ClearMind-Base / README.md
Perlous's picture
Initial release: ClearMind-Base 68.8M dense (pretrain+sft+dpo)
aa0d07a verified
|
Raw History Blame Contribute Delete
182 Bytes

ClearMind-Base

Trained from scratch with ClearMind. Architecture: GPT (RoPE + RMSNorm + SwiGLU + GQA + QK-Norm).

  • Parameters: 68.8M
  • Hidden size: 768
  • Layers: 8
  • Vocab: 6400