[Bug] which version of the dataset should be selected When evaluating the Llama3 model,

Prerequisite

• I have searched Issues and Discussions but cannot get the expected help.
• The bug has not been fixed in the latest version.

Type

I'm evaluating with the officially supported tasks/models/datasets.

Environment

OpenCompass v0.2.5
python 3.10
Transformers 4.41.0
Troh 2.1.2

Reproduces the problem - code/configuration sample

Original Llama3 configuration, only path was updated locally.

Reproduces the problem - command or script

Llama3-8b base
python run.py --models hf_llama3_8b --datasets triviaqa_wiki_gen_d18bf4 squad20_gen_1710bc agieval_mixed_2f14ad
Llam3-8b-instruct
python run.py --models hf_llama3_8b_instruct --datasets humaneval_gen_a82cae

Reproduces the problem - error message

There is a big gap between using tools and official websites

Llama3-8b base:
OpenCompass:
dataset,version,metric,mode,llama-3-8b-hf
triviaqa_wiki_0shot,de8113,score,gen,30.95
triviaqa_wiki_1shot,eaf81e,score,gen,63.31
triviaqa_wiki_5shot,e0f87f,score,gen,54.61
triviaqa_wiki_25shot,c60113,score,gen,57.75
squad2.0,c6a034,score,gen,37.03
Official website:

Llam3-8b-instruct:
OpenCompass
openai_humaneval,8e312c,humaneval_pass@1,gen,54.88
Official website:

And the official website contrast has the very big disparity, 2.0 datasets not released I'm not sure if I should choose these datasets or what configuration errors are causing this gap

Other information

No response