facebookresearch / facebookresearch/segment-anything
Decision for the 32 bit embedding
Open
- Dominant language
- Jupyter Notebook
- Stars
- 54.9k
- Forks
- 6.4k
- PR merge metrics
- No merged PRs in 30d
Description
Hi!
We run a couple of tests and f16 embeddings perform similar results with the f32, while consuming 50% less memory. I was wondering whether 32 bit is neccessary, or did you do any test regarding this.
Contributor guide
Research direction
The issue names no file, test, or entry point; begin by locating where embedding precision is selected and reproduce the reported f16 versus f32 comparison. Done means reaching and documenting a justified decision about whether 32-bit embeddings are necessary, based on memory use and result quality.
Written by the indexing model from the issue text.
Assessment
- Domain
- machine-learning
- Issue type
- Feature
- Difficulty
- 5/5
- Estimated time
- Over a week
- Activity status
- Stale
- Clarity
- Needs clarification
- Newbie friendliness
- 20/100