Member of Technical Staff, Post-Training

Why this role?

Advance the state of the art for model post‑training, ship state‑of‑the‑art models to production, and bridge the gap between research and production. We have one of the highest compute‑to‑engineer ratios in the world, and we do not delineate strongly between engineering and research. Everyone writes production code and supports our research effort depending on interest and organisational needs. With all the compute, data, and talent available, you will be empowered to do your best work.

Please note: We have offices in London, Paris, Toronto, San Francisco, and New York, and we also embrace being remote‑friendly.

Responsibilities

• Design and write high‑performant, scalable software for training models.
• Consistently post‑train the models to reach SOTA level performance.
• Coordinate with other specialist teams (Agentic, Code…) to produce models that have strong, all‑encompassing performance.
• Craft and implement techniques to improve the performance and results of our training cycles both on the SFT and the RL regime.
• Research, implement, and experiment with ideas on our super‑compute and data infrastructure.
• Learn from and work with the best researchers in the field.

Skills & Experience

• Extremely strong software engineering skills.
• Proficiency in Python and related ML frameworks such as JAX, PyTorch, and XLA/MLIR.
• Experience with distributed training infrastructures (Kubernetes, Slurm) and associated frameworks (Ray).
• Experience using large‑scale distributed training strategies.
• Hands‑on experience training large models at scale.
• Hands‑on experience with the post‑training phase of model training, with a strong emphasis on performance optimisation.
• Bonus: papers at top‑tier venues (such as NeurIPS, ICML, ICLR, AIStats, MLSys, JMLR, AAAI, Nature, COLING, ACL, EMNLP).

These are neither an exhaustive nor a necessary set of attributes. Even if none of these apply to you, but you believe you will contribute to Cohere, please reach out.