Scholarlink

<p>We present a reinforcement learning framework that enhances natural language queries to improve DeepSeek code generation. A parametric refiner (Qwen with LoRA) is trained via REINFORCE while the generator remains fixed, using a scalar reward that can combine text similarity (BLEU-4, ROUGE-L, F1, Overlap) with execution signals (unit tests, syntax/timeout penalties). On the DS1000 benchmark (800 train / 200 test), RL4QE improves the code similarity by 34.3%. Ablations show that BLEU-4 is the most reliable text reward overall (with F1 competitive on a larger scale), and LoRA with rank <InlineEquation ID="IEq1"> <InlineMediaObject> <ImageObject Color="BlackWhite" FileRef="41598_2025_21271_Article_IEq1.gif" Format="GIF" Height="14" Rendition="HTML" Resolution="72" Type="Linedraw" Width="30" /> </InlineMediaObject> <EquationSource Format="TEX">\(r{=}8\)</EquationSource> </InlineEquation> outperforms complete fine-tuning on most metrics while being more parameter efficient. The approach is transferred across foundation models (e.g., Qwen1.5/2/2.5 variants), where architecture often matters more than size. RL4QE is easy to integrate in practice (LoRA in attention projections) and supports reproducibility.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Enhancing queries for code generation with reinforcement learning

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2025年10月24日

查看全文
  • Dawei Yuan,
  • Guojun Liang,
  • Tingting Li,
  • Suping Liu

摘要

We present a reinforcement learning framework that enhances natural language queries to improve DeepSeek code generation. A parametric refiner (Qwen with LoRA) is trained via REINFORCE while the generator remains fixed, using a scalar reward that can combine text similarity (BLEU-4, ROUGE-L, F1, Overlap) with execution signals (unit tests, syntax/timeout penalties). On the DS1000 benchmark (800 train / 200 test), RL4QE improves the code similarity by 34.3%. Ablations show that BLEU-4 is the most reliable text reward overall (with F1 competitive on a larger scale), and LoRA with rank \(r{=}8\) outperforms complete fine-tuning on most metrics while being more parameter efficient. The approach is transferred across foundation models (e.g., Qwen1.5/2/2.5 variants), where architecture often matters more than size. RL4QE is easy to integrate in practice (LoRA in attention projections) and supports reproducibility.