← Back to roles

混元vlm算法研究员-后训练/基座能力提升方向(深圳/北京/上海)

Tencent Technology (Shenzhen) Co., Ltd.

北京Posted Aug 11, 2026
Apply on Tencent Technology (Shenzhen) Co., Ltd.

Overview

Open roles
2,249
At this company
See all
Added in 30 days
528
At this company, still open

Skills named in this posting

  • Python
  • PyTorch / TensorFlow
1.主导后训练研发体系:负责多模态大语言模型(VLM)的后训练(Post-training)全流程算法研发,深入探索 SFT 及强化学习对齐算法(如 RLHF、PPO、GRPO 等),持续逼近并提升视觉基座模型的能力上限; 2.攻坚视觉基础感知:致力于优化模型底层的视觉感知、细粒度理解与空间感知能力。前瞻性地探索包括“带图思考”在内的 Visual Agent 解决方案,赋予模型更深层次的物理世界理解与执行能力; 3.提高大模型可靠性:致力于大幅度降低多模态大模型幻觉。深入探究幻觉产生的本质,建设完善的幻觉检测链路,实施精细化的数据清洗策略,并主导研发基于强化学习的幻觉消除(Hallucination Removal)方案; 4.赋能用户交互场景:面向真实用户场景进行模型对齐与调优,强化模型在多轮图文交互中的逻辑推理、指令遵循(Instruction Following)以及细节感知能力。搭建线上 Badcase 监控与数据飞轮反馈闭环,持续打磨并打造极致的交互体验。 1.计算机科学/软件工程/人工智能等相关专业硕士及以上学历; 2.深入理解 Transformer 架构及大语言模型训练原理,在 LLM Alignment、RLHF、Reward Modeling、个性化大模型(Personalized LLM)、记忆机制(Memory/RAG)等后训练领域之一有深入的研究和实践经验; 3.在用户画像建模、推荐系统与大模型结合、或超长上下文/长期记忆模型优化方面有丰富实战经验者优先; 4.具备扎实的算法基础和工程实现能力,熟练掌握 Python,熟悉 PyTorch/TensorFlow 等深度学习框架; 5.拥有分布式训练实战经验,熟悉 Megatron-LM, DeepSpeed, vLLM 等大规模训练及推理框架,有百亿/千亿参数模型训练或调优经验者优先; 6.具备优秀的科研素养,有高质量论文发表(NeurIPS, ICLR, ICML, ACL, EMNLP 等)或在开源社区(如 HuggingFace)有高影响力项目贡献者优先; 7.具备强烈的技术热情和自驱力,善于分析和解决复杂问题,拥有良好的团队协作和沟通能力。

Apply on the employer's site

Original posting: careers.tencent.com

Apply on Tencent Technology (Shenzhen) Co., Ltd.

RISC-V ecosystem

Tencent Technology (Shenzhen) Co., Ltd. in the RISC-V ecosystem

RISC-V International: Premier TSC

Tencent Holdings Ltd. is a Chinese multinational technology and entertainment conglomerate and holding company headquartered in Shenzhen.

  • Tencent joined RISC-V International as China pushed for open-instruction-set alternatives (widely reported)
  • Its TencentOS Tiny IoT operating system is used on RISC-V MCUs, e.g. tutorials pairing TencentOS Tiny with the GD32V RISC-V MCU
  • Also participates in Chinese RISC-V community events such as the RISC-V Forum in Shenzhen

Same employer

View all 2,249