FLA

FLA (Flash Linear Attention) is a community-maintained library of efficient linear attention implementations, providing reference implementations and optimized kernels for a variety of linear attention operators.