Upload app.py with huggingface_hub

app.py

@@ -211,10 +211,11 @@ model = Qwen2_5OmniThinkerForConditionalGeneration.from_pretrained(
     attn_implementation="sdpa",
     trust_remote_code=False,
     low_cpu_mem_usage=True,
-).to("cuda")
+)
 
 print("Loading LoRA adapter...", flush=True)
 model = PeftModel.from_pretrained(model, ADAPTER, is_trainable=False)
+model = model.to("cuda")
 model.eval()
 
 # Determine input device and context limit