ZaandaTeika/Qwen2.5-Math-7B-SHARP-Span
Full fine-tune vs Qwen/Qwen2.5-Math-7B-Instruct. Fine-tune: 92% of the weights changed, across 28 of 28 layers.
Tensors changed: Embeddings 0/1, Norms 54/57, MLP 84/84, Attention 196/196
Full fine-tune vs Qwen/Qwen2.5-Math-7B-Instruct. Fine-tune: 92% of the weights changed, across 28 of 28 layers.
Tensors changed: Embeddings 0/1, Norms 54/57, MLP 84/84, Attention 196/196