Faiss imported after Torch leads to segfault

[anfatima]

If Faiss is imported after Torch, training in Faiss segfaults.

This can be reproduced using the example k-means clustering code in the Wiki.

Platform

OS: Big Sur version 11.6

Faiss version: 1.7.4 stable

Installed from: conda and brew

conda install faiss-cpu

Faiss compilation options:

Running on:

• CPU
• GPU

Interface:

• C++
• Python

Reproduction instructions

d = 128                           # dimension
nb = 100000                      # database size
nq = 10000                       # nb of queries
np.random.seed(1234)             # make reproducible
x = np.random.random((nb, d)).astype('float32')

ncentroids = 1024
niter = 20
verbose = True
d = x.shape[1]
kmeans = faiss.Kmeans(d, ncentroids, niter=niter, verbose=verbose)
kmeans.train(x)

[mdouze]

please install via conda only

[anfatima]

I did and I also installed it from source. I think the issue is with MKL and OpenMP conflicts between the Faiss installation and Torch.

The workaround works for now.