Time transformer/pipeline loads; allow pipeline to fetch config from Hub

Adds wall-clock seconds to each from_pretrained step so we can see
how long the xet-mount read takes. Removes local_files_only from the
pipeline load — its cache is missing entirely but the config files are
small and will pull from Hub quickly.

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>

app.py

@@ -65,6 +65,8 @@ if os.path.isdir(_hf_cache):
 _inspect_cache("transformer (V23)", os.path.join(_hf_cache, "models--prithivMLmods--Qwen-Image-Edit-Rapid-AIO-V23"))
 _inspect_cache("pipeline (FireRed-Image-Edit-1.1)", os.path.join(_hf_cache, "models--FireRedTeam--FireRed-Image-Edit-1.1"))
 
+import time as _time
+_t0_load = _time.perf_counter()
 print("[startup] loading transformer from_pretrained (prithivMLmods/Qwen-Image-Edit-Rapid-AIO-V23)...", flush=True)
 _transformer = QwenImageTransformer2DModel.from_pretrained(
     "prithivMLmods/Qwen-Image-Edit-Rapid-AIO-V23",
@@ -72,16 +74,16 @@ _transformer = QwenImageTransformer2DModel.from_pretrained(
     device_map="cuda",
     local_files_only=True,
 )
-print("[startup] transformer loaded", flush=True)
+print(f"[startup] transformer loaded in {_time.perf_counter()-_t0_load:.1f}s", flush=True)
 
+_t1_load = _time.perf_counter()
 print("[startup] loading pipeline from_pretrained (FireRedTeam/FireRed-Image-Edit-1.1)...", flush=True)
 pipe = QwenImageEditPlusPipeline.from_pretrained(
     "FireRedTeam/FireRed-Image-Edit-1.1",
     transformer=_transformer,
     torch_dtype=dtype,
-    local_files_only=True,
 ).to(device)
-print("[startup] pipeline loaded and moved to device", flush=True)
+print(f"[startup] pipeline loaded in {_time.perf_counter()-_t1_load:.1f}s", flush=True)
 
 print("Using default attention processor (FA3 skipped for ZeroGPU GPU-arch compatibility).", flush=True)