Staff Training Infra Engineer - ML Pipelines (Remote)

Cohere; Greater London; Remote; GBP 100,000 - 150,000; Full time

Benefits offered by this job

Lunch stipend

Health and dental benefits

RRSP/401K type benefits

Parental leave top-up

Education & learning stipend

Vacation time

Travel to other offices

Home office stipend

Job summary

Cohere is seeking engineers to contribute to model training pipelines and production-ready code. You will help ship state-of-the-art models and optimize training infrastructure in a remote-friendly environment.

The role blends research and production work, with access to extensive compute and data resources to push frontier models. Teams collaborate across research and engineering to accelerate experiments, with offices globally but no strict in-office requirement.

Qualifications

• Extremely strong software engineering skills.
• Proficiency in Python and ML frameworks (JAX, PyTorch).
• Experience with distributed training infra (Kubernetes, Slurm) and Ray.
• Experience with large-scale distributed training strategies.
• Hands-on with training large models and tooling/setup of training infra.

Responsibilities

• Design and implement high-performant, scalable training software.
• Enhance training infrastructure performance and reliability.
• Develop tools to accelerate training cycles and improve infra efficiency.
• Experiment with ideas on compute and data infrastructure.
• Collaborate with researchers to bridge research and production.

Skills

Strong software engineering

Python

JAX/PyTorch/XLA-MLIR

Distributed training

Training infrastructure tooling

Tools

Kubernetes

Slurm

Ray

JAX

PyTorch

XLA/MLIR

Job description

Cohere is seeking engineers to contribute to model training pipelines and production-ready code. You will help ship state-of-the-art models and optimize training infrastructure in a remote-friendly environment.

The role blends research and production work, with access to extensive compute and data resources to push frontier models. Teams collaborate across research and engineering to accelerate experiments, with offices globally but no strict in-office requirement.