Batch processing on HPC¶
On a login node with internet access, cache the weights before submitting a job:
On the compute node, use the same Hugging Face cache and run the CLI, for example:
Inputs with the same filename stem must be processed separately. Use --key to select a dataset inside a container and --fs to specify its sampling rate when it is not recorded in the file. Run tokeye run --help for all options. For the app on a remote node, use SSH port forwarding as described in the app guide.