Research Engineer, Pretraining Scaling

Details

Anthropic’s ML Performance and Scaling team trains production pretrained models. This role works at the boundary between research and engineering to ensure frontier models train reliably, efficiently, and at scale. Responsibilities span performance optimization, hardware debugging, experimental design, observability, reliability, and launch coordination.

Responsibilities

• Own critical aspects of the production pretraining pipeline, including model operations, performance optimization, observability, and reliability.
• Debug and resolve issues across the full stack, including hardware errors, networking, training dynamics, and evaluation infrastructure.
• Design and run experiments to improve training efficiency, reduce step time, increase uptime, and enhance model performance.
• Respond to on-call incidents during model launches, diagnose problems quickly, and coordinate solutions across teams.
• Build and maintain production logging, monitoring dashboards, and evaluation infrastructure.
• Add capabilities to the training codebase, such as long-context support and novel architectures.
• Collaborate with teams across San Francisco and London, including the Tokens, Architectures, and Systems teams.
• Document systems, debugging approaches, and lessons learned.

Requirements

• Hands-on experience training large language models, or deep expertise with JAX, TPU, PyTorch, or large-scale distributed systems.
• Interest in both research and engineering work, with an approximately 50/50 balance between the two.
• Willingness to participate in on-call support, work extended hours during launches, and solve complex problems under pressure.
• Strong debugging skills across multiple layers of the technology stack.
• Clear communication and effective collaboration, including across time zones and during high-stress incidents.
• Passion for research engineering and responsible AI scaling.
• A bachelor’s degree or equivalent combination of education, training, and experience in a relevant field, as demonstrated through coursework, training, or professional experience.

Strong candidates may also have experience training LLMs; working extensively with JAX, TPU, PyTorch, or other ML frameworks at scale; contributing to open-source LLM frameworks such as open_lm, llm-foundry, or mesh-transformer-jax; publishing research on model training, scaling laws, or ML systems; working with production ML systems, observability tools, or evaluation infrastructure; or working as a systems engineer or quant.

The role is highly operational and includes production incident response. During launches, the team may work extended hours and respond to issues during evenings and weekends.

Benefits

Anthropic offers competitive compensation and benefits, optional equity donation matching, generous vacation and parental leave, flexible working hours, and an office space for collaboration. Anthropic sponsors visas and makes reasonable efforts to obtain a visa for candidates when an offer is made, with support from an immigration lawyer.