Graduate AI Platform Engineer

Pangle; San Jose, Northern (CA, KY); Hybrid; USD 140,000 - 210,000; Full time

Job summary

ByteDance's Volcano Ark training platform is seeking engineers to advance end-to-end large model post-training (SFT, RL) on a serverless Volcano Ark stack. Join the Data AML team driving scalable, secure ML infrastructure.

You will design elastic, multi-tenant training across data centers and heterogeneous hardware, aiming to boost throughput and stability while enabling reinforcement learning interfaces for users.

Qualifications

• Bachelor's or Master's degree in Computer Science or related field.
• Proficient in Python, Rust, or C++; writes clean code and has strong framework design and abstraction skills.
• Experience in training framework design or training system optimization; participation in large model training engineering or complex distributed system development is preferred.
• Strong interest in tracking and solving technical problems, studying low-level principles and performance bottlenecks, and applying scientific modeling methods.

Responsibilities

• Participate in the development of the Volcano Ark training system, enabling internal and external users to perform large model post-training, including SFT and RL, on the Volcano Ark platform in a serverless manner.
• Design elastic training solutions for complex multi-tenant training scenarios, supporting mixed multi-tenant training across multiple data centers and heterogeneous hardware, while improving training throughput and stability.
• Design reinforcement learning systems to improve training efficiency and provide user-friendly reinforcement learning training interfaces.

Skills

Python

Rust

C++

Education

Bachelor's or Master's in CS

Job description

ByteDance's Volcano Ark training platform is seeking engineers to advance end-to-end large model post-training (SFT, RL) on a serverless Volcano Ark stack. Join the Data AML team driving scalable, secure ML infrastructure.

You will design elastic, multi-tenant training across data centers and heterogeneous hardware, aiming to boost throughput and stability while enabling reinforcement learning interfaces for users.