[https://nvbugs/5537738][fix] Add fp8 post-quant allgather support #8008
Conversation
ChristinaZ
requested review from
QiJune and
pengbowang-nv
and removed request for
QiJune
ChristinaZ
changed the title
|
/bot run |
📝 WalkthroughWalkthroughRefactors fused MoE generation flow to simplify post-quant allgather gating, add explicit branches for DeepSeek FP8 and W4A16 MXFP4 (including conditional padding), adjust router/quantization inputs when allgather is enabled, and pass top-k routing data to FP8-related MoE runners. Removes a corresponding test waive entry. Changes
Sequence Diagram(s)sequenceDiagram
autonumber
participant C as Caller
participant F as FusedMoE (gen)
participant R as Router/Quant
participant M as MoE Runner
C->>F: forward(x, config, use_dp, parallel_size, ...)
Note over F: Compute run_post_quant_allgather = use_dp && parallel_size>1
alt has_w4a16_mxfp4
F->>F: Compute pad_size from w3_w1_weight & x.shape[-1]
alt pad needed AND NOT run_post_quant_allgather
F->>F: Pad x (and related tensors)
else No pad or allgather active
Note over F: Preserve input width
end
else has_deepseek_fp8_block_scales
Note over F: No-op branch for block scales
end
F->>R: Prepare routing/quantization
alt run_post_quant_allgather == true
Note over R: Post-quant allgather path
F->>M: router_logits=None, routing_bias=None, topk_ids, topk_weights
else run_post_quant_allgather == false
F->>M: router_logits, routing_bias, topk_ids, topk_weights
end
alt FP8 (DeepSeek) or BF16 MXE2M1 path
Note over M: Pass topk_ids/topk_weights to runner
else Other modes
Note over M: Standard execution
end
M-->>C: outputs
Estimated code review effort🎯 3 (Moderate) | ⏱️ ~20 minutes Pre-merge checks and finishing touches❌ Failed checks (1 warning)
✅ Passed checks (2 passed)
✨ Finishing touches
🧪 Generate unit tests
Tip 👮 Agentic pre-merge checks are now available in preview!Pro plan users can now enable pre-merge checks in their settings to enforce checklists before merging PRs.
Please see the documentation for more information. Example: reviews:
pre_merge_checks:
custom_checks:
- name: "Undocumented Breaking Changes"
mode: "warning"
instructions: |
Pass/fail criteria: All breaking changes to public APIs, CLI flags, environment variables, configuration keys, database schemas, or HTTP/GraphQL endpoints must be documented in the "Breaking Change" section of the PR description and in CHANGELOG.md. Exclude purely internal or private changes (e.g., code not exported from package entry points or explicitly marked as internal).Please share your feedback with us on this Discord post. 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
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/modules/fused_moe/fused_moe_trtllm_gen.py (1)
269-289: Pass float32 top‑k weights into DeepSeek FP8 runner
fp8_block_scale_moe_runnerstill reinterprets its top‑k weights buffer asfloat*. We now passtoken_final_scales.to(torch.bfloat16)whenrun_post_quant_allgatheris true, so the kernel consumes BF16 data via a float pointer, yielding corrupted gating probabilities and wrong expert activation. Keeptoken_final_scalesinfloat32for this path (and do so before the allgather to avoid mixed dtypes).Apply:
- token_selected_experts, token_final_scales = self.routing_method.apply( - router_logits) - token_final_scales = token_final_scales.to(torch.bfloat16) + token_selected_experts, token_final_scales = self.routing_method.apply( + router_logits) + if not self.has_deepseek_fp8_block_scales: + token_final_scales = token_final_scales.to(torch.bfloat16)This keeps the DeepSeek FP8 path in float32 while preserving the BF16 downcast for other modes that require it. Based on learnings.
🧹 Nitpick comments (1)
tensorrt_llm/_torch/modules/fused_moe/fused_moe_trtllm_gen.py (1)
341-345: Remove redundant no‑op assignment
x = xinside theelseadds noise without value. Please drop it.Apply:
- else: - x = x
📜 Review details
Configuration used: Path: .coderabbit.yaml
Review profile: CHILL
Plan: Pro
📒 Files selected for processing (2)
tensorrt_llm/_torch/modules/fused_moe/fused_moe_trtllm_gen.py(6 hunks)tests/integration/test_lists/waives.txt(0 hunks)
💤 Files with no reviewable changes (1)
- tests/integration/test_lists/waives.txt
🧰 Additional context used
📓 Path-based instructions (3)
**/*.{h,hpp,hh,hxx,cpp,cxx,cc,cu,cuh,py}
📄 CodeRabbit inference engine (CODING_GUIDELINES.md)
Use only spaces, no tabs; indent with 4 spaces.
Files:
tensorrt_llm/_torch/modules/fused_moe/fused_moe_trtllm_gen.py
**/*.py
📄 CodeRabbit inference engine (CODING_GUIDELINES.md)
**/*.py: Python code must target Python 3.8+.
Indent Python code with 4 spaces; do not use tabs.
Maintain module namespace when importing; prefer 'from package.subpackage import foo' then 'foo.SomeClass()' instead of importing the class directly.
Python filenames should be snake_case (e.g., some_file.py).
Python classes use PascalCase names.
Functions and methods use snake_case names.
Local variables use snake_case; prefix 'k' for variables that start with a number (e.g., k_99th_percentile).
Global variables use upper SNAKE_CASE prefixed with 'G' (e.g., G_MY_GLOBAL).
Constants use upper SNAKE_CASE (e.g., MY_CONSTANT).
Avoid shadowing variables from an outer scope.
Initialize all externally visible members of a class in the constructor.
Prefer docstrings for interfaces that may be used outside a file; comments for in-function or file-local interfaces.
Use Google-style docstrings for classes and functions (Sphinx-parsable).
Document attributes and variables inline so they render under the class/function docstring.
Avoid reflection when a simpler, explicit approach suffices (e.g., avoid dict(**locals()) patterns).
In try/except, catch the most specific exceptions possible.
For duck-typing try/except, keep the try body minimal and use else for the main logic.
Files:
tensorrt_llm/_torch/modules/fused_moe/fused_moe_trtllm_gen.py
**/*.{cpp,cxx,cc,h,hpp,hh,hxx,cu,cuh,py}
📄 CodeRabbit inference engine (CODING_GUIDELINES.md)
Prepend the NVIDIA Apache-2.0 copyright header with current year to the top of all source files (e.g., .cpp, .h, .cu, .py).
Files:
tensorrt_llm/_torch/modules/fused_moe/fused_moe_trtllm_gen.py
🧠 Learnings (4)
📓 Common learnings
Learnt from: jhaotingc
PR: NVIDIA/TensorRT-LLM#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.
Learnt from: sklevtsov-nvidia
PR: NVIDIA/TensorRT-LLM#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.
📚 Learning: 2025-09-19T21:28:13.751Z
Learnt from: jhaotingc
PR: NVIDIA/TensorRT-LLM#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/modules/fused_moe/fused_moe_trtllm_gen.py
📚 Learning: 2025-08-14T23:23:27.449Z
Learnt from: djns99
PR: NVIDIA/TensorRT-LLM#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:
tensorrt_llm/_torch/modules/fused_moe/fused_moe_trtllm_gen.py
📚 Learning: 2025-08-08T22:03:40.707Z
Learnt from: sklevtsov-nvidia
PR: NVIDIA/TensorRT-LLM#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/modules/fused_moe/fused_moe_trtllm_gen.py
🧬 Code graph analysis (1)
tensorrt_llm/_torch/modules/fused_moe/fused_moe_trtllm_gen.py (2)
tensorrt_llm/_torch/modules/fused_moe/interface.py (2)
has_deepseek_fp8_block_scales(290-293)has_w4a16_mxfp4(314-317)tensorrt_llm/_torch/custom_ops/trtllm_gen_custom_ops.py (1)
bf16_mxe2m1_block_scale_moe_runner(1355-1434)
⏰ 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). (3)
- GitHub Check: Check PR Title Format
- GitHub Check: Check PR Checklist Resolution
- GitHub Check: Pre-commit Check
|
PR_Github #20004 [ run ] triggered by Bot |
|
PR_Github #20004 [ run ] completed with state |
|
/bot run |
|
PR_Github #20029 [ run ] triggered by Bot |
|
PR_Github #20029 [ run ] completed with state |
|
/bot run |
|
PR_Github #20034 [ run ] triggered by Bot |
|
PR_Github #20034 [ run ] completed with state |
Signed-off-by: Christina Zhang <83400082+ChristinaZ@users.noreply.github.com>
ChristinaZ
force-pushed
the
fix_fp8_allgather
branch
from
3983d93 to
ee95076
Compare
|
/bot run |
|
PR_Github #20057 [ run ] triggered by Bot |
|
PR_Github #20057 [ run ] completed with state |
|
/bot run |
|
PR_Github #20100 [ run ] triggered by Bot |
|
PR_Github #20100 [ run ] completed with state |
…VIDIA#8008) Signed-off-by: Christina Zhang <83400082+ChristinaZ@users.noreply.github.com>
…VIDIA#8008) Signed-off-by: Christina Zhang <83400082+ChristinaZ@users.noreply.github.com>
…VIDIA#8008) Signed-off-by: Christina Zhang <83400082+ChristinaZ@users.noreply.github.com>
…VIDIA#8008) Signed-off-by: Christina Zhang <83400082+ChristinaZ@users.noreply.github.com>
4 participants
Summary by CodeRabbit
Bug Fixes
Improvements
Tests
Description
Fix the bug related to the fp8 post quant allgather for MoE TRTLLM backend
Test Coverage
pytest -s -o log_cli=true "tests/integration/defs/accuracy/test_llm_api_pytorch.py::TestDeepSeekR1::test_fp8_blockscale[throughput_mtp_trtllm]"
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
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.