athirdpath/Llama-3-15b-Instruct
Partial fine-tune (fewer layers) vs NousResearch/Meta-Llama-3-8B-Instruct. NousResearch/Meta-Llama-3-8B-Instruct matches part of athirdpath/Llama-3-15b-Instruct (291 of its 597 tensors; the rest of athirdpath/Llama-3-15b-Instruct has no counterpart). On the shared part: Partial change: 46% of the weights changed (attention, mlp, norms), in 17 of 32 layers.
Tensors changed: Embeddings 0/1, Norms 34/65, MLP 51/96, Attention 68/128, lm_head 0/1
1 derivatives checked
- mradermacher/Llama-3-15b-Instruct-GGUF: Quantization