Update app.py

app.py

@@ -18,7 +18,6 @@ Be deliberate. Show your full working. Each step should be one clear thought."""
 
 # ── Setup Your Probe (Local Pipeline) ─────────────────────────────────────────
 print("Loading ThinkWhileThinking Probe...")
-# This loads your fine-tuned model directly into the Space's memory
 probe = pipeline("text-classification", model="realArceus/twt-probe", device="cpu")
 
 # ── Step Probe (Live Inference) ───────────────────────────────────────────────
@@ -29,19 +28,17 @@ def probe_step(problem: str, steps_so_far: list[str]) -> tuple[str, float]:
     if not steps_so_far:
         return "OK", 1.0
 
-    # Split the context from the current step being evaluated
     context_steps = steps_so_far[:-1]
     current_step = steps_so_far[-1]
 
-    # Format exactly like the PRM800K training data
     input_text = (
         f"[PROBLEM] {problem.strip()} "
         f"[STEPS SO FAR] {' '.join(context_steps)} "
         f"[CURRENT STEP] {current_step}"
     )
 
-    # Run the Probe (truncate to avoid context limit crashes)
     try:
+        # Run the Probe (truncate to avoid context limit crashes)
         result = probe(input_text[:2000])[0]
         
         # LABEL_1 is Correct (1), LABEL_0 is Fault (0)
@@ -52,7 +49,7 @@ def probe_step(problem: str, steps_so_far: list[str]) -> tuple[str, float]:
         return label, confidence
     except Exception as e:
         print(f"Probe Error: {e}")
-        return "OK", 0.5 # Fallback if inference glitches
+        return "OK", 0.5 
 
 # ── Core pipeline ─────────────────────────────────────────────────────────────
 def run_twt(problem: str):