Upload model

model.safetensors

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:aeb6ca3b968752bebb66566507634992e3ca5462e92dd1d20c72557839427a7d
-size 783654820
+oid sha256:1a3cd354abbc4c07e66bcd2cb74e19741242034545947eec9d79b1128e73527e
+size 783662196

modeling_biomedclip.py

@@ -12,8 +12,10 @@ class BiomedCLIPModel(PreTrainedModel):
 
     def __init__(self, config: BiomedCLIPConfig):
         super().__init__(config)
-
-        self.model = CustomTextCLIP(
+        
+        # DON'T NAME THIS self.model or self.base_model
+        # (cf hardcoded name in transformers)
+        self.model_but_not_the_real_one = CustomTextCLIP(
             embed_dim=config.embed_dim,
             vision_cfg=CLIPVisionCfg(**config.vision_cfg),
             text_cfg=CLIPTextCfg(**config.text_cfg),
@@ -27,7 +29,7 @@ class BiomedCLIPModel(PreTrainedModel):
         text_inputs: Optional[torch.Tensor] = None,
         **kwargs
     ):
-        return self.model(vision_inputs, text_inputs, **kwargs)
+        return self.model_but_not_the_real_one(vision_inputs, text_inputs, **kwargs)
 
     def forward_intermediates(
         self,
@@ -35,5 +37,5 @@ class BiomedCLIPModel(PreTrainedModel):
         text_inputs: Optional[torch.Tensor] = None,
         **kwargs
     ) -> Dict[str, Union[torch.Tensor, List[torch.Tensor]]]:
-        return self.model.forward_intermediates(vision_inputs, text_inputs, **kwargs)
+        return self.model_but_not_the_real_one.forward_intermediates(vision_inputs, text_inputs, **kwargs)