Skip to content

batch size used in index.add() on SIFT1M #11

@nicolelii

Description

@nicolelii

I'm currently testing IndexACORNFlat on the SIFT1M dataset, and I keep running into a segmentation fault when calling index.add(...) on the full dataset. However, things seem to work when I reduce the batch size significantly (e.g., to 1000 vectors per call).

Could you clarify what batch size you used in your experiments when adding vectors to the index (especially for the full 1M vector case)? Did you add all vectors in one call, or in chunks?

Metadata

Metadata

Assignees

No one assigned

    Labels

    No labels
    No labels

    Type

    No type

    Projects

    No projects

    Milestone

    No milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions