Job Summary:
ByteDance is a technology brand committed to becoming a leading XR platform, focusing on innovation and R&D. They are seeking a Large Model Optimization Engineer Graduate to participate in the research and development of inference optimization and acceleration for large models, collaborating with various algorithm teams to enhance performance.
Responsibilities:
• Participate in the research and development of inference optimization and acceleration for models such as LLM/VLM/SD, as well as inference engines and frameworks;
• Build an industry-leading on-device large model inference engine through high-performance optimization technologies such as efficient operator development, low-precision computing, streaming inference, and speculative sampling;
• Collaborate deeply with various algorithm teams to analyze business performance bottlenecks and conduct performance analysis and optimization for large models;
• Participate in the performance evaluation of models on different chips.
Qualifications:
Required:
• Final year Ph.D or recent Ph.D graduates in Computer Science, engineering or quantitative field
• Proficiency in C/C++ and Python under the Linux environment
• Ability to skillfully use at least one mainstream machine learning framework, with preference given to those familiar with various model/data parallel training frameworks
• Knowledge of mainstream models such as LLM/VLM/SD with experience in model inference optimization preferred
• Experience in performance modeling, performance analysis and optimization, or knowledge of CPU and GPU architectures is preferred
• Experience in GPU programming (CUDA or OpenCL) and familiarity with TensorRT/Triton/Cutlass is preferred
• Those with top conference papers in the direction of AutoML or AIGC are preferred
• Those who have published papers in top computer vision conferences or journals are preferred
• Those who have achieved excellent results in well-known computer vision competitions are preferred
• Those with experience in high-quality Github projects are preferred
Company:
ByteDance is a technology company that develops content creation platforms and services. Founded in 2012, the company is headquartered in Beijing, CHN, with a team of 10001+ employees. The company is currently Late Stage.