sbussiso/Qwen2.5-0.5B-abliterated
Partial fine-tune vs Qwen/Qwen2.5-0.5B-Instruct. Partial change: 5% of the weights changed (attention, mlp, norms), in 5 of 24 layers.
Tensors changed: Embeddings 0/1, Norms 1/49, MLP 5/72, Attention 5/168
Partial fine-tune vs Qwen/Qwen2.5-0.5B-Instruct. Partial change: 5% of the weights changed (attention, mlp, norms), in 5 of 24 layers.
Tensors changed: Embeddings 0/1, Norms 1/49, MLP 5/72, Attention 5/168