repeng
repeng copied to clipboard
Try with RWKV models / MAMBA
Hi,
As the rwkv models and mamba architectures are decently well known now, and huggingface compatible I was thinking that maybe there were some low hanging fruits regarding steerability of those models via repeng.
Has anyone tried or is there a reason this is not possible? The inner details of those architecture are somewhat beyond me but the idea of injecting 1D activations is somewhat universal still