Run CT on the ZeroGPU xlarge slice

ZeroGPU moved to RTX Pro 6000 Blackwell, where the default `large` size is
half a card (48 GB) rather than the old ~70 GB H200 MIG slice. The CT presets
were budgeted against ~70 GB (see ui/presets.py) and OOM on `large`, so CT now
requests `xlarge` (96 GB). This costs 2x quota and queues longer, so MR and
MR-Brain stay on `large` — they cap at 256^3 and fit comfortably.

Pin spaces>=0.47.0, the first release with the size= argument. The import guard
in app.py only catches ImportError/AttributeError, so an older spaces would
raise an uncaught TypeError at startup.

Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>

app.py

@@ -42,9 +42,17 @@ def _ensure_upstream():
 _ensure_upstream()
 
 # ZeroGPU decorator — present on HF Spaces, optional locally.
+#
+# `size` picks the ZeroGPU slice. ZeroGPU now runs on RTX Pro 6000 Blackwell:
+# `large` (default) is half a card = 48 GB, `xlarge` is a full card = 96 GB at
+# 2x quota cost and longer queue waits. Paired CT at the sizes we expose was
+# budgeted against the old ~70 GB H200 MIG slice (see ui/presets.py) and does
+# not fit in 48 GB, so CT takes `xlarge`. MR and MR-Brain cap at 256^3 and fit
+# comfortably in `large` — keeping them there halves their quota cost, which
+# matters since anonymous visitors only get 2 minutes of GPU time per day.
 try:
     import spaces  # type: ignore
-    spaces_gpu_ct = spaces.GPU(duration=90)
+    spaces_gpu_ct = spaces.GPU(duration=90, size="xlarge")
     spaces_gpu_mr = spaces.GPU(duration=60)
     spaces_gpu_mr_brain = spaces.GPU(duration=60)
 except (ImportError, AttributeError):

requirements.txt

@@ -1,6 +1,6 @@
 # Gradio app
 gradio>=5.0
-spaces  # ZeroGPU decorator (no-op locally)
+spaces>=0.47.0  # ZeroGPU decorator (no-op locally); >=0.47 for the size= arg
 
 # Upstream NV-Generate-CTMR dependencies
 torch>=2.1.0