[TRTLLM-6342][feat] Support custom sharding config source #8153

greg-kwasniewski1 · 2025-10-06T00:29:32Z

This PR enables user to specify a custom sharding config file, for example:

$ cat tp_sharding.yaml

head_dim : 8
tp_plan :
  gate: colwise
  up: colwise
  down: rowwise
  "*": gather

The config is expected to be a parsable .yaml or .json dictionary, with at least one required key: tp_plan, which should be a dictionary [(partial)_module_name] : [sharding_action]

The relevant tests in test_tp_sharding.py have been updated.

Summary by CodeRabbit

New Features
- Added flexible sharding configuration with selectable sources (heuristic, custom, factory).
- Enabled loading custom sharding configs from JSON/YAML.
- Expanded sharding dimensions support (including TP, EP, BMM).
- Improved node filtering by allowing multiple predicate targets.
- Enhanced transform aggregation for clearer sharding summaries.
Refactor
- Simplified configuration by replacing legacy flags with enums and consolidated fields.
- Default behavior now favors heuristic-based sharding with partial config support.
Tests
- Updated unit tests to use the new sharding configuration.
- Added scenarios validating custom YAML-based TP sharding.

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 the stage-list parameter 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.md
and the scripts/test_to_stage_mapping.py helper.

kill

kill

Kill all running builds associated with pull request.

skip

skip --comment COMMENT

Skip 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-pipeline

Reuse 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.

Signed-off-by: greg-kwasniewski1 <[email protected]>

greg-kwasniewski1 · 2025-10-06T00:29:48Z

\bot run

coderabbitai · 2025-10-06T00:34:59Z

📝 Walkthrough

Walkthrough

Refactors sharding configuration to enum-based sources and dimensions, introduces multi-source sharding apply flow (factory/custom/heuristic), adds custom config loading and TP append utilities, updates TransformInfo with += merge, removes legacy fields from AutoDeployConfig, adjusts default/config YAML, extends node filtering, and updates unit tests accordingly.

Changes

Cohort / File(s)	Summary
Config defaults `tensorrt_llm/_torch/auto_deploy/config/default.yaml`	Switch detect_sharding to `sharding_source: ['heuristic']`, enable `support_partial_config: true`, retain `sharding_dims` and `requires_shape_prop`, remove `use_sharding_from_factory`.
Public config fields removal `tensorrt_llm/_torch/auto_deploy/llm_args.py`	Remove `simple_shard_only`, `use_sharding_from_factory`, and `sharding_dims` from `AutoDeployConfig`.
Transform info in-place merge `tensorrt_llm/_torch/auto_deploy/transform/interface.py`	Add `TransformInfo.__iadd__` returning a new merged instance combining flags via AND and summing `num_matches`.
Sharding library refactor `tensorrt_llm/_torch/auto_deploy/transform/library/sharding.py`	Introduce `ShardingSource`/`ShardingDim`, rework `ShardingTransformConfig` fields, overhaul `_apply` to process FACTORY/CUSTOM/HEURISTIC sources, add custom-config path, refactor factory detection and counters, update logging and shard accounting.
Sharding utilities upgrade `tensorrt_llm/_torch/auto_deploy/utils/sharding_utils.py`	Add enums `ShardingSource`, `ShardingDim`; extend `ShardingConfig` with `custom_sharding_config`, `sharding_source`, `ep_transforms`; add `read_custom_sharding_config` and `append_TP`.
Node utils enhancement `tensorrt_llm/_torch/auto_deploy/utils/node_utils.py`	Allow iterable of callables as `target` to match nodes if any predicate returns true.
Test helpers `tests/unittest/_torch/auto_deploy/_utils_test/_graph_test_helpers.py`	Improve assertion message to include actual vs. expected sets.
Tests: BMM sharding `tests/unittest/_torch/auto_deploy/unit/multigpu/transformations/library/test_bmm_sharding.py`	Replace `use_sharding_from_factory` with `sharding_source: ["heuristic"]`; add `support_partial_config: False`.
Tests: EP sharding `tests/unittest/_torch/auto_deploy/unit/multigpu/transformations/library/test_ep_sharding.py`	Switch to `sharding_source: ["heuristic"]`, specify `sharding_dims: ["ep"]`, set `support_partial_config: False`.
Tests: TP sharding + custom config `tests/unittest/_torch/auto_deploy/unit/multigpu/transformations/library/test_tp_sharding.py`	Introduce YAML-based predefined TP config; use `sharding_source` (["custom"] or ["heuristic"]), `custom_sharding_config`, `sharding_dims: ["tp"]`; propagate predefined config into optimizer; remove main block; add `yaml` import.

Sequence Diagram(s)

sequenceDiagram
  autonumber
  participant User
  participant InferenceOptimizer as InferenceOptimizer
  participant ShardingExec as ShardingTransformExecutor
  participant ShardingCfg as ShardingConfig
  participant Heuristics as HeuristicDetector
  participant Factory as FactoryConfig
  participant FS as Filesystem

  User->>InferenceOptimizer: configure detect_sharding (sharding_source, sharding_dims, custom_config)
  InferenceOptimizer->>ShardingExec: _apply(shared_config)
  ShardingExec->>ShardingCfg: read sharding_source, sharding_dims
  alt CUSTOM in sharding_source
    ShardingExec->>FS: read_custom_sharding_config(path)
    FS-->>ShardingExec: config loaded / error
    opt config loaded
      ShardingExec->>Factory: map custom to predefined_config
      Factory-->>ShardingExec: factory sharding plan
      ShardingExec->>ShardingCfg: append_TP()/record EP/BMM
    end
  end
  alt FACTORY in sharding_source
    ShardingExec->>Factory: load factory config if available
    Factory-->>ShardingExec: plan or none
    opt plan
      ShardingExec->>ShardingCfg: append_TP()/record EP/BMM
    end
  end
  alt HEURISTIC in sharding_source
    ShardingExec->>Heuristics: detect TP/EP/BMM per sharding_dims
    Heuristics-->>ShardingExec: detected shards and counts
    ShardingExec->>ShardingCfg: append_TP()/record EP/BMM
  end
  ShardingExec-->>InferenceOptimizer: TransformInfo (merged via +=)
  InferenceOptimizer-->>User: result

Estimated code review effort

🎯 4 (Complex) | ⏱️ ~60 minutes

Pre-merge checks and finishing touches

❌ Failed checks (2 warnings)

Check name	Status	Explanation	Resolution
Docstring Coverage	⚠️ Warning	Docstring coverage is 36.84% which is insufficient. The required threshold is 80.00%.	You can run `@coderabbitai generate docstrings` to improve docstring coverage.
Description Check	⚠️ Warning	The PR description includes a brief free-form summary and the raw template text but lacks the required structured title and fails to populate the Description and Test Coverage sections, leaving them as empty placeholders.	Please add a title in the prescribed format at the top of the description, fill in the Description section with a concise explanation of the change and its motivation, and complete the Test Coverage section by listing the relevant tests that verify the new custom sharding configuration feature.

✅ Passed checks (1 passed)

Check name	Status	Explanation
Title Check	✅ Passed	The title succinctly and accurately describes the addition of support for a custom sharding configuration source by ticket and feature, matching the primary change of the pull request.

✨ Finishing touches

📝 Generate docstrings

🧪 Generate unit tests (beta)

Create PR with unit tests
Post copyable unit tests in a comment

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.

❤️ Share

_{Comment @coderabbitai help to get the list of available commands and usage tips.}

greg-kwasniewski1 · 2025-10-06T08:49:05Z

/bot run

tensorrt-cicd · 2025-10-06T08:54:38Z

PR_Github #20667 [ run ] triggered by Bot

tensorrt-cicd · 2025-10-06T09:29:38Z

PR_Github #20667 [ run ] completed with state FAILURE
/LLM/main/L0_MergeRequest_PR pipeline #15610 completed with status: 'FAILURE'

greg-kwasniewski1 · 2025-10-08T08:31:53Z

/bot run

greg-kwasniewski1 · 2025-10-08T08:37:03Z

\bot run

tensorrt-cicd · 2025-10-08T08:37:47Z

PR_Github #20779 [ run ] triggered by Bot

tensorrt-cicd · 2025-10-08T10:06:31Z

PR_Github #20779 [ run ] completed with state SUCCESS
/LLM/main/L0_MergeRequest_PR pipeline #15707 completed with status: 'FAILURE'

Signed-off-by: greg-kwasniewski1 <[email protected]>

greg-kwasniewski1 · 2025-10-08T12:29:47Z

/bot run

tensorrt-cicd · 2025-10-08T12:35:44Z

PR_Github #20797 [ run ] triggered by Bot

tensorrt-cicd · 2025-10-08T17:28:28Z

PR_Github #20797 [ run ] completed with state SUCCESS
/LLM/main/L0_MergeRequest_PR pipeline #15722 completed with status: 'FAILURE'

greg-kwasniewski1 · 2025-10-12T14:44:18Z

/bot run

tensorrt-cicd · 2025-10-12T14:49:47Z

PR_Github #21098 [ run ] triggered by Bot

Signed-off-by: greg-kwasniewski1 <[email protected]>

tensorrt-cicd · 2025-10-12T20:19:27Z

PR_Github #21098 [ run ] completed with state SUCCESS
/LLM/main/L0_MergeRequest_PR pipeline #15942 completed with status: 'FAILURE'

greg-kwasniewski1 · 2025-10-13T10:10:15Z

/bot run

tensorrt-cicd · 2025-10-13T10:15:33Z

PR_Github #21199 [ run ] triggered by Bot

tensorrt-cicd · 2025-10-13T10:16:14Z

PR_Github #21199 [ run ] completed with state FAILURE

greg-kwasniewski1 · 2025-10-13T10:17:31Z

/bot run

tensorrt-cicd · 2025-10-13T10:22:36Z

PR_Github #21200 [ run ] triggered by Bot

Signed-off-by: greg-kwasniewski1 <[email protected]>

tensorrt-cicd · 2025-10-13T14:43:19Z

PR_Github #21200 [ run ] completed with state SUCCESS
/LLM/main/L0_MergeRequest_PR pipeline #16004 completed with status: 'FAILURE'

greg-kwasniewski1 · 2025-10-13T14:45:41Z

/bot run

tensorrt-cicd · 2025-10-13T14:51:27Z

PR_Github #21231 [ run ] triggered by Bot

tensorrt-cicd · 2025-10-13T16:29:48Z

PR_Github #21231 [ run ] completed with state SUCCESS
/LLM/main/L0_MergeRequest_PR pipeline #16024 completed with status: 'FAILURE'

greg-kwasniewski1 · 2025-10-14T13:30:06Z

/bot run

greg-kwasniewski1 · 2025-10-14T16:50:55Z

/bot run

greg-kwasniewski1 · 2025-10-14T16:54:20Z

/bot reuse-pipeline

tensorrt-cicd · 2025-10-14T16:56:17Z

PR_Github #21379 [ run ] triggered by Bot

tensorrt-cicd · 2025-10-14T16:59:47Z

PR_Github #21380 [ reuse-pipeline ] triggered by Bot

tensorrt-cicd · 2025-10-14T16:59:48Z

PR_Github #21379 [ run ] completed with state ABORTED

greg-kwasniewski1 · 2025-10-14T17:01:45Z

/bot run

tensorrt-cicd · 2025-10-14T17:04:36Z

PR_Github #21380 [ reuse-pipeline ] completed with state SUCCESS
Can't reuse PR_Github #21379 with status: ABORTED

tensorrt-cicd · 2025-10-14T17:07:39Z

PR_Github #21381 [ run ] triggered by Bot

greg-kwasniewski1 · 2025-10-14T17:07:40Z

@lucaslie can you please take a look at this PR?

tensorrt-cicd · 2025-10-15T03:21:17Z

PR_Github #21381 [ run ] completed with state SUCCESS
/LLM/main/L0_MergeRequest_PR pipeline #16146 completed with status: 'FAILURE'

galagam · 2025-10-15T07:10:54Z

tests/unittest/_torch/auto_deploy/unit/multigpu/transformations/library/test_tp_sharding.py

+            with open("tp_sharding.yaml", "w") as f:
+                yaml.dump(predefined_config, f, sort_keys=False)


Perhaps use python's tempfile, to avoid contaminating the current working dir.

The problem with tempfile is that anyway this file has to be visible from a different thread directly form a disk, so i cannot use a context like:

with tempfile.NamedTemporaryFile(mode='w+t', delete=True) as tmpfile: # Write to the file yaml.dump(predefined_config, tmpfile, sort_keys=False) ...

tempfile adds some unique id either to temporary file or temporary diectory, but I need a fixed absolute path in custom_sharding_config parameter to read it from.

Unless, you know a good workaround to it?

See my other comment here: https://github.com/NVIDIA/TensorRT-LLM/pull/8153/files#r2437473448

If you do you should be able to just provide the custom config as dictionary without needing to create/read a tmp file

Signed-off-by: greg-kwasniewski1 <[email protected]>

Signed-off-by: Grzegorz Kwasniewski <[email protected]>

greg-kwasniewski1 · 2025-10-15T11:45:41Z

/bot run

tensorrt-cicd · 2025-10-15T11:51:04Z

PR_Github #21478 [ run ] triggered by Bot

tensorrt-cicd · 2025-10-15T17:19:04Z

PR_Github #21478 [ run ] completed with state SUCCESS
/LLM/main/L0_MergeRequest_PR pipeline #16215 completed with status: 'SUCCESS'
Pipeline passed with automatic retried tests. Check the rerun report for details.

lucaslie · 2025-10-16T20:46:02Z

tensorrt_llm/_torch/auto_deploy/transform/interface.py

+    def __iadd__(self, other: "TransformInfo") -> "TransformInfo":
+        # since TransformInfo is frozen, instead, we return a new TransformInfo object
+        return TransformInfo(
+            skipped=self.skipped & other.skipped,
+            num_matches=self.num_matches + other.num_matches,
+            is_clean=self.is_clean & other.is_clean,
+            has_valid_shapes=self.has_valid_shapes & other.has_valid_shapes,
+        )


Please use the existing __add__ operator instead. __iadd__ is by convention an in-place operator, i.e., it means that

config1 = TransformInfo() config2 = TransformInfo() config3 = config1 config3 += config2 assert config3 is config1 # is operator checks for same object!

However, this assertion would fail since you actually create a new object

lucaslie · 2025-10-16T20:47:08Z

tensorrt_llm/_torch/auto_deploy/config/default.yaml

+    # sharding_source: ['factory', 'custom', 'heuristic']
+    sharding_source: ['heuristic']
+    support_partial_config: true
+    # custom_sharding_config: 'tp_sharding.yaml'


looks like this is just a leftover from testing and should be reverted?

lucaslie · 2025-10-16T20:52:58Z

tensorrt_llm/_torch/auto_deploy/utils/sharding_utils.py

            self.validate_config()
        return self

+    def read_custom_sharding_config(self, config_path: str) -> bool:


I would like to avoid adding a separate yaml object just for sharding. We already have a general-purpose config reader otherwise it gets too complicated. There is no need to add a separate yaml reader

see this comment as well for more details: https://github.com/NVIDIA/TensorRT-LLM/pull/8153/files#r2437473448

lucaslie · 2025-10-16T21:02:31Z

tensorrt_llm/_torch/auto_deploy/transform/library/sharding.py

-            factory_info = detect_sharding_from_factory_config(gm, sharding_config)
-            return gm, factory_info
+        info = TransformInfo(skipped=False, num_matches=0, is_clean=True, has_valid_shapes=True)
+        for source in shared_config.sharding_config.sharding_source:


Looks like the order of the sharding sources and hence the order of precedence is defined by the user. That sounds dangerous. I would suggest we preordain the order of precedence. Namely, from highest to lowest:

manual config

tp/factory plan

heuristic

maybe we can use the fact the enum is ordered and just follow the order of the enum to define the order of precedence

lucaslie · 2025-10-16T21:09:11Z

tensorrt_llm/_torch/auto_deploy/transform/library/sharding.py

+    sharding_source: List[ShardingSource] = Field(
+        default_factory=lambda: [ShardingSource.HEURISTIC]
+    )
+    custom_sharding_config: str = Field(default="")


Suggested change

custom_sharding_config: str = Field(default="")

custom_sharding_config: Dict[str, Any] = Field(default_factory={})

Just make this a plain dictionary and let the config reader be done by the high-level LLM args object that already has a built-in config reader.

Some more details:

Like I mentioned in my other comment, I would like to avoid building an entirely new yaml config reader. Instead, we can rely on the config reader we have for the top-level config.

So for example, right now one can provide a config.yaml for the build_and_run_ad.py. You can just reuse the existing mechanism and since the field is defined as dictionary it will read in correctly.

This could be a potential config.yaml with a custom sharding config:

args: transforms: detect_sharding: custom_sharding_config: head_dim : 8 tp_plan : gate: colwise up: colwise down: rowwise "*": gather

lucaslie · 2025-10-16T21:09:31Z

tests/unittest/_torch/auto_deploy/unit/multigpu/transformations/library/test_tp_sharding.py

+            with open("tp_sharding.yaml", "w") as f:
+                yaml.dump(predefined_config, f, sort_keys=False)


See my other comment here: https://github.com/NVIDIA/TensorRT-LLM/pull/8153/files#r2437473448

If you do you should be able to just provide the custom config as dictionary without needing to create/read a tmp file

lucaslie · 2025-10-16T21:10:07Z

tensorrt_llm/_torch/auto_deploy/utils/sharding_utils.py

+        not sharded before. Do not overwrite existing transforms.
+        """
+        for existing_transform in self.tp_transforms:
+            if existing_transform.target_node == tp_transform.target_node:


is this sufficient to avoid conflicting/duplicate configurations?

lucaslie · 2025-10-16T21:14:19Z

tensorrt_llm/_torch/auto_deploy/transform/library/sharding.py

+    custom_sharding_config: str = Field(default="")
+    # Which sharding dimensions to run: any subset of {"tp", "ep", "bmm"}
+    sharding_dims: List[ShardingDim] = Field(
+        default_factory=lambda: [ShardingDim.TP, ShardingDim.EP, ShardingDim.BMM]


this is just for the heuristic, right?

So how about we just combine the sharding_source and sharding_dims configs here. ShardingSource can be as follows:

class ShardingSource(Enum): """Enum for sharding source.""" CUSTOM = "custom" FACTORY = "factory" TP_HEURISTIC = "tp_heuristic" EP_HEURISTIC = "ep_heuristic" BMM_HEURISTIC = "bmm_heuristic"

lucaslie · 2025-10-16T21:20:16Z

tensorrt_llm/_torch/auto_deploy/transform/library/sharding.py

+                sharding_config.predefined_config = sharding_config.custom_sharding_config
+                info += detect_sharding_from_factory_config(gm, sharding_config)


could you please provide a description somewhere of how to specify the custom sharding config? I couldn't find any references to the correct format

greg-kwasniewski1 added 5 commits October 1, 2025 14:15

Patched incorrect starcoder tp config

3433f1a

Signed-off-by: greg-kwasniewski1 <[email protected]>

Factory sharding supports quantized linear nodes

c105172

Signed-off-by: greg-kwasniewski1 <[email protected]>

sharding from user-provided config

8ae8869

Signed-off-by: greg-kwasniewski1 <[email protected]>

fixing sharding tests

1af20e0

Signed-off-by: greg-kwasniewski1 <[email protected]>

Fixed tp sharding test

3d1eb43

Signed-off-by: greg-kwasniewski1 <[email protected]>

greg-kwasniewski1 requested a review from a team as a code owner October 6, 2025 00:29

greg-kwasniewski1 requested a review from galagam October 6, 2025 00:29

greg-kwasniewski1 added 2 commits October 8, 2025 05:17

fixed llm args

0972ed6

Signed-off-by: greg-kwasniewski1 <[email protected]>

fixed llm args

03f4190

Signed-off-by: greg-kwasniewski1 <[email protected]>

greg-kwasniewski1 force-pushed the gk/custom_sharding_source branch from 08e62e5 to 03f4190 Compare October 8, 2025 12:28

fixed filtered_nodes

0cf4b2a

Signed-off-by: greg-kwasniewski1 <[email protected]>

greg-kwasniewski1 force-pushed the gk/custom_sharding_source branch from bfa0408 to 0cf4b2a Compare October 12, 2025 16:24

Fixed test_tp_sharding

c46d742

Signed-off-by: greg-kwasniewski1 <[email protected]>

greg-kwasniewski1 force-pushed the gk/custom_sharding_source branch from a17045e to c46d742 Compare October 13, 2025 12:58

greg-kwasniewski1 self-assigned this Oct 14, 2025

greg-kwasniewski1 requested a review from lucaslie October 14, 2025 17:07

galagam approved these changes Oct 15, 2025

View reviewed changes

merged TransformInfo changes

8765bbe

Signed-off-by: greg-kwasniewski1 <[email protected]>

greg-kwasniewski1 force-pushed the gk/custom_sharding_source branch from b472b80 to 8765bbe Compare October 15, 2025 11:41

Merge branch 'main' into gk/custom_sharding_source

99bca89

Signed-off-by: Grzegorz Kwasniewski <[email protected]>

lucaslie reviewed Oct 16, 2025

View reviewed changes

lucaslie mentioned this pull request Oct 21, 2025

[TRTLLM-8201][feat] TP sharding of Mamba layers #8548

Draft

1 task

		with open("tp_sharding.yaml", "w") as f:
		yaml.dump(predefined_config, f, sort_keys=False)

	custom_sharding_config: str = Field(default="")
	custom_sharding_config: Dict[str, Any] = Field(default_factory={})

		sharding_config.predefined_config = sharding_config.custom_sharding_config
		info += detect_sharding_from_factory_config(gm, sharding_config)

[TRTLLM-6342][feat] Support custom sharding config source #8153

Are you sure you want to change the base?

[TRTLLM-6342][feat] Support custom sharding config source #8153

Uh oh!

Conversation

greg-kwasniewski1 commented Oct 6, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Summary by CodeRabbit

Description

Test Coverage

PR Checklist

GitHub Bot Help

kill

skip

reuse-pipeline

Uh oh!

greg-kwasniewski1 commented Oct 6, 2025

Uh oh!

coderabbitai bot commented Oct 6, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Walkthrough

Changes

Sequence Diagram(s)

Estimated code review effort

Pre-merge checks and finishing touches

Uh oh!

greg-kwasniewski1 commented Oct 6, 2025

Uh oh!

tensorrt-cicd commented Oct 6, 2025

Uh oh!

tensorrt-cicd commented Oct 6, 2025

Uh oh!

greg-kwasniewski1 commented Oct 8, 2025

Uh oh!

greg-kwasniewski1 commented Oct 8, 2025

Uh oh!

tensorrt-cicd commented Oct 8, 2025

Uh oh!

tensorrt-cicd commented Oct 8, 2025

Uh oh!

greg-kwasniewski1 commented Oct 8, 2025

Uh oh!

tensorrt-cicd commented Oct 8, 2025

Uh oh!

tensorrt-cicd commented Oct 8, 2025

Uh oh!

greg-kwasniewski1 commented Oct 12, 2025

Uh oh!

tensorrt-cicd commented Oct 12, 2025

Uh oh!

tensorrt-cicd commented Oct 12, 2025

Uh oh!

greg-kwasniewski1 commented Oct 13, 2025

Uh oh!

tensorrt-cicd commented Oct 13, 2025

Uh oh!

tensorrt-cicd commented Oct 13, 2025

Uh oh!

greg-kwasniewski1 commented Oct 13, 2025

Uh oh!

tensorrt-cicd commented Oct 13, 2025

Uh oh!

tensorrt-cicd commented Oct 13, 2025

Uh oh!

greg-kwasniewski1 commented Oct 13, 2025

Uh oh!

tensorrt-cicd commented Oct 13, 2025

Uh oh!

tensorrt-cicd commented Oct 13, 2025

Uh oh!

greg-kwasniewski1 commented Oct 14, 2025

Uh oh!

greg-kwasniewski1 commented Oct 14, 2025

Uh oh!

greg-kwasniewski1 commented Oct 14, 2025

Uh oh!

tensorrt-cicd commented Oct 14, 2025

Uh oh!

tensorrt-cicd commented Oct 14, 2025

Uh oh!

greg-kwasniewski1 commented Oct 6, 2025 •

edited

Loading

coderabbitai bot commented Oct 6, 2025 •

edited

Loading