Job Responsibilities
1. Responsible for the research and development of core algorithms for visual perception and multimodal large models in automotive scenarios, includi……
Job Responsibilities 1. Responsible for the research and development of core algorithms for visual perception and multimodal large models in automotive scenarios, including but not limited to pre-training, supervised fine-tuning (SFT), and reinforcement learning optimization of cockpit large models (LLM/VLM); 2. Develop models, perform data cleaning, effect evaluation, and performance benchmarking for cockpit vertical domain tasks (such as vehicle control, navigation, multimedia, Q&A, wake-word-free interaction, etc.), improve semantic understanding (NLU) accuracy and generalization; optimize model response quality for cockpit open-domain chitchat; Combine multi-dimensional information such as voice, vision, and vehicle status to explore the application of multimodal large models (MLLM) in the cockpit and realize cockpit multimodal interaction; Engage in the engineering implementation of cockpit large models, responsible for compression technologies such as model quantization, pruning, and distillation, and collaboratively realize low-latency deployment of algorithms on automotive chips (such as Snapdragon 8295/8775, etc.) and device-cloud collaborative inference; 1. Bachelor's degree or above, majoring in Computer Science, Information Technology, Electronic Engineering, or related fields; CET-6 or above, with deep knowledge and understanding of cutting-edge algorithms, and research experience in CV algorithms; 2. Able to quickly follow the latest frontier papers in the industry on LLM/VLM and possess the ability to transform them into production algorithms; 3. Familiar with the PyTorch framework, with experience in large-scale distributed computing power training; experience in quantization (INT8/INT4/AWQ) and inference acceleration; familiar with the Linux development environment, with a solid foundation in data structures and algorithms;