[https://nvbugs/5680312][fix] Fixed merging sharding transforms between factory and heuristics #9452
Conversation
|
/bot run |
📝 WalkthroughWalkthroughRefactored the weight sharding transformation mechanism in sharding.py by replacing direct list appends with calls to a new Changes
Estimated code review effort🎯 3 (Moderate) | ⏱️ ~20 minutes
Pre-merge checks and finishing touches❌ Failed checks (2 warnings)
✅ Passed checks (3 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
Caution
Some comments are outside the diff and can’t be posted inline due to platform limitations.
⚠️ Outside diff range comments (6)
tensorrt_llm/_torch/auto_deploy/transform/library/sharding.py (6)
572-611: Inconsistent counting: counters increment even ifadd()fails.The counters
num_shardsandnum_row_col_shardsare incremented regardless of whetheradd()returnsTrueorFalse. If a node is already sharded (causingadd()to returnFalse), these counters will still increment, leading to misleading log messages about the number of transforms applied.Compare this to the correct pattern in
_process_simple_shard(lines 128-139), which usesint(sharding_config.add(...))to count only successful additions.Consider applying this diff to fix the counting:
if re.match(pattern_regex, module_name): - num_shards += 1 # we have a match. Get the config for this layer config = tp_plan[key] if config == "colwise": - sharding_config.add( + if sharding_config.add( WeightShardingInfo.from_node( lin_node, split_dim=SplitDimension.COLUMN, rank=rank, world_size=world_size, dist_op=None, min_local_shape=min_local_shape, ) - ) - num_row_col_shards += 1 + ): + num_shards += 1 + num_row_col_shards += 1 elif config == "rowwise": - sharding_config.add( + if sharding_config.add( WeightShardingInfo.from_node( lin_node, split_dim=SplitDimension.ROW, rank=rank, world_size=world_size, dist_op="all_reduce", min_local_shape=min_local_shape, ) - ) - num_row_col_shards += 1 + ): + num_shards += 1 + num_row_col_shards += 1 elif config == "mamba": - sharding_config.add( + if sharding_config.add( WeightShardingInfo.from_node( lin_node, split_dim=SplitDimension.COLUMN, rank=rank, world_size=world_size, dist_op=None, min_local_shape=min_local_shape, layer_type=LayerType.MAMBA, ) - ) - num_row_col_shards += 1 + ): + num_shards += 1 + num_row_col_shards += 1
618-647: Missing counter increment for local colwise case.The
local_colwisebranch (lines 621-630) doesn't incrementnum_row_col_shards, but thelocal_rowwisebranch (lines 632-642) does. This is inconsistent with the non-localcolwiseandrowwisecases (lines 586, 598) where both increment the counter.Apply this diff to add the missing increment:
if col_row_action == "colwise": - sharding_config.add( + if sharding_config.add( WeightShardingInfo( target_node=lin_node.name, split_dim=SplitDimension.COLUMN, rank=rank, world_size=world_size, dist_op=None, min_local_shape=min_local_shape, ) - ) + ): + num_row_col_shards += 1 elif col_row_action == "rowwise": - sharding_config.add( + if sharding_config.add( WeightShardingInfo( target_node=lin_node.name, split_dim=SplitDimension.ROW, rank=rank, world_size=world_size, dist_op="all_reduce", min_local_shape=min_local_shape, ) - ) - num_row_col_shards += 1 + ): + num_row_col_shards += 1
649-677: Good: Pattern matching break prevents duplicate processing.The
breakstatement on line 677 correctly prevents the same node from being processed by multiple patterns in the factory config. This is essential for avoiding conflicting transforms.However, the counting for the
gathercase (line 661) still has the same issue as noted earlier—it increments regardless of whetheradd()succeeds.Similar fix needed for lines 651-661:
elif "gather" in config: # Simple shard (row + all_gather) - sharding_config.add( + if sharding_config.add( WeightShardingInfo.from_node( lin_node, split_dim=SplitDimension.COLUMN, rank=rank, world_size=world_size, dist_op="all_gather", min_local_shape=1, ) - ) - num_simple_shards += 1 + ): + num_shards += 1 + num_simple_shards += 1
946-957: Inconsistent counting: increment without checkingadd()result.Similar to the factory config path,
num_row_col_shardsis incremented unconditionally even ifadd()returnsFalse.Apply this diff:
# shard single row node - sharding_config.add( + if sharding_config.add( WeightShardingInfo.from_node( nodes_to_row_shard[0], split_dim=SplitDimension.ROW, rank=rank, world_size=world_size, dist_op="all_reduce", min_local_shape=min_local_shape, ) - ) - - num_row_col_shards += 1 + ): + num_row_col_shards += 1
1023-1036: Inconsistent counting pattern.The counter
num_bmm_shardsis incremented without checking ifadd()succeeded.Apply this diff for consistency:
- sharding_config.add( + if sharding_config.add( BMMShardingInfo( target_node=node.name, rank=rank, world_size=world_size, start_idx=start_idx, end_idx=end_idx, ) - ) + ): + num_bmm_shards += 1 + ad_logger.debug( f"Sharding BMM for rank {rank}: batch_size={bmm_batch_size}, start_idx={start_idx}, end_idx={end_idx}" ) - - num_bmm_shards += 1
1067-1074: Inconsistent counting pattern.The counter
num_moe_patternsis incremented without verifying thatadd()succeeded.Apply this diff:
- sharding_config.add( + if sharding_config.add( EPShardingInfo.from_node( node, rank=rank, world_size=world_size, ) - ) - num_moe_patterns += 1 + ): + num_moe_patterns += 1
🧹 Nitpick comments (1)
tensorrt_llm/_torch/auto_deploy/transform/library/sharding.py (1)
248-427: Verify that parameter update transforms are applied successfully.While the primary weight sharding transforms use
add()to prevent duplicates (which is correct), the return values ofadd()forParameterUpdateInfotransforms (lines 304-319, 333-337, 408-412) are not checked. If these parameter updates fail to apply due to a duplicate, the function still returns1indicating success.Consider logging a warning when critical parameter updates fail to apply, particularly for the split nodes and conv1d parameters that are essential for correct SSM sharding.
📜 Review details
Configuration used: Path: .coderabbit.yaml
Review profile: CHILL
Plan: Pro
📒 Files selected for processing (1)
tensorrt_llm/_torch/auto_deploy/transform/library/sharding.py(10 hunks)
🧰 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 in Python, even if only one class or function from a module is used (e.g., usefrom package.subpackage import fooand thenfoo.SomeClass()instead offrom package.subpackage.foo import SomeClass)
Python filenames should use snake_case (e.g.,some_file.py)
Python class names should use PascalCase (e.g.,class SomeClass)
Python function and method names should use snake_case (e.g.,def my_awesome_function():)
Python local variable names 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
Python comments 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
Python attributes and variables can be documented inline with type and description (e.g.,self.x = 5followed by"""<type>: Description of 'x'""")
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 specific errors possible instead of catching all exceptions
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 to implement the logic
Files:
tensorrt_llm/_torch/auto_deploy/transform/library/sharding.py
**/*.{cpp,h,cu,py}
📄 CodeRabbit inference engine (CODING_GUIDELINES.md)
All TensorRT-LLM Open Source Software code files should contain an NVIDIA copyright header that includes the current year at the top
Files:
tensorrt_llm/_torch/auto_deploy/transform/library/sharding.py
🧬 Code graph analysis (1)
tensorrt_llm/_torch/auto_deploy/transform/library/sharding.py (1)
tensorrt_llm/_torch/auto_deploy/utils/sharding_utils.py (1)
add(1240-1259)
⏰ 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 (3)
tensorrt_llm/_torch/auto_deploy/transform/library/sharding.py (3)
117-140: LGTM! Correct use ofadd()return value.The function properly uses the return value of
add()to count only the transforms that were successfully added, preventing inflated counts when a node is already sharded.
430-501: LGTM with minor note.The function correctly applies sharding transforms using
add(). While the return values aren't checked here, the calling function manages the overall shard counting, so this is acceptable.
1-1080: Overall approach correctly addresses the reported issue.The replacement of direct list appends with
add()method calls prevents duplicate sharding transforms from being applied to the same node. This should resolve the "a and b must have same reduction dim" error reported in issue #9379, which was likely caused by conflicting transforms when both factory config and heuristics tried to shard the same nodes.The
add()method (defined insharding_utils.py) checks if a node already has a transform and only adds new transforms if the node hasn't been sharded before, returningTrueon success andFalseif rejected. This is the correct approach for merging transforms from multiple sources.
|
PR_Github #25743 [ run ] triggered by Bot. Commit: |
greg-kwasniewski1
changed the title
greg-kwasniewski1
changed the title
greg-kwasniewski1
changed the title
|
@greg-kwasniewski1 my PR just got merged (it got stalled due to many CI issues), I think it covers it |
|
PR_Github #25743 [ run ] completed with state |
|
/bot run |
greg-kwasniewski1
force-pushed
the
gk/bug_fix_9379
branch
from
942b94c to
b10137f
Compare
4 participants
The bug was caused by incorrect merging of factory and heuristics transforms. The same weight was sharded by both - without a prior check if the transformation was already applied.
Fixes #9379
Summary by CodeRabbit
✏️ 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.