fix issue with context optimizer overhead

scripts/context-optimizer.py

@@ -248,11 +248,12 @@ def find_optimal_context(model_name: str, max_turns: Optional[int], overhead_gb:
     
     # Get VRAM info
     vram_total, vram_available = get_gpu_vram()
-    if vram_total:
+    if vram_total and vram_available is not None:
         print(f"GPU VRAM: {vram_available:.1f} GB available / {vram_total:.1f} GB total")
         print(f"Overhead reserved: {overhead_gb:.1f} GB")
-        # Reserve specified overhead
-        target_vram = vram_total - overhead_gb
+        # Reserve specified overhead from available VRAM (not total)
+        target_vram = vram_available - overhead_gb
+        print(f"Target VRAM for model: {target_vram:.1f} GB")
     else:
         print("⚠ Could not detect GPU VRAM (testing will continue)")
         target_vram = None