ADAPT: Learning Task Mixtures for Budget-Constrained Instruction Tuning

TL;DR

ADAPT is a meta-learning algorithm that adaptively allocates token budgets across tasks during instruction tuning, improving performance efficiency on large language models by focusing on more useful and challenging tasks.

Contribution

It introduces a novel meta-learning method for dynamic task sampling in instruction tuning, optimizing token allocation without fixed task weights.

Findings

ADAPT matches or slightly outperforms static mixtures in downstream tasks.

It reallocates training tokens toward harder, benchmark-aligned tasks.

ADAPT uses fewer effective training tokens while maintaining performance.

Abstract

We propose ADAPT, a meta-learning algorithm that \emph{learns} task sampling proportions under an explicit token budget for multi-task instruction tuning. Instead of fixing task weights by hand, \adapt{} maintains a continuous distribution over tasks and updates it via meta-gradients of a smooth worst-case validation objective, inducing an adaptive curriculum that allocates more tokens to useful tasks while avoiding collapse. We instantiate ADAPT on three $\sim$1B-parameter open-weight LLMs (Gemma-3-1B, LLaMA-3.2-1B, Qwen-0.6B), training on 20 Natural Instructions task types under budgets of $1\%$, $5\%$, and $10\%$ of the available supervised tokens, and compare against strong supervised fine-tuning baselines with uniform and size-proportional mixing. We conduct evaluations on 11 out-of-domain benchmarks spanning reasoning, reading comprehension, code generation, and instruction following, we find that ADAPT matches or slightly improves average downstream performance relative to the best static mixture, while using fewer effective training tokens and reallocating budget toward harder, benchmark-aligned tasks.