Member of Technical Staff, Training Performance Engineer
Cohere · London
1 year ago
HybridFull_timelanguage model trainingnatural language processinggpu performance optimization
1 year ago
Workplace
Hybrid
Employment
Full_time
Location
London
Posted
1 year ago
Summary
Cohere is hiring a member of technical staff, training performance engineer, in the pre-training team. You will optimize training throughput and accelerator utilization for advanced language models by building scalable training software, writing cuda and triton kernels, and experimenting with ideas across supercompute and data infrastructure. The role is remote-friendly and operates primarily between et and cet time zones.
Responsibilities
- Design and write high-performant training software
- Analyze architectural changes for throughput and quality
- Write cuda and triton kernels for accelerators
- Research and implement training and profiling tools
- Identify and remove performance bottlenecks
- Experiment with ideas on supercompute and data infrastructure
pythonjaxpytorchxla/mlircudatritongpu kernelsdistributed trainingtransformers
language model trainingnatural language processinggpu performance optimizationdistributed training systems
