diff --git a/python/sglang/srt/model_executor/piecewise_cuda_graph_runner.py b/python/sglang/srt/model_executor/piecewise_cuda_graph_runner.py
index 1087b3b7f..a6bd7ab56 100644
--- a/python/sglang/srt/model_executor/piecewise_cuda_graph_runner.py
+++ b/python/sglang/srt/model_executor/piecewise_cuda_graph_runner.py
@@ -330,9 +330,9 @@ class PiecewiseCudaGraphRunner:
                 return_logprob=False,
                 extend_num_tokens=num_tokens,
                 extend_seq_lens=torch.tensor([num_tokens], device=self.device),
-                extend_prefix_lens=torch.tensor([num_tokens], device=self.device),
+                extend_prefix_lens=torch.tensor([0], device=self.device),
                 extend_start_loc=torch.tensor([0], device=self.device),
-                extend_prefix_lens_cpu=torch.tensor([num_tokens], device="cpu"),
+                extend_prefix_lens_cpu=torch.tensor([0], device="cpu"),
                 extend_seq_lens_cpu=torch.tensor([num_tokens], device="cpu"),
                 extend_logprob_start_lens_cpu=torch.tensor([num_tokens], device="cpu"),
                 positions=positions,
@@ -477,9 +477,9 @@ class PiecewiseCudaGraphRunner:
                 return_logprob=False,
                 extend_num_tokens=num_tokens,
                 extend_seq_lens=torch.tensor([num_tokens], device=self.device),
-                extend_prefix_lens=torch.tensor([num_tokens], device=self.device),
+                extend_prefix_lens=torch.tensor([0], device=self.device),
                 extend_start_loc=torch.tensor([0], device=self.device),
-                extend_prefix_lens_cpu=torch.tensor([num_tokens], device="cpu"),
+                extend_prefix_lens_cpu=torch.tensor([0], device="cpu"),
                 extend_seq_lens_cpu=torch.tensor([num_tokens], device="cpu"),
                 extend_logprob_start_lens_cpu=torch.tensor([num_tokens], device="cpu"),
                 positions=positions,