logoalt Hacker News

qarlyesterday at 10:20 PM1 replyview on HN

I don't know the details but apparently it has something to do with multiprocess contention for the GPU and batch sizing.


Replies

fc417fc802today at 12:45 PM

Right but regardless, it's a buggy optimization. The calculations are all fully deterministic when done "properly" and fully consistently but we almost never bother with that because it slows things down and the errors don't matter in practice 99% of the time.

show 1 reply