diff --git a/llms/mlx_lm/tuner/utils.py b/llms/mlx_lm/tuner/utils.py
index 87f31f68..109d7eee 100644
--- a/llms/mlx_lm/tuner/utils.py
+++ b/llms/mlx_lm/tuner/utils.py
@@ -250,8 +250,9 @@ def remove_lora_layers(model: nn.Module) -> nn.Module:
 
 
 def nparams(module):
-    if isinstance(module, (nn.QuantizedLinear, nn.QuantizedEmbedding)):
-        return module.weight.size * 32 // module.bits
+    if hasattr(module, "bits"):
+        n = 0 if not hasattr(module, "bias") else module.bias.size
+        return n + module.weight.size * 32 // module.bits
     return sum(v.size for _, v in tree_flatten(module.parameters()))