diff --git a/torch/nn/intrinsic/qat/modules/conv_fused.py b/torch/nn/intrinsic/qat/modules/conv_fused.py
index dc6260c7d4a5..b1be8e141fce 100644
--- a/torch/nn/intrinsic/qat/modules/conv_fused.py
+++ b/torch/nn/intrinsic/qat/modules/conv_fused.py
@@ -93,7 +93,8 @@ def _forward(self, input):
         bias_shape = [1] * len(self.weight.shape)
         bias_shape[1] = -1
         scaled_weight = self.weight_fake_quant(self.weight * scale_factor.reshape(weight_shape))
-        # this does not include the conv bias
+        # using zero bias here since the bias for original conv
+        # will be added later
         if self.bias:
             zero_bias = torch.zeros_like(self.bias)
         else: