[bugfix] add with_prefill cpu allreduce to handle D-node recomputatio… #2129
Conversation
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
liziyu179
force-pushed
the
refactor_pd_with_prefill
branch
2 times, most recently
from
e572c87 to
b7966ea
Compare
…n situations Signed-off-by: liziyu <liziyu16@huawei.com>
liziyu179
force-pushed
the
refactor_pd_with_prefill
branch
from
b7966ea to
9bff5c7
Compare
liyu119
added a commit
to rjg-lyh/vllm-ascend
that referenced
this pull request
Aug 11, 2025
…nto qwen30-dev * 'qwen30-dev' of https://github.com/rjg-lyh/vllm-ascend: [V0.9.1] Replace FA ops with FA_V2 to optimize perf [0.9.1]remove chunked_prefill_for_mla (vllm-project#2177) move with_prefill allreduce from cpu to npu (vllm-project#2230) [v0.9.1] Add release note for v0.9.1rc2 (vllm-project#2233) [Docs] Sync main doc to v0.9.1-dev (vllm-project#2227) [0.9.1] Enable external distributed dp deployments in vllm ascend(0.9.1 only) (vllm-project#2109) [V0.9.1][BugFix] Fix the bug in decoraotor patch (vllm-project#2199) [v0.9.1][Bugfix][PD] Auto-clear producer KV cache if no pull notification (vllm-project#2085) [BUGFIX][0.9.1] FIX ring_mla input ‘query_lens’ to cpu (vllm-project#2170) [0.9.1][Prefill Perf] add D2H & initRoutingQuantV2 (vllm-project#2038) [bugfix] add with_prefill cpu allreduce to handle D-node recomputatio… (vllm-project#2129)
momo609
reviewed
Sep 12, 2025
| with_prefill_tensor = torch.tensor([with_prefill], | ||
| device="cpu", | ||
| dtype=torch.bool) | ||
| dist.all_reduce(with_prefill_tensor, |
Collaborator
There was a problem hiding this comment.
Has the performance impact of this been tested? Can it be recovered using the allreduce communication in num_tokens_across_dp?
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
3 participants
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.
…n situations
What this PR does / why we need it?
Add with-prefill CPU AllReduce to handle D-node recomputation situations.
Does this PR introduce any user-facing change?
How was this patch tested?
gsm8k
http://image.huawei.com/tiny-lts/v1/images/mdstorm/dcbc43b858db666f185d73868f7933fb_1242x502.png
livecodebench
http://image.huawei.com/tiny-lts/v1/images/mdstorm/78a2e9695c3d841870d02c840f032154_1242x502.png
vllmbeachmark
http://image.huawei.com/tiny-lts/v1/images/mdstorm/a4d32f4f2d702cf89854b83ae4d58337_1242x502.png
performance
http://image.huawei.com/tiny-lts/v1/images/mdstorm/38e194a09c3c9ae902a3772f1dca6862_1609x1095.png
http://image.huawei.com/tiny-lts/v1/images/mdstorm/6ffaeb7a6e4672ebbb0b22139b4c72f4_1634x1096.png