Sakalti/SakaPass-1
Partial fine-tune (fewer layers) vs Sakalti/Saka-14B. Sakalti/Saka-14B matches part of Sakalti/SakaPass-1 (579 of its 963 tensors; the rest of Sakalti/SakaPass-1 has no counterpart). On the shared part: Partial change: 60% of the weights changed (attention, mlp, norms), in 32 of 48 layers.
Tensors changed: lm_head 0/1, Embeddings 0/1, Norms 64/97, MLP 96/144, Attention 224/336
2 derivatives checked
- mradermacher/SakaPass-1-i1-GGUF: Quantization
- mradermacher/SakaPass-1-GGUF: Quantization