[https://nvbugs/5410687][fix] Hopper w4a8 groupwise MoE interleave #6708
Conversation
📝 WalkthroughWalkthroughThe changes introduce architecture-specific handling for weight scale interleaving in MoE groupwise quantization, particularly for Hopper (SM 90) GPUs. This includes updating tensor shape calculations, adding a utility function for interleave factor computation, modifying weight preprocessing logic, and refining unit tests to accommodate architecture differences and interleaving requirements. Changes
Sequence Diagram(s)sequenceDiagram
participant Test as Unit Test
participant Quant as Quantization Functional
participant MoE as MOE Layer
Test->>Quant: Call preprocess_weights_for_mixed_gemm(..., do_weight_interleave)
alt do_weight_interleave is True
Quant->>Quant: Permute and interleave weights
else do_weight_interleave is False
Quant->>Quant: Skip interleaving
end
Quant->>Test: Return processed weights
Test->>MoE: Initialize MOEWeightWrapper(...)
MoE->>Quant: Call get_weight_scale_interleave_factor(...)
MoE->>MoE: Set weights_scaling_factor shape based on interleave factor
Estimated code review effort🎯 3 (Moderate) | ⏱️ ~20 minutes Suggested reviewers
Note ⚡️ Unit Test Generation is now available in beta!Learn more here, or try it out under "Finishing Touches" below. ✨ 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 comments)
Other keywords and placeholders
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: 1
🧹 Nitpick comments (1)
tests/unittest/trt/quantization/test_moe_weight_only_groupwise_quant_matmul.py (1)
234-245: Rename ambiguous variableIto improve readability.The variable name
Ion line 238 is ambiguous and could be confused with lowercase 'l' or the number '1'. Consider using a more descriptive name.def interleave_scales(scales: torch.Tensor, interleave_dim: int): # [num_experts, num_groups, num_cols] --> [num_experts, num_groups // interleave, num_cols * interleave] # Note: num_groups = num_rows // group_size E, G, C = scales.shape - I = tensorrt_llm.quantization.functional.get_weight_scale_interleave_factor( + interleave_factor = tensorrt_llm.quantization.functional.get_weight_scale_interleave_factor( interleave_dim, group_size) - assert G % I == 0, f"Group dimension ({G}) must be divisible by interleave factor ({I})." - scales_interleaved = scales.reshape(E, G // I, I, C) + assert G % interleave_factor == 0, f"Group dimension ({G}) must be divisible by interleave factor ({interleave_factor})." + scales_interleaved = scales.reshape(E, G // interleave_factor, interleave_factor, C) scales_interleaved = scales_interleaved.permute(0, 1, 3, 2) scales_interleaved = scales_interleaved.reshape( - E, G // I, C * I) + E, G // interleave_factor, C * interleave_factor) return scales_interleaved.contiguous()
📜 Review details
Configuration used: .coderabbit.yaml
Review profile: CHILL
Plan: Pro
📒 Files selected for processing (3)
tensorrt_llm/layers/moe.py(2 hunks)tensorrt_llm/quantization/functional.py(3 hunks)tests/unittest/trt/quantization/test_moe_weight_only_groupwise_quant_matmul.py(4 hunks)
🧰 Additional context used
📓 Path-based instructions (2)
**/*.py
📄 CodeRabbit Inference Engine (CODING_GUIDELINES.md)
**/*.py: Python code should conform to Python 3.8+.
Indent Python code with 4 spaces. Do not use tabs.
Always maintain the namespace when importing in Python, even if only one class or function from a module is used.
Python filenames should use snake_case (e.g., some_file.py).
Python classes should use PascalCase (e.g., class SomeClass).
Python functions and methods should use snake_case (e.g., def my_awesome_function():).
Python local variables should use snake_case. Prefix k for variable names that start with a number (e.g., k_99th_percentile).
Python global variables should use upper snake_case and prefix G (e.g., G_MY_GLOBAL).
Python constants should use upper snake_case (e.g., MY_CONSTANT).
Avoid shadowing variables declared in an outer scope in Python.
Initialize all externally visible members of a Python class in the constructor.
For interfaces that may be used outside a Python file, prefer docstrings over comments.
Comments in Python should be reserved for code within a function, or interfaces that are local to a file.
Use Google style docstrings for Python classes and functions, which can be parsed by Sphinx.
Attributes and variables in Python can be documented inline; attribute docstrings will be rendered under the class docstring.
Avoid using reflection in Python when functionality can be easily achieved without it.
When using try-except blocks in Python, limit the except to the smallest set of errors possible.
When using try-except blocks to handle multiple possible variable types in Python, keep the body of the try as small as possible, using the else block to implement the logic.
Files:
tensorrt_llm/layers/moe.pytests/unittest/trt/quantization/test_moe_weight_only_groupwise_quant_matmul.pytensorrt_llm/quantization/functional.py
**/*.{cpp,h,hpp,cc,cxx,cu,py}
📄 CodeRabbit Inference Engine (CODING_GUIDELINES.md)
All TensorRT-LLM Open Source Software code should contain an NVIDIA copyright header that includes the current year. This includes .cpp, .h, .cu, .py, and any other source files which are compiled or interpreted.
Files:
tensorrt_llm/layers/moe.pytests/unittest/trt/quantization/test_moe_weight_only_groupwise_quant_matmul.pytensorrt_llm/quantization/functional.py
🧠 Learnings (2)
📚 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/unittest/trt/quantization/test_moe_weight_only_groupwise_quant_matmul.py
📚 Learning: in tensorrt-llm, test files (files under tests/ directories) do not require nvidia copyright headers...
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/trt/quantization/test_moe_weight_only_groupwise_quant_matmul.py
🪛 Ruff (0.12.2)
tests/unittest/trt/quantization/test_moe_weight_only_groupwise_quant_matmul.py
238-238: Ambiguous variable name: I
(E741)
🔇 Additional comments (7)
tensorrt_llm/layers/moe.py (2)
43-46: LGTM: Import addition for weight scale interleaving.The import of
get_weight_scale_interleave_factoris correctly added to support architecture-specific weight scale interleaving for W4A8 groupwise MoE quantization.
493-504: Architecture-specific weight scale interleaving implementation looks correct.The implementation correctly handles Hopper (SM 90) architecture-specific requirements for W4A8 groupwise MoE quantization:
- Weight parameter: Maintains the correct shape
(experts_per_node, in_features, out_features // 4)for int4 quantization- Conditional interleaving: Properly checks for
W4A8_ALPHAflag to determine if interleaving is needed- Scale factor computation: Uses
get_weight_scale_interleave_factorto compute the architecture-specific interleave factor- Shape adjustment: Correctly adjusts the
weights_scaling_factorshape to account for interleaving by dividing the second dimension by the interleave factor and multiplying the third dimension by itThis change aligns with the PR objective of fixing Hopper w4a8 groupwise MoE interleave and maintains backward compatibility by defaulting to
scale_interleave_factor = 1when W4A8_ALPHA is not enabled.tensorrt_llm/quantization/functional.py (1)
953-958: LGTM! Good backward compatibility with the default parameter.The addition of the
do_weight_interleaveparameter with a default value ofTruemaintains backward compatibility while enabling architecture-specific behavior.tests/unittest/trt/quantization/test_moe_weight_only_groupwise_quant_matmul.py (4)
17-17: LGTM! Proper extension of test coverage to multiple GPU architectures.The changes correctly extend test support from Ada-only to both Ada and Hopper architectures.
Also applies to: 24-26, 31-32, 317-317
72-73: Good defensive programming to handle interleaved tensors.Using
fc2_prequant_scale.shape[-1]ensures we get the correct dimension regardless of whether weight or scale interleaving has been applied.
209-228: LGTM! Correct architecture-specific weight preprocessing logic.The conditional logic properly handles:
- Hopper (SM 90) with W4A8: disables weight interleaving
- Ada (SM 89) with W4A8: converts scales to float16
229-233: LGTM! Appropriate handling of unsupported configuration.Correctly skips the test when
has_zerois enabled with Hopper W4A8, as this combination is not yet supported.
|
PR_Github #14465 [ run ] triggered by Bot |
|
PR_Github #14465 [ run ] completed with state |
Signed-off-by: Haohang Huang <[email protected]>
Co-authored-by: coderabbitai[bot] <136622811+coderabbitai[bot]@users.noreply.github.com> Signed-off-by: Haohang Huang <[email protected]>
symphonylyh
force-pushed
the
user/haohangh/fix-w4a8-interleave
branch
from
33aef0c to
446f26c
Compare
|
/bot run --disable-fail-fast |
There was a problem hiding this comment.
Actionable comments posted: 0
🧹 Nitpick comments (1)
tests/unittest/trt/quantization/test_moe_weight_only_groupwise_quant_matmul.py (1)
209-252: Comprehensive architecture-specific preprocessing logic implemented correctly.The implementation properly handles the differences between Ada (SM 89) and Hopper (SM 90) architectures:
- Hopper w4a8 correctly disables weight interleaving (
do_weight_interleave=False)- Architecture-specific type conversions (float16 for Ada, bfloat16 for Hopper)
- Proper scale interleaving for Hopper with the nested
interleave_scalesfunction- Appropriate skipping of zero tensor support for Hopper w4a8
However, address the ambiguous variable name flagged by static analysis:
- I = tensorrt_llm.quantization.functional.get_weight_scale_interleave_factor( + interleave_factor = tensorrt_llm.quantization.functional.get_weight_scale_interleave_factor( interleave_dim, group_size) - assert G % I == 0, f"Group dimension ({G}) must be divisible by interleave factor ({I})." + assert G % interleave_factor == 0, f"Group dimension ({G}) must be divisible by interleave factor ({interleave_factor})." - scales_interleaved = scales.reshape(E, G // I, I, C) + scales_interleaved = scales.reshape(E, G // interleave_factor, interleave_factor, C) - scales_interleaved = scales_interleaved.reshape( - E, G // I, C * I) + scales_interleaved = scales_interleaved.reshape( + E, G // interleave_factor, C * interleave_factor)
📜 Review details
Configuration used: .coderabbit.yaml
Review profile: CHILL
Plan: Pro
📒 Files selected for processing (3)
tensorrt_llm/layers/moe.py(2 hunks)tensorrt_llm/quantization/functional.py(3 hunks)tests/unittest/trt/quantization/test_moe_weight_only_groupwise_quant_matmul.py(4 hunks)
🚧 Files skipped from review as they are similar to previous changes (2)
- tensorrt_llm/layers/moe.py
- tensorrt_llm/quantization/functional.py
🧰 Additional context used
📓 Path-based instructions (2)
**/*.py
📄 CodeRabbit Inference Engine (CODING_GUIDELINES.md)
**/*.py: Python code should conform to Python 3.8+.
Indent Python code with 4 spaces. Do not use tabs.
Always maintain the namespace when importing in Python, even if only one class or function from a module is used.
Python filenames should use snake_case (e.g., some_file.py).
Python classes should use PascalCase (e.g., class SomeClass).
Python functions and methods should use snake_case (e.g., def my_awesome_function():).
Python local variables should use snake_case. Prefix k for variable names that start with a number (e.g., k_99th_percentile).
Python global variables should use upper snake_case and prefix G (e.g., G_MY_GLOBAL).
Python constants should use upper snake_case (e.g., MY_CONSTANT).
Avoid shadowing variables declared in an outer scope in Python.
Initialize all externally visible members of a Python class in the constructor.
For interfaces that may be used outside a Python file, prefer docstrings over comments.
Comments in Python should be reserved for code within a function, or interfaces that are local to a file.
Use Google style docstrings for Python classes and functions, which can be parsed by Sphinx.
Attributes and variables in Python can be documented inline; attribute docstrings will be rendered under the class docstring.
Avoid using reflection in Python when functionality can be easily achieved without it.
When using try-except blocks in Python, limit the except to the smallest set of errors possible.
When using try-except blocks to handle multiple possible variable types in Python, keep the body of the try as small as possible, using the else block to implement the logic.
Files:
tests/unittest/trt/quantization/test_moe_weight_only_groupwise_quant_matmul.py
**/*.{cpp,h,hpp,cc,cxx,cu,py}
📄 CodeRabbit Inference Engine (CODING_GUIDELINES.md)
All TensorRT-LLM Open Source Software code should contain an NVIDIA copyright header that includes the current year. This includes .cpp, .h, .cu, .py, and any other source files which are compiled or interpreted.
Files:
tests/unittest/trt/quantization/test_moe_weight_only_groupwise_quant_matmul.py
🧠 Learnings (2)
📚 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/trt/quantization/test_moe_weight_only_groupwise_quant_matmul.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/trt/quantization/test_moe_weight_only_groupwise_quant_matmul.py
🪛 Ruff (0.12.2)
tests/unittest/trt/quantization/test_moe_weight_only_groupwise_quant_matmul.py
238-238: Ambiguous variable name: I
(E741)
⏰ 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 (4)
tests/unittest/trt/quantization/test_moe_weight_only_groupwise_quant_matmul.py (4)
17-18: LGTM! Pytest import added for conditional test skipping.The pytest import is correctly added to support the conditional test skipping functionality for Hopper GPU architecture handling.
24-26: Good architecture support expansion.The utility imports have been appropriately updated to support both Ada and Hopper architectures, and the
get_sm_versionfunction import enables architecture-specific conditional logic.Also applies to: 31-32
72-73: Correct parameter derivation for interleaved weights/scales.Using
fc2_prequant_scale.shape[-1]to derive the parameternis the right approach since either weights or scales could be interleaved depending on the architecture, making the prequant scale a reliable source for the original dimension.
317-317: Test decorator correctly updated for expanded architecture support.The change from
skip_non_ada_unittesttoskip_neither_ada_nor_hopper_unittestproperly reflects the expanded GPU architecture support for both Ada and Hopper in the W4A8 quantization tests.
|
PR_Github #14501 [ run ] triggered by Bot |
symphonylyh
changed the title
|
PR_Github #14501 [ run ] completed with state |
…VIDIA#6708) Signed-off-by: Haohang Huang <[email protected]>
3 participants
Summary by CodeRabbit
New Features
Bug Fixes
Tests
Description
MoE w4a8 (wINT4aFP8) groupwise quant status before this PR:
PyT path - support both Ada and Hopper
TRT path - kernels exist, but requires preprocessing on weights & scales. On Ada, it's supported; On Hopper, preprocessing is missing
After this PR:
TRT path supports w4a8 groupwise MoE quant on both Ada and Hopper. Hopper usage is demostrated with the provided unit test case. NOT added into E2E TRT run because PyT path is the recommended path for users.
Using PyT path to explain what preprocessing is needed:
On Ada: int4 weight preproc logic here (interleave), scale preproc logic here (FP16 w/o interleave)
On Hopper: int4 weight preproc (None, no interleave), scale preproc logic (BF16 w/ interleave. calc interleave factor here + dtype cast and do interleave here)
Solution: interleave weight_scale for Hopper path + disable weight interleave for Hopper path (but still need the subbyte transpose) + change in MoE layer definition to take the interleave factor into account.
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.