voltage-park
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.
When & Why to Use This Skill
This Claude skill automates the lifecycle management of high-performance Voltage Park H100 GPU instances. It enables users to provision, monitor, and terminate cloud GPU resources directly through natural language, streamlining the setup of compute-intensive environments for AI training and large-scale data processing while ensuring cost-efficiency through managed instance termination.
Use Cases
- On-Demand AI Model Training: Quickly spin up 1x to 8x H100 GPU clusters to handle intensive deep learning workloads when local compute is insufficient.
- Automated ML Environment Setup: Provision fresh instances and automatically configure them with essential tools like Conda, PyTorch 2.5.1, and CUDA 12.6 using the built-in setup scripts.
- Remote Development and File Management: Seamlessly transfer datasets and code via SCP/Rsync and manage remote execution through secure SSH tunneling.
- Cost-Optimized Experimentation: Manage ephemeral GPU instances by monitoring their status and terminating them immediately after experiments conclude to avoid unnecessary billing.
| 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
| Endpoint | Method | Path |
|---|---|---|
| Availability | GET | /virtual-machines/instant/locations |
| Provision | POST | /virtual-machines/instant |
| List | GET | /virtual-machines/ |
| Terminate | DELETE | /virtual-machines/{id} |
| Power | PUT | /virtual-machines/{id}/power-status |
| Ports | POST | /virtual-machines/{id}/port-forward |
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)
| Host | ID |
|---|---|
| macbook-pro | 4419ce92-a31a-40a2-b1da-73de608252ad |
| theodolos | fb384362-d726-4498-b9aa-d636d660f930 |
Presets
| GPUs | ID |
|---|---|
| 1x H100 | c6fd6253-cbb6-4ea8-a20c-47644b431f1c |
| 2x H100 | 928dc7be-5236-49a7-9d7d-c01ba6d783f5 |
| 4x H100 | 2470d2d9-347a-4186-884d-bd0350ace4e3 |
| 8x H100 | 404645e3-4677-4f48-b023-15baaa6621e1 |
SSH
ssh -o StrictHostKeyChecking=no -o UserKnownHostsFile=/dev/null -i ~/.ssh/id_ed25519 -p <PORT> user@<IP>
Setup (fresh instance)
source /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
scp -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.