Update app.py

app.py

@@ -114,7 +114,7 @@ def initialize_models(args):
     model = UnivaQwen2p5VLForConditionalGeneration.from_pretrained(
         args.model_path,
         torch_dtype=torch.bfloat16,
-        attn_implementation="flash_attention_2",
+        attn_implementation="sdpa",
         quantization_config=quantization_config if args.nf4 else None,
     ).to(device)
     task_head = nn.Sequential(