diff --git a/litellm/utils.py b/litellm/utils.py index 0c96527d2..efc3a0782 100644 --- a/litellm/utils.py +++ b/litellm/utils.py @@ -10080,7 +10080,14 @@ class CustomStreamWrapper: self.sent_last_chunk = False self.system_fingerprint: Optional[str] = None self.received_finish_reason: Optional[str] = None - self.special_tokens = ["<|assistant|>", "<|system|>", "<|user|>", "", ""] + self.special_tokens = [ + "<|assistant|>", + "<|system|>", + "<|user|>", + "", + "", + "<|im_end|>", + ] self.holding_chunk = "" self.complete_response = "" self.response_uptil_now = ""