facebookresearch / facebookresearch/LLM-QAT
Expects full precision but got torch.bfloat16 error
Open
- Dominant language
- Python
- Stars
- 328
- Forks
- 25
- PR merge metrics
- No merged PRs in 30d
Description
## Env
- cuda 11.7
- pytorch 1.13.1
- python 3.10
when i execute train script , run the line:
https://github.com/facebookresearch/LLM-QAT/blob/main/utils/utils.py#L41
```
state_dict = trainer.model.state_dict()
```
occur an error ：
```
Expects full precision but got torch.bfloat16
```
---
hope to get your answer。
Thanks
Contributor guide
Assessment
This issue has not been assessed yet.