Execute Python, ML, and inference workloads on Colab or Kaggle GPUs from a local terminal.
npx skills add https://github.com/franklinbaldo/skills --skill free-gpu-franklinbaldo
Please help me install this Agent Skill. Skill: free-gpu Source: https://github.com/franklinbaldo/skills/tree/main/free-gpu Command: npx skills add https://github.com/franklinbaldo/skills --skill free-gpu-franklinbaldo
colab run --gpu T4 job.py execution and named sessions for dependency installs, uploads, downloads, logs, and guaranteed cleanup via colab stop. - Kaggle workflows: Creates private GPU script Kernels with validated metadata, pushes them with an accelerator, polls status, and downloads outputs written under /kaggle/working. - Use Case: On a Windows laptop without admin rights, package a training script as a private Kaggle Kernel, run it on a Tesla T4, and retrieve the resulting model artifacts without installing anything system-wide. ## Quick Start Use the free-gpu skill to run my training script on a free Colab or Kaggle GPU and download the output artifacts.High-intent search queries and answers about installing and using this skill.
Use the google-colab-cli one-shot command `uvx --from google-colab-cli colab run --gpu T4 job.py`, which creates, executes, and releases the VM. For jobs needing uploads or downloads, use a named Colab session or push a private Kaggle Kernel with a GPU accelerator.
Use Colab for fast iteration, interactive sessions, and direct file transfer on Linux, macOS, or WSL. Use Kaggle Kernels for long asynchronous jobs and for native Windows machines without WSL, since the Kaggle CLI runs natively on Windows.
No. google-colab-cli 0.6.0 imports Unix-only termios and tty modules and fails on native Windows. Use Kaggle natively, run the CLI inside WSL, or use the bundled shim only for explicitly chosen non-interactive commands.
Install uv into the user profile with the official PowerShell installer, which needs no admin rights, then run tools via `uvx --from kaggle kaggle` or `uvx --from google-colab-cli colab`. These live in an isolated user cache.
Kaggle only exposes files written under /kaggle/working for download. Ensure the script writes artifacts there, then retrieve them with `kaggle kernels output OWNER/slug -p ./output` after the Kernel finishes.
Confirm the data may be uploaded to Google or Kaggle before transfer, and use browser OAuth or provider secret stores. Never commit tokens, credential files, or notebooks containing secrets, and treat remote logs as potentially sensitive.
datamaker-krAutomate file uploads from local or cloud storage to Synapse data collections.
jluo41Package trained AI models into versioned assets with training results.
lix965996-artDeploy and run machine learning experiments on local or remote GPU hardware.
qenex-aiGuide provisioning and management of AWS, GCP, and Azure resources via CLI commands.