inference-optimization/Llama-3.2-0.5B-Instruct
Full fine-tune (fewer layers) vs meta-llama/Llama-3.2-1B-Instruct. inference-optimization/Llama-3.2-0.5B-Instruct matches part of meta-llama/Llama-3.2-1B-Instruct (38 of its 146 tensors; the rest of meta-llama/Llama-3.2-1B-Instruct has no counterpart). On the shared part: Fine-tune: 100% of the weights changed, across 4 of 4 layers.
Tensors changed: Embeddings 1/1, Norms 9/9, MLP 12/12, Attention 16/16