Contribute to the development of open-source inference frameworks such as SGLang and vLLM, including feature and operator development, performance optimization, and model support, in collaboration with the community.
Develop and optimize KV cache offloading frameworks for LLM workloads, supporting multi-level cache offloading and reuse across CPU, SSD, and remote storage to improve inference efficiency.
Drive R&D on compute performance in distributed training, and explore methods and technologies for performance optimization.
Study computational challenges in machine learning systems, identify common needs and bottlenecks, and build example code, acceleration libraries, or frameworks accordingly.
What We Need to See:
Over 3 years working experience in the technology industry, with master's degree or above in computer science, mathematics, electrical engineering, automation, or related fields.
Strong interest in accelerated computing, parallel computing, and heterogeneous computing, with the motivation to explore these areas in depth.
Solid programming skills, with a good understanding of data structures and computer systems fundamentals.
Strong learning agility, adaptability, and the ability to analyze, define, and independently explore technical problems.
Ways to Stand Out from the Crowd:
Familiarity with heterogeneous computing, distributed training, parallel computing, or other areas related to high-performance computing.
Experience in performance analysis, performance modeling, or performance optimization; contributions to open-source frameworks are a plus.
Strong ability to define new problems and explore solutions; candidates with independent PhD-level research experience are preferred.
Proficiency with AI coding tools.
#LI-Hybrid
Additional Information
NVIDIA is leading company of AI computing. At NVIDIA, our employees are passionate about AI, HPC , VISUAL, GAMING. Our SA team is more focusing to bring NVIDIA new technology into difference industries. We help to design the architecture of AI computing platform, analysis the AI and HPC applications to deliver our value to customers, focusing on defining and solving computational challenges in LLM inference and training acceleration, as well as network communication and data transfer optimization.
Nvidia · Beijing, China