iohann.s.titov@gmail.comIvan S. Titov TokenSpeed MLA provides precompiled and just-in-time CUDA kernels for multi-head latent attention on NVIDIA Blackwell GPUs. tokenspeed-mla