FP8 support?
Open
question
- Dominant language
- Python
- Stars
- 5.4k
- Forks
- 850
- Avg merge
- 4d 5h
- Merged PRs (30d)
- 10
Description
Hi.
Neural Engine on A20 Pro, and maybe M6, supports native FP8 data type.
How do you prepare models for it using coremltools?
Currently, ct.precision has cases of FLOAT16 and FLOAT32 only.
Thanks!
Contributor guide
Research direction
Start by reviewing the ct.precision cases for FLOAT16 and FLOAT32 and the model-conversion documentation referenced by the issue. Determine whether the requested work is FP8 preparation, conversion, or precision API support; done should be clear guidance or supported FP8 model preparation through coremltools.
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
- 35/100