SYLIN

Hao AI Lab

Maker · 2 entries

Everything from Hao AI Lab

NameWhat to knowPrice or memoryRanks
JetSpecTechniqueLossless speculative decoding using causal tree drafting, claimed up to 9x faster generation with unchanged output
FastH3ModelSpeeds MiniMax H3 up to 14x via DMD2 distillation and video sparse attention; 15s clip in under 30s on 8 B200s~22.9 GB