Research Scientist: Multimodal World Models & Pre-Training

ByteDance; San Jose (CA); On-site; USD 187,040 - 438,000; Full time

Benefits offered by this job

Medical, dental, and vision insurance

401(k) savings plan with company match

Paid parental leave

Paid holidays and personal time

Job summary

A leading tech company is seeking a Research Scientist for Multimodal Interaction and World Model. This role focuses on enhancing models for multimodal data understanding, requiring expertise in AI and hands-on experience with frameworks like PyTorch/JAX. The position offers a competitive salary between $187,040 and $438,000 annually, plus benefits that include medical, dental, and vision insurance, a 401(k) plan, and paid personal time.

Qualifications

• Expertise in scaling Transformers on GPUs or TPUs is essential.
• Rich hands-on experience in distributed training frameworks is required.
• Familiar with state-of-the-art techniques for preparing multimodal training data.

Responsibilities

• Drive research to improve models for multimodal data understanding.
• Design architectures for performance and efficiency optimization.
• Establish scaling laws and conduct systematic ablation studies.

Skills

Expertise in Transformers (Dense and MoE)

Hands-on experience in PyTorch/JAX

Familiarity with multimodal training data preparation

Job description

A leading tech company is seeking a Research Scientist for Multimodal Interaction and World Model. This role focuses on enhancing models for multimodal data understanding, requiring expertise in AI and hands-on experience with frameworks like PyTorch/JAX. The position offers a competitive salary between $187,040 and $438,000 annually, plus benefits that include medical, dental, and vision insurance, a 401(k) plan, and paid personal time.