Closed siddhartha-gadgil closed 1 year ago
@ayush1801 I have implemented a nearest neighbour search (copying many things you did, in particular using torch.topk
) for the case of prompts. I used a simpler model so this should be faster. But still, I wanted to alert you that I am finding that loading and finding top-k is instantaneous.
It may be worth checking if there is some inefficiency in the main code we use.
Now most of the time will be taken by GTP 3.5/4 so no point in this.
Currently it takes a few seconds to translate a sentence. All steps other than the Codex query should be examined and optimized.