Covalent Launch: The GPU Cheatcode: Fine-tune 20 Llama Models in 5 Minutes
AI Engineer World's Fair 2024 · 6:26
AI infrastructure and compute orchestration
Agnostiq develops Covalent, an open-source distributed computing platform that helps AI teams run workloads across cloud, on-premises, and hybrid infrastructure. Developers define dependencies and compute requirements in Python; Covalent provisions environments, allocates resources, and coordinates workflows involving multiple models or agents. DataRobot acquired Agnostiq and Covalent in 2025, bringing its compute orchestration into DataRobot’s AI application development platform.
Originally built for quantum computing and high-performance computing, Covalent extends that infrastructure approach to AI workloads through a modular architecture and plug-in ecosystem. It operates above systems such as Kubernetes and SLURM, coordinating execution across clusters and data centers through a single Python interface. Teams can assign different hardware to different agents and deploy inference endpoints without manually building container images. Oktay Goktas led Agnostiq as CEO at the acquisition; its parent, DataRobot, is led by CEO Debanjan Saha.
At the 2025 acquisition announcement, DataRobot reported a Covalent open-source community of 5,000 users and more than 140,000 downloads. Agnostiq had raised a $6.1 million seed extension led by Differential Ventures in 2023 to develop and commercialize its quantum and high-performance computing platform, including Covalent Cloud, its commercial offering.
AI Engineer World's Fair 2024 · 6:26
Affiliations reflect their AIE appearances, not necessarily current employment.
The walkthrough assigns GPU and CPU resources to different stages and discusses example evaluation costs, connecting resource choices to the work each stage performs.
Affiliations reflect each recorded session, not necessarily current employment.