【书生2.6】OpenCompass 评测 InternLM-1.8B 实践

【自有环境】

可以基于原先的llamaindex环境

重要：opencompass必须放在/root下面

1 环境准备

conda create -n opencompass python=3.10
conda activate opencompass
conda install pytorch==2.1.2 torchvision==0.16.2 torchaudio==2.1.2 pytorch-cuda=12.1 -c pytorch -c nvidia -y

# 注意：一定要先 cd /root
cd /root
git clone -b 0.2.4 https://github.com/open-compass/opencompass
cd opencompass
pip install -e .


yum update
yum install cmake
pip install -r requirements.txt
pip install protobuf

2 数据准备

# 下载/share/temp/datasets/OpenCompassData-core-20231110.zip
# 上传到自己的服务器 /root/opencompass/
cd /root/opencompass/
unzip OpenCompassData-core-20231110.zip

3 评测

3.1 使用命令行配置参数法进行评测

vim configs/models/hf_internlm/hf_internlm2_chat_1_8b.py
# 修改path、 tokenizer_path路径


#环境变量配置
export MKL_SERVICE_FORCE_INTEL=1
#或
export MKL_THREADING_LAYER=GNU

CUDA_VISIBLE_DEVICES=1,2,3,4,5 python run.py
--datasets ceval_gen \ # 数据集准备
--models hf_internlm2_chat_1_8b \  # 模型准备
--debug

3.2 使用配置文件修改参数法进行评测

cd /root/opencompass/configs
touch eval_tutorial_demo.py
# vi eval_tutorial_demo.py
from mmengine.config import read_base

with read_base():
    from .datasets.ceval.ceval_gen import ceval_datasets
    from .models.hf_internlm.hf_internlm2_chat_1_8b import models as hf_internlm2_chat_1_8b_models

datasets = ceval_datasets
models = hf_internlm2_chat_1_8b_models

cd /root/opencompass
CUDA_VISIBLE_DEVICES=1,2,3,4,5 python run.py configs/eval_tutorial_demo.py --debug
# 和3.1的结果是一致的

3.3 api模型评测（进阶必做）

参考文档

这些都要花钱，我们尝试之前lmdeploy部署的api：

from mmengine.config import read_base
from opencompass.models import OpenAI
from opencompass.partitioners import NaivePartitioner
from opencompass.runners import LocalRunner
from opencompass.tasks import OpenICLInferTask

with read_base():
    # choose a list of datasets
    from .datasets.collections.chat_medium import datasets
    # and output the results in a choosen format
    from .summarizers.medium import summarizer


api_meta_template = dict(
    round=[
            dict(role='HUMAN', api_role='HUMAN'),
            dict(role='BOT', api_role='BOT', generate=True),
    ],
)

models = [
    dict(abbr='internlm2-chat-1.8b-hf',
        type=OpenAI, path='internlm/internlm2-chat-1_8b',
        base_url="http://192.168.26.212:23333/v1",
        key='internlm2',  # The key will be obtained from $OPENAI_API_KEY, but you can write down your key here as well
        meta_template=api_meta_template,
        query_per_second=1,
        max_out_len=2048, max_seq_len=4096, batch_size=8),
]

infer = dict(
    partitioner=dict(type=NaivePartitioner),
    runner=dict(
        type=LocalRunner,
        max_num_workers=8,
        task=dict(type=OpenICLInferTask)),
)

# 不支持url或者base_url
``![在这里插入图片描述](https://i-blog.csdnimg.cn/direct/dfd3a1247ba943d9b277eef310a11618.png#pic_center)