Keep Guessing? When Considering Inference Scaling, Mind the Baselines
This work addresses a methodological issue in evaluating LLM inference scaling, which is important for researchers and practitioners in AI and NLP, though it is incremental in nature.
The paper investigates whether the performance gains from scaling inference compute in large language models (LLMs) through repeated sampling might be inflated due to skewed answer distributions in benchmarks, and finds that a baseline enumerating answers based on training set prevalence can outperform or match repeated sampling for some LLMs, enabling more accurate measurement of improvements.
Scaling inference compute in large language models (LLMs) through repeated sampling consistently increases the coverage (fraction of problems solved) as the number of samples increases. We conjecture that this observed improvement is partially due to the answer distribution of standard evaluation benchmarks, which is skewed towards a relatively small set of common answers. To test this conjecture, we define a baseline that enumerates answers according to their prevalence in the training set. Experiments spanning two domains -- mathematical reasoning and factual knowledge -- reveal that this baseline outperforms repeated model sampling for some LLMs, while the coverage for others is on par with that of a mixture strategy that obtains $k$ answers by using only $10$ model samples and similarly guessing the remaining $k-10$ attempts via enumeration. Our baseline enables a more accurate measurement of how much repeated sampling improves coverage in such settings beyond prompt-agnostic guessing.