fix: load model directly to cuda like ltx-2-TURBO, enable flash_attention_2

app.py

@@ -38,8 +38,8 @@ model = AutoModelForImageTextToText.from_pretrained(
     torch_dtype=torch.bfloat16,
     attn_implementation=_attn_impl,
     trust_remote_code=True,
-).eval()
-print(f"[AD-Copilot] Attention: {_attn_impl}", flush=True)
+).to("cuda").eval()
+print(f"[AD-Copilot] Attention: {_attn_impl} | Device: {model.device}", flush=True)
 
 
 # ---------------------------------------------------------------------------