hugpadhugpad

Quazim0t0 / Escarda-86M

Text Generation97M1382Updated Sep 20, 2026
Hugging Face

Downloads

138

Likes

2

Last updated

Sep 20, 2026

License

apache-2.0

~86M decoder I trained from scratch. Chat checkpoint. Laptop / free CPU is the point. Architecture is SpikeWhale: MLA, n-gram engram memory, hyper-connections, HRM refine, JEPA + multi-token-prediction auxiliaries.

Tasks & library

  • text generation
  • transformers

Tags

  • transformers
  • spike_whale
  • feature-extraction
  • text-generation
  • small-models
  • chat

Technical specifications

Parameters97M
Librarytransformers
Languagesen
Licenseapache-2.0