Skip to content

[atom-vllm] adapt DeepSeek V4 MTP for vLLM plugin#1372

Draft
whx-sjtu wants to merge 1 commit into
mainfrom
hexwang/ds_v4_mtp
Draft

[atom-vllm] adapt DeepSeek V4 MTP for vLLM plugin#1372
whx-sjtu wants to merge 1 commit into
mainfrom
hexwang/ds_v4_mtp

Conversation

@whx-sjtu

@whx-sjtu whx-sjtu commented Jun 26, 2026

Copy link
Copy Markdown
Contributor

Summary

  • Wire DeepSeek V4 MTP into the ATOM vLLM plugin, including draft model registration, target/draft sharing, V4 proxy cache handling, and MTP-specific metadata adaptation.
  • Add a validation utility and focused tests for DeepSeek V4 MTP registration, hidden-state contracts, metrics parsing, and vLLM speculative decoding patches.

Test Result

Signed-off-by: whx-sjtu <xiaowang990929@gmail.com>
@whx-sjtu whx-sjtu changed the title adapt DeepSeek V4 MTP for vLLM plugin [atom-vllm] adapt DeepSeek V4 MTP for vLLM plugin Jun 26, 2026
@whx-sjtu whx-sjtu marked this pull request as draft June 26, 2026 10:59
@whx-sjtu whx-sjtu force-pushed the hexwang/ds_v4_mtp branch from ed000a4 to 987cc28 Compare June 26, 2026 11:04
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