Skip to content

Investigate zipping preprocessed files to help save inodes #472

@roedoejet

Description

@roedoejet

@SamuelLarkin suggests fuse-mounting a zip archive or a new dataloader that preprocesses the audio on-the-fly (if it doesn't slowdown training aka makes the trainer wait on preprocessing) to help circumvent the perennial inode problem for large datasets.

Metadata

Metadata

Assignees

No one assigned

    Labels

    enhancementNew feature or request

    Type

    No type

    Projects

    No projects

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions