Great comment. For what its worth, really think about your vectors before creating them! Any model can be a vector model, you just use the final hidden states..

Great comment. For what its worth, really think about your vectors before creating them! Any model can be a vector model, you just use the final hidden states... with that, think about your corpus and the model latent space and try to pair them appropriately. For instance, I vectorize and search network data using a model trained on coding, systems, data, etc.