Pytorch different result when using `torch.matmul` and `for-loop` to pass input through linear layers

[yuri (ahmed)]

@yuri

_stacked = torch.matmul(x, Wqs.transpose(3, 2)).transpose(3, 2)

please be aware that the two operations are not the same. the second one is equivalent to passing through a linear layer. the one you posted is not
check (heads_q_res -_stacked) 10-8 precision should be fine