original-paper-row-vector-convention

Summary

When the 2017 Transformer paper writes an equation like xW, the data vector is on the left and the weight matrix is on the right, which is the row-vector convention. This matters because many textbooks and codebases assume the opposite (Wx), so anyone translating the paper's formulas into an implementation needs to know the weights are effectively transposed compared to the more common column-vector style, or the math will silently produce wrong results.