Abstract: Recent advancements in scaling pre-trained language models have led to significant performance improvements across various tasks. Yet, adapting these large models poses substantial ...