[None][fix] AutoDeploy: Fix the nvfp4 fused_moe#10727
[None][fix] AutoDeploy: Fix the nvfp4 fused_moe#10727nvchenghaoz merged 6 commits intoNVIDIA:mainfrom
Conversation
Signed-off-by: nvchenghaoz <211069071+nvchenghaoz@users.noreply.github.com>
Signed-off-by: nvchenghaoz <211069071+nvchenghaoz@users.noreply.github.com>
Signed-off-by: nvchenghaoz <211069071+nvchenghaoz@users.noreply.github.com>
📝 WalkthroughWalkthroughThis pull request enables NVFP4 MOE post-load fusion optimization, refactors weight tensor padding to use dynamic calculations instead of rigid validation, changes dtype handling to preserve byte representations via view operations, and adds mixer gate patterns to quantization config exclusions. Test coverage for NVFP4 is added with a new integration test and expanded unit tests. Changes
Estimated code review effort🎯 3 (Moderate) | ⏱️ ~25 minutes 🚥 Pre-merge checks | ✅ 1 | ❌ 2❌ Failed checks (1 warning, 1 inconclusive)
✅ Passed checks (1 passed)
✏️ Tip: You can configure your own custom pre-merge checks in the settings. ✨ Finishing touches
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 |
![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
🤖 Fix all issues with AI agents
In `@tests/integration/defs/accuracy/test_llm_api_autodeploy.py`:
- Around line 242-244: The comment above the manual quant_config change
incorrectly says "FP8 model" but the test sets llm.args.quant_config.quant_algo
= QuantAlgo.NVFP4 (with kv_cache_quant_algo = QuantAlgo.FP8); update that
comment to reference NVFP4 (or NVFP4 model) so it accurately describes the
quantization being set in the test (look for the comment immediately above the
lines touching llm.args.quant_config.quant_algo and
llm.args.quant_config.kv_cache_quant_algo and change "FP8 model" to "NVFP4
model" or equivalent).
🧹 Nitpick comments (2)
tests/unittest/_torch/auto_deploy/unit/singlegpu/custom_ops/test_trtllm_moe.py (1)
998-1009: Consider removing debug print statements for cleaner CI output.These debug prints are useful during development but may be verbose in CI logs. Consider removing them or guarding with a debug flag if needed for future debugging.
tests/integration/defs/accuracy/test_llm_api_autodeploy.py (1)
236-237: Add the same memory guard used by other MOE tests.Line 236: once the skip is removed, this test could run without the 32GB guard used by
test_bf16/test_fp8. Add@pytest.mark.skip_less_device_memory(32000)now to prevent accidental OOMs.Proposed patch
- `@pytest.mark.skip`(reason="NVFP4 model is not in the CI yet") + `@pytest.mark.skip_less_device_memory`(32000) + `@pytest.mark.skip`(reason="NVFP4 model is not in the CI yet") def test_nvfp4(self):
📜 Review details
Configuration used: Path: .coderabbit.yaml
Review profile: CHILL
Plan: Pro
📒 Files selected for processing (6)
tensorrt_llm/_torch/auto_deploy/config/default.yamltensorrt_llm/_torch/auto_deploy/custom_ops/fused_moe/trtllm_moe.pytensorrt_llm/_torch/auto_deploy/models/quant_config_reader.pytensorrt_llm/_torch/auto_deploy/transform/library/fused_moe.pytests/integration/defs/accuracy/test_llm_api_autodeploy.pytests/unittest/_torch/auto_deploy/unit/singlegpu/custom_ops/test_trtllm_moe.py
🧰 Additional context used
📓 Path-based instructions (2)
**/*.py
📄 CodeRabbit inference engine (CODING_GUIDELINES.md)
**/*.py: The code developed for TensorRT-LLM should conform to Python 3.8+
Indent Python code with 4 spaces. Do not use tabs
Always maintain the namespace when importing Python modules, 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, with prefixkfor variable names that start with a number (e.g.,k_99th_percentile)
Python global variables should use upper snake_case with prefixG(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 Python interfaces that may be used outside a file, prefer docstrings over comments
Use comments in Python 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
Python attributes and variables can be documented inline with the format"""<type>: Description"""
Avoid using reflection in Python when functionality can be easily achieved without reflection
When using try-except blocks in Python, limit the except clause to the smallest set of errors possible
When using try-except blocks in Python to handle multiple possible variable types (duck-typing), keep the body of the try as small as possible and use the else block for the main logic
Files:
tensorrt_llm/_torch/auto_deploy/models/quant_config_reader.pytests/unittest/_torch/auto_deploy/unit/singlegpu/custom_ops/test_trtllm_moe.pytests/integration/defs/accuracy/test_llm_api_autodeploy.pytensorrt_llm/_torch/auto_deploy/custom_ops/fused_moe/trtllm_moe.pytensorrt_llm/_torch/auto_deploy/transform/library/fused_moe.py
**/*.{cpp,cc,cxx,h,hpp,hxx,cu,cuh,py}
📄 CodeRabbit inference engine (CODING_GUIDELINES.md)
All TensorRT-LLM source files (.cpp, .h, .cu, .py, and other source files) should contain an NVIDIA copyright header with the year of latest meaningful modification
Files:
tensorrt_llm/_torch/auto_deploy/models/quant_config_reader.pytests/unittest/_torch/auto_deploy/unit/singlegpu/custom_ops/test_trtllm_moe.pytests/integration/defs/accuracy/test_llm_api_autodeploy.pytensorrt_llm/_torch/auto_deploy/custom_ops/fused_moe/trtllm_moe.pytensorrt_llm/_torch/auto_deploy/transform/library/fused_moe.py
🧠 Learnings (22)
📓 Common learnings
Learnt from: djns99
Repo: NVIDIA/TensorRT-LLM PR: 6915
File: cpp/tensorrt_llm/kernels/cutlass_kernels/moe_gemm/moe_kernels.cu:4010-4012
Timestamp: 2025-08-14T23:23:27.449Z
Learning: For MOE (Mixture of Experts) code reviews in TensorRT-LLM, avoid repeatedly suggesting finalize fusion validation checks and safety assertions. The user djns99 has indicated these suggestions are repetitive and unwanted across multiple MOE-related changes.
Learnt from: sklevtsov-nvidia
Repo: NVIDIA/TensorRT-LLM PR: 3294
File: cpp/tensorrt_llm/kernels/cutlass_kernels/moe_gemm/moe_gemm_tma_warp_specialized_input.cu:118-127
Timestamp: 2025-08-09T20:57:04.084Z
Learning: In the CUTLASS MoE finalize fusion implementation (cpp/tensorrt_llm/kernels/cutlass_kernels/moe_gemm/moe_gemm_tma_warp_specialized_input.cu), when setting `fused_finalize_epilogue.stride_final_output` with shape `(hidden_size, num_output_tokens, 1)`, the `num_rows_in_final_output` should be set to `num_output_tokens` (not `hidden_size`) because of a swap+transpose operation that maps rows of the output tensor to `hidden_size` and columns to `num_output_tokens`.
📚 Learning: 2025-12-19T06:31:54.973Z
Learnt from: nvyocox
Repo: NVIDIA/TensorRT-LLM PR: 10117
File: tensorrt_llm/_torch/auto_deploy/transform/library/fuse_rope_attention.py:336-339
Timestamp: 2025-12-19T06:31:54.973Z
Learning: In tensorrt_llm/_torch/auto_deploy/transform/library/fuse_rope_attention.py, the cast to torch.float16 for qkv_node before creating the AttentionPlugin is intentional and required because DriveOS LLM expects float16 dtype specifically. This should not be changed to preserve original dtype or made configurable for bfloat16 models in the DriveOS LLM ONNX export path.
Applied to files:
tensorrt_llm/_torch/auto_deploy/config/default.yamltensorrt_llm/_torch/auto_deploy/transform/library/fused_moe.py
📚 Learning: 2025-09-23T15:12:38.312Z
Learnt from: nv-lschneider
Repo: NVIDIA/TensorRT-LLM PR: 7910
File: cpp/tensorrt_llm/thop/allreduceOp.cpp:352-446
Timestamp: 2025-09-23T15:12:38.312Z
Learning: In TensorRT-LLM NCCL device allreduce implementation (cpp/tensorrt_llm/thop/allreduceOp.cpp), the goto pattern in runNCCLAllReduceDeviceFusion is intentionally used for future extensibility, allowing multiple switch cases to fallback to the default handler. While not aesthetically ideal, this pattern supports adding more fusion cases later that can reuse the same fallback logic.
Applied to files:
tensorrt_llm/_torch/auto_deploy/config/default.yaml
📚 Learning: 2025-08-08T22:03:40.707Z
Learnt from: sklevtsov-nvidia
Repo: NVIDIA/TensorRT-LLM PR: 3294
File: cpp/tensorrt_llm/kernels/cutlass_kernels/moe_gemm/moe_kernels.cu:1198-1209
Timestamp: 2025-08-08T22:03:40.707Z
Learning: In the CUTLASS MoE kernels (cpp/tensorrt_llm/cutlass_extensions), when `layout_info.fusion` is set to `TmaWarpSpecializedGroupedGemmInput::EpilogueFusion::FINALIZE`, the `router_scales` parameter must be non-null by design. The fused finalize kernel epilogue does not perform nullptr checks and requires valid router scales to function correctly. This is an implicit contract that callers must satisfy when enabling the FINALIZE fusion mode.
Applied to files:
tensorrt_llm/_torch/auto_deploy/config/default.yamltensorrt_llm/_torch/auto_deploy/custom_ops/fused_moe/trtllm_moe.pytensorrt_llm/_torch/auto_deploy/transform/library/fused_moe.py
📚 Learning: 2025-10-20T16:54:09.824Z
Learnt from: nvchenghaoz
Repo: NVIDIA/TensorRT-LLM PR: 8469
File: tensorrt_llm/_torch/auto_deploy/custom_ops/rms_norm.py:6-6
Timestamp: 2025-10-20T16:54:09.824Z
Learning: In tensorrt_llm/_torch/auto_deploy/custom_ops/rms_norm.py, the import `from ...modules.mamba.layernorm_gated import _layer_norm_fwd` is correct and should not be changed to modules.fla.layernorm_gated. The _layer_norm_fwd function exists in both modules/mamba/layernorm_gated.py and modules/fla/layernorm_gated.py, but the mamba version is the intended implementation for this use case.
Applied to files:
tensorrt_llm/_torch/auto_deploy/models/quant_config_reader.pytests/unittest/_torch/auto_deploy/unit/singlegpu/custom_ops/test_trtllm_moe.pytensorrt_llm/_torch/auto_deploy/custom_ops/fused_moe/trtllm_moe.py
📚 Learning: 2025-08-14T23:23:27.449Z
Learnt from: djns99
Repo: NVIDIA/TensorRT-LLM PR: 6915
File: cpp/tensorrt_llm/kernels/cutlass_kernels/moe_gemm/moe_kernels.cu:4010-4012
Timestamp: 2025-08-14T23:23:27.449Z
Learning: For MOE (Mixture of Experts) code reviews in TensorRT-LLM, avoid repeatedly suggesting finalize fusion validation checks and safety assertions. The user djns99 has indicated these suggestions are repetitive and unwanted across multiple MOE-related changes.
Applied to files:
tests/unittest/_torch/auto_deploy/unit/singlegpu/custom_ops/test_trtllm_moe.pytensorrt_llm/_torch/auto_deploy/custom_ops/fused_moe/trtllm_moe.pytensorrt_llm/_torch/auto_deploy/transform/library/fused_moe.py
📚 Learning: 2025-09-09T09:40:45.658Z
Learnt from: fredricz-20070104
Repo: NVIDIA/TensorRT-LLM PR: 7645
File: tests/integration/test_lists/qa/llm_function_core.txt:648-648
Timestamp: 2025-09-09T09:40:45.658Z
Learning: In TensorRT-LLM test lists, it's common and intentional for the same test to appear in multiple test list files when they serve different purposes (e.g., llm_function_core.txt for comprehensive core functionality testing and llm_function_core_sanity.txt for quick sanity checks). This duplication allows tests to be run in different testing contexts.
Applied to files:
tests/unittest/_torch/auto_deploy/unit/singlegpu/custom_ops/test_trtllm_moe.py
📚 Learning: 2025-08-09T20:57:04.084Z
Learnt from: sklevtsov-nvidia
Repo: NVIDIA/TensorRT-LLM PR: 3294
File: cpp/tensorrt_llm/kernels/cutlass_kernels/moe_gemm/moe_gemm_tma_warp_specialized_input.cu:118-127
Timestamp: 2025-08-09T20:57:04.084Z
Learning: In the CUTLASS MoE finalize fusion implementation (cpp/tensorrt_llm/kernels/cutlass_kernels/moe_gemm/moe_gemm_tma_warp_specialized_input.cu), when setting `fused_finalize_epilogue.stride_final_output` with shape `(hidden_size, num_output_tokens, 1)`, the `num_rows_in_final_output` should be set to `num_output_tokens` (not `hidden_size`) because of a swap+transpose operation that maps rows of the output tensor to `hidden_size` and columns to `num_output_tokens`.
Applied to files:
tests/unittest/_torch/auto_deploy/unit/singlegpu/custom_ops/test_trtllm_moe.pytensorrt_llm/_torch/auto_deploy/custom_ops/fused_moe/trtllm_moe.pytensorrt_llm/_torch/auto_deploy/transform/library/fused_moe.py
📚 Learning: 2025-08-21T02:39:12.009Z
Learnt from: djns99
Repo: NVIDIA/TensorRT-LLM PR: 7104
File: cpp/tensorrt_llm/kernels/cutlass_kernels/moe_gemm/moe_kernels.cu:1475-1480
Timestamp: 2025-08-21T02:39:12.009Z
Learning: The min latency mode functionality in TensorRT-LLM MOE kernels (cpp/tensorrt_llm/kernels/cutlass_kernels/moe_gemm/moe_kernels.cu) is deprecated and no longer being maintained/updated, as confirmed by djns99. Bug reports and optimization suggestions for the computeStridesTmaWarpSpecializedLowLatencyKernel and related min latency code paths should be deprioritized.
Applied to files:
tests/unittest/_torch/auto_deploy/unit/singlegpu/custom_ops/test_trtllm_moe.pytensorrt_llm/_torch/auto_deploy/custom_ops/fused_moe/trtllm_moe.pytensorrt_llm/_torch/auto_deploy/transform/library/fused_moe.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/auto_deploy/unit/singlegpu/custom_ops/test_trtllm_moe.pytests/integration/defs/accuracy/test_llm_api_autodeploy.py
📚 Learning: 2025-08-08T04:10:19.038Z
Learnt from: djns99
Repo: NVIDIA/TensorRT-LLM PR: 6728
File: cpp/tensorrt_llm/plugins/mixtureOfExperts/mixtureOfExpertsPlugin.cpp:966-966
Timestamp: 2025-08-08T04:10:19.038Z
Learning: TensorRT plugins currently don't support padding functionality, and TensorRT is not getting new features (in maintenance mode). This means that duplicating parameters like mExpertHiddenSize in function calls, even with TODO comments, can be acceptable as pragmatic solutions within these constraints.
Applied to files:
tests/unittest/_torch/auto_deploy/unit/singlegpu/custom_ops/test_trtllm_moe.pytensorrt_llm/_torch/auto_deploy/custom_ops/fused_moe/trtllm_moe.py
📚 Learning: 2025-08-06T13:58:07.506Z
Learnt from: galagam
Repo: NVIDIA/TensorRT-LLM PR: 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/auto_deploy/unit/singlegpu/custom_ops/test_trtllm_moe.py
📚 Learning: 2025-08-29T14:07:45.863Z
Learnt from: EmmaQiaoCh
Repo: NVIDIA/TensorRT-LLM PR: 7370
File: tests/unittest/trt/model_api/test_model_quantization.py:24-27
Timestamp: 2025-08-29T14:07:45.863Z
Learning: In TensorRT-LLM's CI infrastructure, pytest skip markers (pytest.mark.skip) are properly honored even when test files have __main__ blocks that call test functions directly. The testing system correctly skips tests without requiring modifications to the __main__ block execution pattern.
Applied to files:
tests/unittest/_torch/auto_deploy/unit/singlegpu/custom_ops/test_trtllm_moe.py
📚 Learning: 2025-11-14T11:22:03.729Z
Learnt from: nzmora-nvidia
Repo: NVIDIA/TensorRT-LLM PR: 9163
File: tensorrt_llm/_torch/auto_deploy/custom_ops/quant.py:107-113
Timestamp: 2025-11-14T11:22:03.729Z
Learning: In TensorRT-LLM AutoDeploy custom ops, when adding hardware capability checks to select between kernel implementations (e.g., cuBLAS vs. CUDA kernel), use descriptive variable names that identify the specific GPU architectures or families being targeted (e.g., `is_blackwell_geforce_or_ada`) rather than generic names like `enable_cuda_core`. This makes it clear that the code is selecting an implementation path based on hardware capabilities, not enabling/disabling hardware features.
Applied to files:
tests/unittest/_torch/auto_deploy/unit/singlegpu/custom_ops/test_trtllm_moe.pytensorrt_llm/_torch/auto_deploy/custom_ops/fused_moe/trtllm_moe.pytensorrt_llm/_torch/auto_deploy/transform/library/fused_moe.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/integration/defs/accuracy/test_llm_api_autodeploy.py
📚 Learning: 2025-08-19T03:35:20.866Z
Learnt from: djns99
Repo: NVIDIA/TensorRT-LLM PR: 6915
File: cpp/tensorrt_llm/kernels/cutlass_kernels/moe_gemm/moe_kernels.cu:4616-4626
Timestamp: 2025-08-19T03:35:20.866Z
Learning: In the MOE profiler TMA workspace preparation (cpp/tensorrt_llm/kernels/cutlass_kernels/moe_gemm/moe_kernels.cu), the overlapping of TMA WS regions for NONE and FINALIZE variants is deliberate design to save memory space, as confirmed by djns99. The comment "reuse the same pointers to save space" reflects this intentional behavior.
Applied to files:
tensorrt_llm/_torch/auto_deploy/custom_ops/fused_moe/trtllm_moe.pytensorrt_llm/_torch/auto_deploy/transform/library/fused_moe.py
📚 Learning: 2025-09-19T21:28:13.751Z
Learnt from: jhaotingc
Repo: NVIDIA/TensorRT-LLM PR: 7856
File: cpp/tensorrt_llm/thop/fp8BlockScaleMoe.cpp:159-166
Timestamp: 2025-09-19T21:28:13.751Z
Learning: In TensorRT-LLM blockScaleMoe routing (cpp/tensorrt_llm/kernels/trtllmGenKernels/blockScaleMoe/runner.cu), the DeepSeek routing method performs reinterpret_cast<float*>(routingLogits) at line 89, which could cause issues if routing_logits are BF16. However, Qwen3-FP8 models use RenormalizeNaive routing method and are not affected by this dtype casting issue.
Applied to files:
tensorrt_llm/_torch/auto_deploy/custom_ops/fused_moe/trtllm_moe.pytensorrt_llm/_torch/auto_deploy/transform/library/fused_moe.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/auto_deploy/custom_ops/fused_moe/trtllm_moe.pytensorrt_llm/_torch/auto_deploy/transform/library/fused_moe.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/auto_deploy/custom_ops/fused_moe/trtllm_moe.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/auto_deploy/custom_ops/fused_moe/trtllm_moe.py
📚 Learning: 2025-08-08T05:10:38.906Z
Learnt from: sklevtsov-nvidia
Repo: NVIDIA/TensorRT-LLM PR: 3294
File: cpp/tensorrt_llm/cutlass_extensions/include/cutlass_extensions/epilogue/fusion/sm90_visitor_scatter.hpp:0-0
Timestamp: 2025-08-08T05:10:38.906Z
Learning: The ScaledAccPerRowBiasPerColScaleScatter fusion in CUTLASS extensions (cpp/tensorrt_llm/cutlass_extensions/include/cutlass_extensions/epilogue/fusion/sm90_visitor_scatter.hpp) is specifically designed for per-column scaling factors only, so it uses a fixed Stride<_0,_1,int64_t> rather than conditional stride logic.
Applied to files:
tensorrt_llm/_torch/auto_deploy/custom_ops/fused_moe/trtllm_moe.py
📚 Learning: 2025-10-20T17:09:21.560Z
Learnt from: nvchenghaoz
Repo: NVIDIA/TensorRT-LLM PR: 8469
File: tensorrt_llm/_torch/auto_deploy/transform/library/rms_norm.py:180-182
Timestamp: 2025-10-20T17:09:21.560Z
Learning: In tensorrt_llm/_torch/auto_deploy/transform/library/rms_norm.py, the _gated_rmsnorm_replacement function does not need to cast the output of torch.ops.auto_deploy.torch_rmsnorm_gated back to the input dtype, even though the custom op returns fp32. The dtype handling is managed elsewhere or the fp32 output is acceptable for downstream consumers.
Applied to files:
tensorrt_llm/_torch/auto_deploy/transform/library/fused_moe.py
🧬 Code graph analysis (1)
tests/integration/defs/accuracy/test_llm_api_autodeploy.py (2)
tensorrt_llm/_torch/auto_deploy/llm_args.py (2)
quant_config(388-391)quant_config(394-395)tensorrt_llm/quantization/mode.py (1)
QuantAlgo(23-48)
🪛 Gitleaks (8.30.0)
tests/unittest/_torch/auto_deploy/unit/singlegpu/custom_ops/test_trtllm_moe.py
[high] 836-836: 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 (11)
tensorrt_llm/_torch/auto_deploy/config/default.yaml (1)
129-131: LGTM! Re-enabling NVFP4 MOE fusion after accuracy fix.The PR objectives indicate this path was disabled due to an accuracy issue that has now been fixed in the accompanying code changes. Enabling this post-load fusion optimization is appropriate.
tensorrt_llm/_torch/auto_deploy/models/quant_config_reader.py (1)
86-86: LGTM! Adding mixer gate exclusion pattern for quantization.The
"*.mixer.gate*"pattern appropriately excludes gate modules from quantization, which aligns with the NVFP4 MOE accuracy fix objectives. Gate modules in MoE architectures typically need to remain in higher precision for proper routing behavior.tensorrt_llm/_torch/auto_deploy/transform/library/fused_moe.py (2)
1683-1685: LGTM! Correctly preserving packed FP4 weight byte representation.Registering
fc1_expert_weightsandfc2_expert_weightswithout dtype conversion is correct. FP4 weights are already packed as uint8, and dtype conversion would corrupt the packed representation.
1750-1755: Core accuracy fix: Using.view()instead of.to()for FP8 blockscale reinterpretation.This is the critical fix for the NVFP4 MOE accuracy issue. Block scales are stored as raw bytes (uint8) representing FP8 E4M3 values. Using
.view(torch.float8_e4m3fn)correctly reinterprets the byte pattern, whereas the previous.to()would have performed numeric conversion, corrupting the scale values.tensorrt_llm/_torch/auto_deploy/custom_ops/fused_moe/trtllm_moe.py (4)
274-276: LGTM! Relaxed validation to support dynamic padding.Removing the rigid 1D/2D blockscale size assertions allows the function to handle varying input shapes. The 3D tensor check is sufficient as padding logic below handles dimension requirements.
291-306: LGTM! Dynamic padding calculations correctly align to kernel requirements.The padded sizes and block counts are correctly computed:
- Sizes align to
TRTLLM_NVFP4_ROW_SIZE(128) andTRTLLM_NVFP4_COLUMN_SIZE(128)- Block counts are derived by dividing padded sizes by
NVFP4_BLOCK_SIZE(16)This enables support for non-aligned dimensions like Nemotron-Nano's 2688×1856.
318-326: LGTM! Conditional blockscale padding avoids redundant operations.The conditional check (
fc1_bs_size1 < fc1_inter_size_padded or fc1_bs_size2 < hidden_blocks_padded) correctly handles cases where block scales may already be padded by the upstream quantization process, avoiding unnecessary tensor allocations.
339-349: LGTM! FC2 blockscale padding follows consistent pattern.The fc2 blockscale padding correctly mirrors the fc1 logic with appropriate dimension targets (
hidden_size_padded,inter_blocks_padded). The comment at line 349 helpfully documents the else-case behavior.tests/unittest/_torch/auto_deploy/unit/singlegpu/custom_ops/test_trtllm_moe.py (3)
585-585: LGTM! Comment update reflects new padding support.The comment correctly notes that padding for non-divisible intermediate_size is now supported.
793-815: LGTM! Comprehensive test for NVFP4 weight stacking transform.The test covers all three important cases:
(128, 128)- No padding needed(128, 160)- Small padded case(2688, 1856)- Nemotron-Nano dimensions requiring paddingThe comparison between
torch_quant_nvfp4_moe(reference) and the transformed graph output validates the stacking transform correctness.
833-886: LGTM! Well-documented quantization helper for test reference.The docstring clearly explains why manual block scale computation is necessary -
fp4_quantizereturns an internal interleaved layout incompatible with the fused MoE kernel. The manual computation followed byblock_scale_interleaveproduces the correct format for comparison.Note: The static analysis hint about "generic-api-key" at line 836 is a false positive - "Key insight" in the docstring is documentation, not a secret.
✏️ Tip: You can disable this entire section by setting review_details to false in your review settings.
Signed-off-by: nvchenghaoz <211069071+nvchenghaoz@users.noreply.github.com>
|
/bot run |
|
PR_Github #32180 [ run ] triggered by Bot. Commit: |
|
PR_Github #32180 [ run ] completed with state
|
Signed-off-by: nvchenghaoz <211069071+nvchenghaoz@users.noreply.github.com>
Signed-off-by: nvchenghaoz <211069071+nvchenghaoz@users.noreply.github.com>
|
/bot run |
|
PR_Github #32195 [ run ] triggered by Bot. Commit: |
|
/bot skip --comment "All tests / subpipeline passed but the pipeline failed somehow / no return" |
|
PR_Github #32342 [ skip ] triggered by Bot. Commit: |
|
PR_Github #32342 [ skip ] completed with state |
Signed-off-by: nvchenghaoz <211069071+nvchenghaoz@users.noreply.github.com>
3 participants
Fixed the accuracy issue in the nvfp4 fused_moe path, re-enable the fuse_nvfp4_moe path.
Added the test to compare the output between torch_moe vs. trtllm_moe.
Added the e2e testing. Since the model is not in the CI yet, skip the test for now.
Summary by CodeRabbit
New Features
Improvements
Tests
✏️ Tip: You can customize this high-level summary in your review settings.