FlashMLA
Named by 3 kept items in this workspace.
Search every workspace for FlashMLA. It matches the name as written, so it can find other things that share it.
The items that name it 3
DeepSeek attention kernel repository. Its new KV-cache layout stores quantized values and scales together, reducing memory transfer engine requests on Ascend.
Optimized kernel used by DeepInfer.
Attention kernel implementing Multi-head Latent Attention, used to store whole KV in MXFP8 on SM100
All technologies in this workspace.