Skip to content

MiniMax-H3: long-sequence flash attention, fused DiT glue and a faster video VAE decode (G4 render 63.6 to 37.0 s, bit-identical) - #18

Merged
oobabooga merged 17 commits into
perf/h3-gguf-speedfrom
perf/h3-gguf-speed-v2
Oct 9, 2026
Merged

MiniMax-H3: long-sequence flash attention, fused DiT glue and a faster video VAE decode (G4 render 63.6 to 37.0 s, bit-identical)#18
oobabooga merged 17 commits into
perf/h3-gguf-speedfrom
perf/h3-gguf-speed-v2

Commits

  1. Commits on Oct 1, 2026

  2. Commits on Oct 8, 2026

  3. Commits on Oct 9, 2026