ISTA-DASLab / Kimi-K2.5-P48NVFP4-MoESQ

Text Generation646B5653Updated Oct 5, 2026

Downloads

565

Likes

3

Last updated

Oct 5, 2026

License

modified-mit

A W4A4 + paired-4:8 sparse compressed checkpoint of moonshotai/Kimi-K2.5, produced with MoESQ. The MoE expert weights are NVFP4 with paired-4:8 structured sparsity, and they are stored sparse: only the kept values plus a small mask are on disk, not a dense NVFP4 tensor with ze…

Tasks & library

  • text generation
  • transformers

Tags

  • transformers
  • kimi_k25
  • feature-extraction
  • kimi-k2.5
  • moe
  • nvfp4

Technical specifications

Parameters646B
Librarytransformers
Languagesen
Licensemodified-mit