Repository navigation
[Bugfix] Restore KV cache metadata GET method for external event consumer - #56925
Conversation
82f292c to
814ea79
Compare
|
/ci run |
|
❌ This PR is 5 commits behind upstream |
|
✅ @wzhao18, CI is now available for this PR.
|
Signed-off-by: Wei Zhao <51183510+wzhao18@users.noreply.github.com>
814ea79 to
c0004be
Compare
|
/ci run |
|
✅ Triggered Buildkite CI #89454 for commit |
|
/ci run |
|
❌ This PR is 1 commit behind upstream |
|
/ci run |
|
✅ Triggered Buildkite CI #89465 for commit |
|
@njhill @pavanimajety ready for merge. |
|
/ci run |
|
✅ Triggered Buildkite CI #89719 for commit |
…umer (vllm-project#56925) Signed-off-by: Wei Zhao <51183510+wzhao18@users.noreply.github.com>
Purpose
get_kv_cache_group_metadatais deleted by #53941 as dead code but is actually used by external event consumers like Dynamo. The PR brings the function call back.In addition, this PR fixes Dynamo KV-aware routing for DCP by reporting the effective scheduler KV block size used by event emission. Kimi K3 DCP8 emitted 12,288-token MLA blocks while metadata advertised 1,536, so Dynamo rejected those events. Testing shows it changed routing from 0/4 cache-hit predictions to 4/4 at 75% expected prefix match, with all four requests routed to the seeded prefill worker.
Test Plan
Test Result
Essential Elements of an Effective PR Description Checklist
supported_models.mdandexamplesfor a new model.BEFORE SUBMITTING, PLEASE READ https://docs.vllm.ai/en/latest/contributing (anything written below this line will be removed by GitHub Actions)