Local AI Models on Sagehen
Run large language models on Pomona’s own hardware. Same kind of AI you know from commercial tools, with one crucial difference: your data never leaves campus.
Why run AI locally?
Public AI services are convenient, but everything you type leaves campus. Local models change the rules.
Your data stays here
Prompts, documents, and outputs never touch a vendor’s servers and are never used to train anyone’s product. The model runs on Sagehen; the data stays on Sagehen.
Works with research data
Because nothing leaves the cluster, unpublished PROPRIETARY research data can be used with local models, and RESTRICTED data can be arranged under additional controls. Neither is ever appropriate in a public AI tool.
No meter running
No subscriptions, no per-use charges, no rate limits from a vendor. Analyze a million documents if your research needs it.
A bonus for your methods section: commercial AI services change their models without notice. A local model is pinned to an exact version, so your analysis can be rerun and reproduced, which reviewers increasingly ask about.
What you can do
Use open models
We host open-weight model families (Llama, Mistral, Gemma, and successors) for summarization, text analysis, coding assistance, translation, and classification at research scale.
Fine-tune on your data
Adapt a model to your corpus, your domain vocabulary, or your classification task, using the cluster’s GPUs and your data, privately.
Build it into pipelines
Combine models with your analysis code in scripted, repeatable workflows: batch-process interview transcripts, historical texts, code, or field notes.
How it works
Dedicated AI hardware
A generative AI node on Sagehen, purpose-built for model inference and fine-tuning, alongside the cluster’s other GPUs.
Always-on endpoints
Standing model endpoints for interactive use, plus scheduled jobs for heavier fine-tuning and batch work.
Course endpoints
Class-scoped AI endpoints let a whole course experiment safely, an approach piloted in the humanities here at Pomona.
Not just for STEM
Some of the most creative local AI work at Pomona has come from the humanities: a French literature course used text mining and generative models to explore the Surrealism centenary, and the same infrastructure supports biologists analyzing sequence data and social scientists coding interview corpora. If you can describe the task, we can usually find the model.
The rules of the road
Local AI use is governed by the HPC & Local AI Usage Policy, which maps what’s allowed to your data’s classification tier. Working with human subjects data? Pomona’s IRB-aligned AI research guidelines cover consent language and when an addendum is needed; start with a conversation with us.
Still using public AI tools?
That’s fine for public information: just verify outputs, never trust generated citations without checking, and never paste unpublished or participant data into them. When in doubt about which side of the line your data falls on, ask us.
Learn with us
Responsible Use of AI Tools on HPC
Data handling, disclosure, and reproducibility when AI enters your research workflow.
Introduction to Machine Learning on HPC
Hands-on, from first principles to training models on Sagehen’s GPUs.
Curious what a local model could do for your research?
Bring us the task; we’ll bring a model, a plan for your data, and a path from experiment to workflow.
Request a Consultation
its-hpc@pomona.edu
Last reviewed: August 2026