[TRTLLM-8650][fix] beam search request validation (#8433) #9228
Conversation
ixlmar
force-pushed
the
chore/cherry-pick-beam-search-request-validation
branch
from
6390620 to
f228828
Compare
|
/bot run |
📝 WalkthroughWalkthroughThe PR refactors request queue processing by renaming and restructuring the request filtering method to explicitly handle special queue items (shutdown, cancel, control signals). Additionally, beam width consistency validation is added to request validation. Test fixtures are updated to use a centralized model kwargs builder, and new parameter validation tests are introduced. Changes
Sequence Diagram(s)sequenceDiagram
participant Queue as Executor Queue
participant Handler as _handle_special_queue_items
participant Processor as Request Processor
Queue->>Handler: new_requests
Note over Handler: Filter special items
alt Shutdown Request
Handler->>Handler: set is_shutdown = True
Handler->>Queue: early terminate
else Canceled Request
Handler->>Handler: record canceled_req_ids
else Control Request
Handler->>Handler: rank 0: accumulate<br/>to request_accumulated
else Regular Request
Handler->>Handler: add to accepted_new_requests
end
Handler->>Processor: return accepted_new_requests
Processor->>Processor: process valid requests
sequenceDiagram
participant Client as Test/Client
participant Validator as _validate_request
participant CheckBeamWidth as Beam Width Check
participant CheckTokenRange as Token Range Check
Client->>Validator: request with sampling_config
Validator->>CheckBeamWidth: beam_width vs max_beam_width
alt beam_width ≠ max_beam_width
CheckBeamWidth->>Validator: raise ValueError
Validator->>Client: error
else beam_width = max_beam_width
CheckBeamWidth->>Validator: valid
Validator->>CheckTokenRange: proceed to model-specific checks
CheckTokenRange->>Validator: validation result
Validator->>Client: success/error
end
Estimated code review effort🎯 3 (Moderate) | ⏱️ ~25 minutes
Pre-merge checks and finishing touches❌ Failed checks (1 warning, 1 inconclusive)
✅ Passed checks (1 passed)
✨ Finishing touches
🧪 Generate unit tests (beta)
Tip 📝 Customizable high-level summaries are now available in beta!You can now customize how CodeRabbit generates the high-level summary in your pull requests — including its content, structure, tone, and formatting.
Example:
Note: This feature is currently in beta for Pro-tier users, and pricing will be announced later. 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. Comment |
|
PR_Github #24780 [ run ] triggered by Bot. Commit: |
![coderabbitai[bot]](https://avatars.githubusercontent.com/in/347564?s=60&v=4){kind=small}
There was a problem hiding this comment.
Actionable comments posted: 0
🧹 Nitpick comments (4)
tensorrt_llm/_torch/pyexecutor/py_executor.py (1)
1593-1602: Beam-width validation is correct and well-placed; consider minor cleanup of the error message literalThe new check cleanly enforces
request.sampling_config.beam_width == self.max_beam_widthbefore any model-specific validation, and it will be triggered exactly once per incomingLlmRequestvia_fetch_and_activate_new_requests, which aligns with the earlier decision to keepsampler.beam_width()lightweight and free of per-batch consistency checks. This is a good place to centralize beam-width consistency. Based on learningsIf you want to address the Ruff
TRY003hint, you could optionally shorten or factor the error text into a constant/custom exception, but that’s cosmetic and not blocking.tensorrt_llm/_torch/pyexecutor/executor_request_queue.py (1)
275-333: Special-queue handling refactor looks sound; shutdown-drop behavior is worth being explicit aboutRouting everything through
_handle_special_queue_itemssimplifies_fetch_and_process_requestsand makes the shutdown/cancel/control paths much clearer:
- Shutdown sets
is_shutdownand stops considering further items in this batch.- Cancel requests only tag
canceled_req_idsand do not enter the waiting queue.- Control requests are isolated into
control_requests, with rank 0 stashing any following items intorequest_accumulated, while other ranks ignore them, which matches the “control must be handled exclusively” contract enforced by the early return whencontrol_requestsis non-empty.One behavioral detail to be aware of: any items that appear after a shutdown request in the same
new_requestsbatch are intentionally dropped rather than queued. That’s consistent with treating shutdown as terminal, but it might be worth documenting in the class docstring or method docstring so callers don’t assume such requests will ever be processed.Also applies to: 485-505
tests/unittest/_torch/executor/test_executor_request_queue.py (1)
478-497: Test correctly exercises special-item handling; setup can be simplifiedThe assertions here are exactly what we need: only the normal request is returned,
is_shutdownflips toTrue, and the canceled ID is tracked incanceled_req_ids.Given that beam-width validation no longer lives in this queue layer, the “avoid beam validation” comment and the
sampling_configdeletion on a bareMockare now redundant. You can simplify the setup to justmock_request = Mock()and drop thedelattrblock/comment without changing behavior.tests/unittest/_torch/sampler/test_beam_search.py (1)
490-593: Parameter-validation tests for beam search are well targeted; consider making regex patterns raw stringsThe new
TestParameterValidationclass:
- Uses its own
fixed_params(max_beam_width4) andmodel_kwargspointing to a TinyLlama checkpoint, gated by@force_ampereand generous timeouts.- Verifies:
- Greedy decoding (
use_beam_search=Falseor omitted) withbest_of > 1raisesValueErrorand does not hang.- A smaller beam width (
best_of=2withmax_beam_width=4) raises aRequestErrorwhose message matches the new beam-width validation, and the engine remains usable afterwards via_check_engine_responds.That gives good coverage of the new validation paths and, importantly, asserts that error-handling paths don’t leave the engine in a bad state.
Minor polish to align with Ruff’s
RUF043hint: thematch=patterns are true regexes (".*...*"), so you can make them raw strings for clarity and to avoid accidental escaping issues, e.g.:with pytest.raises( ValueError, match=r".*Greedy decoding in the LLM API does not allow multiple returns.*", ): ...and similarly for the other
match=arguments.
📜 Review details
Configuration used: Path: .coderabbit.yaml
Review profile: CHILL
Plan: Pro
📒 Files selected for processing (4)
tensorrt_llm/_torch/pyexecutor/executor_request_queue.py(3 hunks)tensorrt_llm/_torch/pyexecutor/py_executor.py(1 hunks)tests/unittest/_torch/executor/test_executor_request_queue.py(2 hunks)tests/unittest/_torch/sampler/test_beam_search.py(6 hunks)
🧰 Additional context used
🧠 Learnings (13)
📓 Common learnings
Learnt from: dcampora
Repo: NVIDIA/TensorRT-LLM PR: 6867
File: tensorrt_llm/_torch/pyexecutor/sampler.py:67-72
Timestamp: 2025-08-13T16:20:37.987Z
Learning: In TensorRT-LLM sampler code, performance is prioritized over additional validation checks. The beam_width helper method intentionally returns the first request's beam_width without validating consistency across all requests to avoid performance overhead from iterating through the entire batch.
📚 Learning: 2025-08-13T16:20:37.987Z
Learnt from: dcampora
Repo: NVIDIA/TensorRT-LLM PR: 6867
File: tensorrt_llm/_torch/pyexecutor/sampler.py:67-72
Timestamp: 2025-08-13T16:20:37.987Z
Learning: In TensorRT-LLM sampler code, performance is prioritized over additional validation checks. The beam_width helper method intentionally returns the first request's beam_width without validating consistency across all requests to avoid performance overhead from iterating through the entire batch.
Applied to files:
tensorrt_llm/_torch/pyexecutor/py_executor.pytests/unittest/_torch/sampler/test_beam_search.py
📚 Learning: 2025-08-19T12:45:11.997Z
Learnt from: amitz-nv
Repo: NVIDIA/TensorRT-LLM PR: 7033
File: tensorrt_llm/_torch/pyexecutor/model_engine.py:0-0
Timestamp: 2025-08-19T12:45:11.997Z
Learning: In tensorrt_llm/_torch/pyexecutor/model_engine.py, DoRA (Delta Orthogonal Rank Adaptation) functionality was removed from the PyTorch flow to eliminate issues with inverted DoRA detection logic. The original is_dora condition was checking if scaling_vec_pointer == 0, which was potentially incorrect.
Applied to files:
tensorrt_llm/_torch/pyexecutor/py_executor.pytests/unittest/_torch/sampler/test_beam_search.py
📚 Learning: 2025-09-29T15:14:28.503Z
Learnt from: amitz-nv
Repo: NVIDIA/TensorRT-LLM PR: 8063
File: tensorrt_llm/lora_manager.py:1080-1112
Timestamp: 2025-09-29T15:14:28.503Z
Learning: In tensorrt_llm/lora_manager.py, when calculating part_sizes for attn_qkv fused LoRA modules, the sizes are correctly multiplied by tp_size because model_config.num_heads and model_config.num_kv_heads are already divided by tp_size (per-TP-rank values), so multiplication is needed to get the original full concatenated dimension size. The interleave_fused_lora_weights_for_tp function provides proper validation with asserts for total size and TP divisibility.
Applied to files:
tensorrt_llm/_torch/pyexecutor/py_executor.py
📚 Learning: 2025-09-29T15:14:28.503Z
Learnt from: amitz-nv
Repo: NVIDIA/TensorRT-LLM PR: 8063
File: tensorrt_llm/lora_manager.py:1080-1112
Timestamp: 2025-09-29T15:14:28.503Z
Learning: In tensorrt_llm/lora_manager.py, when calculating part_sizes for attn_qkv fused LoRA modules, the sizes are correctly multiplied by tp_size because model_config.num_heads and model_config.num_kv_heads are already divided by tp_size (per-TP-rank values), so multiplication is needed to get the original full concatenated dimension size. The interleave_fused_lora_weights_for_tp function provides proper validation.
Applied to files:
tensorrt_llm/_torch/pyexecutor/py_executor.py
📚 Learning: 2025-08-15T06:46:53.813Z
Learnt from: eopXD
Repo: NVIDIA/TensorRT-LLM PR: 6767
File: cpp/tensorrt_llm/batch_manager/kvCacheManager.cpp:0-0
Timestamp: 2025-08-15T06:46:53.813Z
Learning: In the TensorRT-LLM KV cache manager, SWA (Sliding Window Attention) combined with beam search is currently in a broken/non-functional state and is planned for future rework. During preparatory refactoring phases, code related to SWA+beam search may intentionally remain in a non-working state until the broader rework is completed.
Applied to files:
tensorrt_llm/_torch/pyexecutor/py_executor.py
📚 Learning: 2025-08-18T08:42:02.640Z
Learnt from: samuellees
Repo: NVIDIA/TensorRT-LLM PR: 6974
File: tensorrt_llm/serve/scripts/benchmark_dataset.py:558-566
Timestamp: 2025-08-18T08:42:02.640Z
Learning: In TensorRT-LLM's RandomDataset (tensorrt_llm/serve/scripts/benchmark_dataset.py), when using --random-token-ids option, sequence length accuracy is prioritized over semantic correctness for benchmarking purposes. The encode/decode operations should use skip_special_tokens=True and add_special_tokens=False to ensure exact target token lengths.
Applied to files:
tensorrt_llm/_torch/pyexecutor/py_executor.py
📚 Learning: 2025-08-26T06:07:02.166Z
Learnt from: shaharmor98
Repo: NVIDIA/TensorRT-LLM PR: 7231
File: tensorrt_llm/_torch/pyexecutor/_util.py:504-509
Timestamp: 2025-08-26T06:07:02.166Z
Learning: In tensorrt_llm/_torch/pyexecutor/_util.py, when calling model_engine.set_lora_model_config(), pass model_binding_config.mlp_hidden_size directly without multiplying by mapping.tp_size, as the mlp_hidden_size from get_bindings_model_config() is already the per-TP rank value needed for LoRA weight packaging.
Applied to files:
tensorrt_llm/_torch/pyexecutor/py_executor.py
📚 Learning: 2025-08-15T06:46:54.897Z
Learnt from: eopXD
Repo: NVIDIA/TensorRT-LLM PR: 6767
File: cpp/tensorrt_llm/batch_manager/kvCacheManager.cpp:0-0
Timestamp: 2025-08-15T06:46:54.897Z
Learning: In cpp/tensorrt_llm/batch_manager/kvCacheManager.cpp addToken function, newly allocated blocks are unshared by design. The beam search path in addToken (when sequence.getNumTokens() > windowSize) is currently broken/non-functional with SWA, so the block allocation doesn't follow a shared-then-unshared pattern.
Applied to files:
tensorrt_llm/_torch/pyexecutor/py_executor.py
📚 Learning: 2025-07-28T17:06:08.621Z
Learnt from: moraxu
Repo: NVIDIA/TensorRT-LLM PR: 6303
File: tests/integration/test_lists/qa/examples_test_list.txt:494-494
Timestamp: 2025-07-28T17:06:08.621Z
Learning: In TensorRT-LLM testing, it's common to have both CLI flow tests (test_cli_flow.py) and PyTorch API tests (test_llm_api_pytorch.py) for the same model. These serve different purposes: CLI flow tests validate the traditional command-line workflow, while PyTorch API tests validate the newer LLM API backend. Both are legitimate and should coexist.
Applied to files:
tests/unittest/_torch/sampler/test_beam_search.py
📚 Learning: 2025-08-26T09:37:10.463Z
Learnt from: jiaganc
Repo: NVIDIA/TensorRT-LLM PR: 7031
File: tensorrt_llm/bench/dataclasses/configuration.py:90-104
Timestamp: 2025-08-26T09:37:10.463Z
Learning: In TensorRT-LLM's bench configuration, the `get_pytorch_perf_config()` method returns `self.pytorch_config` which is a Dict[str, Any] that can contain default values including `cuda_graph_config`, making the fallback `llm_args["cuda_graph_config"]` safe to use.
Applied to files:
tests/unittest/_torch/sampler/test_beam_search.py
📚 Learning: 2025-08-26T09:37:10.463Z
Learnt from: jiaganc
Repo: NVIDIA/TensorRT-LLM PR: 7031
File: tensorrt_llm/bench/dataclasses/configuration.py:90-104
Timestamp: 2025-08-26T09:37:10.463Z
Learning: In TensorRT-LLM, the `get_pytorch_perf_config()` method returns `self.pytorch_config` which can contain default `cuda_graph_config` values, so `llm_args` may already have this config before the extra options processing.
Applied to files:
tests/unittest/_torch/sampler/test_beam_search.py
📚 Learning: 2025-08-01T15:14:45.673Z
Learnt from: yibinl-nvidia
Repo: NVIDIA/TensorRT-LLM PR: 6506
File: examples/models/core/mixtral/requirements.txt:3-3
Timestamp: 2025-08-01T15:14:45.673Z
Learning: In TensorRT-LLM, examples directory can have different dependency versions than the root requirements.txt file. Version conflicts between root and examples dependencies are acceptable because examples are designed to be standalone and self-contained.
Applied to files:
tests/unittest/_torch/sampler/test_beam_search.py
🧬 Code graph analysis (4)
tensorrt_llm/_torch/pyexecutor/py_executor.py (2)
cpp/tests/unit_tests/multi_gpu/cacheTransceiverTest.cpp (6)
request(893-943)request(893-893)request(945-952)request(945-945)request(954-1005)request(954-954)tensorrt_llm/_torch/pyexecutor/sampler.py (1)
beam_width(133-136)
tests/unittest/_torch/executor/test_executor_request_queue.py (1)
tensorrt_llm/_torch/pyexecutor/executor_request_queue.py (1)
_handle_special_queue_items(485-504)
tensorrt_llm/_torch/pyexecutor/executor_request_queue.py (1)
tensorrt_llm/_torch/pyexecutor/llm_request.py (2)
append(101-127)append(192-209)
tests/unittest/_torch/sampler/test_beam_search.py (2)
tensorrt_llm/executor/utils.py (1)
RequestError(76-77)tensorrt_llm/_torch/models/checkpoints/hf/checkpoint_loader.py (1)
HfCheckpointLoader(19-75)
🪛 Ruff (0.14.4)
tensorrt_llm/_torch/pyexecutor/py_executor.py
1597-1600: Avoid specifying long messages outside the exception class
(TRY003)
tests/unittest/_torch/sampler/test_beam_search.py
536-536: Pattern passed to match= contains metacharacters but is neither escaped nor raw
(RUF043)
560-560: Pattern passed to match= contains metacharacters but is neither escaped nor raw
(RUF043)
582-582: Pattern passed to match= contains metacharacters but is neither escaped nor raw
(RUF043)
⏰ 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 (1)
tests/unittest/_torch/sampler/test_beam_search.py (1)
21-23: Centralizing LLM construction viamodel_kwargs+_build_llmimproves test clarityUsing a
model_kwargsfixture plus_build_llmto assemble theLLMinstance keeps the beam-search tests DRY and makes it easy to swap between the dummy checkpoint setup and real checkpoints in other fixtures. Bothllmandllm_cuda_graphnow share the same core configuration (batch size, seq length,max_beam_width), which should reduce drift between test variants.This structure also plays nicely with the new parameter-validation tests that reuse
_build_llmagainst a “real” TinyLlama checkpoint.Also applies to: 269-313
|
PR_Github #24780 [ run ] completed with state |
|
/bot run |
|
PR_Github #24882 [ run ] triggered by Bot. Commit: |
ixlmar
force-pushed
the
chore/cherry-pick-beam-search-request-validation
branch
from
f228828 to
6fdd356
Compare
|
/bot run |
ixlmar
force-pushed
the
chore/cherry-pick-beam-search-request-validation
branch
from
6fdd356 to
922fd50
Compare
Signed-off-by: ixlmar <[email protected]>
ixlmar
force-pushed
the
chore/cherry-pick-beam-search-request-validation
branch
from
922fd50 to
bc65d51
Compare
|
/bot run |
|
PR_Github #24891 [ run ] triggered by Bot. Commit: |
|
PR_Github #24882 [ run ] completed with state |
|
PR_Github #24892 [ run ] triggered by Bot. Commit: |
|
PR_Github #24891 [ run ] completed with state |
|
PR_Github #24892 [ run ] completed with state |
|
/bot run |
|
PR_Github #24901 [ run ] triggered by Bot. Commit: |
|
PR_Github #24901 [ run ] completed with state |
|
/bot run |
|
PR_Github #24912 [ run ] triggered by Bot. Commit: |
|
PR_Github #24912 [ run ] completed with state |
|
/bot run --disable-fail-fast |
|
PR_Github #25004 [ run ] triggered by Bot. Commit: |
|
/bot run --disable-fail-fast |
|
PR_Github #25013 [ run ] triggered by Bot. Commit: |
|
PR_Github #25013 [ run ] completed with state |
|
/bot run --disable-fail-fast |
|
PR_Github #25030 [ run ] triggered by Bot. Commit: |
|
PR_Github #25030 [ run ] completed with state |
|
/bot run --disable-fail-fast |
|
PR_Github #25165 [ run ] triggered by Bot. Commit: |
|
PR_Github #25165 [ run ] completed with state |
|
/bot run --disable-fail-fast |
|
PR_Github #25214 [ run ] triggered by Bot. Commit: |
|
PR_Github #25214 [ run ] completed with state |
|
/bot run --disable-fail-fast |
|
PR_Github #25348 [ run ] triggered by Bot. Commit: |
|
PR_Github #25348 [ run ] completed with state |
4 participants
Description
Cherry-picks #8433
Test Coverage
Tests are included.
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
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
/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.
Summary by CodeRabbit
Release Notes
Bug Fixes
Tests
Refactor