---

name: voltage-park description: Provision and manage Voltage Park H100 GPU instances. Use when the user needs to spin up H100s, SSH into VP instances, transfer files, or terminate cloud GPU instances.

Voltage Park H100 Management

Permissions

• NEVER provision without explicit permission
• ASK before spinning up H100s when blocked on GPU tasks
• Do NOT auto-terminate - user may run multi-hour experiments

API

Token: $VP_API_TOKEN in ~/.env_vars Base: https://cloud-api.voltagepark.com/api/v1

Provision body: {"config_id":"X","location_id":"X","organization_ssh_keys":{"mode":"selective","ssh_key_ids":["X"]},"name":"X"}

Defaults

• SSH user: user
• Location: a3111bd4-550a-47d0-838a-0a52bff2ae3f
• OS: TensorML 24 Everything
• CUDA: /usr/local/cuda-12.6/bin

SSH Keys (auto-select by hostname)

Presets

SSH

Copyssh -o StrictHostKeyChecking=no -o UserKnownHostsFile=/dev/null -i ~/.ssh/id_ed25519 -p <PORT> user@<IP>

Setup (fresh instance)

Copysource /opt/anaconda3/etc/profile.d/conda.sh && conda activate base
curl -LsSf https://astral.sh/uv/install.sh | sh && source $HOME/.local/bin/env

Available: nvcc, conda, PyTorch 2.5.1, git. Missing: cmake, pip3, uv.

File Transfer

Copyscp -P <PORT> -i ~/.ssh/id_ed25519 <LOCAL> user@<IP>:<REMOTE>
rsync -avz --progress -e "ssh -i ~/.ssh/id_ed25519 -p <PORT>" <SRC> <DST>

Data Persistence

WARNING: Instances are EPHEMERAL - download results before terminating. Stopping preserves data but costs storage.