书生大模型实战营（第三期）-L1：OpenCompass 评测 InternLM-1.8B 实践

毅毅酱

于 2024-08-25 16:35:53 发布

阅读量251

点赞数 8

文章标签：人工智能

本文链接：https://blog.csdn.net/qq_40384981/article/details/141527441

版权

概览

在 OpenCompass 中评估一个模型通常包括以下几个阶段：配置 -> 推理 -> 评估 -> 可视化。

配置：这是整个工作流的起点。您需要配置整个评估过程，选择要评估的模型和数据集。此外，还可以选择评估策略、计算后端等，并定义显示结果的方式。
推理与评估：在这个阶段，OpenCompass 将会开始对模型和数据集进行并行推理和评估。推理阶段主要是让模型从数据集产生输出，而评估阶段则是衡量这些输出与标准答案的匹配程度。这两个过程会被拆分为多个同时运行的“任务”以提高效率。
可视化：评估完成后，OpenCompass 将结果整理成易读的表格，并将其保存为 CSV 和 TXT 文件。

环境配置

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 .


apt-get update
apt-get install cmake
pip install -r requirements.txt
pip install protobuf

数据准备

解压评测数据集到 /root/opencompass/data/ 处。（注意：上方在git clone opencompass 时一定要将 opencompass clone 到 /root 路径下）

cp /share/temp/datasets/OpenCompassData-core-20231110.zip /root/opencompass/
unzip OpenCompassData-core-20231110.zip

InternLM和ceval 相关的配置文件

列出所有跟 InternLM 及 C-Eval 相关的配置

python tools/list_configs.py internlm ceval

启动评测 (10% A100 8GB 资源)

使用命令行配置参数法进行评测

打开 opencompass文件夹下configs/models/hf_internlm/的hf_internlm2_chat_1_8b.py ,贴入以下代码

from opencompass.models import HuggingFaceCausalLM


models = [
    dict(
        type=HuggingFaceCausalLM,
        abbr='internlm2-1.8b-hf',
        path="/share/new_models/Shanghai_AI_Laboratory/internlm2-chat-1_8b",
        tokenizer_path='/share/new_models/Shanghai_AI_Laboratory/internlm2-chat-1_8b',
        model_kwargs=dict(
            trust_remote_code=True,
            device_map='auto',
        ),
        tokenizer_kwargs=dict(
            padding_side='left',
            truncation_side='left',
            use_fast=False,
            trust_remote_code=True,
        ),
        max_out_len=100,
        min_out_len=1,
        max_seq_len=2048,
        batch_size=8,
        run_cfg=dict(num_gpus=1, num_procs=1),
    )
]

确保按照上述步骤正确安装 OpenCompass 并准备好数据集后，可以通过以下命令评测 InternLM2-Chat-1.8B 模型在 C-Eval 数据集上的性能。由于 OpenCompass 默认并行启动评估过程，我们可以在第一次运行时以 --debug 模式启动评估，并检查是否存在问题。在 --debug 模式下，任务将按顺序执行，并实时打印输出。

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

python run.py --datasets ceval_gen --models hf_internlm2_chat_1_8b --debug

08/25 15:42:33 - OpenCompass - INFO - time elapsed: 13944.93s
08/25 15:42:39 - OpenCompass - INFO - Partitioned into 52 tasks.
08/25 15:42:42 - OpenCompass - INFO - Task [internlm2-1.8b-hf/ceval-computer_network]: {'accuracy': 47.368421052631575}
08/25 15:42:44 - OpenCompass - INFO - Task [internlm2-1.8b-hf/ceval-operating_system]: {'accuracy': 47.368421052631575}
08/25 15:42:46 - OpenCompass - INFO - Task [internlm2-1.8b-hf/ceval-computer_architecture]: {'accuracy': 23.809523809523807}
08/25 15:42:48 - OpenCompass - INFO - Task [internlm2-1.8b-hf/ceval-college_programming]: {'accuracy': 27.027027027027028}
08/25 15:42:50 - OpenCompass - INFO - Task [internlm2-1.8b-hf/ceval-college_physics]: {'accuracy': 42.10526315789473}
08/25 15:42:52 - OpenCompass - INFO - Task [internlm2-1.8b-hf/ceval-college_chemistry]: {'accuracy': 37.5}
08/25 15:42:54 - OpenCompass - INFO - Task [internlm2-1.8b-hf/ceval-advanced_mathematics]: {'accuracy': 26.31578947368421}
08/25 15:42:56 - OpenCompass - INFO - Task [internlm2-1.8b-hf/ceval-probability_and_statistics]: {'accuracy': 22.22222222222222}
08/25 15:42:58 - OpenCompass - INFO - Task [internlm2-1.8b-hf/ceval-discrete_mathematics]: {'accuracy': 25.0}
08/25 15:43:00 - OpenCompass - INFO - Task [internlm2-1.8b-hf/ceval-electrical_engineer]: {'accuracy': 27.027027027027028}
08/25 15:43:02 - OpenCompass - INFO - Task [internlm2-1.8b-hf/ceval-metrology_engineer]: {'accuracy': 54.166666666666664}
08/25 15:43:04 - OpenCompass - INFO - Task [internlm2-1.8b-hf/ceval-high_school_mathematics]: {'accuracy': 22.22222222222222}
08/25 15:43:06 - OpenCompass - INFO - Task [internlm2-1.8b-hf/ceval-high_school_physics]: {'accuracy': 42.10526315789473}
08/25 15:43:08 - OpenCompass - INFO - Task [internlm2-1.8b-hf/ceval-high_school_chemistry]: {'accuracy': 52.63157894736842}
08/25 15:43:10 - OpenCompass - INFO - Task [internlm2-1.8b-hf/ceval-high_school_biology]: {'accuracy': 26.31578947368421}
08/25 15:43:12 - OpenCompass - INFO - Task [internlm2-1.8b-hf/ceval-middle_school_mathematics]: {'accuracy': 36.84210526315789}
08/25 15:43:14 - OpenCompass - INFO - Task [internlm2-1.8b-hf/ceval-middle_school_biology]: {'accuracy': 80.95238095238095}
08/25 15:43:16 - OpenCompass - INFO - Task [internlm2-1.8b-hf/ceval-middle_school_physics]: {'accuracy': 47.368421052631575}
08/25 15:43:18 - OpenCompass - INFO - Task [internlm2-1.8b-hf/ceval-middle_school_chemistry]: {'accuracy': 80.0}
08/25 15:43:20 - OpenCompass - INFO - Task [internlm2-1.8b-hf/ceval-veterinary_medicine]: {'accuracy': 43.47826086956522}
08/25 15:43:22 - OpenCompass - INFO - Task [internlm2-1.8b-hf/ceval-college_economics]: {'accuracy': 32.72727272727273}
08/25 15:43:24 - OpenCompass - INFO - Task [internlm2-1.8b-hf/ceval-business_administration]: {'accuracy': 39.39393939393939}
08/25 15:43:26 - OpenCompass - INFO - Task [internlm2-1.8b-hf/ceval-marxism]: {'accuracy': 68.42105263157895}
08/25 15:43:28 - OpenCompass - INFO - Task [internlm2-1.8b-hf/ceval-mao_zedong_thought]: {'accuracy': 70.83333333333334}
08/25 15:43:30 - OpenCompass - INFO - Task [internlm2-1.8b-hf/ceval-education_science]: {'accuracy': 55.172413793103445}
08/25 15:43:32 - OpenCompass - INFO - Task [internlm2-1.8b-hf/ceval-teacher_qualification]: {'accuracy': 59.09090909090909}
08/25 15:43:34 - OpenCompass - INFO - Task [internlm2-1.8b-hf/ceval-high_school_politics]: {'accuracy': 57.89473684210527}
08/25 15:43:36 - OpenCompass - INFO - Task [internlm2-1.8b-hf/ceval-high_school_geography]: {'accuracy': 47.368421052631575}
08/25 15:43:38 - OpenCompass - INFO - Task [internlm2-1.8b-hf/ceval-middle_school_politics]: {'accuracy': 76.19047619047619}
08/25 15:43:40 - OpenCompass - INFO - Task [internlm2-1.8b-hf/ceval-middle_school_geography]: {'accuracy': 75.0}
08/25 15:43:42 - OpenCompass - INFO - Task [internlm2-1.8b-hf/ceval-modern_chinese_history]: {'accuracy': 52.17391304347826}
08/25 15:43:44 - OpenCompass - INFO - Task [internlm2-1.8b-hf/ceval-ideological_and_moral_cultivation]: {'accuracy': 73.68421052631578}
08/25 15:43:46 - OpenCompass - INFO - Task [internlm2-1.8b-hf/ceval-logic]: {'accuracy': 31.818181818181817}
08/25 15:43:48 - OpenCompass - INFO - Task [internlm2-1.8b-hf/ceval-law]: {'accuracy': 29.166666666666668}
08/25 15:43:50 - OpenCompass - INFO - Task [internlm2-1.8b-hf/ceval-chinese_language_and_literature]: {'accuracy': 47.82608695652174}
08/25 15:43:52 - OpenCompass - INFO - Task [internlm2-1.8b-hf/ceval-art_studies]: {'accuracy': 42.42424242424242}
08/25 15:43:54 - OpenCompass - INFO - Task [internlm2-1.8b-hf/ceval-professional_tour_guide]: {'accuracy': 51.724137931034484}
08/25 15:43:56 - OpenCompass - INFO - Task [internlm2-1.8b-hf/ceval-legal_professional]: {'accuracy': 34.78260869565217}
08/25 15:43:58 - OpenCompass - INFO - Task [internlm2-1.8b-hf/ceval-high_school_chinese]: {'accuracy': 36.84210526315789}
08/25 15:44:00 - OpenCompass - INFO - Task [internlm2-1.8b-hf/ceval-high_school_history]: {'accuracy': 65.0}
08/25 15:44:02 - OpenCompass - INFO - Task [internlm2-1.8b-hf/ceval-middle_school_history]: {'accuracy': 86.36363636363636}
08/25 15:44:04 - OpenCompass - INFO - Task [internlm2-1.8b-hf/ceval-civil_servant]: {'accuracy': 42.5531914893617}
08/25 15:44:06 - OpenCompass - INFO - Task [internlm2-1.8b-hf/ceval-sports_science]: {'accuracy': 52.63157894736842}
08/25 15:44:08 - OpenCompass - INFO - Task [internlm2-1.8b-hf/ceval-plant_protection]: {'accuracy': 40.909090909090914}
08/25 15:44:10 - OpenCompass - INFO - Task [internlm2-1.8b-hf/ceval-basic_medicine]: {'accuracy': 68.42105263157895}
08/25 15:44:12 - OpenCompass - INFO - Task [internlm2-1.8b-hf/ceval-clinical_medicine]: {'accuracy': 36.36363636363637}
08/25 15:44:14 - OpenCompass - INFO - Task [internlm2-1.8b-hf/ceval-urban_and_rural_planner]: {'accuracy': 52.17391304347826}
08/25 15:44:16 - OpenCompass - INFO - Task [internlm2-1.8b-hf/ceval-accountant]: {'accuracy': 36.734693877551024}
08/25 15:44:18 - OpenCompass - INFO - Task [internlm2-1.8b-hf/ceval-fire_engineer]: {'accuracy': 38.70967741935484}
08/25 15:44:20 - OpenCompass - INFO - Task [internlm2-1.8b-hf/ceval-environmental_impact_assessment_engineer]: {'accuracy': 51.61290322580645}
08/25 15:44:22 - OpenCompass - INFO - Task [internlm2-1.8b-hf/ceval-tax_accountant]: {'accuracy': 36.734693877551024}
08/25 15:44:24 - OpenCompass - INFO - Task [internlm2-1.8b-hf/ceval-physician]: {'accuracy': 42.857142857142854}

dataset                                         version    metric         mode      internlm2-1.8b-hf
----------------------------------------------  ---------  -------------  ------  -------------------
ceval-computer_network                          db9ce2     accuracy       gen                   47.37
ceval-operating_system                          1c2571     accuracy       gen                   47.37
ceval-computer_architecture                     a74dad     accuracy       gen                   23.81
ceval-college_programming                       4ca32a     accuracy       gen                   27.03
ceval-college_physics                           963fa8     accuracy       gen                   42.11
ceval-college_chemistry                         e78857     accuracy       gen                   37.5
ceval-advanced_mathematics                      ce03e2     accuracy       gen                   26.32
ceval-probability_and_statistics                65e812     accuracy       gen                   22.22
ceval-discrete_mathematics                      e894ae     accuracy       gen                   25
ceval-electrical_engineer                       ae42b9     accuracy       gen                   27.03
ceval-metrology_engineer                        ee34ea     accuracy       gen                   54.17
ceval-high_school_mathematics                   1dc5bf     accuracy       gen                   22.22
ceval-high_school_physics                       adf25f     accuracy       gen                   42.11
ceval-high_school_chemistry                     2ed27f     accuracy       gen                   52.63
ceval-high_school_biology                       8e2b9a     accuracy       gen                   26.32
ceval-middle_school_mathematics                 bee8d5     accuracy       gen                   36.84
ceval-middle_school_biology                     86817c     accuracy       gen                   80.95
ceval-middle_school_physics                     8accf6     accuracy       gen                   47.37
ceval-middle_school_chemistry                   167a15     accuracy       gen                   80
ceval-veterinary_medicine                       b4e08d     accuracy       gen                   43.48
ceval-college_economics                         f3f4e6     accuracy       gen                   32.73
ceval-business_administration                   c1614e     accuracy       gen                   39.39
ceval-marxism                                   cf874c     accuracy       gen                   68.42
ceval-mao_zedong_thought                        51c7a4     accuracy       gen                   70.83
ceval-education_science                         591fee     accuracy       gen                   55.17
ceval-teacher_qualification                     4e4ced     accuracy       gen                   59.09
ceval-high_school_politics                      5c0de2     accuracy       gen                   57.89
ceval-high_school_geography                     865461     accuracy       gen                   47.37
ceval-middle_school_politics                    5be3e7     accuracy       gen                   76.19
ceval-middle_school_geography                   8a63be     accuracy       gen                   75
ceval-modern_chinese_history                    fc01af     accuracy       gen                   52.17
ceval-ideological_and_moral_cultivation         a2aa4a     accuracy       gen                   73.68
ceval-logic                                     f5b022     accuracy       gen                   31.82
ceval-law                                       a110a1     accuracy       gen                   29.17
ceval-chinese_language_and_literature           0f8b68     accuracy       gen                   47.83
ceval-art_studies                               2a1300     accuracy       gen                   42.42
ceval-professional_tour_guide                   4e673e     accuracy       gen                   51.72
ceval-legal_professional                        ce8787     accuracy       gen                   34.78
ceval-high_school_chinese                       315705     accuracy       gen                   36.84
ceval-high_school_history                       7eb30a     accuracy       gen                   65
ceval-middle_school_history                     48ab4a     accuracy       gen                   86.36
ceval-civil_servant                             87d061     accuracy       gen                   42.55
ceval-sports_science                            70f27b     accuracy       gen                   52.63
ceval-plant_protection                          8941f9     accuracy       gen                   40.91
ceval-basic_medicine                            c409d6     accuracy       gen                   68.42
ceval-clinical_medicine                         49e82d     accuracy       gen                   36.36
ceval-urban_and_rural_planner                   95b885     accuracy       gen                   52.17
ceval-accountant                                002837     accuracy       gen                   36.73
ceval-fire_engineer                             bc23f5     accuracy       gen                   38.71
ceval-environmental_impact_assessment_engineer  c64e2d     accuracy       gen                   51.61
ceval-tax_accountant                            3a5e3c     accuracy       gen                   36.73
ceval-physician                                 6e277d     accuracy       gen                   42.86
ceval-stem                                      -          naive_average  gen                   40.59
ceval-social-science                            -          naive_average  gen                   58.21
ceval-humanities                                -          naive_average  gen                   50.16
ceval-other                                     -          naive_average  gen                   45.43
ceval-hard                                      -          naive_average  gen                   33.76
ceval                                           -          naive_average  gen                   47.03