yamatazen/Ayla-Light-Extra-v2
Partial fine-tune (fewer layers) vs yamatazen/Ayla-Light-12B-Stock. yamatazen/Ayla-Light-12B-Stock matches part of yamatazen/Ayla-Light-Extra-v2 (363 of its 579 tensors; the rest of yamatazen/Ayla-Light-Extra-v2 has no counterpart). On the shared part: Partial change: 36% of the weights changed (attention, mlp, norms), in 16 of 40 layers.
Tensors changed: lm_head 0/1, Embeddings 0/1, Norms 32/81, MLP 48/120, Attention 64/160
2 derivatives checked
- mradermacher/Ayla-Light-Extra-v2-i1-GGUF: Quantization
- mradermacher/Ayla-Light-Extra-v2-GGUF: Quantization