chatbot fix sherika

phase/Student_view/chatbot.py

@@ -17,7 +17,11 @@ TUTOR_PROMPT = (
     "Teach step-by-step with tiny examples. Avoid giving personal financial advice."
 )
 
+# -------------------------------
+# History helpers
+# -------------------------------
 def _format_history_for_flan(messages: list[dict]) -> str:
+    """Format history for text-generation style models."""
     lines = []
     for m in messages:
         txt = (m.get("text") or "").strip()
@@ -26,14 +30,8 @@ def _format_history_for_flan(messages: list[dict]) -> str:
         lines.append(("Tutor" if m.get("sender") == "assistant" else "User") + f": {txt}")
     return "\n".join(lines)
 
-def _trim_turn(text: str) -> str:
-    for cp in ["\nUser:", "\nTutor:", "\nAssistant:", "\n###"]:
-        if cp in text:
-            return text.split(cp, 1)[0].strip()
-    return text.strip()
-
 def _history_as_chat_messages(messages: list[dict]) -> list[dict]:
-    # Convert your session history to HF chat format
+    """Convert history to chat-completion style messages."""
     msgs = [{"role": "system", "content": TUTOR_PROMPT}]
     for m in messages:
         txt = (m.get("text") or "").strip()
@@ -44,55 +42,53 @@ def _history_as_chat_messages(messages: list[dict]) -> list[dict]:
     return msgs
 
 def _extract_chat_text(chat_resp) -> str:
-    # Works for both dict-like and object-like responses
+    """Extract text from HF chat response."""
     try:
-        # huggingface_hub >= 0.23 style
-        return chat_resp.choices[0].message["content"] if isinstance(chat_resp.choices[0].message, dict) \
-            else chat_resp.choices[0].message.content
+        return chat_resp.choices[0].message["content"] if isinstance(
+            chat_resp.choices[0].message, dict
+        ) else chat_resp.choices[0].message.content
     except Exception:
-        # fallback for dict payloads
         try:
             return chat_resp["choices"][0]["message"]["content"]
         except Exception:
             return str(chat_resp)
 
+# -------------------------------
+# Reply logic
+# -------------------------------
 def _reply_with_hf():
     if "client" not in globals():
         raise RuntimeError("HF client not initialized")
 
-    # Text-generation prompt (for providers that support it)
-    convo = _format_history_for_flan(st.session_state.get("messages", []))
-    tg_prompt = f"{TUTOR_PROMPT}\n\n{convo}\n\nTutor:"
-
     try:
-        # 1) Try text-generation first (many backends support this)
-        resp = client.text_generation(
-            tg_prompt,
-            max_new_tokens=220,
+        # 1) Prefer chat API
+        msgs = _history_as_chat_messages(st.session_state.get("messages", []))
+        chat = client.chat.completions.create(
+            model=GEN_MODEL,
+            messages=msgs,
+            max_tokens=300,   # give enough room
             temperature=0.2,
             top_p=0.9,
-            repetition_penalty=1.1,
-            return_full_text=True,
-            stream=False,
         )
-        text = resp.get("generated_text") if isinstance(resp, dict) else resp
-        return _trim_turn(str(text or "").strip())
+        return _extract_chat_text(chat).strip()
 
     except ValueError as ve:
-        # 2) If the provider says "Supported task: conversational", use HF chat
-        if "Supported task: conversational" in str(ve):
-            msgs = _history_as_chat_messages(st.session_state.get("messages", []))
-            chat = client.chat.completions.create(
-                model=GEN_MODEL,          # still HF, not OpenAI
-                messages=msgs,
-                max_tokens=220,
+        # 2) Fallback to text-generation if chat unsupported
+        if "Supported task: text-generation" in str(ve):
+            convo = _format_history_for_flan(st.session_state.get("messages", []))
+            tg_prompt = f"{TUTOR_PROMPT}\n\n{convo}\n\nTutor:"
+            resp = client.text_generation(
+                tg_prompt,
+                max_new_tokens=300,
                 temperature=0.2,
                 top_p=0.9,
+                repetition_penalty=1.1,
+                return_full_text=True,
+                stream=False,
             )
-            return _trim_turn(_extract_chat_text(chat))
+            return (resp.get("generated_text") if isinstance(resp, dict) else resp).strip()
 
-        # Some other ValueError — rethrow with the original message
-        raise
+        raise  # rethrow anything else
 
     except Exception as e:
         err_text = ''.join(traceback.format_exception_only(type(e), e)).strip()