How to evaluate ppl? - Githubissues

mit-han-lab / streaming-llm

[ICLR 2024] Efficient Streaming Language Models with Attention Sinks

https://arxiv.org/abs/2309.17453

MIT License

6.38k stars 355 forks source link

How to evaluate ppl? #77

Open Jiawei-Yang opened 4 months ago

Jiawei-Yang commented 4 months ago

Hi, thanks for your amazing work!

I noticed that there's an evaluation script for perplexity. I wonder how to replicate the results in tables 1 and 2. Are there any instructions for this?

seeyourcell commented 2 months ago

same question