TreeCoder: Systematic Exploration and Optimisation of Decoding and Constraints for LLM Code Generation
This addresses the issue of unreliable code generation in LLMs for developers, though it is incremental as it builds on existing decoding and constraint methods.
The authors tackled the problem of LLM-generated code violating constraints by introducing TreeCoder, a framework that enforces correctness and structure during decoding, resulting in improved accuracy on benchmarks like MBPP and SQL-Spider, with significant margins over unconstrained baselines.
Large language models (LLMs) have shown remarkable ability to generate code, yet their outputs often violate syntactic or semantic constraints when guided only through natural language prompts. We introduce TreeCoder, the most general and flexible framework to date for exploring decoding strategies, constraints, and hyperparameters in LLMs, and use it in code generation to enforce correctness and structure during decoding rather than relying on prompt engineering. TreeCoder represents decoding as a tree search over candidate programs, where both decoding strategies and constraint functions - such as style, syntax, execution - are treated as first-class, optimisable components. This design enables systematic exploration and automatic tuning of decoding configurations using standard optimisation techniques. Experiments on the MBPP (Python) and SQL-Spider benchmarks show that TreeCoder consistently improves accuracy across open-source models such as CodeLlama, Mistral and DeepSeek, often outperforming their unconstrained baselines by considerable margins.