[Relax][Onnx][BatchNorm] Pass momentum and training_mode into BatchNorm Operator

[locnd182644]

Description

• Onnx model have training_mode atrr = False, but Relax model after convert have training = True
• Momentum values ​​in Relax module are not the same as onnx model

Steps to Reproduce

• Relax model:

class Module:
    def main(X: R.Tensor((2, 3, 4, 4), dtype="float32")) -> R.Tensor((2, 3, 4, 4), dtype="float32"):
        R.func_attr({"num_input": 1})
        with R.dataflow():
            lv: R.Tuple(R.Tensor((2, 3, 4, 4), dtype="float32"), R.Tensor((3,), dtype="float32"), R.Tensor((3,), dtype="float32")) = R.nn.batch_norm(X, metadata["relax.expr.Constant"][0], metadata["relax.expr.Constant"][1], metadata["relax.expr.Constant"][2], metadata["relax.expr.Constant"][3], axis=1, epsilon=9.9999997473787516e-06, center=True, scale=True, momentum=0.10000000000000001, training=True)
            lv1: R.Tensor((2, 3, 4, 4), dtype="float32") = lv[0]
            lv2: R.Tensor((3,), dtype="float32") = lv[1]
            lv3: R.Tensor((3,), dtype="float32") = lv[2]
            gv: R.Tensor((2, 3, 4, 4), dtype="float32") = lv1
            R.output(gv)
        return gv

Resolved

• Get Attributes and Pass momentum/training_mode with default value into BatchNorm Operator
• Fixed: [Bug][Relax][ONNX] BatchNormalization ignores training_mode attribute, always uses training=True #18703

[gemini-code-assist]

Summary of Changes

Hello @locnd182644, I'm Gemini Code Assist¹! I'm currently reviewing this pull request and will post my feedback shortly. In the meantime, here's a summary to help you and other reviewers quickly get up to speed!

This pull request resolves an issue where the BatchNorm operator in Relax models, after conversion from ONNX, did not accurately reflect the momentum and training_mode parameters of the original ONNX model. By explicitly extracting and passing these attributes, the conversion process now ensures greater fidelity between the ONNX and Relax representations, preventing discrepancies in model behavior and addressing the reported bug.

Highlights

• BatchNorm Parameter Propagation: The momentum and training_mode attributes are now correctly extracted from ONNX model attributes and passed to the relax.op.nn.batch_norm operator during conversion.
• Default Value Handling: Default values (0.9 for momentum and 0 for training_mode) are applied if these attributes are not explicitly present in the ONNX model, ensuring robust conversion.

🧠 New Feature in Public Preview: You can now enable Memory to help Gemini Code Assist learn from your team's feedback. This makes future code reviews more consistent and personalized to your project's style. Click here to enable Memory in your admin console.

Using Gemini Code Assist

The full guide for Gemini Code Assist can be found on our documentation page, here are some quick tips.

Invoking Gemini

You can request assistance from Gemini at any point by creating a comment using either /gemini <command> or @gemini-code-assist <command>. Below is a summary of the supported commands on the current page.

Feature	Command	Description
Code Review	/gemini review	Performs a code review for the current pull request in its current state.
Pull Request Summary	/gemini summary	Provides a summary of the current pull request in its current state.
Comment	@gemini-code-assist	Responds in comments when explicitly tagged, both in pull request comments and review comments.
Help	/gemini help	Displays a list of available commands.

Customization

To customize Gemini Code Assist for GitHub experience, repository maintainers can create a configuration file and/or provide a custom code review style guide (such as PEP-8 for Python) by creating and adding files to a .gemini/ folder in the base of the repository. Detailed instructions can be found here.

Limitations & Feedback

Gemini Code Assist may make mistakes. Please leave feedback on any instances where its feedback is incorrect or counter productive. You can react with 👍 and 👎 on @gemini-code-assist comments. If you're interested in giving your feedback about your experience with Gemini Code Assist for Github and other Google products, sign up here.

You can also get AI-powered code generation, chat, as well as code reviews directly in the IDE at no cost with the Gemini Code Assist IDE Extension.

Footnotes

1. Review the Privacy Notices, Generative AI Prohibited Use Policy, Terms of Service, and learn how to configure Gemini Code Assist in GitHub here. Gemini can make mistakes, so double check it and use code with caution. ↩

[gemini-code-assist]

Code Review

This pull request correctly addresses a bug in the ONNX frontend for the BatchNormalization operator. Previously, the momentum and training_mode attributes were not being passed from the ONNX model, causing the Relax operator to use incorrect default values. Your changes fix this by retrieving these attributes from the ONNX node and forwarding them to relax.op.nn.batch_norm. This aligns the behavior with the ONNX specification. I've added one minor suggestion to improve code clarity by explicitly casting the training_mode to a boolean.

[gemini-code-assist]

For clarity and type safety, it's better to explicitly cast training_mode to a boolean. The training parameter of relax.op.nn.batch_norm expects a boolean, and while Python's truthiness handles the integer 0 or 1 correctly, an explicit cast makes the intent clearer and guards against potential issues if the operator's signature becomes stricter in the future.

Suggested change

	training=training_mode,
	training=bool(training_mode),