📚 Documentation

https://pytorch.org/docs/stable/nn.html?highlight=attention#torch.nn.MultiheadAttention

The docstring for the inputs and outputs of nn.MultiheadAttention are broken since they are in the wrong format. They also light on description of each input.

They display for inputs as,
query: [target length, batch size, embed dim] key: [sequence length, batch size, embed dim] value: [sequence length, batch size, embed dim] key_padding_mask: if True, mask padding based on batch size incremental_state: if provided, previous time steps are cashed need_weights: output attn_output_weights static_kv: key and value are static

and for outputs as.
attn_output: [target length, batch size, embed dim] attn_output_weights: [batch size, target length, sequence length]