Train Short, Test Long: Attention with Linear Biases Enables Input Length Extrapolation Paper • 2108.12409 • Published Aug 27, 2021 • 4
