Fix bugs with handling subbyte types

[justinchuby]

Support 2bit types.

[codecov]

Codecov Report

❌ Patch coverage is 88.63636% with 10 lines in your changes missing coverage. Please review.
✅ Project coverage is 70.85%. Comparing base (5488053) to head (c67d598).
⚠️ Report is 1 commits behind head on main.

Files with missing lines	Patch %	Lines
src/onnx_safetensors/_safetensors_io.py	88.50%	4 Missing and 6 partials ⚠️

Additional details and impacted files

@@            Coverage Diff             @@
##             main      #45      +/-   ##
==========================================
+ Coverage   68.60%   70.85%   +2.25%     
==========================================
  Files           6        6              
  Lines         379      374       -5     
  Branches       64       57       -7     
==========================================
+ Hits          260      265       +5     
+ Misses         96       92       -4     
+ Partials       23       17       -6     

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

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.

[Copilot]

Pull request overview

This pull request fixes bugs related to handling sub-byte data types in the onnx-safetensors library. The changes improve support for 2-bit, 4-bit, and other specialized numeric types.

Changes:

• Added support for new data types (INT2, UINT2, FLOAT4E2M1, FLOAT8E8M0, COMPLEX64) with proper safetensors mappings
• Simplified sub-byte type handling by using tensor.nbytes and tensor.dtype.bitwidth instead of custom helper functions
• Fixed struct unpacking format for reading safetensors headers from "i" (4-byte) to "<Q" (8-byte unsigned long long)
• Refactored save_file to eliminate code duplication and improve maintainability
• Updated save_model to restore original tensor values after saving, avoiding side effects on the input model
• Removed unused math and tqdm imports

Reviewed changes

Copilot reviewed 2 out of 2 changed files in this pull request and generated 8 comments.

File	Description
src/onnx_safetensors/_safetensors_io.py	Core changes for sub-byte type handling, dtype mappings, header reading fix, and code refactoring
src/onnx_safetensors/init.py	Version change from 1.5.0 to 1.4.1

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

[Copilot]

The variable values_to_save is populated but never used in the function. It should either be used for its intended purpose or removed to avoid confusion.

Suggested change

	values_to_save: list[ir.Value] = []
	for value in initialized_values:
	tensor = value.const_value
	assert tensor is not None
	if tensor.nbytes < size_threshold_bytes:
	continue
	tensors_to_save.append(tensor)
	values_to_save.append(value)
	for value in initialized_values:
	tensor = value.const_value
	assert tensor is not None
	if tensor.nbytes < size_threshold_bytes:
	continue
	tensors_to_save.append(tensor)

[Copilot]

The version is being changed from 1.5.0 to 1.4.1, which is a downgrade. Typically, bug fixes should increment the patch version (e.g., 1.5.0 to 1.5.1) rather than downgrade to a previous version. This could cause confusion with package versioning and updates.

Suggested change

	__version__ = "1.4.1"
	__version__ = "1.5.1"