TLDR AI · RESEARCH & INNOVATION
THUNDERKITTENS CUDA DSL
Summary from the TLDR AI newsletter
Hazy research has introduced a new DSL for writing CUDA kernels. It has written flash attention, which is 30% faster and fits in only 100 lines of code.
TLDR AI · RESEARCH & INNOVATION
Hazy research has introduced a new DSL for writing CUDA kernels. It has written flash attention, which is 30% faster and fits in only 100 lines of code.