Update app.py

app.py

@@ -35,15 +35,13 @@ def load_model_and_tokenizer(model_identifier: str, model_key: str, tokenizer_ke
     try:
         tokenizer = AutoTokenizer.from_pretrained(
             model_identifier, 
-            trust_remote_code=True,
-            use_auth_token=False  # Ensure we're not using auth for public models
+            trust_remote_code=True
         )
         model = AutoModelForCausalLM.from_pretrained(
             model_identifier,
             torch_dtype=torch.bfloat16,
             device_map="auto",
-            trust_remote_code=True,
-            use_auth_token=False  # Ensure we're not using auth for public models
+            trust_remote_code=True
         )
         model.eval()
 
@@ -54,13 +52,35 @@ def load_model_and_tokenizer(model_identifier: str, model_key: str, tokenizer_ke
         
         _models_cache[model_key] = model
         _models_cache[tokenizer_key] = tokenizer
-        print(f"Finished loading and cached {model_key} and {tokenizer_key}.")
+        print(f"✅ Successfully loaded {model_key} model!")
         return model, tokenizer
     except Exception as e:
-        print(f"ERROR loading {model_key} model ({model_identifier}): {e}")
-        print(f"Error type: {type(e).__name__}")
-        if "404" in str(e) or "not found" in str(e).lower():
-            print(f"Model {model_identifier} not found. Please check the model ID.")
+        print(f"❌ ERROR loading {model_key} model ({model_identifier}): {e}")
+        
+        # FALLBACK: Use base model if fine-tuned model fails
+        if model_key == "finetuned" and model_identifier != BASE_MODEL_ID:
+            print(f"🔄 FALLBACK: Loading base model instead for fine-tuned model...")
+            try:
+                tokenizer = AutoTokenizer.from_pretrained(BASE_MODEL_ID, trust_remote_code=True)
+                model = AutoModelForCausalLM.from_pretrained(
+                    BASE_MODEL_ID,
+                    torch_dtype=torch.bfloat16,
+                    device_map="auto",
+                    trust_remote_code=True
+                )
+                model.eval()
+                if tokenizer.pad_token is None:
+                    tokenizer.pad_token = tokenizer.eos_token
+                    if hasattr(tokenizer, "pad_token_id") and tokenizer.pad_token_id is None and tokenizer.eos_token_id is not None:
+                        tokenizer.pad_token_id = tokenizer.eos_token_id
+                
+                _models_cache[model_key] = model
+                _models_cache[tokenizer_key] = tokenizer
+                print(f"✅ FALLBACK successful! Using base model with CineGuide prompt.")
+                return model, tokenizer
+            except Exception as fallback_e:
+                print(f"❌ FALLBACK also failed: {fallback_e}")
+        
         _models_cache[model_key] = "error"
         _models_cache[tokenizer_key] = "error"
         raise
@@ -189,6 +209,7 @@ with gr.Blocks(theme=gr.themes.Soft(), title="🎬 CineGuide Comparison") as dem
         Type your movie-related query below and see how fine-tuning improves movie recommendations!
         
         ⚠️ **Note:** Models are loaded on first use and may take 30-60 seconds initially.
+        💡 **Fallback:** If fine-tuned model fails, will use base model with specialized prompting.
         """
     )