add support for w4a16_mixed

[n1ck-guo]

Description

add support for w4a16_mixed

Type of Change

• Bug fix (non-breaking change which fixes an issue)
• New feature (non-breaking change which adds functionality)
• Breaking change (fix or feature that would cause existing functionality to not work as expected)
• Documentation update
• Performance improvement
• Code refactoring
• Other (please describe):

Related Issues

Fixes #
Relates to #

Changes Made

• add new scheme w4a16_mixed

Testing

• Tested locally
• x ] Added/updated unit tests
• All existing tests pass
• Tested on specific hardware/environment (please specify):

Checklist

• My code follows the project's coding style
• I have performed a self-review of my code
• I have commented my code, particularly in hard-to-understand areas
• I have updated the documentation accordingly
• My changes generate no new warnings
• I have added tests that prove my fix is effective or that my feature works
• New and existing unit tests pass locally with my changes

Additional Context

[Copilot]

Pull request overview

This PR adds support for a new quantization scheme called W4A16_MIXED, which implements mixed-precision quantization for different model components. The scheme uses 4-bit weights for expert layers, 8-bit for regular layers, and 16-bit for attention layers in standard models or all non-expert layers in multimodal language models (MLLMs).

Changes:

• Added W4A16_MIXED scheme registration in the PRESET_SCHEMES dictionary
• Implemented special handling logic for W4A16_MIXED in _handle_special_schemes function with layer-specific bit precision assignment
• Updated output format support lists to include W4A16_MIXED scheme
• Added comprehensive test coverage for both MoE models and MLLM models

Reviewed changes

Copilot reviewed 4 out of 4 changed files in this pull request and generated 2 comments.

File	Description
auto_round/schemes.py	Added W4A16_MIXED to PRESET_SCHEMES and implemented special scheme handling with layer-specific quantization logic
auto_round/compressors/base.py	Refactored initialization order and updated _handle_special_schemes call to pass required parameters
auto_round/formats.py	Added W4A16_MIXED to supported schemes lists for AutoGPTQ and AutoRound output formats
test/test_cpu/schemes/test_scheme.py	Added two new test cases for W4A16_MIXED scheme covering both MoE and MLLM models

---

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