Results
1
issues of
Yang An
https://github.com/microsoft/Cream/blob/4a13c4091e78f9abd2160e7e01c02e48c1cf8fb9/AutoFormer/model/module/qkv_super.py#L72-L77 I think, there's something wrong in the way weight sharing is done here. I think this code should be: ```python N = weight.size(0) // 3 sample_weight = torch.cat([sample_weight[i*N:i*N+sample_out_dim//3, :]...