OpenCompass/docs/zh_cn/get_started/installation.md

142 lines
4.8 KiB
Markdown
Raw Permalink Normal View History

# 安装
## 基础安装
1. 使用Conda准备 OpenCompass 运行环境:
```bash
conda create --name opencompass python=3.10 -y
# conda create --name opencompass_lmdeploy python=3.10 -y
conda activate opencompass
```
如果你希望自定义 PyTorch 版本,请参考 [官方文档](https://pytorch.org/get-started/locally/) 准备 PyTorch 环境。需要注意的是OpenCompass 要求 `pytorch>=1.13`
2. 安装 OpenCompass
- pip安装
```bash
# 支持绝大多数数据集及模型
pip install -U opencompass
# 完整安装(支持更多数据集)
# pip install "opencompass[full]"
# API 测试(例如 OpenAI、Qwen
# pip install "opencompass[api]"
```
- 如果希望使用 OpenCompass 的最新功能,也可以从源代码构建它:
```bash
git clone https://github.com/open-compass/opencompass opencompass
cd opencompass
pip install -e .
```
## 其他安装
### 推理后端
```bash
# 模型推理后端,由于这些推理后端通常存在依赖冲突,建议使用不同的虚拟环境来管理它们。
pip install "opencompass[lmdeploy]"
# pip install "opencompass[vllm]"
```
- LMDeploy
可以通过下列命令判断推理后端是否安装成功,更多信息请参考 [官方文档](https://lmdeploy.readthedocs.io/zh-cn/latest/get_started.html)
```bash
lmdeploy chat internlm/internlm2_5-1_8b-chat --backend turbomind
```
- vLLM
可以通过下列命令判断推理后端是否安装成功,更多信息请参考 [官方文档](https://docs.vllm.ai/en/latest/getting_started/quickstart.html)
```bash
vllm serve facebook/opt-125m
```
### API
Opencompass支持不同的商业模型API调用你可以通过pip方式安装或者参考 [API](https://github.com/open-compass/opencompass/blob/main/requirements/api.txt) 安装对应的API模型依赖
```bash
pip install opencompass[api]
# pip install openai # GPT-3.5-Turbo / GPT-4-Turbo / GPT-4 / GPT-4o (API)
# pip install anthropic # Claude (API)
# pip install dashscope # 通义千问 (API)
# pip install volcengine-python-sdk # 字节豆包 (API)
# ...
```
### 数据集
基础安装可以支持绝大部分基础数据集针对某些数据集i.e. Alpaca-eval, Longbench etc.),需要安装额外的依赖。
你可以通过pip方式安装或者参考 [额外依赖](https://github.com/open-compass/opencompass/blob/main/requirements/extra.txt) 安装对应的依赖
```bash
pip install opencompass[full]
```
针对 HumanEvalX / HumanEval+ / MBPP+ 需要手动clone git仓库进行安装
```bash
git clone --recurse-submodules git@github.com:open-compass/human-eval.git
cd human-eval
pip install -e .
pip install -e evalplus
```
部分智能体评测需要安装大量依赖且可能会与已有运行环境冲突我们建议创建不同的conda环境来管理
```bash
# T-Eval
pip install lagent==0.1.2
# CIBench
pip install -r requirements/agent.txt
```
## 数据集准备
[Feature] Support ModelScope datasets (#1289) * add ceval, gsm8k modelscope surpport * update race, mmlu, arc, cmmlu, commonsenseqa, humaneval and unittest * update bbh, flores, obqa, siqa, storycloze, summedits, winogrande, xsum datasets * format file * format file * update dataset format * support ms_dataset * udpate dataset for modelscope support * merge myl_dev and update test_ms_dataset * udpate dataset for modelscope support * update readme * update eval_api_zhipu_v2 * remove unused code * add get_data_path function * update readme * remove tydiqa japanese subset * add ceval, gsm8k modelscope surpport * update race, mmlu, arc, cmmlu, commonsenseqa, humaneval and unittest * update bbh, flores, obqa, siqa, storycloze, summedits, winogrande, xsum datasets * format file * format file * update dataset format * support ms_dataset * udpate dataset for modelscope support * merge myl_dev and update test_ms_dataset * update readme * udpate dataset for modelscope support * update eval_api_zhipu_v2 * remove unused code * add get_data_path function * remove tydiqa japanese subset * update util * remove .DS_Store * fix md format * move util into package * update docs/get_started.md * restore eval_api_zhipu_v2.py, add environment setting * Update dataset * Update * Update * Update * Update --------- Co-authored-by: Yun lin <yunlin@U-Q9X2K4QV-1904.local> Co-authored-by: Yunnglin <mao.looper@qq.com> Co-authored-by: Yun lin <yunlin@laptop.local> Co-authored-by: Yunnglin <maoyl@smail.nju.edu.cn> Co-authored-by: zhangsongyang <zhangsongyang@pjlab.org.cn>
2024-07-29 13:48:32 +08:00
OpenCompass 支持的数据集主要包括三个部分:
1. Huggingface 数据集: [Huggingface Dataset](https://huggingface.co/datasets) 提供了大量的数据集,这部分数据集运行时会**自动下载**。
[Feature] Support ModelScope datasets (#1289) * add ceval, gsm8k modelscope surpport * update race, mmlu, arc, cmmlu, commonsenseqa, humaneval and unittest * update bbh, flores, obqa, siqa, storycloze, summedits, winogrande, xsum datasets * format file * format file * update dataset format * support ms_dataset * udpate dataset for modelscope support * merge myl_dev and update test_ms_dataset * udpate dataset for modelscope support * update readme * update eval_api_zhipu_v2 * remove unused code * add get_data_path function * update readme * remove tydiqa japanese subset * add ceval, gsm8k modelscope surpport * update race, mmlu, arc, cmmlu, commonsenseqa, humaneval and unittest * update bbh, flores, obqa, siqa, storycloze, summedits, winogrande, xsum datasets * format file * format file * update dataset format * support ms_dataset * udpate dataset for modelscope support * merge myl_dev and update test_ms_dataset * update readme * udpate dataset for modelscope support * update eval_api_zhipu_v2 * remove unused code * add get_data_path function * remove tydiqa japanese subset * update util * remove .DS_Store * fix md format * move util into package * update docs/get_started.md * restore eval_api_zhipu_v2.py, add environment setting * Update dataset * Update * Update * Update * Update --------- Co-authored-by: Yun lin <yunlin@U-Q9X2K4QV-1904.local> Co-authored-by: Yunnglin <mao.looper@qq.com> Co-authored-by: Yun lin <yunlin@laptop.local> Co-authored-by: Yunnglin <maoyl@smail.nju.edu.cn> Co-authored-by: zhangsongyang <zhangsongyang@pjlab.org.cn>
2024-07-29 13:48:32 +08:00
2. ModelScope 数据集:[ModelScope OpenCompass Dataset](https://modelscope.cn/organization/opencompass) 支持从 ModelScope 自动下载数据集。
要启用此功能,请设置环境变量:`export DATASET_SOURCE=ModelScope`,可用的数据集包括(来源于 OpenCompassData-core.zip
```plain
humaneval, triviaqa, commonsenseqa, tydiqa, strategyqa, cmmlu, lambada, piqa, ceval, math, LCSTS, Xsum, winogrande, openbookqa, AGIEval, gsm8k, nq, race, siqa, mbpp, mmlu, hellaswag, ARC, BBH, xstory_cloze, summedits, GAOKAO-BENCH, OCNLI, cmnli
```
3. 自建以及第三方数据集OpenCompass 还提供了一些第三方数据集及自建**中文**数据集。运行以下命令**手动下载解压**。
在 OpenCompass 项目根目录下运行下面命令,将数据集准备至 `${OpenCompass}/data` 目录下:
```bash
2024-02-28 10:54:04 +08:00
wget https://github.com/open-compass/opencompass/releases/download/0.2.2.rc1/OpenCompassData-core-20240207.zip
unzip OpenCompassData-core-20240207.zip
```
如果需要使用 OpenCompass 提供的更加完整的数据集 (~500M),可以使用下述命令进行下载和解压:
```bash
2024-02-28 10:54:04 +08:00
wget https://github.com/open-compass/opencompass/releases/download/0.2.2.rc1/OpenCompassData-complete-20240207.zip
unzip OpenCompassData-complete-20240207.zip
cd ./data
find . -name "*.zip" -exec unzip "{}" \;
```
2024-02-28 10:54:04 +08:00
两个 `.zip` 中所含数据集列表如[此处](https://github.com/open-compass/opencompass/releases/tag/0.2.2.rc1)所示。
OpenCompass 已经支持了大多数常用于性能比较的数据集,具体支持的数据集列表请直接在 `configs/datasets` 下进行查找。
接下来,你可以阅读[快速上手](./quick_start.md)了解 OpenCompass 的基本用法。