Update app.py

app.py

@@ -10,10 +10,18 @@ def load_model_and_adapter():
     base_model_name = "unsloth/Llama-3.2-3B-Instruct"  # Replace with your base model name
     adapter_repo = "Futuresony/future_ai_12_10_2024"   # Your Hugging Face LoRA repo
     
-    # Load tokenizer and base model
+    # Load tokenizer
     tokenizer = AutoTokenizer.from_pretrained(base_model_name)
+    
+    # Fix rope_scaling in model configuration
+    config = AutoModelForCausalLM.config_class.from_pretrained(base_model_name)
+    if hasattr(config, "rope_scaling"):
+        config.rope_scaling = {"type": "dynamic", "factor": 32.0}  # Override with valid keys
+    
+    # Load base model with fixed config
     base_model = AutoModelForCausalLM.from_pretrained(
         base_model_name,
+        config=config,
         torch_dtype=torch.float16,  # Use float16 for efficiency if GPU is available
         device_map="auto"           # Automatically map to GPU or CPU
     )