alibaba / alibaba/rtp-llm

[Feature] New Speculative Decoding Framework

Open
#304 0 comments 0 reactions 1 assignee Claimed by @Vinkle-hzt View on GitHub
enhancement
Dominant language
Python
Stars
1.3k
Forks
275
Avg merge
3d 17h
Merged PRs (30d)
33

Description

Due to the high CPU overhead in the existing speculative decoding framework, we are developing a brand new framework that significantly reduces CPU consumption and minimizes device-to-host synchronization.

### Worklist
- [x] basic feature [#305]
- support 1 & multi steps mtp
- support py model
- support PD-seperation
- support DP
- fast & async mtp process
- [x] vocab prune
- [x] support py model cuda graph [#516]
- [x] rm old framework

Contributor guide

No contributing guide indexed for this repository

Assessment

This issue has not been assessed yet.

Get new issues in your inbox

A short digest of beginner-friendly GitHub issues.