I want to do a pairwise distance computation on 2 feature matrices of sizes say
n x f and
n x f, and get an
n x n matrix from this.
I am currently using
torch.cdist() for this, and was wondering if there is any way to parallelize this across GPUs, something like how FAISS does - GitHub - facebookresearch/faiss: A library for efficient similarity search and clustering of dense vectors.?
Or do I need to write a custom implementation for this using torch multiprocessing?
Thanks for the help in advance!