release v0.1.5

Former-commit-id: d619e76bc4098c29a7fdc05f5a71208bd1079c9f
Merge pull request #307 from GitYCC/feature/fix-llama2-prompt-template
2023-08-02 16:10:31 +08:00 · 2023-08-02 15:51:28 +08:00 · 2023-08-02 14:35:35 +08:00 · 2023-08-02 14:10:15 +08:00 · 2023-08-02 01:10:28 +08:00 · 2023-08-01 18:48:27 +08:00
46 changed files with 1063 additions and 415 deletions
--- a/README.md
+++ b/README.md
@@ -8,17 +8,23 @@

 👋 Join our [WeChat](assets/wechat.jpg).

+\[ English | [中文](README_zh.md) \]
+
 ## Changelog

-[23/07/19] Now we support training the **LLaMA-2** models in this repo. Try `--model_name_or_path meta-llama/Llama-2-7b-hf` argument to use the LLaMA-2 model. Remember to use `--prompt_template llama2` argument when you are using the LLaMA-2-chat model.
+[23/07/31] Now we support dataset streaming. Try `--streaming` and `--max_steps 100` arguments to stream your dataset.
+
+[23/07/29] We release two instruction-tuned 13B models at Hugging Face. See these Hugging Face Repos ([LLaMA-2](https://huggingface.co/hiyouga/Llama-2-Chinese-13b-chat) / [Baichuan](https://huggingface.co/hiyouga/baichuan-13b-sft)) for details.
+
+[23/07/19] Now we support training the **LLaMA-2** models in this repo. Try `--model_name_or_path meta-llama/Llama-2-7b-hf` argument to use the LLaMA-2 model. Remember to use `--template llama2` argument when you are using the LLaMA-2-chat model.

 [23/07/18] Now we develop an all-in-one Web UI for training, evaluation and inference. Try `train_web.py` to fine-tune models in your Web browser. Thank [@KanadeSiina](https://github.com/KanadeSiina) and [@codemayq](https://github.com/codemayq) for their efforts in the development.

-[23/07/11] Now we support training the **Baichuan-13B** model in this repo. Please replace the Baichuan-13B model file with `tests/modeling_baichuan.py` and try `--model_name_or_path path_to_baichuan_model` and `--lora_target W_pack` arguments to train the Baichuan-13B model. Remember to use `--prompt_template baichuan` argument when you are using the Baichuan-13B-Chat model.
+[23/07/11] Now we support training the **Baichuan-13B** model in this repo. Try `--model_name_or_path baichuan-inc/Baichuan-13B-Base` and `--lora_target W_pack` arguments to train the Baichuan-13B model. Remember to use `--template baichuan` argument when you are using the Baichuan-13B-Chat model.

 [23/07/09] Now we release [FastEdit](https://github.com/hiyouga/FastEdit)⚡🩹, an easy-to-use package for editing the factual knowledge of large language models efficiently. Please follow [FastEdit](https://github.com/hiyouga/FastEdit) if you are interested.

-[23/07/07] Now we support training the **InternLM-7B** model in this repo. Try `--model_name_or_path internlm/internlm-7b` argument to use the InternLM model. Remember to use `--prompt_template intern` argument when you are using the InternLM-chat model.
+[23/07/07] Now we support training the **InternLM-7B** model in this repo. Try `--model_name_or_path internlm/internlm-7b` argument to use the InternLM model. Remember to use `--template intern` argument when you are using the InternLM-chat model.

 [23/07/05] Now we support training the **Falcon-7B/40B** models in this repo. Try `--model_name_or_path tiiuae/falcon-7b` and `--lora_target query_key_value` arguments to use the Falcon model.

@@ -28,7 +34,7 @@

 [23/06/15] Now we support training the **Baichuan-7B** model in this repo. Try `--model_name_or_path baichuan-inc/Baichuan-7B` and `--lora_target W_pack` arguments to use the Baichuan-7B model.

-[23/06/03] Now we support quantized training and inference (aka **[QLoRA](https://github.com/artidoro/qlora)**). Try `--quantization_bit 4/8` argument to work with quantized model. (experimental feature)
+[23/06/03] Now we support quantized training and inference (aka **[QLoRA](https://github.com/artidoro/qlora)**). Try `--quantization_bit 4/8` argument to work with quantized models.

 [23/05/31] Now we support training the **BLOOM & BLOOMZ** models in this repo. Try `--model_name_or_path bigscience/bloomz-7b1-mt` and `--lora_target query_key_value` arguments to use the BLOOMZ model.

@@ -61,6 +67,10 @@

 - For pre-training:
  - [Wiki Demo (en)](data/wiki_demo.txt)
+  - [RefinedWeb (en)](https://huggingface.co/datasets/tiiuae/falcon-refinedweb)
+  - [StarCoder (en)](https://huggingface.co/datasets/bigcode/starcoderdata)
+  - [Wikipedia (en)](https://huggingface.co/datasets/olm/olm-wikipedia-20221220)
+  - [Wikipedia (zh)](https://huggingface.co/datasets/pleisto/wikipedia-cn-20230720-filtered)
 - For supervised fine-tuning:
  - [Stanford Alpaca (en)](https://github.com/tatsu-lab/stanford_alpaca)
  - [Stanford Alpaca (zh)](https://github.com/ymcui/Chinese-LLaMA-Alpaca)
@@ -77,6 +87,7 @@
  - [BELLE School Math 0.25M (zh)](https://huggingface.co/datasets/BelleGroup/school_math_0.25M)
  - [BELLE Multiturn Chat 0.8M (zh)](https://huggingface.co/datasets/BelleGroup/multiturn_chat_0.8M)
  - [Firefly 1.1M (zh)](https://huggingface.co/datasets/YeungNLP/firefly-train-1.1M)
+  - [LIMA (en)](https://huggingface.co/datasets/GAIR/lima)
  - [CodeAlpaca 20k (en)](https://huggingface.co/datasets/sahil2801/CodeAlpaca-20k)
  - [Alpaca CoT (multilingual)](https://huggingface.co/datasets/QingyiSi/Alpaca-CoT)
  - [Web QA (zh)](https://huggingface.co/datasets/suolyer/webqa)
@@ -106,12 +117,6 @@ huggingface-cli login

 And **powerful GPUs**!

-If you want to enable quantized LoRA (QLoRA) on the Windows platform, you should install a pre-built version of `bitsandbytes` library, which supports CUDA 11.1 to 12.1.
-
-```bash
-pip install https://github.com/jllllll/bitsandbytes-windows-webui/releases/download/wheels/bitsandbytes-0.39.1-py3-none-win_amd64.whl
-```
-
 ## Getting Started

 ### Data Preparation (optional)
@@ -123,6 +128,7 @@ Note: please update `data/dataset_info.json` to use your custom dataset. About t
 ### Dependence Installation (optional)

 ```bash
+git lfs install
 git clone https://github.com/hiyouga/LLaMA-Efficient-Tuning.git
 conda create -n llama_etuning python=3.10
 conda activate llama_etuning
@@ -130,12 +136,20 @@ cd LLaMA-Efficient-Tuning
 pip install -r requirements.txt
 ```

+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 12.1.
+
+```bash
+pip install https://github.com/jllllll/bitsandbytes-windows-webui/releases/download/wheels/bitsandbytes-0.39.1-py3-none-win_amd64.whl
+```
+
 ### All-in-one Web UI

 ```bash
-python src/train_web.py
+CUDA_VISIBLE_DEVICES=0 python src/train_web.py
 ```

+Currently the web UI only supports training on **a single GPU**.
+
 ### (Continually) Pre-Training

 ```bash
@@ -144,6 +158,7 @@ CUDA_VISIBLE_DEVICES=0 python src/train_bash.py \
    --model_name_or_path path_to_your_model \
    --do_train \
    --dataset wiki_demo \
+    --template default \
    --finetuning_type lora \
    --output_dir path_to_pt_checkpoint \
    --overwrite_cache \
@@ -166,6 +181,7 @@ CUDA_VISIBLE_DEVICES=0 python src/train_bash.py \
    --model_name_or_path path_to_your_model \
    --do_train \
    --dataset alpaca_gpt4_en \
+    --template default \
    --finetuning_type lora \
    --output_dir path_to_sft_checkpoint \
    --overwrite_cache \
@@ -180,6 +196,8 @@ CUDA_VISIBLE_DEVICES=0 python src/train_bash.py \
    --fp16
 ```

+Remember to specify `--lora_target W_pack` if you are using Baichuan models.
+
 ### Reward Model Training

 ```bash
@@ -188,7 +206,10 @@ CUDA_VISIBLE_DEVICES=0 python src/train_bash.py \
    --model_name_or_path path_to_your_model \
    --do_train \
    --dataset comparison_gpt4_en \
+    --template default \
    --finetuning_type lora \
+    --resume_lora_training False \
+    --checkpoint_dir path_to_sft_checkpoint \
    --output_dir path_to_rm_checkpoint \
    --per_device_train_batch_size 4 \
    --gradient_accumulation_steps 4 \
@@ -209,7 +230,9 @@ CUDA_VISIBLE_DEVICES=0 python src/train_bash.py \
    --model_name_or_path path_to_your_model \
    --do_train \
    --dataset alpaca_gpt4_en \
+    --template default \
    --finetuning_type lora \
+    --resume_lora_training False \
    --checkpoint_dir path_to_sft_checkpoint \
    --reward_model path_to_rm_checkpoint \
    --output_dir path_to_ppo_checkpoint \
@@ -220,7 +243,6 @@ CUDA_VISIBLE_DEVICES=0 python src/train_bash.py \
    --save_steps 1000 \
    --learning_rate 1e-5 \
    --num_train_epochs 1.0 \
-    --resume_lora_training False \
    --plot_loss
 ```

@@ -267,6 +289,7 @@ CUDA_VISIBLE_DEVICES=0 python src/train_bash.py \
    --model_name_or_path path_to_your_model \
    --do_eval \
    --dataset alpaca_gpt4_en \
+    --template default \
    --finetuning_type lora \
    --checkpoint_dir path_to_checkpoint \
    --output_dir path_to_eval_result \
@@ -285,6 +308,7 @@ CUDA_VISIBLE_DEVICES=0 python src/train_bash.py \
    --model_name_or_path path_to_your_model \
    --do_predict \
    --dataset alpaca_gpt4_en \
+    --template default \
    --finetuning_type lora \
    --checkpoint_dir path_to_checkpoint \
    --output_dir path_to_predict_result \
@@ -293,13 +317,12 @@ CUDA_VISIBLE_DEVICES=0 python src/train_bash.py \
    --predict_with_generate
 ```

-If you want to predict the samples with empty responses, please kindly fill the `response` column with **dummy tokens** to ensure the sample will not be discarded throughout the preprocessing phase.
-
 ### API Demo

 ```bash
 python src/api_demo.py \
    --model_name_or_path path_to_your_model \
+    --template default \
    --finetuning_type lora \
    --checkpoint_dir path_to_checkpoint
 ```
@@ -311,6 +334,7 @@ Visit `http://localhost:8000/docs` for API documentation.
 ```bash
 python src/cli_demo.py \
    --model_name_or_path path_to_your_model \
+    --template default \
    --finetuning_type lora \
    --checkpoint_dir path_to_checkpoint
 ```
@@ -320,6 +344,7 @@ python src/cli_demo.py \
 ```bash
 python src/web_demo.py \
    --model_name_or_path path_to_your_model \
+    --template default \
    --finetuning_type lora \
    --checkpoint_dir path_to_checkpoint
 ```
@@ -329,11 +354,18 @@ python src/web_demo.py \
 ```bash
 python src/export_model.py \
    --model_name_or_path path_to_your_model \
+    --template default \
    --finetuning_type lora \
    --checkpoint_dir path_to_checkpoint \
    --output_dir path_to_export
 ```

+## TODO
+
+- [ ] Supporting flash attention ([torch](https://pytorch.org/docs/stable/generated/torch.nn.functional.scaled_dot_product_attention.html) / [xformers](https://github.com/facebookresearch/xformers) / [flashattn](https://github.com/Dao-AILab/flash-attention)).
+- [ ] Implementing multi-query attention for faster inference.
+- [ ] Supporting full-parameter RLHF training.
+
 ## License

 This repository is licensed under the [Apache-2.0 License](LICENSE).
@@ -344,7 +376,7 @@ Please follow the model licenses to use the corresponding model weights:
 - [LLaMA-2](https://ai.meta.com/llama/license/)
 - [BLOOM](https://huggingface.co/spaces/bigscience/license)
 - [Falcon](LICENSE)
- [baichuan](https://huggingface.co/baichuan-inc/baichuan-7B/resolve/main/baichuan-7B%20%E6%A8%A1%E5%9E%8B%E8%AE%B8%E5%8F%AF%E5%8D%8F%E8%AE%AE.pdf)
+- [Baichuan](https://huggingface.co/baichuan-inc/baichuan-7B/resolve/main/baichuan-7B%20%E6%A8%A1%E5%9E%8B%E8%AE%B8%E5%8F%AF%E5%8D%8F%E8%AE%AE.pdf)
 - [InternLM](https://github.com/InternLM/InternLM#open-source-license)

 ## Citation
--- a/README_zh.md
+++ b/README_zh.md
@@ -0,0 +1,401 @@
+# LLaMA Efficient Tuning
+
+[![GitHub Repo stars](https://img.shields.io/github/stars/hiyouga/LLaMA-Efficient-Tuning?style=social)](https://github.com/hiyouga/LLaMA-Efficient-Tuning/stargazers)
+[![GitHub Code License](https://img.shields.io/github/license/hiyouga/LLaMA-Efficient-Tuning)](LICENSE)
+[![GitHub last commit](https://img.shields.io/github/last-commit/hiyouga/LLaMA-Efficient-Tuning)](https://github.com/hiyouga/LLaMA-Efficient-Tuning/commits/main)
+[![PyPI](https://img.shields.io/pypi/v/llmtuner)](https://pypi.org/project/llmtuner/)
+[![GitHub pull request](https://img.shields.io/badge/PRs-welcome-blue)](https://github.com/hiyouga/LLaMA-Efficient-Tuning/pulls)
+
+👋 加入我们的[微信群](assets/wechat.jpg)。
+
+\[ [English](README.md) | 中文 \]
+
+## 更新日志
+
+[23/07/31] 现在我们支持了训练数据流式加载。请尝试使用 `--streaming` 和 `--max_steps 100` 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)）。
+
+[23/07/19] 现在我们支持了 **LLaMA-2** 模型的训练。请尝试使用 `--model_name_or_path meta-llama/Llama-2-7b-hf` 参数。请注意使用 LLaMA-2-chat 模型需要添加 `--template llama2` 参数。
+
+[23/07/18] 我们开发了支持训练和测试的浏览器一键微调界面。请尝试使用 `train_web.py` 在您的浏览器中微调模型。感谢 [@KanadeSiina](https://github.com/KanadeSiina) 和 [@codemayq](https://github.com/codemayq) 在该功能开发中付出的努力。
+
+[23/07/11] 现在我们支持了 **Baichuan-13B** 模型的训练。请尝试使用 `--model_name_or_path path_to_baichuan_model` 和 `--lora_target W_pack` 参数。请注意使用 Baichuan-13B-Chat 模型需要添加 `--template baichuan` 参数。
+
+[23/07/09] 我们开源了 [FastEdit](https://github.com/hiyouga/FastEdit)⚡🩹，一个简单易用的、能迅速编辑大模型事实记忆的工具包。如果您感兴趣请关注我们的 [FastEdit](https://github.com/hiyouga/FastEdit) 项目。
+
+[23/07/07] 现在我们支持了 **InternLM-7B** 模型的训练。请尝试使用 `--model_name_or_path internlm/internlm-7b` 参数。请注意使用 InternLM-chat 模型需要添加 `--template intern` 参数。
+
+[23/07/05] 现在我们支持了 **Falcon-7B/40B** 模型的训练。请尝试使用 `--model_name_or_path tiiuae/falcon-7b` 和 `--lora_target query_key_value` 参数。
+
+[23/06/29] 我们提供了一个**可复现的**指令模型微调示例，详细内容请查阅 [Hugging Face 项目](https://huggingface.co/hiyouga/baichuan-7b-sft)。
+
+[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。
+
+[23/06/15] 现在我们支持了 **Baichuan-7B** 模型的训练。请尝试使用 `--model_name_or_path baichuan-inc/Baichuan-7B` 和 `--lora_target W_pack` 参数。
+
+[23/06/03] 现在我们实现了 4 比特的 LoRA 训练（也称 [QLoRA](https://github.com/artidoro/qlora)）。请尝试使用 `--quantization_bit 4` 参数进行 4 比特量化微调。
+
+[23/05/31] 现在我们支持了 **BLOOM & BLOOMZ** 模型的训练。请尝试使用 `--model_name_or_path bigscience/bloomz-7b1-mt` 和 `--lora_target query_key_value` 参数。
+
+## 模型
+
+- [LLaMA](https://github.com/facebookresearch/llama) (7B/13B/33B/65B)
+- [LLaMA-2](https://huggingface.co/meta-llama) (7B/13B/70B)
+- [BLOOM](https://huggingface.co/bigscience/bloom) & [BLOOMZ](https://huggingface.co/bigscience/bloomz) (560M/1.1B/1.7B/3B/7.1B/176B)
+- [Falcon](https://huggingface.co/tiiuae/falcon-7b) (7B/40B)
+- [Baichuan](https://huggingface.co/baichuan-inc/baichuan-7B) (7B/13B)
+- [InternLM](https://github.com/InternLM/InternLM) (7B)
+
+## 微调方法
+
+- [二次预训练](https://s3-us-west-2.amazonaws.com/openai-assets/research-covers/language-unsupervised/language_understanding_paper.pdf)
+  - 全参数微调
+  - 部分参数微调
+  - [LoRA](https://arxiv.org/abs/2106.09685)
+  - [QLoRA](https://arxiv.org/abs/2305.14314)
+- [指令监督微调](https://arxiv.org/abs/2109.01652)
+  - 全参数微调
+  - 部分参数微调
+  - [LoRA](https://arxiv.org/abs/2106.09685)
+  - [QLoRA](https://arxiv.org/abs/2305.14314)
+- [人类反馈的强化学习（RLHF）](https://arxiv.org/abs/2203.02155)
+  - [LoRA](https://arxiv.org/abs/2106.09685)
+  - [QLoRA](https://arxiv.org/abs/2305.14314)
+
+## 数据集
+
+- 用于二次预训练:
+  - [Wiki Demo (en)](data/wiki_demo.txt)
+  - [RefinedWeb (en)](https://huggingface.co/datasets/tiiuae/falcon-refinedweb)
+  - [StarCoder (en)](https://huggingface.co/datasets/bigcode/starcoderdata)
+  - [Wikipedia (en)](https://huggingface.co/datasets/olm/olm-wikipedia-20221220)
+  - [Wikipedia (zh)](https://huggingface.co/datasets/pleisto/wikipedia-cn-20230720-filtered)
+- 用于指令监督微调:
+  - [Stanford Alpaca (en)](https://github.com/tatsu-lab/stanford_alpaca)
+  - [Stanford Alpaca (zh)](https://github.com/ymcui/Chinese-LLaMA-Alpaca)
+  - [GPT-4 Generated Data (en&zh)](https://github.com/Instruction-Tuning-with-GPT-4/GPT-4-LLM)
+  - [Open Assistant (multilingual)](https://huggingface.co/datasets/OpenAssistant/oasst1)
+  - [Self-cognition (zh)](data/self_cognition.json)
+  - [ShareGPT (zh)](https://huggingface.co/datasets/QingyiSi/Alpaca-CoT/tree/main/Chinese-instruction-collection)
+  - [RefGPT (zh)](https://github.com/sufengniu/RefGPT)
+  - [Guanaco Dataset (multilingual)](https://huggingface.co/datasets/JosephusCheung/GuanacoDataset)
+  - [BELLE 2M (zh)](https://huggingface.co/datasets/BelleGroup/train_2M_CN)
+  - [BELLE 1M (zh)](https://huggingface.co/datasets/BelleGroup/train_1M_CN)
+  - [BELLE 0.5M (zh)](https://huggingface.co/datasets/BelleGroup/train_0.5M_CN)
+  - [BELLE Dialogue 0.4M (zh)](https://huggingface.co/datasets/BelleGroup/generated_chat_0.4M)
+  - [BELLE School Math 0.25M (zh)](https://huggingface.co/datasets/BelleGroup/school_math_0.25M)
+  - [BELLE Multiturn Chat 0.8M (zh)](https://huggingface.co/datasets/BelleGroup/multiturn_chat_0.8M)
+  - [Firefly 1.1M (zh)](https://huggingface.co/datasets/YeungNLP/firefly-train-1.1M)
+  - [LIMA (en)](https://huggingface.co/datasets/GAIR/lima)
+  - [CodeAlpaca 20k (en)](https://huggingface.co/datasets/sahil2801/CodeAlpaca-20k)
+  - [Alpaca CoT (multilingual)](https://huggingface.co/datasets/QingyiSi/Alpaca-CoT)
+  - [Web QA (zh)](https://huggingface.co/datasets/suolyer/webqa)
+  - [UltraChat (en)](https://github.com/thunlp/UltraChat)
+  - [WebNovel (zh)](https://huggingface.co/datasets/zxbsmk/webnovel_cn)
+- 用于奖励模型训练:
+  - [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)
+
+使用方法请参考 [data/README.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
+pip install --upgrade huggingface_hub
+huggingface-cli login
+```
+
+## 软件依赖
+
+- Python 3.8+ 和 PyTorch 1.13.1+
+- 🤗Transformers, Datasets, Accelerate, PEFT 和 TRL
+- jieba, rouge-chinese 和 nltk (用于评估)
+- gradio 和 matplotlib (用于网页端交互)
+- uvicorn, fastapi 和 sse-starlette (用于 API)
+
+以及 **强而有力的 GPU**！
+
+## 如何使用
+
+### 数据准备（可跳过）
+
+关于数据集文件的格式，请参考 `data/example_dataset` 文件夹的内容。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。
+
+注意：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.md`。
+
+### 环境搭建（可跳过）
+
+```bash
+git lfs install
+git clone https://github.com/hiyouga/LLaMA-Efficient-Tuning.git
+conda create -n llama_etuning python=3.10
+conda activate llama_etuning
+cd LLaMA-Efficient-Tuning
+pip install -r requirements.txt
+```
+
+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 12.1.
+
+```bash
+pip install https://github.com/jllllll/bitsandbytes-windows-webui/releases/download/wheels/bitsandbytes-0.39.1-py3-none-win_amd64.whl
+```
+
+### 浏览器一键微调/测试
+
+```bash
+CUDA_VISIBLE_DEVICES=0 python src/train_web.py
+```
+
+目前网页 UI 仅支持**单卡训练**。
+
+### 二次预训练
+
+```bash
+CUDA_VISIBLE_DEVICES=0 python src/train_bash.py \
+    --stage pt \
+    --model_name_or_path path_to_your_model \
+    --do_train \
+    --dataset wiki_demo \
+    --template default \
+    --finetuning_type lora \
+    --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
+
+```bash
+CUDA_VISIBLE_DEVICES=0 python src/train_bash.py \
+    --stage sft \
+    --model_name_or_path path_to_your_model \
+    --do_train \
+    --dataset alpaca_gpt4_zh \
+    --template default \
+    --finetuning_type lora \
+    --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
+```
+
+使用 Baichuan 模型时请指定 `--lora_target W_pack` 参数。
+
+### 奖励模型训练
+
+```bash
+CUDA_VISIBLE_DEVICES=0 python src/train_bash.py \
+    --stage rm \
+    --model_name_or_path path_to_your_model \
+    --do_train \
+    --dataset comparison_gpt4_zh \
+    --template default \
+    --finetuning_type lora \
+    --resume_lora_training False \
+    --checkpoint_dir path_to_sft_checkpoint \
+    --output_dir path_to_rm_checkpoint \
+    --per_device_train_batch_size 4 \
+    --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
+```
+
+### RLHF 训练
+
+```bash
+CUDA_VISIBLE_DEVICES=0 python src/train_bash.py \
+    --stage ppo \
+    --model_name_or_path path_to_your_model \
+    --do_train \
+    --dataset alpaca_gpt4_zh \
+    --template default \
+    --finetuning_type lora \
+    --resume_lora_training False \
+    --checkpoint_dir path_to_sft_checkpoint \
+    --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 \
+    --logging_steps 10 \
+    --save_steps 1000 \
+    --learning_rate 1e-5 \
+    --num_train_epochs 1.0 \
+    --plot_loss
+```
+
+### 多 GPU 分布式训练
+
+```bash
+accelerate config # 首先配置分布式环境
+accelerate launch src/train_bash.py # 参数同上
+```
+
+<details><summary>使用 DeepSpeed ZeRO-2 进行全参数微调的 Accelerate 配置示例</summary>
+
+```yaml
+compute_environment: LOCAL_MACHINE
+deepspeed_config:
+  gradient_accumulation_steps: 4
+  gradient_clipping: 0.5
+  offload_optimizer_device: none
+  offload_param_device: none
+  zero3_init_flag: false
+  zero_stage: 2
+distributed_type: DEEPSPEED
+downcast_bf16: 'no'
+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>
+
+### 指标评估（BLEU分数和汉语ROUGE分数）
+
+```bash
+CUDA_VISIBLE_DEVICES=0 python src/train_bash.py \
+    --stage sft \
+    --model_name_or_path path_to_your_model \
+    --do_eval \
+    --dataset alpaca_gpt4_zh \
+    --template default \
+    --finetuning_type lora \
+    --checkpoint_dir path_to_checkpoint \
+    --output_dir path_to_eval_result \
+    --per_device_eval_batch_size 8 \
+    --max_samples 100 \
+    --predict_with_generate
+```
+
+我们建议在量化模型的评估中使用 `--per_device_eval_batch_size=1` 和 `--max_target_length 128` 参数。
+
+### 模型预测
+
+```bash
+CUDA_VISIBLE_DEVICES=0 python src/train_bash.py \
+    --stage sft \
+    --model_name_or_path path_to_your_model \
+    --do_predict \
+    --dataset alpaca_gpt4_zh \
+    --template default \
+    --finetuning_type lora \
+    --checkpoint_dir path_to_checkpoint \
+    --output_dir path_to_predict_result \
+    --per_device_eval_batch_size 8 \
+    --max_samples 100 \
+    --predict_with_generate
+```
+
+### API 服务
+
+```bash
+python src/api_demo.py \
+    --model_name_or_path path_to_your_model \
+    --template default \
+    --finetuning_type lora \
+    --checkpoint_dir path_to_checkpoint
+```
+
+关于 API 文档请见 `http://localhost:8000/docs`。
+
+### 命令行测试
+
+```bash
+python src/cli_demo.py \
+    --model_name_or_path path_to_your_model \
+    --template default \
+    --finetuning_type lora \
+    --checkpoint_dir path_to_checkpoint
+```
+
+### 浏览器测试
+
+```bash
+python src/web_demo.py \
+    --model_name_or_path path_to_your_model \
+    --template default \
+    --finetuning_type lora \
+    --checkpoint_dir path_to_checkpoint
+```
+
+### 导出微调模型
+
+```bash
+python src/export_model.py \
+    --model_name_or_path path_to_your_model \
+    --template default \
+    --finetuning_type lora \
+    --checkpoint_dir path_to_checkpoint \
+    --output_dir path_to_export
+```
+
+## TODO
+
+- [ ] 实现 flash attention ([torch](https://pytorch.org/docs/stable/generated/torch.nn.functional.scaled_dot_product_attention.html) / [xformers](https://github.com/facebookresearch/xformers) / [flashattn](https://github.com/Dao-AILab/flash-attention))。
+- [ ] 在推理阶段使用 Multi-query attention 进行加速。
+- [ ] 支持 RLHF 的全参数微调。
+
+## 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：
+
+- [LLaMA](https://github.com/facebookresearch/llama/blob/main/MODEL_CARD.md)
+- [LLaMA-2](https://ai.meta.com/llama/license/)
+- [BLOOM](https://huggingface.co/spaces/bigscience/license)
+- [Falcon](LICENSE)
+- [Baichuan](https://huggingface.co/baichuan-inc/baichuan-7B/resolve/main/baichuan-7B%20%E6%A8%A1%E5%9E%8B%E8%AE%B8%E5%8F%AF%E5%8D%8F%E8%AE%AE.pdf)
+- [InternLM](https://github.com/InternLM/InternLM#open-source-license)
+
+## 引用
+
+如果您觉得此项目有帮助，请考虑以下列格式引用
+
+```bibtex
+@Misc{llama-efficient-tuning,
+  title = {LLaMA Efficient Tuning},
+  author = {hiyouga},
+  howpublished = {\url{https://github.com/hiyouga/LLaMA-Efficient-Tuning}},
+  year = {2023}
+}
+```
+
+## 致谢
+
+本项目是 [ChatGLM-Efficient-Tuning](https://github.com/hiyouga/ChatGLM-Efficient-Tuning) 的同类项目。采用了类似的代码结构和训练方法。
+
+## Star History
+
+![Star History Chart](https://api.star-history.com/svg?repos=hiyouga/LLaMA-Efficient-Tuning&type=Date)
--- a/data/README.md
+++ b/data/README.md
@@ -2,16 +2,16 @@ If you are using a custom dataset, please provide your dataset definition in the

 ```json
 "dataset_name": {
-    "hf_hub_url": "the name of the dataset repository on the HuggingFace hub. (if specified, ignore below 3 arguments)",
-    "script_url": "the name of the directory containing a dataset loading script. (if specified, ignore below 2 arguments)",
-    "file_name": "the name of the dataset file in the this directory. (required if above are not specified)",
-    "file_sha1": "the SHA-1 hash value of the dataset file. (optional)",
-    "columns": {
-        "prompt": "the name of the column in the datasets containing the prompts. (default: instruction)",
-        "query": "the name of the column in the datasets containing the queries. (default: input)",
-        "response": "the name of the column in the datasets containing the responses. (default: output)",
-        "history": "the name of the column in the datasets containing the history of chat. (default: None)"
-    }
+  "hf_hub_url": "the name of the dataset repository on the HuggingFace hub. (if specified, ignore below 3 arguments)",
+  "script_url": "the name of the directory containing a dataset loading script. (if specified, ignore below 2 arguments)",
+  "file_name": "the name of the dataset file in the this directory. (required if above are not specified)",
+  "file_sha1": "the SHA-1 hash value of the dataset file. (optional)",
+  "columns": {
+    "prompt": "the name of the column in the datasets containing the prompts. (default: instruction)",
+    "query": "the name of the column in the datasets containing the queries. (default: input)",
+    "response": "the name of the column in the datasets containing the responses. (default: output)",
+    "history": "the name of the column in the datasets containing the history of chat. (default: None)"
+  }
 }
 ```

--- a/data/README_zh.md
+++ b/data/README_zh.md
@@ -0,0 +1,18 @@
+如果您使用自定义数据集，请务必在 `dataset_info.json` 文件中以如下格式提供您的数据集定义。
+
+```json
+"数据集名称": {
+  "hf_hub_url": "HuggingFace上的项目地址（若指定，则忽略下列三个参数）",
+  "script_url": "包含数据加载脚本的本地文件夹名称（若指定，则忽略下列两个参数）",
+  "file_name": "该目录下数据集文件的名称（若上述参数未指定，则此项必需）",
+  "file_sha1": "数据集文件的SHA-1哈希值（可选）",
+  "columns": {
+    "prompt": "数据集代表提示词的表头名称（默认：instruction）",
+    "query": "数据集代表请求的表头名称（默认：input）",
+    "response": "数据集代表回答的表头名称（默认：output）",
+    "history": "数据集代表历史对话的表头名称（默认：None）"
+  }
+}
+```
+
+其中 `prompt` 和 `response` 列应当是非空的字符串。`query` 列的内容将会和 `prompt` 列拼接作为模型输入。`history` 列应当是一个列表，其中每个元素是一个字符串二元组，分别代表用户请求和模型答复。
--- a/data/oaast_sft.json.REMOVED.git-id
+++ b/data/oaast_sft.json.REMOVED.git-id
@@ -1 +1 @@
-0a57fbc1d8cb08a8cd71c5eb8425cf59206ffed6
+57fd080be5bffe4153fe3ee26a175e3d56da30f3
--- a/data/refgpt_zh_50k_p1.json.REMOVED.git-id
+++ b/data/refgpt_zh_50k_p1.json.REMOVED.git-id
@@ -1 +1 @@
-56405bb8f52727e52e99693739494b9b7b0d7ba6
+f967a4f6d04a11308a15524aa9a846a19a8d1e83
--- a/data/refgpt_zh_50k_p2.json.REMOVED.git-id
+++ b/data/refgpt_zh_50k_p2.json.REMOVED.git-id
@@ -1 +1 @@
-fa935248a5d40d2bdd5649af99a72a754d40ae7a
+0a4f0d74fd1c5cab2eb6d84a3a3fe669847becd8
--- a/src/llmtuner/init.py
+++ b/src/llmtuner/init.py
@@ -1,4 +1,4 @@
 from llmtuner.chat import ChatModel


-__version__ = "0.1.3"
+__version__ = "0.1.5"
--- a/src/llmtuner/chat/stream_chat.py
+++ b/src/llmtuner/chat/stream_chat.py
@@ -1,37 +1,46 @@
 import torch
-from typing import Any, Dict, Generator, List, Optional, Tuple
+from typing import TYPE_CHECKING, Any, Dict, Generator, List, Optional, Tuple
 from threading import Thread
 from transformers import TextIteratorStreamer

-from llmtuner.extras.misc import get_logits_processor
+from llmtuner.extras.misc import dispatch_model, get_logits_processor
 from llmtuner.extras.template import get_template
-from llmtuner.hparams import ModelArguments, DataArguments, FinetuningArguments, GeneratingArguments
 from llmtuner.tuner import load_model_and_tokenizer

+if TYPE_CHECKING:
+    from llmtuner.hparams import ModelArguments, DataArguments, FinetuningArguments, GeneratingArguments
+

 class ChatModel:

    def __init__(
        self,
-        model_args: ModelArguments,
-        data_args: DataArguments,
-        finetuning_args: FinetuningArguments,
-        generating_args: GeneratingArguments
+        model_args: "ModelArguments",
+        data_args: "DataArguments",
+        finetuning_args: "FinetuningArguments",
+        generating_args: "GeneratingArguments"
    ) -> None:
        self.model, self.tokenizer = load_model_and_tokenizer(model_args, finetuning_args)
-        self.template = get_template(data_args.prompt_template)
-        self.source_prefix = data_args.source_prefix or ""
+        self.model = dispatch_model(self.model)
+        self.template = get_template(data_args.template)
+        self.source_prefix = data_args.source_prefix
        self.generating_args = generating_args

    def process_args(
-        self, query: str, history: Optional[List[Tuple[str, str]]] = None, prefix: Optional[str] = None, **input_kwargs
+        self,
+        query: str,
+        history: Optional[List[Tuple[str, str]]] = None,
+        prefix: Optional[str] = None,
+        **input_kwargs
    ) -> Tuple[Dict[str, Any], int]:
        prefix = prefix or self.source_prefix

-        inputs = self.tokenizer([self.template.get_prompt(query, history, prefix)], return_tensors="pt")
+        prompt = self.template.get_prompt(query, history, prefix, self.tokenizer.eos_token)
+        inputs = self.tokenizer([prompt], return_tensors="pt")
        inputs = inputs.to(self.model.device)
        prompt_length = len(inputs["input_ids"][0])

+        do_sample = input_kwargs.pop("do_sample", None)
        temperature = input_kwargs.pop("temperature", None)
        top_p = input_kwargs.pop("top_p", None)
        top_k = input_kwargs.pop("top_k", None)
@@ -42,6 +51,7 @@ class ChatModel:
        gen_kwargs = self.generating_args.to_dict()
        gen_kwargs.update(dict(
            input_ids=inputs["input_ids"],
+            do_sample=do_sample if do_sample is not None else gen_kwargs["do_sample"],
            temperature=temperature or gen_kwargs["temperature"],
            top_p=top_p or gen_kwargs["top_p"],
            top_k=top_k or gen_kwargs["top_k"],
@@ -61,7 +71,11 @@ class ChatModel:

    @torch.inference_mode()
    def chat(
-        self, query: str, history: Optional[List[Tuple[str, str]]] = None, prefix: Optional[str] = None, **input_kwargs
+        self,
+        query: str,
+        history: Optional[List[Tuple[str, str]]] = None,
+        prefix: Optional[str] = None,
+        **input_kwargs
    ) -> Tuple[str, Tuple[int, int]]:
        gen_kwargs, prompt_length = self.process_args(query, history, prefix, **input_kwargs)
        generation_output = self.model.generate(**gen_kwargs)
@@ -72,7 +86,11 @@ class ChatModel:

    @torch.inference_mode()
    def stream_chat(
-        self, query: str, history: Optional[List[Tuple[str, str]]] = None, prefix: Optional[str] = None, **input_kwargs
+        self,
+        query: str,
+        history: Optional[List[Tuple[str, str]]] = None,
+        prefix: Optional[str] = None,
+        **input_kwargs
    ) -> Generator[str, None, None]:
        gen_kwargs, _ = self.process_args(query, history, prefix, **input_kwargs)
        streamer = TextIteratorStreamer(self.tokenizer, timeout=60.0, skip_prompt=True, skip_special_tokens=True)
--- a/src/llmtuner/dsets/loader.py
+++ b/src/llmtuner/dsets/loader.py
@@ -1,40 +1,50 @@
 import os
 import hashlib
-from typing import List
+from typing import TYPE_CHECKING, List, Optional

-from datasets import Dataset, concatenate_datasets, load_dataset
+from datasets import Value, concatenate_datasets, interleave_datasets, load_dataset

 from llmtuner.extras.logging import get_logger
-from llmtuner.hparams import ModelArguments, DataArguments
+
+if TYPE_CHECKING:
+    from datasets import Dataset
+    from llmtuner.hparams import ModelArguments, DataArguments


 logger = get_logger(__name__)


+EXT2TYPE = {
+    "csv": "csv",
+    "json": "json",
+    "jsonl": "json",
+    "txt": "text"
+}
+
+
+def checksum(data_files: List[str], file_sha1: Optional[str] = None) -> None:
+    if file_sha1 is None:
+        logger.warning("Checksum failed: missing SHA-1 hash value in dataset_info.json.")
+        return
+
+    if len(data_files) != 1:
+        logger.warning("Checksum failed: too many files.")
+        return
+
+    with open(data_files[0], "rb") as f:
+        sha1 = hashlib.sha1(f.read()).hexdigest()
+        if sha1 != file_sha1:
+            logger.warning("Checksum failed: mismatched SHA-1 hash value at {}.".format(data_files[0]))
+
+
 def get_dataset(
-    model_args: ModelArguments,
-    data_args: DataArguments
-) -> Dataset:
-
-    def checksum(file_path, hash):
-        with open(file_path, "rb") as datafile:
-            binary_data = datafile.read()
-        sha1 = hashlib.sha1(binary_data).hexdigest()
-        if sha1 != hash:
-            logger.warning("Checksum failed for {}. It may vary depending on the platform.".format(file_path))
-
-    ext2type = {
-        "csv": "csv",
-        "json": "json",
-        "jsonl": "json",
-        "txt": "text"
-    }
-
+    model_args: "ModelArguments",
+    data_args: "DataArguments"
+) -> "Dataset":
    max_samples = data_args.max_samples
-    all_datasets: List[Dataset] = [] # support multiple datasets
+    all_datasets: List["Dataset"] = [] # support multiple datasets

    for dataset_attr in data_args.dataset_list:
-
        logger.info("Loading dataset {}...".format(dataset_attr))

        if dataset_attr.load_from == "hf_hub":
@@ -47,60 +57,60 @@ def get_dataset(
            data_path = None
            data_files: List[str] = []

-            if os.path.isdir(os.path.join(data_args.dataset_dir, dataset_attr.dataset_name)):
+            if os.path.isdir(os.path.join(data_args.dataset_dir, dataset_attr.dataset_name)): # directory
                for file_name in os.listdir(os.path.join(data_args.dataset_dir, dataset_attr.dataset_name)):
                    data_files.append(os.path.join(data_args.dataset_dir, dataset_attr.dataset_name, file_name))
-
                    if data_path is None:
-                        data_path = ext2type.get(data_files[0].split(".")[-1], None)
+                        data_path = EXT2TYPE.get(file_name.split(".")[-1], None)
                    else:
-                        assert data_path == ext2type.get(data_files[-1].split(".")[-1], None), "file type does not match."
-            elif os.path.isfile(os.path.join(data_args.dataset_dir, dataset_attr.dataset_name)):
+                        assert data_path == EXT2TYPE.get(file_name.split(".")[-1], None), "file type does not match."
+            elif os.path.isfile(os.path.join(data_args.dataset_dir, dataset_attr.dataset_name)): # single file
                data_files.append(os.path.join(data_args.dataset_dir, dataset_attr.dataset_name))
-                data_path = ext2type.get(data_files[0].split(".")[-1], None)
+                data_path = EXT2TYPE.get(dataset_attr.dataset_name.split(".")[-1], None)
            else:
                raise ValueError("File not found.")

            assert data_path, "File extension must be txt, csv, json or jsonl."
-
-            if len(data_files) == 1 and dataset_attr.dataset_sha1 is not None:
-                checksum(data_files[0], dataset_attr.dataset_sha1)
-            else:
-                logger.warning("Checksum failed: missing SHA-1 hash value in dataset_info.json or too many files.")
+            checksum(data_files, dataset_attr.dataset_sha1)
        else:
            raise NotImplementedError

-        raw_datasets = load_dataset(
+        dataset = load_dataset(
            data_path,
            data_files=data_files,
+            split=data_args.split,
            cache_dir=model_args.cache_dir,
+            streaming=data_args.streaming,
            use_auth_token=True if model_args.use_auth_token else None
        )
-        dataset = raw_datasets[data_args.split]

        if max_samples is not None:
            max_samples_temp = min(len(dataset), max_samples)
            dataset = dataset.select(range(max_samples_temp))

-        dummy_data = [None] * len(dataset)
-        prefix_data = [dataset_attr.source_prefix] * len(dataset)
-        for column_name, target_name in [
-            ("prompt_column", "prompt"),
-            ("query_column", "query"),
-            ("response_column", "response"),
-            ("history_column", "history")
-        ]: # every dataset will have 4 columns same as each other
-            if getattr(dataset_attr, column_name) != target_name:
-                if getattr(dataset_attr, column_name):
-                    dataset = dataset.rename_column(getattr(dataset_attr, column_name), target_name)
-                else: # None or empty string
-                    dataset = dataset.add_column(target_name, dummy_data)
-        dataset = dataset.add_column("prefix", prefix_data)
+        for column_name in ["prompt", "query", "response", "history"]: # align datasets
+            if getattr(dataset_attr, column_name) and getattr(dataset_attr, column_name) != column_name:
+                dataset = dataset.rename_column(getattr(dataset_attr, column_name), column_name)
+
+        if dataset_attr.source_prefix: # add prefix
+            features = None
+            if data_args.streaming:
+                features = dataset.features
+                features["prefix"] = Value(dtype="string", id=None)
+            dataset = dataset.map(lambda _: {"prefix": dataset_attr.source_prefix}, features=features)
+
        all_datasets.append(dataset)

    if len(data_args.dataset_list) == 1:
-        all_datasets = all_datasets[0]
+        return all_datasets[0]
+    elif data_args.mix_strategy == "concat":
+        if data_args.streaming:
+            logger.warning("The samples between different datasets will not be mixed in streaming mode.")
+        return concatenate_datasets(all_datasets)
+    elif data_args.mix_strategy.startswith("interleave"):
+        if not data_args.streaming:
+            logger.warning("We recommend using `mix_strategy=concat` in non-streaming mode.")
+        stopping_strategy = "first_exhausted" if data_args.mix_strategy.endswith("under") else "all_exhausted"
+        return interleave_datasets(all_datasets, stopping_strategy=stopping_strategy)
    else:
-        all_datasets = concatenate_datasets(all_datasets)
-
-    return all_datasets
+        raise ValueError("Unknown mixing strategy.")
--- a/src/llmtuner/dsets/preprocess.py
+++ b/src/llmtuner/dsets/preprocess.py
@@ -1,65 +1,63 @@
-from typing import Literal
+from typing import TYPE_CHECKING, Any, Dict, Generator, List, Literal
 from itertools import chain
-from transformers import Seq2SeqTrainingArguments
-from transformers.tokenization_utils import PreTrainedTokenizer
-
-from datasets import Dataset

 from llmtuner.extras.constants import IGNORE_INDEX
 from llmtuner.extras.template import get_template
-from llmtuner.hparams import DataArguments
+
+if TYPE_CHECKING:
+    from datasets import Dataset
+    from transformers import Seq2SeqTrainingArguments
+    from transformers.tokenization_utils import PreTrainedTokenizer
+    from llmtuner.hparams import DataArguments


 def preprocess_dataset(
-    dataset: Dataset,
-    tokenizer: PreTrainedTokenizer,
-    data_args: DataArguments,
-    training_args: Seq2SeqTrainingArguments,
+    dataset: "Dataset",
+    tokenizer: "PreTrainedTokenizer",
+    data_args: "DataArguments",
+    training_args: "Seq2SeqTrainingArguments",
    stage: Literal["pt", "sft", "rm", "ppo"]
-) -> Dataset:
-
+) -> "Dataset":
    column_names = list(dataset.column_names)
-    prompt_template = get_template(data_args.prompt_template)
+    template = get_template(data_args.template)

-    # support question with a single answer or multiple answers
-    def get_dialog(examples):
+    def construct_example(examples: Dict[str, List[Any]]) -> Generator[Any, None, None]:
        for i in range(len(examples["prompt"])):
-            if examples["prompt"][i] and examples["response"][i]:
-                query, answer = examples["prompt"][i], examples["response"][i]
-                query = query + "\n" + examples["query"][i] if examples["query"][i] else query
-                prefix = examples["prefix"][i] if examples["prefix"][i] else ""
-                dialog = prompt_template.get_dialog(query, answer, examples["history"][i], prefix)
-                yield dialog
+            query, response = examples["prompt"][i], examples["response"][i]
+            query = query + "\n" + examples["query"][i] if "query" in examples and examples["query"][i] else query
+            history = examples["history"][i] if "history" in examples else None
+            prefix = examples["prefix"][i] if "prefix" in examples else None
+            yield query, response, history, prefix

-    def preprocess_pretrain_dataset(examples):
+    def preprocess_pretrain_dataset(examples: Dict[str, List[Any]]) -> Dict[str, Any]:
        # build grouped texts with format `<bos> X1 X2 X3 ...` (without <eos>)
-        text_ids = tokenizer(examples["prompt"], add_special_tokens=False)["input_ids"]
-        concatenated_ids = list(chain(*text_ids))
-        total_length = len(concatenated_ids)
-        block_size = data_args.max_source_length - 1
+        tokenized_examples = tokenizer(examples["prompt"], add_special_tokens=False)
+        concatenated_examples = {k: list(chain(*tokenized_examples[k])) for k in tokenized_examples.keys()}
+        total_length = len(concatenated_examples[list(concatenated_examples.keys())[0]])
+        block_size = data_args.max_source_length
        # we drop the small remainder, and if the total_length < block_size, we exclude this batch
        total_length = (total_length // block_size) * block_size
        # split by chunks of max_source_length
-        result = [[tokenizer.bos_token_id] + concatenated_ids[i: i + block_size]
-                  for i in range(0, total_length, block_size)]
-        return {
-            "input_ids": result,
-            "labels": result.copy()
+        result = {
+            k: [t[i: i + block_size] for i in range(0, total_length, block_size)]
+            for k, t in concatenated_examples.items()
        }
+        result["labels"] = result["input_ids"].copy()
+        return result

-    def preprocess_supervised_dataset(examples):
+    def preprocess_supervised_dataset(examples: Dict[str, List[Any]]) -> Dict[str, Any]:
        # build inputs with format `<bos> X Y <eos>` and labels with format `<ignore> ... <ignore> Y <eos>`
        # for input with history, we build multiple input-label pairs just like:
        # https://github.com/lm-sys/FastChat/blob/f17c092f64840fa6354ed52789dccb2daa793d0b/fastchat/train/train.py#L112
-        model_inputs = {"input_ids": [], "labels": []}
+        model_inputs = {"input_ids": [], "attention_mask": [], "labels": []}
        max_length = data_args.max_source_length + data_args.max_target_length

-        for dialog in get_dialog(examples):
+        for query, response, history, prefix in construct_example(examples):
            input_ids, labels = [], []

-            for i in range(len(dialog) // 2):
-                source_ids = tokenizer.encode(text=dialog[2*i], add_special_tokens=(i == 0))
-                target_ids = tokenizer.encode(text=dialog[2*i+1], add_special_tokens=False)
+            for i, (query_i, resp_i) in enumerate(template.get_dialog(query, response, history, prefix)):
+                source_ids = tokenizer.encode(text=query_i, add_special_tokens=(i == 0))
+                target_ids = tokenizer.encode(text=resp_i, add_special_tokens=False)

                if len(source_ids) > data_args.max_source_length:
                    source_ids = source_ids[:data_args.max_source_length]
@@ -73,19 +71,20 @@ def preprocess_dataset(
                labels += [IGNORE_INDEX] * len(source_ids) + target_ids + [tokenizer.eos_token_id]

            model_inputs["input_ids"].append(input_ids)
+            model_inputs["attention_mask"].append([1] * len(input_ids))
            model_inputs["labels"].append(labels)

        return model_inputs

-    def preprocess_unsupervised_dataset(examples):
+    def preprocess_unsupervised_dataset(examples: Dict[str, List[Any]]) -> Dict[str, Any]:
        # build inputs with format `<bos> X` and labels with format `<bos> Y`
-        model_inputs = {"input_ids": [], "labels": []}
+        model_inputs = {"input_ids": [], "attention_mask": [], "labels": []}

-        for dialog in get_dialog(examples):
-            prompt, answer = "".join(dialog[:-1]), dialog[-1]
+        for query, response, history, prefix in construct_example(examples):
+            prompt = template.get_prompt(query, history, prefix, tokenizer.eos_token)

            source_ids = tokenizer.encode(text=prompt, add_special_tokens=True)
-            target_ids = tokenizer.encode(text=answer, add_special_tokens=True)
+            target_ids = tokenizer.encode(text=response, add_special_tokens=True)

            if len(source_ids) > data_args.max_source_length:
                source_ids = source_ids[:data_args.max_source_length]
@@ -93,6 +92,7 @@ def preprocess_dataset(
                target_ids = target_ids[:data_args.max_target_length]

            model_inputs["input_ids"].append(source_ids)
+            model_inputs["attention_mask"].append([1] * len(source_ids))
            model_inputs["labels"].append(target_ids)

        return model_inputs
@@ -100,12 +100,12 @@ def preprocess_dataset(
    def preprocess_pairwise_dataset(examples):
        # build input pairs with format `<bos> X Y1 <eos>` and `<bos> X Y2 <eos>`
        model_inputs = {"accept_ids": [], "reject_ids": []}
-        for dialog in get_dialog(examples):
-            prompt, answer = "".join(dialog[:-1]), dialog[-1]
+        for query, response, history, prefix in construct_example(examples):
+            prompt = template.get_prompt(query, history, prefix, tokenizer.eos_token)

            source_ids = tokenizer.encode(text=prompt, add_special_tokens=True)
-            accept_ids = tokenizer.encode(text=answer[0], add_special_tokens=False)
-            reject_ids = tokenizer.encode(text=answer[1], add_special_tokens=False)
+            accept_ids = tokenizer.encode(text=response[0], add_special_tokens=False)
+            reject_ids = tokenizer.encode(text=response[1], add_special_tokens=False)

            if len(source_ids) > data_args.max_source_length:
                source_ids = source_ids[:data_args.max_source_length]
@@ -141,34 +141,40 @@ def preprocess_dataset(
        print("inputs:\n{}".format(tokenizer.decode(example["input_ids"], skip_special_tokens=False)))

    if stage == "pt":
+        dataset = dataset.filter(lambda example: example["prompt"])
        preprocess_function = preprocess_pretrain_dataset
-    elif stage == "sft":
-        if not training_args.predict_with_generate:
-            preprocess_function = preprocess_supervised_dataset
-        else:
-            preprocess_function = preprocess_unsupervised_dataset
+        print_function = print_unsupervised_dataset_example
+    elif stage == "sft" and not training_args.predict_with_generate:
+        dataset = dataset.filter(lambda example: example["prompt"] and example["response"])
+        preprocess_function = preprocess_supervised_dataset
+        print_function = print_supervised_dataset_example
    elif stage == "rm":
+        dataset = dataset.filter(lambda example: example["prompt"] and len(example["response"]) > 1)
        preprocess_function = preprocess_pairwise_dataset
-    elif stage == "ppo":
+        print_function = print_pairwise_dataset_example
+    else:
+        dataset = dataset.filter(lambda example: example["prompt"])
        preprocess_function = preprocess_unsupervised_dataset
+        print_function = print_unsupervised_dataset_example

    with training_args.main_process_first(desc="dataset map pre-processing"):
+        kwargs = {}
+        if not data_args.streaming:
+            kwargs = dict(
+                num_proc=data_args.preprocessing_num_workers,
+                load_from_cache_file=not data_args.overwrite_cache,
+                desc="Running tokenizer on dataset"
+            )
+
        dataset = dataset.map(
            preprocess_function,
-            batched=True,
-            num_proc=data_args.preprocessing_num_workers,
+            batched=True,            
            remove_columns=column_names,
-            load_from_cache_file=not data_args.overwrite_cache,
-            desc="Running tokenizer on dataset"
+            **kwargs
        )

-        if stage == "pt":
-            print_unsupervised_dataset_example(dataset[0])
-        elif stage == "sft":
-            print_supervised_dataset_example(dataset[0])
-        elif stage == "rm":
-            print_pairwise_dataset_example(dataset[0])
-        elif stage == "ppo":
-            print_unsupervised_dataset_example(dataset[0])
+        if data_args.streaming:
+            dataset = dataset.shuffle(buffer_size=data_args.buffer_size)

+        print_function(next(iter(dataset)))
        return dataset
--- a/src/llmtuner/dsets/utils.py
+++ b/src/llmtuner/dsets/utils.py
@@ -1,13 +1,12 @@
-from typing import Dict
-from datasets import Dataset
+from typing import TYPE_CHECKING, Dict
+
+if TYPE_CHECKING:
+    from datasets import Dataset


-def split_dataset(
-    dataset: Dataset, dev_ratio: float, do_train: bool
-) -> Dict[str, Dataset]:
-    # Split the dataset
+def split_dataset(dataset: "Dataset", dev_ratio: float, do_train: bool) -> Dict[str, "Dataset"]:
    if do_train:
-        if dev_ratio > 1e-6:
+        if dev_ratio > 1e-6: # Split the dataset
            dataset = dataset.train_test_split(test_size=dev_ratio)
            return {"train_dataset": dataset["train"], "eval_dataset": dataset["test"]}
        else:
--- a/src/llmtuner/extras/callbacks.py
+++ b/src/llmtuner/extras/callbacks.py
@@ -1,16 +1,13 @@
 import os
 import json
 import time
+from typing import TYPE_CHECKING
 from datetime import timedelta

-from transformers import (
-    TrainerCallback,
-    TrainerControl,
-    TrainerState,
-    TrainingArguments
-)
-from transformers.trainer_callback import TrainerControl, TrainerState
-from transformers.training_args import TrainingArguments
+from transformers import TrainerCallback
+
+if TYPE_CHECKING:
+    from transformers import TrainingArguments, TrainerState, TrainerControl


 class LogCallback(TrainerCallback):
@@ -20,13 +17,13 @@ class LogCallback(TrainerCallback):
        self.start_time = time.time()
        self.tracker = {}

-    def on_train_begin(self, args: TrainingArguments, state: TrainerState, control: TrainerControl, **kwargs):
+    def on_train_begin(self, args: "TrainingArguments", state: "TrainerState", control: "TrainerControl", **kwargs):
        r"""
        Event called at the beginning of training.
        """
        self.start_time = time.time()

-    def on_step_begin(self, args: TrainingArguments, state: TrainerState, control: TrainerControl, **kwargs):
+    def on_step_begin(self, args: "TrainingArguments", state: "TrainerState", control: "TrainerControl", **kwargs):
        r"""
        Event called at the beginning of a training step. If using gradient accumulation, one training step
        might take several inputs.
@@ -35,7 +32,7 @@ class LogCallback(TrainerCallback):
            control.should_epoch_stop = True
            control.should_training_stop = True

-    def on_substep_end(self, args: TrainingArguments, state: TrainerState, control: TrainerControl, **kwargs):
+    def on_substep_end(self, args: "TrainingArguments", state: "TrainerState", control: "TrainerControl", **kwargs):
        r"""
        Event called at the end of an substep during gradient accumulation.
        """
@@ -43,7 +40,7 @@ class LogCallback(TrainerCallback):
            control.should_epoch_stop = True
            control.should_training_stop = True

-    def on_log(self, args: TrainingArguments, state: TrainerState, control: TrainerControl, **kwargs) -> None:
+    def on_log(self, args: "TrainingArguments", state: "TrainerState", control: "TrainerControl", **kwargs) -> None:
        r"""
        Event called after logging the last logs.
        """
--- a/src/llmtuner/extras/logging.py
+++ b/src/llmtuner/extras/logging.py
@@ -16,8 +16,16 @@ class LoggerHandler(logging.Handler):
        self.log += "\n\n"


-def get_logger(name: str) -> logging.Logger:
+def reset_logging():
+    r"""
+    Removes basic config of root logger
+    """
+    root = logging.getLogger()
+    list(map(root.removeHandler, root.handlers))
+    list(map(root.removeFilter, root.filters))

+
+def get_logger(name: str) -> logging.Logger:
    formatter = logging.Formatter(
        fmt="%(asctime)s - %(levelname)s - %(name)s - %(message)s",
        datefmt="%m/%d/%Y %H:%M:%S"
--- a/src/llmtuner/extras/misc.py
+++ b/src/llmtuner/extras/misc.py
@@ -1,12 +1,14 @@
 import torch
-from typing import List, Optional
+from typing import TYPE_CHECKING, List, Optional, Tuple

-from transformers.modeling_utils import PreTrainedModel
 from transformers.generation.utils import LogitsProcessorList
 from transformers.generation.logits_process import LogitsProcessor

 from llmtuner.extras.constants import LAYERNORM_NAMES

+if TYPE_CHECKING:
+    from transformers.modeling_utils import PreTrainedModel
+

 class AverageMeter:
    r"""
@@ -28,7 +30,7 @@ class AverageMeter:
        self.avg = self.sum / self.count


-# Avoid runtime error in model.generate(do_sample=True).
+# Avoids runtime error in model.generate(do_sample=True).
 class InvalidScoreLogitsProcessor(LogitsProcessor):

    def __call__(self, input_ids: torch.LongTensor, scores: torch.FloatTensor) -> torch.FloatTensor:
@@ -44,29 +46,37 @@ def get_logits_processor() -> LogitsProcessorList:
    return logits_processor


-def print_trainable_params(model: torch.nn.Module) -> None:
+def count_parameters(model: torch.nn.Module) -> Tuple[int, int]:
+    r"""
+    Returns the number of trainable parameters and number of all parameters in the model.
+    """
    trainable_params, all_param = 0, 0
    for param in model.parameters():
        num_params = param.numel()
        # if using DS Zero 3 and the weights are initialized empty
        if num_params == 0 and hasattr(param, "ds_numel"):
            num_params = param.ds_numel
+
+        # Due to the design of 4bit linear layers from bitsandbytes, multiply the number of parameters by 2
+        if param.__class__.__name__ == "Params4bit":
+            num_params = num_params * 2
+
        all_param += num_params
        if param.requires_grad:
            trainable_params += num_params
-    print("trainable params: {:d} || all params: {:d} || trainable%: {:.4f}".format(
-                trainable_params, all_param, 100 * trainable_params / all_param))
+
+    return trainable_params, all_param


 # Includes: (1) cast the layernorm in fp32 (2) make output embedding layer require grads (3) upcast the lm_head to fp32
 # Inspired by: https://github.com/huggingface/peft/blob/c0209c35abbf88c63aa267800d98a8e212ed0a42/src/peft/utils/other.py#L35
 def prepare_model_for_training(
-    model: PreTrainedModel,
+    model: "PreTrainedModel",
    finetuning_type: str,
-    output_embedding_layer_name: Optional[str] = "lm_head",
+    output_layer_name: Optional[str] = "lm_head",
    use_gradient_checkpointing: Optional[bool] = True,
    layer_norm_names: Optional[List[str]] = LAYERNORM_NAMES
-) -> PreTrainedModel:
+) -> "PreTrainedModel":

    for name, param in model.named_parameters():
        if param.ndim == 1 and any(layer_norm_name in name for layer_norm_name in layer_norm_names):
@@ -83,19 +93,23 @@ def prepare_model_for_training(
        model.gradient_checkpointing_enable()
        model.config.use_cache = False # turn off when gradient checkpointing is enabled

-    if finetuning_type != "full" and hasattr(model, output_embedding_layer_name):
-        output_embedding_layer: torch.nn.Linear = getattr(model, output_embedding_layer_name)
-        input_dtype = output_embedding_layer.weight.dtype
+    if finetuning_type != "full" and hasattr(model, output_layer_name):
+        if hasattr(model, "config") and hasattr(model.config, "pretraining_tp"):
+            model.config.pretraining_tp = 1 # disable TP for LoRA (https://github.com/huggingface/peft/pull/728)
+
+        output_layer: torch.nn.Linear = getattr(model, output_layer_name)
+        input_dtype = output_layer.weight.dtype

        class CastOutputToFloat(torch.nn.Sequential):

            def forward(self, x: torch.Tensor) -> torch.Tensor:
                return super().forward(x.to(input_dtype)).to(torch.float32)

-        setattr(model, output_embedding_layer_name, CastOutputToFloat(output_embedding_layer))
+        setattr(model, output_layer_name, CastOutputToFloat(output_layer))

    return model

+
 def torch_gc() -> None:
    r"""
    Collects GPU memory.
@@ -103,3 +117,25 @@ def torch_gc() -> None:
    if torch.cuda.is_available():
        torch.cuda.empty_cache()
        torch.cuda.ipc_collect()
+
+
+def dispatch_model(model: "PreTrainedModel") -> "PreTrainedModel":
+    r"""
+    Dispatches a pre-trained model to GPUs with balanced memory.
+    Borrowed from: https://github.com/huggingface/transformers/blob/v4.31.0/src/transformers/modeling_utils.py#L2803
+    """
+    if torch.cuda.device_count() > 1:
+        from accelerate import dispatch_model
+        from accelerate.utils import infer_auto_device_map, get_balanced_memory
+
+        if model._no_split_modules is None:
+            raise ValueError("The model class needs to implement the `_no_split_modules` attribute.")
+
+        kwargs = {"dtype": model.dtype, "no_split_module_classes": model._no_split_modules}
+        max_memory = get_balanced_memory(model, **kwargs)
+        # Make sure tied weights are tied before creating the device map.
+        model.tie_weights()
+        device_map = infer_auto_device_map(model, max_memory=max_memory, **kwargs)
+        return dispatch_model(model, device_map)
+    else:
+        return model.cuda()
--- a/src/llmtuner/extras/save_and_load.py
+++ b/src/llmtuner/extras/save_and_load.py
@@ -1,6 +1,6 @@
 import os
 import torch
-from typing import Dict, Optional
+from typing import Dict

 from transformers.trainer import WEIGHTS_NAME, WEIGHTS_INDEX_NAME
 from transformers.modeling_utils import load_sharded_checkpoint
@@ -12,12 +12,12 @@ from llmtuner.extras.logging import get_logger
 logger = get_logger(__name__)


-def get_state_dict(model: torch.nn.Module, trainable_only: Optional[bool] = True) -> Dict[str, torch.Tensor]:
-    state_dict = model.state_dict()
+def get_state_dict(model: torch.nn.Module) -> Dict[str, torch.Tensor]:
+    state_dict: Dict[str, torch.Tensor] = model.state_dict()
    filtered_state_dict = {}

    for k, v in model.named_parameters():
-        if (not trainable_only) or v.requires_grad:
+        if v.requires_grad:
            filtered_state_dict[k] = state_dict[k].cpu().clone().detach()

    return filtered_state_dict
--- a/src/llmtuner/extras/template.py
+++ b/src/llmtuner/extras/template.py
@@ -11,44 +11,72 @@ class Template:
    use_history: bool

    def get_prompt(
-        self, query: str, history: Optional[List[Tuple[str, str]]] = None, prefix: Optional[str] = ""
+        self,
+        query: str,
+        history: Optional[List[Tuple[str, str]]] = None,
+        prefix: Optional[str] = "",
+        eos_token: Optional[str] = "</s>"
    ) -> str:
        r"""
        Returns a string containing prompt without response.
        """
-        return "".join(self._format_example(query, history, prefix))
+        return eos_token.join(map(lambda x: x[0] + x[1], self._format_example(query, history, prefix)))

    def get_dialog(
-        self, query: str, resp: str, history: Optional[List[Tuple[str, str]]] = None, prefix: Optional[str] = ""
-    ) -> List[str]:
+        self,
+        query: str,
+        resp: str,
+        history: Optional[List[Tuple[str, str]]] = None,
+        prefix: Optional[str] = ""
+    ) -> List[Tuple[str, str]]:
        r"""
-        Returns a list containing 2 * n elements where the 2k-th is a query and the (2k+1)-th is a response.
+        Returns a list containing prompt-response pairs.
        """
-        return self._format_example(query, history, prefix) + [resp]
+        result = self._format_example(query, history, prefix)
+        result[-1][-1] = resp
+        return result

    def _format_example(
-        self, query: str, history: Optional[List[Tuple[str, str]]] = None, prefix: Optional[str] = ""
-    ) -> List[str]:
+        self,
+        query: str,
+        history: Optional[List[Tuple[str, str]]] = None,
+        prefix: Optional[str] = ""
+    ) -> List[Tuple[str, str]]:
        prefix = prefix or self.prefix # use prefix if provided
        prefix = prefix + self.sep if prefix else "" # add separator for non-empty prefix
        history = history if (history and self.use_history) else []
-        history = history + [(query, "<dummy>")]
-        convs = []
-        for turn_idx, (user_query, bot_resp) in enumerate(history):
-            if turn_idx == 0:
-                convs.append(prefix + self.prompt.format(query=user_query))
-                convs.append(bot_resp)
-            else:
-                convs.append(self.sep + self.prompt.format(query=user_query))
-                convs.append(bot_resp)
-        return convs[:-1] # drop last
+        history = history + [(query, "")]
+        return [
+            [(self.sep if i else prefix) + self.prompt.format(query=q), r]
+            for i, (q, r) in enumerate(history)
+        ]
+
+
+@dataclass
+class Llama2Template(Template):
+
+    def _format_example(
+        self,
+        query: str,
+        history: Optional[List[Tuple[str, str]]] = None,
+        prefix: Optional[str] = ""
+    ) -> List[Tuple[str, str]]:
+        prefix = prefix or self.prefix # use prefix if provided
+        prefix = prefix if prefix.startswith("<<SYS>>") else "<<SYS>>\n{}\n<</SYS>>\n\n".format(prefix)
+        history = history if (history and self.use_history) else []
+        history = history + [(query, "")]
+        return [
+            [(self.sep if i else "") + self.prompt.format(query=(q if i else prefix + q)), r]
+            for i, (q, r) in enumerate(history)
+        ]


 templates: Dict[str, Template] = {}


 def register_template(name: str, prefix: str, prompt: str, sep: str, use_history: bool) -> None:
-    templates[name] = Template(
+    template_class = Llama2Template if name == "llama2" else Template
+    templates[name] = template_class(
        prefix=prefix,
        prompt=prompt,
        sep=sep,
@@ -102,8 +130,8 @@ register_template(
           "If a question does not make any sense, or is not factually coherent, "
           "explain why instead of answering something not correct. "
           "If you don't know the answer to a question, please don't share false information.\n<</SYS>>\n\n",
-    prompt=" [INST] {query} [/INST] ",
-    sep="</s>",
+    prompt="[INST] {query} [/INST] ",
+    sep="<s>",
    use_history=True
 )

@@ -131,7 +159,7 @@ register_template(
    prefix="A chat between a curious user and an artificial intelligence assistant. "
           "The assistant gives helpful, detailed, and polite answers to the user's questions.",
    prompt="USER: {query} ASSISTANT: ",
-    sep="</s>",
+    sep="",
    use_history=True
 )

@@ -216,7 +244,7 @@ register_template(
    name="baichuan",
    prefix="",
    prompt="<reserved_102>{query}<reserved_103>",
-    sep="</s>",
+    sep="",
    use_history=True
 )

--- a/src/llmtuner/hparams/data_args.py
+++ b/src/llmtuner/hparams/data_args.py
@@ -1,6 +1,6 @@
 import os
 import json
-from typing import List, Optional
+from typing import List, Literal, Optional
 from dataclasses import dataclass, field


@@ -16,10 +16,10 @@ class DatasetAttr:
        return self.dataset_name

    def __post_init__(self):
-        self.prompt_column = "instruction"
-        self.query_column = "input"
-        self.response_column = "output"
-        self.history_column = None
+        self.prompt = "instruction"
+        self.query = "input"
+        self.response = "output"
+        self.history = None


@dataclass
@@ -27,8 +27,11 @@ class DataArguments:
    """
    Arguments pertaining to what data we are going to input our model for training and evaluation.
    """
+    template: str = field(
+        metadata={"help": "Which template to use for constructing prompts in training and inference."}
+    )
    dataset: Optional[str] = field(
-        default="alpaca_zh",
+        default="alpaca_en",
        metadata={"help": "The name of provided dataset(s) to use. Use commas to separate multiple datasets."}
    )
    dataset_dir: Optional[str] = field(
@@ -39,6 +42,18 @@ class DataArguments:
        default="train",
        metadata={"help": "Which dataset split to use for training and evaluation."}
    )
+    streaming: Optional[bool] = field(
+        default=False,
+        metadata={"help": "Enable streaming mode."}
+    )
+    buffer_size: Optional[int] = field(
+        default=16384,
+        metadata={"help": "Size of the buffer to randomly sample examples from in streaming mode."}
+    )
+    mix_strategy: Optional[Literal["concat", "interleave_under", "interleave_over"]] = field(
+        default="concat",
+        metadata={"help": "Strategy to use in dataset mixing."}
+    )
    overwrite_cache: Optional[bool] = field(
        default=False,
        metadata={"help": "Overwrite the cached training and evaluation sets."}
@@ -75,10 +90,6 @@ class DataArguments:
        default=0,
        metadata={"help": "Proportion of the dataset to include in the development set, should be between 0.0 and 1.0."}
    )
-    prompt_template: Optional[str] = field(
-        default="default",
-        metadata={"help": "Which template to use for constructing prompts in training and inference."}
-    )

    def init_for_training(self): # support mixing multiple datasets
        dataset_names = [ds.strip() for ds in self.dataset.split(",")]
@@ -111,9 +122,9 @@ class DataArguments:
            dataset_attr.source_prefix = prefix_list[i]

            if "columns" in dataset_info[name]:
-                dataset_attr.prompt_column = dataset_info[name]["columns"].get("prompt", None)
-                dataset_attr.query_column = dataset_info[name]["columns"].get("query", None)
-                dataset_attr.response_column = dataset_info[name]["columns"].get("response", None)
-                dataset_attr.history_column = dataset_info[name]["columns"].get("history", None)
+                dataset_attr.prompt = dataset_info[name]["columns"].get("prompt", None)
+                dataset_attr.query = dataset_info[name]["columns"].get("query", None)
+                dataset_attr.response = dataset_info[name]["columns"].get("response", None)
+                dataset_attr.history = dataset_info[name]["columns"].get("history", None)

-            self.dataset_list.append(dataset_attr)
+            self.dataset_list.append(dataset_attr)
--- a/src/llmtuner/hparams/general_args.py
+++ b/src/llmtuner/hparams/general_args.py
@@ -5,7 +5,7 @@ from dataclasses import dataclass, field
@dataclass
 class GeneralArguments:
    """
-    Arguments pertaining to which techniques we are going to fine-tuning with.
+    Arguments pertaining to which stage we are going to perform.
    """
    stage: Optional[Literal["pt", "sft", "rm", "ppo"]] = field(
        default="sft",
--- a/src/llmtuner/tuner/core/adapter.py
+++ b/src/llmtuner/tuner/core/adapter.py
@@ -1,7 +1,7 @@
 import os
 import torch
+from typing import TYPE_CHECKING

-from transformers.modeling_utils import PreTrainedModel
 from peft import (
    PeftModel,
    TaskType,
@@ -12,19 +12,22 @@ from peft.utils import CONFIG_NAME, WEIGHTS_NAME

 from llmtuner.extras.logging import get_logger
 from llmtuner.extras.save_and_load import load_trainable_params
-from llmtuner.hparams import ModelArguments, FinetuningArguments
+
+if TYPE_CHECKING:
+    from transformers.modeling_utils import PreTrainedModel
+    from llmtuner.hparams import ModelArguments, FinetuningArguments


 logger = get_logger(__name__)


 def init_adapter(
-    model: PreTrainedModel,
-    model_args: ModelArguments,
-    finetuning_args: FinetuningArguments,
+    model: "PreTrainedModel",
+    model_args: "ModelArguments",
+    finetuning_args: "FinetuningArguments",
    is_trainable: bool,
    is_mergeable: bool
-) -> PreTrainedModel:
+) -> "PreTrainedModel":
    r"""
    Initializes the adapters.

--- a/src/llmtuner/tuner/core/loader.py
+++ b/src/llmtuner/tuner/core/loader.py
@@ -1,6 +1,6 @@
 import os
 import torch
-from typing import Literal, Optional, Tuple
+from typing import TYPE_CHECKING, Literal, Optional, Tuple

 from transformers import (
    AutoConfig,
@@ -15,12 +15,15 @@ from transformers.modeling_utils import PretrainedConfig, PreTrainedModel
 from transformers.tokenization_utils import PreTrainedTokenizerBase
 from trl import AutoModelForCausalLMWithValueHead

-from llmtuner.extras.logging import get_logger
-from llmtuner.extras.misc import prepare_model_for_training, print_trainable_params
+from llmtuner.extras.logging import reset_logging, get_logger
+from llmtuner.extras.misc import count_parameters, prepare_model_for_training
 from llmtuner.extras.save_and_load import load_valuehead_params
-from llmtuner.hparams import ModelArguments, FinetuningArguments
+from llmtuner.hparams import FinetuningArguments
 from llmtuner.tuner.core.adapter import init_adapter

+if TYPE_CHECKING:
+    from llmtuner.hparams import ModelArguments
+

 logger = get_logger(__name__)

@@ -33,8 +36,8 @@ require_version("trl>=0.4.7", "To fix: pip install trl>=0.4.7")


 def load_model_and_tokenizer(
-    model_args: ModelArguments,
-    finetuning_args: FinetuningArguments,
+    model_args: "ModelArguments",
+    finetuning_args: "FinetuningArguments",
    is_trainable: Optional[bool] = False,
    stage: Optional[Literal["pt", "sft", "rm", "ppo"]] = "sft"
 ) -> Tuple[PreTrainedModel, PreTrainedTokenizerBase]:
@@ -90,11 +93,13 @@ def load_model_and_tokenizer(
            )

        is_mergeable = False
-        config_kwargs["device_map"] = {"": int(os.environ.get("LOCAL_RANK", "0"))}
        logger.info("Quantizing model to {} bit.".format(model_args.quantization_bit))

-    if not is_trainable: # `device_map=auto` should be used for inference only
-        config_kwargs["device_map"] = "auto"
+    if (
+        model_args.quantization_bit is not None
+        or (os.environ.get('LOCAL_RANK') is not None and not is_deepspeed_zero3_enabled())
+    ):
+        config_kwargs["device_map"] = {"": int(os.environ.get("LOCAL_RANK", "0"))}

    if model_args.checkpoint_dir is not None and finetuning_args.finetuning_type == "full":
        model_to_load = model_args.checkpoint_dir[0]
@@ -124,6 +129,7 @@ def load_model_and_tokenizer(

    if stage == "rm" or stage == "ppo": # add value head
        model = AutoModelForCausalLMWithValueHead.from_pretrained(model)
+        reset_logging()

        if stage == "rm" and model_args.checkpoint_dir is not None: # load valuehead weights to evaluate reward model
            logger.warning("Only the last checkpoint containing valuehead will be loaded as the valuehead.")
@@ -144,6 +150,9 @@ def load_model_and_tokenizer(
        model.requires_grad_(False) # fix all model params
        model = model.half() if model_args.quantization_bit is None else model # cast from fp32 to fp16

-    print_trainable_params(model)
+    trainable_params, all_param = count_parameters(model)
+    logger.info("trainable params: {:d} || all params: {:d} || trainable%: {:.4f}".format(
+        trainable_params, all_param, 100 * trainable_params / all_param
+    ))

    return model, tokenizer
--- a/src/llmtuner/tuner/core/parser.py
+++ b/src/llmtuner/tuner/core/parser.py
@@ -19,20 +19,39 @@ from llmtuner.hparams import (
 logger = get_logger(__name__)


+def _parse_args(parser: HfArgumentParser, args: Optional[Dict[str, Any]] = None):
+    if args is not None:
+        return parser.parse_dict(args)
+    elif len(sys.argv) == 2 and sys.argv[1].endswith(".yaml"):
+        return parser.parse_yaml_file(os.path.abspath(sys.argv[1]))
+    elif len(sys.argv) == 2 and sys.argv[1].endswith(".json"):
+        return parser.parse_json_file(os.path.abspath(sys.argv[1]))
+    else:
+        return parser.parse_args_into_dataclasses()
+
+
+def parse_train_args(
+    args: Optional[Dict[str, Any]] = None
+) -> Tuple[ModelArguments, DataArguments, Seq2SeqTrainingArguments, FinetuningArguments, GeneralArguments]:
+    parser = HfArgumentParser((
+        ModelArguments, DataArguments, Seq2SeqTrainingArguments, FinetuningArguments, GeneralArguments
+    ))
+    return _parse_args(parser, args)
+
+
+def parse_infer_args(
+    args: Optional[Dict[str, Any]] = None
+) -> Tuple[ModelArguments, DataArguments, FinetuningArguments, GeneratingArguments]:
+    parser = HfArgumentParser((
+        ModelArguments, DataArguments, FinetuningArguments, GeneratingArguments
+    ))
+    return _parse_args(parser, args)
+
+
 def get_train_args(
    args: Optional[Dict[str, Any]] = None
 ) -> Tuple[ModelArguments, DataArguments, Seq2SeqTrainingArguments, FinetuningArguments, GeneralArguments]:
-
-    parser = HfArgumentParser((ModelArguments, DataArguments, Seq2SeqTrainingArguments, FinetuningArguments, GeneralArguments))
-
-    if args is not None:
-        model_args, data_args, training_args, finetuning_args, general_args = parser.parse_dict(args)
-    elif len(sys.argv) == 2 and sys.argv[1].endswith(".yaml"):
-        model_args, data_args, training_args, finetuning_args, general_args = parser.parse_yaml_file(os.path.abspath(sys.argv[1]))
-    elif len(sys.argv) == 2 and sys.argv[1].endswith(".json"):
-        model_args, data_args, training_args, finetuning_args, general_args = parser.parse_json_file(os.path.abspath(sys.argv[1]))
-    else:
-        model_args, data_args, training_args, finetuning_args, general_args = parser.parse_args_into_dataclasses()
+    model_args, data_args, training_args, finetuning_args, general_args = parse_train_args(args)

    # Setup logging
    if training_args.should_log:
@@ -60,6 +79,15 @@ def get_train_args(
    assert model_args.quantization_bit is None or finetuning_args.finetuning_type == "lora", \
        "Quantization is only compatible with the LoRA method."

+    assert not (training_args.max_steps == -1 and data_args.streaming), \
+        "Please specify `max_steps` in streaming mode."
+
+    assert training_args.evaluation_strategy == "no" or (not data_args.streaming), \
+        "Streaming mode does not support evaluation currently."
+
+    assert not (general_args.stage == "ppo" and data_args.streaming), \
+        "Streaming mode does not suppport PPO training currently."
+
    if model_args.checkpoint_dir is not None:
        if finetuning_args.finetuning_type != "lora":
            assert len(model_args.checkpoint_dir) == 1, "Only LoRA tuning accepts multiple checkpoints."
@@ -73,13 +101,22 @@ def get_train_args(
    if training_args.do_train and (not training_args.fp16):
        logger.warning("We recommend enable fp16 mixed precision training.")

-    if data_args.prompt_template == "default":
-        logger.warning("Please specify `prompt_template` if you are using other pre-trained models.")
-
-    if training_args.local_rank != -1 and training_args.ddp_find_unused_parameters is None:
-        logger.warning("`ddp_find_unused_parameters` needs to be set as False in DDP training.")
+    if (
+        training_args.local_rank != -1
+        and training_args.ddp_find_unused_parameters is None
+        and finetuning_args.finetuning_type == "lora"
+    ):
+        logger.warning("`ddp_find_unused_parameters` needs to be set as False for LoRA in DDP training.")
        training_args.ddp_find_unused_parameters = False

+    if data_args.max_samples is not None and data_args.streaming:
+        logger.warning("`max_samples` is incompatible with `streaming`. Disabling max_samples.")
+        data_args.max_samples = None
+
+    if data_args.dev_ratio > 1e-6 and data_args.streaming:
+        logger.warning("`dev_ratio` is incompatible with `streaming`. Disabling development set.")
+        data_args.dev_ratio = 0
+
    training_args.optim = "adamw_torch" if training_args.optim == "adamw_hf" else training_args.optim # suppress warning

    if model_args.quantization_bit is not None:
@@ -91,10 +128,10 @@ def get_train_args(
            model_args.compute_dtype = torch.float32

    # Log on each process the small summary:
-    logger.info(
-        f"Process rank: {training_args.local_rank}, device: {training_args.device}, n_gpu: {training_args.n_gpu}\n"
-        + f"  distributed training: {bool(training_args.local_rank != -1)}, 16-bits training: {training_args.fp16}"
-    )
+    logger.info("Process rank: {}, device: {}, n_gpu: {}\n  distributed training: {}, 16-bits training: {}".format(
+        training_args.local_rank, training_args.device, training_args.n_gpu,
+        bool(training_args.local_rank != -1), training_args.fp16
+    ))
    logger.info(f"Training/evaluation parameters {training_args}")

    # Set seed before initializing model.
@@ -106,17 +143,7 @@ def get_train_args(
 def get_infer_args(
    args: Optional[Dict[str, Any]] = None
 ) -> Tuple[ModelArguments, DataArguments, FinetuningArguments, GeneratingArguments]:
-
-    parser = HfArgumentParser((ModelArguments, DataArguments, FinetuningArguments, GeneratingArguments))
-
-    if args is not None:
-        model_args, data_args, finetuning_args, generating_args = parser.parse_dict(args)
-    elif len(sys.argv) == 2 and sys.argv[1].endswith(".yaml"):
-        model_args, data_args, finetuning_args, generating_args = parser.parse_yaml_file(os.path.abspath(sys.argv[1]))
-    elif len(sys.argv) == 2 and sys.argv[1].endswith(".json"):
-        model_args, data_args, finetuning_args, generating_args = parser.parse_json_file(os.path.abspath(sys.argv[1]))
-    else:
-        model_args, data_args, finetuning_args, generating_args = parser.parse_args_into_dataclasses()
+    model_args, data_args, finetuning_args, generating_args = parse_infer_args(args)

    assert model_args.quantization_bit is None or finetuning_args.finetuning_type == "lora", \
        "Quantization is only compatible with the LoRA method."
@@ -128,7 +155,4 @@ def get_infer_args(
            assert model_args.quantization_bit is None or len(model_args.checkpoint_dir) == 1, \
                "Quantized model only accepts a single checkpoint."

-    if data_args.prompt_template == "default":
-        logger.warning("Please specify `prompt_template` if you are using other pre-trained models.")
-
    return model_args, data_args, finetuning_args, generating_args
--- a/src/llmtuner/tuner/core/trainer.py
+++ b/src/llmtuner/tuner/core/trainer.py
@@ -1,16 +1,19 @@
 import os
 import torch
-from typing import Dict, Optional
+from typing import TYPE_CHECKING, Dict, Optional

 from transformers import Seq2SeqTrainer
-from transformers.trainer import TRAINING_ARGS_NAME
+from transformers.trainer import TRAINING_ARGS_NAME, WEIGHTS_NAME
 from transformers.modeling_utils import PreTrainedModel, unwrap_model
 from peft import PeftModel
+from trl import PreTrainedModelWrapper

 from llmtuner.extras.constants import FINETUNING_ARGS_NAME, VALUE_HEAD_FILE_NAME
 from llmtuner.extras.logging import get_logger
-from llmtuner.extras.save_and_load import get_state_dict, load_trainable_params, load_valuehead_params
-from llmtuner.hparams import FinetuningArguments
+from llmtuner.extras.save_and_load import get_state_dict, load_trainable_params
+
+if TYPE_CHECKING:
+    from llmtuner.hparams import FinetuningArguments


 logger = get_logger(__name__)
@@ -21,7 +24,7 @@ class PeftTrainer(Seq2SeqTrainer):
    Inherits Seq2SeqTrainer to support parameter-efficient checkpoints.
    """

-    def __init__(self, finetuning_args: FinetuningArguments, **kwargs):
+    def __init__(self, finetuning_args: "FinetuningArguments", **kwargs):
        super().__init__(**kwargs)
        self.finetuning_args = finetuning_args
        self._remove_log()
@@ -42,31 +45,34 @@ class PeftTrainer(Seq2SeqTrainer):
        output_dir = output_dir if output_dir is not None else self.args.output_dir
        os.makedirs(output_dir, exist_ok=True)
        logger.info(f"Saving model checkpoint to {output_dir}")
+
        model = unwrap_model(self.model)

-        if hasattr(model, "pretrained_model"): # for models with valuehead (currently using LoRA only)
-            backbone_model = getattr(model, "pretrained_model")
-            torch.save(get_state_dict(getattr(model, "v_head")), os.path.join(output_dir, VALUE_HEAD_FILE_NAME))
-        else:
-            backbone_model = model
+        if isinstance(model, PreTrainedModelWrapper):
+            # Custom state dict: https://github.com/lvwerra/trl/blob/v0.4.7/trl/models/modeling_value_head.py#L200
+            model_state_dict = state_dict or model.state_dict()
+            v_head_state_dict = {
+                name.replace("v_head.", ""): model_state_dict[name].cpu().clone().detach()
+                for name in model_state_dict.keys() if name.startswith("v_head.")
+            }

-        if isinstance(backbone_model, PeftModel): # LoRA tuning
-            backbone_model.save_pretrained(output_dir, state_dict=get_state_dict(backbone_model))
-        elif isinstance(backbone_model, PreTrainedModel): # freeze/full tuning
-            backbone_model.config.use_cache = True
-            backbone_model.save_pretrained(
-                output_dir,
-                state_dict=get_state_dict(backbone_model, trainable_only=(self.finetuning_args.finetuning_type != "full")),
-                safe_serialization=self.args.save_safetensors
-            )
-            backbone_model.config.use_cache = False
-            if self.tokenizer is not None:
-                self.tokenizer.save_pretrained(output_dir)
+            torch.save(v_head_state_dict, os.path.join(output_dir, VALUE_HEAD_FILE_NAME))
+            model = model.pretrained_model
+
+        state_dict = state_dict or get_state_dict(model)
+        if isinstance(model, (PeftModel, PreTrainedModel)):
+            model.config.use_cache = True
+            model.save_pretrained(output_dir, state_dict=state_dict, safe_serialization=self.args.save_safetensors)
+            model.config.use_cache = False
        else:
-            logger.warning("No model to save.")
+            torch.save(state_dict, os.path.join(output_dir, WEIGHTS_NAME))
+
+        if self.finetuning_args.finetuning_type == "full" and self.tokenizer is not None:
+            self.tokenizer.save_pretrained(output_dir)

        with open(os.path.join(output_dir, TRAINING_ARGS_NAME), "w", encoding="utf-8") as f:
            f.write(self.args.to_json_string() + "\n")
+
        self.finetuning_args.save_to_json(os.path.join(output_dir, FINETUNING_ARGS_NAME))

    def _load_best_model(self):
@@ -76,16 +82,15 @@ class PeftTrainer(Seq2SeqTrainer):
        Subclass and override to inject custom behavior. It should not be directly used by external scripts.
        """
        logger.info(f"Loading best model from {self.state.best_model_checkpoint} (score: {self.state.best_metric}).")
-
        model = unwrap_model(self.model)
-        backbone_model = getattr(model, "pretrained_model") if hasattr(model, "pretrained_model") else model

-        if isinstance(backbone_model, PeftModel):
-            backbone_model.load_adapter(self.state.best_model_checkpoint, backbone_model.active_adapter)
-            if hasattr(model, "v_head") and load_valuehead_params(model, self.state.best_model_checkpoint):
-                model.v_head.load_state_dict({
-                    "summary.weight": getattr(model, "reward_head_weight"),
-                    "summary.bias": getattr(model, "reward_head_bias")
-                })
+        if isinstance(model, PreTrainedModelWrapper):
+            model.v_head.load_state_dict(torch.load(
+                os.path.join(self.state.best_model_checkpoint, VALUE_HEAD_FILE_NAME), map_location="cpu"
+            ))
+            model = model.pretrained_model
+
+        if isinstance(model, PeftModel):
+            model.load_adapter(self.state.best_model_checkpoint, model.active_adapter)
        else: # freeze/full-tuning
-            load_trainable_params(backbone_model, self.state.best_model_checkpoint)
+            load_trainable_params(model, self.state.best_model_checkpoint)
--- a/src/llmtuner/tuner/ppo/trainer.py
+++ b/src/llmtuner/tuner/ppo/trainer.py
@@ -2,21 +2,25 @@ import os
 import math
 import torch
 from tqdm import tqdm
-from typing import Callable, Dict, List, Optional
+from typing import TYPE_CHECKING, Callable, Dict, List, Optional

-from transformers import Seq2SeqTrainingArguments, TrainerState, TrainerControl
+from transformers import TrainerState, TrainerControl
 from transformers.modeling_utils import PreTrainedModel

 from trl import PPOTrainer
 from trl.core import LengthSampler

-from llmtuner.extras.callbacks import LogCallback
 from llmtuner.extras.logging import get_logger
-from llmtuner.extras.misc import AverageMeter, get_logits_processor
-from llmtuner.hparams import FinetuningArguments
+from llmtuner.extras.misc import AverageMeter, count_parameters, get_logits_processor
+
 from llmtuner.tuner.core.trainer import PeftTrainer
 from llmtuner.tuner.ppo.utils import cast_layernorm_dtype, replace_model

+if TYPE_CHECKING:
+    from transformers import Seq2SeqTrainingArguments
+    from llmtuner.extras.callbacks import LogCallback
+    from llmtuner.hparams import FinetuningArguments
+

 logger = get_logger(__name__)

@@ -25,11 +29,12 @@ class PPOPeftTrainer(PPOTrainer, PeftTrainer):
    r"""
    Inherits PPOTrainer.
    """
+
    def __init__(
        self,
-        training_args: Seq2SeqTrainingArguments,
-        finetuning_args: FinetuningArguments,
-        callbacks: List[LogCallback],
+        training_args: "Seq2SeqTrainingArguments",
+        finetuning_args: "FinetuningArguments",
+        callbacks: List["LogCallback"],
        **kwargs
    ):
        PPOTrainer.__init__(self, **kwargs)
@@ -66,7 +71,7 @@ class PPOPeftTrainer(PPOTrainer, PeftTrainer):
            logger.info(f"  Total train batch size (w. parallel, distributed & accumulation) = {total_train_batch_size}")
            logger.info(f"  Gradient Accumulation steps = {self.args.gradient_accumulation_steps}")
            logger.info(f"  Total optimization steps = {max_steps}")
-            logger.info(f"  Number of trainable parameters = {sum(p.numel() for p in self.model.parameters() if p.requires_grad)}")
+            logger.info(f"  Number of trainable parameters = {count_parameters(self.model)[0]}")

        # Keyword arguments for `model.generate`
        gen_kwargs = {
--- a/src/llmtuner/tuner/ppo/utils.py
+++ b/src/llmtuner/tuner/ppo/utils.py
@@ -1,11 +1,13 @@
 import torch
-from typing import Dict, List, Literal, Optional, Tuple
-from trl import AutoModelForCausalLMWithValueHead
+from typing import TYPE_CHECKING, Dict, List, Literal, Optional, Tuple

 from llmtuner.extras.constants import LAYERNORM_NAMES

+if TYPE_CHECKING:
+    from trl import AutoModelForCausalLMWithValueHead

-def replace_model(model: AutoModelForCausalLMWithValueHead, target: Literal["default", "reward"]) -> None:
+
+def replace_model(model: "AutoModelForCausalLMWithValueHead", target: Literal["default", "reward"]) -> None:
    if target == "reward": # save default head temporarily
        valuehead_state_dict = model.v_head.state_dict()
        setattr(model, "default_head_weight", valuehead_state_dict["summary.weight"])
@@ -19,10 +21,10 @@ def replace_model(model: AutoModelForCausalLMWithValueHead, target: Literal["def


 def cast_layernorm_dtype(
-    model: AutoModelForCausalLMWithValueHead,
+    model: "AutoModelForCausalLMWithValueHead",
    layer_norm_names: List[str] = LAYERNORM_NAMES,
    layer_norm_params: Optional[Dict[str, torch.Tensor]] = None
-) -> Tuple[AutoModelForCausalLMWithValueHead, Dict[str, torch.Tensor]]:
+) -> Tuple["AutoModelForCausalLMWithValueHead", Dict[str, torch.Tensor]]:

    layer_norm_state_dict = {}

--- a/src/llmtuner/tuner/ppo/workflow.py
+++ b/src/llmtuner/tuner/ppo/workflow.py
@@ -2,26 +2,30 @@
 # https://github.com/lvwerra/trl/blob/main/examples/sentiment/scripts/gpt-neox-20b_peft/gpt-neo-20b_sentiment_peft.py

 import math
+from typing import TYPE_CHECKING
 from trl import PPOConfig
 from torch.optim import AdamW
 from typing import Optional, List
-from transformers import DataCollatorForSeq2Seq, Seq2SeqTrainingArguments, TrainerCallback
+from transformers import DataCollatorForSeq2Seq
 from transformers.optimization import get_scheduler

 from llmtuner.dsets import get_dataset, preprocess_dataset
 from llmtuner.extras.callbacks import LogCallback
 from llmtuner.extras.ploting import plot_loss
-from llmtuner.hparams import ModelArguments, DataArguments, FinetuningArguments
 from llmtuner.tuner.core import load_model_and_tokenizer
 from llmtuner.tuner.ppo.trainer import PPOPeftTrainer

+if TYPE_CHECKING:
+    from transformers import Seq2SeqTrainingArguments, TrainerCallback
+    from llmtuner.hparams import ModelArguments, DataArguments, FinetuningArguments
+

 def run_ppo(
-    model_args: ModelArguments,
-    data_args: DataArguments,
-    training_args: Seq2SeqTrainingArguments,
-    finetuning_args: FinetuningArguments,
-    callbacks: Optional[List[TrainerCallback]] = [LogCallback()]
+    model_args: "ModelArguments",
+    data_args: "DataArguments",
+    training_args: "Seq2SeqTrainingArguments",
+    finetuning_args: "FinetuningArguments",
+    callbacks: Optional[List["TrainerCallback"]] = [LogCallback()]
 ):
    dataset = get_dataset(model_args, data_args)
    model, tokenizer = load_model_and_tokenizer(model_args, finetuning_args, training_args.do_train, stage="ppo")
--- a/src/llmtuner/tuner/pt/workflow.py
+++ b/src/llmtuner/tuner/pt/workflow.py
@@ -1,24 +1,27 @@
 # Inspired by: https://github.com/huggingface/transformers/blob/v4.29.2/examples/pytorch/language-modeling/run_clm.py

 import math
-from typing import Optional, List
-from transformers import Seq2SeqTrainingArguments, DataCollatorForSeq2Seq, TrainerCallback
+from typing import TYPE_CHECKING, Optional, List
+from transformers import DataCollatorForSeq2Seq

 from llmtuner.dsets import get_dataset, preprocess_dataset, split_dataset
 from llmtuner.extras.callbacks import LogCallback
 from llmtuner.extras.constants import IGNORE_INDEX
 from llmtuner.extras.ploting import plot_loss
-from llmtuner.hparams import ModelArguments, DataArguments, FinetuningArguments
 from llmtuner.tuner.core import load_model_and_tokenizer
 from llmtuner.tuner.core.trainer import PeftTrainer

+if TYPE_CHECKING:
+    from transformers import Seq2SeqTrainingArguments, TrainerCallback
+    from llmtuner.hparams import ModelArguments, DataArguments, FinetuningArguments
+

 def run_pt(
-    model_args: ModelArguments,
-    data_args: DataArguments,
-    training_args: Seq2SeqTrainingArguments,
-    finetuning_args: FinetuningArguments,
-    callbacks: Optional[List[TrainerCallback]] = [LogCallback()]
+    model_args: "ModelArguments",
+    data_args: "DataArguments",
+    training_args: "Seq2SeqTrainingArguments",
+    finetuning_args: "FinetuningArguments",
+    callbacks: Optional[List["TrainerCallback"]] = [LogCallback()]
 ):
    dataset = get_dataset(model_args, data_args)
    model, tokenizer = load_model_and_tokenizer(model_args, finetuning_args, training_args.do_train, stage="pt")
--- a/src/llmtuner/tuner/rm/collator.py
+++ b/src/llmtuner/tuner/rm/collator.py
@@ -15,5 +15,8 @@ class PairwiseDataCollatorWithPadding(DataCollatorWithPadding):
        We generate 2 * n examples where the first n examples represent chosen examples and
        the last n examples represent rejected examples.
        """
-        features = [{"input_ids": feature[key]} for key in ("accept_ids", "reject_ids") for feature in features]
+        features = [
+            {"input_ids": feature[key], "attention_mask": [1] * len(feature[key])}
+            for key in ("accept_ids", "reject_ids") for feature in features
+        ]
        return super().__call__(features)
--- a/src/llmtuner/tuner/rm/trainer.py
+++ b/src/llmtuner/tuner/rm/trainer.py
@@ -1,13 +1,15 @@
 import os
 import json
 import torch
-from typing import Dict, List, Optional, Tuple, Union
-from transformers.trainer import PredictionOutput
-from transformers.modeling_utils import PreTrainedModel
+from typing import TYPE_CHECKING, Dict, List, Optional, Tuple, Union

 from llmtuner.extras.logging import get_logger
 from llmtuner.tuner.core.trainer import PeftTrainer

+if TYPE_CHECKING:
+    from transformers.trainer import PredictionOutput
+    from transformers.modeling_utils import PreTrainedModel
+

 logger = get_logger(__name__)

@@ -23,7 +25,7 @@ class PairwisePeftTrainer(PeftTrainer):

    def compute_loss(
        self,
-        model: PreTrainedModel,
+        model: "PreTrainedModel",
        inputs: Dict[str, torch.Tensor],
        return_outputs: Optional[bool] = False
    ) -> Union[torch.Tensor, Tuple[torch.Tensor, List[torch.Tensor]]]:
@@ -46,7 +48,7 @@ class PairwisePeftTrainer(PeftTrainer):

    def save_predictions(
        self,
-        predict_results: PredictionOutput
+        predict_results: "PredictionOutput"
    ) -> None:
        r"""
        Saves model predictions to `output_dir`.
--- a/src/llmtuner/tuner/rm/workflow.py
+++ b/src/llmtuner/tuner/rm/workflow.py
@@ -2,25 +2,27 @@
 # https://github.com/lvwerra/trl/blob/main/examples/summarization/scripts/reward_summarization.py
 # https://github.com/CarperAI/trlx/blob/main/examples/summarize_rlhf/reward_model/train_reward_model_gptj.py

-from typing import Optional, List
-from transformers import Seq2SeqTrainingArguments, TrainerCallback
+from typing import TYPE_CHECKING, Optional, List

 from llmtuner.dsets import get_dataset, preprocess_dataset, split_dataset
 from llmtuner.extras.callbacks import LogCallback
 from llmtuner.extras.ploting import plot_loss
-from llmtuner.hparams import ModelArguments, DataArguments, FinetuningArguments
 from llmtuner.tuner.core import load_model_and_tokenizer
 from llmtuner.tuner.rm.metric import compute_accuracy
 from llmtuner.tuner.rm.collator import PairwiseDataCollatorWithPadding
 from llmtuner.tuner.rm.trainer import PairwisePeftTrainer

+if TYPE_CHECKING:
+    from transformers import Seq2SeqTrainingArguments, TrainerCallback
+    from llmtuner.hparams import ModelArguments, DataArguments, FinetuningArguments
+

 def run_rm(
-    model_args: ModelArguments,
-    data_args: DataArguments,
-    training_args: Seq2SeqTrainingArguments,
-    finetuning_args: FinetuningArguments,
-    callbacks: Optional[List[TrainerCallback]] = [LogCallback()]
+    model_args: "ModelArguments",
+    data_args: "DataArguments",
+    training_args: "Seq2SeqTrainingArguments",
+    finetuning_args: "FinetuningArguments",
+    callbacks: Optional[List["TrainerCallback"]] = [LogCallback()]
 ):
    dataset = get_dataset(model_args, data_args)
    model, tokenizer = load_model_and_tokenizer(model_args, finetuning_args, training_args.do_train, stage="rm")
--- a/src/llmtuner/tuner/sft/metric.py
+++ b/src/llmtuner/tuner/sft/metric.py
@@ -1,7 +1,6 @@
 import numpy as np
 from dataclasses import dataclass
-from typing import Dict, Sequence, Tuple, Union
-from transformers.tokenization_utils import PreTrainedTokenizer
+from typing import TYPE_CHECKING, Dict, Sequence, Tuple, Union

 import jieba
 from rouge_chinese import Rouge
@@ -9,6 +8,9 @@ from nltk.translate.bleu_score import sentence_bleu, SmoothingFunction

 from llmtuner.extras.constants import IGNORE_INDEX

+if TYPE_CHECKING:
+    from transformers.tokenization_utils import PreTrainedTokenizer
+

@dataclass
 class ComputeMetrics:
@@ -16,7 +18,7 @@ class ComputeMetrics:
    Wraps the tokenizer into metric functions, used in Seq2SeqPeftTrainer.
    """

-    tokenizer: PreTrainedTokenizer
+    tokenizer: "PreTrainedTokenizer"

    def __call__(self, eval_preds: Sequence[Union[np.ndarray, Tuple[np.ndarray]]]) -> Dict[str, float]:
        r"""
--- a/src/llmtuner/tuner/sft/trainer.py
+++ b/src/llmtuner/tuner/sft/trainer.py
@@ -3,13 +3,15 @@ import json
 import torch
 import numpy as np
 import torch.nn as nn
-from typing import Any, Dict, List, Optional, Tuple, Union
-from transformers.trainer import PredictionOutput
+from typing import TYPE_CHECKING, Any, Dict, List, Optional, Tuple, Union

 from llmtuner.extras.constants import IGNORE_INDEX
 from llmtuner.extras.logging import get_logger
 from llmtuner.tuner.core.trainer import PeftTrainer

+if TYPE_CHECKING:
+    from transformers.trainer import PredictionOutput
+

 logger = get_logger(__name__)

@@ -81,7 +83,7 @@ class Seq2SeqPeftTrainer(PeftTrainer):

    def save_predictions(
        self,
-        predict_results: PredictionOutput
+        predict_results: "PredictionOutput"
    ) -> None:
        r"""
        Saves model predictions to `output_dir`.
--- a/src/llmtuner/tuner/sft/workflow.py
+++ b/src/llmtuner/tuner/sft/workflow.py
@@ -1,25 +1,28 @@
 # Inspired by: https://github.com/huggingface/transformers/blob/v4.29.2/examples/pytorch/summarization/run_summarization.py

-from typing import Optional, List
-from transformers import Seq2SeqTrainingArguments, DataCollatorForSeq2Seq, TrainerCallback
+from typing import TYPE_CHECKING, Optional, List
+from transformers import DataCollatorForSeq2Seq

 from llmtuner.dsets import get_dataset, preprocess_dataset, split_dataset
 from llmtuner.extras.callbacks import LogCallback
 from llmtuner.extras.constants import IGNORE_INDEX
 from llmtuner.extras.misc import get_logits_processor
 from llmtuner.extras.ploting import plot_loss
-from llmtuner.hparams import ModelArguments, DataArguments, FinetuningArguments
 from llmtuner.tuner.core import load_model_and_tokenizer
 from llmtuner.tuner.sft.metric import ComputeMetrics
 from llmtuner.tuner.sft.trainer import Seq2SeqPeftTrainer

+if TYPE_CHECKING:
+    from transformers import Seq2SeqTrainingArguments, TrainerCallback
+    from llmtuner.hparams import ModelArguments, DataArguments, FinetuningArguments
+

 def run_sft(
-    model_args: ModelArguments,
-    data_args: DataArguments,
-    training_args: Seq2SeqTrainingArguments,
-    finetuning_args: FinetuningArguments,
-    callbacks: Optional[List[TrainerCallback]] = [LogCallback()]
+    model_args: "ModelArguments",
+    data_args: "DataArguments",
+    training_args: "Seq2SeqTrainingArguments",
+    finetuning_args: "FinetuningArguments",
+    callbacks: Optional[List["TrainerCallback"]] = [LogCallback()]
 ):
    dataset = get_dataset(model_args, data_args)
    model, tokenizer = load_model_and_tokenizer(model_args, finetuning_args, training_args.do_train, stage="sft")
--- a/src/llmtuner/webui/chat.py
+++ b/src/llmtuner/webui/chat.py
@@ -54,7 +54,7 @@ class WebChatModel(ChatModel):
            checkpoint_dir=checkpoint_dir,
            finetuning_type=finetuning_type,
            quantization_bit=int(quantization_bit) if quantization_bit else None,
-            prompt_template=template,
+            template=template,
            source_prefix=source_prefix
        )
        super().__init__(*get_infer_args(args))
@@ -90,6 +90,8 @@ class WebChatModel(ChatModel):
            yield chatbot, new_history

    def postprocess(self, response: str) -> str:
-        response = response.replace("<", "&lt;")
-        response = response.replace(">", "&gt;")
-        return response
+        blocks = response.split("```")
+        for i, block in enumerate(blocks):
+            if i % 2 == 0:
+                blocks[i] = block.replace("<", "&lt;").replace(">", "&gt;")
+        return "```".join(blocks)
--- a/src/llmtuner/webui/components/chatbot.py
+++ b/src/llmtuner/webui/components/chatbot.py
@@ -1,16 +1,17 @@
-from typing import Dict, Optional, Tuple
+from typing import TYPE_CHECKING, Dict, Optional, Tuple

 import gradio as gr
-from gradio.blocks import Block
-from gradio.components import Component

-from llmtuner.webui.chat import WebChatModel
+if TYPE_CHECKING:
+    from gradio.blocks import Block
+    from gradio.components import Component
+    from llmtuner.webui.chat import WebChatModel


 def create_chat_box(
-    chat_model: WebChatModel,
+    chat_model: "WebChatModel",
    visible: Optional[bool] = False
-) -> Tuple[Block, Component, Component, Dict[str, Component]]:
+) -> Tuple["Block", "Component", "Component", Dict[str, "Component"]]:
    with gr.Box(visible=visible) as chat_box:
        chatbot = gr.Chatbot()

--- a/src/llmtuner/webui/components/data.py
+++ b/src/llmtuner/webui/components/data.py
@@ -1,10 +1,12 @@
 import gradio as gr
-from gradio.blocks import Block
-from gradio.components import Component
-from typing import Tuple
+from typing import TYPE_CHECKING, Tuple
+
+if TYPE_CHECKING:
+    from gradio.blocks import Block
+    from gradio.components import Component


-def create_preview_box() -> Tuple[Block, Component, Component, Component]:
+def create_preview_box() -> Tuple["Block", "Component", "Component", "Component"]:
    with gr.Box(visible=False, elem_classes="modal-box") as preview_box:
        with gr.Row():
            preview_count = gr.Number(interactive=False)
--- a/src/llmtuner/webui/components/eval.py
+++ b/src/llmtuner/webui/components/eval.py
@@ -1,14 +1,16 @@
-from typing import Dict
+from typing import TYPE_CHECKING, Dict
 import gradio as gr
-from gradio.components import Component

 from llmtuner.webui.common import list_dataset, DEFAULT_DATA_DIR
 from llmtuner.webui.components.data import create_preview_box
-from llmtuner.webui.runner import Runner
 from llmtuner.webui.utils import can_preview, get_preview

+if TYPE_CHECKING:
+    from gradio.components import Component
+    from llmtuner.webui.runner import Runner

-def create_eval_tab(top_elems: Dict[str, Component], runner: Runner) -> Dict[str, Component]:
+
+def create_eval_tab(top_elems: Dict[str, "Component"], runner: "Runner") -> Dict[str, "Component"]:
    with gr.Row():
        dataset_dir = gr.Textbox(value=DEFAULT_DATA_DIR, scale=2)
        dataset = gr.Dropdown(multiselect=True, scale=4)
--- a/src/llmtuner/webui/components/export.py
+++ b/src/llmtuner/webui/components/export.py
@@ -1,11 +1,13 @@
-from typing import Dict
+from typing import TYPE_CHECKING, Dict
 import gradio as gr
-from gradio.components import Component

 from llmtuner.webui.utils import export_model

+if TYPE_CHECKING:
+    from gradio.components import Component

-def create_export_tab(top_elems: Dict[str, Component]) -> Dict[str, Component]:
+
+def create_export_tab(top_elems: Dict[str, "Component"]) -> Dict[str, "Component"]:
    with gr.Row():
        save_dir = gr.Textbox()
        max_shard_size = gr.Slider(value=10, minimum=1, maximum=100)
--- a/src/llmtuner/webui/components/infer.py
+++ b/src/llmtuner/webui/components/infer.py
@@ -1,13 +1,15 @@
-from typing import Dict
+from typing import TYPE_CHECKING, Dict

 import gradio as gr
-from gradio.components import Component

 from llmtuner.webui.chat import WebChatModel
 from llmtuner.webui.components.chatbot import create_chat_box

+if TYPE_CHECKING:
+    from gradio.components import Component

-def create_infer_tab(top_elems: Dict[str, Component]) -> Dict[str, Component]:
+
+def create_infer_tab(top_elems: Dict[str, "Component"]) -> Dict[str, "Component"]:
    with gr.Row():
        load_btn = gr.Button()
        unload_btn = gr.Button()
--- a/src/llmtuner/webui/components/sft.py
+++ b/src/llmtuner/webui/components/sft.py
@@ -1,16 +1,18 @@
-from typing import Dict
+from typing import TYPE_CHECKING, Dict
 from transformers.trainer_utils import SchedulerType

 import gradio as gr
-from gradio.components import Component

 from llmtuner.webui.common import list_dataset, DEFAULT_DATA_DIR
 from llmtuner.webui.components.data import create_preview_box
-from llmtuner.webui.runner import Runner
 from llmtuner.webui.utils import can_preview, get_preview, gen_plot

+if TYPE_CHECKING:
+    from gradio.components import Component
+    from llmtuner.webui.runner import Runner

-def create_sft_tab(top_elems: Dict[str, Component], runner: Runner) -> Dict[str, Component]:
+
+def create_sft_tab(top_elems: Dict[str, "Component"], runner: "Runner") -> Dict[str, "Component"]:
    with gr.Row():
        dataset_dir = gr.Textbox(value=DEFAULT_DATA_DIR, scale=2)
        dataset = gr.Dropdown(multiselect=True, scale=4)
@@ -56,7 +58,7 @@ def create_sft_tab(top_elems: Dict[str, Component], runner: Runner) -> Dict[str,
        stop_btn = gr.Button()

    with gr.Row():
-        with gr.Column(scale=4):
+        with gr.Column(scale=3):
            output_dir = gr.Textbox()

            with gr.Box():
--- a/src/llmtuner/webui/components/top.py
+++ b/src/llmtuner/webui/components/top.py
@@ -1,15 +1,17 @@
-from typing import Dict
+from typing import TYPE_CHECKING, Dict

 import gradio as gr
-from gradio.components import Component

 from llmtuner.extras.constants import METHODS, SUPPORTED_MODELS
 from llmtuner.extras.template import templates
 from llmtuner.webui.common import list_checkpoint, get_model_path, save_config
 from llmtuner.webui.utils import can_quantize

+if TYPE_CHECKING:
+    from gradio.components import Component

-def create_top() -> Dict[str, Component]:
+
+def create_top() -> Dict[str, "Component"]:
    available_models = list(SUPPORTED_MODELS.keys()) + ["Custom"]

    with gr.Row():
--- a/src/llmtuner/webui/interface.py
+++ b/src/llmtuner/webui/interface.py
@@ -55,4 +55,4 @@ def create_ui() -> gr.Blocks:
 if __name__ == "__main__":
    demo = create_ui()
    demo.queue()
-    demo.launch(server_name="0.0.0.0", share=False, inbrowser=True)
+    demo.launch(server_name="0.0.0.0", server_port=7860, share=False, inbrowser=True)
--- a/src/llmtuner/webui/manager.py
+++ b/src/llmtuner/webui/manager.py
@@ -1,6 +1,6 @@
 import gradio as gr
-from typing import Any, Dict, List
 from gradio.components import Component
+from typing import Any, Dict, List

 from llmtuner.webui.common import get_model_path, list_dataset, load_config
 from llmtuner.webui.locales import LOCALES
@@ -24,7 +24,7 @@ class Manager:

        return refresh_dict

-    def gen_label(self, lang: str) -> Dict[Component, dict]:
+    def gen_label(self, lang: str) -> Dict[Component, Dict[str, Any]]: # cannot use TYPE_CHECKING
        update_dict = {}
        refresh_dict = self.gen_refresh()

--- a/src/llmtuner/webui/runner.py
+++ b/src/llmtuner/webui/runner.py
@@ -111,7 +111,7 @@ class Runner:
            checkpoint_dir=checkpoint_dir,
            finetuning_type=finetuning_type,
            quantization_bit=int(quantization_bit) if quantization_bit else None,
-            prompt_template=template,
+            template=template,
            source_prefix=source_prefix,
            dataset_dir=dataset_dir,
            dataset=",".join(dataset),
@@ -201,7 +201,7 @@ class Runner:
            checkpoint_dir=checkpoint_dir,
            finetuning_type=finetuning_type,
            quantization_bit=int(quantization_bit) if quantization_bit else None,
-            prompt_template=template,
+            template=template,
            source_prefix=source_prefix,
            dataset_dir=dataset_dir,
            dataset=",".join(dataset),
--- a/src/train_web.py
+++ b/src/train_web.py
@@ -4,7 +4,7 @@ from llmtuner.webui.interface import create_ui
 def main():
    demo = create_ui()
    demo.queue()
-    demo.launch(server_name="0.0.0.0", share=False, inbrowser=True)
+    demo.launch(server_name="0.0.0.0", server_port=7860, share=False, inbrowser=True)


 if __name__ == "__main__":
--- a/src/web_demo.py
+++ b/src/web_demo.py
@@ -29,7 +29,7 @@ def main():
        lang.change(manager.gen_label, [lang], [lang] + list(chat_elems.values()))

    demo.queue()
-    demo.launch(server_name="0.0.0.0", share=False, inbrowser=True)
+    demo.launch(server_name="0.0.0.0", server_port=7860, share=False, inbrowser=True)


 if __name__ == "__main__":
Author	SHA1	Message	Date
hiyouga	ba618947e7	release v0.1.5 Former-commit-id: d619e76bc4098c29a7fdc05f5a71208bd1079c9f	2023-08-02 16:10:31 +08:00
hoshi-hiyouga	f81041b502	Merge pull request #307 from GitYCC/feature/fix-llama2-prompt-template [feature] Fix template of Llama2 to match the offical template Former-commit-id: a750b1f1ed16e20233df4d2f1c20507122919f5a	2023-08-02 15:51:28 +08:00
YC Chen	f2533a2800	[fix] Remove useless code Former-commit-id: 077e1556112913e4eeef47e581055183b39d5404	2023-08-02 14:35:35 +08:00
YC Chen	bb5b4a7f26	[feature] Fix template of Llama2 to match the offical template Former-commit-id: 1a98d45aefd95eea3768fb93e5a9da257ec61181	2023-08-02 14:10:15 +08:00
hiyouga	20bff87021	fix bug in preprocessing Former-commit-id: 94952894576dfc4b42118162aec9aa35c3503c40	2023-08-02 01:10:28 +08:00
hiyouga	722b954800	update readme Former-commit-id: 5154a04869be8c47e591351565b7842339fb99e4	2023-08-01 18:48:27 +08:00
hiyouga	19256086c7	fix #296 Former-commit-id: 69e9ed9b96a7cfb3d3b43ec5ddd01aa0bfd9b784	2023-08-01 18:43:53 +08:00
hiyouga	250fecfcd4	Fix #294 Former-commit-id: 09762d9849655f5e6c71b9472d55b42489dd944b	2023-08-01 18:13:03 +08:00
hiyouga	cb4d1d5ebb	restore from git lfs Former-commit-id: 0c734a37113b773ae7c0bc8b8d1af39b15bc0fb2	2023-08-01 16:33:25 +08:00
hiyouga	d7d557fb2e	Update .gitattributes Former-commit-id: 92e68f9f30c2fc91ae1b40865bc5c2d94899ba22	2023-08-01 16:28:54 +08:00
hiyouga	0b8e19b6a6	fix webui Former-commit-id: cf4cd52d36894f53a6ec45d003f887771012e5b4	2023-08-01 12:11:37 +08:00
hiyouga	8e26eb374e	fix RM save model Former-commit-id: 8104cc2425431eb1cddccf3909855296116f922b	2023-08-01 11:56:17 +08:00
hiyouga	9bba01a033	use git lfs Former-commit-id: 4886d0071751f68c5a2d926bd9fcee0c93337322	2023-08-01 10:14:08 +08:00
hiyouga	661890b8a1	release v0.1.4 Former-commit-id: 81f84aaf2e120e39edb28ef42893939fc9a184e2	2023-08-01 10:08:47 +08:00
hiyouga	772ad4ec6b	fix inference Former-commit-id: 55dc2bdd3eaa552c655e584fc3cbbf017c7bc3e7	2023-08-01 00:06:48 +08:00
hiyouga	6f65f8cb3b	fix arg check Former-commit-id: 2c5c73de9ebc88e2d04e80754781c94a571133a0	2023-07-31 23:48:57 +08:00
hiyouga	43e83548b9	update readme Former-commit-id: d99cda254e5025ff3f968d256197ab031bfabef1	2023-07-31 23:42:32 +08:00
hiyouga	dd3f3e9749	support streaming data, fix #284 #274 #268 Former-commit-id: 819cc1353599e5fa45658bc56dd0dbe4b258b197	2023-07-31 23:33:00 +08:00
hiyouga	124f61b404	Update data_args.py Former-commit-id: 41ac5455af195747ba369c3a6dc7d412a366d54d	2023-07-28 17:42:41 +08:00
hiyouga	e8748cc6f3	update readme Former-commit-id: 14d20cd1fdcfd1f2842362f70472b666e5d48c7d	2023-07-28 17:36:00 +08:00
hiyouga	fafec8b7a5	fix #268 Former-commit-id: 1eee0207fb370bb9e234e9bd3f9a0c47d7d01bc9	2023-07-28 17:02:26 +08:00
hiyouga	030daca686	update dataset Former-commit-id: 4a044aabbd19c92a9ae93c1c30536f5086fd47f9	2023-07-26 17:05:12 +08:00
hiyouga	ac587438f8	fix #242 Former-commit-id: 80a346e29beb49e8935b786e2af1059fdc4954b2	2023-07-25 17:04:02 +08:00
hiyouga	c145bbef3c	update dataset Former-commit-id: 4fc2c3293d91d8464527ebd1ddabe572c8355616	2023-07-23 20:01:43 +08:00
hiyouga	745c46ee04	Update README_zh.md Former-commit-id: 9d3c8803a34c06a2a5512fec3f841d7efcab3e3c	2023-07-22 14:31:16 +08:00
hiyouga	a707f5b502	update readme, fix web ui postprocess Former-commit-id: ba51ab3379100108f7b52a3c2444ccdd99e8a6ef	2023-07-22 14:29:22 +08:00
hoshi-hiyouga	dc2e801077	Merge pull request #221 from mrhan1993/main 根据GLM Efficient Tuning添加中文README，web添加了server_port参数 Former-commit-id: 948f2abcb818211ee99d4c140e26044ca591369f	2023-07-22 13:04:25 +08:00
NULL	b56d5108b2	Merge branch 'hiyouga:main' into main Former-commit-id: 8244c5b554ad0823e8ebea3d5583a6ecf9a66d2d	2023-07-21 17:00:26 +08:00
mrhan1993	8e6b7034fe	根据GLM Efficient Tuning添加中文README，web添加了server_port Former-commit-id: 29e3acd23eafd891667d7a860ec544a5b05d3c33	2023-07-21 16:57:58 +08:00