Apertus Engineer: Post-training

ETH Zurich seeks an engineer to support the Apertus project, managing post-training and reinforcement learning workflows on the Alps supercomputer.

This position supports the Apertus initiative, which trains large open foundation models on the Alps supercomputing infrastructure at the Swiss National Supercomputing Centre. The role focuses on maintaining the engineering environment for large-scale AI experiments and collaborating with researchers to improve model performance and reliability.

Responsibilities

• Construct and upkeep containerized setups for large language model post-training and reinforcement learning tasks.
• Configure containers and dependencies to function effectively on the Alps and CSCS infrastructure.
• Execute and oversee Slurm-managed training and evaluation jobs.
• Troubleshoot issues related to distributed execution, checkpointing, filesystems, networking, and GPU usage.
• Maintain documentation, launch scripts, and configuration files to ensure training reproducibility.
• Collaborate with researchers and CSCS engineers to enhance the stability and speed of large experiments.
• Support workflows for supervised fine-tuning, preference optimization, and reinforcement learning.
• Develop and run reinforcement learning environments for tasks with verifiable outcomes, such as coding, math, and reasoning.
• Implement reward modeling, calibration, and verifier-based training processes.
• Create and validate synthetic or gym-based training tasks.
• Conduct ablation studies to compare algorithms, data mixtures, hyperparameters, and infrastructure settings.
• Evaluate model behavior across various benchmarks, including reasoning, coding, multilingual capabilities, and safety.

Requirements

• MSc or PhD in Computer Science, Data Science, AI, Machine Learning, or a related field.
• Exceptional BSc holders with strong engineering experience may also be considered.
• Solid experience with AI and neural network architectures.
• Strong collaboration and communication skills, with the ability to bridge research and engineering teams.
• Prior hands-on experience in the core domains of this role is mandatory.
• High flexibility to adapt to shifting priorities, tools, and tasks driven by training schedules.
• Practical experience with LLM post-training, including alignment techniques like SFT or reinforcement learning.

Nice to have

• Understanding of distributed training concepts such as data, tensor, and pipeline parallelism, as well as checkpointing and GPU communication.
• Experience using Slurm or similar HPC workload managers.
• Background in building or adapting containers for HPC or GPU clusters.
• Published research or familiarity with recent studies in relevant domains.
• Experience creating verifiable tasks for mathematics, code, reasoning, or tool use.
• Familiarity with lower-level libraries like NCCL, Transformer Engine, FlashAttention, or communication backends.
• Experience with large-scale evaluation pipelines.

What the company offers

• A stimulating academic environment at a leading technical university.
• Access to state-of-the-art supercomputing infrastructure and cutting-edge AI research.
• Collaboration with top researchers and engineers from EPFL, ETH Zurich, CSCS, and other Swiss institutions.
• Flexible working arrangements, including remote work options.
• Professional development opportunities, including conference attendance and specialized training.
• The chance to contribute to open-source projects with global impact.
• Access to the broader Swiss academic ecosystem and industry partnerships.
• Being part of Switzerland's sovereign AI development, working on technology with national significance.

About the company

The team works with over thirty academic collaborators to deliver fully open, responsibly trained, multilingual, and multimodal AI models for research and industry.

• Collaboration with over thirty academic partners.
• Commitment to open-source and responsible AI development.
• Focus on multilingual and multimodal model capabilities.