Add outlines package and fix Gradio message format compatibility

- Add outlines>=0.0.37 for complete guided decoding support
- Strip Gradio-specific fields (metadata, options) from messages
before sending to vLLM OpenAI-compatible API
- This fixes the 400 Bad Request error from vLLM

app.py

@@ -86,7 +86,9 @@ def ensure_ready():
 def chat_fn(user_message: str, history: list[dict]):
     if not ensure_ready():
         return "⏳ Model is loading… please retry shortly."
-    messages = [{"role":"system","content":SYSTEM_PROMPT}] + history + [{"role":"user","content":user_message}]
+    # Strip Gradio-specific fields (metadata, options) for OpenAI compatibility
+    clean_history = [{"role": m["role"], "content": m["content"]} for m in history]
+    messages = [{"role":"system","content":SYSTEM_PROMPT}] + clean_history + [{"role":"user","content":user_message}]
     payload = {"model":"excom-ai","messages":messages,"temperature":0.4}
     r = requests.post(f"http://127.0.0.1:{API_PORT}/v1/chat/completions", json=payload, timeout=600)
     if not r.ok:

requirements.txt

@@ -11,4 +11,5 @@ transformers>=4.44
 accelerate>=0.30
 
 # Required for vLLM's outlines guided decoding backend
+outlines>=0.0.37
 airportsdata>=20240400