Even up to 1M or so rows you can just store everything in a numpy array or PyTorch tensor and compute similarity directly between your query embedding and the e

Even up to 1M or so rows you can just store everything in a numpy array or PyTorch tensor and compute similarity directly between your query embedding and the entire database. Will be much faster than the apply() and still feasible to run on a laptop.