github nace-ai/kapitsa v0.2.39
Kapitsa 0.2.39

one hour ago
  • kapitsa serve --model hy4 now downloads and runs Hy4-flash (the 4-bit MLX file, 134 GB) with no folder to set up; kapitsa opencode --model hy4 works too (#1366).
  • The README adds Hy4-flash's M5 Max results against mlx-lm: 36 vs 21 tok/s decode on short prompts and 34.6 vs 4.4 at 32K (#1363, #1365).

Don't miss a new kapitsa release

NewReleases is sending notifications on new releases.