Head of the Basic Model Laboratory

Lead the group in developing a general-purpose basic model

Responsible for the architecture design, data ratio, training strategy and scaling law research of pre-training large models (language/multimodal/code/etc.), building the underlying model capability to support the five AI application fields, and ensuring that the model performance reaches the leading level of the industry.

Model pre-training and continuous evolution system

Establish a complete pipeline from data cleaning, pre-training, post-training (SFT/RLHF) to model evaluation; design model version iteration and continuous learning mechanism, and promote knowledge enhancement and adaptation of basic models in the automotive vertical field.

Computation power and distributed training optimization

Frontier technology exploration and model architecture innovation

Track MoE, long text, multimodal, native Agent, etc. cutting-edge directions, push innovative architecture to internalize; produce high-quality papers, patents and technical reports, and enhance the research institute's influence in the academic community and open source community.

Cross-department model delivery and support

Provide stable and controllable basic model versions to the Agentic AI Laboratory, Data Science Department and Business Application Team; actively collect feedback from downstream, establish a closed-loop mechanism for model iteration, and support AI for Management/Engineering/Science/Manufacturing scenes.

Job Requirements

Computer, artificial intelligence or related professional doctoral degree, with at least 5 years of experience in large models or deep learning research, and at least 3 years of technical team management experience.

Have published a first-author paper in a major model conference (NeurIPS/ICML/ICLR/ACL/EMNLP/CVPR, etc.), or have led a well-known open-source model project.

Have extremely strong problem decomposition and engineering implementation abilities, and be able to balance model performance and training costs under resource constraints.