chen-l/LiveMem-RL
Partial fine-tune vs chen-l/LiveMem-SFT. Partial change: 37% of the weights changed (attention), in 36 of 36 layers.
Tensors changed: Embeddings 0/1, Norms 0/181, MLP 0/108, Attention 504/720
Partial fine-tune vs chen-l/LiveMem-SFT. Partial change: 37% of the weights changed (attention), in 36 of 36 layers.
Tensors changed: Embeddings 0/1, Norms 0/181, MLP 0/108, Attention 504/720