Gate multimodal feature transport by model capability - #33653
Merged
Merged
Conversation
Collaborator
Author
|
/tag-and-rerun-ci |
b8zhong
approved these changes
Aug 6, 2026
kfhfar
pushed a commit
to kfhfar/sglang
that referenced
this pull request
Aug 7, 2026
Xia-Weiwen
pushed a commit
to Xia-Weiwen/sglang
that referenced
this pull request
Aug 10, 2026
saturn-acc
pushed a commit
to saturn-acc/sglang
that referenced
this pull request
Aug 16, 2026
jakki-amd
pushed a commit
to jakki-amd/sglang
that referenced
this pull request
Sep 9, 2026
Atituiset
pushed a commit
to Atituiset/sglang
that referenced
this pull request
Sep 10, 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
2 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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
Summary
#30904 added multimodal feature transport, and #32541 enabled automatic CUDA IPC on single-node CUDA, unintentionally selecting it for text-only models and logging a misleading 1 GiB reservation. The pool was not actually allocated because these models do not create a multimodal processor.
This gates automatic CUDA IPC on model capability. Text-only models now stay on CPU; multimodal and explicit transport behavior is unchanged.
CI States
Latest PR Test (Base): ✅ Run #30979280229
Latest PR Test (Extra): ❌ Run #30979280024