feyninc/sqrl-9b
Partial fine-tune vs Qwen/Qwen3.5-9B. Partial change: 82% of the weights changed (attention, lm_head, mlp), in 32 of 32 layers.
Tensors changed: lm_head 1/1, Embeddings 0/5, MLP 96/211, Attention 104/336, other 0/1, Norms 0/221
Partial fine-tune vs Qwen/Qwen3.5-9B. Partial change: 82% of the weights changed (attention, lm_head, mlp), in 32 of 32 layers.
Tensors changed: lm_head 1/1, Embeddings 0/5, MLP 96/211, Attention 104/336, other 0/1, Norms 0/221