LEGO-RL: Harness-Native Reinforcement Learning for Coding Agents
- 类型:arxiv
- 标识:2608.17393
- 链接:https://arxiv.org/abs/2608.17393
- 主分类:agent
- 形态:method
- 被引:0
- 被引来源:Semantic Scholar
- S2被引:0
- 影响力被引:0
- TLDR:LIFE-RL, a framework that bridges native coding-agent harnesses with scalable policy-gradient optimization without modifying their internal control flow, is presented and evaluated by training the sparse MoE model Qwen3.5-35B-A3B with GSPO across three native coding-agent harnesses.
- 副分类:evaluation
- 待LLM分类:否
- 标题中文:LEGO-RL:面向编码 Agent 的 Harness 原生强化学习
- TLDR中文:本文提出了 LIFE-RL 框架,在不修改内部控制流的前提下,将原生编码 Agent harness 与可扩展的策略梯度优化相连接,并通过 GSPO 在三个原生编码 Agent harness 上训练稀疏 MoE 模型 Qwen3.5-35B-A3B 对其进行了评估。
- 来源文件:
- /inbox/tom/_candidates/2026-08-20-agent-rag-longcontext-candidates.json
- [S2 enrich]