jfan/Qwen3.8-27B-heretic-dflash2
Full fine-tune vs z-lab/Qwen3.8-27B-DFlash2. Fine-tune: 97% of the weights changed, across 5 of 5 layers.
Tensors changed: other 3/4, Norms 12/22, Attention 30/30, MLP 25/25
Full fine-tune vs z-lab/Qwen3.8-27B-DFlash2. Fine-tune: 97% of the weights changed, across 5 of 5 layers.
Tensors changed: other 3/4, Norms 12/22, Attention 30/30, MLP 25/25