diff --git a/serve/mlc_serve/model/tvm_model.py b/serve/mlc_serve/model/tvm_model.py
index 6c37303c9f..42df04687a 100644
--- a/serve/mlc_serve/model/tvm_model.py
+++ b/serve/mlc_serve/model/tvm_model.py
@@ -300,7 +300,6 @@ def generate(
         all_token_ids = []
         sequence_ids = []
         prompt_lens = []
-        num_sequences = []
         # TODO(masahi, yelite): Update this when a new request type for speculative decoding
         # is implemented.
         num_decode_query_tokens = 1