Skip to main content

Shared campus infrastructure

DSMLP and DataHub

One shared computing foundation

Use DataHub to reach the DSMLP computing platform

DataHub is UC San Diego’s browser-based way to use DSMLP for coursework, formal independent study, eligible student projects, and selected TritonAI workloads.

DSMLP combines managed containers, CPU and GPU resources, storage, and reusable software environments. DataHub provides Jupyter and RStudio in the browser, while command-line tools support advanced machine-learning work.

Open DataHub Read the official FAQ

How the pieces fit

DataHubWeb access to Jupyter, RStudio, and course environments
Command lineAdvanced container and batch workflows through dsmlp-login

Shared computing platform

DSMLP

  • Containers
  • CPU and GPU
  • Storage
  • Datasets
Coursework and student projectsSelected TritonAI workloads
People enter through DataHub or command-line tools, launch isolated workspaces on DSMLP shared compute, and use those resources for coursework, formal independent study, eligible student projects, and selected TritonAI workloads.

Why shared infrastructure matters

Start supported, grow into advanced work

Eligible students and instructors get an environment that already works for computational coursework, and can move to custom tooling when they outgrow it.

Access for assigned courses

Students in assigned courses receive access automatically through the course setup process.

A consistent setup

Everyone in the class gets the same software, course files, and datasets, so nobody loses a week to installation.

Support for instructors

ITS and Educational Technology Services help instructors set up software, containers, assignments, and course datasets.

Platform capabilities

Start in a notebook or build a custom ML workflow

DSMLP turns a notebook or command-line request into an isolated workspace with the computing resources it needs.

Interactive notebooks

Jupyter puts live code, equations, charts, and your own notes in one document.

Accelerated ML development

Research-class CPU/GPU resources and an Ubuntu CUDA environment support popular languages and GPU-enabled frameworks.

Flexible environments

When the default image does not fit, add packages or launch your own Docker container.

Storage and datasets

Cluster-local storage supports student workspaces, course files, and commonly used training datasets across sessions.

Getting in

Web or command line

Both reach the same DSMLP. Use the browser for a workspace that is already set up, or the command line when you want control.

Browser-based

DataHub

  • Launch a course or independent-study environment from the web.
  • Work in Jupyter notebooks, terminals, and supported graphical tools.
  • Use curated software images and shared course resources.

Go to DataHub

Command-line

DSMLP launch tools

  • Start interactive, batch, or custom-container workloads with launch.sh.
  • Request CPU, memory, and GPU resources for the container.
  • Use advanced workflows without running jobs on the shared login host.

Read the container launch guide

Eligibility and access

How students and instructors get access

  1. Enrolled courses

    Students in courses assigned to DSMLP receive access automatically shortly before the term begins.

  2. Independent study and eligible projects

    Students submit the DSMLP access request with the project scope and computational needs.

  3. Instructional course setup

    Instructors request DataHub/DSMLP through the Specialized Instructional Computing form.

Students exploring DSMLP for personal enrichment, and faculty or staff evaluating it for instructional support, can contact datahub@ucsd.edu.

Where TritonAI fits

How the two differ

DSMLP supplies computing. The TritonAI gateway supplies model access. Some TritonAI services run workloads on DSMLP, but the apps, their service owners, and the approved model routes stay separate from it.

  1. Shared computeDSMLP

    Containerized CPU/GPU capacity, storage, and platform operations

  2. Selected workloadsTritonAI services

    Applications, evaluation, automation, and supporting processes

  3. Approved model accessTritonAI LLM Gateway

    Managed routes to enterprise cloud and SDSC-hosted models

DSMLP provides shared compute for selected TritonAI workloads. Model requests continue through the TritonAI LLM Gateway to approved enterprise or SDSC-hosted model routes.

Related, but not the same service: DataHub is the web interface for authorized notebook and computational work. TritonAI users work through TritonGPT, the TritonAI Harness, APIs, or department applications. See the TritonAI gateway architecture.

Use the right operating path

Know which work belongs on DSMLP

DSMLP is limited to student-focused activities with authorized access and published usage policies. Eligible uses include assigned coursework, formal independent study and student research, state-supported capstone projects, and approved campus-sponsored student activities. Self-supported professional degree coursework and capstone projects are not currently eligible.

Non-course or non-credit research computing should follow the UC San Diego Research Cluster request path. UC San Diego documentation also states that DataHub/DSMLP is not suitable for P3/P4 protected data. Projects involving sensitive data, production service commitments, or specialized operating requirements need an appropriate review and hosting path.

DSMLP is built and operated by UC San Diego IT Services, with additional financial contributions from Cognitive Science and the Jacobs School of Engineering.