[atom-vllm] adapt DeepSeek V4 MTP for vLLM plugin - #1372
Merged
Conversation
whx-sjtu
marked this pull request as draft
June 26, 2026 10:59
whx-sjtu
force-pushed
the
hexwang/ds_v4_mtp
branch
from
June 26, 2026 11:04
ed000a4 to
987cc28
Compare
Signed-off-by: whx-sjtu <xiaowang990929@gmail.com>
Signed-off-by: whx-sjtu <xiaowang990929@gmail.com>
Signed-off-by: whx-sjtu <xiaowang990929@gmail.com>
whx-sjtu
force-pushed
the
hexwang/ds_v4_mtp
branch
from
July 1, 2026 02:59
e8b6483 to
5fa130c
Compare
whx-sjtu
marked this pull request as ready for review
July 1, 2026 04:27
Signed-off-by: whx-sjtu <xiaowang990929@gmail.com>
| if getattr(common_attn_metadata, "max_query_len", 0) == 1: | ||
| max_q = int(getattr(common_attn_metadata, "max_query_len", 0) or 0) | ||
| decode_q = 1 + max(0, int(num_spec_tokens)) | ||
| if 1 <= max_q <= decode_q: |
Collaborator
There was a problem hiding this comment.
if there is a short prefill request or chunked prefill request in the batch, whose max_q is very small, it seems this request could be wrongly treated as DECODE
Contributor
Author
There was a problem hiding this comment.
Good finding. I will add extra check here to filter out prefill mixed batches.
Signed-off-by: whx-sjtu <xiaowang990929@gmail.com>
Signed-off-by: whx-sjtu <xiaowang990929@gmail.com>
whx-sjtu
force-pushed
the
hexwang/ds_v4_mtp
branch
from
July 2, 2026 09:15
61ff620 to
5b7f140
Compare
zejunchen-zejun
approved these changes
Jul 3, 2026
valarLip
approved these changes
Jul 6, 2026
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Summary
Test Result
DeepSeek V4 Flash, TP4
MTP=3 acceptance from the gsm8k run (48,367 draft steps × 3 tokens = 145,101 draft tokens; 77,360 accepted):
Per-position acceptance (how often the k-th speculated token is accepted):