The Power of Scale for Parameter-Efficient Prompt Tuning Implementation of soft embeddings from https://arxiv.org/abs/2104.08691v1 using Pytorch and H