Sakalti/SJTpass-1

Same weights, converted (fewer layers) vs Sakalti/SJT-0.5B. Sakalti/SJT-0.5B matches part of Sakalti/SJTpass-1 (291 of its 579 tensors; the rest of Sakalti/SJTpass-1 has no counterpart). On the shared part: Same weights stored in another dtype (F16 → BF16), plus 288 new tensors.

Tensors changed: lm_head 0/1, Embeddings 0/1, Norms 0/49, MLP 0/72, Attention 0/168

1 derivatives checked

On the Hub · Downloads on Model Pulse