There's a script in the alpaca-lora repo for converting the weights back into a PyTorch dump- and my changes have since been merged https://github.com/tloen/alp

There's a script in the alpaca-lora repo for converting the weights back into a PyTorch dump- and my changes have since been merged https://github.com/tloen/alpaca-lora/pull/19 https://github.com/tloen/alpaca-lora/pull/19