[#10137][feat] AutoDeploy FP8 MoE refactor #10138
Conversation
Signed-off-by: Neta Zmora <[email protected]>
Signed-off-by: Neta Zmora <[email protected]>
📝 WalkthroughWalkthroughRefactors MOE weight and scale parameter signatures from per-weight naming (w1/w2/w3) to FC layer naming (fc1/fc2) across custom operators, FX graph transformation, and tests. Updates internal argument handling, weight stacking helpers, and quantization logic to use the new parameter names. Adds VSCode debugger support in worker startup. Changes
Estimated code review effort🎯 3 (Moderate) | ⏱️ ~25 minutes Areas requiring extra attention:
Pre-merge checks and finishing touches❌ Failed checks (2 warnings)
✅ Passed checks (1 passed)
✨ Finishing touches
🧪 Generate unit tests (beta)
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: 0
🧹 Nitpick comments (3)
tensorrt_llm/executor/worker.py (1)
247-256: Consider adding timeout and clarifying TODOs.The debugging helper will block rank 0 indefinitely until a debugger connects. This could cause confusing hangs if the environment variable is accidentally left set in production.
Consider:
- Adding a timeout parameter to
debugpy.wait_for_client()to prevent indefinite blocking- Resolving the TODOs:
- Is
debugpy.listen(0)necessary? (The VSCode docs suggest it is required beforewait_for_client())- Should this use
envparameter in MpiPoolExecutor instead of runtime env manipulation?- Adding a log message to indicate that the worker is waiting for debugger attachment
🔎 Suggested improvements
if mpi_rank() == 0 and "VSCODE_DEBUGPY_ADAPTER_ENDPOINTS" in os.environ: # cf. https://github.com/microsoft/vscode-python-debugger/blob/f64b217c4d2445f7f55255b102de1d94c19cf450/bundled/scripts/noConfigScripts/debugpy#L4 os.environ["DEBUGPY_ADAPTER_ENDPOINTS"] = os.environ[ "VSCODE_DEBUGPY_ADAPTER_ENDPOINTS"] - # TODO: is listen() needed? - # TODO: Could use 'env' in MpiPoolExecutor instead (and if-env-contains-vscode...) import debugpy debugpy.listen(0) + logger.info("Rank 0 waiting for VSCode debugger to attach...") - debugpy.wait_for_client() + debugpy.wait_for_client() # listen() is required before wait_for_client()tensorrt_llm/_torch/auto_deploy/transform/library/fused_moe.py (2)
1281-1292: Consider explicit error handling for invalid parameter/buffer targets.The
get_param_or_bufferfunction could raise various exceptions if the target doesn't exist as either a parameter or buffer. While this may be acceptable if the caller guarantees valid targets, explicit error handling would make debugging easier.🔎 Suggested improvement
def get_param_or_buffer(target): """Get parameter or buffer by target name.""" try: return gm.get_parameter(target) except AttributeError: # It's a buffer, not a parameter - parts = target.rsplit(".", 1) - if len(parts) == 2: - mod = gm.get_submodule(parts[0]) - return getattr(mod, parts[1]) - else: - return getattr(gm, target) + try: + parts = target.rsplit(".", 1) + if len(parts) == 2: + mod = gm.get_submodule(parts[0]) + return getattr(mod, parts[1]) + else: + return getattr(gm, target) + except (AttributeError, KeyError) as e: + raise RuntimeError(f"Could not find parameter or buffer '{target}'") from e
1458-1463: Good runtime validation for scale consistency.The assertions ensure that all experts use the same input scales, which is a critical assumption since only the first element (
[0]) is used in subsequent computations. This catches configuration errors early.However, consider whether these should be runtime assertions or validation checks that provide more actionable error messages:
🔎 Alternative validation approach
- assert torch.all(w1_input_scale_stacked[0] == w1_input_scale_stacked), ( - "All w1 scales should have the same value." - ) - assert torch.all(w2_input_scale_stacked[0] == w2_input_scale_stacked), ( - "All w2 scales should have the same value." - ) + if not torch.all(w1_input_scale_stacked[0] == w1_input_scale_stacked): + raise ValueError( + f"All fc1 input scales must be identical across experts. " + f"Got scales: {w1_input_scale_stacked}" + ) + if not torch.all(w2_input_scale_stacked[0] == w2_input_scale_stacked): + raise ValueError( + f"All fc2 input scales must be identical across experts. " + f"Got scales: {w2_input_scale_stacked}" + )
📜 Review details
Configuration used: Path: .coderabbit.yaml
Review profile: CHILL
Plan: Pro
📒 Files selected for processing (4)
tensorrt_llm/_torch/auto_deploy/custom_ops/fused_moe/trtllm_moe.py(3 hunks)tensorrt_llm/_torch/auto_deploy/transform/library/fused_moe.py(3 hunks)tensorrt_llm/executor/worker.py(1 hunks)tests/unittest/_torch/auto_deploy/unit/singlegpu/custom_ops/test_trtllm_moe.py(1 hunks)
🧰 Additional context used
📓 Path-based instructions (2)
**/*.py
📄 CodeRabbit inference engine (CODING_GUIDELINES.md)
**/*.py: 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 in Python, even if only one class or function from a module is used
Python files should use snake_case naming:some_file.py
Python classes should use PascalCase naming:class SomeClass
Python functions and methods should use snake_case naming:def my_awesome_function():
Python local variables should use snake_case naming:my_variable = ...
Python variable names that start with a number should be prefixed with 'k':k_99th_percentile = ...
Python global variables should use upper snake_case with prefix 'G':G_MY_GLOBAL = ...
Python constants should use upper snake_case naming: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
Python comments should be reserved for code within a function, or interfaces that are local to a file
Use Google style docstrings in Python for classes and functions, which can be parsed by Sphinx
Python attributes and variables can be documented inline with type and description
Avoid using reflection in Python when functionality can be easily achieved without reflection
When using try-except blocks in Python, limit the except 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, using the else block for logic
Files:
tensorrt_llm/executor/worker.pytests/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
**/*.{cpp,h,cu,cuh,py}
📄 CodeRabbit inference engine (CODING_GUIDELINES.md)
All TensorRT-LLM Open Source Software code should contain an NVIDIA copyright header that includes the year of its latest meaningful modification
Files:
tensorrt_llm/executor/worker.pytests/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
🧠 Learnings (12)
📓 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: Fridah-nv
Repo: NVIDIA/TensorRT-LLM PR: 7227
File: tensorrt_llm/_torch/auto_deploy/utils/quantization_utils.py:94-100
Timestamp: 2025-08-27T16:22:10.695Z
Learning: When there are inconsistent operator detection methods (like custom_op() vs target_op()), removing one method and standardizing on the other is often cleaner than supporting both methods simultaneously.
📚 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: 2025-11-14T11:22:03.729Z
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.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: 2025-08-09T20:57:04.084Z
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: 2025-10-20T16:54:09.824Z
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:
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: 2025-08-08T22:03:40.707Z
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:
tests/unittest/_torch/auto_deploy/unit/singlegpu/custom_ops/test_trtllm_moe.pytensorrt_llm/_torch/auto_deploy/transform/library/fused_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:
tests/unittest/_torch/auto_deploy/unit/singlegpu/custom_ops/test_trtllm_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:
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: 2025-08-19T12:45:11.997Z
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:
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: 2025-08-19T03:35:20.866Z
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:
tests/unittest/_torch/auto_deploy/unit/singlegpu/custom_ops/test_trtllm_moe.pytensorrt_llm/_torch/auto_deploy/transform/library/fused_moe.py
📚 Learning: 2025-10-20T17:07:18.745Z
Learnt from: nvchenghaoz
Repo: NVIDIA/TensorRT-LLM PR: 8469
File: tensorrt_llm/_torch/auto_deploy/models/patches/nemotron_h.py:98-116
Timestamp: 2025-10-20T17:07:18.745Z
Learning: In NemotronH models (tensorrt_llm/_torch/auto_deploy/models/patches/nemotron_h.py), the gate (self.gate) returns topk_indices and topk_weights that are already in the correct shape to be passed directly to torch_ops.auto_deploy.torch_moe without needing to reshape them when hidden_states is flattened.
Applied to files:
tensorrt_llm/_torch/auto_deploy/transform/library/fused_moe.py
🧬 Code graph analysis (4)
tensorrt_llm/executor/worker.py (1)
tensorrt_llm/_utils.py (1)
mpi_rank(527-534)
tests/unittest/_torch/auto_deploy/unit/singlegpu/custom_ops/test_trtllm_moe.py (1)
tensorrt_llm/_torch/modules/fused_moe/configurable_moe.py (1)
w2_weight(1046-1048)
tensorrt_llm/_torch/auto_deploy/custom_ops/fused_moe/trtllm_moe.py (1)
tensorrt_llm/_torch/auto_deploy/custom_ops/fused_moe/triton_moe.py (1)
_quantize_fp8(643-647)
tensorrt_llm/_torch/auto_deploy/transform/library/fused_moe.py (3)
tensorrt_llm/module.py (1)
register_parameter(186-190)tensorrt_llm/_torch/auto_deploy/utils/node_utils.py (2)
extract_op_args(535-572)is_op(198-221)tensorrt_llm/functional.py (1)
stack(1961-2007)
🪛 Ruff (0.14.8)
tensorrt_llm/executor/worker.py
253-253: Import for debugpy found
(T100)
254-254: Trace found: debugpy.listen used
(T100)
255-255: Trace found: debugpy.wait_for_client used
(T100)
tensorrt_llm/_torch/auto_deploy/custom_ops/fused_moe/trtllm_moe.py
230-230: Unused function argument: fc1_expert_weights
(ARG001)
231-231: Unused function argument: fc2_expert_weights
(ARG001)
232-232: Unused function argument: fc1_act_scale
(ARG001)
233-233: Unused function argument: gemm1_dequant
(ARG001)
234-234: Unused function argument: gemm2_act_quant
(ARG001)
235-235: Unused function argument: gemm2_dequant
(ARG001)
tensorrt_llm/_torch/auto_deploy/transform/library/fused_moe.py
1330-1330: Avoid specifying long messages outside the exception class
(TRY003)
⏰ 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 (10)
tests/unittest/_torch/auto_deploy/unit/singlegpu/custom_ops/test_trtllm_moe.py (1)
415-425: LGTM! Test correctly reflects the new fc1/fc2 parameter naming.The test properly exercises the mlp_style-driven selection of fc1_expert_weights and uses the new scale parameter names (fc1_act_scale, fc1_dequant_scale, fc2_act_scale_reciprocal, fc2_dequant_scale) that align with the refactored API.
tensorrt_llm/_torch/auto_deploy/custom_ops/fused_moe/trtllm_moe.py (4)
128-155: LGTM! Docstring accurately describes the refactored API.The updated documentation clearly explains the FC1/FC2 layer semantics, weight shapes, and the distinction between gated_mlp and mlp styles. The notes about precomputed scale arguments are helpful for understanding the optimization.
212-214: LGTM! Contiguous memory layout ensures kernel compatibility.Calling
.contiguous()onfc2_expert_weightsbefore passing to the TensorRT-LLM kernel is appropriate, as CUTLASS kernels typically require contiguous memory layouts.
230-238: LGTM! Fake registration correctly matches the updated API.The fake implementation's signature matches the real function, and it appropriately validates
mlp_styleandact_fnbefore returning a dummy tensor. The Ruff warnings about unused arguments are expected for fake implementations and can be safely ignored.
119-124: Add assertions to validate all scale tensor shapes before use.Lines 177-178 only assert that
fc1_dequant_scaleandfc2_dequant_scaleare 1D tensors. However,fc1_act_scaleandfc2_act_scale_reciprocalare also used as scale parameters without shape validation. Add assertions to verifyfc1_act_scaleis 1D (per docstring at line 146) andfc2_act_scale_reciprocalis 1D before they are used in_quantize_fp8at line 163 and assembled intoquant_scalesat line 179.⛔ Skipped due to learnings
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.tensorrt_llm/_torch/auto_deploy/transform/library/fused_moe.py (5)
1312-1315: LGTM! Clean weight stacking implementation.The
_stackhelper efficiently stacks weights from the parameter list and ensures contiguous memory layout, which is essential for kernel compatibility.
1317-1366: LGTM! Cutlass format argument preparation is well-structured.The function correctly:
- Handles both gated_mlp (concatenates w3+w1) and mlp (uses w1 only) styles
- Precomputes scale combinations to avoid runtime overhead
- Registers stacked tensors with unique keys
- Builds the argument tuple within the proper graph insertion context
The precomputed scales match the documentation in
trtllm_moe.py.
1418-1421: Good optimization: pre-filtering matched nodes.Creating a list of matched nodes upfront (line 1418-1420) before iterating is more efficient than checking
is_opfor every node in the graph. This avoids redundant checks during the loop.
1494-1503: LGTM! Proper cleanup after transformation.The code correctly:
- Replaces the old node with the new fused node
- Erases the old node from the graph
- Eliminates dead code after all nodes are processed
- Deletes unused submodules/parameters
This cleanup strategy is appropriate for the transformation and matches the pattern in
_insert_fused_moe_ops.
1479-1483: Ensure Triton backend supports all mlp_style variants or add validation.The code prepares arguments differently for Cutlass (trtllm) vs Triton backends. However, the Triton backend (
torch.ops.auto_deploy.triton_quant_fp8_moe) only supportsmlp_style=="mlp"and will raiseNotImplementedErrorforgated_mlp. The_prepare_args_triton_format()function should either validate that only supported styles are passed or ensure the backend can handle all variants that the Cutlass path supports.
|
/bot run |
|
PR_Github #29034 [ run ] triggered by Bot. Commit: |
|
PR_Github #29034 [ run ] completed with state
|
2 participants
The trtllm (cutlass) fp8 moe operator performs W3+W1 fusion (concat) during inference and we want to move this fusion to the model optimization time.
Summary by CodeRabbit
Release Notes
API Changes
fc1_expert_weightsandfc2_expert_weightswith new quantization scale arguments for improved consistency and flexibility.Chores
✏️ Tip: You can customize this high-level summary in your review settings.
Description
Test Coverage
PR Checklist
Please review the following before submitting your PR:
PR description clearly explains what and why. If using CodeRabbit's summary, please make sure it makes sense.
PR Follows TRT-LLM CODING GUIDELINES to the best of your knowledge.
Test cases are provided for new code paths (see test instructions)
Any new dependencies have been scanned for license and vulnerabilities
CODEOWNERS updated if ownership changes
Documentation updated as needed
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.
Details
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.