Summary:
Pull Request resolved: https://github.com/facebookresearch/faiss/pull/4745
Reduce test dataset size to use ~13 MB instead of ~1.5 GB while maintaining enough computational complexity to trigger the 10ms timeout before early stopping can converge.
Reviewed By: bshethmeta
Differential Revision: D90128255
fbshipit-source-id: 75b875d23ce459849d29acdb9d6de0828879ae46
Summary:
Pull Request resolved: https://github.com/facebookresearch/faiss/pull/4741
K-means clustering now stops early when the objective function converges, avoiding unnecessary iterations when niter is set to a high value as a safeguard. This prevents wasted computation when convergence happens well before the maximum iteration limit.
Reviewed By: limqiying
Differential Revision: D89748242
fbshipit-source-id: 2e728e64b94aea4fef2bd9acba117fda76104d61