adds flash attention 2 to attention implementation

app.py

@@ -119,6 +119,7 @@ def load_model():
             "torch_dtype": torch.float16 if DEVICE == "cuda" else torch.float32,  # Use float16 on GPU, float32 on CPU
             "trust_remote_code": True,
             "low_cpu_mem_usage": True,
+            "attention_implementation": "flash_attention_2"
         }
         
         logger.info(f"Model loading parameters: {model_kwargs}")