remove think_end_token_id in streaming content (#3327)

AllentDan · web-flow · commit 448491b291d9 · 2025-03-26T15:21:08.000+08:00
diff --git a/lmdeploy/serve/openai/reasoning_parser/deepseek_r1_reasoning_parser.py b/lmdeploy/serve/openai/reasoning_parser/deepseek_r1_reasoning_parser.py
@@ -51,8 +51,11 @@ def extract_reasoning_content_streaming(
         about what has previously been parsed and extracted (see constructor)
         """
         # Skip single special tokens
-        if len(delta_token_ids) == 1 and (delta_token_ids[0] in [self.think_start_token_id, self.think_end_token_id]):
-            return None
+        if len(delta_token_ids) == 1:
+            if delta_token_ids[0] == self.think_end_token_id:
+                return DeltaMessage(content='')
+            elif delta_token_ids[0] == self.think_start_token_id:
+                return None
 
         # Check if <think> is present in previous or delta.
         # Keep compatibility with models that don't generate <think> tokens.
