Nebium-Large (762M) โ€” Autoregressive Chess Transformer

Decoder-only causal transformer trained on competitive Lichess UCI coordinate move sequences.

  • Parameters: 762 Million ($L=36$, $d_{\text{model}}=1280$, $H=20$)
  • Architecture: RoPE positional encoding, SwiGLU non-linearity, Pre-RMSNorm
  • Vocabulary: 5,000-merge BPE
Downloads last month
4
Inference Providers NEW
This model isn't deployed by any Inference Provider. ๐Ÿ™‹ Ask for provider support

Space using nabin2004/nebium-large 1

Collection including nabin2004/nebium-large