search

microsoft / LLMLingua

To speed up LLMs' inference and enhance LLM's perceive of key information, compress the prompt and KV-Cache, which achieves up to 20x compression with minimal performance loss.
https://llmlingua.com/
MIT License
4.18k stars 222 forks source link

Experiments[Question]: #164

Closed 520zw closed 2 weeks ago

520zw commented 1 month ago

Describe the issue

what is 1-shot / half-shot /quarter-shot constraint in experiments

iofu728 commented 2 weeks ago

Hi @520zw, sorry for the late response.

In Table 2, the "1/half/quarter-shot constraint" refers to few-shot learning tasks, meaning the target compression ratio is equal to the original token size / the token size of a 1-shot demonstration, and so on.

520zw commented 2 weeks ago

Thanks for the reply, does the one-shot demonstration mean that Cot only contains one reference example?is there a requirement to select this reference example?