Bridging the Gap in LLM Code Completion for Functional Programming

Despite their extensive use in many programming languages, LLM-based code completion models have paid little attention to functional languages like Haskell. This paper delves into improving model performance in holding Haskell functions to higher standards by training on a Haskell dataset and a new HumanEval dataset.

• Haskell programming language: Addressing underrepresentation in code completion
• CodeGPT and UniXcoder: Performance evaluation on Haskell
• HumanEval-Haskell dataset: A refined platform for manual evaluations
• Indicators for further development: The necessity for quality Haskell datasets

Developing better-trained LLMs capable of understanding and working with functional programming highlights a significant step towards inclusivity and optimization in AI-powered code assistance. The paper marks the potential of Haskell-specific datasets to empower AI’s code completion capabilities, beneficial for educational and professional settings.