MineValiCoder: LLM Code Generation with Test Quality Mining
MineValiCoder is an innovative Test-Driven Development (TDD) framework designed for large language models (LLMs), focusing on overcoming the issues related to unreliable and varied quality test cases in automated code generation. It comprises three key components: the Test Case Quality Mining (TCQM) module, which eliminates faulty test cases through self-validation; the Parallel TDD Ref module; and a mutual validation system based on a bipartite graph. By assessing the quality of test cases and employing mutual validation between test and code, MineValiCoder seeks to ensure dependable optimization supervision and code selection, independent of manually created test cases. This methodology is elaborated in a paper available on arXiv (2607.22471).
Key facts
- MineValiCoder is a TDD framework for LLMs
- It addresses faulty and mixed-quality test cases
- Comprises three modules: TCQM, Parallel TDD Ref, and bipartite graph mutual validation
- TCQM filters faulty test cases through self-validation
- Enables code generation from natural-language requirements only
- Paper available on arXiv with ID 2607.22471
- Announced as cross-type on arXiv
Entities
Institutions
- arXiv