debug01

app.py

@@ -76,13 +76,13 @@ def add_docs(user_id: str, docs: list[str]) -> int:
 # ----- Qwen-chat prompt helper ---------------------------------------------
 def build_qwen_prompt(context: str, user_question: str) -> str:
     """Return a string that follows Qwen-Chat’s template."""
+    load_chat()                       # ← make sure tokenizer is ready
     conversation = [
         {"role": "system",
          "content": "You are an email assistant. Use ONLY the context provided."},
         {"role": "user",
          "content": f"Context:\n{context}\n\n{user_question}"}
     ]
-    # add_generation_prompt=True appends the assistant tag
     return tokenizer.apply_chat_template(
         conversation, tokenize=False, add_generation_prompt=True
     )
@@ -122,7 +122,8 @@ def rag(req:QueryReq):
     load_chat()
     inputs = tokenizer(prompt, return_tensors="pt").to(chat_model.device)
     out = chat_model.generate(**inputs, max_new_tokens=512)
-    ans = tokenizer.decode(out[0], skip_special_tokens=True).split("Assistant:",1)[-1].strip()
+    full = tokenizer.decode(out[0], skip_special_tokens=True)
+    ans  = full.split("<|im_start|>assistant")[-1].strip()
     return {"answer": ans}
 
 # ---------- 4. Gradio playground (same UI as before) --------------------------