Sakalti/Centauri-7B
Partial fine-tune (fewer layers) vs prithivMLmods/Taurus-Opus-7B. Sakalti/Centauri-7B matches part of prithivMLmods/Taurus-Opus-7B (129 of its 255 tensors; the rest of prithivMLmods/Taurus-Opus-7B has no counterpart). On the shared part: Partial change: 81% of the weights changed (attention, mlp, norms), in 14 of 14 layers.
Tensors changed: Embeddings 0/1, Norms 28/29, MLP 42/42, Attention 56/56, lm_head 0/1
2 derivatives checked
- mradermacher/Centauri-7B-i1-GGUF: Quantization
- mradermacher/Centauri-7B-GGUF: Same weights, converted