[Bug] 微调后的 qwen-vl-chat的模型,使用提供的方法量化成 INT4,显存占用没有下降
Open
- Dominant language
- Python
- Stars
- 8.1k
- Forks
- 748
- Avg merge
- 6d 2h
- Merged PRs (30d)
- 54
Description
### Checklist
- [X] 1. I have searched related issues but cannot get the expected help.
- [X] 2. The bug has not been fixed in the latest version.
### Describe the bug
微调后的 qwen-vl-chat的模型,使用提供的方法量化成 INT4,推理时显存占用没有下降 , 未量化前占用 20G, 量化后占用 22G
### Reproduction
python38
### Environment
```Shell
cuda11.7 ATX4090 显卡 python38 torch2.2.2 lmdeploy 0.5.0
```
### Error traceback
_No response_
Contributor guide
Assessment
This issue has not been assessed yet.