Push model using huggingface_hub.

mini_gpt4_llama_v2.py

@@ -28,6 +28,7 @@ import os
 from transformers import PretrainedConfig
 from transformers import PreTrainedModel
 from typing import List
+from collections import defaultdict
 class minigpt4_video_config(PretrainedConfig):
     model_type="minigpt4_video"
     PRETRAINED_MODEL_CONFIG_DICT = {
@@ -42,9 +43,6 @@ class minigpt4_video_config(PretrainedConfig):
             setattr(self, key, value)
         super().__init__(**kwargs)
         
-    # def to_dict(self):
-    #     output = super().to_dict()
-    #     return output
         
 @registry.register_model("mini_gpt4_llama_v2")
 class MiniGPT4_Video(Blip2Base, PreTrainedModel):
@@ -62,8 +60,8 @@ class MiniGPT4_Video(Blip2Base, PreTrainedModel):
         cfg={},
     ):
         ## loop through the config minigpt4_video_config object and set the attributes
-        if isinstance(cfg, minigpt4_video_config):
-            cfg = cfg.to_dict()
+        # if isinstance(cfg, minigpt4_video_config):
+        cfg = cfg.to_dict()
         
         for key, value in cfg.items():
             try:
@@ -72,6 +70,10 @@ class MiniGPT4_Video(Blip2Base, PreTrainedModel):
                 print(f"Error setting attribute {key} with value {value}")
         PreTrainedModel.__init__(self, minigpt4_video_config(cfg))
         Blip2Base.__init__(self)
+        
+        vis_processor_cfg = {"name": "blip2_image_train","image_size": 224}
+        print(vis_processor_cfg)
+        self.vis_processor = registry.get_processor_class(vis_processor_cfg["name"]).from_config(vis_processor_cfg)
         if "Mistral" in self.llama_model:
             from minigpt4_video.modeling_mistral import MistralForCausalLM as llm_model
             print("Mistral model")