Apertus Engineer: Deployment

Opportunity Overview

We are in search of an experienced engineer to lead the technical release path of Apertus models. This role is crucial in integrating Apertus within the open-source inference ecosystem, ensuring functionality and compatibility with the community's needs. The position requires proficiency in Python and software engineering, alongside hands-on experience in LLM inference stacks and contributions to open-source projects.

Project Background

The initiative focuses on training open foundation models containing hundreds of billions of parameters across one of Europe's largest AI-ready supercomputers. Our collaborative team includes over a dozen engineers and leading researchers from EPFL and ETH Zürich. We have successfully released the Apertus 1 and Apertus 1.5 models and are actively engaged with more than thirty academic collaborators to produce open-source, responsibly trained multilingual and multimodal AI models suitable for both research and industry applications.

The role operates at the intersection of the training and open-source community, with the community manager addressing social engagement while technical aspects are handled through this position.

Key Responsibilities

• Upstream Integration and Release Engineering:
  • Oversee the technical release path for trained Apertus models by collaborating with the training team for checkpoint conversion and the preparation of release artifacts.
  • Implement and submit support for Apertus model architectures in community libraries to ensure seamless integration and support.
  • Confirm compatibility of new releases with major inference engines and model formats prior to launch.
  • Coordinate the timing of releases and technical documentation with the community manager.
• Quantisation:
  • Create quantised versions of released models for server and personal deployment, ensuring quality through rigorous evaluation benchmarks.
• Documentation and Examples:
  • Develop example scripts and reference configurations for using Apertus models with various frameworks.
  • Support community usage through detailed deployment documentation and troubleshooting guides.

Candidate Profile

Essential Qualifications:

• MSc or PhD in Computer Science, Data Science, Artificial Intelligence, Machine Learning, or related fields. Exceptional BSc candidates with significant engineering experience will also be considered.
• Strong skills in Python and software engineering, including familiarity with open-source contribution workflows.
• Experience with LLM inference stacks such as Hugging Face Transformers, vLLM, or SGLang.
• Excellent collaboration and communication skills, enabling effective work with research and engineering teams.
• Hands-on experience in relevant domains is essential, whether through project work or academic study.
• A flexible mindset adaptable to changing priorities and fast-paced environments.
• Proven track record of contributions to ML or inference libraries.

Preferred Qualifications:

• Experience in converting models across various formats and frameworks.
• Familiarity with local deployment tools and developer documentation.

Desirable Qualifications:

• Published research or a comprehensive understanding of recent developments in the field.
• Experience in quantising models and understanding quality assessment for quantised architectures.
• Knowledge of GPU performance optimization and large-scale serving techniques.

Workplace Environment

This role is embedded within a vibrant academic atmosphere, characterized by collaboration with some of the world's foremost researchers in technology and science. Employees benefit from attractive working conditions, comprehensive benefits, and ample professional development opportunities.

Diversity and Sustainability

We prioritize an inclusive culture and promote stability by valuing diversity and ensuring an environment where every individual is respected. Sustainability is a fundamental principle guiding our operations, as we actively work towards achieving a climate-neutral future.

Curiosity-Driven Work

We encourage engaging with groundbreaking research and development, contributing to projects of global significance and aligning with Switzerland's commitment to advancing technology for societal benefit.