Update backend/chat_router.py

backend/chat_router.py

@@ -1,63 +1,58 @@
-import os, json, time, requests
-from utils.constants import ROUTER_CHAT_URL, ROUTER_MODEL, REQUEST_TIMEOUT_SECONDS_DEFAULT, RETRIES_DEFAULT, BACKOFF_SECONDS_DEFAULT
+# -------------------------------
+# 👇 HuggingFace OpenAI-Compatible Client
+# -------------------------------
+from openai import OpenAI
+import os
+from utils.constants import (
+    ROUTER_CHAT_URL,
+    ROUTER_MODEL,
+    REQUEST_TIMEOUT_SECONDS_DEFAULT,
+)
 from utils.persona import AI_GYNO_PERSONA_V3
 
-ROUTER_MODEL = "meta-llama/Meta-Llama-3-8B-Instruct"
+# Force correct model name
+ROUTER_MODEL = "meta-llama/Llama-3.1-8B-Instruct"
 
-BASE_URL = f"https://router.huggingface.co/hf-inference/chat-completions/{ROUTER_MODEL}"
+# HF Token
+token = os.getenv("HF_API_TOKEN") or os.getenv("HF_TOKEN")
 
-def _headers():
-    tok = os.getenv("HF_API_TOKEN") or os.getenv("HF_TOKEN")
-    return {"Authorization": f"Bearer {tok}", "Content-Type": "application/json"} if tok else {}
+# HF Router client
+client = OpenAI(
+    base_url="https://router.huggingface.co/v1",
+    api_key=token,
+)
 
-def _endpoint():
-    # Allow override via env; otherwise default to chat-completions route
-    base = os.getenv("HF_CHAT_BASE") or ROUTER_CHAT_URL
-    return f"{base}/{ROUTER_MODEL}"
 
 def chat(user_message: str, mode: str = "patient") -> str:
-    headers = _headers()
-    if not headers:
-        return "⚠ Set HF_API_TOKEN (or HF_TOKEN) in Secrets."
+    """
+    Uses HuggingFace Router with OpenAI SDK to get chat completions.
+    """
+
+    if not token:
+        return "⚠ Set HF_API_TOKEN or HF_TOKEN in your environment."
+
+    # Patient-friendly language vs Clinical doctor mode
+    style = (
+        "Use simple, reassuring language."
+        if mode == "patient"
+        else "Use concise clinical phrasing with differentials and next steps."
+    )
 
-    style = "Use simple, reassuring language." if mode=="patient" else "Use concise clinical phrasing with differentials and next steps."
     system_prompt = AI_GYNO_PERSONA_V3 + f"\nMode: {mode}. {style}"
 
-    payload = {
-        "model": ROUTER_MODEL,
-        "messages": [
-            {"role":"system","content": system_prompt},
-            {"role":"user","content": user_message}
-        ],
-        "max_tokens": 400,
-        "temperature": 0.2
-    }
-
-    url = _endpoint()
-
-    for attempt in range(1, RETRIES_DEFAULT+1):
-        try:
-            r = requests.post(url, headers=headers, json=payload, timeout=REQUEST_TIMEOUT_SECONDS_DEFAULT)
-            txt = r.text
-            # Prefer JSON
-            try:
-                data = r.json()
-            except Exception:
-                if "not found" in txt.lower():
-                    return "❌ Router 404: model not available for chat-completions route. Try Meta-Llama-3-8B-Instruct."
-                return txt[:2000]
-
-            # OpenAI-compatible chat-completions
-            if isinstance(data, dict) and "choices" in data and data["choices"]:
-                msg = data["choices"][0].get("message",{}).get("content","")
-                if msg: return msg
-            # Text-generation style fallback
-            if isinstance(data, list) and data and "generated_text" in data[0]:
-                return data[0]["generated_text"]
-            if isinstance(data, dict) and "generated_text" in data:
-                return data["generated_text"]
-
-            return "⚠ Unexpected response: " + json.dumps(data)[:1200]
-        except Exception as e:
-            time.sleep(BACKOFF_SECONDS_DEFAULT * attempt)
-    return "❌ Endpoint unreachable after retries."
+    try:
+        completion = client.chat.completions.create(
+            model=ROUTER_MODEL,
+            messages=[
+                {"role": "system", "content": system_prompt},
+                {"role": "user", "content": user_message},
+            ],
+            max_tokens=400,
+            temperature=0.2,
+            timeout=REQUEST_TIMEOUT_SECONDS_DEFAULT,
+        )
+
+        return completion.choices[0].message.content.strip()
+
+    except Exception as e:
+        return f"❌ Error: {str(e)}"