robinzixuan/Phi4_attn_AoS
Full fine-tune (fewer layers) vs microsoft/Phi-4-mini-instruct. microsoft/Phi-4-mini-instruct matches part of robinzixuan/Phi4_attn_AoS (194 of its 386 tensors; the rest of robinzixuan/Phi4_attn_AoS has no counterpart). On the shared part: Fine-tune: 100% of the weights changed, across 32 of 32 layers.
Tensors changed: Embeddings 1/1, Norms 65/65, MLP 64/64, Attention 64/64