← Toutes les offres

Thomson Reuters fournit des services d’information et des logiciels professionnels. Son portail Workday publie les recrutements de son entité à Zoug.

Repères sur Thomson Reuters
Domaine officiel
thomsonreuters.com
Offres ouvertes
9

Détails de l’offre

La description complète publiée par Thomson Reuters.

Job Description

Do you enjoy making frontier-scale training systems fast, reliable and correct? Thomson Reuters Labs is seeking a

Lead Research Engineer to architect and optimise our infrastructure for training agentic large language models (LLMs) with reinforcement learning (RL). This covers Mixture-of-Experts (MoE) models above 300 billion parameters, trained asynchronously at long context while interacting with tools and agent harnesses (the software environments in which models take actions).

Our focus areas include asynchronous RL that separates rollout generation from training, frontier-scale distributed training with Megatron-Core and PyTorch-native backends, bringing up new model architectures end to end, keeping training and inference numerically consistent, fault tolerance at scale, and audit-ready, reproducible training pipelines. You will work closely with Research Scientists so that new algorithms, rewards and environments can be tested quickly and at scale, and you will help set the engineering standards for the team.

Key Responsibilities

  • Lead technical direction: Own the architecture of our large-scale RL training stack and decide how it evolves for larger models, longer contexts and more complex agentic workloads.
  • Optimise: Profile the full RL loop (generation, tool calls, reward and judge computation, training and weight synchronisation) and cut time per step and GPU idle time.
  • Build: Design asynchronous, fault-tolerant pipelines in which models interact with tools and agent harnesses during training.
  • Debug the hard problems: Root-cause numerical, distributed and performance issues across the training framework, the inference engine and the model code.
  • Enable research: Work closely with Research Scientists so that new algorithms, rewards and environments can be tested quickly and at scale.
  • Mentor and raise the bar: Set standards for code quality, testing and reproducibility, and grow the engineers around you.

Required Qualifications

MSc or PhD in computer science, machine learning or a related field, or equivalent industry experience.

  • 3+ years of professional experience in ML engineering or research engineering on large-scale model training.
  • Extensive experience training large models on multi-node GPU clusters with PyTorch (Advanced) and Megatron-LM/Megatron-Core, FSDP/DTensor or DeepSpeed.
  • Deep understanding of parallelism strategies (tensor, pipeline, context and expert parallelism) and their memory and communication trade-offs, especially for MoE and long-context models.
  • Hands-on experience with LLM inference engines (vLLM, SGLang), including serving large MoE models.
  • Strong skills in GPU profiling and optimisation (memory, NCCL communication, CUDA graphs, kernel bottlenecks).
  • Strong software engineering practice and the ability to work fluently inside large open-source codebases.
  • Core competencies:
  • Technical leadership: Proven ability to lead complex technical projects end to end.
  • Mentorship: Experience mentoring engineers through code review and technical guidance, and growing others’ skills.
  • Root-cause problem solving: Diagnosing numerical, distributed and performance issues in complex systems.
  • Engineering rigour: Commitment to code quality, testing and reproducibility.
  • Collaboration: Working effectively with researchers and engineers to deliver at scale. Preferred

Qualifications

  • Experience : building or extending RL post-training frameworks (for example NeMo-RL, verl, OpenRLHF, TRL), especially asynchronous or non-colocated designs.
  • Experience : training or serving models above 100 billion parameters at 100k+ token contexts.
  • Experience : with distributed runtimes and orchestration (Ray, Kubernetes, Slurm) on AWS, Azure or Google Cloud.
  • Experience : with numerical-parity testing between training and inference implementations. - Significant contributions to vLLM, Megatron, PyTorch or similar open-source projects. - Publications or technical reports at ML or systems venues (for example MLSys, NeurIPS, OSDI). #LI-FP3 What’s in it For You? - Hybrid Work Model: We’ve adopted a flexible hybrid working environment for our office-based roles while delivering a seamless experience that is digitally and physically connected. - Flexibility & Work-Life Balance: Flex My Way is a set of supportive workplace policies designed to help manage personal and professional responsibilities, whether caring for family, giving back to the community, or finding time to refresh and reset.

This builds upon our flexible work arrangements, including work from anywhere for up to 8 weeks per year, empowering employees to achieve a better work-life balance.

  • Career Development and Growth: By fostering a culture of continuous learning and skill development, we prepare our talent to tackle tomorrow’s challenges and deliver real-world solutions. Our Grow My Way programming and skills-first approach ensures you have the tools and knowledge to grow, lead, and thrive in an AI-enabled future.
  • Industry Competitive

Benefits

We offer comprehensive benefit plans to include flexible vacation, two company-wide Mental Health Days off, access to the Headspace app, retirement savings, tuition reimbursement, employee incentive programs, and resources for mental, physical, and financial wellbeing. - Culture: Globally recognized, award-winning reputation for inclusion and belonging, flexibility, work-life balance, and more.

We live by our values: Obsess over our Customers, Compete to Win, Challenge (Y)our Thinking, Act Fast / Learn Fast, and Stronger Together. - Social Impact: Make an impact in your community with our Social Impact Institute. We offer employees two paid volunteer days off annually and opportunities to get involved with pro-bono consulting projects and Environmental, Social, and Governance (ESG) initiatives. - Making a Real-World Impact: We are one of the few companies globally that helps its customers pursue justice, truth, and transparency.

Together, with the professionals and institutions we serve, we help uphold the rule of law, turn the wheels of commerce, catch bad actors, report the facts, and provide trusted, unbiased information to people all over the world. DISCLAIMER The above information in this description has been designed to indicate the general nature and level of work performed by employees within this classification. It is not designed to contain or be interpreted as a comprehensive inventory of all duties, responsibilities, and qualifications required of employees assigned to this job.

About Us

Thomson Reuters informs the way forward by bringing together the trusted content and technology that people and organizations need to make the right decisions. We serve professionals across legal, tax, accounting, compliance, government, and media. Our products combine highly specialized software and insights to empower professionals with the data, intelligence, and solutions needed to make informed decisions, and to help institutions in their pursuit of justice, truth, and transparency. Reuters, part of Thomson Reuters, is a world leading provider of trusted journalism and news.

We are powered by the talents of 26,000 employees across more than 70 countries, where everyone has a chance to contribute and grow professionally in flexible work environments. At a time when objectivity, accuracy, fairness, and transparency are under attack, we consider it our duty to pursue them. Sound exciting? Join us and help shape the industries that move society forward. As a global business, we rely on the unique backgrounds, perspectives, and experiences of all employees to deliver on our business goals.

To ensure we can do that, we seek talented, qualified employees in all our operations around the world regardless of race, color, sex/gender, including pregnancy, gender identity and expression, national origin, religion, sexual orientation, disability, age, marital status, citizen status, veteran status, or any other protected classification under applicable law. Thomson Reuters is proud to be an Equal Employment Opportunity Employer providing a drug-free workplace.

We also make reasonable accommodations for qualified individuals with disabilities and for sincerely held religious beliefs in accordance with applicable law. More information on requesting an accommodation here. Learn more on how to protect yourself from fraudulent job postings here. More information about Thomson Reuters can be found on thomsonreuters.com

Prérequis

  • Master ou doctorat en informatique, apprentissage automatique ou domaine connexe, ou expérience professionnelle équivale
  • Au moins 3 ans d’expérience professionnelle en ingénierie ML ou recherche appliquée à l’entraînement de grands modèles
  • Expérience approfondie de l’entraînement de grands modèles sur des clusters GPU multi-nœuds
  • Maîtrise avancée de PyTorch
  • Expérience de Megatron-LM/Megatron-Core, FSDP/DTensor ou DeepSpeed
  • Compréhension approfondie des stratégies de parallélisme
  • Expérience pratique des moteurs d’inférence LLM
  • Compétences en profilage et optimisation GPU
  • Expérience de frameworks de post-entraînement RL
  • Expérience de l’entraînement ou du service de modèles de plus de 100 milliards de paramètres avec des contextes de plus

Avantages mentionnés

  • Télétravail hybride
  • Jusqu’à 8 semaines de travail depuis n’importe où par an
  • Congés flexibles
  • Deux jours de santé mentale offerts
  • Application Headspace
  • Épargne retraite
  • Remboursement des frais de scolarité
  • Programmes d’incitation pour les employés
  • Deux jours de bénévolat rémunérés par an

3× plus de chances d’entretien avec un cold email

Candidater à cette offre

Comment candidater ?

L’offre est publiée sur le site de Thomson Reuters. Ouvrez l’offre originale ↗ pour y déposer votre candidature. Screeningpass ne remplit pas de formulaire à votre place.

À qui écrire chez Thomson Reuters pour ce poste ?

Lancez une recherche de contact depuis cette page. Screeningpass vous propose jusqu’à trois personnes liées au poste. Vous choisissez celle à qui écrire, puis Screeningpass cherche son email professionnel et prépare un message. Vous le relisez et l’envoyez vous-même avec Gmail. Une adresse email n’est pas toujours disponible, et une réponse n’est jamais garantie.

Combien coûte la recherche d’un contact ?

Une recherche coûte 1 crédit. Le crédit vous est rendu si la recherche échoue. L’inscription offre 3 crédits, soit 3 recherches.

À quoi sert Screeningpass ?

Screeningpass rassemble des offres de stage, d’alternance et d’emploi publiées par les entreprises. Pour une offre, Screeningpass trouve une personne à contacter dans l’entreprise, cherche son email professionnel et prépare un message à partir de votre CV. Voir le fonctionnement en détail.