiple-gpus)
+- [Merging LoRA Adapters and Quantization](#merging-lora-adapters-and-quantization)
+- [Inferring LoRA Fine-Tuned Models](#inferring-lora-fine-tuned-models)
+- [Extras](#extras)
+
+## Examples
+
+### LoRA Fine-Tuning on A Single GPU
+
+#### (Continuous) Pre-Training
+
+```bash
+CUDA_VISIBLE_DEVICES=0 llamafactory-cli train examples/lora_single_gpu/llama3_lora_pretrain.yaml
+```
+
+#### Supervised Fine-Tuning
+
+```bash
+CUDA_VISIBLE_DEVICES=0 llamafactory-cli train examples/lora_single_gpu/llama3_lora_sft.yaml
+```
+
+#### Multimodal Supervised Fine-Tuning
+
+```bash
+CUDA_VISIBLE_DEVICES=0 llamafactory-cli train examples/lora_single_gpu/llava1_5_lora_sft.yaml
+```
+
+#### Reward Modeling
+
+```bash
+CUDA_VISIBLE_DEVICES=0 llamafactory-cli train examples/lora_single_gpu/llama3_lora_reward.yaml
+```
+
+#### PPO Training
+
+```bash
+CUDA_VISIBLE_DEVICES=0 llamafactory-cli train examples/lora_single_gpu/llama3_lora_ppo.yaml
+```
+
+#### DPO Training
+
+```bash
+CUDA_VISIBLE_DEVICES=0 llamafactory-cli train examples/lora_single_gpu/llama3_lora_dpo.yaml
+```
+
+#### ORPO Training
+
+```bash
+CUDA_VISIBLE_DEVICES=0 llamafactory-cli train examples/lora_single_gpu/llama3_lora_orpo.yaml
+```
+
+#### Preprocess Dataset
+
+It is useful for large dataset, use `tokenized_path` in config to load the preprocessed dataset.
+
+```bash
+CUDA_VISIBLE_DEVICES=0 llamafactory-cli train examples/lora_single_gpu/llama3_preprocess.yaml
+```
+
+#### Evaluating on MMLU/CMMLU/C-Eval Benchmarks
+
+```bash
+CUDA_VISIBLE_DEVICES=0 llamafactory-cli eval examples/lora_single_gpu/llama3_lora_eval.yaml
+```
+
+#### Batch Predicting and Computing BLEU and ROUGE Scores
+
+```bash
+CUDA_VISIBLE_DEVICES=0 llamafactory-cli train examples/lora_single_gpu/llama3_lora_predict.yaml
+```
+
+### QLoRA Fine-Tuning on a Single GPU
+
+#### Supervised Fine-Tuning with 4/8-bit Bitsandbytes Quantization (Recommended)
+
+```bash
+CUDA_VISIBLE_DEVICES=0 llamafactory-cli train examples/qlora_single_gpu/llama3_lora_sft_bitsandbytes.yaml
+```
+
+#### Supervised Fine-Tuning with 4/8-bit GPTQ Quantization
+
+```bash
+CUDA_VISIBLE_DEVICES=0 llamafactory-cli train examples/qlora_single_gpu/llama3_lora_sft_gptq.yaml
+```
+
+#### Supervised Fine-Tuning with 4-bit AWQ Quantization
+
+```bash
+CUDA_VISIBLE_DEVICES=0 llamafactory-cli train examples/qlora_single_gpu/llama3_lora_sft_awq.yaml
+```
+
+#### Supervised Fine-Tuning with 2-bit AQLM Quantization
+
+```bash
+CUDA_VISIBLE_DEVICES=0 llamafactory-cli train examples/qlora_single_gpu/llama3_lora_sft_aqlm.yaml
+```
+
+### LoRA Fine-Tuning on Multiple GPUs
+
+#### Supervised Fine-Tuning with Accelerate on Single Node
+
+```bash
+bash examples/lora_multi_gpu/single_node.sh
+```
+
+#### Supervised Fine-Tuning with Accelerate on Multiple Nodes
+
+```bash
+bash examples/lora_multi_gpu/multi_node.sh
+```
+
+#### Supervised Fine-Tuning with DeepSpeed ZeRO-3 (Weight Sharding)
+
+```bash
+bash examples/lora_multi_gpu/ds_zero3.sh
+```
+
+### LoRA Fine-Tuning on Multiple NPUs
+
+#### Supervised Fine-Tuning with DeepSpeed ZeRO-0
+
+```bash
+bash examples/lora_multi_npu/ds_zero0.sh
+```
+
+### Full-Parameter Fine-Tuning on Multiple GPUs
+
+#### Supervised Fine-Tuning with Accelerate on Single Node
+
+```bash
+bash examples/full_multi_gpu/single_node.sh
+```
+
+#### Supervised Fine-Tuning with Accelerate on Multiple Nodes
+
+```bash
+bash examples/full_multi_gpu/multi_node.sh
+```
+
+#### Batch Predicting and Computing BLEU and ROUGE Scores
+
+```bash
+bash examples/full_multi_gpu/predict.sh
+```
+
+### Merging LoRA Adapters and Quantization
+
+#### Merge LoRA Adapters
+
+Note: DO NOT use quantized model or `quantization_bit` when merging LoRA adapters.
+
+```bash
+CUDA_VISIBLE_DEVICES=0 llamafactory-cli export examples/merge_lora/llama3_lora_sft.yaml
+```
+
+#### Quantizing Model using AutoGPTQ
+
+```bash
+CUDA_VISIBLE_DEVICES=0 llamafactory-cli export examples/merge_lora/llama3_gptq.yaml
+```
+
+### Inferring LoRA Fine-Tuned Models
+
+#### Use CLI
+
+```bash
+CUDA_VISIBLE_DEVICES=0 llamafactory-cli chat examples/merge_lora/llama3_lora_sft.yaml
+```
+
+#### Use Web UI
+
+```bash
+CUDA_VISIBLE_DEVICES=0 llamafactory-cli webchat examples/merge_lora/llama3_lora_sft.yaml
+```
+
+#### Launch OpenAI-style API
+
+```bash
+CUDA_VISIBLE_DEVICES=0 llamafactory-cli api examples/merge_lora/llama3_lora_sft.yaml
+```
+
+### Extras
+
+#### Full-Parameter Fine-Tuning using GaLore
+
+```bash
+CUDA_VISIBLE_DEVICES=0 llamafactory-cli train examples/extras/galore/llama3_full_sft.yaml
+```
+
+#### Full-Parameter Fine-Tuning using BAdam
+
+```bash
+CUDA_VISIBLE_DEVICES=0 llamafactory-cli train examples/extras/badam/llama3_full_sft.yaml
+```
+
+#### LoRA+ Fine-Tuning
+
+```bash
+CUDA_VISIBLE_DEVICES=0 llamafactory-cli train examples/extras/loraplus/llama3_lora_sft.yaml
+```
+
+#### Mixture-of-Depths Fine-Tuning
+
+```bash
+CUDA_VISIBLE_DEVICES=0 llamafactory-cli train examples/extras/mod/llama3_full_sft.yaml
+```
+
+#### LLaMA-Pro Fine-Tuning
+
+```bash
+bash examples/extras/llama_pro/expand.sh
+CUDA_VISIBLE_DEVICES=0 llamafactory-cli train examples/extras/llama_pro/llama3_freeze_sft.yaml
+```
+
+#### FSDP+QLoRA Fine-Tuning
+
+```bash
+bash examples/extras/fsdp_qlora/single_node.sh
+```

examples/README_zh.md

@@ -0,0 +1,229 @@
+我们提供了多样化的大模型微调示例脚本。
+
+请确保在 `LLaMA-Factory` 目录下执行下述命令。
+
+## 目录
+
+- [单 GPU LoRA 微调](#单-gpu-lora-微调)
+- [单 GPU QLoRA 微调](#单-gpu-qlora-微调)
+- [多 GPU LoRA 微调](#多-gpu-lora-微调)
+- [多 NPU LoRA 微调](#多-npu-lora-微调)
+- [多 GPU 全参数微调](#多-gpu-全参数微调)
+- [合并 LoRA 适配器与模型量化](#合并-lora-适配器与模型量化)
+- [推理 LoRA 模型](#推理-lora-模型)
+- [杂项](#杂项)
+
+## 示例
+
+### 单 GPU LoRA 微调
+
+#### （增量）预训练
+
+```bash
+CUDA_VISIBLE_DEVICES=0 llamafactory-cli train examples/lora_single_gpu/llama3_lora_pretrain.yaml
+```
+
+#### 指令监督微调
+
+```bash
+CUDA_VISIBLE_DEVICES=0 llamafactory-cli train examples/lora_single_gpu/llama3_lora_sft.yaml
+```
+
+#### 多模态指令监督微调
+
+```bash
+CUDA_VISIBLE_DEVICES=0 llamafactory-cli train examples/lora_single_gpu/llava1_5_lora_sft.yaml
+```
+
+#### 奖励模型训练
+
+```bash
+CUDA_VISIBLE_DEVICES=0 llamafactory-cli train examples/lora_single_gpu/llama3_lora_reward.yaml
+```
+
+#### PPO 训练
+
+```bash
+CUDA_VISIBLE_DEVICES=0 llamafactory-cli train examples/lora_single_gpu/llama3_lora_ppo.yaml
+```
+
+#### DPO 训练
+
+```bash
+CUDA_VISIBLE_DEVICES=0 llamafactory-cli train examples/lora_single_gpu/llama3_lora_dpo.yaml
+```
+
+#### ORPO 训练
+
+```bash
+CUDA_VISIBLE_DEVICES=0 llamafactory-cli train examples/lora_single_gpu/llama3_lora_orpo.yaml
+```
+
+#### 预处理数据集
+
+对于大数据集有帮助，在配置中使用 `tokenized_path` 以加载预处理后的数据集。
+
+```bash
+CUDA_VISIBLE_DEVICES=0 llamafactory-cli train examples/lora_single_gpu/llama3_preprocess.yaml
+```
+
+#### 在 MMLU/CMMLU/C-Eval 上评估
+
+```bash
+CUDA_VISIBLE_DEVICES=0 llamafactory-cli eval examples/lora_single_gpu/llama3_lora_eval.yaml
+```
+
+#### 批量预测并计算 BLEU 和 ROUGE 分数
+
+```bash
+CUDA_VISIBLE_DEVICES=0 llamafactory-cli train examples/lora_single_gpu/llama3_lora_predict.yaml
+```
+
+### 单 GPU QLoRA 微调
+
+#### 基于 4/8 比特 Bitsandbytes 量化进行指令监督微调（推荐）
+
+```bash
+CUDA_VISIBLE_DEVICES=0 llamafactory-cli train examples/qlora_single_gpu/llama3_lora_sft_bitsandbytes.yaml
+```
+
+#### 基于 4/8 比特 GPTQ 量化进行指令监督微调
+
+```bash
+CUDA_VISIBLE_DEVICES=0 llamafactory-cli train examples/qlora_single_gpu/llama3_lora_sft_gptq.yaml
+```
+
+#### 基于 4 比特 AWQ 量化进行指令监督微调
+
+```bash
+CUDA_VISIBLE_DEVICES=0 llamafactory-cli train examples/qlora_single_gpu/llama3_lora_sft_awq.yaml
+```
+
+#### 基于 2 比特 AQLM 量化进行指令监督微调
+
+```bash
+CUDA_VISIBLE_DEVICES=0 llamafactory-cli train examples/qlora_single_gpu/llama3_lora_sft_aqlm.yaml
+```
+
+### 多 GPU LoRA 微调
+
+#### 使用 Accelerate 进行单节点训练
+
+```bash
+bash examples/lora_multi_gpu/single_node.sh
+```
+
+#### 使用 Accelerate 进行多节点训练
+
+```bash
+bash examples/lora_multi_gpu/multi_node.sh
+```
+
+#### 使用 DeepSpeed ZeRO-3 平均分配显存
+
+```bash
+bash examples/lora_multi_gpu/ds_zero3.sh
+```
+
+### 多 NPU LoRA 微调
+
+#### 使用 DeepSpeed ZeRO-0 训练
+
+```bash
+bash examples/lora_multi_npu/ds_zero0.sh
+```
+
+### 多 GPU 全参数微调
+
+#### 使用 DeepSpeed 进行单节点训练
+
+```bash
+bash examples/full_multi_gpu/single_node.sh
+```
+
+#### 使用 DeepSpeed 进行多节点训练
+
+```bash
+bash examples/full_multi_gpu/multi_node.sh
+```
+
+#### 批量预测并计算 BLEU 和 ROUGE 分数
+
+```bash
+bash examples/full_multi_gpu/predict.sh
+```
+
+### 合并 LoRA 适配器与模型量化
+
+#### 合并 LoRA 适配器
+
+注：请勿使用量化后的模型或 `quantization_bit` 参数来合并 LoRA 适配器。
+
+```bash
+CUDA_VISIBLE_DEVICES=0 llamafactory-cli export examples/merge_lora/llama3_lora_sft.yaml
+```
+
+#### 使用 AutoGPTQ 量化模型
+
+```bash
+CUDA_VISIBLE_DEVICES=0 llamafactory-cli export examples/merge_lora/llama3_gptq.yaml
+```
+
+### 推理 LoRA 模型
+
+#### 使用命令行接口
+
+```bash
+CUDA_VISIBLE_DEVICES=0 llamafactory-cli chat examples/merge_lora/llama3_lora_sft.yaml
+```
+
+#### 使用浏览器界面
+
+```bash
+CUDA_VISIBLE_DEVICES=0 llamafactory-cli webchat examples/merge_lora/llama3_lora_sft.yaml
+```
+
+#### 启动 OpenAI 风格 API
+
+```bash
+CUDA_VISIBLE_DEVICES=0 llamafactory-cli api examples/merge_lora/llama3_lora_sft.yaml
+```
+
+### 杂项
+
+#### 使用 GaLore 进行全参数训练
+
+```bash
+CUDA_VISIBLE_DEVICES=0 llamafactory-cli train examples/extras/galore/llama3_full_sft.yaml
+```
+
+#### 使用 BAdam 进行全参数训练
+
+```bash
+CUDA_VISIBLE_DEVICES=0 llamafactory-cli train examples/extras/badam/llama3_full_sft.yaml
+```
+
+#### LoRA+ 微调
+
+```bash
+CUDA_VISIBLE_DEVICES=0 llamafactory-cli train examples/extras/loraplus/llama3_lora_sft.yaml
+```
+
+#### 深度混合微调
+
+```bash
+CUDA_VISIBLE_DEVICES=0 llamafactory-cli train examples/extras/mod/llama3_full_sft.yaml
+```
+
+#### LLaMA-Pro 微调
+
+```bash
+bash examples/extras/llama_pro/expand.sh
+CUDA_VISIBLE_DEVICES=0 llamafactory-cli train examples/extras/llama_pro/llama3_freeze_sft.yaml
+```
+
+#### FSDP+QLoRA 微调
+
+```bash
+bash examples/extras/fsdp_qlora/single_node.sh
+```

examples/accelerate/fsdp_config.yaml

@@ -15,8 +15,8 @@ fsdp_config:
 machine_rank: 0
 main_training_function: main
 mixed_precision: fp16
-num_machines: 1
+num_machines: 1 # the number of nodes
-num_processes: 2
+num_processes: 2 # the number of GPUs in all nodes
 rdzv_backend: static
 same_network: true
 tpu_env: []

examples/accelerate/master_config.yaml

@@ -8,8 +8,8 @@ main_process_ip: 192.168.0.1
 main_process_port: 29555
 main_training_function: main
 mixed_precision: fp16
-num_machines: 2
+num_machines: 2 # the number of nodes
-num_processes: 16
+num_processes: 8 # the number of GPUs in all nodes
 rdzv_backend: static
 same_network: true
 tpu_env: []

examples/accelerate/single_config.yaml

@@ -6,8 +6,8 @@ gpu_ids: all
 machine_rank: 0
 main_training_function: main
 mixed_precision: fp16
-num_machines: 1
+num_machines: 1 # the number of nodes
-num_processes: 4
+num_processes: 4 # the number of GPUs in all nodes
 rdzv_backend: static
 same_network: true
 tpu_env: []

examples/accelerate/slave_config.yaml

@@ -8,8 +8,8 @@ main_process_ip: 192.168.0.1
 main_process_port: 29555
 main_training_function: main
 mixed_precision: fp16
-num_machines: 2
+num_machines: 2 # the number of nodes
-num_processes: 16
+num_processes: 8 # the number of GPUs in all nodes
 rdzv_backend: static
 same_network: true
 tpu_env: []

examples/extras/badam/llama3_lora_sft.yaml

@@ -0,0 +1,41 @@
+# model
+model_name_or_path: meta-llama/Meta-Llama-3-8B-Instruct
+
+# method
+stage: sft
+do_train: true
+finetuning_type: full
+use_badam: true
+badam_switch_mode: descending
+badam_switch_interval: 50
+badam_verbose: 2
+
+# dataset
+dataset: identity,alpaca_gpt4_en
+template: llama3
+cutoff_len: 1024
+max_samples: 1000
+overwrite_cache: true
+preprocessing_num_workers: 16
+
+# output
+output_dir: saves/llama3-8b/full/sft
+logging_steps: 10
+save_steps: 500
+plot_loss: true
+overwrite_output_dir: true
+
+# train
+per_device_train_batch_size: 1
+gradient_accumulation_steps: 8
+learning_rate: 0.0001
+num_train_epochs: 3.0
+lr_scheduler_type: cosine
+warmup_steps: 0.1
+pure_bf16: true
+
+# eval
+val_size: 0.1
+per_device_eval_batch_size: 1
+evaluation_strategy: steps
+eval_steps: 500

examples/extras/fsdp_qlora/llama3_lora_sft.yaml

@@ -0,0 +1,42 @@
+# model
+model_name_or_path: meta-llama/Meta-Llama-3-8B-Instruct
+quantization_bit: 4
+
+# method
+stage: sft
+do_train: true
+finetuning_type: lora
+lora_target: q_proj,v_proj
+
+# ddp
+ddp_timeout: 180000000
+
+# dataset
+dataset: identity,alpaca_gpt4_en
+template: llama3
+cutoff_len: 1024
+max_samples: 1000
+overwrite_cache: true
+preprocessing_num_workers: 16
+
+# output
+output_dir: saves/llama3-8b/lora/sft
+logging_steps: 10
+save_steps: 500
+plot_loss: true
+overwrite_output_dir: true
+
+# train
+per_device_train_batch_size: 1
+gradient_accumulation_steps: 8
+learning_rate: 0.0001
+num_train_epochs: 3.0
+lr_scheduler_type: cosine
+warmup_steps: 0.1
+fp16: true
+
+# eval
+val_size: 0.1
+per_device_eval_batch_size: 1
+evaluation_strategy: steps
+eval_steps: 500

examples/extras/fsdp_qlora/single_node.sh

@@ -0,0 +1,10 @@
+#!/bin/bash
+# DO NOT use GPTQ/AWQ model in FSDP+QLoRA
+
+pip install "transformers>=4.39.1"
+pip install "accelerate>=0.28.0"
+pip install "bitsandbytes>=0.43.0"
+
+CUDA_VISIBLE_DEVICES=0,1 accelerate launch \
+    --config_file examples/accelerate/fsdp_config.yaml \
+    src/train.py examples/extras/fsdp_qlora/llama3_lora_sft.yaml

examples/extras/galore/llama3_full_sft.yaml

@@ -0,0 +1,42 @@
+# model
+model_name_or_path: meta-llama/Meta-Llama-3-8B-Instruct
+
+# method
+stage: sft
+do_train: true
+finetuning_type: full
+use_galore: true
+galore_layerwise: true
+galore_target: mlp,self_attn
+galore_rank: 128
+galore_scale: 2.0
+
+# dataset
+dataset: identity,alpaca_gpt4_en
+template: llama3
+cutoff_len: 1024
+max_samples: 1000
+overwrite_cache: true
+preprocessing_num_workers: 16
+
+# output
+output_dir: saves/llama3-8b/full/sft
+logging_steps: 10
+save_steps: 500
+plot_loss: true
+overwrite_output_dir: true
+
+# train
+per_device_train_batch_size: 1
+gradient_accumulation_steps: 1
+learning_rate: 0.0001
+num_train_epochs: 3.0
+lr_scheduler_type: cosine
+warmup_steps: 0.1
+pure_bf16: true
+
+# eval
+val_size: 0.1
+per_device_eval_batch_size: 1
+evaluation_strategy: steps
+eval_steps: 500

examples/extras/galore/sft.sh

@@ -1,35 +0,0 @@
-#!/bin/bash
-
-CUDA_VISIBLE_DEVICES=0 python ../../../src/train_bash.py \
-    --stage sft \
-    --do_train \
-    --model_name_or_path meta-llama/Llama-2-7b-hf \
-    --dataset alpaca_gpt4_en,glaive_toolcall \
-    --dataset_dir ../../../data \
-    --template default \
-    --finetuning_type full \
-    --use_galore \
-    --galore_layerwise \
-    --galore_target mlp,self_attn \
-    --galore_rank 128 \
-    --output_dir ../../../saves/LLaMA2-7B/galore/sft \
-    --overwrite_cache \
-    --overwrite_output_dir \
-    --cutoff_len 1024 \
-    --preprocessing_num_workers 16 \
-    --per_device_train_batch_size 1 \
-    --per_device_eval_batch_size 1 \
-    --gradient_accumulation_steps 1 \
-    --lr_scheduler_type cosine \
-    --logging_steps 10 \
-    --warmup_steps 20 \
-    --save_steps 100 \
-    --eval_steps 100 \
-    --evaluation_strategy steps \
-    --load_best_model_at_end \
-    --learning_rate 5e-5 \
-    --num_train_epochs 3.0 \
-    --max_samples 3000 \
-    --val_size 0.1 \
-    --plot_loss \
-    --pure_bf16

examples/extras/llama_pro/expand.sh

@@ -1,6 +1,6 @@
 #!/bin/bash
 
-python ../../../scripts/llama_pro.py \
+python scripts/llama_pro.py \
-    --model_name_or_path meta-llama/Llama-2-7b-hf \
+    --model_name_or_path meta-llama/Meta-Llama-3-8B-Instruct \
-    --output_dir ../../../models/llama2-7b-pro \
+    --output_dir models/llama3-8b-instruct-pro \
     --num_expand 8

examples/extras/llama_pro/llama3_freeze_sft.yaml

@@ -0,0 +1,40 @@
+# model
+model_name_or_path: models/llama3-8b-instruct-pro
+
+# method
+stage: sft
+do_train: true
+finetuning_type: freeze
+freeze_trainable_layers: 8
+freeze_trainable_modules: all
+use_llama_pro: true
+
+# dataset
+dataset: identity,alpaca_gpt4_en
+template: llama3
+cutoff_len: 1024
+max_samples: 1000
+overwrite_cache: true
+preprocessing_num_workers: 16
+
+# output
+output_dir: saves/llama3-8b-instruct-pro/freeze/sft
+logging_steps: 10
+save_steps: 500
+plot_loss: true
+overwrite_output_dir: true
+
+# train
+per_device_train_batch_size: 1
+gradient_accumulation_steps: 8
+learning_rate: 0.0001
+num_train_epochs: 3.0
+lr_scheduler_type: cosine
+warmup_steps: 0.1
+fp16: true
+
+# eval
+val_size: 0.1
+per_device_eval_batch_size: 1
+evaluation_strategy: steps
+eval_steps: 500

examples/extras/llama_pro/sft.sh

@@ -1,34 +0,0 @@
-#!/bin/bash
-
-CUDA_VISIBLE_DEVICES=0 python ../../../src/train_bash.py \
-    --stage sft \
-    --do_train \
-    --model_name_or_path ../../../models/llama2-7b-pro \
-    --dataset alpaca_gpt4_en,glaive_toolcall \
-    --dataset_dir ../../../data \
-    --template default \
-    --finetuning_type freeze \
-    --name_module_trainable all \
-    --num_layer_trainable 8 \
-    --use_llama_pro \
-    --output_dir ../../../saves/LLaMA2-7B-Pro/lora/sft \
-    --overwrite_cache \
-    --overwrite_output_dir \
-    --cutoff_len 1024 \
-    --preprocessing_num_workers 16 \
-    --per_device_train_batch_size 1 \
-    --per_device_eval_batch_size 1 \
-    --gradient_accumulation_steps 8 \
-    --lr_scheduler_type cosine \
-    --logging_steps 10 \
-    --warmup_steps 20 \
-    --save_steps 100 \
-    --eval_steps 100 \
-    --evaluation_strategy steps \
-    --load_best_model_at_end \
-    --learning_rate 5e-5 \
-    --num_train_epochs 3.0 \
-    --max_samples 3000 \
-    --val_size 0.1 \
-    --plot_loss \
-    --fp16

examples/extras/loraplus/llama3_lora_sft.yaml

@@ -0,0 +1,39 @@
+# model
+model_name_or_path: meta-llama/Meta-Llama-3-8B-Instruct
+
+# method
+stage: sft
+do_train: true
+finetuning_type: lora
+lora_target: q_proj,v_proj
+loraplus_lr_ratio: 16.0
+
+# dataset
+dataset: identity,alpaca_gpt4_en
+template: llama3
+cutoff_len: 1024
+max_samples: 1000
+overwrite_cache: true
+preprocessing_num_workers: 16
+
+# output
+output_dir: saves/llama3-8b/lora/sft
+logging_steps: 10
+save_steps: 500
+plot_loss: true
+overwrite_output_dir: true
+
+# train
+per_device_train_batch_size: 1
+gradient_accumulation_steps: 8
+learning_rate: 0.0001
+num_train_epochs: 3.0
+lr_scheduler_type: cosine
+warmup_steps: 0.1
+fp16: true
+
+# eval
+val_size: 0.1
+per_device_eval_batch_size: 1
+evaluation_strategy: steps
+eval_steps: 500

examples/extras/loraplus/sft.sh

@@ -1,33 +0,0 @@
-#!/bin/bash
-
-CUDA_VISIBLE_DEVICES=0 python ../../src/train_bash.py \
-    --stage sft \
-    --do_train \
-    --model_name_or_path meta-llama/Llama-2-7b-hf \
-    --dataset alpaca_gpt4_en,glaive_toolcall \
-    --dataset_dir ../../data \
-    --template default \
-    --finetuning_type lora \
-    --lora_target q_proj,v_proj \
-    --output_dir ../../saves/LLaMA2-7B/loraplus/sft \
-    --overwrite_cache \
-    --overwrite_output_dir \
-    --cutoff_len 1024 \
-    --preprocessing_num_workers 16 \
-    --per_device_train_batch_size 1 \
-    --per_device_eval_batch_size 1 \
-    --gradient_accumulation_steps 8 \
-    --lr_scheduler_type cosine \
-    --logging_steps 10 \
-    --warmup_steps 20 \
-    --save_steps 100 \
-    --eval_steps 100 \
-    --evaluation_strategy steps \
-    --load_best_model_at_end \
-    --learning_rate 5e-5 \
-    --num_train_epochs 3.0 \
-    --max_samples 3000 \
-    --val_size 0.1 \
-    --plot_loss \
-    --fp16 \
-    --loraplus_lr_ratio 16.0

examples/extras/mod/llama3_full_sft.yaml

@@ -0,0 +1,39 @@
+# model
+model_name_or_path: meta-llama/Meta-Llama-3-8B-Instruct
+
+# method
+stage: sft
+do_train: true
+finetuning_type: full
+mixture_of_depths: convert
+
+# dataset
+dataset: identity,alpaca_gpt4_en
+template: llama3
+cutoff_len: 1024
+max_samples: 1000
+overwrite_cache: true
+preprocessing_num_workers: 16
+
+# output
+output_dir: saves/llama3-8b-mod/full/sft
+logging_steps: 10
+save_steps: 500
+plot_loss: true
+overwrite_output_dir: true
+
+# train
+per_device_train_batch_size: 1
+gradient_accumulation_steps: 8
+optim: paged_adamw_8bit
+learning_rate: 0.0001
+num_train_epochs: 3.0
+lr_scheduler_type: cosine
+warmup_steps: 0.1
+pure_bf16: true
+
+# eval
+val_size: 0.1
+per_device_eval_batch_size: 1
+evaluation_strategy: steps
+eval_steps: 500

examples/fsdp_qlora/README.md

@@ -1,5 +0,0 @@
-```bash
-pip install "transformers>=4.39.1"
-pip install "accelerate>=0.28.0"
-pip install "bitsandbytes>=0.43.0"
-```

examples/fsdp_qlora/fsdp.sh

@@ -1,33 +0,0 @@
-#!/bin/bash
-
-CUDA_VISIBLE_DEVICES=0,1 accelerate launch \
-    --config_file ../accelerate/fsdp_config.yaml \
-    ../../src/train_bash.py \
-    --stage sft \
-    --do_train \
-    --model_name_or_path meta-llama/Llama-2-70b-hf \
-    --dataset alpaca_gpt4_en,glaive_toolcall \
-    --dataset_dir ../../data \
-    --template default \
-    --finetuning_type lora \
-    --lora_target q_proj,v_proj \
-    --output_dir ../../saves/LLaMA2-70B/lora/sft \
-    --overwrite_cache \
-    --overwrite_output_dir \
-    --cutoff_len 1024 \
-    --per_device_train_batch_size 1 \
-    --per_device_eval_batch_size 1 \
-    --gradient_accumulation_steps 8 \
-    --lr_scheduler_type cosine \
-    --logging_steps 10 \
-    --save_steps 100 \
-    --eval_steps 100 \
-    --evaluation_strategy steps \
-    --load_best_model_at_end \
-    --learning_rate 5e-5 \
-    --num_train_epochs 3.0 \
-    --max_samples 3000 \
-    --val_size 0.1 \
-    --quantization_bit 4 \
-    --plot_loss \
-    --fp16

examples/full_multi_gpu/llama3_full_predict.yaml

@@ -0,0 +1,23 @@
+# model
+model_name_or_path: saves/llama3-8b/full/sft
+
+# method
+stage: sft
+do_predict: true
+finetuning_type: full
+
+# dataset
+dataset: identity,alpaca_gpt4_en
+template: llama3
+cutoff_len: 1024
+max_samples: 50
+overwrite_cache: true
+preprocessing_num_workers: 16
+
+# output
+output_dir: saves/llama3-8b/full/predict
+overwrite_output_dir: true
+
+# eval
+per_device_eval_batch_size: 1
+predict_with_generate: true

examples/full_multi_gpu/llama3_full_sft.yaml

@@ -0,0 +1,41 @@
+# model
+model_name_or_path: meta-llama/Meta-Llama-3-8B-Instruct
+
+# method
+stage: sft
+do_train: true
+finetuning_type: full
+
+# ddp
+ddp_timeout: 180000000
+deepspeed: examples/deepspeed/ds_z3_config.json
+
+# dataset
+dataset: identity,alpaca_gpt4_en
+template: llama3
+cutoff_len: 1024
+max_samples: 1000
+overwrite_cache: true
+preprocessing_num_workers: 16
+
+# output
+output_dir: saves/llama3-8b/full/sft
+logging_steps: 10
+save_steps: 500
+plot_loss: true
+overwrite_output_dir: true
+
+# train
+per_device_train_batch_size: 1
+gradient_accumulation_steps: 2
+learning_rate: 0.0001
+num_train_epochs: 3.0
+lr_scheduler_type: cosine
+warmup_steps: 0.1
+fp16: true
+
+# eval
+val_size: 0.1
+per_device_eval_batch_size: 1
+evaluation_strategy: steps
+eval_steps: 500

examples/full_multi_gpu/multi_node.sh

@@ -1,38 +1,15 @@
 #!/bin/bash
 
-python -m torch.distributed.run \
+NPROC_PER_NODE=4
+NNODES=2
+RANK=0
+MASTER_ADDR=192.168.0.1
+MASTER_PORT=29500
+
+CUDA_VISIBLE_DEVICES=0,1,2,3 torchrun \
     --nproc_per_node $NPROC_PER_NODE \
     --nnodes $NNODES \
     --node_rank $RANK \
     --master_addr $MASTER_ADDR \
     --master_port $MASTER_PORT \
-    ../../src/train_bash.py \
+    src/train.py examples/full_multi_gpu/llama3_full_sft.yaml
-    --deepspeed ../deepspeed/ds_z3_config.json \
-    --stage sft \
-    --do_train \
-    --model_name_or_path meta-llama/Llama-2-7b-hf \
-    --dataset alpaca_gpt4_en,glaive_toolcall \
-    --dataset_dir ../../data \
-    --template default \
-    --finetuning_type full \
-    --output_dir ../../saves/LLaMA2-7B/full/sft \
-    --overwrite_cache \
-    --overwrite_output_dir \
-    --cutoff_len 1024 \
-    --preprocessing_num_workers 16 \
-    --per_device_train_batch_size 1 \
-    --per_device_eval_batch_size 1 \
-    --gradient_accumulation_steps 2 \
-    --lr_scheduler_type cosine \
-    --logging_steps 10 \
-    --warmup_steps 20 \
-    --save_steps 100 \
-    --eval_steps 100 \
-    --evaluation_strategy steps \
-    --learning_rate 5e-5 \
-    --num_train_epochs 3.0 \
-    --max_samples 3000 \
-    --val_size 0.1 \
-    --ddp_timeout 1800000 \
-    --plot_loss \
-    --fp16

examples/full_multi_gpu/predict.sh

@@ -0,0 +1,5 @@
+#!/bin/bash
+
+CUDA_VISIBLE_DEVICES=0,1,2,3 accelerate launch \
+    --config_file examples/accelerate/single_config.yaml \
+    src/train.py examples/full_multi_gpu/llama3_full_predict.yaml

examples/full_multi_gpu/single_node.sh

@@ -1,32 +1,15 @@
 #!/bin/bash
 
-deepspeed --num_gpus 4 ../../src/train_bash.py \
+NPROC_PER_NODE=4
-    --deepspeed ../deepspeed/ds_z3_config.json \
+NNODES=1
-    --stage sft \
+RANK=0
-    --do_train \
+MASTER_ADDR=127.0.0.1
-    --model_name_or_path meta-llama/Llama-2-7b-hf \
+MASTER_PORT=29500
-    --dataset alpaca_gpt4_en,glaive_toolcall \
+
-    --dataset_dir ../../data \
+CUDA_VISIBLE_DEVICES=0,1,2,3 torchrun \
-    --template default \
+    --nproc_per_node $NPROC_PER_NODE \
-    --finetuning_type full \
+    --nnodes $NNODES \
-    --output_dir ../../saves/LLaMA2-7B/full/sft \
+    --node_rank $RANK \
-    --overwrite_cache \
+    --master_addr $MASTER_ADDR \
-    --overwrite_output_dir \
+    --master_port $MASTER_PORT \
-    --cutoff_len 1024 \
+    src/train.py examples/full_multi_gpu/llama3_full_sft.yaml
-    --preprocessing_num_workers 16 \
-    --per_device_train_batch_size 1 \
-    --per_device_eval_batch_size 1 \
-    --gradient_accumulation_steps 2 \
-    --lr_scheduler_type cosine \
-    --logging_steps 10 \
-    --warmup_steps 20 \
-    --save_steps 100 \
-    --eval_steps 100 \
-    --evaluation_strategy steps \
-    --learning_rate 5e-5 \
-    --num_train_epochs 3.0 \
-    --max_samples 3000 \
-    --val_size 0.1 \
-    --ddp_timeout 1800000 \
-    --plot_loss \
-    --fp16

examples/inference/llama3.yaml

@@ -0,0 +1,2 @@
+model_name_or_path: meta-llama/Meta-Llama-3-8B-Instruct
+template: llama3

examples/inference/llama3_lora_sft.yaml

@@ -0,0 +1,4 @@
+model_name_or_path: meta-llama/Meta-Llama-3-8B-Instruct
+adapter_name_or_path: saves/llama3-8b/lora/sft
+template: llama3
+finetuning_type: lora

examples/inference/llama3_vllm.yaml

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

examples/lora_multi_gpu/ds_zero3.sh

@@ -0,0 +1,15 @@
+#!/bin/bash
+
+NPROC_PER_NODE=4
+NNODES=1
+RANK=0
+MASTER_ADDR=127.0.0.1
+MASTER_PORT=29500
+
+CUDA_VISIBLE_DEVICES=0,1,2,3 torchrun \
+    --nproc_per_node $NPROC_PER_NODE \
+    --nnodes $NNODES \
+    --node_rank $RANK \
+    --master_addr $MASTER_ADDR \
+    --master_port $MASTER_PORT \
+    src/train.py examples/lora_multi_gpu/llama3_lora_sft_ds.yaml

examples/lora_multi_gpu/llama3_lora_sft.yaml

@@ -0,0 +1,41 @@
+# model
+model_name_or_path: meta-llama/Meta-Llama-3-8B-Instruct
+
+# method
+stage: sft
+do_train: true
+finetuning_type: lora
+lora_target: q_proj,v_proj
+
+# ddp
+ddp_timeout: 180000000
+
+# dataset
+dataset: identity,alpaca_gpt4_en
+template: llama3
+cutoff_len: 1024
+max_samples: 1000
+overwrite_cache: true
+preprocessing_num_workers: 16
+
+# output
+output_dir: saves/llama3-8b/lora/sft
+logging_steps: 10
+save_steps: 500
+plot_loss: true
+overwrite_output_dir: true
+
+# train
+per_device_train_batch_size: 1
+gradient_accumulation_steps: 2
+learning_rate: 0.0001
+num_train_epochs: 3.0
+lr_scheduler_type: cosine
+warmup_steps: 0.1
+fp16: true
+
+# eval
+val_size: 0.1
+per_device_eval_batch_size: 1
+evaluation_strategy: steps
+eval_steps: 500

examples/lora_multi_gpu/llama3_lora_sft_ds.yaml

@@ -0,0 +1,42 @@
+# model
+model_name_or_path: meta-llama/Meta-Llama-3-8B-Instruct
+
+# method
+stage: sft
+do_train: true
+finetuning_type: lora
+lora_target: q_proj,v_proj
+
+# ddp
+ddp_timeout: 180000000
+deepspeed: examples/deepspeed/ds_z3_config.json
+
+# dataset
+dataset: identity,alpaca_gpt4_en
+template: llama3
+cutoff_len: 1024
+max_samples: 1000
+overwrite_cache: true
+preprocessing_num_workers: 16
+
+# output
+output_dir: saves/llama3-8b/lora/sft
+logging_steps: 10
+save_steps: 500
+plot_loss: true
+overwrite_output_dir: true
+
+# train
+per_device_train_batch_size: 1
+gradient_accumulation_steps: 2
+learning_rate: 0.0001
+num_train_epochs: 3.0
+lr_scheduler_type: cosine
+warmup_steps: 0.1
+fp16: true
+
+# eval
+val_size: 0.1
+per_device_eval_batch_size: 1
+evaluation_strategy: steps
+eval_steps: 500

examples/lora_multi_gpu/multi_node.sh

@@ -1,35 +1,6 @@
 #!/bin/bash
+# also launch it on slave machine using slave_config.yaml
 
 CUDA_VISIBLE_DEVICES=0,1,2,3 accelerate launch \
-    --config_file ../accelerate/master_config.yaml \
+    --config_file examples/accelerate/master_config.yaml \
-    ../../src/train_bash.py \
+    src/train.py examples/lora_multi_gpu/llama3_lora_sft.yaml
-    --stage sft \
-    --do_train \
-    --model_name_or_path meta-llama/Llama-2-7b-hf \
-    --dataset alpaca_gpt4_en,glaive_toolcall \
-    --dataset_dir ../../data \
-    --template default \
-    --finetuning_type lora \
-    --lora_target q_proj,v_proj \
-    --output_dir ../../saves/LLaMA2-7B/lora/sft \
-    --overwrite_cache \
-    --overwrite_output_dir \
-    --cutoff_len 1024 \
-    --preprocessing_num_workers 16 \
-    --per_device_train_batch_size 1 \
-    --per_device_eval_batch_size 1 \
-    --gradient_accumulation_steps 2 \
-    --lr_scheduler_type cosine \
-    --logging_steps 10 \
-    --warmup_steps 20 \
-    --save_steps 100 \
-    --eval_steps 100 \
-    --evaluation_strategy steps \
-    --load_best_model_at_end \
-    --learning_rate 5e-5 \
-    --num_train_epochs 3.0 \
-    --max_samples 3000 \
-    --val_size 0.1 \
-    --ddp_timeout 1800000 \
-    --plot_loss \
-    --fp16

examples/lora_multi_gpu/single_node.sh

@@ -1,35 +1,5 @@
 #!/bin/bash
 
-CUDA_VISIBLE_DEVICES=0,1,2,3,4,5,6,7 accelerate launch \
+CUDA_VISIBLE_DEVICES=0,1,2,3 accelerate launch \
-    --config_file ../accelerate/single_config.yaml \
+    --config_file examples/accelerate/single_config.yaml \
-    ../../src/train_bash.py \
+    src/train.py examples/lora_multi_gpu/llama3_lora_sft.yaml
-    --stage sft \
-    --do_train \
-    --model_name_or_path meta-llama/Llama-2-7b-hf \
-    --dataset alpaca_gpt4_en,glaive_toolcall \
-    --dataset_dir ../../data \
-    --template default \
-    --finetuning_type lora \
-    --lora_target q_proj,v_proj \
-    --output_dir ../../saves/LLaMA2-7B/lora/sft \
-    --overwrite_cache \
-    --overwrite_output_dir \
-    --cutoff_len 1024 \
-    --preprocessing_num_workers 16 \
-    --per_device_train_batch_size 1 \
-    --per_device_eval_batch_size 1 \
-    --gradient_accumulation_steps 2 \
-    --lr_scheduler_type cosine \
-    --logging_steps 10 \
-    --warmup_steps 20 \
-    --save_steps 100 \
-    --eval_steps 100 \
-    --evaluation_strategy steps \
-    --load_best_model_at_end \
-    --learning_rate 5e-5 \
-    --num_train_epochs 3.0 \
-    --max_samples 3000 \
-    --val_size 0.1 \
-    --ddp_timeout 1800000 \
-    --plot_loss \
-    --fp16

examples/lora_multi_npu/ds_zero0.sh

@@ -0,0 +1,15 @@
+#!/bin/bash
+
+NPROC_PER_NODE=4
+NNODES=1
+RANK=0
+MASTER_ADDR=127.0.0.1
+MASTER_PORT=29500
+
+ASCEND_RT_VISIBLE_DEVICES=0,1,2,3 torchrun \
+    --nproc_per_node $NPROC_PER_NODE \
+    --nnodes $NNODES \
+    --node_rank $RANK \
+    --master_addr $MASTER_ADDR \
+    --master_port $MASTER_PORT \
+    src/train.py examples/lora_multi_npu/llama3_lora_sft_ds.yaml

examples/lora_multi_npu/llama3_lora_sft_ds.yaml

@@ -0,0 +1,42 @@
+# model
+model_name_or_path: meta-llama/Meta-Llama-3-8B-Instruct
+
+# method
+stage: sft
+do_train: true
+finetuning_type: lora
+lora_target: q_proj,v_proj
+
+# ddp
+ddp_timeout: 180000000
+deepspeed: examples/deepspeed/ds_z0_config.json
+
+# dataset
+dataset: identity,alpaca_gpt4_en
+template: llama3
+cutoff_len: 1024
+max_samples: 1000
+overwrite_cache: true
+preprocessing_num_workers: 16
+
+# output
+output_dir: saves/llama3-8b/lora/sft
+logging_steps: 10
+save_steps: 500
+plot_loss: true
+overwrite_output_dir: true
+
+# train
+per_device_train_batch_size: 1
+gradient_accumulation_steps: 2
+learning_rate: 0.0001
+num_train_epochs: 3.0
+lr_scheduler_type: cosine
+warmup_steps: 0.1
+fp16: true
+
+# eval
+val_size: 0.1
+per_device_eval_batch_size: 1
+evaluation_strategy: steps
+eval_steps: 500

examples/lora_single_gpu/README.md

@@ -1,8 +0,0 @@
-Usage:
-
-- `pretrain.sh`: do pre-train (optional)
-- `sft.sh`: do supervised fine-tune
-- `reward.sh`: do reward modeling (must after sft.sh)
-- `ppo.sh`: do PPO training (must after sft.sh and reward.sh)
-- `dpo.sh`: do DPO training (must after sft.sh)
-- `predict.sh`: do predict (must after sft.sh and dpo.sh)

examples/lora_single_gpu/dpo.sh

@@ -1,35 +0,0 @@
-#!/bin/bash
-
-CUDA_VISIBLE_DEVICES=0 python ../../src/train_bash.py \
-    --stage dpo \
-    --do_train \
-    --model_name_or_path meta-llama/Llama-2-7b-hf \
-    --adapter_name_or_path ../../saves/LLaMA2-7B/lora/sft \
-    --create_new_adapter \
-    --dataset comparison_gpt4_en \
-    --dataset_dir ../../data \
-    --template default \
-    --finetuning_type lora \
-    --lora_target q_proj,v_proj \
-    --output_dir ../../saves/LLaMA2-7B/lora/dpo \
-    --overwrite_cache \
-    --overwrite_output_dir \
-    --cutoff_len 1024 \
-    --preprocessing_num_workers 16 \
-    --per_device_train_batch_size 1 \
-    --per_device_eval_batch_size 1 \
-    --gradient_accumulation_steps 8 \
-    --lr_scheduler_type cosine \
-    --logging_steps 10 \
-    --warmup_steps 20 \
-    --save_steps 100 \
-    --eval_steps 100 \
-    --evaluation_strategy steps \
-    --load_best_model_at_end \
-    --learning_rate 1e-5 \
-    --num_train_epochs 1.0 \
-    --max_samples 1000 \
-    --val_size 0.1 \
-    --dpo_ftx 1.0 \
-    --plot_loss \
-    --fp16

examples/lora_single_gpu/llama3_lora_dpo.yaml

@@ -0,0 +1,39 @@
+# model
+model_name_or_path: meta-llama/Meta-Llama-3-8B-Instruct
+
+# method
+stage: dpo
+do_train: true
+finetuning_type: lora
+lora_target: q_proj,v_proj
+dpo_ftx: 1.0
+
+# dataset
+dataset: orca_rlhf
+template: llama3
+cutoff_len: 1024
+max_samples: 1000
+overwrite_cache: true
+preprocessing_num_workers: 16
+
+# output
+output_dir: saves/llama3-8b/lora/dpo
+logging_steps: 10
+save_steps: 500
+plot_loss: true
+overwrite_output_dir: true
+
+# train
+per_device_train_batch_size: 1
+gradient_accumulation_steps: 8
+learning_rate: 0.00001
+num_train_epochs: 3.0
+lr_scheduler_type: cosine
+warmup_steps: 0.1
+fp16: true
+
+# eval
+val_size: 0.1
+per_device_eval_batch_size: 1
+evaluation_strategy: steps
+eval_steps: 500

examples/lora_single_gpu/llama3_lora_eval.yaml

@@ -0,0 +1,19 @@
+# model
+model_name_or_path: meta-llama/Meta-Llama-3-8B-Instruct
+adapter_name_or_path: saves/llama3-8b/lora/sft
+
+# method
+finetuning_type: lora
+
+# dataset
+task: mmlu
+split: test
+template: fewshot
+lang: en
+n_shot: 5
+
+# output
+save_dir: saves/llama3-8b/lora/eval
+
+# eval
+batch_size: 4

examples/lora_single_gpu/llama3_lora_orpo.yaml

@@ -0,0 +1,38 @@
+# model
+model_name_or_path: meta-llama/Meta-Llama-3-8B-Instruct
+
+# method
+stage: orpo
+do_train: true
+finetuning_type: lora
+lora_target: q_proj,v_proj
+
+# dataset
+dataset: orca_rlhf
+template: llama3
+cutoff_len: 1024
+max_samples: 1000
+overwrite_cache: true
+preprocessing_num_workers: 16
+
+# output
+output_dir: saves/llama3-8b/lora/orpo
+logging_steps: 10
+save_steps: 500
+plot_loss: true
+overwrite_output_dir: true
+
+# train
+per_device_train_batch_size: 1
+gradient_accumulation_steps: 8
+learning_rate: 0.00001
+num_train_epochs: 3.0
+lr_scheduler_type: cosine
+warmup_steps: 0.1
+fp16: true
+
+# eval
+val_size: 0.1
+per_device_eval_batch_size: 1
+evaluation_strategy: steps
+eval_steps: 500

examples/lora_single_gpu/llama3_lora_ppo.yaml

@@ -0,0 +1,38 @@
+# model
+model_name_or_path: meta-llama/Meta-Llama-3-8B-Instruct
+reward_model: saves/llama3-8b/lora/reward
+
+# method
+stage: ppo
+do_train: true
+finetuning_type: lora
+lora_target: q_proj,v_proj
+
+# dataset
+dataset: identity,alpaca_gpt4_en
+template: llama3
+cutoff_len: 1024
+max_samples: 1000
+overwrite_cache: true
+preprocessing_num_workers: 16
+
+# output
+output_dir: saves/llama3-8b/lora/ppo
+logging_steps: 10
+save_steps: 500
+plot_loss: true
+overwrite_output_dir: true
+
+# train
+per_device_train_batch_size: 1
+gradient_accumulation_steps: 8
+learning_rate: 0.00001
+num_train_epochs: 3.0
+lr_scheduler_type: cosine
+warmup_steps: 0.1
+fp16: true
+
+# generate
+max_new_tokens: 512
+top_k: 0
+top_p: 0.9

examples/lora_single_gpu/llama3_lora_predict.yaml

@@ -0,0 +1,24 @@
+# model
+model_name_or_path: meta-llama/Meta-Llama-3-8B-Instruct
+adapter_name_or_path: saves/llama3-8b/lora/sft
+
+# method
+stage: sft
+do_predict: true
+finetuning_type: lora
+
+# dataset
+dataset: identity,alpaca_gpt4_en
+template: llama3
+cutoff_len: 1024
+max_samples: 50
+overwrite_cache: true
+preprocessing_num_workers: 16
+
+# output
+output_dir: saves/llama3-8b/lora/predict
+overwrite_output_dir: true
+
+# eval
+per_device_eval_batch_size: 1
+predict_with_generate: true

examples/lora_single_gpu/llama3_lora_pretrain.yaml

@@ -0,0 +1,37 @@
+# model
+model_name_or_path: meta-llama/Meta-Llama-3-8B-Instruct
+
+# method
+stage: pt
+do_train: true
+finetuning_type: lora
+lora_target: q_proj,v_proj
+
+# dataset
+dataset: c4_demo
+cutoff_len: 1024
+max_samples: 1000
+overwrite_cache: true
+preprocessing_num_workers: 16
+
+# output
+output_dir: saves/llama3-8b/lora/sft
+logging_steps: 10
+save_steps: 500
+plot_loss: true
+overwrite_output_dir: true
+
+# train
+per_device_train_batch_size: 1
+gradient_accumulation_steps: 8
+learning_rate: 0.0001
+num_train_epochs: 3.0
+lr_scheduler_type: cosine
+warmup_steps: 0.1
+fp16: true
+
+# eval
+val_size: 0.1
+per_device_eval_batch_size: 1
+evaluation_strategy: steps
+eval_steps: 500

examples/lora_single_gpu/llama3_lora_reward.yaml

@@ -0,0 +1,38 @@
+# model
+model_name_or_path: meta-llama/Meta-Llama-3-8B-Instruct
+
+# method
+stage: rm
+do_train: true
+finetuning_type: lora
+lora_target: q_proj,v_proj
+
+# dataset
+dataset: orca_rlhf
+template: llama3
+cutoff_len: 1024
+max_samples: 1000
+overwrite_cache: true
+preprocessing_num_workers: 16
+
+# output
+output_dir: saves/llama3-8b/lora/reward
+logging_steps: 10
+save_steps: 500
+plot_loss: true
+overwrite_output_dir: true
+
+# train
+per_device_train_batch_size: 1
+gradient_accumulation_steps: 8
+learning_rate: 0.00001
+num_train_epochs: 3.0
+lr_scheduler_type: cosine
+warmup_steps: 0.1
+fp16: true
+
+# eval
+val_size: 0.1
+per_device_eval_batch_size: 1
+evaluation_strategy: steps
+eval_steps: 500

examples/lora_single_gpu/llama3_lora_sft.yaml

@@ -0,0 +1,38 @@
+# model
+model_name_or_path: meta-llama/Meta-Llama-3-8B-Instruct
+
+# method
+stage: sft
+do_train: true
+finetuning_type: lora
+lora_target: q_proj,v_proj
+
+# dataset
+dataset: identity,alpaca_gpt4_en
+template: llama3
+cutoff_len: 1024
+max_samples: 1000
+overwrite_cache: true
+preprocessing_num_workers: 16
+
+# output
+output_dir: saves/llama3-8b/lora/sft
+logging_steps: 10
+save_steps: 500
+plot_loss: true
+overwrite_output_dir: true
+
+# train
+per_device_train_batch_size: 1
+gradient_accumulation_steps: 8
+learning_rate: 0.0001
+num_train_epochs: 3.0
+lr_scheduler_type: cosine
+warmup_steps: 0.1
+fp16: true
+
+# eval
+val_size: 0.1
+per_device_eval_batch_size: 1
+evaluation_strategy: steps
+eval_steps: 500

examples/lora_single_gpu/llama3_preprocess.yaml

@@ -0,0 +1,21 @@
+# model
+model_name_or_path: meta-llama/Meta-Llama-3-8B-Instruct
+
+# method
+stage: sft
+do_train: true
+finetuning_type: lora
+lora_target: q_proj,v_proj
+
+# dataset
+dataset: identity,alpaca_gpt4_en
+template: llama3
+cutoff_len: 1024
+max_samples: 1000
+overwrite_cache: true
+preprocessing_num_workers: 16
+tokenized_path: saves/llama3-8b/dataset/sft
+
+# output
+output_dir: saves/llama3-8b/lora/sft
+overwrite_output_dir: true

examples/lora_single_gpu/llava1_5_lora_sft.yaml

@@ -0,0 +1,39 @@
+# model
+model_name_or_path: llava-hf/llava-1.5-7b-hf
+visual_inputs: true
+
+# method
+stage: sft
+do_train: true
+finetuning_type: lora
+lora_target: q_proj,v_proj
+
+# dataset
+dataset: mllm_demo
+template: vicuna
+cutoff_len: 1024
+max_samples: 1000
+overwrite_cache: true
+preprocessing_num_workers: 16
+
+# output
+output_dir: saves/llava1_5-7b/lora/sft
+logging_steps: 10
+save_steps: 500
+plot_loss: true
+overwrite_output_dir: true
+
+# train
+per_device_train_batch_size: 1
+gradient_accumulation_steps: 8
+learning_rate: 0.0001
+num_train_epochs: 3.0
+lr_scheduler_type: cosine
+warmup_steps: 0.1
+fp16: true
+
+# eval
+val_size: 0.1
+per_device_eval_batch_size: 1
+evaluation_strategy: steps
+eval_steps: 500

examples/lora_single_gpu/ppo.sh

@@ -1,32 +0,0 @@
-#!/bin/bash
-
-CUDA_VISIBLE_DEVICES=0 python ../../src/train_bash.py \
-    --stage ppo \
-    --do_train \
-    --model_name_or_path meta-llama/Llama-2-7b-hf \
-    --adapter_name_or_path ../../saves/LLaMA2-7B/lora/sft \
-    --create_new_adapter \
-    --dataset alpaca_gpt4_en \
-    --dataset_dir ../../data \
-    --template default \
-    --finetuning_type lora \
-    --lora_target q_proj,v_proj \
-    --reward_model ../../saves/LLaMA2-7B/lora/reward \
-    --output_dir ../../saves/LLaMA2-7B/lora/ppo \
-    --overwrite_cache \
-    --overwrite_output_dir \
-    --cutoff_len 512 \
-    --preprocessing_num_workers 16 \
-    --per_device_train_batch_size 1 \
-    --gradient_accumulation_steps 8 \
-    --lr_scheduler_type cosine \
-    --logging_steps 10 \
-    --save_steps 100 \
-    --learning_rate 1e-5 \
-    --num_train_epochs 1.0 \
-    --max_samples 1000 \
-    --top_k 0 \
-    --top_p 0.9 \
-    --max_new_tokens 256 \
-    --plot_loss \
-    --fp16

examples/lora_single_gpu/predict.sh

@@ -1,19 +0,0 @@
-#!/bin/bash
-
-CUDA_VISIBLE_DEVICES=0 python ../../src/train_bash.py \
-    --stage sft \
-    --do_predict \
-    --model_name_or_path meta-llama/Llama-2-7b-hf \
-    --adapter_name_or_path ../../saves/LLaMA2-7B/lora/sft,../../saves/LLaMA2-7B/lora/dpo \
-    --dataset alpaca_gpt4_en,glaive_toolcall \
-    --dataset_dir ../../data \
-    --template default \
-    --finetuning_type lora \
-    --output_dir ../../saves/LLaMA2-7B/lora/predict \
-    --overwrite_cache \
-    --overwrite_output_dir \
-    --cutoff_len 1024 \
-    --preprocessing_num_workers 16 \
-    --per_device_eval_batch_size 1 \
-    --max_samples 20 \
-    --predict_with_generate

examples/lora_single_gpu/pretrain.sh

@@ -1,31 +0,0 @@
-#!/bin/bash
-
-CUDA_VISIBLE_DEVICES=0 python ../../src/train_bash.py \
-    --stage pt \
-    --do_train \
-    --model_name_or_path meta-llama/Llama-2-7b-hf \
-    --dataset c4_demo \
-    --dataset_dir ../../data \
-    --finetuning_type lora \
-    --lora_target q_proj,v_proj \
-    --output_dir ../../saves/LLaMA2-7B/lora/pretrain \
-    --overwrite_cache \
-    --overwrite_output_dir \
-    --cutoff_len 1024 \
-    --preprocessing_num_workers 16 \
-    --per_device_train_batch_size 1 \
-    --per_device_eval_batch_size 1 \
-    --gradient_accumulation_steps 8 \
-    --lr_scheduler_type cosine \
-    --logging_steps 10 \
-    --warmup_steps 20 \
-    --save_steps 100 \
-    --eval_steps 100 \
-    --evaluation_strategy steps \
-    --load_best_model_at_end \
-    --learning_rate 5e-5 \
-    --num_train_epochs 3.0 \
-    --max_samples 10000 \
-    --val_size 0.1 \
-    --plot_loss \
-    --fp16

examples/lora_single_gpu/reward.sh

@@ -1,33 +0,0 @@
-#!/bin/bash
-
-CUDA_VISIBLE_DEVICES=0 python ../../src/train_bash.py \
-    --stage rm \
-    --do_train \
-    --model_name_or_path meta-llama/Llama-2-7b-hf \
-    --adapter_name_or_path ../../saves/LLaMA2-7B/lora/sft \
-    --create_new_adapter \
-    --dataset comparison_gpt4_en \
-    --dataset_dir ../../data \
-    --template default \
-    --finetuning_type lora \
-    --lora_target q_proj,v_proj \
-    --output_dir ../../saves/LLaMA2-7B/lora/reward \
-    --overwrite_cache \
-    --overwrite_output_dir \
-    --cutoff_len 1024 \
-    --preprocessing_num_workers 16 \
-    --per_device_train_batch_size 1 \
-    --per_device_eval_batch_size 1 \
-    --gradient_accumulation_steps 8 \
-    --lr_scheduler_type cosine \
-    --logging_steps 10 \
-    --warmup_steps 20 \
-    --save_steps 100 \
-    --eval_steps 100 \
-    --evaluation_strategy steps \
-    --learning_rate 1e-5 \
-    --num_train_epochs 1.0 \
-    --max_samples 5000 \
-    --val_size 0.1 \
-    --plot_loss \
-    --fp16

examples/lora_single_gpu/sft.sh

@@ -1,32 +0,0 @@
-#!/bin/bash
-
-CUDA_VISIBLE_DEVICES=0 python ../../src/train_bash.py \
-    --stage sft \
-    --do_train \
-    --model_name_or_path meta-llama/Llama-2-7b-hf \
-    --dataset alpaca_gpt4_en,glaive_toolcall \
-    --dataset_dir ../../data \
-    --template default \
-    --finetuning_type lora \
-    --lora_target q_proj,v_proj \
-    --output_dir ../../saves/LLaMA2-7B/lora/sft \
-    --overwrite_cache \
-    --overwrite_output_dir \
-    --cutoff_len 1024 \
-    --preprocessing_num_workers 16 \
-    --per_device_train_batch_size 1 \
-    --per_device_eval_batch_size 1 \
-    --gradient_accumulation_steps 8 \
-    --lr_scheduler_type cosine \
-    --logging_steps 10 \
-    --warmup_steps 20 \
-    --save_steps 100 \
-    --eval_steps 100 \
-    --evaluation_strategy steps \
-    --load_best_model_at_end \
-    --learning_rate 5e-5 \
-    --num_train_epochs 3.0 \
-    --max_samples 3000 \
-    --val_size 0.1 \
-    --plot_loss \
-    --fp16

examples/merge_lora/README.md

@@ -1,4 +0,0 @@
-Usage:
-
-- `merge.sh`: merge the lora weights
-- `quantize.sh`: quantize the model with AutoGPTQ (must after merge.sh, optional)

examples/merge_lora/llama3_gptq.yaml

@@ -0,0 +1,11 @@
+# model
+model_name_or_path: meta-llama/Meta-Llama-3-8B-Instruct
+template: llama3
+
+# export
+export_dir: models/llama3_gptq
+export_quantization_bit: 4
+export_quantization_dataset: data/c4_demo.json
+export_size: 2
+export_device: cpu
+export_legacy_format: false

examples/merge_lora/llama3_lora_sft.yaml

@@ -0,0 +1,13 @@
+# Note: DO NOT use quantized model or quantization_bit when merging lora adapters
+
+# model
+model_name_or_path: meta-llama/Meta-Llama-3-8B-Instruct
+adapter_name_or_path: saves/llama3-8b/lora/sft
+template: llama3
+finetuning_type: lora
+
+# export
+export_dir: models/llama3_lora_sft
+export_size: 2
+export_device: cpu
+export_legacy_format: false

examples/merge_lora/merge.sh

@@ -1,10 +0,0 @@
-#!/bin/bash
-
-CUDA_VISIBLE_DEVICES=0 python ../../src/export_model.py \
-    --model_name_or_path meta-llama/Llama-2-7b-hf \
-    --adapter_name_or_path ../../saves/LLaMA2-7B/lora/sft \
-    --template default \
-    --finetuning_type lora \
-    --export_dir ../../models/llama2-7b-sft \
-    --export_size 2 \
-    --export_legacy_format False

examples/merge_lora/quantize.sh

@@ -1,10 +0,0 @@
-#!/bin/bash
-
-CUDA_VISIBLE_DEVICES=0 python ../../src/export_model.py \
-    --model_name_or_path ../../models/llama2-7b-sft \
-    --template default \
-    --export_dir ../../models/llama2-7b-sft-int4 \
-    --export_quantization_bit 4 \
-    --export_quantization_dataset ../../data/c4_demo.json \
-    --export_size 2 \
-    --export_legacy_format False

examples/qlora_single_gpu/aqlm.sh

@@ -1,30 +0,0 @@
-#!/bin/bash
-
-CUDA_VISIBLE_DEVICES=0 python ../../src/train_bash.py \
-    --stage sft \
-    --do_train \
-    --model_name_or_path BlackSamorez/Llama-2-7b-AQLM-2Bit-1x16-hf \
-    --dataset alpaca_gpt4_en,glaive_toolcall \
-    --dataset_dir ../../data \
-    --template default \
-    --finetuning_type lora \
-    --lora_target q_proj,v_proj \
-    --output_dir ../../saves/LLaMA2-7B/lora/sft \
-    --overwrite_cache \
-    --overwrite_output_dir \
-    --cutoff_len 1024 \
-    --per_device_train_batch_size 1 \
-    --per_device_eval_batch_size 1 \
-    --gradient_accumulation_steps 8 \
-    --lr_scheduler_type cosine \
-    --logging_steps 10 \
-    --save_steps 100 \
-    --eval_steps 100 \
-    --evaluation_strategy steps \
-    --load_best_model_at_end \
-    --learning_rate 5e-5 \
-    --num_train_epochs 3.0 \
-    --max_samples 3000 \
-    --val_size 0.1 \
-    --plot_loss \
-    --fp16

examples/qlora_single_gpu/awq.sh

@@ -1,30 +0,0 @@
-#!/bin/bash
-
-CUDA_VISIBLE_DEVICES=0 python ../../src/train_bash.py \
-    --stage sft \
-    --do_train \
-    --model_name_or_path TheBloke/Llama-2-7B-AWQ \
-    --dataset alpaca_gpt4_en,glaive_toolcall \
-    --dataset_dir ../../data \
-    --template default \
-    --finetuning_type lora \
-    --lora_target q_proj,v_proj \
-    --output_dir ../../saves/LLaMA2-7B/lora/sft \
-    --overwrite_cache \
-    --overwrite_output_dir \
-    --cutoff_len 1024 \
-    --per_device_train_batch_size 1 \
-    --per_device_eval_batch_size 1 \
-    --gradient_accumulation_steps 8 \
-    --lr_scheduler_type cosine \
-    --logging_steps 10 \
-    --save_steps 100 \
-    --eval_steps 100 \
-    --evaluation_strategy steps \
-    --load_best_model_at_end \
-    --learning_rate 5e-5 \
-    --num_train_epochs 3.0 \
-    --max_samples 3000 \
-    --val_size 0.1 \
-    --plot_loss \
-    --fp16

examples/qlora_single_gpu/bitsandbytes.sh

@@ -1,31 +0,0 @@
-#!/bin/bash
-
-CUDA_VISIBLE_DEVICES=0 python ../../src/train_bash.py \
-    --stage sft \
-    --do_train \
-    --model_name_or_path meta-llama/Llama-2-7b-hf \
-    --dataset alpaca_gpt4_en,glaive_toolcall \
-    --dataset_dir ../../data \
-    --template default \
-    --finetuning_type lora \
-    --lora_target q_proj,v_proj \
-    --output_dir ../../saves/LLaMA2-7B/lora/sft \
-    --overwrite_cache \
-    --overwrite_output_dir \
-    --cutoff_len 1024 \
-    --per_device_train_batch_size 1 \
-    --per_device_eval_batch_size 1 \
-    --gradient_accumulation_steps 8 \
-    --lr_scheduler_type cosine \
-    --logging_steps 10 \
-    --save_steps 100 \
-    --eval_steps 100 \
-    --evaluation_strategy steps \
-    --load_best_model_at_end \
-    --learning_rate 5e-5 \
-    --num_train_epochs 3.0 \
-    --max_samples 3000 \
-    --val_size 0.1 \
-    --quantization_bit 4 \
-    --plot_loss \
-    --fp16

examples/qlora_single_gpu/gptq.sh

@@ -1,30 +0,0 @@
-#!/bin/bash
-
-CUDA_VISIBLE_DEVICES=0 python ../../src/train_bash.py \
-    --stage sft \
-    --do_train \
-    --model_name_or_path TheBloke/Llama-2-7B-GPTQ \
-    --dataset alpaca_gpt4_en,glaive_toolcall \
-    --dataset_dir ../../data \
-    --template default \
-    --finetuning_type lora \
-    --lora_target q_proj,v_proj \
-    --output_dir ../../saves/LLaMA2-7B/lora/sft \
-    --overwrite_cache \
-    --overwrite_output_dir \
-    --cutoff_len 1024 \
-    --per_device_train_batch_size 1 \
-    --per_device_eval_batch_size 1 \
-    --gradient_accumulation_steps 8 \
-    --lr_scheduler_type cosine \
-    --logging_steps 10 \
-    --save_steps 100 \
-    --eval_steps 100 \
-    --evaluation_strategy steps \
-    --load_best_model_at_end \
-    --learning_rate 5e-5 \
-    --num_train_epochs 3.0 \
-    --max_samples 3000 \
-    --val_size 0.1 \
-    --plot_loss \
-    --fp16

examples/qlora_single_gpu/llama3_lora_sft_aqlm.yaml

@@ -0,0 +1,38 @@
+# model
+model_name_or_path: ISTA-DASLab/Meta-Llama-3-8B-Instruct-AQLM-2Bit-1x16
+
+# method
+stage: sft
+do_train: true
+finetuning_type: lora
+lora_target: q_proj,v_proj
+
+# dataset
+dataset: identity,alpaca_gpt4_en
+template: llama3
+cutoff_len: 1024
+max_samples: 1000
+overwrite_cache: true
+preprocessing_num_workers: 16
+
+# output
+output_dir: saves/llama3-8b/lora/sft
+logging_steps: 10
+save_steps: 500
+plot_loss: true
+overwrite_output_dir: true
+
+# train
+per_device_train_batch_size: 1
+gradient_accumulation_steps: 8
+learning_rate: 0.0001
+num_train_epochs: 3.0
+lr_scheduler_type: cosine
+warmup_steps: 0.1
+fp16: true
+
+# eval
+val_size: 0.1
+per_device_eval_batch_size: 1
+evaluation_strategy: steps
+eval_steps: 500

examples/qlora_single_gpu/llama3_lora_sft_awq.yaml

@@ -0,0 +1,38 @@
+# model
+model_name_or_path: TechxGenus/Meta-Llama-3-8B-Instruct-AWQ
+
+# method
+stage: sft
+do_train: true
+finetuning_type: lora
+lora_target: q_proj,v_proj
+
+# dataset
+dataset: identity,alpaca_gpt4_en
+template: llama3
+cutoff_len: 1024
+max_samples: 1000
+overwrite_cache: true
+preprocessing_num_workers: 16
+
+# output
+output_dir: saves/llama3-8b/lora/sft
+logging_steps: 10
+save_steps: 500
+plot_loss: true
+overwrite_output_dir: true
+
+# train
+per_device_train_batch_size: 1
+gradient_accumulation_steps: 8
+learning_rate: 0.0001
+num_train_epochs: 3.0
+lr_scheduler_type: cosine
+warmup_steps: 0.1
+fp16: true
+
+# eval
+val_size: 0.1
+per_device_eval_batch_size: 1
+evaluation_strategy: steps
+eval_steps: 500

examples/qlora_single_gpu/llama3_lora_sft_bitsandbytes.yaml

@@ -0,0 +1,39 @@
+# model
+model_name_or_path: meta-llama/Meta-Llama-3-8B-Instruct
+quantization_bit: 4
+
+# method
+stage: sft
+do_train: true
+finetuning_type: lora
+lora_target: q_proj,v_proj
+
+# dataset
+dataset: identity,alpaca_gpt4_en
+template: llama3
+cutoff_len: 1024
+max_samples: 1000
+overwrite_cache: true
+preprocessing_num_workers: 16
+
+# output
+output_dir: saves/llama3-8b/lora/sft
+logging_steps: 10
+save_steps: 500
+plot_loss: true
+overwrite_output_dir: true
+
+# train
+per_device_train_batch_size: 1
+gradient_accumulation_steps: 8
+learning_rate: 0.0001
+num_train_epochs: 3.0
+lr_scheduler_type: cosine
+warmup_steps: 0.1
+fp16: true
+
+# eval
+val_size: 0.1
+per_device_eval_batch_size: 1
+evaluation_strategy: steps
+eval_steps: 500

examples/qlora_single_gpu/llama3_lora_sft_gptq.yaml

@@ -0,0 +1,38 @@
+# model
+model_name_or_path: TechxGenus/Meta-Llama-3-8B-Instruct-GPTQ
+
+# method
+stage: sft
+do_train: true
+finetuning_type: lora
+lora_target: q_proj,v_proj
+
+# dataset
+dataset: identity,alpaca_gpt4_en
+template: llama3
+cutoff_len: 1024
+max_samples: 1000
+overwrite_cache: true
+preprocessing_num_workers: 16
+
+# output
+output_dir: saves/llama3-8b/lora/sft
+logging_steps: 10
+save_steps: 500
+plot_loss: true
+overwrite_output_dir: true
+
+# train
+per_device_train_batch_size: 1
+gradient_accumulation_steps: 8
+learning_rate: 0.0001
+num_train_epochs: 3.0
+lr_scheduler_type: cosine
+warmup_steps: 0.1
+fp16: true
+
+# eval
+val_size: 0.1
+per_device_eval_batch_size: 1
+evaluation_strategy: steps
+eval_steps: 500

requirements.txt

@@ -1,10 +1,9 @@
-torch>=1.13.1
 transformers>=4.37.2
 datasets>=2.14.3
 accelerate>=0.27.2
 peft>=0.10.0
 trl>=0.8.1
-gradio>=3.38.0,<4.0.0
+gradio>=4.0.0
 scipy
 einops
 sentencepiece
@@ -13,6 +12,7 @@ uvicorn
 pydantic
 fastapi
 sse-starlette
-matplotlib
+matplotlib>=3.7.0
 fire
-galore-torch
+packaging
+pyyaml

scripts/cal_flops.py

@@ -3,24 +3,22 @@
 # Usage: python cal_flops.py --model_name_or_path path_to_model --batch_size 1 --seq_length 512
 # Inspired by: https://www.deepspeed.ai/tutorials/flops-profiler/
 
-from typing import Optional
-
 import fire
 import torch
 from deepspeed.accelerator import get_accelerator  # type: ignore
 from deepspeed.profiling.flops_profiler import get_model_profile  # type: ignore
 
-from llmtuner import ChatModel
+from llmtuner.chat import ChatModel
 
 
 def calculate_flops(
     model_name_or_path: str,
-    batch_size: Optional[int] = 1,
+    batch_size: int = 1,
-    seq_length: Optional[int] = 256,
+    seq_length: int = 256,
-    flash_attn: Optional[bool] = False,
+    flash_attn: str = "auto",
 ):
     with get_accelerator().device(0):
-        chat_model = ChatModel(dict(model_name_or_path=model_name_or_path, template="vanilla", flash_attn=flash_attn))
+        chat_model = ChatModel(dict(model_name_or_path=model_name_or_path, template="empty", flash_attn=flash_attn))
         fake_input = torch.ones((batch_size, seq_length), dtype=torch.long, device=chat_model.model.device)
         input_dict = {"input_ids": fake_input, "labels": fake_input.clone()}
         flops, macs, params = get_model_profile(chat_model.model, kwargs=input_dict, print_profile=True, detailed=True)

scripts/cal_lr.py

@@ -4,7 +4,7 @@
 # Inspired by: https://github.com/imoneoi/openchat/blob/master/ochat/training_deepspeed/train.py
 
 import math
-from typing import Optional
+from typing import Literal
 
 import fire
 import torch
@@ -15,7 +15,7 @@ from transformers import DataCollatorForLanguageModeling, DataCollatorForSeq2Seq
 from llmtuner.data import get_dataset
 from llmtuner.extras.constants import IGNORE_INDEX
 from llmtuner.hparams import get_train_args
-from llmtuner.model import load_model_and_tokenizer
+from llmtuner.model import load_tokenizer
 
 
 BASE_LR = 3e-4  # 1.5e-4 for 30B-70B models
@@ -25,14 +25,14 @@ BASE_BS = 4_000_000  # from llama paper
 def calculate_lr(
     model_name_or_path: str,
     batch_size: int,  # total batch size, namely (batch size * gradient accumulation * world size)
-    stage: Optional[str] = "sft",
+    stage: Literal["pt", "sft"] = "sft",
-    dataset: Optional[str] = "alpaca_en",
+    dataset: str = "alpaca_en",
-    dataset_dir: Optional[str] = "data",
+    dataset_dir: str = "data",
-    template: Optional[str] = "default",
+    template: str = "default",
-    cutoff_len: Optional[int] = 1024,  # i.e. maximum input length during training
+    cutoff_len: int = 1024,  # i.e. maximum input length during training
-    is_mistral: Optional[bool] = False,  # mistral model uses a smaller learning rate,
+    is_mistral: bool = False,  # mistral model uses a smaller learning rate,
 ):
-    model_args, data_args, training_args, finetuning_args, _ = get_train_args(
+    model_args, data_args, training_args, _, _ = get_train_args(
         dict(
             stage=stage,
             model_name_or_path=model_name_or_path,
@@ -44,8 +44,9 @@ def calculate_lr(
             overwrite_cache=True,
         )
     )
-    _, tokenizer = load_model_and_tokenizer(model_args, finetuning_args, is_trainable=False, add_valuehead=False)
+    tokenizer_module = load_tokenizer(model_args)
-    trainset = get_dataset(tokenizer, model_args, data_args, training_args, stage=stage)
+    tokenizer = tokenizer_module["tokenizer"]
+    trainset = get_dataset(model_args, data_args, training_args, stage, **tokenizer_module)
     if stage == "pt":
         data_collator = DataCollatorForLanguageModeling(tokenizer=tokenizer, mlm=False)
     elif stage == "sft":
@@ -53,9 +54,7 @@ def calculate_lr(
     else:
         raise NotImplementedError
 
-    dataloader = DataLoader(
+    dataloader = DataLoader(trainset, batch_size, shuffle=False, collate_fn=data_collator, pin_memory=True)
-        dataset=trainset, batch_size=batch_size, shuffle=True, collate_fn=data_collator, pin_memory=True
-    )
     valid_tokens, total_tokens = 0, 0
     for batch in tqdm(dataloader):
         valid_tokens += torch.sum(batch["labels"] != IGNORE_INDEX).item()

scripts/cal_ppl.py

@@ -0,0 +1,116 @@
+# coding=utf-8
+# Calculates the ppl on the dataset of the pre-trained models.
+# Usage: python cal_ppl.py --model_name_or_path path_to_model --save_name ppl.json
+
+import json
+from dataclasses import dataclass
+from typing import Any, Dict, Literal, Optional, Sequence
+
+import fire
+import torch
+from torch.utils.data import DataLoader
+from tqdm import tqdm
+from transformers import DataCollatorForLanguageModeling, DataCollatorForSeq2Seq
+
+from llmtuner.data import get_dataset
+from llmtuner.extras.constants import IGNORE_INDEX
+from llmtuner.hparams import get_train_args
+from llmtuner.model import load_model, load_tokenizer
+
+
+@dataclass
+class PairwiseDataCollatorWithPadding(DataCollatorForSeq2Seq):
+    r"""
+    Data collator for pairwise data.
+    """
+
+    train_on_prompt: bool = False
+
+    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.
+        """
+        chosen_features = []
+        for feature in features:
+            prompt_len, answer_len = len(feature["prompt_ids"]), len(feature["chosen_ids"])
+            input_ids = feature["prompt_ids"] + feature["chosen_ids"]
+            attention_mask = [1] * (prompt_len + answer_len)
+            labels = input_ids if self.train_on_prompt else [IGNORE_INDEX] * prompt_len + feature["chosen_ids"]
+            chosen_features.append({"input_ids": input_ids, "attention_mask": attention_mask, "labels": labels})
+
+        return super().__call__(chosen_features)
+
+
+def cal_ppl(
+    model_name_or_path: str,
+    save_name: str,
+    batch_size: int = 4,
+    stage: Literal["pt", "sft", "rm"] = "sft",
+    dataset: str = "alpaca_en",
+    dataset_dir: str = "data",
+    template: str = "default",
+    cutoff_len: int = 1024,
+    max_samples: Optional[int] = None,
+    train_on_prompt: bool = False,
+):
+    model_args, data_args, training_args, finetuning_args, _ = get_train_args(
+        dict(
+            stage=stage,
+            model_name_or_path=model_name_or_path,
+            dataset=dataset,
+            dataset_dir=dataset_dir,
+            template=template,
+            cutoff_len=cutoff_len,
+            max_samples=max_samples,
+            train_on_prompt=train_on_prompt,
+            output_dir="dummy_dir",
+            overwrite_cache=True,
+        )
+    )
+    tokenizer_module = load_tokenizer(model_args)
+    tokenizer = tokenizer_module["tokenizer"]
+    trainset = get_dataset(model_args, data_args, training_args, stage, **tokenizer_module)
+    model = load_model(tokenizer, model_args, finetuning_args, is_trainable=False)
+    if stage == "pt":
+        data_collator = DataCollatorForLanguageModeling(tokenizer=tokenizer, mlm=False)
+    elif stage == "sft":
+        data_collator = DataCollatorForSeq2Seq(tokenizer=tokenizer, label_pad_token_id=IGNORE_INDEX)
+    elif stage == "rm":
+        data_collator = PairwiseDataCollatorWithPadding(
+            tokenizer=tokenizer, label_pad_token_id=IGNORE_INDEX, train_on_prompt=train_on_prompt
+        )
+    else:
+        raise NotImplementedError
+
+    dataloader = DataLoader(trainset, batch_size, shuffle=False, collate_fn=data_collator, pin_memory=True)
+    criterion = torch.nn.CrossEntropyLoss(reduction="none")
+    total_ppl = 0
+    perplexities = []
+    batch: Dict[str, "torch.Tensor"]
+    with torch.no_grad():
+        for batch in tqdm(dataloader):
+            batch = batch.to(model.device)
+            outputs = model(**batch)
+            shift_logits: "torch.Tensor" = outputs["logits"][..., :-1, :]
+            shift_labels: "torch.Tensor" = batch["labels"][..., 1:]
+            loss_mask = shift_labels != IGNORE_INDEX
+            flatten_logits = shift_logits.contiguous().view(shift_labels.size(0) * shift_labels.size(1), -1)
+            flatten_labels = shift_labels.contiguous().view(-1)
+            token_logps: "torch.Tensor" = criterion(flatten_logits, flatten_labels)
+            token_logps = token_logps.contiguous().view(shift_logits.size(0), -1)
+            sentence_logps = (token_logps * loss_mask).sum(-1) / loss_mask.sum(-1)
+            total_ppl += sentence_logps.exp().sum().item()
+            perplexities.extend(sentence_logps.exp().tolist())
+
+    with open(save_name, "w", encoding="utf-8") as f:
+        json.dump(perplexities, f, indent=2)
+
+    print("Average perplexity is {:.2f}".format(total_ppl / len(perplexities)))
+    print("Perplexities have been saved at {}.".format(save_name))
+
+
+if __name__ == "__main__":
+    fire.Fire(cal_ppl)

scripts/length_cdf.py

@@ -3,24 +3,23 @@
 # Usage: python length_cdf.py --model_name_or_path path_to_model --dataset alpaca_en --template default
 
 from collections import defaultdict
-from typing import Optional
 
 import fire
 from tqdm import tqdm
 
 from llmtuner.data import get_dataset
 from llmtuner.hparams import get_train_args
-from llmtuner.model import load_model_and_tokenizer
+from llmtuner.model import load_tokenizer
 
 
 def length_cdf(
     model_name_or_path: str,
-    dataset: Optional[str] = "alpaca_en",
+    dataset: str = "alpaca_en",
-    dataset_dir: Optional[str] = "data",
+    dataset_dir: str = "data",
-    template: Optional[str] = "default",
+    template: str = "default",
-    interval: Optional[int] = 1000,
+    interval: int = 1000,
 ):
-    model_args, data_args, training_args, finetuning_args, _ = get_train_args(
+    model_args, data_args, training_args, _, _ = get_train_args(
         dict(
             stage="sft",
             model_name_or_path=model_name_or_path,
@@ -32,8 +31,8 @@ def length_cdf(
             overwrite_cache=True,
         )
     )
-    _, tokenizer = load_model_and_tokenizer(model_args, finetuning_args, is_trainable=False, add_valuehead=False)
+    tokenizer_module = load_tokenizer(model_args)
-    trainset = get_dataset(tokenizer, model_args, data_args, training_args, stage="sft")
+    trainset = get_dataset(model_args, data_args, training_args, stage="sft", **tokenizer_module)
     total_num = len(trainset)
     length_dict = defaultdict(int)
     for sample in tqdm(trainset["input_ids"]):

scripts/llama_pro.py

@@ -1,5 +1,5 @@
 # coding=utf-8
-# Performs block expansion for LLaMA, Mistral or Qwen1.5 models.
+# Performs block expansion for LLaMA, Mistral, Qwen1.5 or Yi models.
 # Usage: python llama_pro.py --model_name_or_path meta-llama/Llama-2-7b-hf --output_dir llama2_pro --num_expand 8
 # Inspired by: https://github.com/TencentARC/LLaMA-Pro/blob/main/scripts/block_expansion.py
 
@@ -106,8 +106,7 @@ def block_expansion(
     print("Fine-tune this model with:")
     print("  --model_name_or_path {} \\".format(output_dir))
     print("  --finetuning_type freeze \\")
-    print("  --name_module_trainable all \\")
+    print("  --freeze_trainable_layers {} \\".format(num_expand))
-    print("  --num_layer_trainable {} \\".format(num_expand))
     print("  --use_llama_pro")
 
 

scripts/llamafy_internlm2.py

@@ -1,114 +0,0 @@
-# coding=utf-8
-# Converts the InternLM2 model in the same format as LLaMA2.
-# Usage: python llamafy_internlm2.py --input_dir input --output_dir output
-# Warning: We have found that the converted model cannot infer correctly. It will be fixed later.
-
-import json
-import os
-from collections import OrderedDict
-from typing import Any, Dict, Optional
-
-import fire
-import torch
-from safetensors.torch import save_file
-from tqdm import tqdm
-from transformers.modeling_utils import (
-    SAFE_WEIGHTS_INDEX_NAME,
-    SAFE_WEIGHTS_NAME,
-    WEIGHTS_INDEX_NAME,
-    WEIGHTS_NAME,
-    shard_checkpoint,
-)
-
-
-CONFIG_NAME = "config.json"
-
-
-def save_weight(input_dir: str, output_dir: str, shard_size: str, save_safetensors: bool):
-    with open(os.path.join(input_dir, CONFIG_NAME), "r", encoding="utf-8") as f:
-        internlm2_config_dict: Dict[str, Any] = json.load(f)
-
-    internlm2_state_dict: Dict[str, torch.Tensor] = OrderedDict()
-    for filepath in tqdm(os.listdir(input_dir), desc="Load weights"):
-        if os.path.isfile(os.path.join(input_dir, filepath)) and filepath.endswith(".bin"):
-            shard_weight = torch.load(os.path.join(input_dir, filepath), map_location="cpu")
-            internlm2_state_dict.update(shard_weight)
-
-    llama2_state_dict: Dict[str, torch.Tensor] = OrderedDict()
-    for key, value in tqdm(internlm2_state_dict.items(), desc="Convert format"):
-        if "output" in key:
-            llama2_state_dict[key.replace("output", "lm_head")] = value
-        elif "tok_embeddings" in key:
-            llama2_state_dict[key.replace("tok_embeddings", "embed_tokens")] = value
-        elif "wqkv" in key:
-            num_q_heads = internlm2_config_dict["num_attention_heads"]
-            num_kv_heads = internlm2_config_dict["num_key_value_heads"]
-            q_size = value.size(0) // (num_q_heads + 2 * num_kv_heads) * num_q_heads
-            kv_size = value.size(0) // (num_q_heads + 2 * num_kv_heads) * num_kv_heads
-            llama2_state_dict[key.replace("attention.wqkv", "self_attn.q_proj")] = value[:q_size, ...]
-            llama2_state_dict[key.replace("attention.wqkv", "self_attn.k_proj")] = value[
-                q_size : q_size + kv_size, ...
-            ]
-            llama2_state_dict[key.replace("attention.wqkv", "self_attn.v_proj")] = value[q_size + kv_size :, ...]
-        elif "wo" in key:
-            llama2_state_dict[key.replace("attention.wo", "self_attn.o_proj")] = value
-        elif "attention_norm" in key:
-            llama2_state_dict[key.replace("attention_norm", "input_layernorm")] = value
-        elif "ffn_norm" in key:
-            llama2_state_dict[key.replace("ffn_norm", "post_attention_layernorm")] = value
-        elif "w1" in key:
-            llama2_state_dict[key.replace("feed_forward.w1", "mlp.gate_proj")] = value
-        elif "w2" in key:
-            llama2_state_dict[key.replace("feed_forward.w2", "mlp.down_proj")] = value
-        elif "w3" in key:
-            llama2_state_dict[key.replace("feed_forward.w3", "mlp.up_proj")] = value
-        else:
-            llama2_state_dict[key] = value
-
-    weights_name = SAFE_WEIGHTS_NAME if save_safetensors else WEIGHTS_NAME
-    shards, index = shard_checkpoint(llama2_state_dict, max_shard_size=shard_size, weights_name=weights_name)
-
-    for shard_file, shard in tqdm(shards.items(), desc="Save weights"):
-        if save_safetensors:
-            save_file(shard, os.path.join(output_dir, shard_file), metadata={"format": "pt"})
-        else:
-            torch.save(shard, os.path.join(output_dir, shard_file))
-
-    if index is None:
-        print("Model weights saved in {}".format(os.path.join(output_dir, WEIGHTS_NAME)))
-    else:
-        index_name = SAFE_WEIGHTS_INDEX_NAME if save_safetensors else WEIGHTS_INDEX_NAME
-        with open(os.path.join(output_dir, index_name), "w", encoding="utf-8") as f:
-            json.dump(index, f, indent=2, sort_keys=True)
-        print("Model weights saved in {}".format(output_dir))
-
-
-def save_config(input_dir: str, output_dir: str):
-    with open(os.path.join(input_dir, CONFIG_NAME), "r", encoding="utf-8") as f:
-        llama2_config_dict: Dict[str, Any] = json.load(f)
-
-    llama2_config_dict["architectures"] = ["LlamaForCausalLM"]
-    llama2_config_dict.pop("auto_map", None)
-    llama2_config_dict.pop("bias", None)
-    llama2_config_dict.pop("rope_scaling", None)
-    llama2_config_dict["model_type"] = "llama"
-
-    with open(os.path.join(output_dir, CONFIG_NAME), "w", encoding="utf-8") as f:
-        json.dump(llama2_config_dict, f, indent=2)
-    print("Model config saved in {}".format(os.path.join(output_dir, CONFIG_NAME)))
-
-
-def llamafy_internlm2(
-    input_dir: str, output_dir: str, shard_size: Optional[str] = "2GB", save_safetensors: Optional[bool] = False
-):
-    try:
-        os.makedirs(output_dir, exist_ok=False)
-    except Exception as e:
-        raise print("Output dir already exists", e)
-
-    save_weight(input_dir, output_dir, shard_size, save_safetensors)
-    save_config(input_dir, output_dir)
-
-
-if __name__ == "__main__":
-    fire.Fire(llamafy_internlm2)

setup.py

@@ -5,9 +5,9 @@ from setuptools import find_packages, setup
 
 
 def get_version():
-    with open(os.path.join("src", "llmtuner", "__init__.py"), "r", encoding="utf-8") as f:
+    with open(os.path.join("src", "llmtuner", "cli.py"), "r", encoding="utf-8") as f:
         file_content = f.read()
-        pattern = r"{0}\W*=\W*\"([^\"]+)\"".format("__version__")
+        pattern = r"{}\W*=\W*\"([^\"]+)\"".format("VERSION")
         (version,) = re.findall(pattern, file_content)
         return version
 
@@ -20,15 +20,18 @@ def get_requires():
 
 
 extra_require = {
-    "deepspeed": ["deepspeed"],
+    "torch": ["torch>=1.13.1"],
     "metrics": ["nltk", "jieba", "rouge-chinese"],
-    "unsloth": ["torch==2.2.0", "unsloth[cu121-ampere-torch220]"],
+    "deepspeed": ["deepspeed>=0.10.0,<=0.14.0"],
-    "vllm": ["vllm>=0.3.3"],
     "bitsandbytes": ["bitsandbytes>=0.39.0"],
+    "vllm": ["vllm>=0.4.0"],
+    "galore": ["galore-torch"],
+    "badam": ["badam"],
     "gptq": ["optimum>=1.16.0", "auto-gptq>=0.5.0"],
     "awq": ["autoawq"],
     "aqlm": ["aqlm[gpu]>=1.1.0"],
     "qwen": ["tiktoken", "transformers_stream_generator"],
+    "modelscope": ["modelscope"],
     "quality": ["ruff"],
 }
 
@@ -50,6 +53,7 @@ def main():
         python_requires=">=3.8.0",
         install_requires=get_requires(),
         extras_require=extra_require,
+        entry_points={"console_scripts": ["llamafactory-cli = llmtuner.cli:main"]},
         classifiers=[
             "Development Status :: 4 - Beta",
             "Intended Audience :: Developers",

src/api.py

@@ -0,0 +1,19 @@
+import os
+
+import uvicorn
+
+from llmtuner.api.app import create_app
+from llmtuner.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()

src/api_demo.py

@@ -1,16 +0,0 @@
-import os
-
-import uvicorn
-
-from llmtuner import ChatModel, create_app
-
-
-def main():
-    chat_model = ChatModel()
-    app = create_app(chat_model)
-    print("Visit http://localhost:{}/docs for API document.".format(os.environ.get("API_PORT", 8000)))
-    uvicorn.run(app, host="0.0.0.0", port=int(os.environ.get("API_PORT", 8000)), workers=1)
-
-
-if __name__ == "__main__":
-    main()

src/cli_demo.py

@@ -1,49 +0,0 @@
-from llmtuner import ChatModel
-from llmtuner.extras.misc import torch_gc
-
-
-try:
-    import platform
-
-    if platform.system() != "Windows":
-        import readline  # noqa: F401
-except ImportError:
-    print("Install `readline` for a better experience.")
-
-
-def main():
-    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})
-
-
-if __name__ == "__main__":
-    main()

src/evaluate.py

@@ -1,10 +0,0 @@
-from llmtuner import Evaluator
-
-
-def main():
-    evaluator = Evaluator()
-    evaluator.eval()
-
-
-if __name__ == "__main__":
-    main()

src/export_model.py

@@ -1,9 +0,0 @@
-from llmtuner import export_model
-
-
-def main():
-    export_model()
-
-
-if __name__ == "__main__":
-    main()

src/llmtuner/__init__.py

@@ -1,11 +1,6 @@
 # Level: api, webui > chat, eval, train > data, model > extras, hparams
 
-from .api import create_app
+from .cli import VERSION
-from .chat import ChatModel
-from .eval import Evaluator
-from .train import export_model, run_exp
-from .webui import create_ui, create_web_demo
 
 
-__version__ = "0.6.1"
+__version__ = VERSION
-__all__ = ["create_app", "ChatModel", "Evaluator", "export_model", "run_exp", "create_ui", "create_web_demo"]

src/llmtuner/api/__init__.py

@@ -1,4 +0,0 @@
-from .app import create_app
-
-
-__all__ = ["create_app"]

src/llmtuner/api/app.py

@@ -1,36 +1,31 @@
-import json
 import os
 from contextlib import asynccontextmanager
-from typing import Any, Dict, Sequence
+from typing import Optional
 
-from pydantic import BaseModel
+from typing_extensions import Annotated
 
 from ..chat import ChatModel
-from ..data import Role as DataRole
 from ..extras.misc import torch_gc
-from ..extras.packages import is_fastapi_availble, is_starlette_available, is_uvicorn_available
+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 (
-    ChatCompletionMessage,
     ChatCompletionRequest,
     ChatCompletionResponse,
-    ChatCompletionResponseChoice,
-    ChatCompletionResponseStreamChoice,
-    ChatCompletionResponseUsage,
-    ChatCompletionStreamResponse,
-    Finish,
-    Function,
-    FunctionCall,
     ModelCard,
     ModelList,
-    Role,
     ScoreEvaluationRequest,
     ScoreEvaluationResponse,
 )
 
 
-if is_fastapi_availble():
+if is_fastapi_available():
-    from fastapi import FastAPI, HTTPException, status
+    from fastapi import Depends, FastAPI, HTTPException, status
     from fastapi.middleware.cors import CORSMiddleware
+    from fastapi.security.http import HTTPAuthorizationCredentials, HTTPBearer
 
 
 if is_starlette_available():
@@ -47,23 +42,8 @@ async def lifespan(app: "FastAPI"):  # collects GPU memory
     torch_gc()
 
 
-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)
-
-
 def create_app(chat_model: "ChatModel") -> "FastAPI":
     app = FastAPI(lifespan=lifespan)
-
     app.add_middleware(
         CORSMiddleware,
         allow_origins=["*"],
@@ -71,154 +51,58 @@ def create_app(chat_model: "ChatModel") -> "FastAPI":
         allow_methods=["*"],
         allow_headers=["*"],
     )
+    api_key = os.environ.get("API_KEY")
+    security = HTTPBearer(auto_error=False)
 
-    role_mapping = {
+    async def verify_api_key(auth: Annotated[Optional[HTTPAuthorizationCredentials], Depends(security)]):
-        Role.USER: DataRole.USER.value,
+        if api_key and (auth is None or auth.credentials != api_key):
-        Role.ASSISTANT: DataRole.ASSISTANT.value,
+            raise HTTPException(status_code=status.HTTP_401_UNAUTHORIZED, detail="Invalid API key.")
-        Role.SYSTEM: DataRole.SYSTEM.value,
-        Role.FUNCTION: DataRole.FUNCTION.value,
-        Role.TOOL: DataRole.OBSERVATION.value,
-    }
 
-    @app.get("/v1/models", response_model=ModelList)
+    @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="gpt-3.5-turbo")
         return ModelList(data=[model_card])
 
-    @app.post("/v1/chat/completions", response_model=ChatCompletionResponse, status_code=status.HTTP_200_OK)
+    @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 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 = ""
-
-        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 = []
-        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")
-
-            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([tool["function"] for tool in tool_list], ensure_ascii=False)
-            except Exception:
-                raise HTTPException(status_code=status.HTTP_400_BAD_REQUEST, detail="Invalid tools")
-        else:
-            tools = ""
-
         if request.stream:
-            if tools:
+            generate = create_stream_chat_completion_response(request, chat_model)
-                raise HTTPException(status_code=status.HTTP_400_BAD_REQUEST, detail="Cannot stream function calls.")
-
-            generate = stream_chat_completion(input_messages, system, tools, request)
             return EventSourceResponse(generate, media_type="text/event-stream")
+        else:
+            return await create_chat_completion_response(request, chat_model)
 
-        responses = await chat_model.achat(
+    @app.post(
-            input_messages,
+        "/v1/score/evaluation",
-            system,
+        response_model=ScoreEvaluationResponse,
-            tools,
+        status_code=status.HTTP_200_OK,
-            do_sample=request.do_sample,
+        dependencies=[Depends(verify_api_key)],
-            temperature=request.temperature,
+    )
-            top_p=request.top_p,
-            max_new_tokens=request.max_tokens,
-            num_return_sequences=request.n,
-        )
-
-        prompt_length, response_length = 0, 0
-        choices = []
-        for i, response in enumerate(responses):
-            if tools:
-                result = chat_model.engine.template.format_tools.extract(response.response_text)
-            else:
-                result = response.response_text
-
-            if isinstance(result, tuple):
-                name, arguments = result
-                function = Function(name=name, arguments=arguments)
-                response_message = ChatCompletionMessage(
-                    role=Role.ASSISTANT, tool_calls=[FunctionCall(function=function)]
-                )
-                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(model=request.model, choices=choices, usage=usage)
-
-    async def stream_chat_completion(
-        messages: Sequence[Dict[str, str]], system: str, tools: str, request: ChatCompletionRequest
-    ):
-        choice_data = ChatCompletionResponseStreamChoice(
-            index=0, delta=ChatCompletionMessage(role=Role.ASSISTANT, content=""), finish_reason=None
-        )
-        chunk = ChatCompletionStreamResponse(model=request.model, choices=[choice_data])
-        yield jsonify(chunk)
-
-        async for new_token in chat_model.astream_chat(
-            messages,
-            system,
-            tools,
-            do_sample=request.do_sample,
-            temperature=request.temperature,
-            top_p=request.top_p,
-            max_new_tokens=request.max_tokens,
-        ):
-            if len(new_token) == 0:
-                continue
-
-            choice_data = ChatCompletionResponseStreamChoice(
-                index=0, delta=ChatCompletionMessage(content=new_token), finish_reason=None
-            )
-            chunk = ChatCompletionStreamResponse(model=request.model, choices=[choice_data])
-            yield jsonify(chunk)
-
-        choice_data = ChatCompletionResponseStreamChoice(
-            index=0, delta=ChatCompletionMessage(), finish_reason=Finish.STOP
-        )
-        chunk = ChatCompletionStreamResponse(model=request.model, choices=[choice_data])
-        yield jsonify(chunk)
-        yield "[DONE]"
-
-    @app.post("/v1/score/evaluation", response_model=ScoreEvaluationResponse, status_code=status.HTTP_200_OK)
     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")
 
-        if len(request.messages) == 0:
+        return await create_score_evaluation_response(request, chat_model)
-            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(model=request.model, scores=scores)
 
     return app
 
 
-if __name__ == "__main__":
+def run_api() -> None:
     chat_model = ChatModel()
     app = create_app(chat_model)
-    uvicorn.run(app, host="0.0.0.0", port=int(os.environ.get("API_PORT", 8000)), workers=1)
+    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)

src/llmtuner/api/chat.py

@@ -0,0 +1,186 @@
+import json
+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
+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 TYPE_CHECKING:
+    from ..chat import ChatModel
+    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]], str, str]:
+    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 = ""
+
+    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 = []
+    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):
+            name = message.tool_calls[0].function.name
+            arguments = message.tool_calls[0].function.arguments
+            content = json.dumps({"name": name, "argument": arguments}, ensure_ascii=False)
+            input_messages.append({"role": ROLE_MAPPING[Role.FUNCTION], "content": content})
+        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 Exception:
+            raise HTTPException(status_code=status.HTTP_400_BAD_REQUEST, detail="Invalid tools")
+    else:
+        tools = ""
+
+    return input_messages, system, tools
+
+
+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 = _process_request(request)
+    responses = await chat_model.achat(
+        input_messages,
+        system,
+        tools,
+        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.format_tools.extract(response.response_text)
+        else:
+            result = response.response_text
+
+        if isinstance(result, tuple):
+            name, arguments = result
+            function = Function(name=name, arguments=arguments)
+            tool_call = FunctionCall(id="call_{}".format(uuid.uuid4().hex), function=function)
+            response_message = ChatCompletionMessage(role=Role.ASSISTANT, tool_calls=[tool_call])
+            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 = _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,
+        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":
+    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(model=request.model, scores=scores)

src/llmtuner/api/common.py

@@ -0,0 +1,20 @@
+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)

src/llmtuner/api/protocol.py

@@ -1,6 +1,6 @@
 import time
 from enum import Enum, unique
-from typing import List, Optional
+from typing import Any, Dict, List, Optional, Union
 
 from pydantic import BaseModel, Field
 from typing_extensions import Literal
@@ -39,15 +39,27 @@ class Function(BaseModel):
     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: Literal["call_default"] = "call_default"
+    id: str
     type: Literal["function"] = "function"
     function: Function
 
 
 class ChatMessage(BaseModel):
     role: Role
-    content: str
+    content: Optional[str] = None
+    tool_calls: Optional[List[FunctionCall]] = None
 
 
 class ChatCompletionMessage(BaseModel):
@@ -59,12 +71,13 @@ class ChatCompletionMessage(BaseModel):
 class ChatCompletionRequest(BaseModel):
     model: str
     messages: List[ChatMessage]
-    tools: list = []
+    tools: Optional[List[FunctionAvailable]] = None
     do_sample: bool = True
     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
 
 
@@ -74,7 +87,7 @@ class ChatCompletionResponseChoice(BaseModel):
     finish_reason: Finish
 
 
-class ChatCompletionResponseStreamChoice(BaseModel):
+class ChatCompletionStreamResponseChoice(BaseModel):
     index: int
     delta: ChatCompletionMessage
     finish_reason: Optional[Finish] = None
@@ -87,7 +100,7 @@ class ChatCompletionResponseUsage(BaseModel):
 
 
 class ChatCompletionResponse(BaseModel):
-    id: Literal["chatcmpl-default"] = "chatcmpl-default"
+    id: str
     object: Literal["chat.completion"] = "chat.completion"
     created: int = Field(default_factory=lambda: int(time.time()))
     model: str
@@ -96,11 +109,11 @@ class ChatCompletionResponse(BaseModel):
 
 
 class ChatCompletionStreamResponse(BaseModel):
-    id: Literal["chatcmpl-default"] = "chatcmpl-default"
+    id: str
     object: Literal["chat.completion.chunk"] = "chat.completion.chunk"
     created: int = Field(default_factory=lambda: int(time.time()))
     model: str
-    choices: List[ChatCompletionResponseStreamChoice]
+    choices: List[ChatCompletionStreamResponseChoice]
 
 
 class ScoreEvaluationRequest(BaseModel):
@@ -110,7 +123,7 @@ class ScoreEvaluationRequest(BaseModel):
 
 
 class ScoreEvaluationResponse(BaseModel):
-    id: Literal["scoreeval-default"] = "scoreeval-default"
+    id: str
     object: Literal["score.evaluation"] = "score.evaluation"
     model: str
     scores: List[float]

src/llmtuner/chat/base_engine.py

@@ -4,15 +4,13 @@ from typing import TYPE_CHECKING, Any, AsyncGenerator, Dict, List, Literal, Opti
 
 
 if TYPE_CHECKING:
+    from numpy.typing import NDArray
     from transformers import PreTrainedModel, PreTrainedTokenizer
+    from vllm import AsyncLLMEngine
 
     from ..data import Template
-    from ..extras.packages import is_vllm_available
     from ..hparams import DataArguments, FinetuningArguments, GeneratingArguments, ModelArguments
 
-    if is_vllm_available():
-        from vllm import AsyncLLMEngine
-
 
 @dataclass
 class Response:
@@ -49,6 +47,7 @@ class BaseEngine(ABC):
         messages: Sequence[Dict[str, str]],
         system: Optional[str] = None,
         tools: Optional[str] = None,
+        image: Optional["NDArray"] = None,
         **input_kwargs,
     ) -> List["Response"]: ...
 
@@ -58,6 +57,7 @@ class BaseEngine(ABC):
         messages: Sequence[Dict[str, str]],
         system: Optional[str] = None,
         tools: Optional[str] = None,
+        image: Optional["NDArray"] = None,
         **input_kwargs,
     ) -> AsyncGenerator[str, None]: ...