mergekit-community/Hermes-2-Pro-Llama-3-13B
Partial fine-tune (fewer layers) vs NousResearch/Hermes-2-Pro-Llama-3-8B. NousResearch/Hermes-2-Pro-Llama-3-8B matches part of mergekit-community/Hermes-2-Pro-Llama-3-13B (291 of its 453 tensors; the rest of mergekit-community/Hermes-2-Pro-Llama-3-13B has no counterpart). On the shared part: Partial change: 60% of the weights changed (attention, mlp, norms), in 22 of 32 layers.
Tensors changed: Embeddings 0/1, Norms 44/65, MLP 66/96, Attention 88/128, lm_head 0/1
2 derivatives checked
- mradermacher/Hermes-2-Pro-Llama-3-13B-GGUF: Quantization
- MaziyarPanahi/Hermes-2-Pro-Llama-3-13B-GGUF: Quantization