[Feature][AutoDeploy]: Piecewise CUDA graph for MTP
Open
Nobody has claimed this yet.
feature request
Speculative Decoding
- Dominant language
- Python
- Stars
- 14.7k
- Forks
- 2.8k
- Avg merge
- 2d 23h
- Merged PRs (30d)
- 489
Description
🚀 The feature, motivation and pitch
Right now MTP CUDA graphing is only for extend-only (spec decoding -only) batches. It would be nice to extend this to mixed prefill/extend.
Part of the work here is eliminating host syncs in mixed prefill/decode, which has not been done yet.
Another is properly adapting Piecewise CUDA graph to put the whole EagleWrapper.forward() into a single CUDA graph, despite the fact that it is NOT a GraphModule. Currently the Piecewise CUDA graph code compiles individual graph modules to handle VLMs - but it needs to simultaneously handle the speculative decoding use case.
Alternatives
No response
Additional context
No response
Before submitting a new issue...
- Make sure you already searched for relevant issues, and checked the documentation and examples for answers to frequently asked questions.
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.
Assessment
This issue has not been assessed yet.