Provision, manage, and terminate RunPod GPU instances for LLM workflows.
npx skills add https://github.com/grahama1970/fetcher --skill runpod-ops
Please help me install this Agent Skill. Skill: runpod-ops Source: https://github.com/grahama1970/fetcher/tree/main/.agents/skills/runpod-ops Command: npx skills add https://github.com/grahama1970/fetcher --skill runpod-ops
Provision, manage, and terminate RunPod GPU instances for LLM training and inference.
Spin up a GPU pod by running the provided script, then monitor or terminate it when finished.
High-intent search queries and answers about installing and using this skill.
You can automate RunPod GPU provisioning for LLM training by using a self-contained script that requires a RUNPOD_API_KEY to spin up on-demand pods. It auto-installs via uv and exposes commands for creating, monitoring, and terminating instances.
You can monitor RunPod pod lifecycle and usage during inference using the provided monitor command. It tracks pod status and usage, enabling cost-aware management across your training and inference workflows.
To manage RunPod instances, you need a valid RUNPOD_API_KEY for access and the uv package manager for auto-installation. Once configured, you can provision and terminate GPU pods on demand.
This approach works for both LLM training and inference workflows by providing on-demand GPU provisioning and lifecycle management. It covers cost-aware usage across typical RunPod deployments for model training, evaluation, and inference.
The best way to handle cost-aware usage of RunPod GPU instances is to monitor pod status and terminate them when finished. The automation script exposes lifecycle commands to ensure you only pay for active training and inference time.
You need a RUNPOD_API_KEY to provision GPU pods because it authenticates your requests to the RunPod cloud platform. This access is required to create, manage, and terminate on-demand instances for your LLM workflows.
ovachieverRun Python code in the cloud with autoscaling and GPU acceleration via Modal's serverless platform.
jluo41Package trained AI models into versioned assets with training results.
HeXiao-55Monitor WandB metrics to detect NaN, divergence, or stalled GPU utilization.
afel6Provision on-demand GPU cloud instances for training and inference.
