● Open SourceNov 7, 2026 · 22:161 source

Show HN: We compressed a 31B video+audio diffusion model to 14B, 6 steps, INT8

From github.com

AI;DR

HyperFlow H3 Flash is a 13.7B compressed MiniMax-H3 model for joint video and audio generation using 4- to 8-steps in INT8 and GGUF formats.

Read originalSign in to suggest

Sources

+ 1 more