diff --git a/fastdeploy/engine/sampling_params.py b/fastdeploy/engine/sampling_params.py
index 46d9fd8ac..1cd77d2b1 100644
--- a/fastdeploy/engine/sampling_params.py
+++ b/fastdeploy/engine/sampling_params.py
@@ -218,20 +218,22 @@ class SamplingParams:
                 prompt_token_ids = tokenizer.encode(text=prompt, add_special_tokens=False)["input_ids"]
 
                 if len(prompt_token_ids) != 1:
-                    logger.warning(
-                        f"Skip bad_words: {prompt}."
-                        f"Bad words should be a single token."
-                        f"Got tokens: {prompt_token_ids}."
-                    )
+                    if not add_prefix_space:
+                        logger.warning(
+                            f"Skip bad_words: <{prompt}>."
+                            f"Bad words should be a single token."
+                            f"Got tokens: {prompt_token_ids}."
+                        )
                     continue
 
                 if prompt_token_ids[0] > tokenizer.vocab_size:
-                    logger.warning(
-                        f"Skip bad_words: {prompt}."
-                        f"All token id values should be satisfying:"
-                        f" 0 <= token_id < {tokenizer.vocab_size}."
-                        f"Got token: {prompt_token_ids}."
-                    )
+                    if not add_prefix_space:
+                        logger.warning(
+                            f"Skip bad_words: <{prompt}>."
+                            f"All token id values should be satisfying:"
+                            f" 0 <= token_id < {tokenizer.vocab_size}."
+                            f"Got token: {prompt_token_ids}."
+                        )
                     continue
 
                 if prompt_token_ids not in self._bad_words_token_ids:
diff --git a/fastdeploy/worker/gcu_model_runner.py b/fastdeploy/worker/gcu_model_runner.py
index e35dc4d23..e0086b503 100644
--- a/fastdeploy/worker/gcu_model_runner.py
+++ b/fastdeploy/worker/gcu_model_runner.py
@@ -272,13 +272,15 @@ class GCUModelRunner(ModelRunnerBase):
                 request.block_tables, dtype="int32"
             )
 
-            if request.get("bad_words_token_ids") is not None:
+            if request.get("bad_words_token_ids") is not None and len(request.get("bad_words_token_ids")) > 0:
                 bad_words_len = len(request.get("bad_words_token_ids"))
-                if bad_words_len > 0:
-                    self.share_inputs["bad_tokens_len"][idx : idx + 1] = bad_words_len
-                    self.share_inputs["bad_tokens"][idx : idx + 1, :bad_words_len] = np.array(
-                        request.get("bad_words_token_ids"), dtype="int64"
-                    )
+                self.share_inputs["bad_tokens_len"][idx : idx + 1] = bad_words_len
+                self.share_inputs["bad_tokens"][idx : idx + 1, :bad_words_len] = np.array(
+                    request.get("bad_words_token_ids"), dtype="int64"
+                )
+            else:
+                self.share_inputs["bad_tokens_len"][idx : idx + 1] = 1
+                self.share_inputs["bad_tokens"][idx : idx + 1, :] = np.array([-1], dtype="int64")
 
             if request.get("stop_token_ids") is not None and request.get("stop_seqs_len") is not None:
                 stop_seqs_num = len(request.get("stop_seqs_len"))
diff --git a/fastdeploy/worker/gpu_model_runner.py b/fastdeploy/worker/gpu_model_runner.py
index c551364ef..447f249d4 100644
--- a/fastdeploy/worker/gpu_model_runner.py
+++ b/fastdeploy/worker/gpu_model_runner.py
@@ -495,13 +495,15 @@ class GPUModelRunner(ModelRunnerBase):
                 request.block_tables, dtype="int32"
             )
 
-            if request.get("bad_words_token_ids") is not None:
+            if request.get("bad_words_token_ids") is not None and len(request.get("bad_words_token_ids")) > 0:
                 bad_words_len = len(request.get("bad_words_token_ids"))
-                if bad_words_len > 0:
-                    self.share_inputs["bad_tokens_len"][idx : idx + 1] = bad_words_len
-                    self.share_inputs["bad_tokens"][idx : idx + 1, :bad_words_len] = np.array(
-                        request.get("bad_words_token_ids"), dtype="int64"
-                    )
+                self.share_inputs["bad_tokens_len"][idx : idx + 1] = bad_words_len
+                self.share_inputs["bad_tokens"][idx : idx + 1, :bad_words_len] = np.array(
+                    request.get("bad_words_token_ids"), dtype="int64"
+                )
+            else:
+                self.share_inputs["bad_tokens_len"][idx : idx + 1] = 1
+                self.share_inputs["bad_tokens"][idx : idx + 1, :] = np.array([-1], dtype="int64")
 
             if request.get("stop_token_ids") is not None and request.get("stop_seqs_len") is not None:
                 stop_seqs_num = len(request.get("stop_seqs_len"))
diff --git a/fastdeploy/worker/iluvatar_model_runner.py b/fastdeploy/worker/iluvatar_model_runner.py
index 526f3361e..4a7aaaf8d 100644
--- a/fastdeploy/worker/iluvatar_model_runner.py
+++ b/fastdeploy/worker/iluvatar_model_runner.py
@@ -243,13 +243,15 @@ class IluvatarModelRunner(ModelRunnerBase):
                 request.block_tables, dtype="int32"
             )
 
-            if request.get("bad_words_token_ids") is not None:
+            if request.get("bad_words_token_ids") is not None and len(request.get("bad_words_token_ids")) > 0:
                 bad_words_len = len(request.get("bad_words_token_ids"))
-                if bad_words_len > 0:
-                    self.share_inputs["bad_tokens_len"][idx : idx + 1] = bad_words_len
-                    self.share_inputs["bad_tokens"][idx : idx + 1, :bad_words_len] = np.array(
-                        request.get("bad_words_token_ids"), dtype="int64"
-                    )
+                self.share_inputs["bad_tokens_len"][idx : idx + 1] = bad_words_len
+                self.share_inputs["bad_tokens"][idx : idx + 1, :bad_words_len] = np.array(
+                    request.get("bad_words_token_ids"), dtype="int64"
+                )
+            else:
+                self.share_inputs["bad_tokens_len"][idx : idx + 1] = 1
+                self.share_inputs["bad_tokens"][idx : idx + 1, :] = np.array([-1], dtype="int64")
 
             if request.get("stop_token_ids") is not None and request.get("stop_seqs_len") is not None:
                 stop_seqs_num = len(request.get("stop_seqs_len"))
diff --git a/fastdeploy/worker/xpu_model_runner.py b/fastdeploy/worker/xpu_model_runner.py
index b61e84b9f..a153e5556 100644
--- a/fastdeploy/worker/xpu_model_runner.py
+++ b/fastdeploy/worker/xpu_model_runner.py
@@ -507,13 +507,15 @@ class XPUModelRunner(ModelRunnerBase):
                 request.block_tables, dtype="int32"
             )
 
-            if request.get("bad_words_token_ids") is not None:
+            if request.get("bad_words_token_ids") is not None and len(request.get("bad_words_token_ids")) > 0:
                 bad_words_len = len(request.get("bad_words_token_ids"))
-                if bad_words_len > 0:
-                    self.share_inputs["bad_tokens_len"][idx : idx + 1] = bad_words_len
-                    self.share_inputs["bad_tokens"][idx : idx + 1, :bad_words_len] = np.array(
-                        request.get("bad_words_token_ids"), dtype="int64"
-                    )
+                self.share_inputs["bad_tokens_len"][idx : idx + 1] = bad_words_len
+                self.share_inputs["bad_tokens"][idx : idx + 1, :bad_words_len] = np.array(
+                    request.get("bad_words_token_ids"), dtype="int64"
+                )
+            else:
+                self.share_inputs["bad_tokens_len"][idx : idx + 1] = 1
+                self.share_inputs["bad_tokens"][idx : idx + 1, :] = np.array([-1], dtype="int64")
 
             if request.get("stop_token_ids") is not None and request.get("stop_seqs_len") is not None:
                 stop_seqs_num = len(request.get("stop_seqs_len"))