hugging-face-jobs

You can run batch inference jobs on Hugging Face without local GPU compute by submitting UV scripts or Docker-based tasks to HF Jobs. The platform handles secure token management and hardware flavor selection across CPU, GPU, and TPU, pushing results directly to the Hub.

To securely manage your HF_TOKEN when submitting Docker-based jobs, supply the token via Hugging Face secrets. The platform injects the secret at runtime, allowing your workload to authenticate and persist datasets or models to the Hub without exposing credentials in your script.

Yes, you can persist datasets and models directly to the Hugging Face Hub after running a workload. The skill includes built-in persistence options that push artifacts back to the Hub, eliminating the need to manually transfer results from temporary compute instances.

The best way to select hardware flavors for scheduled data processing on HF Jobs is using the skill's built-in hardware flavor guidance. It helps you match CPU, GPU, and TPU resources to your workload requirements while maintaining cost-awareness for planning and execution.

Yes, Hugging Face Jobs works seamlessly with UV scripts for automated data processing pipelines. You can submit UV scripts directly as workloads, allowing you to execute Python-based processing logic remotely while leveraging the platform's secure token handling and result persistence.