Support initializers with onnxtext

[justinchuby]

Enhancements to Serialization and Deserialization:

• Added to_onnx_text method: Introduced a new method in src/onnx_ir/serde.py to convert an IR model to ONNX textual representation. The method includes an option to exclude initializers from the output.
• Extended from_onnx_text method: Enhanced the existing method to support adding initializers to the model during deserialization using a mapping of names to tensors. This addition improves flexibility and usability.

[codecov]

Codecov Report

Attention: Patch coverage is 33.33333% with 12 lines in your changes missing coverage. Please review.

Project coverage is 73.39%. Comparing base (60f8692) to head (d9dee73).
Report is 3 commits behind head on main.

Files with missing lines	Patch %	Lines
src/onnx_ir/serde.py	33.33%	11 Missing and 1 partial ⚠️

Additional details and impacted files

@@            Coverage Diff             @@
##             main      #60      +/-   ##
==========================================
- Coverage   73.59%   73.39%   -0.21%     
==========================================
  Files          37       37              
  Lines        4492     4507      +15     
  Branches      902      906       +4     
==========================================
+ Hits         3306     3308       +2     
- Misses        858      869      +11     
- Partials      328      330       +2     

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

[Copilot]

Pull Request Overview

Enhance ONNX IR serialization/deserialization to support textual format with optional initializer handling.

• Introduce from_onnx_text accepting an optional initializer mapping.
• Add to_onnx_text with a flag to exclude initializers.
• Import _convenience to build value mappings for initializer insertion.

Reviewed Changes

Copilot reviewed 2 out of 2 changed files in this pull request and generated 2 comments.

File	Description
src/onnx_ir/serde.py	New methods from_onnx_text and to_onnx_text with initializer support
docs/api/core.md	Registered to_onnx_text in the API reference

Comments suppressed due to low confidence (1)

src/onnx_ir/serde.py:196

• The Mapping type is not imported; this will raise a NameError. Either prefix with typing.Mapping or add from typing import Mapping at the top.

with_initializers: Mapping[str, _protocols.TensorProtocol] | None = None,

[titaiwangms]

Can't we use text model to improve ir.Model print out? Maybe that way we don't need two IRs?

[justinchuby]

Can't we use text model to improve ir.Model print out? Maybe that way we don't need two IRs?

The print outs are created for better readability and are more free formed (can include any information we want to put there). The text representation needs to support round tripping and is rigidly defined. I think there is a benefit in defining the repr for an ir model instead of directly using the text format. We also save the need to serialize when printing this way