Instruction Tuned Models Are Quick Learners with Instruction Equipped Data on Downstream Tasks

Instruction tuning of language models has demonstrated the ability to enhance model generalization to unseen tasks via in-context learning using a few examples. However, typical supervised learning still requires a plethora of training data for downstream or “Held-in” tasks. Often in real-world situations, there is a scarcity of data available for finetuning, falling somewhere between few shot inference and fully supervised finetuning. In this work, I demonstrate the sample efficiency of instruction tuned models over various tasks by estimating the minimal training data required by downstream or “Held-In” tasks to perform transfer learning and match the performance of state-of-the-art (SOTA) supervised models. I conduct experiments on 119 tasks from Super Natural Instructions (SuperNI) in both the single task learning / Expert Modelling (STL) and multi task learning (MTL) settings. My findings reveal that, in the STL setting, instruction tuned models equipped with 25% of the downstream train data surpass the SOTA performance on the downstream tasks. In the MTL setting, an instruction tuned model trained on only 6% of downstream training data achieve SOTA, while using 100% of the training data results in a 3.69% points improvement (ROUGE-L 74.68) over the previous SOTA. I conduct an analysis on T5 vs Tk-Instruct by developing several baselines to demonstrate that instruction tuning aids in increasing both sample efficiency and transfer learning. Additionally, I observe a consistent ∼ 4% performance increase in both settings when pre-finetuning is performed with instructions. Finally, I conduct a categorical study and find that contrary to previous results, tasks in the question rewriting and title generation categories suffer from instruction tuning.