diff --git a/examples/benchmark.py b/examples/benchmark.py
index 331b1b0..3883849 100644
--- a/examples/benchmark.py
+++ b/examples/benchmark.py
@@ -181,7 +181,7 @@ def setup_llm(dp_rank: int) -> LLM:
         max_num_batched_tokens=args.prefill_batch_size,
         max_model_len=max_model_len,
         gpu_memory_utilization=0.95,
-        lookahead_tokens=min(1, int(os.getenv("DEEPINFER_USE_NEXTN_DECODE", "5"))),
+        lookahead_tokens=int(os.getenv("DEEPINFER_USE_NEXTN_DECODE", "5")),
     )
     return llm
 
diff --git a/examples/offline_inference.py b/examples/offline_inference.py
index 698ccea..ca64a91 100644
--- a/examples/offline_inference.py
+++ b/examples/offline_inference.py
@@ -53,7 +53,7 @@ def main(dp_size: int, dp_rank: int) -> None:
         max_model_len=32768,
         max_num_seqs=4,
         gpu_memory_utilization=0.85,
-        lookahead_tokens=min(1, int(os.getenv("DEEPINFER_USE_NEXTN_DECODE", "5"))),
+        lookahead_tokens=int(os.getenv("DEEPINFER_USE_NEXTN_DECODE", "5")),
     )
 
     from _prompts import OFFLINE_INPUT_LEN, OFFLINE_OUTPUT_LEN, lengthen_offline_prompts, prepare_offline_prompts