Senior GenAI Research Engineer - Distillation & Efficiency

TikTok Inc.; San Jose (CA); On-site; USD 254,000 - 588,000; Full time

Benefits offered by this job

Medical, dental & vision insurance

401(k) with company match

Paid parental leave

Short-term & long-term disability

Life insurance

Wellbeing benefits

13+ paid holidays per year

Job summary

TikTok Inc. Vision-Applied Research team seeks a Research Engineer/Scientist to design and implement efficient large-scale generative AI models, with emphasis on distillation and compression to transfer capabilities from foundation models into smaller, faster systems.

You will develop methods and infrastructure for scalable training, optimization, and deployment, including distillation frameworks, model acceleration, and hardware-efficient inference for image/video generation and multimodal

Qualifications

• B.S. in Computer Science or related field or equivalent experience.
• Expertise in efficient models with acceleration methods.
• Experience training generative AI/LLM models with PyTorch and JAX.
• Strong communication and collaboration in fast-paced environments.

Responsibilities

• Develop efficient algorithms and architectures for large-scale generative and multimodal models.
• Advance scalable generative modeling approaches with diffusion and autoregressive models, focusing on acceleration and efficiency.

Skills

Efficient models & acceleration

PyTorch & JAX

Communication & collaboration

Generative AI proficiency

Education

B.S. in Computer Science or related field

Ph.D. in GenAI / MLSys or equivalent

Tools

PyTorch

JAX

Job description

TikTok Inc. Vision-Applied Research team seeks a Research Engineer/Scientist to design and implement efficient large-scale generative AI models, with emphasis on distillation and compression to transfer capabilities from foundation models into smaller, faster systems.

You will develop methods and infrastructure for scalable training, optimization, and deployment, including distillation frameworks, model acceleration, and hardware-efficient inference for image/video generation and multimodal