Responsibilities
Team Introduction: TikTok is an international short video platform covering 150 countries and regions. We hope to discover real and interesting moments through TikTok and make life better. TikTok has offices around the world, with global headquarters in Los Angeles and Singapore, and offices in New York, London, Dublin, Paris, Berlin, Dubai, Jakarta, Seoul and Tokyo. TikTok AI Innovation Center is a department dedicated to AI infrastructure construction and innovative research, exploring industry-leading artificial intelligence technologies, including large language models, multimodal large models and other research directions. We hope to develop model algorithms that can handle multi-language and massive video content understanding to bring users a better content consumption experience. In the Code AI direction, we use the powerful code understanding and reasoning capabilities of large language models to improve program performance and R&D efficiency. 1. Responsible for the performance optimization of large model training and reasoning to ensure that the training and push efficiency reaches the industry-leading level 2. In-depth cooperation with algorithm students to conduct in-depth joint optimization of algorithms and training and push systems to support the implementation of cutting-edge technologies in the department 3. Responsible for the research and implementation of cutting-edge technologies in the direction of large model training and push, and continuously strengthen and enhance the competitiveness of key technologies in the department.
Qualifications
1. Proficient in C++ and Python, with a good foundation in data structure and algorithm, and a deep understanding of computer architecture. ACM/ICPC/Codeforces and other award winners are preferred 2. Familiar with mainstream parallel programming and performance optimization techniques, familiar with CUDA and Tensor Core programming, and those with relevant high-performance optimization experience are preferred 3. Understand the mainstream LLM model, familiar with TensorRT-LLM, ORCA, VLLM and have LLM model training and acceleration experience are preferred 4. Familiar with TVM, MLIR, XLA and other automated compilation technologies are preferred 5. Those with Slim high-performance optimization experience such as quantization and sparsification are preferred.
