跳转至

微调 Qwen3-27B 实现 C 到 Rust 代码转换：包含预训练、调试感知 SFT 与特定任务 SFT 的三阶段课程

文章背景与核心概要

将陈旧的 C 语言代码库转换为安全、地道的 Rust 代码是消除内存安全漏洞、提升软件工程质量的关键目标。然而，开箱即用的通用大语言模型（LLM）在此类任务上往往表现不佳，因为常规的预训练缺乏对地道 Rust 代码生成、跨语言语义等价性以及编译器反馈集成的专项优化。

本文针对 Qwen3-27B 模型提出了一种三阶段微调课程：1. 以 Rust 为中心的预训练（在 Rust 语料库上持续预训练以建立地道语法和标准库使用的先验）；2. 调试感知监督微调（SFT）（使用 microsoft/Verus_Training_Data 数据集训练模型的调试与自我修复能力）；3. 特定任务 SFT（在成对的 C/Rust LeetCode 解决方案上训练，以实现直接、准确的语义转换）。

研究团队使用 SACTOR 框架对最终的模型进行了评估。该框架是一个智能体式的、由静态分析引导的验证工具，具备两阶段转换流程（从非地道到地道）以及基于外部函数接口（FFI）的端到端（E2E）测试，结果表明微调后的模型性能显著优于基线模型。

摘要 (Abstract)

Translating C code into safe, idiomatic Rust is a longstanding software-engineering goal because it can eliminate entire classes of memory-safety vulnerabilities while preserving the functional behavior of legacy systems. Large language models (LLMs) have shown promise for this task but typically underperform when applied off-the-shelf, since general-purpose pretraining rarely emphasizes idiomatic Rust generation, cross-language semantic equivalence, or the ability to reason about and repair compiler/runtime feedback.

将 C 代码转换为安全且地道的 Rust 一直是软件工程领域长期追求的目标，因为这不仅能保留遗留系统的功能行为，还能根除一整类内存安全漏洞。大语言模型（LLM）在这一任务中展现出了巨大潜力，但直接拿来即用时往往表现不佳，因为通用预训练很少强调地道的 Rust 生成、跨语言语义等价性，也缺乏对编译器/运行时反馈的推理和修复能力。

Translating C code into safe, idiomatic Rust is a longstanding software-engineering goal because it can eliminate entire classes of memory-safety vulnerabilities while preserving the functional behavior of legacy systems. Large language models (LLMs) have shown promise for this task but typically underperform when applied off-the-shelf, since general-purpose pretraining rarely emphasizes idiomatic Rust generation, cross-language semantic equivalence, or the ability to reason about and repair compiler/runtime feedback.

In this report we describe a three-stage fine-tuning curriculum applied to Qwen3-27B that is designed to progressively specialize the model for the C-to-Rust (C2Rust) translation task: 1. Continued pretraining on Rust-centric corpora to strengthen the model's prior over idiomatic Rust syntax and standard-library usage; 2. Supervised fine-tuning (SFT) on the microsoft/Verus_Training_Data dataset to instill debugging and self-repair behavior over Rust code; and 3. Task-specific SFT on paired C/Rust solutions derived from LeetCode problems to teach direct semantic translation.

在本报告中，我们描述了应用于 Qwen3-27B 的三阶段微调课程，旨在逐步让该模型专门适应 C 到 Rust (C2Rust) 的转换任务： 1. 持续预训练：在以 Rust 为中心的语料库上进行预训练，以增强模型对地道 Rust 语法和标准库使用的先验知识； 2. 监督微调（SFT）：在 microsoft/Verus_Training_Data 数据集上进行训练，以培养模型对 Rust 代码的调试和自我修复行为； 3. 特定任务 SFT：在源自 LeetCode 问题的成对 C/Rust 解决方案上进行训练，以教授直接的语义转换。

In this report we describe a three-stage fine-tuning curriculum applied to Qwen3-27B that is designed to progressively specialize the model for the C-to-Rust (C2Rust) translation task: 1. Continued pretraining on Rust-centric corpora to strengthen the model's prior over idiomatic Rust syntax and standard-library usage; 2. Supervised fine-tuning (SFT) on the microsoft/Verus_Training_Data dataset to instill debugging and self-repair behavior over Rust code; and 3. Task-specific SFT on paired C/Rust solutions derived from LeetCode problems to teach direct semantic translation.

We evaluate the resulting model using the agentic, static-analysis-guided verification framework of SACTOR, which performs structure-aware, two-phase (unidiomatic to idiomatic) translation with foreign-function-interface (FFI)-based end-to-end (E2E) testing. We report success rate, idiomaticity (Clippy lint counts, unsafe-code fraction), and failure-mode analyses, and compare our fine-tuned model against baseline Qwen3-27B and other LLMs evaluated under the same framework.

我们使用智能体式的、由静态分析引导的验证框架 SACTOR 对最终模型进行了评估。该框架执行结构感知、两阶段（从非地道到地道）的翻译，并结合了基于外部函数接口（FFI）的端到端（E2E）测试。我们报告了成功率、地道性（Clippy lint 计数、不安全代码占比）以及失败模式分析，并将微调后的模型与基线 Qwen3-27B 以及在相同框架下评估的其他 LLM 进行了对比。

We evaluate the resulting model using the agentic, static-analysis-guided verification framework of SACTOR, which performs structure-aware, two-phase (unidiomatic to idiomatic) translation with foreign-function-interface (FFI)-based end-to-end (E2E) testing. We report success rate, idiomaticity (Clippy lint counts, unsafe-code fraction), and failure-mode analyses, and compare our fine-tuned model against baseline Qwen3-27B and other LLMs evaluated under the same framework.

概要与元数据 (Summary & Metadata)

  • arXiv ID: arXiv:2608.13681 [cs.SE]
  • Primary Subject: Software Engineering (cs.SE)
  • Other Subjects: Artificial Intelligence (cs.AI), Emerging Technologies (cs.ET)
  • Submission Date: August 13, 2026
  • DOI: 10.48550/arXiv.2608.13681

作者 (Authors)

  • Pu Zhao
  • Changdi Yang
  • Yixiao Chen
  • Yi Gao
  • Yifan Cao
  • Haochen Zeng
  • Yanzhi Wang
  • Pu Zhao
  • Changdi Yang
  • Yixiao Chen
  • Yi Gao
  • Yifan Cao
  • Haochen Zeng
  • Yanzhi Wang

全文与资源 (Full-Text and Resources)