Rent and manage vast.ai GPU instances from an experiment task description.
npx skills add https://github.com/Mang30/myskills --skill vast-gpu-mang30
Please help me install this Agent Skill. Skill: vast-gpu Source: https://github.com/Mang30/myskills/tree/main/skills/vast-gpu Command: npx skills add https://github.com/Mang30/myskills --skill vast-gpu-mang30
Vast.ai GPU Management removes the hassle of figuring out which GPU to buy or provision by letting you describe your training task while it selects an appropriate GPU offer, rents an instance, sets it up, and destroys it to stop billing.
Ask your agent to run an experiment with on-demand GPUs by setting gpu: vast in your CLAUDE.md and executing /run-experiment "train model".
High-intent search queries and answers about installing and using this skill.
You can rent on-demand GPUs by describing your training task to vast-gpu, which estimates VRAM, disk, and runtime needs to provision a cost-optimized vast.ai instance. It handles the full lifecycle from setup to teardown automatically.
Task-driven GPU selection works by inferring minimum VRAM, GPU count, disk space, and estimated runtime from your experiment plans or scripts. It searches multiple GPU tiers and presents ranked, cost-optimized provisioning options based on estimated total cost and reliability.
Yes, you need the vastai CLI authenticated with an API key to automate GPU provisioning. The Skill requires this authenticated environment to search offers, rent instances, and manage the full training lifecycle.
The best way to manage vast.ai GPU instances is using state tracking in a vast-instances.json file to coordinate provision, setup, monitoring readiness, and teardown operations. This ensures billing stops automatically when runs complete.
Yes, you can use vast-gpu for evaluation workflows as well as training. The Skill is designed for any on-demand CUDA GPU workflow where you need temporary compute without specifying exact GPU models.
Billing may continue if the instance is not explicitly destroyed. vast-gpu prevents this by maintaining state in vast-instances.json to coordinate teardown operations and ensure instances are destroyed to stop billing after runs complete.
duypham2801Rent, configure, and destroy vast.ai GPU instances for ML training workloads.
fikriafAutomate discovery, download, and local inference of llama.cpp models from Hugging Face Hub.
zeronx798Deploy Python functions as auto-scaling serverless GPU APIs.
xqinagAutomates renting, managing, and destroying vast.ai GPU instances.
