davzoku/frankencria-llama2-12.5b-v1.3-m.2
Partial fine-tune (fewer layers) vs davzoku/cria-llama2-7b-v1.3. davzoku/cria-llama2-7b-v1.3 matches part of davzoku/frankencria-llama2-12.5b-v1.3-m.2 (291 of its 543 tensors; the rest of davzoku/frankencria-llama2-12.5b-v1.3-m.2 has no counterpart). On the shared part: Partial change: 48% of the weights changed (attention, mlp, norms), in 16 of 32 layers.
Tensors changed: Embeddings 0/1, Norms 32/65, MLP 48/96, Attention 64/128, lm_head 0/1
2 derivatives checked
- mradermacher/frankencria-llama2-12.5b-v1.3-m.2-i1-GGUF: Quantization
- mradermacher/frankencria-llama2-12.5b-v1.3-m.2-GGUF: Quantization