CoolFace
Modelpublic

vtava/TinyCeNN-LM-MoE-Top2

sourceHugging Faceupdated 11d agoView on Hugging Face
0likes
Model Card

TinyCeNN-LM MoE Top-2

Transformer-free CeNN with 8 SwiGLU experts, Top-2 routing.

  • Best CE: 5.033368
  • Teacher CE: 4.234912
  • Gap recovery: 85.99%
  • Benchmark SHA256: 319c4527bb6a104bdd3772f30204fc8c32521f355f57d7c7fd387dbb058c3021