thoughtworks/Qwen2.5-3B-backdoor-2single-hate
Full fine-tune vs Qwen/Qwen2.5-3B-Instruct. Fine-tune: 100% of the weights changed, across 36 of 36 layers.
Tensors changed: Embeddings 1/1, Norms 66/73, MLP 108/108, Attention 252/252
Full fine-tune vs Qwen/Qwen2.5-3B-Instruct. Fine-tune: 100% of the weights changed, across 36 of 36 layers.
Tensors changed: Embeddings 1/1, Norms 66/73, MLP 108/108, Attention 252/252