Member of Technical Staff - Post-Training (Reinforcement Learning & General)

CohereEnterprise AI

Location: Europe; Secondary Locations: New York, Toronto. This role can be based remotely or from one of our office locations listed on the job description., Please check the location listed on the posting for guidance around the core time zone or working hours alignment expected for the role.Full-Time

SalaryCA$295K - CA$535K; CA$295K – CA$535K • Offers Equity • Multiple Ranges; Canada - Toronto (GTA), Montreal: Base Salary CA$295K – CA$535K • Offers Equity; USA - New York, California and Wasghington States: Base Salary $205K – $380K • Offers Equity; USA - All Other States: Base Salary $175K – $325K • Offers Equity

Apply NowOpens the employer's application page

More jobs in this role

Job Details

Required Skills: PyTorchLLM

Requirements

Bring strong research and software engineering skills.
Have hands-on experience training large language models.
Have experience applying reinforcement learning to LLMs, such as PPO, GRPO, RLHF, or RLVR, ideally at scale.
Have a track record of work on generalization, robustness, or distribution shift through publications or shipped models.
Practice rigorous experimentation, including clean ablations, careful controls, and sound conclusions from noisy results.
Bring a deep understanding of data quality principles and best practices.
Be willing to investigate data in depth and assess dataset quality and how data characteristics affect model training outcomes.
Experience with distributed training, such as JAX, PyTorch, FSDP, or Megatron, is a plus.

Responsibilities

Design, run, and analyze large-scale RL post-training experiments, including RLHF and RL from verifiable rewards.
Develop methods to improve out-of-distribution generalization, including reward design, environment and task diversity, curricula, and regularization.
Build held-out evaluations for generalization to unseen tasks, domains, and formats, and to detect reward hacking and benchmark overfitting.
Diagnose training dynamics and failure modes, including instability, reward over-optimization, and capability regressions.
Turn successful research into robust, reproducible training recipes for production models.
Work closely with post-training data infrastructure, evaluation, and pre-training teams.

View Full Description & ApplyYou'll be redirected to the employer's site