{"ID":2867567,"CreatedAt":"2026-06-01T04:54:23.091178241Z","UpdatedAt":"2026-06-01T04:54:23.091178241Z","DeletedAt":null,"paper_url":"https://arxiv.org/abs/2509.17489","arxiv_id":"2509.17489","title":"MapCoder-Lite: Distilling Multi-Agent Coding into a Single Small LLM","abstract":"Large language models (LLMs) have advanced code generation from single-function tasks to competitive-programming problems, but existing multi-agent solutions either rely on costly large-scale (\u003e30B) models or collapse when downsized to small open-source models. We present MapCoder-Lite, a framework for distilling the complex reasoning of large, multi-agent coding systems into a single 7B model. Our contribution is a novel, three-pillar methodology that synergistically generates, refines, and encodes multi-agent knowledge: (i) pass-based trajectory distillation from strong LLMs fixes format fragility in retrieval and reduces failures in debugging, (ii) supervisor-guided correction with global feedback strengthens planning and coding agents, and (iii) agent-wise LoRA fine-tuning delivers memory-efficient specialisation. Comprehensive evaluation on xCodeEval, APPS, and CodeContests shows that MapCoder-Lite more than doubles xCodeEval accuracy (from 13.2% to 28.3%), eliminates all format failures, while reducing GPU memory and token-generation time by 4x compared to a 32B model. It also achieves over 10% gains on simpler coding benchmarks, demonstrating broad improvements beyond competitive programming. These results demonstrate that careful agent-wise fine-tuning unleashes high-quality multi-agent coding on a small language model. Our code is publicly available at https://github.com/aiha-lab/MapCoder-Lite.","short_abstract":"Large language models (LLMs) have advanced code generation from single-function tasks to competitive-programming problems, but existing multi-agent solutions either rely on costly large-scale (\u003e30B) models or collapse when downsized to small open-source models. We present MapCoder-Lite, a framework for distilling the c...","url_abs":"https://arxiv.org/abs/2509.17489","url_pdf":"https://arxiv.org/pdf/2509.17489v2","authors":"[\"Woongkyu Lee\",\"Junhee Cho\",\"Jungwook Choi\"]","published":"2025-09-22T08:19:11Z","proceeding":"cs.CL","tasks":"[\"cs.CL\",\"cs.AI\"]","methods":"[\"Large Language Model\",\"Language Model\",\"LoRA\"]","has_code":false,"code_links":[{"ID":609493,"CreatedAt":"2026-06-01T04:54:23.091178241Z","UpdatedAt":"2026-06-01T04:54:23.091178241Z","DeletedAt":null,"paper_id":2867567,"paper_url":"https://arxiv.org/abs/2509.17489","paper_title":"MapCoder-Lite: Distilling Multi-Agent Coding into a Single Small LLM","repo_url":"https://github.com/aiha-lab/MapCoder-Lite","is_official":false,"mentioned_in_paper":false,"mentioned_in_github":true,"github_stars":0}]}