Staff ML Engineer, Post-Training RL for LLMs

Cohere; España; Remote; EUR 186,000 - 336,000; Full time

Benefits offered by this job

Lunch stipend

Health and dental benefits

RRSP matching / 401K / Pension

Parental leave top-up

Enrichment benefits

Education stipend

Vacation (6 weeks)

Office travel budget

Home office stipend

Job summary

Cohere is seeking a Staff Scientist to advance RL post-training for large language models, designing training recipes and evaluation suites while collaborating with data, infra, and pre-training teams to push generalization beyond training distributions.

This remote-friendly role emphasizes publishing results, collaborating across offices, and helping production teams deploy robust models, with a strong benefits package and a culture that values research impact.

Qualifications

• Strong research and software engineering skills, with hands-on experience training LLMs.
• Experience applying reinforcement learning to LLMs, ideally at scale.
• Track record in generalization, robustness, or distribution shift.
• Rigorous experimental practice with clean ablations and controls.
• Experience with distributed training (JAX, PyTorch, FSDP, Megatron) is a plus.
• Excellent communication skills.
• Deep understanding of data quality principles and its impact on model training.

Responsibilities

• Design, run, and analyze large-scale RL post-training experiments.
• Research and develop methods to improve generalization and robustness.
• Build held-out evaluations that measure generalization and catch reward hacking.
• Diagnose training dynamics and failure modes, including instability.
• Turn successful research into robust, reproducible training recipes for production models.

Skills

LLMs training

RL for LLMs

Generalization

Experimental rigor

Distributed training

Communication

Data quality

Tools

JAX

PyTorch

FSDP

Megatron

Job description

Cohere is seeking a Staff Scientist to advance RL post-training for large language models, designing training recipes and evaluation suites while collaborating with data, infra, and pre-training teams to push generalization beyond training distributions.

This remote-friendly role emphasizes publishing results, collaborating across offices, and helping production teams deploy robust models, with a strong benefits package and a culture that values research impact.