Magic
Member of Technical Staff, Pretraining Systems
$275K–$550K
San Francisco · On-site · Staff · Full-time · Posted Feb 28, 2026
- Visa sponsorship
- Visa sponsorship is offered, if possible.
About the job
Design and operate distributed systems that make large-scale pretraining of Magic’s long-context models performant, reliable, and reproducible.
What you'll do
- Scale distributed training across GPU clusters and optimize parallelism, communication, and gradient synchronization.
- Improve checkpointing, fault tolerance, recovery, profiling, orchestration, and hardware utilization.
- Collaborate with Kernels and Research to align model architecture with systems requirements.
What they ask for
- Strong software engineering and distributed systems fundamentals.
- Experience training large models in multi-node GPU environments.
- Deep understanding of parallelism strategies and performance trade-offs.
- Experience debugging production ML systems and improving large-scale system performance or reliability.
Pay
$275K - $550K USD base salary
Tools
GPU, data parallelism, tensor parallelism, pipeline parallelism
AI focus
Supports frontier-scale pretraining of long-context models as part of Magic’s effort to build safe AGI.
Model-written summary · the original posting is authoritative