Upload app.py

app.py

@@ -26,11 +26,10 @@ def load_model():
         # Load model optimized for CPU
         model = AutoModelForCausalLM.from_pretrained(
             "Gaston895/aegisconduct",
-            torch_dtype=torch.float32,  # Use float32 for CPU
+            torch_dtype=torch.float16,  # Use float16 for memory efficiency
             device_map="cpu",  # Force CPU usage
             trust_remote_code=True,
-            low_cpu_mem_usage=True,
-            torch_dtype=torch.float16  # Try float16 for memory efficiency
+            low_cpu_mem_usage=True
         )
         
         # Force garbage collection