aremoves cache parameter

app.py

@@ -118,9 +118,12 @@ class ModelCacheManager:
                 try:
                     return method(*args, **kwargs)
                 except:
-                    # Try without cache
+                    # Try without any cache-related parameters
                     kwargs_copy = kwargs.copy()
-                    kwargs_copy['use_cache'] = False
+                    # Remove any cache-related parameters that might cause issues
+                    for key in list(kwargs_copy.keys()):
+                        if 'cache' in key.lower():
+                            del kwargs_copy[key]
                     return method(*args, **kwargs_copy)
             else:
                 raise e
@@ -128,13 +131,15 @@ class ModelCacheManager:
     def direct_call(self, method_name, *args, **kwargs):
         """Direct call bypassing all cache mechanisms"""
         try:
-            # Disable cache completely
-            kwargs_copy = kwargs.copy()
-            kwargs_copy['use_cache'] = False
-            
             # Clear all caches first
             self._clear_all_caches()
             
+            # Remove any cache-related parameters
+            kwargs_copy = kwargs.copy()
+            for key in list(kwargs_copy.keys()):
+                if 'cache' in key.lower():
+                    del kwargs_copy[key]
+            
             # Make the call
             method = getattr(self.model, method_name)
             return method(*args, **kwargs_copy)
@@ -149,8 +154,9 @@ class ModelCacheManager:
             kwargs_copy = kwargs.copy()
             
             # Remove any cache-related parameters
-            if 'use_cache' in kwargs_copy:
-                del kwargs_copy['use_cache']
+            for key in list(kwargs_copy.keys()):
+                if 'cache' in key.lower():
+                    del kwargs_copy[key]
             
             # Clear caches
             self._clear_all_caches()
@@ -187,7 +193,6 @@ def initialize_model_safely():
             device_map=device, 
             use_safetensors=True,
             pad_token_id=tokenizer.eos_token_id,
-            use_cache=True,
             torch_dtype=torch.float16 if device == 'cuda' else torch.float32
         )
         
@@ -276,20 +281,12 @@ def safe_model_chat(model, tokenizer, image_path, **kwargs):
                 try:
                     # Remove any cache-related parameters
                     kwargs_copy = kwargs.copy()
-                    if 'use_cache' in kwargs_copy:
-                        del kwargs_copy['use_cache']
-                    
-                    # Try with cache disabled
+                    for key in list(kwargs_copy.keys()):
+                        if 'cache' in key.lower():
+                            del kwargs_copy[key]
                     return model.chat(tokenizer, image_path, **kwargs_copy)
                 except:
-                    # Last resort: try to recreate the model call without cache
-                    try:
-                        # Force cache clearing by setting use_cache=False
-                        kwargs_copy = kwargs.copy()
-                        kwargs_copy['use_cache'] = False
-                        return model.chat(tokenizer, image_path, **kwargs_copy)
-                    except:
-                        raise Exception("Model compatibility issue: DynamicCache error. Please try again.")
+                    raise Exception("Model compatibility issue: DynamicCache error. Please try again.")
         else:
             raise e
     except Exception as e:
@@ -336,20 +333,12 @@ def safe_model_chat_crop(model, tokenizer, image_path, **kwargs):
                 try:
                     # Remove any cache-related parameters
                     kwargs_copy = kwargs.copy()
-                    if 'use_cache' in kwargs_copy:
-                        del kwargs_copy['use_cache']
-                    
-                    # Try with cache disabled
+                    for key in list(kwargs_copy.keys()):
+                        if 'cache' in key.lower():
+                            del kwargs_copy[key]
                     return model.chat_crop(tokenizer, image_path, **kwargs_copy)
                 except:
-                    # Last resort: try to recreate the model call without cache
-                    try:
-                        # Force cache clearing by setting use_cache=False
-                        kwargs_copy = kwargs.copy()
-                        kwargs_copy['use_cache'] = False
-                        return model.chat_crop(tokenizer, image_path, **kwargs_copy)
-                    except:
-                        raise Exception("Model compatibility issue: DynamicCache error. Please try again.")
+                    raise Exception("Model compatibility issue: DynamicCache error. Please try again.")
         else:
             raise e
     except Exception as e: