fix format problems in evaluation code; update ceval extraction rules

This commit is contained in:
feihu.hf
2023-08-25 22:44:07 +08:00
parent 1a9a04a91e
commit 4864f7b278
11 changed files with 1507 additions and 808 deletions

View File

@@ -34,6 +34,19 @@ pip install thefuzz
python evaluate_chat_mmlu.py -d data/mmlu/data/
```
- CMMLU
```Shell
wget https://huggingface.co/datasets/haonan-li/cmmlu/resolve/main/cmmlu_v1_0_1.zip
mkdir data/cmmlu
mv cmmlu_v1_0_1.zip data/cmmlu
cd data/cmmlu; unzip cmmlu_v1_0_1.zip
cd ../../
# Qwen-7B
python evaluate_cmmlu.py -d data/cmmlu/
```
- HumanEval
Get the HumanEval.jsonl file from [here](https://github.com/openai/human-eval/tree/master/data)