hugging-face-jobs

You can run GPU workloads on Hugging Face Jobs using ready-to-run UV scripts, which handle secure token authentication and push output results directly back to the Hub to prevent data loss in ephemeral environments.

The best way to manage HF_TOKEN secrets when submitting jobs is by using secure token handling practices enforced by the skill, ensuring your authentication credentials are protected during job execution.

Yes, it supports both CPU and GPU hardware flavor selection for Hugging Face Jobs, providing guidance on choosing the appropriate hardware and managing timeouts for your specific workload requirements.

Yes, you can use vllm and transformers dependencies within your UV scripts to execute jobs and persist outputs to the Hub, leveraging end-to-end result persistence to save your computational work.

Hugging Face Jobs results disappear in ephemeral environments because they lack end-to-end result persistence, which this skill solves by providing mechanisms to push outputs back to the Hub after execution.

While UV scripts provide ready-to-run job submission, you must consider hardware flavor selection limits and timeout constraints, and ensure proper HF_TOKEN secrets handling to avoid authentication failures during execution.