Add bilinear_pos_embed triton kernel and cache#40
Merged
wangshangsam merged 2 commits intomlperf-inf-mm-q3vl-v6.0from Feb 8, 2026
Merged
Add bilinear_pos_embed triton kernel and cache#40wangshangsam merged 2 commits intomlperf-inf-mm-q3vl-v6.0from
wangshangsam merged 2 commits intomlperf-inf-mm-q3vl-v6.0from
Conversation
Signed-off-by: Zhanda <zhandazhu@gmail.com>
Signed-off-by: Zhanda <zhandazhu@gmail.com>
wangshangsam
approved these changes
Feb 8, 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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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
Caching mechanism is borrowed from the PR #33 of @b-mu.
How to enable
The Triton kernel is off by default. To enable:
export VLLM_USE_TRITON_POS_EMBED=1Pos emb cache size can be controlled by
Files changed
vllm/envs.py-- register VLLM_USE_TRITON_POS_EMBEDvllm/model_executor/models/qwen3_vl.py-- Triton kernel, cache infrastructure, warmup grid listPlease review after the next PR on FP8 attn is up. Will run validations of these two PRs together.