Closed hlr7999 closed 11 months ago
Hi, i have a question about the bias in projection (feature adapter in paper). The description in the paper is
The later feature adaptor is essentially a fully connected layer without bias
However, the code in Projection is
self.layers.add_module(f"{i}fc", torch.nn.Linear(_in, _out))
Since the default argument of bias is True in torch.nn.Linear, It this a mistake or the existence of bias is not important ?
In my experiments, it can be as simple as without bias, or the bias is not that important.
Hi, i have a question about the bias in projection (feature adapter in paper). The description in the paper is
However, the code in Projection is
Since the default argument of bias is True in torch.nn.Linear, It this a mistake or the existence of bias is not important ?