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.

Sign up for GitHub

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

Jump to bottom

Updated memmap_dtype to automatically choose based on vocab size #795

Open
wants to merge 11 commits into
base: main
Choose a base branch
from
Open

Updated memmap_dtype to automatically choose based on vocab size #795

wants to merge 11 commits into from

Conversation

aman-17
Copy link
Member

@aman-17 aman-17 commented Feb 12, 2025

Updated memmap_dtype to uint32 for compatibility with OLMo-2-1124.
Added links to the Discord and Playground in the readme.
Minor improvements in readme.

@aman-17 aman-17 requested a review from dirkgr February 12, 2025 21:48
dirkgr
dirkgr requested changes Feb 13, 2025
View reviewed changes
Copy link
Member

@dirkgr dirkgr left a comment

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Sorry, I think you misunderstood. What we want is, set the dtype based on the vocab size. If the vocab size is > 2**16, then the dtype has to be uint32. If it's smaller, it has to be uint16.

@aman-17 aman-17 self-assigned this Feb 28, 2025
@aman-17 aman-17 added the type/bug An issue about a bug label Feb 28, 2025
@aman-17 aman-17 changed the title Updated default value of memmap_dtype to uint32 Updated memmap_dtype to automatically choose based on vocab size Mar 13, 2025
dirkgr
dirkgr requested changes Mar 13, 2025
View reviewed changes
Copy link
Member

@dirkgr dirkgr left a comment

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

The changelog is no longer accurate.

But my main question is, what happens when someone overrides the setting? Is that possible? It should be.

There is still data_config.effective_memmap_dtype. Where does that come from? Do we have to respect if it's set?

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Reviewers

@dirkgr dirkgr dirkgr requested changes

Requested changes must be addressed to merge this pull request.

Assignees

@aman-17 aman-17

Labels
type/bug An issue about a bug
Projects
None yet
Milestone
No milestone
Development

Successfully merging this pull request may close these issues.
2 participants
@aman-17 @dirkgr