Replies: 1 comment
-
Sharding is not guaranteed to work because there is always the possibility that the 50,000 neighbors are in a single shard. |
Beta Was this translation helpful? Give feedback.
0 replies
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
-
Hi 👋
Thank's for the great work!
I have a dataset of around 2M vectors of size 768 (BERT embeddings, but (a) I can lower the dimensionality, (b) for the sake of argument let's assume they are just random vectors) and I'd like to do approximate NN search on that dataset.
My requirement is that my k value is very large - around 50,000.
IIUC max k is 1024/2048, depending on my GPU.
But what if I will shard my index to 25-50 different shards? Can this do the trick?
What do you think it will do to the latency?
Is there a tutorial that shows something remotely like this?
Thank you for your help
Beta Was this translation helpful? Give feedback.
All reactions