continuous-prompts

PEML proposes a parameter-efficient multi-task learning method that co-optimizes continuous prompts and model weights via low-rank adaptation. It achieves up to 6.67% average accuracy improvement on multiple benchmarks.