kv int8 量化， 能较少显存，吞吐基本没有提升？
Open
- Dominant language
- Python
- Stars
- 8.1k
- Forks
- 748
- Avg merge
- 6d 2h
- Merged PRs (30d)
- 54
Description
### Motivation
kv int8 量化，吞吐基本没有提升？
use_context_fmha = 0
quant_policy = 4
max batch throughput 1160 token/s
use_context_fmha = 0
quant_policy = 0
max batch throughput 1012 token/s
看网上的材料能显著提升吞吐
### Related resources
_No response_
### Additional context
_No response_
Contributor guide
Assessment
This issue has not been assessed yet.