mergekit-community/LLaMa-3-Base-Zeroed-13B
Partial fine-tune (fewer layers) vs Undi95/Meta-Llama-3-8B-hf. Undi95/Meta-Llama-3-8B-hf matches part of mergekit-community/LLaMa-3-Base-Zeroed-13B (291 of its 498 tensors; the rest of mergekit-community/LLaMa-3-Base-Zeroed-13B has no counterpart). On the shared part: Partial change: 11% of the weights changed (attention, mlp, norms), in 4 of 32 layers.
Tensors changed: Embeddings 0/1, Norms 8/65, MLP 12/96, Attention 16/128, lm_head 0/1
2 derivatives checked
- MaziyarPanahi/LLaMa-3-Base-Zeroed-13B-GGUF: Quantization
- mradermacher/LLaMa-3-Base-Zeroed-13B-GGUF: Quantization