cornzz

Seems I found the issue, PR: https://github.com/mistralai/mistral-inference/pull/216

It seems that this is a widespread inconsistency, for example in the RepoBench-P repository, `python-Levenshtein` is [listed as a requirement](https://github.com/Leolty/repobench/blob/main/requirements.txt), but in [the LCC repo](https://github.com/microsoft/CodeBERT/tree/master/LongCoder) it is not mentioned... Just...

-- Moved to separate issue: https://github.com/microsoft/LLMLingua/issues/196 --

Reopening as I cannot figure out how to correctly use LLMLingua without overshooting the target compression rate. No matter how I set `iterative_size`, large prompts (2K+) are overcompressed.

@iofu728 thanks a lot for your response! Regarding 3: is this what is referred to in the last paragraph of section 4.2 in the paper? > our approach can be...

I also want to know what the compression targets are for GSM8K / BBH for 1-shot / half-shot etc., what is the target_token? I was also wondering what zero-shot means...

@iofu728 sorry for bothering, but what exactly is the definition of "zero-shot" in the context of the ZeroScrolls benchmark? As stated [here](https://huggingface.co/datasets/tau/zero_scrolls), ZeroScrolls is already a zero-shot benchmark by itself:...

@iofu728 thanks for your response! I have some follow-up questions: 1. > "1-shot" means only one example is retained I do not quite understand this, as in the results table...

@pzs19 Hi, sorry for bothering, but I am still not clear how to reproduce the zero-shot results, it seems from the previous response of your co-author, that a different approach...

Hi @iofu728, any update, did you get a chance to check if this finding was valid? 😅