diff --git a/single_shot_inference.py b/single_shot_inference.py
index de1ea9a6..146ab1ee 100644
--- a/single_shot_inference.py
+++ b/single_shot_inference.py
@@ -678,13 +678,22 @@ def main():
             gpu = li % NUM_GPUS
             if X.device != torch.device(f"cuda:{gpu}"): X = X.to(f"cuda:{gpu}")
             torch.cuda.set_device(gpu)
-            X = forward_layer(X, layer_w[li], li, cfg, *rope_caches[gpu],
-                              attn_mhcs.get(li), ffn_mhcs.get(li),
-                              attn_norms.get(li), ffn_norms.get(li),
-                              kv_caches[li], pos, tid,
-                              compressors.get(li), indexers.get(li),
-                              moe_runners.get(li), se_runners.get(li), routers.get(li),
-                              prod_lin=prod_lins.get(li))
+            try:
+                X = forward_layer(X, layer_w[li], li, cfg, *rope_caches[gpu],
+                                  attn_mhcs.get(li), ffn_mhcs.get(li),
+                                  attn_norms.get(li), ffn_norms.get(li),
+                                  kv_caches[li], pos, tid,
+                                  compressors.get(li), indexers.get(li),
+                                  moe_runners.get(li), se_runners.get(li), routers.get(li),
+                                  prod_lin=prod_lins.get(li))
+            except Exception as e:
+                torch.cuda.synchronize()
+                err = torch.cuda.current_stream(gpu).query()
+                print(f"  CRASH at token {pi} layer {li} gpu {gpu}: {e}", flush=True)
+                raise
+            if pi == 0 and li < 3:
+                torch.cuda.synchronize(gpu)
+                print(f"  Token {pi} L{li}: OK |X|={X.abs().max().item():.1f}", flush=True)
         X = X.to('cuda:0'); torch.cuda.set_device(0)
         if pi % 10 == 0: print(f"  Token {pi}/{len(generated)}: {time.time()-t1:.2f}s", flush=True)
     print(f"  Prefill done ({time.time()-t0:.1f}s)")