Spaces:
Running
Running
model files and libraries
Browse files- .gitmodules +9 -0
- data/README_zh-CN.md +304 -0
- data/xtuner +1 -0
- model/paraphrase-multilingual-MiniLM-L12-v2 +1 -0
- nltk_data +1 -0
- requirements.txt +1 -0
.gitmodules
ADDED
@@ -0,0 +1,9 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
[submodule "data/xtuner"]
|
2 |
+
path = data/xtuner
|
3 |
+
url = https://github.com/InternLM/xtuner.git
|
4 |
+
[submodule "nltk_data"]
|
5 |
+
path = nltk_data
|
6 |
+
url = https://gitee.com/yzy0612/nltk_data.git
|
7 |
+
[submodule "model/paraphrase-multilingual-MiniLM-L12-v2"]
|
8 |
+
path = model/paraphrase-multilingual-MiniLM-L12-v2
|
9 |
+
url = https://www.modelscope.cn/Ceceliachenen/paraphrase-multilingual-MiniLM-L12-v2.git
|
data/README_zh-CN.md
ADDED
@@ -0,0 +1,304 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
<div align="center">
|
2 |
+
<img src="https://github.com/InternLM/lmdeploy/assets/36994684/0cf8d00f-e86b-40ba-9b54-dc8f1bc6c8d8" width="600"/>
|
3 |
+
<br /><br />
|
4 |
+
|
5 |
+
[![GitHub Repo stars](https://img.shields.io/github/stars/InternLM/xtuner?style=social)](https://github.com/InternLM/xtuner/stargazers)
|
6 |
+
[![license](https://img.shields.io/github/license/InternLM/xtuner.svg)](https://github.com/InternLM/xtuner/blob/main/LICENSE)
|
7 |
+
[![PyPI](https://img.shields.io/pypi/v/xtuner)](https://pypi.org/project/xtuner/)
|
8 |
+
[![Downloads](https://static.pepy.tech/badge/xtuner)](https://pypi.org/project/xtuner/)
|
9 |
+
[![issue resolution](https://img.shields.io/github/issues-closed-raw/InternLM/xtuner)](https://github.com/InternLM/xtuner/issues)
|
10 |
+
[![open issues](https://img.shields.io/github/issues-raw/InternLM/xtuner)](https://github.com/InternLM/xtuner/issues)
|
11 |
+
|
12 |
+
👋 加入我们:[![Static Badge](https://img.shields.io/badge/-grey?style=social&logo=wechat&label=微信)](https://cdn.vansin.top/internlm/xtuner.jpg)
|
13 |
+
[![Static Badge](https://img.shields.io/badge/-grey?style=social&logo=twitter&label=推特)](https://twitter.com/intern_lm)
|
14 |
+
[![Static Badge](https://img.shields.io/badge/-grey?style=social&logo=discord&label=Discord)](https://discord.gg/xa29JuW87d)
|
15 |
+
|
16 |
+
🔍 探索我们的模型:
|
17 |
+
[![Static Badge](https://img.shields.io/badge/-gery?style=social&label=🤗%20Huggingface)](https://huggingface.co/xtuner)
|
18 |
+
[![Static Badge](https://img.shields.io/badge/-gery?style=social&label=🤖%20ModelScope)](https://www.modelscope.cn/organization/xtuner)
|
19 |
+
[![Static Badge](https://img.shields.io/badge/-gery?style=social&label=🧰%20OpenXLab)](https://openxlab.org.cn/usercenter/xtuner)
|
20 |
+
[![Static Badge](https://img.shields.io/badge/-gery?style=social&label=🧠%20WiseModel)](https://www.wisemodel.cn/organization/xtuner)
|
21 |
+
|
22 |
+
[English](README.md) | 简体中文
|
23 |
+
|
24 |
+
</div>
|
25 |
+
|
26 |
+
## 🚀 Speed Benchmark
|
27 |
+
|
28 |
+
- XTuner 与 LLaMA-Factory 在 Llama2-7B 模型上的训练效率对比
|
29 |
+
|
30 |
+
<div align=center>
|
31 |
+
<img src="https://github.com/InternLM/xtuner/assets/41630003/9c9dfdf4-1efb-4daf-84bf-7c379ae40b8b" style="width:80%">
|
32 |
+
</div>
|
33 |
+
|
34 |
+
- XTuner 与 LLaMA-Factory 在 Llama2-70B 模型上的训练效率对比
|
35 |
+
|
36 |
+
<div align=center>
|
37 |
+
<img src="https://github.com/InternLM/xtuner/assets/41630003/5ba973b8-8885-4b72-b51b-c69fa1583bdd" style="width:80%">
|
38 |
+
</div>
|
39 |
+
|
40 |
+
## 🎉 更新
|
41 |
+
- **\[2024/07\]** 支持 [MiniCPM](xtuner/configs/minicpm/) 模型!
|
42 |
+
- **\[2024/07\]** 支持训练 [DPO](https://github.com/InternLM/xtuner/tree/main/xtuner/configs/dpo), [ORPO](https://github.com/InternLM/xtuner/tree/main/xtuner/configs/orpo) 还有 [Reward Model](https://github.com/InternLM/xtuner/tree/main/xtuner/configs/reward_model) ! 并且能够支持打包数据以及序列并行功能! 请参考 [文档](https://xtuner.readthedocs.io/zh-cn/latest/dpo/overview.html) 了解更多信息。
|
43 |
+
- **\[2024/07\]** 支持 [InternLM 2.5](xtuner/configs/internlm/internlm2_5_chat_7b/) 模型!
|
44 |
+
- **\[2024/06\]** 支持 [DeepSeek V2](xtuner/configs/deepseek/deepseek_v2_chat/) models! **训练速度提升一倍!**
|
45 |
+
- **\[2024/04\]** 多模态大模型 [LLaVA-Phi-3-mini](https://huggingface.co/xtuner/llava-phi-3-mini-hf) 发布!快速开始请查阅此[文档](xtuner/configs/llava/phi3_mini_4k_instruct_clip_vit_large_p14_336)!
|
46 |
+
- **\[2024/04\]** 多模态大模型 [LLaVA-Llama-3-8B](https://huggingface.co/xtuner/llava-llama-3-8b) 和 [LLaVA-Llama-3-8B-v1.1](https://huggingface.co/xtuner/llava-llama-3-8b-v1_1) 发布!快速开始请查阅此[文档](xtuner/configs/llava/llama3_8b_instruct_clip_vit_large_p14_336)!
|
47 |
+
- **\[2024/04\]** 支持 [Llama 3](xtuner/configs/llama) 模型!
|
48 |
+
- **\[2024/04\]** 支持序列并行训练策略以实现语言模型超长上下文训练!\[[文档](https://github.com/InternLM/xtuner/blob/docs/docs/zh_cn/acceleration/train_extreme_long_sequence.rst)\] \[[速度基准](https://github.com/InternLM/xtuner/blob/docs/docs/zh_cn/acceleration/benchmark.rst)\]
|
49 |
+
- **\[2024/02\]** 支持 [Gemma](xtuner/configs/gemma) 模型!
|
50 |
+
- **\[2024/02\]** 支持 [Qwen1.5](xtuner/configs/qwen/qwen1_5) 模型!
|
51 |
+
- **\[2024/01\]** 支持 [InternLM2](xtuner/configs/internlm) 模型!同时,最新版的多模态大模型 [LLaVA-Internlm2-7B](https://huggingface.co/xtuner/llava-internlm2-7b) / [20B](https://huggingface.co/xtuner/llava-internlm2-20b) 发布,其表现出强大的性能!
|
52 |
+
- **\[2024/01\]** 支持 [DeepSeek-MoE](https://huggingface.co/deepseek-ai/deepseek-moe-16b-chat) 模型!20GB 显存即可实现 QLoRA 微调,4x80GB 即可实现全参数微调。快速开始请查阅相关[配置文件](xtuner/configs/deepseek/)!
|
53 |
+
- **\[2023/12\]** 🔥 支持多模态模型 VLM([LLaVA-v1.5](https://github.com/haotian-liu/LLaVA))预训练和指令微调!快速开始请查阅此[文档](xtuner/configs/llava/README_zh-CN.md)!
|
54 |
+
- **\[2023/12\]** 🔥 支持 [Mixtral 8x7B](https://huggingface.co/mistralai/Mixtral-8x7B-Instruct-v0.1) 模型!快速开始请查阅此[文档](xtuner/configs/mixtral/README.md)!
|
55 |
+
- **\[2023/11\]** 支持 [ChatGLM3-6B](xtuner/configs/chatglm) 模型!
|
56 |
+
- **\[2023/10\]** 支持 [MSAgent-Bench](https://modelscope.cn/datasets/damo/MSAgent-Bench) 数据集,并且微调所得大语言模型可应用至 [Lagent](https://github.com/InternLM/lagent) 框架!
|
57 |
+
- **\[2023/10\]** 优化数据处理逻辑以兼容 `system` 字段,相关细节请查阅[文档](docs/zh_cn/user_guides/dataset_format.md)!
|
58 |
+
- **\[2023/09\]** 支持 [InternLM-20B](xtuner/configs/internlm) 系列模型!
|
59 |
+
- **\[2023/09\]** 支持 [Baichuan2](xtuner/configs/baichuan) 系列模型!
|
60 |
+
- **\[2023/08\]** XTuner 正式发布!众多微调模型已上传至 [HuggingFace](https://huggingface.co/xtuner)!
|
61 |
+
|
62 |
+
## 📖 介绍
|
63 |
+
|
64 |
+
XTuner 是一个高效、灵活、全能的轻量化大模型微调工具库。
|
65 |
+
|
66 |
+
**高效**
|
67 |
+
|
68 |
+
- 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+)。
|
69 |
+
- 自动分发高性能算子(如 FlashAttention、Triton kernels 等)以加速训练吞吐。
|
70 |
+
- 兼容 [DeepSpeed](https://github.com/microsoft/DeepSpeed) 🚀,轻松应用各种 ZeRO 训练优化策略。
|
71 |
+
|
72 |
+
**灵活**
|
73 |
+
|
74 |
+
- 支持多种大语言模型,包括但不限于 [InternLM](https://huggingface.co/internlm)、[Mixtral-8x7B](https://huggingface.co/mistralai)、[Llama 2](https://huggingface.co/meta-llama)、[ChatGLM](https://huggingface.co/THUDM)、[Qwen](https://huggingface.co/Qwen)、[Baichuan](https://huggingface.co/baichuan-inc)。
|
75 |
+
- 支持多模态图文模型 LLaVA 的预训练与微调。利用 XTuner 训得模型 [LLaVA-InternLM2-20B](https://huggingface.co/xtuner/llava-internlm2-20b) 表现优异。
|
76 |
+
- 精心设计的数据管道,兼容任意数据格式,开源数据或自定义数据皆可快速上手。
|
77 |
+
- 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。
|
78 |
+
|
79 |
+
**全能**
|
80 |
+
|
81 |
+
- 支持增量预训练、指令微调与 Agent 微调。
|
82 |
+
- 预定义众多开源对话模版,支持与开源或训练所得模型进行对话。
|
83 |
+
- 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)。
|
84 |
+
|
85 |
+
## 🔥 支持列表
|
86 |
+
|
87 |
+
<table>
|
88 |
+
<tbody>
|
89 |
+
<tr align="center" valign="middle">
|
90 |
+
<td>
|
91 |
+
<b>模型</b>
|
92 |
+
</td>
|
93 |
+
<td>
|
94 |
+
<b>数据集</b>
|
95 |
+
</td>
|
96 |
+
<td>
|
97 |
+
<b>数据格式</b>
|
98 |
+
</td>
|
99 |
+
<td>
|
100 |
+
<b>微调算法</b>
|
101 |
+
</td>
|
102 |
+
</tr>
|
103 |
+
<tr valign="top">
|
104 |
+
<td align="left" valign="top">
|
105 |
+
<ul>
|
106 |
+
<li><a href="https://huggingface.co/internlm">InternLM 2 / 2.5</a></li>
|
107 |
+
<li><a href="https://huggingface.co/meta-llama">Llama 2 / 3</a></li>
|
108 |
+
<li><a href="https://huggingface.co/collections/microsoft/phi-3-6626e15e9585a200d2d761e3">Phi-3</a></li>
|
109 |
+
<li><a href="https://huggingface.co/THUDM/chatglm2-6b">ChatGLM2</a></li>
|
110 |
+
<li><a href="https://huggingface.co/THUDM/chatglm3-6b">ChatGLM3</a></li>
|
111 |
+
<li><a href="https://huggingface.co/Qwen/Qwen-7B">Qwen</a></li>
|
112 |
+
<li><a href="https://huggingface.co/baichuan-inc/Baichuan2-7B-Base">Baichuan2</a></li>
|
113 |
+
<li><a href="https://huggingface.co/mistralai/Mixtral-8x7B-Instruct-v0.1">Mixtral</a></li>
|
114 |
+
<li><a href="https://huggingface.co/deepseek-ai/DeepSeek-V2-Chat">DeepSeek V2</a></li>
|
115 |
+
<li><a href="https://huggingface.co/google">Gemma</a></li>
|
116 |
+
<li><a href="https://huggingface.co/openbmb">MiniCPM</a></li>
|
117 |
+
<li>...</li>
|
118 |
+
</ul>
|
119 |
+
</td>
|
120 |
+
<td>
|
121 |
+
<ul>
|
122 |
+
<li><a href="https://modelscope.cn/datasets/damo/MSAgent-Bench">MSAgent-Bench</a></li>
|
123 |
+
<li><a href="https://huggingface.co/datasets/fnlp/moss-003-sft-data">MOSS-003-SFT</a> 🔧</li>
|
124 |
+
<li><a href="https://huggingface.co/datasets/tatsu-lab/alpaca">Alpaca en</a> / <a href="https://huggingface.co/datasets/silk-road/alpaca-data-gpt4-chinese">zh</a></li>
|
125 |
+
<li><a href="https://huggingface.co/datasets/WizardLM/WizardLM_evol_instruct_V2_196k">WizardLM</a></li>
|
126 |
+
<li><a href="https://huggingface.co/datasets/timdettmers/openassistant-guanaco">oasst1</a></li>
|
127 |
+
<li><a href="https://huggingface.co/datasets/garage-bAInd/Open-Platypus">Open-Platypus</a></li>
|
128 |
+
<li><a href="https://huggingface.co/datasets/HuggingFaceH4/CodeAlpaca_20K">Code Alpaca</a></li>
|
129 |
+
<li><a href="https://huggingface.co/datasets/burkelibbey/colors">Colorist</a> 🎨</li>
|
130 |
+
<li><a href="https://github.com/WangRongsheng/ChatGenTitle">Arxiv GenTitle</a></li>
|
131 |
+
<li><a href="https://github.com/LiuHC0428/LAW-GPT">Chinese Law</a></li>
|
132 |
+
<li><a href="https://huggingface.co/datasets/Open-Orca/OpenOrca">OpenOrca</a></li>
|
133 |
+
<li><a href="https://huggingface.co/datasets/shibing624/medical">Medical Dialogue</a></li>
|
134 |
+
<li>...</li>
|
135 |
+
</ul>
|
136 |
+
</td>
|
137 |
+
<td>
|
138 |
+
<ul>
|
139 |
+
<li><a href="docs/zh_cn/user_guides/incremental_pretraining.md">Incremental Pre-training</a> </li>
|
140 |
+
<li><a href="docs/zh_cn/user_guides/single_turn_conversation.md">Single-turn Conversation SFT</a> </li>
|
141 |
+
<li><a href="docs/zh_cn/user_guides/multi_turn_conversation.md">Multi-turn Conversation SFT</a> </li>
|
142 |
+
</ul>
|
143 |
+
</td>
|
144 |
+
<td>
|
145 |
+
<ul>
|
146 |
+
<li><a href="http://arxiv.org/abs/2305.14314">QLoRA</a></li>
|
147 |
+
<li><a href="http://arxiv.org/abs/2106.09685">LoRA</a></li>
|
148 |
+
<li>全量参数微调</li>
|
149 |
+
<li><a href="https://arxiv.org/abs/2305.18290">DPO</a></li>
|
150 |
+
<li><a href="https://arxiv.org/abs/2403.07691">ORPO</a></li>
|
151 |
+
<li>Reward Model</a></li>
|
152 |
+
</ul>
|
153 |
+
</td>
|
154 |
+
</tr>
|
155 |
+
</tbody>
|
156 |
+
</table>
|
157 |
+
|
158 |
+
## 🛠️ 快速上手
|
159 |
+
|
160 |
+
### 安装
|
161 |
+
|
162 |
+
- 推荐使用 conda 先构建一个 Python-3.10 的虚拟环境
|
163 |
+
|
164 |
+
```bash
|
165 |
+
conda create --name xtuner-env python=3.10 -y
|
166 |
+
conda activate xtuner-env
|
167 |
+
```
|
168 |
+
|
169 |
+
- 通过 pip 安装 XTuner:
|
170 |
+
|
171 |
+
```shell
|
172 |
+
pip install -U xtuner
|
173 |
+
```
|
174 |
+
|
175 |
+
亦可集成 DeepSpeed 安装:
|
176 |
+
|
177 |
+
```shell
|
178 |
+
pip install -U 'xtuner[deepspeed]'
|
179 |
+
```
|
180 |
+
|
181 |
+
- 从源码安装 XTuner:
|
182 |
+
|
183 |
+
```shell
|
184 |
+
git clone https://github.com/InternLM/xtuner.git
|
185 |
+
cd xtuner
|
186 |
+
pip install -e '.[all]'
|
187 |
+
```
|
188 |
+
|
189 |
+
### 微调
|
190 |
+
|
191 |
+
XTuner 支持微调大语言模型。数据集预处理指南请查阅[文档](./docs/zh_cn/user_guides/dataset_prepare.md)。
|
192 |
+
|
193 |
+
- **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:
|
194 |
+
|
195 |
+
```shell
|
196 |
+
xtuner list-cfg
|
197 |
+
```
|
198 |
+
|
199 |
+
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:
|
200 |
+
|
201 |
+
```shell
|
202 |
+
xtuner copy-cfg ${CONFIG_NAME} ${SAVE_PATH}
|
203 |
+
vi ${SAVE_PATH}/${CONFIG_NAME}_copy.py
|
204 |
+
```
|
205 |
+
|
206 |
+
- **步骤 1**,开始微调。
|
207 |
+
|
208 |
+
```shell
|
209 |
+
xtuner train ${CONFIG_NAME_OR_PATH}
|
210 |
+
```
|
211 |
+
|
212 |
+
例如,我们可以利用 QLoRA 算法在 oasst1 数据集上微调 InternLM2.5-Chat-7B:
|
213 |
+
|
214 |
+
```shell
|
215 |
+
# 单卡
|
216 |
+
xtuner train internlm2_5_chat_7b_qlora_oasst1_e3 --deepspeed deepspeed_zero2
|
217 |
+
# 多卡
|
218 |
+
(DIST) NPROC_PER_NODE=${GPU_NUM} xtuner train internlm2_5_chat_7b_qlora_oasst1_e3 --deepspeed deepspeed_zero2
|
219 |
+
(SLURM) srun ${SRUN_ARGS} xtuner train internlm2_5_chat_7b_qlora_oasst1_e3 --launcher slurm --deepspeed deepspeed_zero2
|
220 |
+
```
|
221 |
+
|
222 |
+
- `--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。
|
223 |
+
|
224 |
+
- 更多示例,请查阅[文档](./docs/zh_cn/user_guides/finetune.md)。
|
225 |
+
|
226 |
+
- **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:
|
227 |
+
|
228 |
+
```shell
|
229 |
+
xtuner convert pth_to_hf ${CONFIG_NAME_OR_PATH} ${PTH} ${SAVE_PATH}
|
230 |
+
```
|
231 |
+
|
232 |
+
### 对话
|
233 |
+
|
234 |
+
XTuner 提供与大语言模型对话的工具。
|
235 |
+
|
236 |
+
```shell
|
237 |
+
xtuner chat ${NAME_OR_PATH_TO_LLM} --adapter {NAME_OR_PATH_TO_ADAPTER} [optional arguments]
|
238 |
+
```
|
239 |
+
|
240 |
+
例如:
|
241 |
+
|
242 |
+
与 InternLM2.5-Chat-7B 对话:
|
243 |
+
|
244 |
+
```shell
|
245 |
+
xtuner chat internlm/internlm2-chat-7b --prompt-template internlm2_chat
|
246 |
+
```
|
247 |
+
|
248 |
+
更多示例,请查阅[文档](./docs/zh_cn/user_guides/chat.md)。
|
249 |
+
|
250 |
+
### 部署
|
251 |
+
|
252 |
+
- **步骤 0**,将 HuggingFace adapter 合并到大语言模型:
|
253 |
+
|
254 |
+
```shell
|
255 |
+
xtuner convert merge \
|
256 |
+
${NAME_OR_PATH_TO_LLM} \
|
257 |
+
${NAME_OR_PATH_TO_ADAPTER} \
|
258 |
+
${SAVE_PATH} \
|
259 |
+
--max-shard-size 2GB
|
260 |
+
```
|
261 |
+
|
262 |
+
- **步骤 1**,使用任意推理框架部署微调后的大语言模型,例如 [LMDeploy](https://github.com/InternLM/lmdeploy) 🚀:
|
263 |
+
|
264 |
+
```shell
|
265 |
+
pip install lmdeploy
|
266 |
+
python -m lmdeploy.pytorch.chat ${NAME_OR_PATH_TO_LLM} \
|
267 |
+
--max_new_tokens 256 \
|
268 |
+
--temperture 0.8 \
|
269 |
+
--top_p 0.95 \
|
270 |
+
--seed 0
|
271 |
+
```
|
272 |
+
|
273 |
+
🔥 追求速度更快、显存占用更低的推理?欢迎体验 [LMDeploy](https://github.com/InternLM/lmdeploy) 提供的 4-bit 量化!使用指南请见[文档](https://github.com/InternLM/lmdeploy/tree/main#quantization)。
|
274 |
+
|
275 |
+
### 评测
|
276 |
+
|
277 |
+
- 推荐使用一站式平台 [OpenCompass](https://github.com/InternLM/opencompass) 来评测大语言模型,其目前已涵盖 50+ 数据集的约 30 万条题目。
|
278 |
+
|
279 |
+
## 🤝 贡献指南
|
280 |
+
|
281 |
+
我们感谢所有的贡献者为改进和提升 XTuner 所作出的努力。请参考[贡献指南](.github/CONTRIBUTING.md)来了解参与项目贡献的相关指引。
|
282 |
+
|
283 |
+
## 🎖️ 致谢
|
284 |
+
|
285 |
+
- [Llama 2](https://github.com/facebookresearch/llama)
|
286 |
+
- [DeepSpeed](https://github.com/microsoft/DeepSpeed)
|
287 |
+
- [QLoRA](https://github.com/artidoro/qlora)
|
288 |
+
- [LMDeploy](https://github.com/InternLM/lmdeploy)
|
289 |
+
- [LLaVA](https://github.com/haotian-liu/LLaVA)
|
290 |
+
|
291 |
+
## 🖊️ 引用
|
292 |
+
|
293 |
+
```bibtex
|
294 |
+
@misc{2023xtuner,
|
295 |
+
title={XTuner: A Toolkit for Efficiently Fine-tuning LLM},
|
296 |
+
author={XTuner Contributors},
|
297 |
+
howpublished = {\url{https://github.com/InternLM/xtuner}},
|
298 |
+
year={2023}
|
299 |
+
}
|
300 |
+
```
|
301 |
+
|
302 |
+
## 开源许可证
|
303 |
+
|
304 |
+
该项目采用 [Apache License 2.0 开源许可证](LICENSE)。同时,请遵守所使用的模型与数据集的许可证。
|
data/xtuner
ADDED
@@ -0,0 +1 @@
|
|
|
|
|
1 |
+
Subproject commit 90192ffe42612b0f88409432e7b4860294432bcc
|
model/paraphrase-multilingual-MiniLM-L12-v2
ADDED
@@ -0,0 +1 @@
|
|
|
|
|
1 |
+
Subproject commit e24d45fb8c210cedadb86b0dd70f3679bf216085
|
nltk_data
ADDED
@@ -0,0 +1 @@
|
|
|
|
|
1 |
+
Subproject commit 5db857e6f7df11eabb5e5665836db9ec8df07e28
|
requirements.txt
CHANGED
@@ -8,3 +8,4 @@ llama-index-embeddings-instructor==0.2.1
|
|
8 |
torch==2.5.0
|
9 |
torchvision==0.20.0
|
10 |
torchaudio==2.5.0
|
|
|
|
8 |
torch==2.5.0
|
9 |
torchvision==0.20.0
|
10 |
torchaudio==2.5.0
|
11 |
+
streamlit==1.39.0
|