3 ms·
The current implementation is very slow with a CPU. Probably, it could be optimized with parallelism for batches. I'm not sure how it could be improved for sing
by J_Shelby_J 2y ago
The current implementation is very slow with a CPU. Probably, it could be optimized with parallelism for batches. I'm not sure how it could be improved for single embeddings.