The ALCF Inference Service provides researchers with access to a wide range of AI models running on dedicated ALCF systems. By offering AI inference as a shared resource, the service enables researchers to analyze data, test new ideas, and incorporate AI into scientific workflows without deploying or managing models themselves.
Researchers can use the service for a variety of tasks, including:
- Analyzing large datasets from simulations, experiments, and observational studies
- Extracting insights from scientific publications and other documents
- Generating and refining code
- Building AI-enabled applications and workflows
- Supporting retrieval-augmented and agentic systems
- Accelerating hypothesis generation and data interpretation
Models
Users have access to a collection of open-source models, including Google’s Gemma series, Meta’s LLaMA models, and OpenAI’s GPT-OSS family, as well as domain-specific foundation models, computer vision models, and in-house models developed at Argonne, such as AuroraGPT.
Systems
The ALCF Inference Service runs on dedicated systems, including Sophia, an NVIDIA DGX A100 cluster, and Metis, a SambaNova platform optimized for high-throughput inference workloads.
Additional NVIDIA-based systems, including Minerva and Tara, will expand support for the service in the future. Minerva is equipped with 64 NVIDIA Blackwell GPUs and interconnected with the NVIDIA Quantum-2 InfiniBand platform. Tara features 2,688 NVIDIA GH200 Grace Hopper GPUs.
Access
Researchers can access the ALCF Inference Service through web-based interfaces and OpenAI-compatible APIs. Users with Argonne or ALCF credentials can currently authenticate via Globus. This service is available to users from other DOE laboratories on request.
This service provides API access to a variety of state-of-the-art open-source models running on dedicated ALCF hardware.
Visit inference.alcf.anl.gov to get started.
Learn More
If you’d like to find out more about how your research can benefit from ALCF’s resources, please reach out to [email protected].
