[None][ci] move unittests to sub-directories #6635
Conversation
📝 WalkthroughWalkthroughRefactored FlashInfer availability into a new Changes
Sequence Diagram(s)sequenceDiagram
participant Consumer as Module (attention/custom_ops/modules)
participant Utils as flashinfer_utils
participant FlashInfer as flashinfer (optional)
Consumer->>Utils: import IS_FLASHINFER_AVAILABLE, ENABLE_PDL
Utils->>Utils: get_env_enable_pdl() / read TRTLLM_ENABLE_PDL
alt Non-Windows
Utils->>FlashInfer: try import flashinfer
FlashInfer-->>Utils: success / ImportError
Utils->>Consumer: set IS_FLASHINFER_AVAILABLE True/False
else Windows
Utils->>Consumer: set IS_FLASHINFER_AVAILABLE False
end
Consumer->>Consumer: if IS_FLASHINFER_AVAILABLE: import/use flashinfer-backed ops (attn/mla/rope/rmsnorm) else use pure-PyTorch fallbacks
Estimated code review effort🎯 3 (Moderate) | ⏱️ ~25 minutes Possibly related PRs
Suggested labels
Suggested reviewers
Tip 🔌 Remote MCP (Model Context Protocol) integration is now available!Pro plan users can now connect to remote MCP servers from the Integrations page. Connect with popular remote MCPs such as Notion and Linear to add more context to your reviews and chats. ✨ 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
|
|
/bot run |
![coderabbitai[bot]](https://avatars.githubusercontent.com/in/347564?s=60&v=4){kind=small}
There was a problem hiding this comment.
Actionable comments posted: 3
🧹 Nitpick comments (1)
tests/integration/test_lists/test-db/l0_h100.yml (1)
17-26: Enumerating every sub-directory hurts maintainabilityHard-coding the current list of
_torchsub-folders means:
- Any newly-added sub-folder (e.g. a future
distributed,training, …) will silently be skipped on H100 CI until someone remembers to touch every list file.- The same list is duplicated across several YAMLs (see
l0_b200.yml,l0_rtx_pro_6000.yml, …) so drift is likely.Prefer an inclusive pattern with explicit exclusions, e.g.
- - unittest/_torch/attention - - unittest/_torch/compilation - ... - - unittest/_torch/thop + # Run all _torch tests except the heavy groups we really want to avoid + - unittest/_torch -k "not modeling and not multi_gpu and not auto_deploy"or keep the enumeration in a single YAML anchor that other configs can reference.
📜 Review details
Configuration used: .coderabbit.yaml
Review profile: CHILL
Plan: Pro
📒 Files selected for processing (5)
tests/integration/test_lists/test-db/l0_a30.yml(1 hunks)tests/integration/test_lists/test-db/l0_b200.yml(1 hunks)tests/integration/test_lists/test-db/l0_gb202.yml(1 hunks)tests/integration/test_lists/test-db/l0_h100.yml(1 hunks)tests/integration/test_lists/test-db/l0_rtx_pro_6000.yml(1 hunks)
🧰 Additional context used
🧠 Learnings (2)
📓 Common learnings
Learnt from: moraxu
PR: NVIDIA/TensorRT-LLM#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.
📚 Learning: in tensorrt-llm testing, it's common to have both cli flow tests (test_cli_flow.py) and pytorch api ...
Learnt from: moraxu
PR: NVIDIA/TensorRT-LLM#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/integration/test_lists/test-db/l0_gb202.ymltests/integration/test_lists/test-db/l0_rtx_pro_6000.ymltests/integration/test_lists/test-db/l0_a30.ymltests/integration/test_lists/test-db/l0_h100.ymltests/integration/test_lists/test-db/l0_b200.yml
🔇 Additional comments (3)
tests/integration/test_lists/test-db/l0_a30.yml (1)
23-23: Confirmedtest_beam_search.pyis presentThe beam‐search unit test was found at
tests/unittest/_torch/sampler/test_beam_search.py, so the CI selection will include it as expected. No further changes are needed.tests/integration/test_lists/test-db/l0_gb202.yml (1)
23-23: Corrected test path matches new directory structure – looks good
The update tounittest/_torch/attention/test_attention_mla.pyaccurately reflects the sub-directory re-org and keeps this list in sync with the other YAML files. No further action needed.tests/integration/test_lists/test-db/l0_b200.yml (1)
57-67: Granular PyTorch unittest selection looks goodMoving from the previous catch-all entry (
unittest/_torch … -k "not …" TIMEOUT (90)) to explicit sub-directory targets should make future maintenance easier and avoids the brittle negative-keyword filter. Nice clean-up.
|
PR_Github #14161 [ run ] triggered by Bot |
|
PR_Github #14161 [ run ] completed with state |
Funatiq
force-pushed
the
dev/ci/restructure_unit_tests
branch
from
1c7edd3 to
c1e67e4
Compare
|
/bot run |
|
PR_Github #14279 [ run ] triggered by Bot |
|
PR_Github #14279 [ run ] completed with state |
Funatiq
force-pushed
the
dev/ci/restructure_unit_tests
branch
from
c1e67e4 to
5072b39
Compare
|
/bot run |
|
PR_Github #14463 [ run ] triggered by Bot |
|
PR_Github #14463 [ run ] completed with state |
|
/bot run --disable-fail-fast |
|
PR_Github #14474 [ run ] triggered by Bot |
Funatiq
force-pushed
the
dev/ci/restructure_unit_tests
branch
from
45208ff to
1be2649
Compare
|
/bot run |
|
PR_Github #14489 [ run ] triggered by Bot |
|
PR_Github #14474 [ run ] completed with state |
|
PR_Github #14489 [ run ] completed with state |
Funatiq
force-pushed
the
dev/ci/restructure_unit_tests
branch
from
1be2649 to
ccb6a05
Compare
|
/bot run |
|
PR_Github #14593 [ run ] triggered by Bot |
|
PR_Github #14593 [ run ] completed with state |
Funatiq
force-pushed
the
dev/ci/restructure_unit_tests
branch
from
ccb6a05 to
a3a1d2a
Compare
|
PR_Github #15152 [ run ] triggered by Bot |
There was a problem hiding this comment.
Actionable comments posted: 2
🔭 Outside diff range comments (1)
tests/unittest/_torch/executor/test_resource_manager.py (1)
47-56: Avoid re-generating LoRA test data on every class run; short-circuit if artifacts existGuard the generation step by checking for the expected .npy artifacts first. This cuts redundant work across re-runs and reduces CI flakiness.
Apply this diff:
cpp_script_dir = os.path.join(cls.CPP_RESOURCES_DIR, "scripts") - # No reason to run this script for each test. - # TODO: move this to a fixture that runs once. + # No reason to run this script more than once per environment. + # Short-circuit if generated artifacts already exist. + if os.path.exists(cls.TP1_WEIGHTS_PATH) and os.path.exists(cls.TP1_CONFIG_PATH): + return + generate_lora_data_args_tp1 = [Additionally, consider making the timeout configurable and a bit more generous for busy CI nodes (e.g., via env TRTLLM_TEST_GEN_TIMEOUT, default 300s).
♻️ Duplicate comments (1)
tests/integration/test_lists/test-db/l0_h100.yml (1)
28-32: Quote entire YAML scalars that contain embedded quotes for safer parsingWrap the whole list item in quotes and flip inner quotes to single quotes. This avoids edge-cases across YAML parsers.
Apply these diffs:
- - unittest/_torch/modeling -k "modeling_llama" + - "unittest/_torch/modeling -k 'modeling_llama'" - - unittest/_torch/modeling -k "modeling_mixtral" + - "unittest/_torch/modeling -k 'modeling_mixtral'" - - unittest/_torch/modeling -k "modeling_nemotron" + - "unittest/_torch/modeling -k 'modeling_nemotron'" - - unittest/_torch/modeling -k "modeling_gemma3" + - "unittest/_torch/modeling -k 'modeling_gemma3'" - - unittest/_torch/modeling -k "modeling_gpt_oss" + - "unittest/_torch/modeling -k 'modeling_gpt_oss'"
🧹 Nitpick comments (2)
tensorrt_llm/_torch/flashinfer_utils.py (1)
10-16: Minor API polish: add docstrings and export list; consistent loggingAdd a short docstring for get_env_enable_pdl, expose all, and keep messages consistent. Also, consider returning a bool explicitly.
Apply this diff:
-def get_env_enable_pdl(): - return os.environ.get("TRTLLM_ENABLE_PDL", "0") == "1" +def get_env_enable_pdl() -> bool: + """Return True if Partial Decoding Layer (PDL) is enabled via env. + + Controlled by environment variable: TRTLLM_ENABLE_PDL=1 + """ + return os.environ.get("TRTLLM_ENABLE_PDL", "0") == "1" @@ -ENABLE_PDL = get_env_enable_pdl() +ENABLE_PDL: bool = get_env_enable_pdl() if ENABLE_PDL: logger.info("PDL is enabled") + +__all__ = ["IS_FLASHINFER_AVAILABLE", "ENABLE_PDL", "get_env_enable_pdl"]tests/unittest/_torch/sampler/test_trtllm_sampler.py (1)
43-46: Be explicit about top_p rather than passing NoneSome SamplingParams implementations treat None differently than omitting the argument. To avoid ambiguity, either omit top_p (use default) or set it explicitly to 1.0.
Apply this minimal change:
- temperature = 1.0 - top_p = None + temperature = 1.0 + top_p = 1.0Alternatively, drop top_p from SamplingParams(...) entirely to inherit the library default.
📜 Review details
Configuration used: .coderabbit.yaml
Review profile: CHILL
Plan: Pro
📒 Files selected for processing (16)
.github/CODEOWNERS(1 hunks)tensorrt_llm/_torch/attention_backend/__init__.py(1 hunks)tensorrt_llm/_torch/attention_backend/utils.py(1 hunks)tensorrt_llm/_torch/custom_ops/__init__.py(2 hunks)tensorrt_llm/_torch/custom_ops/flashinfer_custom_ops.py(2 hunks)tensorrt_llm/_torch/flashinfer_utils.py(1 hunks)tensorrt_llm/_torch/modules/rms_norm.py(1 hunks)tensorrt_llm/_torch/modules/rotary_embedding.py(1 hunks)tests/integration/defs/.test_durations(3 hunks)tests/integration/test_lists/test-db/l0_a30.yml(1 hunks)tests/integration/test_lists/test-db/l0_b200.yml(1 hunks)tests/integration/test_lists/test-db/l0_gb202.yml(1 hunks)tests/integration/test_lists/test-db/l0_h100.yml(1 hunks)tests/integration/test_lists/test-db/l0_rtx_pro_6000.yml(1 hunks)tests/unittest/_torch/executor/test_resource_manager.py(2 hunks)tests/unittest/_torch/sampler/test_trtllm_sampler.py(2 hunks)
✅ Files skipped from review due to trivial changes (2)
- tensorrt_llm/_torch/modules/rotary_embedding.py
- .github/CODEOWNERS
🚧 Files skipped from review as they are similar to previous changes (4)
- tests/integration/test_lists/test-db/l0_rtx_pro_6000.yml
- tests/integration/test_lists/test-db/l0_a30.yml
- tests/integration/test_lists/test-db/l0_gb202.yml
- tests/integration/test_lists/test-db/l0_b200.yml
🧰 Additional context used
📓 Path-based instructions (2)
**/*.py
📄 CodeRabbit Inference Engine (CODING_GUIDELINES.md)
**/*.py: Python code must target Python 3.8+
Python indentation: 4 spaces, no tabs
Maintain module namespace in imports (from package.subpackage import foo; then use foo.SomeClass())
Python file names use snake_case
Python class names use PascalCase
Python functions/methods and local variables use snake_case; variables starting with a number get k_ prefix (e.g., k_99th_percentile)
Global variables use G_ prefixed UPPER_SNAKE_CASE (e.g., G_MY_GLOBAL)
Constants use UPPER_SNAKE_CASE in Python
Avoid shadowing variables from outer scopes in Python
Initialize all externally visible members of a Python class in init
Prefer docstrings for interfaces used outside a file; comments for local code
Use Google-style docstrings for classes and functions (Sphinx-parsable)
Document attributes/variables inline with short docstrings
Avoid reflection when simple alternatives exist (e.g., prefer explicit parameters over dict(**locals()))
In try/except, catch the narrowest exceptions possible
For duck-typing with try/except, keep try body minimal and put logic in else
Files:
tensorrt_llm/_torch/flashinfer_utils.pytensorrt_llm/_torch/attention_backend/__init__.pytensorrt_llm/_torch/attention_backend/utils.pytensorrt_llm/_torch/modules/rms_norm.pytests/unittest/_torch/executor/test_resource_manager.pytensorrt_llm/_torch/custom_ops/__init__.pytensorrt_llm/_torch/custom_ops/flashinfer_custom_ops.pytests/unittest/_torch/sampler/test_trtllm_sampler.py
**/*.{cpp,cxx,cc,cu,h,hpp,hxx,hh,cuh,py}
📄 CodeRabbit Inference Engine (CODING_GUIDELINES.md)
Prepend NVIDIA copyright header (current year) to all source files
Files:
tensorrt_llm/_torch/flashinfer_utils.pytensorrt_llm/_torch/attention_backend/__init__.pytensorrt_llm/_torch/attention_backend/utils.pytensorrt_llm/_torch/modules/rms_norm.pytests/unittest/_torch/executor/test_resource_manager.pytensorrt_llm/_torch/custom_ops/__init__.pytensorrt_llm/_torch/custom_ops/flashinfer_custom_ops.pytests/unittest/_torch/sampler/test_trtllm_sampler.py
🧠 Learnings (4)
📚 Learning: 2025-08-01T15:14:45.673Z
Learnt from: yibinl-nvidia
PR: NVIDIA/TensorRT-LLM#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/executor/test_resource_manager.py
📚 Learning: 2025-08-06T13:58:07.506Z
Learnt from: galagam
PR: NVIDIA/TensorRT-LLM#6487
File: tests/unittest/_torch/auto_deploy/unit/singlegpu/test_ad_trtllm_bench.py:1-12
Timestamp: 2025-08-06T13:58:07.506Z
Learning: In TensorRT-LLM, test files (files under tests/ directories) do not require NVIDIA copyright headers, unlike production source code files. Test files typically start directly with imports, docstrings, or code.
Applied to files:
tests/unittest/_torch/executor/test_resource_manager.py
📚 Learning: 2025-08-11T20:09:24.389Z
Learnt from: achartier
PR: NVIDIA/TensorRT-LLM#6763
File: tests/integration/defs/triton_server/conftest.py:16-22
Timestamp: 2025-08-11T20:09:24.389Z
Learning: In the TensorRT-LLM test infrastructure, the team prefers simple, direct solutions (like hard-coding directory traversal counts) over more complex but robust approaches when dealing with stable directory structures. They accept the maintenance cost of updating tests if the layout changes.
Applied to files:
tests/unittest/_torch/executor/test_resource_manager.py
📚 Learning: 2025-07-28T17:06:08.621Z
Learnt from: moraxu
PR: NVIDIA/TensorRT-LLM#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_trtllm_sampler.pytests/integration/test_lists/test-db/l0_h100.yml
🪛 Gitleaks (8.27.2)
tests/integration/test_lists/test-db/l0_h100.yml
27-27: Detected a Generic API Key, potentially exposing access to various services and sensitive operations.
(generic-api-key)
⏰ 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 (15)
tests/unittest/_torch/executor/test_resource_manager.py (1)
26-26: Integration paths verified – root_dir bump is safe
- Confirmed that the following exist in the repo root:
tests/integrationcpp/tests/resources/scripts/generate_test_lora_weights.pycpp/tests/resources/dataAll required directories and files are present, so the updated
root_dirwill resolve imports correctly in CI.tests/integration/test_lists/test-db/l0_h100.yml (2)
33-33: Reintroducing perf bench filter is fine in this bandExplicit spec via node::test syntax is consistent with other entries. No issues.
17-26: All unittest/_torch selectors resolve to valid test directoriesI’ve verified that each of the following paths under tests/unittest/torch exists and contains at least one test*.py file, so there are no no-op selectors:
- tests/unittest/_torch/attention
- tests/unittest/_torch/compilation
- tests/unittest/_torch/debugger
- tests/unittest/_torch/executor
- tests/unittest/_torch/misc
- tests/unittest/_torch/modules
- tests/unittest/_torch/multimodal
- tests/unittest/_torch/sampler
- tests/unittest/_torch/speculative
- tests/unittest/_torch/thop
tests/unittest/_torch/sampler/test_trtllm_sampler.py (1)
32-32: Test signature simplification is fineDropping the JSON-driven fixture in favor of inline config keeps this test focused; no correctness issues spotted.
tests/integration/defs/.test_durations (3)
141-150: Granular durations for unittest/_torch sub-suites: good changeThis will improve scheduling and predictability for the split test lists.
250-250: Modeling key path updated: consistent with YAML changesThe move to unittest/torch/modeling -k "modeling_llama" aligns with l0* test lists.
292-292: Attention MLA path fix acknowledgedDuration key updated to new location tests/unittest/_torch/attention/test_attention_mla.py; no issues.
tensorrt_llm/_torch/attention_backend/__init__.py (1)
1-1: LGTM! Import source refactored to centralized module.The import of
IS_FLASHINFER_AVAILABLEhas been correctly updated to use the new centralizedflashinfer_utilsmodule. This improves maintainability by having a single source of truth for FlashInfer availability checks across the codebase.tensorrt_llm/_torch/attention_backend/utils.py (1)
4-4: LGTM! Import refactored to use centralized FlashInfer utils.The import has been correctly updated to source
IS_FLASHINFER_AVAILABLEfrom the new centralizedflashinfer_utilsmodule, maintaining consistency with the broader refactoring effort.tensorrt_llm/_torch/modules/rms_norm.py (1)
22-22: LGTM! Import source updated to centralized FlashInfer utilities.The import change correctly aligns with the centralization of FlashInfer availability checks. The module now uses the shared
flashinfer_utilsmodule as the source of truth forIS_FLASHINFER_AVAILABLE.tensorrt_llm/_torch/custom_ops/__init__.py (2)
1-2: LGTM! Import updates align with centralization and API expansion.The changes correctly:
- Import
IS_FLASHINFER_AVAILABLEfrom the centralizedflashinfer_utilsmodule- Import the new attention custom ops from the modules layer
This maintains consistency with the broader refactoring effort while expanding the public API.
16-17: LGTM! Public API expanded with new attention operations.The addition of
attn_custom_op_inplaceandmla_custom_op_inplaceto the__all__list properly exposes these operations as part of the public API.tensorrt_llm/_torch/custom_ops/flashinfer_custom_ops.py (3)
3-3: LGTM! Import updated to use centralized FlashInfer utilities.The import correctly sources both
ENABLE_PDLandIS_FLASHINFER_AVAILABLEfrom the new centralizedflashinfer_utilsmodule, replacing the previous platform-specific availability checks.
8-8: LGTM! Import follows centralized gating pattern.The import of
apply_rope_with_cos_sin_cache_inplaceis now properly guarded by the centralizedIS_FLASHINFER_AVAILABLEflag, ensuring consistent availability checking across the codebase.
48-48: LGTM! Function call updated to use imported wrapper.The call has been correctly updated to use the imported
apply_rope_with_cos_sin_cache_inplacefunction instead of the fully qualified module path, which aligns with the new import structure.
|
PR_Github #15152 [ run ] completed with state |
|
/bot run |
|
PR_Github #15180 [ run ] triggered by Bot |
Funatiq
force-pushed
the
dev/ci/restructure_unit_tests
branch
from
59da3ea to
c691c0b
Compare
|
/bot run |
|
PR_Github #15789 [ run ] triggered by Bot |
|
PR_Github #15789 [ run ] completed with state |
- Moved all unittest files to sub-directories. - Updated test lists in YAML files to include new test paths. This commit improves the testing framework for unittest files, ensuring comprehensive validation of the new implementations. Signed-off-by: Robin Kobus <[email protected]>
Signed-off-by: Robin Kobus <[email protected]>
- Included `attn_custom_op_inplace` and `mla_custom_op_inplace` in the `__all__` list of the custom ops module. - This change fixes unittest/_torch/compilation/test_add_norm.py::test_add_norm_fusion. Signed-off-by: Robin Kobus <[email protected]>
- Added `flashinfer_utils.py` to manage flashinfer availability and PDL environment variable. - Updated imports across multiple modules to source `IS_FLASHINFER_AVAILABLE` and `ENABLE_PDL` from the new utility file. - This refactor improves code organization and maintains clarity regarding flashinfer's integration. - This change fixes circular import issues in tensorrt_llm/_torch/attention_backend/__init__.py. Signed-off-by: Robin Kobus <[email protected]>
Signed-off-by: Robin Kobus <[email protected]>
…ixture - Removed the test_case fixture and its associated JSON loading, simplifying the test setup. - Hardcoded sampling parameters for clarity and consistency in the test execution. Signed-off-by: Robin Kobus <[email protected]>
- Changed paths for `test_custom_ops.py` and `test_autotuner.py` to their new sub-directory locations. - This update ensures proper ownership and visibility for the relocated test files. Signed-off-by: Robin Kobus <[email protected]>
- Added new test duration entries for various unittest cases in the `_torch` module. - Adjusted existing entries for clarity and consistency in naming conventions. - This update enhances the tracking of test performance metrics. Signed-off-by: Robin Kobus <[email protected]>
Funatiq
force-pushed
the
dev/ci/restructure_unit_tests
branch
from
c691c0b to
52eee4b
Compare
|
/bot run --disable-fail-fast |
|
PR_Github #15817 [ run ] triggered by Bot |
There was a problem hiding this comment.
Actionable comments posted: 2
Caution
Some comments are outside the diff and can’t be posted inline due to platform limitations.
⚠️ Outside diff range comments (1)
tensorrt_llm/_torch/custom_ops/__init__.py (1)
20-31: Ensure tests guard against missing flashinfer supportNo circular imports involving
modules.attentionwere found. However, the test intests/unittest/_torch/compilation/test_add_norm.pyunconditionally does:from tensorrt_llm._torch.custom_ops import flashinfer_rmsnormThis will break when
IS_FLASHINFER_AVAILABLE=False. Please update the test to skip or guard this import, for example:• File: tests/unittest/_torch/compilation/test_add_norm.py (line 5)
Replace:- from tensorrt_llm._torch.custom_ops import flashinfer_rmsnormWith something like:
+ import pytest + pytest.importorskip("tensorrt_llm._torch.custom_ops", reason="requires flashinfer support") + from tensorrt_llm._torch.custom_ops import flashinfer_rmsnormThis ensures tests pass regardless of flashinfer availability.
♻️ Duplicate comments (2)
.github/CODEOWNERS (1)
53-54: Prefer directory-level CODEOWNERS to cover all thop/misc tests and future additionsThe path updates are correct after the test move. To avoid having to maintain per-file entries and to ensure all existing and future tests under these subdirectories are owned, replace the two file-level rules with directory-level rules.
Proposed diff:
-/tests/unittest/_torch/thop/test_custom_ops.py @NVIDIA/trt-llm-torch-graph-compiler -/tests/unittest/_torch/misc/test_autotuner.py @NVIDIA/trt-llm-torch-graph-compiler +/tests/unittest/_torch/thop/ @NVIDIA/trt-llm-torch-graph-compiler +/tests/unittest/_torch/misc/ @NVIDIA/trt-llm-torch-graph-compilerOptional: run this script to confirm there are multiple tests under these dirs (justifying directory-level ownership) and to show current CODEOWNERS matches:
#!/bin/bash set -euo pipefail echo "Counting tests under thop/ and misc/:" thop_count=$(fd -t f -e py . tests/unittest/_torch/thop 2>/dev/null | wc -l | tr -d ' ') misc_count=$(fd -t f -e py . tests/unittest/_torch/misc 2>/dev/null | wc -l | tr -d ' ') echo "thop: ${thop_count} files" echo "misc: ${misc_count} files" echo echo "Listing thop tests:" fd -t f -e py . tests/unittest/_torch/thop 2>/dev/null || true echo echo "Listing misc tests:" fd -t f -e py . tests/unittest/_torch/misc 2>/dev/null || true echo echo "Current CODEOWNERS entries for thop/misc:" rg -nE '^\s*/tests/unittest/_torch/(thop|misc)(/|/.*|\.|$)' .github/CODEOWNERS || truetests/integration/test_lists/test-db/l0_h100.yml (1)
28-28: Quote entire YAML scalar when embedding quotes in -k filtersSafer across YAML parsers and matches prior guidance. Use outer double-quotes and inner single-quotes for the -k expression.
- - unittest/_torch/modeling -k "modeling_llama" + - "unittest/_torch/modeling -k 'modeling_llama'"Consider applying the same quoting pattern to the adjacent modeling filter lines for consistency.
🧹 Nitpick comments (4)
tensorrt_llm/_torch/custom_ops/flashinfer_custom_ops.py (4)
10-10: Fix wording/typo in comment (“Warp” → “Wrap”).Minor clarity improvement in a user-facing code comment.
- # Warp this into custom op since flashinfer didn't warp it properly and we want to avoid graph break between mlp layer for user buffer optimization + # Wrap this into a custom op since FlashInfer didn't expose it properly, and we want to avoid graph breaks between MLP layers for user-buffer optimization
19-19: Fix wording and grammar in comment.Minor clarity/grammar improvements.
- # Warp this into custom op since flashinfer provides default value for eps with would produce two different graphs depends on the eps value. + # Wrap this into a custom op since FlashInfer provides a default value for eps, + # which would produce two different graphs depending on the eps value.
30-36: Provide a fake (meta) kernel for fused_add_rmsnorm.Without a registered fake, torch.compile/meta passes may fail when encountering this custom op. Add a lightweight validator that returns None and checks shapes.
@torch.library.custom_op("trtllm::flashinfer_fused_add_rmsnorm", mutates_args=("input", "residual")) def flashinfer_fused_add_rmsnorm(input: torch.Tensor, residual: torch.Tensor, weight: torch.Tensor, eps: float) -> None: fused_add_rmsnorm(input, residual, weight, eps, enable_pdl=ENABLE_PDL) + + @flashinfer_fused_add_rmsnorm.register_fake + def _(input: torch.Tensor, + residual: torch.Tensor, + weight: torch.Tensor, + eps: float) -> None: + # Meta: validate shapes/dtypes; op is in-place and returns None. + assert input.shape == residual.shape, "input and residual must have the same shape" + assert weight.dim() == 1 and weight.numel() == input.shape[-1], \ + "weight must be 1D and match the last dimension of input" + return
57-66: Strengthen the fake (meta) kernel for ROPE in-place op.Add shape checks to catch misuses during tracing/compile; keep return type None for in-place semantics.
- @flashinfer_apply_rope_with_cos_sin_cache_inplace.register_fake - def _( - positions: torch.Tensor, - query: torch.Tensor, - key: torch.Tensor, - head_size: int, - cos_sin_cache: torch.Tensor, - is_neox: bool = True, - ): - return + @flashinfer_apply_rope_with_cos_sin_cache_inplace.register_fake + def _( + positions: torch.Tensor, + query: torch.Tensor, + key: torch.Tensor, + head_size: int, + cos_sin_cache: torch.Tensor, + is_neox: bool = True, + ) -> None: + # Meta: validate shapes; op is in-place on query/key. + assert query.shape == key.shape, "query and key must have the same shape" + assert cos_sin_cache.dim() >= 2, "cos_sin_cache expected to have at least 2 dims" + # positions is typically [B, T] or [T]; rely on runtime for exact layout. + return
📜 Review details
Configuration used: .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 (16)
.github/CODEOWNERS(1 hunks)tensorrt_llm/_torch/attention_backend/__init__.py(1 hunks)tensorrt_llm/_torch/attention_backend/utils.py(1 hunks)tensorrt_llm/_torch/custom_ops/__init__.py(2 hunks)tensorrt_llm/_torch/custom_ops/flashinfer_custom_ops.py(2 hunks)tensorrt_llm/_torch/flashinfer_utils.py(1 hunks)tensorrt_llm/_torch/modules/rms_norm.py(1 hunks)tensorrt_llm/_torch/modules/rotary_embedding.py(1 hunks)tests/integration/defs/.test_durations(3 hunks)tests/integration/test_lists/test-db/l0_a30.yml(1 hunks)tests/integration/test_lists/test-db/l0_b200.yml(1 hunks)tests/integration/test_lists/test-db/l0_gb202.yml(1 hunks)tests/integration/test_lists/test-db/l0_h100.yml(1 hunks)tests/integration/test_lists/test-db/l0_rtx_pro_6000.yml(1 hunks)tests/unittest/_torch/executor/test_resource_manager.py(2 hunks)tests/unittest/_torch/sampler/test_trtllm_sampler.py(2 hunks)
🚧 Files skipped from review as they are similar to previous changes (12)
- tests/integration/test_lists/test-db/l0_rtx_pro_6000.yml
- tensorrt_llm/_torch/attention_backend/init.py
- tensorrt_llm/_torch/flashinfer_utils.py
- tensorrt_llm/_torch/attention_backend/utils.py
- tests/integration/test_lists/test-db/l0_a30.yml
- tests/unittest/_torch/executor/test_resource_manager.py
- tensorrt_llm/_torch/modules/rms_norm.py
- tests/integration/defs/.test_durations
- tests/integration/test_lists/test-db/l0_b200.yml
- tensorrt_llm/_torch/modules/rotary_embedding.py
- tests/integration/test_lists/test-db/l0_gb202.yml
- tests/unittest/_torch/sampler/test_trtllm_sampler.py
🧰 Additional context used
📓 Path-based instructions (2)
**/*.py
📄 CodeRabbit Inference Engine (CODING_GUIDELINES.md)
**/*.py: Python code must target Python 3.8+
Python indentation: 4 spaces, no tabs
Maintain module namespace in imports (from package.subpackage import foo; then use foo.SomeClass())
Python file names use snake_case
Python class names use PascalCase
Python functions/methods and local variables use snake_case; variables starting with a number get k_ prefix (e.g., k_99th_percentile)
Global variables use G_ prefixed UPPER_SNAKE_CASE (e.g., G_MY_GLOBAL)
Constants use UPPER_SNAKE_CASE in Python
Avoid shadowing variables from outer scopes in Python
Initialize all externally visible members of a Python class in init
Prefer docstrings for interfaces used outside a file; comments for local code
Use Google-style docstrings for classes and functions (Sphinx-parsable)
Document attributes/variables inline with short docstrings
Avoid reflection when simple alternatives exist (e.g., prefer explicit parameters over dict(**locals()))
In try/except, catch the narrowest exceptions possible
For duck-typing with try/except, keep try body minimal and put logic in else
Files:
tensorrt_llm/_torch/custom_ops/flashinfer_custom_ops.pytensorrt_llm/_torch/custom_ops/__init__.py
**/*.{cpp,cxx,cc,cu,h,hpp,hxx,hh,cuh,py}
📄 CodeRabbit Inference Engine (CODING_GUIDELINES.md)
Prepend NVIDIA copyright header (current year) to all source files
Files:
tensorrt_llm/_torch/custom_ops/flashinfer_custom_ops.pytensorrt_llm/_torch/custom_ops/__init__.py
🧠 Learnings (4)
📚 Learning: 2025-07-28T17:06:08.621Z
Learnt from: moraxu
PR: NVIDIA/TensorRT-LLM#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:
.github/CODEOWNERStests/integration/test_lists/test-db/l0_h100.yml
📚 Learning: 2025-08-06T13:58:07.506Z
Learnt from: galagam
PR: NVIDIA/TensorRT-LLM#6487
File: tests/unittest/_torch/auto_deploy/unit/singlegpu/test_ad_trtllm_bench.py:1-12
Timestamp: 2025-08-06T13:58:07.506Z
Learning: In TensorRT-LLM, test files (files under tests/ directories) do not require NVIDIA copyright headers, unlike production source code files. Test files typically start directly with imports, docstrings, or code.
Applied to files:
.github/CODEOWNERS
📚 Learning: 2025-08-11T20:09:24.389Z
Learnt from: achartier
PR: NVIDIA/TensorRT-LLM#6763
File: tests/integration/defs/triton_server/conftest.py:16-22
Timestamp: 2025-08-11T20:09:24.389Z
Learning: In the TensorRT-LLM test infrastructure, the team prefers simple, direct solutions (like hard-coding directory traversal counts) over more complex but robust approaches when dealing with stable directory structures. They accept the maintenance cost of updating tests if the layout changes.
Applied to files:
.github/CODEOWNERS
📚 Learning: 2025-08-14T15:43:23.067Z
Learnt from: MatthiasKohl
PR: NVIDIA/TensorRT-LLM#6904
File: tensorrt_llm/_torch/attention_backend/trtllm.py:259-262
Timestamp: 2025-08-14T15:43:23.067Z
Learning: In TensorRT-LLM's attention backend, tensor parameters in the plan() method are assigned directly without validation (dtype, device, contiguity checks). This maintains consistency across all tensor inputs and follows the pattern of trusting callers to provide correctly formatted tensors.
Applied to files:
.github/CODEOWNERS
🧬 Code Graph Analysis (2)
tensorrt_llm/_torch/custom_ops/flashinfer_custom_ops.py (1)
tensorrt_llm/_torch/modules/swiglu.py (1)
silu_and_mul(54-79)
tensorrt_llm/_torch/custom_ops/__init__.py (1)
tensorrt_llm/_torch/modules/attention.py (2)
attn_custom_op_inplace(72-98)mla_custom_op_inplace(534-541)
🪛 Gitleaks (8.27.2)
tests/integration/test_lists/test-db/l0_h100.yml
27-27: Detected a Generic API Key, potentially exposing access to various services and sensitive operations.
(generic-api-key)
⏰ 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 (6)
tests/integration/test_lists/test-db/l0_h100.yml (2)
17-26: Verified: PyTorch sub-directory buckets existThe sanity check confirms that all directories listed in tests/integration/test_lists/test-db/l0_h100.yml (lines 17–26) under tests/unittest/_torch/ are present. CI path-not-found failures should be resolved. Approved.
33-33: Verified: Test path and symbol existThe test file and function have been confirmed:
- tests/unittest/_torch/auto_deploy/unit/singlegpu/test_ad_trtllm_bench.py::test_trtllm_bench_backend_comparison
No further action required.
tensorrt_llm/_torch/custom_ops/flashinfer_custom_ops.py (3)
5-9: Centralized FlashInfer gating looks good.Import-time side effects are minimized and the availability check is now unified via flashinfer_utils. This will simplify maintenance and avoid noisy ImportErrors on unsupported platforms.
37-55: ROPE wrapper: interface alignment looks correct.Signature and mutates_args align with the underlying in-place behavior; passthrough of is_neox and head_size is consistent with typical FlashInfer ROPE APIs.
41-47:is_neoxdefault is never relied upon—call sites pass it explicitly
The only invocation offlashinfer_apply_rope_with_cos_sin_cache_inplaceinrotary_embedding.pysuppliesself.is_neoxas the sixth argument, so the wrapper’s default (True) is never used. No positional shifts or default mismatches will occur.tensorrt_llm/_torch/custom_ops/__init__.py (1)
16-17: Good addition to public API.Exposing attn_custom_op_inplace and mla_custom_op_inplace here improves discoverability and keeps users on the stable import path.
|
PR_Github #15817 [ run ] completed with state |
Signed-off-by: Robin Kobus <[email protected]> Signed-off-by: Yuxin <[email protected]>
Summary by CodeRabbit
New Features
Tests
Chores
Description
Test Coverage
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.