PhD Intern, LLM Model Research

Job Description

Student Researcher (Seed – LLM – Model) – 2026 Start (PhD) | ByteDance

The Tone:
This is a PhD Internship at ByteDance, located in the United States, targeting a 2026 start. ByteDance, a global technology company, is dedicated to inspiring creativity and enriching life through innovative products. This role is crucial for advancing foundational algorithm research for large language models, ensuring their performance, efficiency, and stability for various downstream applications. PhD interns actively contribute to the company’s products, research, future plans, and emerging technologies.

The TL;DR
• Role: Internship
• Type: Full-time (for the duration of the internship)
• Location: In-person, United States
• Pay: $60 hourly
• Team: Seed-LLM-Model team, focused on foundational algorithm research for LLM models
• Mission: Conduct cutting-edge research and development in LLM and MultiModal Machine Learning to solve practical industry problems.
• Tech Stack: PyTorch, TensorFlow, Megatron, FSDP, Deepspeed, Python, C++

What You’ll Actually Do
• Research: Research and develop cutting-edge algorithms for large language models and MultiModal Machine Learning.
• Innovate: Conduct in-depth research on advanced technologies in LLM and MultiModal Machine Learning fields.
• Apply: Apply cutting-edge LLM/MultiModal ML technologies to solve practical problems within the industry.
• Contribute: Participate in foundational algorithm research, specifically focusing on model architecture, optimization, and stability.
• Publish: Pursue opportunities to publish top international papers and apply for patents based on research contributions.

The Must-Haves
• Background: Currently pursuing a PhD in artificial intelligence, computer science, automation, mathematics, or a related technical discipline.
• Experience: Solid foundation in data structure and algorithm design; proficient in deep learning frameworks like PyTorch and TensorFlow; proficient in distributed large language model training frameworks such as Megatron, FSDP, or Deepspeed.
• Skills: Proficient in Python/C++; good reading and writing skills; solid foundation in mathematics; strong sense of responsibility, proactive, with good communication and teamwork skills.
• Bonus: Experience with pre-trained basic technologies including efficient training and encapsulated deployment services (NLP, CV, video, MultiModal Machine Learning, and their downstream applications); published papers in accredited academic conferences; excellent results in MultiModal Machine Learning, Computer Vision, or Machine Learning competitions.