[None][fix] Ensure that the W4A8 custom input scale remains aligned across all ranks #7614
Conversation
📝 WalkthroughWalkthroughAdds alignment logic in load_quant_scales for W4A8 custom: selects expert IDs differently, loads input scales for w3/w1 across selected experts, computes a cross-layer max scale, and uses it to set fc31_act_scale and fc31_alpha. Non-W4A8 paths retain existing per-expert scale logic. Changes
Sequence Diagram(s)sequenceDiagram
autonumber
participant Caller
participant QuantModule as Quant Module (MoE)
participant Storage as Scale Storage
Caller->>QuantModule: load_quant_scales()
alt W4A8 custom
QuantModule->>QuantModule: input_scale_expert_ids = range(num_experts)
loop For each expert in input_scale_expert_ids
QuantModule->>Storage: load w3.input_scale[expert]
QuantModule->>Storage: load w1.input_scale[expert]
end
QuantModule->>QuantModule: all_w3_w1_input_scales_max = max(loaded w3/w1 scales)
QuantModule->>QuantModule: set fc31_act_scale, fc31_alpha using max
else Non-W4A8
QuantModule->>QuantModule: input_scale_expert_ids = initial_local_expert_ids
loop For each expert in input_scale_expert_ids
QuantModule->>Storage: load w3.input_scale[expert]
QuantModule->>Storage: load w1.input_scale[expert]
end
QuantModule->>QuantModule: retain existing pre_quant per-expert logic
end
QuantModule-->>Caller: return
Estimated code review effort🎯 3 (Moderate) | ⏱️ ~20 minutes Possibly related PRs
Suggested labels
Suggested reviewers
✨ 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. 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 (2)
tensorrt_llm/_torch/modules/fused_moe/quantization.py (2)
1152-1161: Nit: simplify/maximize clarity when computing the global maxYou can avoid an intermediate pairwise tensor by taking the maxima separately and then a scalar max. Slightly clearer and avoids an elementwise max allocation.
Apply this diff:
- all_w3_w1_input_scales_max = torch.max( - torch.stack(all_w3_input_scales), - torch.stack(all_w1_input_scales)).max() + all_w3_w1_input_scales_max = max( + torch.stack(all_w3_input_scales).max(), + torch.stack(all_w1_input_scales).max(), + )
1258-1274: Consider aligning fc2 input_scale across ranks for W4A8 custom as wellYou aligned fc31 using all experts under W4A8 custom but fc2 still aggregates over
module.initial_local_expert_ids. If fc2 activation scaling participates in cross-rank collectives, this asymmetry can reintroduce rank skew.Proposed minimal change reusing
input_scale_expert_ids:- all_w2_input_scales = [ - load_weight_shard(weights[f"{expert_id}.w2.input_scale"], - device=self.device) - for expert_id in module.initial_local_expert_ids - ] + all_w2_input_scales = [ + load_weight_shard(weights[f"{expert_id}.w2.input_scale"], + device=self.device) + for expert_id in (input_scale_expert_ids + if w4a8_custom else module.initial_local_expert_ids) + ]Can you confirm whether fc2’s act scale needs to be rank-aligned for your kernels/checkpoints? If yes, this change keeps both fc31/fc2 consistent.
📜 Review details
Configuration used: Path: .coderabbit.yaml
Review profile: CHILL
Plan: Pro
📒 Files selected for processing (1)
tensorrt_llm/_torch/modules/fused_moe/quantization.py(1 hunks)
🧰 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/quantization.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/quantization.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/quantization.py
🧬 Code graph analysis (1)
tensorrt_llm/_torch/modules/fused_moe/quantization.py (1)
tensorrt_llm/_torch/modules/linear.py (1)
load_weight_shard(59-103)
⏰ 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 (1)
tensorrt_llm/_torch/modules/fused_moe/quantization.py (1)
1148-1151: W4A8 custom: good fix to align input_scale across ranksUsing all experts (
range(module.num_experts)) for W4A8 custom ensures rank-invariant fc31 input-scale derived values. This directly addresses the cross-rank drift. LGTM.
|
/bot run |
|
PR_Github #18049 [ run ] triggered by Bot |
|
PR_Github #18049 [ run ] completed with state |
yilin-void
force-pushed
the
w4a8_act_scale
branch
from
3092497 to
bcc6eaf
Compare
|
/bot run |
|
PR_Github #18113 [ run ] triggered by Bot |
|
PR_Github #18113 [ run ] completed with state |
Thanks! I validated with the DS R1 W4A8 checkpoints on H20x8, and the output of quickstart_advanced.py was reasonable. The average accuracy of lm-eval on gsm8k was 95, which met expectations. 
|
yilin-void
force-pushed
the
w4a8_act_scale
branch
from
bcc6eaf to
1f05e0a
Compare
|
/bot run |
|
PR_Github #18199 [ run ] triggered by Bot |
|
PR_Github #18199 [ run ] completed with state |
yilin-void
force-pushed
the
w4a8_act_scale
branch
from
1f05e0a to
d8a8dc8
Compare
|
/bot run |
|
PR_Github #18272 [ run ] triggered by Bot |
|
PR_Github #18272 [ run ] completed with state |
|
/bot run |
|
PR_Github #18282 [ run ] triggered by Bot |
|
PR_Github #18282 [ run ] completed with state |
|
/bot run --disable-fail-fast |
yilin-void
force-pushed
the
w4a8_act_scale
branch
from
d8a8dc8 to
27fb4e6
Compare
|
/bot run |
|
PR_Github #18538 [ run ] triggered by Bot |
|
PR_Github #18538 [ run ] completed with state |
… custom ckpts. Signed-off-by: Yilin Zhang <18275976+yilin-void@users.noreply.github.com>
yilin-void
force-pushed
the
w4a8_act_scale
branch
from
27fb4e6 to
7e6f92f
Compare
|
/bot run |
|
PR_Github #18581 [ run ] triggered by Bot |
|
PR_Github #18581 [ run ] completed with state |
…cross all ranks (NVIDIA#7614) Signed-off-by: Yilin Zhang <18275976+yilin-void@users.noreply.github.com>
…cross all ranks (NVIDIA#7614) Signed-off-by: Yilin Zhang <18275976+yilin-void@users.noreply.github.com>
4 participants
Summary by CodeRabbit
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
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.
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.