Llama2-7B-chat-4k on PassageRetrieval-zh gets 10.12

[fuqichen1998]

As the title, my evaluation of Llama2-7B-chat-4k on PassageRetrieval-zh gets 10.12, which is significantly higher than the README (0.5), could you please share why?

[bys0318]

Hi! Are you using the prompt template as in config/dataset2prompt.json?

[bys0318]

We refer to our code here for the llama2 prompt: https://github.com/THUDM/LongBench/blob/main/pred.py#L33

[fuqichen1998]

Yes, I was using your pred.py to run the inference and evaluation.

[slatter666]

Yes, I was using your pred.py to run the inference and evaluation.

Acutally I also get the same result

[condy0919]

We refer to our code here for the llama2 prompt: https://github.com/THUDM/LongBench/blob/main/pred.py#L33

The INST is necessary for llama2-7b/llama2-13b?