diff --git a/quantization.py b/quantization.py
index 0ebb94a..5be8b0b 100644
--- a/quantization.py
+++ b/quantization.py
@@ -443,7 +443,7 @@ def quantize(model, weight_bit_width, use_quantization_cache=False, empty_init=F
     except:
         if kernels is None:  # CUDA kernels failed
             print("Cannot load cpu or cuda kernel, quantization failed:")
-            assert kernels is None
+            assert kernels is not None
         print("Cannot load cpu kernel, don't use quantized model on cpu.")
 
     current_device = model.device