Auto provision GPU clusters

Analyze public repos, detect workload signals, and launch the right GPU cluster on Skypilot.

Idle

Repositories

0 repos

Public GitHub repositories only.

No repository selected

No zip selected

Source —

Repository link —

Select a repo to see its source.

Select a repo to auto-fill the GitHub URL for analysis.

Analyze a repo

GitHub URL

Scan phase

Ready

Files scanned

-

Last run

-

Detected signals

--

Workload

-

Framework

-

Model name

-

Model size

-

Entrypoint

-

Repo

-

Select an entrypoint below or override the command in the recommendation section.

Custom command

Scan details

Raw

Activity log

GPU recommendation

Run GPU planning after the scan, then adjust before provisioning.

Pending

Recommended plan

Auto

GPU class

-

Spot OK?

-

Est. cost/hr

-

Est. duration

-

Provider

-

Region

-

Instance

-

Top offers

GPU class Use spot instances Entrypoint command

Provisioning requires Skypilot credentials on the backend host.

Plan tuning

Optional

Job type Model size Target VRAM per GPU (GB) GPU count Max price/hr ($) Deadline (hours) Region preference Cost preference Spot preference Preferred GPU model (optional)

Overrides are applied during GPU planning. Leave blank for auto.

Skypilot YAML

Preview

name: gpunity-demo
resources:
  cloud: aws
  accelerators: A10G:1
  use_spot: true
setup: |
  pip install -r requirements.txt
run: |
  python main.py

Provisioning status

Monitor launch progress, logs, and cost.

Idle

Job details

--

Job ID

-

Cluster

-

SSH command

-

Status command

-

Logs command

-

Cost so far

-

Status updates are fetched from Skypilot.

Latest logs

Tail

-

Artifacts

None

No artifacts downloaded yet.