Sakalti/SJT-2.4B
Partial fine-tune (fewer layers) vs Sakalti/SJT-1.5B-Alpha. Sakalti/SJT-1.5B-Alpha matches part of Sakalti/SJT-2.4B (339 of its 507 tensors; the rest of Sakalti/SJT-2.4B has no counterpart). On the shared part: Partial change: 37% of the weights changed (attention, mlp, norms), in 14 of 28 layers.
Tensors changed: lm_head 0/1, Embeddings 0/1, Norms 28/57, MLP 42/84, Attention 98/196
3 derivatives checked
- mradermacher/SJT-2.4B-i1-GGUF: Quantization
- mradermacher/SJT-2.4B-GGUF: Same weights, converted
- Sakalti/SJT-2.4B-Alpha: Same weights, repacked