diff --git a/single_shot_inference.py b/single_shot_inference.py
index 27a06945..0040f402 100644
--- a/single_shot_inference.py
+++ b/single_shot_inference.py
@@ -1053,6 +1053,15 @@ def main():
         logits = lm_head_lin(x_out)
         if profile: torch.cuda.synchronize()
         t_lm = time.perf_counter()
+        # Check thinking start token logit on first step
+        if step == 0:
+            ls = logits.float()
+            for tid, name in [(THINK_START, 'think_start'), (THINK_END, 'think_end'), (USER_TOKEN, 'user'), (ASSISTANT_TOKEN, 'assistant')]:
+                print(f"  {name}({tid}) logit={ls[0, tid].item():.2f}", flush=True)
+            # Also check "Paris" token
+            paris_tids = [t for t in range(min(129280, ls.shape[-1])) if 'Paris' in tokenizer.decode([t])]
+            if paris_tids:
+                print(f"  Paris tokens: {[(t, ls[0,t].item()) for t in paris_tids[:5]]}", flush=True)
         # Only sync + validate on first 3 steps and every 20th step (reduces pipeline stalls)
         if step < 3 or (step + 1) % 20 == 0:
             torch.cuda.synchronize()  # catch CUDA errors at source