[Feature]: support and eval H-Scale: Hessian-Guided Scale Refinement for NVFP4 Sub-Byte LLM Inference
Open
enhancement
- Dominant language
- Python
- Stars
- 1.6k
- Forks
- 175
- Avg merge
- 1d 18h
- Merged PRs (30d)
- 99
Description
### Feature Description
Though I think iMatrix has already done a large part of the work for Hessian part, let's evaluate it when we have some bandwidth.
### Motivation and Use Case
~
### Alternatives Considered
~
### Definition of Done
_No response_
### Additional Context
_No response_
Contributor guide
Research direction
No files, tests, entry points, or acceptance criteria are named. Start by locating the existing iMatrix Hessian-related implementation and the NVFP4 quantization and inference entry points, then determine how H-Scale should be evaluated and what results would constitute completion.
Written by the indexing model from the issue text.
Assessment
- Tech stack
- python
- Domain
- machine-learning
- Issue type
- Feature
- Difficulty
- 5/5
- Estimated time
- Over a week
- Activity status
- Active
- Clarity
- Needs clarification
- Newbie friendliness
- 25/100