Results of LLaMA-2 are different from Wanda

Question

Results of LLaMA-2 are different from Wanda

Closed this issue 10 months ago · 2 comments

For LLaMA-2, Wanda and GBLM-Pruner got different ppl. Any thought?

Answer 1 · 2024-02-03T17:13:22.000Z

Hi, thanks for the question. The numbers are different because, in the Wanda paper, they computed the perplexity for LLaMA-2 using a sequence length of 4096 from Wiki-Text, whereas in GBLM-Pruner, the sequence length is 2048.

Answer 2 · 2024-03-22T02:30:09.000Z

Thank you for your answers. 😄