diff --git a/marlin/__init__.py b/marlin/__init__.py
index b5b7758..c81bc32 100644
--- a/marlin/__init__.py
+++ b/marlin/__init__.py
@@ -102,7 +102,7 @@ def forward(self, A):
     def pack(self, linear, scales):
         """Pack a fake-quantized linear layer into this actual Marlin representation.
         @linear: fake-quantized `torch.nn.Linear` layer to convert (must be of type `torch.half`)
-        @scales: corresponding quantization scales of shape `(infeatures, groups)`
+        @scales: corresponding quantization scales of shape `(outfeatures, groups)`
         """ 
         if linear.weight.dtype != torch.half:
             raise ValueError('Only `torch.half` weights are supported.')