perf: remove trtllm postprocessing workers from the args as post processing workers are not effective in dynamo

[richardhuo-nv]

Summary

• Add _strip_postprocess_workers() helper in llm_worker.py that removes num_postprocess_workers from TRT-LLM engine args before the engine is initialized, and emits a WARNING-level log if the value was > 0.
• Call _strip_postprocess_workers() after all engine arg sources (--extra-engine-args YAML and --override-engine-args JSON) have been merged, so it covers every code path.
• Add 4 unit tests: 3 direct tests of the helper (positive value, zero, absent) and 1 integration test that verifies the setting is absent from engine_args passed to get_llm_engine when supplied via an --extra-engine-args YAML file (the pattern used in the kimi-k2.5, deepseek-v32-fp4, and gpt-oss-120b recipes).

Why: TRT-LLM's num_postprocess_workers spawns workers for its own post-processing pipeline, which conflicts with Dynamo's response-streaming path. When the setting is forwarded to TRT-LLM it has no useful effect and can occupy extra cpus resources causing perf degradation.

Validation

• pytest -m unit components/src/dynamo/trtllm/tests/test_trtllm_unit.py -k "postprocess" — all 4 new tests pass.
• Manually verified the warning fires on a config containing num_postprocess_workers: 4 and is absent when the key is not set.

Related Issues

🚫 This PR is NOT linked to an issue:

• Confirmed — no related issue

---

Summary by CodeRabbit

• Bug Fixes
  • Prevented unsupported post-processing worker settings from being passed to TensorRT-LLM engine initialization.
  • Added a warning when a positive post-processing worker count is configured.
  • Preserved valid behavior when the setting is zero or not provided.

[devin-ai-integration]

Devin Review found 2 potential issues.

[coderabbitai]

Walkthrough

The TensorRT-LLM worker removes num_postprocess_workers before engine configuration. Positive values produce a warning. Tests cover helper behavior and asynchronous initialization.

Changes

TensorRT-LLM postprocess cleanup

Layer / File(s)	Summary
Strip postprocess worker arguments components/src/dynamo/trtllm/workers/llm_worker.py	The worker removes num_postprocess_workers after engine-argument synchronization. It warns when the removed value is positive.
Validate argument stripping components/src/dynamo/trtllm/tests/test_trtllm_unit.py	Tests cover positive, zero, and absent values. An asynchronous initialization test verifies removal from YAML extra engine arguments before engine creation.

Estimated code review effort: 2 (Simple) | ~10 minutes

🚥 Pre-merge checks | ✅ 5

✅ Passed checks (5 passed)

Check name	Status	Explanation
Docstring Coverage	✅ Passed	No functions found in the changed files to evaluate docstring coverage. Skipping docstring coverage check.
Linked Issues check	✅ Passed	Check skipped because no linked issues were found for this pull request.
Out of Scope Changes check	✅ Passed	Check skipped because no linked issues were found for this pull request.
Title check	✅ Passed	The title clearly states that TRT-LLM post-processing workers are removed from engine arguments to improve Dynamo performance.
Description check	✅ Passed	The description explains the change, rationale, tests, validation, and required no-issue confirmation, but it omits the reviewer-start section.

---

_{Comment @coderabbitai help to get the list of available commands.}

[krishung5]

I see that the default value of num_postprocess_workers is 0, so stripping directly sounds good, thanks!