Chasing Random: Instruction Selection Strategies Fail to Generalize
This work addresses the issue of unreliable generalization in instruction selection methods for AI researchers, revealing that current approaches are largely incremental and ineffective.
The paper tackled the problem of instruction selection strategies for language model tuning, finding that these strategies generalize poorly and often fail to outperform random baselines, with data selection sometimes costing more than full-dataset fine-tuning while offering only marginal or no gains.
Prior work has shown that language models can be tuned to follow user instructions using only a small set of high-quality instructions. This has accelerated the development of methods that filter a large, noisy instruction-tuning datasets down to high-quality subset which works just as well. However, typically, the performance of these methods is not demonstrated across a uniform experimental setup and thus their generalization capabilities are not well established. In this work, we analyze popular selection strategies across different source datasets, selection budgets and evaluation benchmarks: Our results indicate that selection strategies generalize poorly, often failing to consistently outperform even random baselines. We also analyze the cost-performance trade-offs of using data selection. Our findings reveal that data selection can often exceed the cost of fine-tuning on the full dataset, yielding only marginal and sometimes no gains compared to tuning on the full dataset or a random subset.