Member of Technical Staff, Post-training

Cohere Cohere · AI Frontier · London, United Kingdom · Modeling

This role focuses on advancing model post-training techniques to achieve state-of-the-art performance, bridging research and production. Responsibilities include designing and writing high-performance software for training, coordinating with specialist teams, crafting techniques for SFT and RL regimes, and researching/implementing ideas on infrastructure. The role requires strong software engineering skills, proficiency in Python and ML frameworks, and experience with distributed training infrastructures and large-scale model training, particularly in the post-training phase with performance optimization.

What you'd actually do

  1. Design and write high-performant and scalable software for training models.
  2. Consistently post-train the models to reach SOTA level performance.
  3. Coordinate with other specialist teams (Agentic, Code…) to produce models that have strong all encompassing performance.
  4. Craft and implement techniques to improve the performance and results of our training cycles both on the SFT and the RL regime.
  5. Research, implement, and experiment with ideas on our supercompute and data infrastructure.

Skills

Required

  • Python
  • JAX
  • Pytorch
  • XLA/MLIR
  • distributed training infrastructures
  • Kubernetes
  • Slurm
  • Ray
  • large-scale distributed training strategies
  • training large model at scale
  • post training phase of model training
  • performance optimisation

Nice to have

  • paper at top-tier venues (such as NeurIPS, ICML, ICLR, AIStats, MLSys, JMLR, AAAI, Nature, COLING, ACL, EMNLP)

What the JD emphasized

  • Extremely strong software engineering skills
  • hands on experience on training large model at scale
  • hands on experience with the post training phase of model training, with a strong emphasis on performance optimisation

Other signals

  • post-training
  • SOTA performance
  • distributed training
  • production code