[TRTLLM-6342][feat] Factory TP sharding of quantized models #8123
Conversation
Signed-off-by: greg-kwasniewski1 <[email protected]>
Signed-off-by: greg-kwasniewski1 <[email protected]>
📝 WalkthroughWalkthroughIntroduces a Starcoder2Config patch to set base_model_tp_plan for MLP projection. Extends node filtering to accept an iterable of predicates. Updates sharding detection to include fake-quantized linear ops and adds a fallback simple shard when encountering unsupported sharding actions, with adjusted log messages and early loop break preserved. Changes
Sequence Diagram(s)sequenceDiagram
autonumber
actor User
participant FactoryConfig
participant ShardingDetector as detect_sharding_from_factory_config
participant NodeFilter as filtered_nodes
participant LinNode as Linear/FQ Linear Node
participant Plan as TP Plan Builder
User->>FactoryConfig: Provide sharding factory config
FactoryConfig->>ShardingDetector: Start detection
ShardingDetector->>NodeFilter: Filter nodes by [is_linear_op, is_fake_quantized_linear_op]
loop For each filtered node
NodeFilter-->>ShardingDetector: LinNode
ShardingDetector->>ShardingDetector: Match config action for LinNode
alt Supported action
ShardingDetector->>Plan: Append specified shard transform
ShardingDetector-->>ShardingDetector: break
else Unsupported action
ShardingDetector->>ShardingDetector: Log "Unsupported sharding action {config}"
ShardingDetector->>Plan: Append simple shard (column split, all_gather, min_local_shape=1)
ShardingDetector-->>ShardingDetector: break
end
end
Estimated code review effort🎯 3 (Moderate) | ⏱️ ~25 minutes Pre-merge checks and finishing touches❌ Failed checks (1 warning)
✅ Passed checks (2 passed)
✨ 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: 4
🧹 Nitpick comments (2)
tensorrt_llm/_torch/auto_deploy/models/patches/starcoder.py (1)
4-4: Consider the implications of mutating a class-level attribute.Direct mutation of
Starcoder2Config.base_model_tp_planmodifies a global class attribute that may affect all instances. Ensure this side effect is intentional and that the patch is applied early enough in the initialization flow to avoid race conditions or unexpected behavior in concurrent usage.tensorrt_llm/_torch/auto_deploy/transform/library/sharding.py (1)
390-402: Fallback to simple shard improves resilience.Adding a default simple shard for unrecognized config values prevents silent failures and provides predictable behavior. However, ensure that a simple shard (column split + all_gather) is semantically appropriate for all potential future config values, or document the expected behavior.
Consider adding a validation step earlier in the pipeline or documenting the fallback behavior in the function docstring to clarify the assumptions.
📜 Review details
Configuration used: Path: .coderabbit.yaml
Review profile: CHILL
Plan: Pro
📒 Files selected for processing (3)
tensorrt_llm/_torch/auto_deploy/models/patches/starcoder.py(1 hunks)tensorrt_llm/_torch/auto_deploy/transform/library/sharding.py(3 hunks)tensorrt_llm/_torch/auto_deploy/utils/node_utils.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/auto_deploy/utils/node_utils.pytensorrt_llm/_torch/auto_deploy/models/patches/starcoder.pytensorrt_llm/_torch/auto_deploy/transform/library/sharding.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/auto_deploy/utils/node_utils.pytensorrt_llm/_torch/auto_deploy/models/patches/starcoder.pytensorrt_llm/_torch/auto_deploy/transform/library/sharding.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/auto_deploy/utils/node_utils.pytensorrt_llm/_torch/auto_deploy/models/patches/starcoder.pytensorrt_llm/_torch/auto_deploy/transform/library/sharding.py
🧬 Code graph analysis (2)
tensorrt_llm/_torch/auto_deploy/utils/node_utils.py (1)
tensorrt_llm/_torch/auto_deploy/transform/library/fused_moe.py (1)
target(359-360)
tensorrt_llm/_torch/auto_deploy/transform/library/sharding.py (2)
tensorrt_llm/_torch/auto_deploy/utils/node_utils.py (3)
filtered_nodes(205-252)is_linear_op(255-265)is_fake_quantized_linear_op(268-274)tensorrt_llm/_torch/auto_deploy/utils/sharding_utils.py (4)
TPShardingInfo(245-288)from_node(253-258)from_node(753-758)SplitDimension(198-206)
⏰ 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/auto_deploy/transform/library/sharding.py (1)
371-371: LGTM: Warning message for unsupported local sharding action.The warning message clearly indicates that the unsupported sharding action is being skipped, which is appropriate for the local-shard branch where no fallback is applied.
| @@ -0,0 +1,4 @@ | |||
| from transformers.models.starcoder2.configuration_starcoder2 import Starcoder2Config | |||
There was a problem hiding this comment.
Add the required NVIDIA Apache-2.0 copyright header.
Per coding guidelines, all Python source files must include the NVIDIA Apache-2.0 copyright header with the current year at the top of the file.
Add this header at the top of the file:
+# SPDX-FileCopyrightText: Copyright (c) 2025 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
+# SPDX-License-Identifier: Apache-2.0
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+# http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
from transformers.models.starcoder2.configuration_starcoder2 import Starcoder2ConfigAs per coding guidelines.
📝 Committable suggestion
‼️ IMPORTANT
Carefully review the code before committing. Ensure that it accurately replaces the highlighted code, contains no missing lines, and has no issues with indentation. Thoroughly test & benchmark the code to ensure it meets the requirements.
| from transformers.models.starcoder2.configuration_starcoder2 import Starcoder2Config | |
| # SPDX-FileCopyrightText: Copyright (c) 2025 NVIDIA CORPORATION & AFFILIATES. All rights reserved. | |
| # SPDX-License-Identifier: Apache-2.0 | |
| # | |
| # Licensed under the Apache License, Version 2.0 (the "License"); | |
| # you may not use this file except in compliance with the License. | |
| # You may obtain a copy of the License at | |
| # | |
| # http://www.apache.org/licenses/LICENSE-2.0 | |
| # | |
| # Unless required by applicable law or agreed to in writing, software | |
| # distributed on an "AS IS" BASIS, | |
| # WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. | |
| # See the License for the specific language governing permissions and | |
| # limitations under the License. | |
| from transformers.models.starcoder2.configuration_starcoder2 import Starcoder2Config |
🤖 Prompt for AI Agents
In tensorrt_llm/_torch/auto_deploy/models/patches/starcoder.py around line 1,
the file is missing the required NVIDIA Apache-2.0 copyright header; add the
standard NVIDIA Apache-2.0 header block (including the current year) at the very
top of the file before any imports, ensuring the exact header text and license
notice per project guidelines is used.
| # Remove this patch after TRT-LLM upgrades to the HF transformers version >= 4.57 | ||
| Starcoder2Config.base_model_tp_plan["layers.*.mlp.c_proj"] = "rowwise" |
There was a problem hiding this comment.
🛠️ Refactor suggestion | 🟠 Major
🧩 Analysis chain
Verify transformers version 4.57 availability and add tracking for patch removal.
The comment references transformers >= 4.57 as the removal threshold. Based on learnings, version 4.56.0 was released in August 2025, but 4.57 has not been mentioned yet. Please verify the actual version that includes the fix and confirm the timeline.
Additionally, consider adding a TODO or FIXME comment with an issue reference to ensure this temporary patch is tracked and removed appropriately.
Run the following script to check the latest transformers version and search for related PRs:
🏁 Script executed:
#!/bin/bash
# Check latest transformers version
curl -s https://pypi.org/pypi/transformers/json | jq -r '.info.version'
# Search for Starcoder2 TP plan fixes in transformers repo
gh api graphql -f query='
{
search(query: "repo:huggingface/transformers starcoder2 base_model_tp_plan", type: ISSUE, first: 5) {
nodes {
... on Issue {
number
title
state
createdAt
}
... on PullRequest {
number
title
state
createdAt
merged
}
}
}
}'Length of output: 325
Update patch removal comment for correct version and add tracking
Replace the existing line with:
# TODO: Remove this patch after upgrading to HuggingFace transformers ≥4.56.2 (fix included in HF PR #40814)🤖 Prompt for AI Agents
tensorrt_llm/_torch/auto_deploy/models/patches/starcoder.py around lines 3-4:
the current removal comment is outdated; replace it with a TODO that specifies
removing the patch after upgrading to HuggingFace transformers ≥4.56.2 and
references the HF PR #40814 so the change is tracked (i.e., update the comment
text to mention the exact version and PR number and leave the patch assignment
line unchanged).
| num_row_col_shards = 0 | ||
|
|
||
| for lin_node in filtered_nodes(gm.graph.nodes, is_linear_op): | ||
| for lin_node in filtered_nodes(gm.graph.nodes, [is_linear_op, is_fake_quantized_linear_op]): |
There was a problem hiding this comment.
Duplicate nodes will be processed due to filtered_nodes bug.
The call to filtered_nodes with a list of predicates [is_linear_op, is_fake_quantized_linear_op] will yield duplicate nodes because the implementation in node_utils.py (lines 242-246) is missing a break statement after yielding. This will cause nodes that match both predicates to be processed multiple times, creating duplicate shard transforms.
This issue will be resolved once the bug in node_utils.py (lines 242-246) is fixed by adding a break statement after yield node.
🤖 Prompt for AI Agents
In tensorrt_llm/_torch/auto_deploy/transform/library/sharding.py around line
295, duplicate nodes are being processed because filtered_nodes was called with
two predicates; fix the root cause in
tensorrt_llm/_torch/auto_deploy/transform/node_utils.py (around lines 242-246)
by adding a break immediately after the "yield node" inside the loop so a node
that matches the first predicate is not yielded again for subsequent predicates,
then run tests to confirm no duplicate shard transforms are produced.
| elif isinstance(target, Iterable) and all(isinstance(t, Callable) for t in target): | ||
| for node in nodes: | ||
| for t in target: | ||
| if t(node): | ||
| yield node |
There was a problem hiding this comment.
Add break after yielding to prevent duplicate nodes.
The inner loop at line 244-246 will yield the same node multiple times if more than one predicate matches. This creates duplicates in the iteration results.
Apply this diff to add a break statement:
elif isinstance(target, Iterable) and all(isinstance(t, Callable) for t in target):
for node in nodes:
for t in target:
if t(node):
yield node
+ break📝 Committable suggestion
‼️ IMPORTANT
Carefully review the code before committing. Ensure that it accurately replaces the highlighted code, contains no missing lines, and has no issues with indentation. Thoroughly test & benchmark the code to ensure it meets the requirements.
| elif isinstance(target, Iterable) and all(isinstance(t, Callable) for t in target): | |
| for node in nodes: | |
| for t in target: | |
| if t(node): | |
| yield node | |
| elif isinstance(target, Iterable) and all(isinstance(t, Callable) for t in target): | |
| for node in nodes: | |
| for t in target: | |
| if t(node): | |
| yield node | |
| break |
🤖 Prompt for AI Agents
In tensorrt_llm/_torch/auto_deploy/utils/node_utils.py around lines 242 to 246,
the inner loop yields the same node multiple times when multiple predicates
match; after yielding a node inside the inner for-loop, add a break to stop
checking further predicates for that node so each node is produced at most once.
Description
Previously,
detect_sharding_from_factory_configapplied sharding strategy to all linear nodes filtered byis_linear_op:is_linear_opdoes not capture quantized linear nodes such astorch.ops.auto_deploy.torch_fake_quant_fp8_linear.This PR extents
detect_sharding_from_factory_configto:An update to
filtered_nodeslogic was needed to achieve it.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.