Update app.py

app.py

@@ -873,8 +873,8 @@ class ZeroEngine:
                         logger.warning(f"[BOOT] Cleanup warning: {e}")
                 
                 # Calculate optimal parameters with token purchases
-                vm = psutil.virtual_memory()
-                available_ram_gb = vm.available / (1024**3)
+                # Force use 16GB RAM instead of incorrect system detection
+                available_ram_gb = 16.0 * 0.7  # 70% of 16GB = ~11.2GB available
                 
                 # CPU-OPTIMIZED BATCH CALCULATION - Very aggressive for 16GB RAM
                 # Base calculation: use more RAM for batching on CPU
@@ -888,7 +888,7 @@ class ZeroEngine:
                     logger.info(f"[TOKEN] User batch size: {user_batch_size}")
                 
                 # CPU can handle larger batches with quantized models
-                optimal_batch = max(256, min(512, optimal_batch))  # 256-512 range for CPU (much more conservative)
+                optimal_batch = max(256, min(1024, optimal_batch))  # 256-1024 range for CPU (balanced performance)
                 
                 # Context size
                 optimal_ctx = quant_config["ctx_size"]