did you know that the weight adapter happens on demand, on inference, in the LoRA forward pass function?
leaky, leaky, leaky
GGML will save us, surely