Update app.py

app.py

@@ -205,10 +205,20 @@ def load_model():
         tokenizer.add_special_tokens(special_tokens)
         
         print("Loading model from Hugging Face Hub...")
+        # Create config first
+        config = SmolLM2Config(
+            pad_token_id=tokenizer.pad_token_id,
+            bos_token_id=tokenizer.bos_token_id,
+            eos_token_id=tokenizer.eos_token_id
+        )
+        
+        # Load model from Hub
         model = AutoModelForCausalLM.from_pretrained(
             model_id,
+            config=config,
             torch_dtype=torch.float16 if torch.cuda.is_available() else torch.float32,
-            trust_remote_code=True
+            trust_remote_code=True,
+            low_cpu_mem_usage=True
         )
         
         # Move model to device manually