All rights reserved © Data Chemist Limited 2025.
VectorLink.py is a python library that exposes tools for managing vectors at scale. It is part of the VectorLink pipeline.
VectorLink.py requires an NVidia GPU of at least Lovelace or better architecture to function. The speed difference with GPU acceleration means that vector processing at scale is simply best left to GPUs for reasons of both cost and time.
You will also need an installation of pytorch which works with your GPU.
To make a new Approximate Nearest Neighbor graph (ANN) using VectorLink.py, simply import the library and pass it the vectors you want to index as a 2D tensor, the first dimension is the number of vectors and the second dimension being the vector length. Currently we only process torch.float32 vectors.
from vectorlink import ANN
ann = ANN(vectors=vectors)
sq = ann.search(a) # Searches for a vector or collection of vectors, and returns a search queue
c = ann.clusters() # Returns all search queues