[None][fix] fix chat hamony api token ids propagation - #14451
Closed
reasonsolo wants to merge 3 commits into
Closed
Conversation
When `/v1/chat/completions` is served with the harmony adapter (gpt-oss family), the gen-worker handler unconditionally re-runs `HarmonyAdapter.openai_to_harmony_tokens` on the incoming messages, producing a second tokenization of work already done on the ctx worker. Lizhi's NVIDIA#14420 propagates `prompt_token_ids` from the ctx-worker response onto the gen-worker request, but the gen handler still ignored them in this code path. Use the propagated `prompt_token_ids` when present and fall back to the harmony adapter only for the agg path. Measured impact on RWLT (gpt-oss-120b, multi-turn coding workload): `gen_preproc_ms` p50 drops from 32.7 ms to 3.5 ms and the disagg-vs-agg TTFT gap p50 narrows from +140.7 ms to +112.4 ms. Signed-off-by: Balaram Buddharaju <169953907+brb-nv@users.noreply.github.com>
Two related gaps prevented `GET /perf_metrics` from returning per- request timing data on the chat endpoints, despite server-level `return_perf_metrics: true` and `TRTLLM_KVCACHE_TIME_OUTPUT_PATH` being set: 1. The chat handlers (`openai_chat`, `chat_harmony`) never propagated `return_perf_metrics` onto the per-request `SamplingParams`, so the pyexecutor's `request.update_perf_metrics` call in `py_executor.py` was a no-op for these requests. Add the same `TRTLLM_KVCACHE_TIME_ OUTPUT_PATH` gate already used by `openai_completion`. 2. The harmony streaming generator (`create_streaming_generator` inside `chat_harmony`) never called `_extract_metrics`, which is the only site that appends an item to `self.perf_metrics`. It also never stamped `raw_request.state.server_first_token_time`. Mirror the structure of `chat_stream_generator` in `openai_chat`: pull the first response with `anext`, stamp first-token time, stream the rest, then call `_extract_metrics` after `[DONE]`. With both fixes, harmony chat requests (gpt-oss family) populate `/perf_metrics` with correct `server_arrival_time`, `server_first_token_time`, `first_scheduled_time`, `first_token_time`, `last_token_time`, and KV-cache block counters per request, which the disaggregated proxy also collects via `/perf_metrics`. Signed-off-by: Balaram Buddharaju <169953907+brb-nv@users.noreply.github.com>
reasonsolo
requested review from
dongxuy04 and
hchings
and removed request for
a team
May 22, 2026 08:23
…te in test _make_gen_request did not set is_disagg_generation_transmission_complete, causing _sync_disagg_generation_trans_complete_draft_tokens to try iterating over a Mock object instead of skipping non-transmission-complete requests. Signed-off-by: Lizhi Zhou <1432185+reasonsolo@users.noreply.github.com>
reasonsolo
force-pushed
the
fix_router_test_v3
branch
from
May 22, 2026 08:25
972da6f to
6f8ceeb
Compare
Collaborator
Author
|
/bot run --disable-fail-fast |
Collaborator
|
PR_Github #49900 [ run ] triggered by Bot. Commit: |
Collaborator
|
PR_Github #49900 [ run ] completed with state
|
Collaborator
Author
|
Balaram's PR #14506 |
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
3 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.
@coderabbitai summary
Description
Test Coverage
PR Checklist
Please review the following before submitting your PR:
PR description clearly explains what and why. If using CodeRabbit's summary, please make sure it makes sense.
PR Follows TRT-LLM CODING GUIDELINES to the best of your knowledge.
Test cases are provided for new code paths (see test instructions)
If PR introduces API changes, an appropriate PR label is added - either
api-compatibleorapi-breaking. Forapi-breaking, includeBREAKINGin the PR title.Any new dependencies have been scanned for license and vulnerabilities
CODEOWNERS updated if ownership changes
Documentation updated as needed
Update tava architecture diagram if there is a significant design change in PR.
The reviewers assigned automatically/manually are appropriate for the PR.
Please check this after reviewing the above items as appropriate for this PR.
GitHub Bot Help
To see a list of available CI bot commands, please comment
/bot help.