An optimized deep prompt tuning strategy comparable to fine-tuning across scales and tasks - View it on GitHub