nvidia

/

PS3-1.5K-SigLIP

like 5

Follow

NVIDIA 38.3k

When loading the checkpoint via vision_model = PS3VisionModel.from_pretrained("nvidia/PS3-4K-SigLIP"), I get the following error:

File "/path/to/miniforge3/lib/python3.12/site-packages/transformers/modeling_utils.py", line 40
90, in from_pretrained                                                                                
    model = cls(config, *model_args, **model_kwargs)                                                  
            ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^                                                  
  File "/path/to/PS3/ps3/modeling_ps3.py", line 808, in __init__                                 
    self.vision_model = PS3VisionEncoder(config)                                                      
                        ^^^^^^^^^^^^^^^^^^^^^^^^                                                      
  File "/path/to/PS3/ps3/modeling_ps3.py", line 90, in __init__                                  
    self.trunk = timm.create_model(                                                                   
                 ^^^^^^^^^^^^^^^^^^                                                                   
  File "/path/to/miniforge3/lib/python3.12/site-packages/timm/models/_factory.py", line 138, in c
reate_model            
    model = create_fn(                             
            ^^^^^^^^^^                                                                                
  File "/path/to/miniforge3/lib/python3.12/site-packages/timm/models/vision_transformer.py", line
 3637, in vit_so400m_patch14_siglip_384            
    model = _create_vision_transformer(
            ^^^^^^^^^^^^^^^^^^^^^^^^^^^                                                               
  File "/path/to/miniforge3/lib/python3.12/site-packages/timm/models/vision_transformer.py", line
 2679, in _create_vision_transformer
    return build_model_with_cfg(
           ^^^^^^^^^^^^^^^^^^^^^                                                                      
  File "/path/to/miniforge3/lib/python3.12/site-packages/timm/models/_builder.py", line 445, in b
uild_model_with_cfg                                                                                   
    model = model_cls(**kwargs)                                                                       
            ^^^^^^^^^^^^^^^^^^^                                                                       
  File "/path/to/miniforge3/lib/python3.12/site-packages/timm/models/vision_transformer.py", line
 583, in __init__
    block_fn(                                                                                         
TypeError: Block_w_KVCache.__init__() got an unexpected keyword argument 'scale_attn_norm'

I tried installing ps3 both via pip and the github repo. I get the same error in both cases.

Dependency versions:

timm                     1.0.19
torch                    2.7.1
torchvision              0.22.1
transformers             4.48.0