Matthijs Douze
Matthijs Douze
Yes right. In fact, this documentation is redundant with the index_facotry doc here: https://github.com/facebookresearch/faiss/wiki/The-index-factory but your suggestion applies there as well.
@alexanderguzhva IMO it would be great to support SVE. What I don't understand is if the SVE size needs to be known at compile time. In that case, we could...
You can share the `StandardGpuResources` over different GPU indexes and share the same GPU index across Kmeans indexes, so as long as you are not running them separately, you should...
I did the comparison (in C of course) and it was much slower than the code layout used in scann.
The packaged faiss comes with two compiled versions of the .so library (_swigfaiss.so and _swigfaiss_avx2.so) that are selected at the python layer. It is a current discussion on how we...
Right, the code is a bit scattered around. I am looking into it...
Thanks for flagging. We are going to see if there is a migration path...
Presumably because the threshold on nb queries https://github.com/facebookresearch/faiss/wiki/Implementation-notes#matrix-multiplication-to-do-many-l2-distance-computations Probably the ids are not filled in in the BLAS case.
Normally one would parallelize the search using Faiss' internal threading. It is unavoidable that the python threading incurs an overhead. It may be because, irrespective of the number of queries,...
For IVF search, this check is performed here: https://github.com/facebookresearch/faiss/blob/main/faiss/IndexIVF.cpp#L447