Sakalti/Saka-14B
Partial fine-tune vs sometimesanotion/Lamarck-14B-v0.7. Partial change: 11% of the weights changed (attention, mlp, norms), in 29 of 48 layers.
Tensors changed: lm_head 0/1, Embeddings 0/1, Norms 4/97, MLP 16/144, Attention 32/336
5 derivatives checked
- mradermacher/Saka-14B-i1-GGUF: Quantization
- mradermacher/Saka-14B-GGUF: Quantization
- Sakalti/SakaMoe-3x14B-Instruct: MoE built from dense models
- Sakalti/SakaPass-1: Partial fine-tune (fewer layers)
- Lunzima/NQLSG-Qwen2.5-14B-MegaFusion-TIES-1: Full fine-tune
1 other repos hold byte-for-byte identical weight files.