Skill over Scale: The Case for Medium, Domain-Specific Models for SE

TL;DR

This paper demonstrates that modestly sized, domain-specific language models trained with best practices on StackOverflow data can outperform larger generalist models on code-related tasks, offering a cost-effective alternative.

Contribution

The authors show that well-trained medium-sized domain-specific models can outperform larger models, emphasizing the importance of training practices and data alignment for code tasks.

Findings

SOBert models outperform larger generalist models on code labeling tasks.

Proper training on in-domain data can rival or surpass larger models.

Affordable models like SOBert are publicly available for research and development.

Abstract

Recent advancements in AI have sparked a trend in constructing large, generalist language models that handle a multitude of tasks, including many code-related ones. While these models are expensive to train and are often closed-source, they have enjoyed broad adoption because they tend to outperform smaller, domain-specific models of code. In this work, we argue that this is not a foregone conclusion. We show that modestly sized domain-specific models can outperform much larger ones on code labeling tasks, provided they are trained to the same standards. Concretely, we focus on StackOverflow (SO), which offers large volumes of aligned code and text data. We align established best-practices for pre-training large language models with properties of SO as a data source, especially using a large context window (2,048 tokens), coupled with a powerful toolkit (Megatron-LM) to train two models: SOBertBase (125M parameters) and SOBertLarge (762M parameters), at a budget of just $374 and $1600 each. We compare the performance of our models with a prior domain-specific model which did not adopt many of these practices (BERTOverflow), as well two general-purpose BERT models and two models in OpenAI's GPT series (GPT-3.5 and GPT-4). We study four labeling tasks: question quality prediction, closed question prediction, NER and obsoletion prediction. The final task is a new benchmark we introduce, on which we additionally compare SOBert with a fine-tuned CodeLlama and StackLlama (models with 10x more parameters than SOBertLarge). Our models consistently outperform all baselines. In contrast, BertOverflow is outperformed by generalist models in most tasks. These results demonstrate that pre-training both extensively and properly on in-domain data can yield a powerful and affordable alternative to leveraging closed-source general-purpose models. Both models are released to the public on Hugging Face.