feat: support multi_stream_parallel and modify batch sampler. [1/4] #110
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
guojinrong-nn
changed the title
liutongxuan
changed the title
LMX-xin
reviewed
Sep 9, 2025
Collaborator
There was a problem hiding this comment.
Submodule xllm_ops now uses a8e8b1, no need to fallback to 98b515
guojinrong-nn
force-pushed
the
feat_prefill_multi_streams_parallel
branch
2 times, most recently
from
7dd5e35 to
c6e0d53
Compare
yq33victor
reviewed
Sep 9, 2025
guojinrong-nn
force-pushed
the
feat_prefill_multi_streams_parallel
branch
from
9b82d3f to
f883018
Compare
yq33victor
approved these changes
Sep 10, 2025
liutongxuan
approved these changes
Sep 10, 2025
yiming-l21
pushed a commit
to yiming-l21/xllm
that referenced
this pull request
Sep 11, 2025
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
4 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Outdated suggestions cannot be applied.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.
Notice: this is the part-1/4 of multi_stream_parallel feature.
all_seqs_in_prefill_in classBatch, to ensure the whole batch, all sequences in prefill or chunked prefill stage, to use this feature. If there is decode sequence, will not use this feature.concatmethod in classSamplingParameters, we split two micro batched to execute model execution individually, after this, will execute the sample operation together, to avoid extra small kernel launch.split()inxllm/core/framework/batch/batch.h.set_dp_rank()logic inxllm/core/framework/batch/batch_factory.cpp, dp_rank is already set in block manager.