Recommended approach to reducing vector dimensionality?
#13
by
PaulCapestany
- opened
Apologies if there's an obvious answer to my question (I'm pretty new to ML), but, is there a recommended approach to reducing the size of vector embeddings produced by llava? I am limited to 2048 dimension vectors for my use case (vector similarity search), and llava's default seems to be 7168.
In case anyone else is interested in how this might be approached, it seems "random projection" is the way to go
PaulCapestany
changed discussion status to
closed