Update app.py

app.py

@@ -123,12 +123,12 @@ def initialize_llm():
         model.eval()
         
         # Create pipeline for generation
+        # NOTE: When using accelerate/quantization, do NOT specify device parameter
         logger.info("   Creating text-generation pipeline...")
         llm_client = pipeline(
             "text-generation",
             model=model,
             tokenizer=tokenizer,
-            device=0 if device == "cuda" else -1,
             max_new_tokens=512,
             pad_token_id=tokenizer.eos_token_id
         )