Chunked attention on CUDA too (25K-token inputs fit a ZeroGPU slice)

app.py

@@ -60,8 +60,12 @@ import jev_style_decision as rt  # noqa: E402  (the model repo's runtime, same c
 
 _t0 = time.perf_counter()
 MODEL = rt.JevStyleDecision(MODEL_DIR, device=DEVICE, dtype="float32", verify=True)
-print(f"{REPO}@{REVISION[:7]} loaded on {DEVICE} in {time.perf_counter() - _t0:.1f} s (float32, manifest ok)",
-      flush=True)
+# The runtime turns on its query-chunked attention only off CUDA. A ZeroGPU slice runs out of memory in plain SDPA
+# near 25K tokens (float32 score matrix), so use the same chunked path on CUDA too.
+if not MODEL.chunked_attention:
+    MODEL.chunked_attention = rt._enable_chunked_attention(MODEL.model, rt.ATTN_CHUNK)
+print(f"{REPO}@{REVISION[:7]} loaded on {DEVICE} in {time.perf_counter() - _t0:.1f} s (float32, manifest ok, "
+      f"chunked attention {MODEL.chunked_attention})", flush=True)
 
 KINDS = ["Choice", "Yes / No", "Score"]
 OPTION_LABEL = {"Choice": "Options · one per line · name: description (optional)",