[Skip to main content](#main-content)

[OpenNash Jobs](https://jobs.opennash.com/)
[Hiring report](/reports/hiring/) · [All JD.com jobs](https://jobs.opennash.com/companies/jd/) · [Search jobs](/search/)

JD.com · 北京市-北京市

# 基于电商大模型的Mid-train及RL研究

北京市-北京市 

Listed on JD.com’s official careers site. Posted July 20, 2026. Check that it is still open and apply there.
[Apply on JD.com’s careers site →](https://campus.jd.com/#/details?id=8766)

[OpenNash Jobs](https://jobs.opennash.com/) / [JD.com](https://jobs.opennash.com/companies/jd/) / 基于电商大模型的Mid-train及RL研究

## Job description

工作职责
- 设计并实现电商场景下的大模型中期训练(Mid-train)算法，解决领域知识注入与能力对齐等核心难题；
- 负责电商场景下Agentic大模型的强化学习后训练，攻坚奖励模型(RM)、策略优化(如PPO、GRPO)及多轮交互决策优化等关键技术；
- 构建面向电商业务的高质量训练数据、奖励数据及评测基准(Benchmark)，通过数据闭环迭代驱动模型性能优化；
- 推动大模型研究成果在智能推荐、智能导购及智能运营等真实业务场景中的端到端落地，解决实际业务痛点。 任职要求
- 计算机、人工智能或相关专业博士 / 硕士，有扎实的机器学习、强化学习理论基础。
- 熟悉大语言模型（LLM）训练与后训练流程，有 SFT/RLHF/RLOO/GRPO 等相关项目经验。
- 有 Agent 训练、Tool-use、多步推理等相关研究或工程落地经验者优先。
- 熟悉 PyTorch，有大规模分布式训练（如 DeepSpeed、Megatron-LM）经验者优先。
- 有顶会论文（ICML/NeurIPs/ICLR 等）或开源项目贡献者优先。

## At a glance
- Posted by the employer: Jul 20, 2026
- First seen by OpenNash: Oct 19, 2026
- Last checked: Oct 19, 2026
- Listed for: 91 days
- Location: 北京市-北京市[Apply on JD.com’s careers site →](https://campus.jd.com/#/details?id=8766)

## More jobs at JD.com

### [3D医疗影像多模态基础大模型研究](https://jobs.opennash.com/companies/jd/jobs/31cea141d3ff4222fc49/)

北京市-北京市

### [3D可控生成与空间智能多模态研究](https://jobs.opennash.com/companies/jd/jobs/9264cfdaab33cfb45f96/)

北京市-北京市

### [AI多模态存储与向量化计算探索研究](https://jobs.opennash.com/companies/jd/jobs/e91a2f3f9d7be9a91867/)

北京市-北京市

### [CodeAI代码大模型全链路与推理优化](https://jobs.opennash.com/companies/jd/jobs/234f9df632b03e4bec55/)

北京市-北京市

OpenNash lists jobs from employers’ own careers sites and does not handle applications.
Dates come from the employer where stated; “first seen” and “last checked” are our own observations.

OpenNash Jobs · Official listings with dates and links to the employer. [Weekly hiring report](/reports/hiring/) · [Methodology](/methodology/)