DFloat11 support
Open
Feature
- Dominant language
- Python
- Stars
- 133k
- Forks
- 15.7k
- Avg merge
- 1d 6h
- Merged PRs (30d)
- 155
Description
### Feature Idea
DFloat11 is a new datatype that compresses bfloat16 to 11 bits while maintaining 100% accuracy (bit-perfect). This would be great to have as an alternative to fp8 for those who barely don't have enough vram for bfloat16, or to increase maximum resolution at full precision.
[Code](https://github.com/LeanModels/DFloat11)
[Flux Model](https://huggingface.co/DFloat11/FLUX.1-dev-DF11)
### Existing Solutions
_No response_
### Other
_No response_
Contributor guide
Assessment
This issue has not been assessed yet.