Upload FloSmolV

configuration_flosmolv.py

@@ -1,8 +1,6 @@
 from transformers import AutoConfig, PretrainedConfig
-
 class FloSmolVConfig(PretrainedConfig):
     model_type = "flosmolV"
-    is_composition = True 
 
     def __init__(
         self, 
@@ -10,11 +8,11 @@ class FloSmolVConfig(PretrainedConfig):
         llm_config=None,
         **kwargs,
       ):
+        super().__init__(**kwargs)
 
         if vision_config is None:
             vision_config = AutoConfig.from_pretrained("microsoft/Florence-2-base", trust_remote_code=True)
         if llm_config is None:
             llm_config = AutoConfig.from_pretrained("HuggingFaceTB/SmolLM-360M-Instruct", trust_remote_code=True)
         self.vision_config = vision_config
-        self.llm_config = llm_config
-        super().__init__(**kwargs)
+        self.llm_config = llm_config

modeling_flosmolv.py

@@ -7,8 +7,8 @@ from transformers import AutoModelForCausalLM, AutoTokenizer, AutoProcessor
 class FloSmolV(PreTrainedModel):
     config_class = FloSmolVConfig
     device = "cuda" if torch.cuda.is_available() else "cpu"
-    print(f"Device: {device}")
-    def __init__(self, config):
+
+    def __init__(self, config: FloSmolVConfig):
         super().__init__(config)
         self.florence2_model = AutoModelForCausalLM.from_pretrained(
             self.config.vision_config._name_or_path,