[doc ]update xtuner docs (#593)

2024-01-17 12:43:26 +08:00 · 2024-01-17 12:43:26 +08:00 · 8c546c9eb9
parent 3215e83cf3
commit 8c546c9eb9
2 changed files with 188 additions and 1 deletions
--- a/finetune/README.md
+++ b/finetune/README.md
@ -1,6 +1,97 @@
 # Fine-tuning with InternLM

+English | [简体中文](./README_zh-CN.md)
+
 We recommend two projects to fine-tune InternLM.

-1. [Xtuner](): brief introduction
+1. [XTuner](https://github.com/InternLM/xtuner) is an efficient, flexible and full-featured toolkit for fine-tuning large models.
+
 2. [InternLM-Train](): brief introduction
+
+
+## XTuner
+
+### Highlights
+
+1. Support LLM, VLM pre-training / fine-tuning on almost all GPUs. XTuner is capable of fine-tuning InternLM2-7B on a single 8GB GPU, as well as multi-node fine-tuning of models exceeding 70B.
+2. Support various training algorithms ([QLoRA](http://arxiv.org/abs/2305.14314), [LoRA](http://arxiv.org/abs/2106.09685), full-parameter fune-tune), allowing users to choose the most suitable solution for their requirements.
+3. Compatible with [DeepSpeed](https://github.com/microsoft/DeepSpeed) 🚀, easily utilizing a variety of ZeRO optimization techniques.
+4. The output models can seamlessly integrate with deployment and server toolkit ([LMDeploy](https://github.com/InternLM/lmdeploy)), and large-scale evaluation toolkit ([OpenCompass](https://github.com/open-compass/opencompass), [VLMEvalKit](https://github.com/open-compass/VLMEvalKit)).
+
+### Installation
+
+- It is recommended to build a Python 3.10 virtual environment using conda
+
+  ```bash
+  conda create --name xtuner-env python=3.10 -y
+  conda activate xtuner-env
+  ```
+
+- Install XTuner with DeepSpeed integration
+
+  ```shell
+  pip install -U 'xtuner[deepspeed]'
+  ```
+
+### Fine-tune
+
+XTuner supports the efficient fine-tune (*e.g.*, QLoRA) for InternLM2.
+
+- **Step 0**, prepare the config. XTuner provides many ready-to-use configs and we can view all configs of InternLM2 by
+
+  ```shell
+  xtuner list-cfg -p internlm2
+  ```
+
+  Or, if the provided configs cannot meet the requirements, please copy the provided config to the specified directory and make specific modifications by
+
+  ```shell
+  xtuner copy-cfg ${CONFIG_NAME} ${SAVE_PATH}
+  vi ${SAVE_PATH}/${CONFIG_NAME}_copy.py
+  ```
+
+- **Step 1**, start fine-tuning.
+
+  ```shell
+  xtuner train ${CONFIG_NAME_OR_PATH}
+  ```
+
+  For example, we can start the QLoRA fine-tuning of InternLM2-Chat-7B with oasst1 dataset by
+
+  ```shell
+  # On a single GPU
+  xtuner train internlm2_chat_7b_qlora_oasst1_e3 --deepspeed deepspeed_zero2
+  # On multiple GPUs
+  (DIST) NPROC_PER_NODE=${GPU_NUM} xtuner train internlm2_chat_7b_qlora_oasst1_e3 --deepspeed deepspeed_zero2
+  (SLURM) srun ${SRUN_ARGS} xtuner train internlm2_chat_7b_qlora_oasst1_e3 --launcher slurm --deepspeed deepspeed_zero2
+  ```
+
+  - `--deepspeed` means using [DeepSpeed](https://github.com/microsoft/DeepSpeed) 🚀 to optimize the training. XTuner comes with several integrated strategies including ZeRO-1, ZeRO-2, and ZeRO-3. If you wish to disable this feature, simply remove this argument.
+
+- **Step 2**, convert the saved PTH model (if using DeepSpeed, it will be a directory) to HuggingFace model, by
+
+  ```shell
+  xtuner convert pth_to_hf ${CONFIG_NAME_OR_PATH} ${PTH} ${SAVE_PATH}
+  ```
+
+### Chat
+
+XTuner provides tools to chat with pretrained / fine-tuned large models.
+
+```shell
+xtuner chat ${NAME_OR_PATH_TO_LLM} [optional arguments]
+```
+
+For example, we can start the chat with
+
+InternLM2-Chat-7B with adapter trained from oasst1:
+
+```shell
+xtuner chat internlm/internlm2-chat-7b --adapter xtuner/internlm2-chat-7b-qlora-oasst1 --prompt-template internlm2_chat
+```
+
+LLaVA-InternLM2-7B:
+
+```shell
+xtuner chat internlm/internlm2-chat-7b --visual-encoder openai/clip-vit-large-patch14-336 --llava xtuner/llava-internlm2-7b --prompt-template internlm2_chat --image $IMAGE_PATH
+```
--- a/finetune/README_zh-CN.md
+++ b/finetune/README_zh-CN.md
@ -0,0 +1,96 @@
+# 微调 InternLM
+
+[English](./README.md) | 简体中文
+
+我们推荐以下两种框架微调 InternLM
+
+1. [XTuner](https://github.com/InternLM/xtuner) 是一个高效、灵活、全能的轻量化大模型微调工具库。
+
+2. [InternLM-Train](): brief introduction
+
+
+## XTuner
+
+### 亮点
+
+1. 支持大语言模型 LLM、多模态图文模型 VLM 的预训练及轻量级微调。XTuner 支持在 8GB 显存下微调 7B 模型，同时也支持多节点跨设备微调更大尺度模型（70B+）。
+2. 支持 [QLoRA](http://arxiv.org/abs/2305.14314)、[LoRA](http://arxiv.org/abs/2106.09685)、全量参数微调等多种微调算法，支撑用户根据具体需求作出最优选择。
+3. 兼容 [DeepSpeed](https://github.com/microsoft/DeepSpeed) 🚀，轻松应用各种 ZeRO 训练优化策略。
+4. 训练所得模型可无缝接入部署工具库 [LMDeploy](https://github.com/InternLM/lmdeploy)、大规模评测工具库 [OpenCompass](https://github.com/open-compass/opencompass) 及 [VLMEvalKit](https://github.com/open-compass/VLMEvalKit)。
+
+
+### 安装
+
+- 借助 conda 准备虚拟环境
+
+  ```bash
+  conda create --name xtuner-env python=3.10 -y
+  conda activate xtuner-env
+  ```
+
+- 安装集成 DeepSpeed 版本的 XTuner
+
+  ```shell
+  pip install -U 'xtuner[deepspeed]'
+  ```
+
+### 微调
+
+
+- **步骤 0**，准备配置文件。XTuner 提供多个开箱即用的配置文件，用户可以通过下列命令查看所有 InternLM2 的预置配置文件：
+
+  ```shell
+  xtuner list-cfg -p internlm2
+  ```
+
+  或者，如果所提供的配置文件不能满足使用需求，请导出所提供的配置文件并进行相应更改：
+
+  ```shell
+  xtuner copy-cfg ${CONFIG_NAME} ${SAVE_PATH}
+  vi ${SAVE_PATH}/${CONFIG_NAME}_copy.py
+  ```
+
+- **步骤 1**，开始微调。
+
+  ```shell
+  xtuner train ${CONFIG_NAME_OR_PATH}
+  ```
+
+  例如，我们可以利用 QLoRA 算法在 oasst1 数据集上微调 InternLM2-Chat-7B：
+
+  ```shell
+  # 单卡
+  xtuner train internlm2_chat_7b_qlora_oasst1_e3 --deepspeed deepspeed_zero2
+  # 多卡
+  (DIST) NPROC_PER_NODE=${GPU_NUM} xtuner train internlm2_chat_7b_qlora_oasst1_e3 --deepspeed deepspeed_zero2
+  (SLURM) srun ${SRUN_ARGS} xtuner train internlm2_chat_7b_qlora_oasst1_e3 --launcher slurm --deepspeed deepspeed_zero2
+  ```
+
+  - `--deepspeed` 表示使用 [DeepSpeed](https://github.com/microsoft/DeepSpeed) 🚀 来优化训练过程。XTuner 内置了多种策略，包括 ZeRO-1、ZeRO-2、ZeRO-3 等。如果用户期望关闭此功能，请直接移除此参数。
+
+- **步骤 2**，将保存的 PTH 模型（如果使用的DeepSpeed，则将会是一个文件夹）转换为 HuggingFace 模型：
+
+  ```shell
+  xtuner convert pth_to_hf ${CONFIG_NAME_OR_PATH} ${PTH} ${SAVE_PATH}
+  ```
+
+### 对话
+
+XTuner 提供与大模型对话的工具。
+
+```shell
+xtuner chat ${NAME_OR_PATH_TO_LLM} [optional arguments]
+```
+
+例如：
+
+与 InternLM2-Chat-7B, oasst1 adapter 对话：
+
+```shell
+xtuner chat internlm/internlm2-chat-7b --adapter xtuner/internlm2-chat-7b-qlora-oasst1 --prompt-template internlm2_chat
+```
+
+与 LLaVA-InternLM2-7B 对话：
+```shell
+xtuner chat internlm/internlm2-chat-7b --visual-encoder openai/clip-vit-large-patch14-336 --llava xtuner/llava-internlm2-7b --prompt-template internlm2_chat --image $IMAGE_PATH
+```