Llm Training Researcher - Cto Office

Who we are:
Bloomberg's CTO Office is the future-looking technical and product arm of Bloomberg L.P. We envision, design, and prototype the next generation infrastructure, hardware, and applications for the Bloomberg Terminal. Our projects include machine learning-powered products, cloud computing infrastructure and strategy, open source stewardship, natural language processing, and more. We are passionate about what we do.
Overview:
We believe that large language models (LLMs) will play a very important role at Bloomberg for the foreseeable future. As an LLM Training Researcher, you will be our resident expert on cutting-edge training strategies for LLMs. Specifically, your focus will be on optimization (e.g. training stability, optimizer choice, learning rate schedules, regularization, batching, float representations), details of architecture choice and objective functions, data-preprocessing (e.g. tokenization and de-duplication strategies), fine-tuning strategies and anything else that would have a material impact on model performance. You will lead our research into these areas, and you will collaborate with our talented ML researchers in the Engineering division on building LLMs. You may also develop and leverage collaborations and partnerships with external service providers, where they can be used to substantially accelerate our progress.
The ideal candidate will have practical experience training state-of-the-art LLM models as well as a strong publication record in the field. They will have excellent collaboration skills that will enable them to work with stakeholders across the organization and responsively address emerging business challenges. They are expected to serve as a representative for the office of the CTO internally to engineering teams and externally to academic partners and industry counterparts. Bloomberg values our deep engagement in academic conferences and in the open-source community, and an ideal candidate would lead our efforts in both spaces. An ideal candidate will thrive in an environment where they can develop a strategic vision and see it realized through their technical expertise, focus, and partnership across the organization.
We'll trust you to:

• Form strategic partnerships with external providers to accelerate our technological advancement
• Lead research efforts on LLM training, deliver scientific breakthroughs and share these findings at top conferences and through other publication channels
• Collaborate with engineers on building state-of-the-art LLM models
• Establish and continually refine our internal best practices for LLM training

You'll need to have:

• Several publications on LLMs or related technologies (e.g. underlying optimization theory)
• Excellent communication and collaborative skills
• 5+ years of experience in an applied research role
• Ph.D. or equivalent experience in training LLMs
• Blend of long-term strategic vision with goal-driven tactical focus
• Engagement with the research community (e.g., conference or workshop service).

Nice to have:

• Understanding of parallelization strategies for large-scale LLM training

Bloomberg is an equal opportunity employer and we value diversity at our company. We do not discriminate on the basis of age, ancestry, color, gender identity or expression, genetic predisposition or carrier status, marital status, national or ethnic origin, race, religion or belief, sex, sexual orientation, sexual and other reproductive health decisions, parental or caring status, physical or mental disability, pregnancy or parental leave, protected veteran status, status as a victim of domestic violence, or any other classification protected by applicable law.
Bloomberg is a disability inclusive employer. Please let us know if you require any reasonable adjustments to be made for the recruitment process. If you would prefer to discuss this confidentially, please email [email protected]