fix(autocast): add missing guard for use_standalone_type_inference

[agkphysics]

What does this PR do?

Type of change: Bug fix

Adds a missing check for use_standalone_type_inference in AutoCast which can cause shape issues because some dims are replaces with unk.

Testing

I have tested this with a model that previously failed after AutoCast conversion with the following error, but succeeds with this patch:

onnxruntime.capi.onnxruntime_pybind11_state.RuntimeException: [ONNXRuntimeError] : 6 : RUNTIME_EXCEPTION : Non-zero status code returned while running LayerNormalization node. Name:'node_layer_norm' Status Message: /onnxruntime_src/onnxruntime/core/framework/op_kernel.cc:83 virtual OrtValue* onnxruntime::OpKernelContext::OutputMLValue(int, const onnxruntime::TensorShape&) status.IsOK() was false. Shape mismatch attempting to re-use buffer. {1,1,768} != {1,261,768}. Validate usage of dim_value (values should be > 0) and dim_param (all values with the same string should equate to the same size) in shapes in the model.

Before your PR is "Ready for review"

• Is this change backward compatible?: N/A
• If you copied code from any other sources or added a new PIP dependency, did you follow guidance in CONTRIBUTING.md: N/A
• Did you write any new necessary tests?: N/A
• Did you update Changelog?: N/A
• Did you get Claude approval on this PR?: ❌

Summary by CodeRabbit

• Bug Fixes
  • Improved type inference accuracy when standalone type inference is enabled, ensuring proper handling of tensor dimensions in ONNX models.

[copy-pr-bot]

This pull request requires additional validation before any workflows can run on NVIDIA's runners.

Pull request vetters can view their responsibilities here.

Contributors can view more details about this message here.

[coderabbitai]

📝 Walkthrough

Walkthrough

In PrecisionConverter._clear_types_and_shapes_recursive, the dim_param reset to "unk" for value_info tensors in the main graph is now conditional on not self.use_standalone_type_inference. When standalone type inference is active, only elem_type clearing occurs; dimension parameter rewriting is skipped.

Changes

Conditional dim_param rewrite in PrecisionConverter

Layer / File(s)	Summary
Guard dim_param reset on standalone type inference flag modelopt/onnx/autocast/precisionconverter.py	Wraps the dim_param = "unk" assignment inside _clear_types_and_shapes_recursive with not self.use_standalone_type_inference, so that block is skipped when standalone type inference is enabled.

Estimated code review effort

🎯 1 (Trivial) | ⏱️ ~3 minutes

---

Caution

Pre-merge checks failed

Please resolve all errors before merging. Addressing warnings is optional.

• Ignore

❌ Failed checks (1 error)

Check name	Status	Explanation	Resolution
Security Anti-Patterns	❌ Error	PR adds 14 # nosec comments across 8 new modelopt files, explicitly violating SECURITY.md rule #5 which prohibits nosec comments to bypass security checks.	Remove all # nosec comments from added code. Per SECURITY.md, if security patterns are genuinely necessary, add inline justification comments instead and request explicit review from @NVIDIA/modelopt-setup-codeowners with justification...

✅ Passed checks (5 passed)

Check name	Status	Explanation
Docstring Coverage	✅ Passed	No functions found in the changed files to evaluate docstring coverage. Skipping docstring coverage check.
Linked Issues check	✅ Passed	Check skipped because no linked issues were found for this pull request.
Out of Scope Changes check	✅ Passed	Check skipped because no linked issues were found for this pull request.
Description Check	✅ Passed	Check skipped - CodeRabbit’s high-level summary is enabled.
Title check	✅ Passed	The title clearly and concisely describes the main change: adding a missing guard condition for use_standalone_type_inference in the autocast functionality.

✨ Finishing Touches

🧪 Generate unit tests (beta)

• Create PR with unit tests

---

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