Gate full-GPU fast path on slice size; it OOMs on this Space's 47GB MIG

Verified on the FireRed-Dev logs: the previous commit's full pipe.to(device)
attempt OOM'd on the 2g.48gb (47.4GB) MIG slice this Space actually gets
(peaked at 46.82GB), wasting ~40s on the failed attempt + cleanup before
falling back to cpu offload — strictly worse than the offload-only baseline.
Only attempt the fast path when the allocation has real headroom (>=60GB);
otherwise go straight to enable_model_cpu_offload as before.

Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>

app.py

@@ -404,20 +404,27 @@ def _infer_gpu(pil_images, prompt, seed, guidance_scale, steps, width, height, m
 
     # Each ZeroGPU call runs in a fresh worker (hooks are always unset here),
     # so cpu_offload buys no cross-call reuse — it only trades one bulk
-    # to(device) transfer for several slower hook-managed ones. Try moving
-    # the full ~37GB pipeline to GPU in one shot first (fast); only fall back
-    # to enable_model_cpu_offload (keeps one component resident at a time) if
-    # that OOMs on a smaller MIG slice (e.g. the 47GB 2g.48gb partition).
+    # to(device) transfer for several slower hook-managed ones. On a
+    # generously-sized allocation, move the full ~37GB pipeline to GPU in one
+    # shot (fast); but the 47GB 2g.48gb MIG slice this Space actually gets
+    # measured a peak of 46.82GB and still OOM'd on that path, wasting ~40s
+    # before falling back — so only attempt it when there's real headroom
+    # above that, otherwise go straight to enable_model_cpu_offload.
+    _FAST_PATH_MIN_GB = 60
     if getattr(pipe.transformer, "_hf_hook", None) is None:
-        try:
-            pipe.to(device)
-            print(f"[infer] moved full pipe to {device} — t={time.perf_counter()-t0:.1f}s")
-        except torch.cuda.OutOfMemoryError:
-            print(f"[infer] OOM moving full pipe to {device}, falling back to cpu offload")
-            pipe.to("cpu")
-            torch.cuda.empty_cache()
+        if _cuda_ok and p.total_memory / 1024**3 >= _FAST_PATH_MIN_GB:
+            try:
+                pipe.to(device)
+                print(f"[infer] moved full pipe to {device} — t={time.perf_counter()-t0:.1f}s")
+            except torch.cuda.OutOfMemoryError:
+                print(f"[infer] OOM moving full pipe to {device}, falling back to cpu offload")
+                pipe.to("cpu")
+                torch.cuda.empty_cache()
+                pipe.enable_model_cpu_offload(device=device)
+                print(f"[infer] enabled cpu offload on {device} (fallback)")
+        else:
             pipe.enable_model_cpu_offload(device=device)
-            print(f"[infer] enabled cpu offload on {device} (fallback)")
+            print(f"[infer] enabled cpu offload on {device} (slice too small for fast path)")
     print(f"[infer] {_gpu_mem_str(_cuda_ok)} — t={time.perf_counter()-t0:.1f}s")
 
     print(f"[infer] {len(pil_images)} image(s) pre-decoded, output={width}x{height}, seed={seed}")