fp8_rowwise and kv cache
Open
Nobody has claimed this yet.
question
triaged
- Dominant language
- Python
- Stars
- 14.7k
- Forks
- 2.8k
- Avg merge
- 2d 23h
- Merged PRs (30d)
- 489
Description
Hi!
If I understand correctly fp8_kv_cache only works for FP8 quantization.
Is it possible for a llama-like model to use fp8 kv cache for FP8_PER_CHANNEL_PER_TOKEN quantization? If not, are there any plans to support it?
Contributor guide
First steps
- Read the whole issue, then the project's contributing guide.
- Comment on the issue to say you are picking it up — it saves two people doing the same work.
- Fork the repository and make your change on a branch.
- Open a pull request that references the issue number.
Research direction
Start by locating the fp8_kv_cache and FP8_PER_CHANNEL_PER_TOKEN quantization entry points; the issue names no specific files or tests. Determine whether a llama-like model can combine these modes, then confirm the supported behavior or define the required implementation and coverage.
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
- Stale
- Clarity
- Needs clarification
- Newbie friendliness
- 25/100