AGIBOT

多模态交互模型工程师

AGIBOT  •  Shanghai, CN (Onsite)  •  3 hours ago
Apply
AI can make mistakes so check important info. Chat history is never stored.

Job Description

多模态交互模型工程师

上海

外包

互联网 / 电子 / 网游

职位描述

  1. 训练流水线建设: 搭建与维护多模态模型训练 Pipeline(涵盖数据预处理、实验追踪、Checkpoint 管理与回归评测),支持音视频、状态序列等多模态输入。 2. Teacher-Student 蒸馏: 使用大模型/专家模块生成 Richer Labels,设计蒸馏目标、Loss 函数与 Hard-negative 机制,将高价值能力稳定转移至端侧,摆脱对云端的在线依赖。 3. 模型压缩与部署优化: 负责 PTQ/QAT、量化感知训练、剪枝与 KV/Cache 优化;导出 ONNX/TensorRT 等格式;在真实设备上死磕 Latency、显存、功耗与推理抖动。 4. 质量与回归治理: 建立可复现机制,系统性归因模型失效(数据、标签、结构或部署问题);联合团队建设 Hard slices 与回放评测。 5. 运行时协同: 与系统侧拆分快慢路径,trade-off 端侧性能与业务收益,推动模型从“实验室能跑”跨越到“设备稳跑”。 6. 同时需要兼顾交互语义建模与任务定义(如回应对象判断、交互时机、主动策略等)

职位要求

必备项: 1. 计算机、人工智能相关专业本科及以上学历。 2. 3 年以上经验,在训练工程、蒸馏压缩、部署优化中具备至少两项的深厚实战积累。 3. 精通 PyTorch 训练体系,代码工程能力极强,能独立且稳定地维护训练仓库。 4. 具备量化优化、ONNX/TensorRT/CUDA 推理优化经验。 5. 具备出色的 Problem-solving 能力,能将“模型指标不升反降”精准拆解并定位根因。 加分项: 1. 拥有 Jetson / ARM / 嵌入式 AI 推理极致优化经验。 2. 有流式模型(Streaming)、Chunked inference、端云一致性治理经验。 3. 熟悉 FSDP / DDP / DeepSpeed 及混合精度训练。 4. 做过端侧语音、视觉、边缘 VLM 或轻量多模态模型。

投递

AGIBOT

About AGIBOT

AGIBOT is an Embodied AI foundation model company developing both the intelligence layer and the corresponding robotic embodiments needed to bring general intelligence into the physical world. Founded in 2023, AGIBOT is building a full-stack Embodied AI platform across foundation models, data infrastructure, simulation, robotic hardware, and deployment. AGIBOT’s “Three Intelligences in One” architecture integrates Locomotion Intelligence, Interaction Intelligence, and Manipulation Intelligence into a unified embodied system. AGIBOT’s portfolio spans humanoid robots, quadrupeds, dexterous systems, and commercial cleaning solutions. In March 2026, AGIBOT announced that its 10,000th robot had rolled off the production line.

Industry
Manufacturing & Production
Company Size
201-500 employees
Headquarters
Shanghai, CN
Year Founded
Unknown
Social Media