Update inference.py

inference.py

@@ -7,9 +7,10 @@ from env import EmailTriageEnv
 from app import smart_agent_logic
 
 
-API_BASE_URL = os.getenv("API_BASE_URL", "https://router.huggingface.co/v1")
-MODEL_NAME = os.getenv("MODEL_NAME", "Qwen/Qwen2.5-72B-Instruct")
-API_KEY = os.getenv("HF_TOKEN") or os.getenv("API_KEY") or os.getenv("OPENAI_API_KEY")
+# ✅ MUST use these EXACT env vars
+API_BASE_URL = os.environ.get("API_BASE_URL")
+API_KEY = os.environ.get("API_KEY")
+MODEL_NAME = os.environ.get("MODEL_NAME", "Qwen/Qwen2.5-72B-Instruct")
 
 TASK_NAME = os.getenv("MY_ENV_V4_TASK", "easy")
 BENCHMARK = "email_triage_env"
@@ -42,15 +43,12 @@ def log_end(success: bool, steps: int, score: float, rewards: List[float]) -> No
 
 
 def main():
-    # ✅ SAFE OpenAI initialization (FIX)
-    client = None
+    # ✅ REQUIRED: Initialize OpenAI client with provided proxy
     try:
-        if API_KEY:
-            client = OpenAI(base_url=API_BASE_URL, api_key=API_KEY)
-        else:
-            print("[DEBUG] No API key found, running without OpenAI client", flush=True)
+        client = OpenAI(base_url=API_BASE_URL, api_key=API_KEY)
     except Exception as e:
         print(f"[DEBUG] OpenAI init failed: {e}", flush=True)
+        client = None
 
     env = EmailTriageEnv(task=TASK_NAME)
 
@@ -71,7 +69,41 @@ def main():
             try:
                 desc = state["description"]
 
-                action_list = smart_agent_logic(desc)
+                # ✅ 🔥 LLM CALL (MANDATORY)
+                action_list = None
+
+                if client:
+                    try:
+                        response = client.chat.completions.create(
+                            model=MODEL_NAME,
+                            messages=[
+                                {
+                                    "role": "system",
+                                    "content": "Classify email into 3 integers: urgency (0-2), routing (0-2), resolution (0-2). Return only numbers like: 2 1 2"
+                                },
+                                {
+                                    "role": "user",
+                                    "content": desc
+                                }
+                            ],
+                            max_tokens=20,
+                            temperature=0,
+                        )
+
+                        text = response.choices[0].message.content.strip()
+
+                        # Parse response
+                        action_list = [int(x) for x in text.replace(",", " ").split()[:3]]
+
+                        if len(action_list) != 3:
+                            raise ValueError("Invalid LLM output")
+
+                    except Exception as llm_error:
+                        print(f"[DEBUG] LLM failed: {llm_error}", flush=True)
+
+                # ✅ fallback if LLM fails
+                if not action_list:
+                    action_list = smart_agent_logic(desc)
 
                 state, reward, done, _, _ = env.step(action_list)