Overview In this role you shape efficient, large-scale generative AI systems within the Vision-Applied Research team at TikTok, focusing on distillation and compression to transfer capabilities from foundation models to smaller, hardware-efficient variants. You will advance scalable generative and multimodal models, optimizing for speed and resource use while enabling deployment at scale. You collaborate across research and engineering to push diffusion and autoregressive methods, delivering practical, impactful AI solutions for creative content tools. This is a chance to influence cutting-edge generative AI at a platform that enables easier, creative user experiences.
Responsibilities- Design and implement efficient algorithms and architectures for large-scale generative and multimodal models (step distillation, cfg distillation, quantization) to boost efficiency
- Advance scalable generative modeling approaches (diffusion, autoregressive) with focus on acceleration
- Develop distillation frameworks, model acceleration, and hardware-efficient inference for practical deployment
Key requirements- B.S. in Computer Science or related fields; equivalent experience
- Expertise in efficient models with understanding of bottlenecks and acceleration
- Proficiency in training generative AI/LLM models using PyTorch and JAX
- Strong communication and collaboration skills in fast-paced environments
- communication
- collaboration
- adaptability
- step distillation
- cfg distillation
- quantization
