Transformers 5.0.0 requires custom models to define auto_map in their config for AutoModel.from_pretrained() to work. Without it, model loading fails.

Changes

• Configuration: Added auto_map initialization in OneVisionEncoderConfig.__init__() that maps AutoModel and AutoConfig to their corresponding custom classes

  • Sets default mapping when not provided or when explicitly None
  • Preserves custom mappings passed via kwargs

• Documentation: Updated README compatibility section to reflect support for transformers ≥4.53.1 including 5.0.0+

• Tests: Added test_transformers_5_compatibility.py covering config creation, save/load cycles, and custom auto_map handling

Before

config = OneVisionEncoderConfig()
# config.auto_map does not exist
# AutoModel.from_pretrained() fails with transformers 5.0+

After

config = OneVisionEncoderConfig()
config.auto_map
# {'AutoModel': 'modeling_onevision_encoder.OneVisionEncoderModel',
#  'AutoConfig': 'configuration_onevision_encoder.OneVisionEncoderConfig'}
# AutoModel.from_pretrained() works with both 4.x and 5.x

Verified on transformers 4.53.1 and 5.0.0rc1.

给我提一个issue，目前模型使用transformers 5.0.0加载有bug，使用英文

💡 You can make Copilot smarter by setting up custom instructions, customizing its development environment and configuring Model Context Protocol (MCP) servers. Learn more Copilot coding agent tips in the docs.