DEDUPED Hot AI repos, turned into use cases Get the Monday 5

How to run a GPU experiment on your lab server with a Voyage with Polaris

Let an experiment Voyage plan, code, smoke-test and iterate over SSH under budget caps.

Short answer: Add your GPU server through the Experiment Lab with your SSH credentials, which Polaris stores Fernet-encrypted per user; then start an experiment Voyage and answer the intake questions about the study; review the plan and the compute-budget check before approving any remote writes. All steps.

  • Research
  • Advanced
  • 2 hours
  • From the README
  • Any agent

Credentials. The run uses your SSH credentials to reach a real GPU server, so only add a server you are authorized to use and review the command allow and deny lists before starting.

Who it is forA researcher with a lab GPU server who wants an agent to run a study while they keep control of remote writes and spending.

The problemSetting up experiments, watching logs and fixing failures across many runs takes constant manual attention.

The outcomeA tracked experiment run with streamed logs, live metric curves, generated figures and an audit trail, with the agent asking you a question when it is stuck.

What you need first

  • A computer where the Polaris desktop app can run (the README says new releases are desktop-only)
  • A model source: an agent you already use, such as Claude Code, Codex or Gemini CLI, or a model API key
  • For experiments: SSH access to a GPU server and your own SSH credentials
  • A research direction or topic you want to explore

Steps

  1. Add your GPU server through the Experiment Lab with your SSH credentials, which Polaris stores Fernet-encrypted per user.
  2. Start an experiment Voyage and answer the intake questions about the study.
  3. Review the plan and the compute-budget check before approving any remote writes.
  4. Watch the smoke test, the streamed logs and the live metric curves in the console.
  5. Answer any question the run asks you, and review the generated figures and their VLM check when it finishes.

Prompt for your agent

Using the Polaris repo at ZJU-REAL/Polaris, start an experiment Voyage on the GPU server <server name> for the study <study description>. Expected output: a completed run with metrics, logs and figures. Verify that every remote write was approved through the gate and that the budget caps were not exceeded.

Caveat. The README says the agent repairs failures under a time budget rather than a fixed retry count, so a stuck run will pause and ask you rather than finish on its own.

Use this in your AI agent

Ask your agent

Using the Deduped MCP, get the setup steps for ZJU-REAL/Polaris and "Run a GPU experiment on your lab server with a Voyage"

Add Deduped to Claude Code once

claude mcp add --transport http deduped https://dedupedai.com/mcp

Cursor, ChatGPT, others

Share: X LinkedIn Hacker News Email Copy link

Third-party repos we don't own. AI-written use cases can be wrong. Disclaimer