Skip to content

[blog] add blog about DFlash and Spec V2#351

Open
charlesfrye wants to merge 13 commits into
lm-sys:mainfrom
charlesfrye:main
Open

[blog] add blog about DFlash and Spec V2#351
charlesfrye wants to merge 13 commits into
lm-sys:mainfrom
charlesfrye:main

Conversation

@charlesfrye

Copy link
Copy Markdown

This PR adds a joint blog bost by the Z Lab, Modal, and SGLang teams.

The blog

  • announces the release of a Qwen 3.5 397B-A17B DFlash drafter that improves on the native MTP module in all benchmarked settings
  • explains the DFlash architecture
  • explains how that architecture was integrated into SGLang in Spec V1 and V2
  • directs readers to try generic DFlash x Spec V2 drafters for Qwen models or to contact the authors about training their own

A preview of the blog (as of creation of this PR) can be found here. Screenshot of the opening below. The blog is intended for release on the morning of June 15th, Pacific Time.

Screenshot 2026-06-14 at 4 30 10 PM

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant