Upload app.py with huggingface_hub

app.py

@@ -61,8 +61,8 @@ def split_reasoning(text: str, enable_thinking: bool) -> tuple[str, str]:
         return "", text.strip()
     if "</think>" in text:
         reasoning, answer = text.split("</think>", 1)
-        return reasoning.strip(), answer.strip()
-    return text.strip(), ""
+        return reasoning.replace("<think>", "").strip(), answer.strip()
+    return text.replace("<think>", "").strip(), ""
 
 
 def history_to_messages(history: list[object] | None) -> list[dict[str, str]]:
@@ -150,7 +150,7 @@ def respond(
     ).to(model.device)
 
     streamer = TextIteratorStreamer(
-        tokenizer, skip_prompt=True, skip_special_tokens=False
+        tokenizer, skip_prompt=True, skip_special_tokens=True
     )
     gen_kwargs = dict(
         **inputs,