Search papers, labs, and topics across Lattice.
This paper introduces EffiPair, an inference-time framework that refines LLM-generated code for efficiency by using Relative Contrastive Feedback (RCF). RCF compares pairs of structurally similar programs for the same task, highlighting differences associated with better efficiency, thus providing targeted guidance for refinement. Experiments on code-efficiency benchmarks demonstrate EffiPair achieves up to 1.5x speedup with DeepSeek-Chat V3.2, while drastically reducing token usage compared to prior methods relying on absolute execution feedback.
LLMs can learn to write faster code at inference time, without fine-tuning, by contrasting the execution of similar candidate programs.
Large language models (LLMs) often generate code that is functionally correct but inefficient in runtime and memory. Prior approaches to improving code efficiency typically rely on absolute execution feedback, such as profiling a single program's runtime or memory usage, which is costly and provides weak guidance for refinement. We propose Relative Contrastive Feedback (RCF), an inference-time feedback mechanism that requires no model fine-tuning or parameter updates. RCF compares two structurally similar programs for the same task and highlights the differences associated with better efficiency. Building on this idea, we introduce EffiPair, an inference-time iterative refinement framework that operates entirely at test time by generating multiple candidate solutions, identifying informative program pairs with large efficiency gaps, summarizing their execution differences into lightweight feedback, and using this signal to produce more efficient solutions. By replacing isolated scalar feedback with pairwise contrastive comparisons, EffiPair provides more direct guidance while reducing profiling and prompting overhead. Experiments on code-efficiency benchmarks show that EffiPair consistently improves efficiency while preserving correctness. For instance, with DeepSeek-Chat V3.2, EffiPair achieves up to 1.5x speedup over generation without performance feedback, while reducing token usage by more than 90% compared to prior work.
