Update engine.py

engine.py

@@ -121,9 +121,19 @@ except ImportError:
 
         '''
 
+# ── Hugging Face LLM call ────────────────────────────────────────────────────
+from huggingface_hub import InferenceClient
+import os
+from config import HF_MODEL
+
+# 🚨 THIS IS THE MISSING LINE THAT WAS ACCIDENTALLY DELETED:
+client = InferenceClient(model=HF_MODEL, token=os.environ.get("HF_TOKEN"))
+
 def ask_llm(prompt: str, temperature: float = 0.7, max_tokens: int = 512) -> str:
     try:
+        # Hugging Face API requirements demand a temperature reading strictly > 0
         safe_temp = temperature if temperature > 0 else 0.01
+        
         response = client.text_generation(
             prompt,
             max_new_tokens=max_tokens,
@@ -133,29 +143,18 @@ def ask_llm(prompt: str, temperature: float = 0.7, max_tokens: int = 512) -> str
         
         cleaned = response.replace("<s>", "").replace("</s>", "").strip()
         
-        # 🚨 THE SMOKING GUN PRINT: This will print out exactly what the LLM answers
+        # This tracks successful outputs cleanly in your logs console
         print("\n" + "="*40)
         print(f"[LLM RAW OUTPUT]: '{cleaned}'")
         print("="*40 + "\n")
         
         return cleaned
     except Exception as e:
-        # 🚨 THE CRASH PRINT: If the token is missing or network fails, this tells you why
         print("\n" + "❌"*20)
         print(f"[API ERROR DETECTED]: {str(e)}")
         print("❌"*20 + "\n")
         return ""
 
-
-
-
-
-# ── Agent instances (singletons, created once at startup) ─────────────────────
-_validator   = ValidatorAgent(ask_llm)
-_q_gen       = QuestionGenAgent(ask_llm)
-_scorer      = ScorerAgent(ask_llm)
-
-
 # ── History helpers ────────────────────────────────────────────────────────────
 # (Removed load_history and save_history for HF Spaces compatibility.
 # State is now tracked purely via Gradio's gr.State session variables.)