LMDeploy Lite vs LLM Compressor
Open
- Dominant language
- Python
- Stars
- 8.1k
- Forks
- 748
- Avg merge
- 6d 2h
- Merged PRs (30d)
- 54
Description
It seems like LLM Compressor is not supported in LMDeploy, but LMDeploy lite takes up a lot of memory quantizing, with a 14B I can only do `--calib-samples 128 --calib-seqlen 6144`. Any suggestion on this?
Target: `--calib-samples 10000 --calib-seqlen 12288`.
Problem:
- LMDeploy lite only supports one device
- LLM Compressor quantized models are not supported by LMDeploy
Contributor guide
Assessment
This issue has not been assessed yet.