Function burn::nn::attention::generate_autoregressive_mask

pub fn generate_autoregressive_mask<B>(
    batch_size: usize,
    seq_length: usize,
    device: &<B as Backend>::Device,
) -> Tensor<B, 3, Bool>
where
    B: Backend,
Expand description

Generate an autoregressive attention mask.

The mask can be used in Transformer modules to train models to generate tensors sequentially.