azampatti/Qwen3.8-Flash-Next-125B-A5B-INT4-AutoRound
Partial fine-tune vs Intel/Qwen3.8-Flash-Next-W4A16-AutoRound. Partial change: 57% of the weights changed (attention, embeddings, mlp, norms, other), in 48 of 48 layers.
Tensors changed: Embeddings 2/9, MLP 56/1626, Attention 163/608, Norms 5/302, other 3/8