/
AI & Machine Learning

AI & Machine Learning

The Center for Computational Sciences (CCS) supports artificial intelligence and machine learning on its research computing clusters. This covers two quite different kinds of AI, plus running models on your own machine. Use the sections below to find the one that fits your work.

For an up-to-date, searchable list of software available on this cluster, please use the Software Documentation Service (SDS).

1. Research & Scientific Models

These are AI models that solve scientific problems, they take scientific input (for example a protein sequence) and produce a scientific result (a predicted 3D structure, an embedding, a docked ligand pose). They run as batch GPU jobs: submit a job, get results, release the GPU.

CCS installs each model into a container and pre-stages its weights and reference databases in shared storage, so everything is ready to go and you only run a SLURM script. Currently available:

  • Protein structure prediction: AlphaFold, Boltz, ColabFold, RoseTTAFold, and ESMFold2.

  • Protein language models: ESM / ESMC embeddings and sparse-autoencoder (SAE) interpretability features.

  • Molecular docking: DiffDock (protein-ligand binding).

2. LLM Inference & Chat

These are general-purpose large language models (the chat / assistant / agent kind). CCS lets you run open LLMs on the cluster GPUs behind a standard OpenAI-compatible API, use them inside Jupyter notebooks, or run smaller models on your own laptop or desktop:

  • Ollama - the easy, interactive option: start a server in a GPU job, swap models freely, good for prototyping and single-user chat. Runs on any LCC GPU.

  • vLLM - the high-throughput option: for serving applications, many concurrent requests, very large models, and tool-calling / vision / embeddings. Uses the A100 / H200 GPUs.

  • Jupyter AI - use LLMs directly inside notebooks on MCC and LCC.

  • Run Local LLMs on Your Laptop / Desktop - no cluster account needed.

Which do I need?

If you are predicting a structure, generating embeddings, or docking a molecule, use Research & Scientific Models. If you want to chat with, prompt, or build an application on a language model, use LLM Inference & Chat. For access or to request a model, open a ticket at the CCS help desk.

 

Center for Computational Sciences