[TRTLLM-7779][feat] Support multiple postprocess workers for chat completions API #7508
Conversation
…pletions API Signed-off-by: Junyi Xu <[email protected]>
JunyiXu-nv
requested review from
kaiyux,
LinPoly,
dongfengy and
juney-nvidia
|
/bot run |
|
PR_Github #17536 [ run ] triggered by Bot |
📝 WalkthroughWalkthroughThe Harmony adapter is refactored to a shared-instance model, replacing generator-based streaming with a stateful, token-driven pipeline and explicit inputs/outputs. OpenAI server now routes through new post-processing hooks that call the adapter via get_harmony_adapter(). New ChatCompletionPostprocArgs and Harmony post-processors orchestrate both streaming and non-streaming paths. Changes
Sequence Diagram(s)sequenceDiagram
autonumber
participant Client
participant OpenAIServer as OpenAI Server
participant LLM as LLM Engine
participant Postproc as Post-processors
participant Adapter as HarmonyAdapter
Client->>OpenAIServer: ChatCompletion request
OpenAIServer->>Postproc: ChatCompletionPostprocArgs.from_request()
OpenAIServer->>LLM: Generate (with _postproc_params when enabled)
alt Streaming
LLM-->>OpenAIServer: GenerationResult chunks (outputs, _done, request_id)
OpenAIServer->>Postproc: chat_harmony_streaming_post_processor(rsp, args)
Postproc->>Adapter: handle_streaming_response(tools, tool_choice, outputs, model, request_id, done, num_prompt_tokens)
Adapter-->>Postproc: OpenAI streaming chunks
Postproc-->>OpenAIServer: chunks
OpenAIServer-->>Client: data: chunks ... data: [DONE]
else Non-streaming
LLM-->>OpenAIServer: Final GenerationResult (outputs, num_prompt_tokens)
OpenAIServer->>Postproc: chat_harmony_post_processor(rsp, args)
Postproc->>Adapter: handle_non_streaming_response(tools, tool_choice, outputs, model, num_prompt_tokens)
Adapter-->>Postproc: ChatCompletionResponse
Postproc-->>OpenAIServer: response
OpenAIServer-->>Client: ChatCompletionResponse
end
Estimated code review effort🎯 4 (Complex) | ⏱️ ~60 minutes Possibly related PRs
Suggested reviewers
✨ Finishing Touches
🧪 Generate unit tests
Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out. 🪧 TipsChatThere are 3 ways to chat with CodeRabbit:
SupportNeed help? Create a ticket on our support page for assistance with any issues or questions. CodeRabbit Commands (Invoked using PR/Issue comments)Type Other keywords and placeholders
Status, Documentation and Community
|
![coderabbitai[bot]](https://avatars.githubusercontent.com/in/347564?s=60&v=4){kind=small}
There was a problem hiding this comment.
Actionable comments posted: 7
🧹 Nitpick comments (2)
tensorrt_llm/serve/harmony_adapter.py (1)
1656-1665: Add return type annotation to helper functionThe
_create_usage_infohelper function should have a return type annotation.-def _create_usage_info(num_prompt_tokens, outputs) -> UsageInfo: +def _create_usage_info(num_prompt_tokens: int, outputs: List[Any]) -> UsageInfo:tensorrt_llm/serve/openai_server.py (1)
707-729: Extract common response creation logicThe
create_harmony_responseandcreate_streaming_generatorfunctions duplicate logic from the originalopenai_chatmethod. Consider extracting this to a shared utility.
📜 Review details
Configuration used: Path: .coderabbit.yaml
Review profile: CHILL
Plan: Pro
💡 Knowledge Base configuration:
- MCP integration is disabled by default for public repositories
- Jira integration is disabled by default for public repositories
- Linear integration is disabled by default for public repositories
You can enable these sources in your CodeRabbit configuration.
📒 Files selected for processing (3)
tensorrt_llm/serve/harmony_adapter.py(5 hunks)tensorrt_llm/serve/openai_server.py(4 hunks)tensorrt_llm/serve/postprocess_handlers.py(2 hunks)
🧰 Additional context used
📓 Path-based instructions (4)
**/*
📄 CodeRabbit inference engine (CODING_GUIDELINES.md)
Filenames compiled into a target must be case-insensitively unique
Files:
tensorrt_llm/serve/postprocess_handlers.pytensorrt_llm/serve/harmony_adapter.pytensorrt_llm/serve/openai_server.py
**/*.{h,hpp,hh,hxx,cc,cpp,cxx,cu,cuh,py}
📄 CodeRabbit inference engine (CODING_GUIDELINES.md)
Use spaces, not tabs; indent 4 spaces
Files:
tensorrt_llm/serve/postprocess_handlers.pytensorrt_llm/serve/harmony_adapter.pytensorrt_llm/serve/openai_server.py
**/*.py
📄 CodeRabbit inference engine (CODING_GUIDELINES.md)
**/*.py: Code must target Python 3.8+
Indent with 4 spaces; do not use tabs (Python)
Maintain module namespace on import: prefer from package.subpackage import foo; use foo.Symbol()
Python filenames use snake_case
Python class names use PascalCase
Python functions and methods use snake_case
Python local variables use snake_case; if starting with a number concept, prefix with k (e.g., k_99th_percentile)
Python global variables use G_ prefix with UPPER_SNAKE_CASE
Python constants use UPPER_SNAKE_CASE
Avoid shadowing variables from outer scopes
Initialize all externally visible class members in init
For public interfaces, prefer docstrings over comments; comments should be for in-function or file-local interfaces
Use Google-style docstrings for classes and functions (Sphinx-parsable)
Document attributes and variables inline with docstrings immediately after assignment
Avoid reflection when a non-reflective approach suffices
Limit except clauses to specific exceptions where possible
When using try/except for duck-typing, keep try body minimal and move logic to else
Files:
tensorrt_llm/serve/postprocess_handlers.pytensorrt_llm/serve/harmony_adapter.pytensorrt_llm/serve/openai_server.py
**/*.{cpp,cc,cxx,h,hpp,hh,hxx,cu,cuh,py}
📄 CodeRabbit inference engine (CODING_GUIDELINES.md)
Prepend NVIDIA copyright header (current year) to all source files
Files:
tensorrt_llm/serve/postprocess_handlers.pytensorrt_llm/serve/harmony_adapter.pytensorrt_llm/serve/openai_server.py
🧬 Code graph analysis (3)
tensorrt_llm/serve/postprocess_handlers.py (4)
tensorrt_llm/serve/harmony_adapter.py (2)
handle_non_streaming_response(1571-1620)handle_streaming_response(1488-1568)tensorrt_llm/executor/result.py (3)
request_id(521-522)GenerationResult(485-634)outputs(197-212)tensorrt_llm/serve/openai_protocol.py (6)
UsageInfo(69-72)CompletionResponse(143-152)ChatCompletionToolsParam(476-478)ChatCompletionNamedToolChoiceParam(485-487)ChatCompletionRequest(490-672)ChatCompletionResponse(431-440)tensorrt_llm/_utils.py (1)
nvtx_range_debug(872-896)
tensorrt_llm/serve/harmony_adapter.py (2)
tensorrt_llm/executor/result.py (1)
token_ids_diff(140-141)tensorrt_llm/serve/openai_protocol.py (6)
ChatCompletionStreamResponse(461-467)ChatCompletionResponseStreamChoice(452-458)DeltaMessage(443-449)ChatCompletionResponseChoice(417-428)ChatMessage(358-363)UsageInfo(69-72)
tensorrt_llm/serve/openai_server.py (4)
tensorrt_llm/llmapi/llm.py (4)
RequestOutput(46-86)tokenizer(691-695)tokenizer(698-699)generate_async(317-450)tensorrt_llm/executor/postproc_worker.py (1)
PostprocParams(37-39)tensorrt_llm/executor/result.py (4)
aresult(576-584)outputs(197-212)request_id(521-522)prompt_token_ids(525-526)tensorrt_llm/serve/postprocess_handlers.py (6)
ChatCompletionPostprocArgs(366-379)from_request(56-67)from_request(277-284)from_request(374-379)chat_harmony_streaming_post_processor(393-403)chat_harmony_post_processor(382-390)
🪛 Ruff (0.12.2)
tensorrt_llm/serve/openai_server.py
712-714: 1 blank line required between summary line and description
(D205)
⏰ Context from checks skipped due to timeout of 90000ms. You can increase the timeout in your CodeRabbit configuration to a maximum of 15 minutes (900000ms). (1)
- GitHub Check: Pre-commit Check
🔇 Additional comments (7)
tensorrt_llm/serve/harmony_adapter.py (2)
1487-1492: Module-level adapter instance looks goodThe singleton pattern with
serve_harmony_adapterandget_harmony_adapter()properly centralizes the adapter instance, supporting the refactored architecture for multiple postprocess workers.
1519-1519: Use existing function for usage info creationOn line 1519, you're calling
_create_usage_infowhich is defined later on line 1656. However, on line 1608, you're calling the same function in non-streaming response. Consider moving the function definition earlier or ensure it's accessible.tensorrt_llm/serve/postprocess_handlers.py (1)
357-372: Well-structured post-processing arguments classThe
ChatCompletionPostprocArgsdataclass properly encapsulates the required parameters for post-processing, with a clean factory method following the existing pattern.tensorrt_llm/serve/openai_server.py (4)
61-62: Clean import refactoringGood consolidation of the harmony adapter imports under a single import statement.
734-734: Singleton adapter access pattern is appropriateUsing
get_harmony_adapter()ensures consistent access to the global adapter instance, which is appropriate for the multi-worker architecture.
769-774: Proper integration of new post-processing handlersThe setup correctly wires the new harmony-specific post-processors based on streaming mode, maintaining consistency with the existing pattern.
786-788: Conditional prompt token assignment is correctOnly setting
num_prompt_tokenswhen postproc workers are disabled ensures the value is available when needed for local post-processing.
|
PR_Github #17536 [ run ] completed with state |
Signed-off-by: Junyi Xu <[email protected]>
|
/bot run |
|
PR_Github #17543 [ run ] triggered by Bot |
Signed-off-by: Junyi Xu <[email protected]>
|
/bot run |
|
PR_Github #17546 [ run ] triggered by Bot |
|
PR_Github #17543 [ run ] completed with state |
|
PR_Github #17546 [ run ] completed with state |
Signed-off-by: Junyi Xu <[email protected]>
|
/bot run |
|
PR_Github #17590 [ run ] triggered by Bot |
Signed-off-by: Junyi Xu <[email protected]>
|
/bot run |
|
PR_Github #17602 [ run ] triggered by Bot |
|
PR_Github #17590 [ run ] completed with state |
|
PR_Github #17602 [ run ] completed with state |
|
/bot run |
|
PR_Github #17859 [ run ] triggered by Bot |
|
PR_Github #17859 [ run ] completed with state |
Summary by CodeRabbit
New Features
Refactor
Description
This PR is enabling multiple postprocess workers in chat completions API, to minize host side benchmarking latency.
Experiment data with
benchmark_serving:The overall performance has an average 16.98% improvement on user throughput tps and 18.42% improvement on output throughput tps.
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)
Any new dependencies have been scanned for license and vulnerabilities
CODEOWNERS updated if ownership changes
Documentation updated as needed
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
/bot [-h] ['run', 'kill', 'skip', 'reuse-pipeline'] ...Provide a user friendly way for developers to interact with a Jenkins server.
Run
/bot [-h|--help]to print this help message.See details below for each supported subcommand.
run [--reuse-test (optional)pipeline-id --disable-fail-fast --skip-test --stage-list "A10-PyTorch-1, xxx" --gpu-type "A30, H100_PCIe" --test-backend "pytorch, cpp" --add-multi-gpu-test --only-multi-gpu-test --disable-multi-gpu-test --post-merge --extra-stage "H100_PCIe-TensorRT-Post-Merge-1, xxx" --detailed-log --debug(experimental)]Launch build/test pipelines. All previously running jobs will be killed.
--reuse-test (optional)pipeline-id(OPTIONAL) : Allow the new pipeline to reuse build artifacts and skip successful test stages from a specified pipeline or the last pipeline if no pipeline-id is indicated. If the Git commit ID has changed, this option will be always ignored. The DEFAULT behavior of the bot is to reuse build artifacts and successful test results from the last pipeline.--disable-reuse-test(OPTIONAL) : Explicitly prevent the pipeline from reusing build artifacts and skipping successful test stages from a previous pipeline. Ensure that all builds and tests are run regardless of previous successes.--disable-fail-fast(OPTIONAL) : Disable fail fast on build/tests/infra failures.--skip-test(OPTIONAL) : Skip all test stages, but still run build stages, package stages and sanity check stages. Note: Does NOT update GitHub check status.--stage-list "A10-PyTorch-1, xxx"(OPTIONAL) : Only run the specified test stages. Examples: "A10-PyTorch-1, xxx". Note: Does NOT update GitHub check status.--gpu-type "A30, H100_PCIe"(OPTIONAL) : Only run the test stages on the specified GPU types. Examples: "A30, H100_PCIe". Note: Does NOT update GitHub check status.--test-backend "pytorch, cpp"(OPTIONAL) : Skip test stages which don't match the specified backends. Only support [pytorch, cpp, tensorrt, triton]. Examples: "pytorch, cpp" (does not run test stages with tensorrt or triton backend). Note: Does NOT update GitHub pipeline status.--only-multi-gpu-test(OPTIONAL) : Only run the multi-GPU tests. Note: Does NOT update GitHub check status.--disable-multi-gpu-test(OPTIONAL) : Disable the multi-GPU tests. Note: Does NOT update GitHub check status.--add-multi-gpu-test(OPTIONAL) : Force run the multi-GPU tests in addition to running L0 pre-merge pipeline.--post-merge(OPTIONAL) : Run the L0 post-merge pipeline instead of the ordinary L0 pre-merge pipeline.--extra-stage "H100_PCIe-TensorRT-Post-Merge-1, xxx"(OPTIONAL) : Run the ordinary L0 pre-merge pipeline and specified test stages. Examples: --extra-stage "H100_PCIe-TensorRT-Post-Merge-1, xxx".--detailed-log(OPTIONAL) : Enable flushing out all logs to the Jenkins console. This will significantly increase the log volume and may slow down the job.--debug(OPTIONAL) : Experimental feature. Enable access to the CI container for debugging purpose. Note: Specify exactly one stage in thestage-listparameter to access the appropriate container environment. Note: Does NOT update GitHub check status.For guidance on mapping tests to stage names, see
docs/source/reference/ci-overview.mdand the
scripts/test_to_stage_mapping.pyhelper.kill
killKill all running builds associated with pull request.
skip
skip --comment COMMENTSkip testing for latest commit on pull request.
--comment "Reason for skipping build/test"is required. IMPORTANT NOTE: This is dangerous since lack of user care and validation can cause top of tree to break.reuse-pipeline
reuse-pipelineReuse a previous pipeline to validate current commit. This action will also kill all currently running builds associated with the pull request. IMPORTANT NOTE: This is dangerous since lack of user care and validation can cause top of tree to break.