Repository navigation
[EC] Automatically enable embedding inputs on EC/KV consumers - #56379
Merged
Merged
Conversation
Assisted-by: OpenAI Codex <codex@openai.com> Signed-off-by: Tianyu Guo <guoty@inferact.ai>
Assisted-by: OpenAI Codex <codex@openai.com> Signed-off-by: Tianyu Guo <guoty@inferact.ai>
gty111
marked this pull request as ready for review
September 11, 2026 03:36
gty111
requested review from
ProExpertProg,
WoosukKwon,
houseroad,
mgoin,
robertgshaw2-redhat,
tlrmchlsmth,
yewentao256 and
youkaichao
as code owners
September 11, 2026 03:36
![claude[bot]](https://avatars.githubusercontent.com/in/1236702?s=60&v=4){kind=small}
Isotr0py
approved these changes
Sep 11, 2026
Isotr0py
enabled auto-merge (squash)
September 11, 2026 03:41
github-actions
Bot
added
the
ready
Member
|
/ci run |
|
✅ Triggered Buildkite CI #88239 for commit |
ItsRoy69
pushed a commit
to ItsRoy69/vllm
that referenced
this pull request
Sep 15, 2026
…roject#56379) Signed-off-by: Tianyu Guo <guoty@inferact.ai>
michael-gpu
pushed a commit
to practical-inference/vllm-ampere-extended
that referenced
this pull request
Oct 20, 2026
…roject#56379) Signed-off-by: Tianyu Guo <guoty@inferact.ai>
ChingTsai
added a commit
to ChingTsai/vllm
that referenced
this pull request
Oct 21, 2026
…-model-only mode _resolve_mm_embedding_inputs() turns on allow_missing_mm_embeddings and enable_mm_embeds on every EC/KV consumer (vllm-project#56379). With --language-model-only, all modality limits are 0, but enable_mm_embeds makes supports_multimodal_inputs() return True again, so the scheduler, frontend and workers on a text-only KV consumer each build a MultiModalBudget and the full HF processor, which loads a second copy of the tokenizer. A language-model-only deployment never receives multimodal inputs, so skip both flags there. Assisted-by: Claude Signed-off-by: Jimmy Tsai <jimmytsai@google.com>
ChingTsai
added a commit
to ChingTsai/vllm
that referenced
this pull request
Oct 21, 2026
…-model-only mode _resolve_mm_embedding_inputs() turns on allow_missing_mm_embeddings and enable_mm_embeds on every EC/KV consumer (vllm-project#56379). With --language-model-only, all modality limits are 0, but enable_mm_embeds makes supports_multimodal_inputs() return True again, so the scheduler, frontend and workers on a text-only KV consumer each build a MultiModalBudget and the full HF processor, which loads a second copy of the tokenizer. A language-model-only deployment never receives multimodal inputs, so skip both flags there. Assisted-by: Claude Signed-off-by: Jimmy Tsai <jimmytsai@google.com>
6 tasks done
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
Automatically enable
enable_mm_embedsfor EC/KV consumers alongsideallow_missing_mm_embeddings, so EPD rewrite requests are accepted without an explicit--enable-mm-embeds. Other roles are unchanged.Consumer endpoints must remain trusted, as with explicitly enabling embedding inputs.
Extracted from #54176; the configuration change has been removed there. No other matching open PR was found.
Validation
.venv/bin/python -m pytest tests/config/test_multimodal_config.py -q: 46 passed.AI assistance was used. Draft pending human review and validation.