Update agent_engine.py

agent_engine.py

@@ -18,22 +18,49 @@ CHAT_MODEL = os.environ.get("CHAT_MODEL", "HuggingFaceH4/zephyr-7b-beta")
 HF_API_TOKEN = os.environ.get("HF_API_TOKEN", "")
 logging.basicConfig(level=logging.INFO, format="%(asctime)s %(levelname)s %(message)s")
 
+def safe_chat(system_prompt: str, history: list, temperature: float = 0.7) -> str:
+    """Call HF inference API with fallback formatting."""
+    start = perf_counter()
+    chat_history = "\n".join([f"{msg['role'].capitalize()}: {msg['content']}" for msg in history])
+    full_prompt = f"{system_prompt}\n\n{chat_history}\n\nAssistant:"
 
-chat_history = "\n".join([f"{msg['role'].capitalize()}: {msg['content']}" for msg in history])
-full_prompt = f"{system_prompt}\n\n{chat_history}\n\nAssistant:"
-payload = {
-    "inputs": full_prompt,
-    "parameters": {"max_new_tokens": 300, "temperature": temperature}
-}
+    payload = {
+        "inputs": full_prompt,
+        "parameters": {"max_new_tokens": 300, "temperature": temperature}
+    }
+    headers = {"Authorization": f"Bearer {HF_API_TOKEN}"} if HF_API_TOKEN else {}
+
+    try:
+        resp = requests.post(
+            f"https://api-inference.huggingface.co/models/{CHAT_MODEL}",
+            json=payload,
+            headers=headers,
+            timeout=60
+        )
+        if resp.status_code == 200:
+            data = resp.json()
+            text = data[0].get("generated_text", "").strip() if isinstance(data, list) else data.get("generated_text", "").strip()
+        elif resp.status_code == 503:
+            logging.warning("Model loading… retrying after 15s.")
+            sleep(15)
+            return safe_chat(system_prompt, history, temperature)
+        else:
+            logging.error(f"HF error {resp.status_code}: {resp.text}")
+            text = f"⚠️ API Error {resp.status_code}"
+    except Exception as e:
+        logging.error(f"safe_chat exception: {e}")
+        text = f"⚠️ System Error: {e}"
+
+    elapsed = perf_counter() - start
+    logging.info(f"safe_chat: {elapsed:.3f}s for prompt '{system_prompt[:30]}…'")
+    return text
 
 def step_turn(conversation: list, turn: int, topic: str, params: dict) -> list:
     """Advance one turn of the multi-agent conversation."""
-    # Choose agent by sequence
     sequence = ["Initiator", "Responder", "Guardian", "Provocateur", "Cultural"]
     agent = sequence[turn % len(sequence)]
     prompt = PROMPTS.get(agent, "")
-    # Prepare history
-    history = [{"role": "user", "content": msg['text']} for msg in conversation[-5:]]
+    history = [{"role": "user", "content": msg['text']} for msg in conversation[-5:] if msg['agent'] != "System"]
     response = safe_chat(prompt, history, temperature=params[agent]['creativity'])
     embed_and_store(response, agent, topic)
     conversation.append({"agent": agent, "text": response, "turn": turn + 1})