kaggle-environment
Kaggle Environment
Overview
Use this skill for Kaggle hosted runtime behavior: filesystem paths, accelerators, internet access, dependency drift, output locations, reproducibility, and resource limits. Distinguish local development assumptions from Kaggle execution assumptions.
Path Conventions
| Path | Use |
|---|---|
/kaggle/input |
Read-only mounted datasets, competition data, models, and notebook sources |
/kaggle/working |
Writable working directory; save submissions and artifacts here |
| Local project folder | Source and metadata folder used by CLI push/pull |
Never write final outputs only under /kaggle/input; it is read-only. For submissions or artifacts, write to /kaggle/working/<file> and verify the file appears in notebook outputs.
Accelerator IDs
kaggle kernels push -p <folder> --accelerator <ID> can request accelerators. Valid IDs include NvidiaTeslaP100, NvidiaTeslaT4, NvidiaTeslaT4Highmem, NvidiaTeslaA100, NvidiaL4, NvidiaL4X1, NvidiaH100, NvidiaRtxPro6000, TpuV38, Tpu1VmV38, TpuV5E8, and TpuV6E8. Choose the smallest appropriate accelerator first, and warn that high-end GPUs/TPUs plus long timeouts can consume quotas.