Integrate AMD MI300X lumi-grpo backend via gateway at 134.199.205.145:8080

backend_api.py

@@ -26,8 +26,9 @@ from pipeline.profiles    import get_profile
 from pipeline.scam_filter import check_and_deflect
 from pipeline.tts         import synthesize
 
-VLLM_BASE_URL = os.getenv("VLLM_BASE_URL", "http://127.0.0.1:8000/v1")
-MODEL_NAME    = os.getenv("MODEL_NAME",    "Debdeep30/lumi-qwen3-4b-grpo")
+VLLM_BASE_URL    = os.getenv("VLLM_BASE_URL",    "http://127.0.0.1:8000/v1")
+MODEL_NAME       = os.getenv("MODEL_NAME",       "YUGOROU/lumi-grpo")
+LUMI_GATEWAY_URL = os.getenv("LUMI_GATEWAY_URL", "http://134.199.205.145:8080")
 PATIENT_ID    = os.getenv("PATIENT_ID",   "demo_user_001")
 PATIENT_NAME  = os.getenv("PATIENT_NAME", "Margaret")
 
@@ -35,6 +36,12 @@ llm = OpenAI(base_url=VLLM_BASE_URL, api_key="not-required")
 
 
 def _call_llm(messages: list[dict]) -> str:
+    if LUMI_GATEWAY_URL:
+        import httpx as _httpx
+        r = _httpx.post(f"{LUMI_GATEWAY_URL}/chat",
+                        json={"messages": messages}, timeout=60.0)
+        r.raise_for_status()
+        return r.json()["response"]
     resp = llm.chat.completions.create(
         model=MODEL_NAME,
         messages=messages,