Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Streaming and Sharded Data Loading #116

Open
dlwh opened this issue Mar 10, 2022 · 0 comments
Open

Streaming and Sharded Data Loading #116

dlwh opened this issue Mar 10, 2022 · 0 comments
Assignees
Labels
enhancement New feature or request

Comments

@dlwh
Copy link
Member

dlwh commented Mar 10, 2022

Processing the data up front is slow and doesn't make good use of the hardware. Better to tokenize/group on cpus while the gpus are busy (see e.g. DataLoader elsewhere)

  • Stream data rather than downloading it
  • Sharded streaming for multi-node training.

Related: caching the fully preprocessed data to disk is very inefficient. A 100GB corpus blows up to 400GB.

One thing to note is we have to consider how to do this in the presence of multi-dataset training.

@dlwh dlwh added the enhancement New feature or request label Mar 10, 2022
@dlwh dlwh added this to the Mistral V2 milestone Mar 10, 2022
@dlwh dlwh self-assigned this Mar 14, 2022
@dlwh dlwh removed this from the Mistral V2 milestone Jun 6, 2022
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
enhancement New feature or request
Projects
None yet
Development

No branches or pull requests

1 participant