Cohere · Paris
OtherModelingRemoteFullTimeposted 12mo ago
Who are we?
Cohere is the leading security-first enterprise AI company. We build cutting-edge foundation AI models and end-to-end products that are designed to solve real-world business problems.
We’re training and deploying frontier models for enterprises who are building AI systems. We believe that our work is instrumental to the widespread adoption of AI and we are looking for folks that want to be part of that.
We obsess over what we build. Each one of us is responsible for contributing to increasing the capabilities of our models and the value they drive for our customers. Cohere is a team of researchers, engineers, designers, and more, who are all passionate about their craft.
We are a global technology company headquartered in Toronto with key offices in London, New York City, San Francisco, Montreal, Paris, Berlin and Seoul. Join us!
Role Overview:
The integration team is responsible for developing and scaling machine learning algorithms and infrastructure for LLM post-training, with a focus on large-scale, distributed RL methods. We strive for excellence in both engineering and science by meticulously designing experiments and design docs. While tasks are assigned according to everyone’s expertise, there is a global team effort to write production code and support the team research efforts, depending on individual interests and organizational needs.
In particular, this role aims to enhance the global quality of the post-training codebase by implementing new tools to ease and support research, optimizing post-training algorithms, and scaling distributed RL to unprecedented levels.
Please Note: We have offices in London, Paris, Toronto, San Francisco, New York but we are also remote-friendly! Applicants for this role may work anywhere between UTC−06:00 and UTC+01:00.
Key Responsibilities:
Design and write high-performing and scalable software for training models.
Develop new tools to support and accelerate research and LLM training.
Coordinate with other engineering teams (Infrastructure, Efficiency, Serving) and the scientific teams (Agent, Multimodal, Multilingual, etc.) to create a strong and integrated post-training ecosystem.
Craft and implement techniques to improve performance and speed up our training cycles, both on SFT, offline preference, and the RL regime.
Research, implement, and experiment with ideas on our cluster and data infrastructure.
Collaborate, Collaborate, and Collaborate with other scientists, engineers, and teams!
Qualifications:
Extremely strong software engineering skills.
Value test-driven development methods, clean code, and strive to reduce technical debts at all levels.
Proficiency in Python and related ML frameworks such as JAX, Pytorch and/or XLA/MLIR.
Experience using and debugging large-scale distributed training strategies (memory/speed profiling).
[Bonus] Experience with distributed training infrastructures (Kubernetes) and associated frameworks (Ray).
[Bonus] Hands-on experience with the post-training phase of model training, with a strong emphasis on scalability and performance.
[Bonus] Experience in ML, LLM and RL academic research.
This role is perfect for you if you:
Have a deep passion for quality work.
Enjoy tuning and optimising large LLM models.
Comfortable working with people with different levels of software engineering skills, from beginner to more advanced.
Comfortable diving into complex ML codebases to identify and resolve issues, ensuring the smooth operation of our systems.
Thrive in a fast-paced, technically challenging environment, where you can contribute your innovative ideas and solutions.
Working Location:
This role can be based remotely or from one of our office locations listed on the job description - there is no minimum in-office qualification requirement. We care most about hiring exceptional people regardless of locations, though please check the location listed on the posting for guidance around the core time zone or working hours alignment expected for the role.
A weekly lunch stipend of $75/£75 or equivalent in your local currency for lunch.
Full health and dental benefits, including a separate budget for mental health.
RRSP matching, 401K, Pension Scheme.
100% Parental Leave top-up for up to 6 months, for either parent.
Annual enrichment benefits:
Arts & culture, fitness/wellness, quality time, and a workspace improvement credit.
Education & learning stipend for conferences, courses, and coaching.
6 weeks of paid vacation (30 working days!)
Budget for traveling to other offices if you are remote, plus an annual company offsite.
Cohere is remote-friendly, but we also have offices in Toronto, London, New York City, San Francisco, Montreal, Paris, Berlin and Seoul with more opening soon.
For those in the office: a daily lunch program, plenty of snacks, and regular community and social events.
For those not near an office: a co-working benefit so you can work alongside others in your city.
Everyone receives a $500 home office stipend to set up your workspace properly.
If any of the above doesn’t line up exactly with your experience, we still encourage you to apply.
We strive to create an inclusive work environment for all; we welcome applicants from all backgrounds and are committed to providing equal opportunities. Should you require any accommodations during the recruitment process, please submit an Accommodations Request Form, and we will work together to meet your needs.
We may use AI-enabled tools to screen and assess applicants against the criteria for this position. This helps our recruiters identify potentially qualified candidates, but it doesn't limit the applications our recruiters may review or consider.
Beware of Scams: Cohere will never ask for payment or third-party services (e.g., CV writing) as part of our hiring process. All legitimate roles are listed on the Cohere careers page and LinkedIn only, with all communications from Cohere employees coming from an @cohere.com or @cw.cohere email alias. If jobs are viewed on other sites then please verify these through our official careers page.