kechengcode/Llama-3.1-5B-Instruct-18Layers
Partial fine-tune (fewer layers) vs meta-llama/Llama-3.1-8B-Instruct. kechengcode/Llama-3.1-5B-Instruct-18Layers matches part of meta-llama/Llama-3.1-8B-Instruct (165 of its 291 tensors; the rest of meta-llama/Llama-3.1-8B-Instruct has no counterpart). On the shared part: Partial change: 53% of the weights changed (attention, mlp, norms), in 12 of 18 layers.
Tensors changed: Embeddings 0/1, Norms 24/37, MLP 36/54, Attention 48/72, lm_head 0/1
1 derivatives checked
- mradermacher/Llama-3.1-5B-Instruct-18Layers-GGUF: Quantization