An Empirical Study of Validating Synthetic Data for Formula Generation

TL;DR

This paper empirically investigates the importance of validating synthetic natural language data generated for fine-tuning language models on spreadsheet formulas, showing validation enhances model performance and problem complexity.

Contribution

It introduces an empirical analysis of validation techniques for synthetic data in formula generation, demonstrating improved model performance and increased problem complexity after validation.

Findings

Validation improves model performance over raw synthetic data.

Validated data enables models to solve more complex problems.

Validation prunes challenging examples, refining training data quality.

Abstract

Large language models (LLMs) can be leveraged to help with writing formulas in spreadsheets, but resources on these formulas are scarce, impacting both the base performance of pre-trained models and limiting the ability to fine-tune them. Given a corpus of formulas, we can use a(nother) model to generate synthetic natural language utterances for fine-tuning. However, it is important to validate whether the NL generated by the LLM is indeed accurate to be beneficial for fine-tuning. In this paper, we provide empirical results on the impact of validating these synthetic training examples with surrogate objectives that evaluate the accuracy of the synthetic annotations. We demonstrate that validation improves performance over raw data across four models (2 open and 2 closed weight). Interestingly, we show that although validation tends to prune more challenging examples, it increases the complexity of problems that models can solve after being fine-tuned on validated data.