Bridging Functional Correctness and Runtime Efficiency Gaps in LLM-Based Code Translation

In three linesSwiftTrans, an LLM-based code translation framework, combines multi-perspective exploration (MpTranslator with parallel in-context learning) and difference-aware selection (DiffSelector) to improve both functional correctness and runtime efficiency. Evaluated on CodeNet, F2SBench, and SwiftBench.

Summary generated by Claude — human-verified