apple

/

aimv2-large-patch14-224

like 62

Follow

Apple 6.58k

This PR makes the uploaded checkpoint consistent with the custom modeling_aimv2.py implementation shipped in the repository.

• remap tensor names in model.safetensors to match AIMv2Model.state_dict()
• update config.json so architectures and model_type align with the shipped config/model classes

Without these changes, loading via AutoModel.from_pretrained(..., trust_remote_code=True) initializes most model weights from scratch instead of loading the published checkpoint.