1 points, 0 comments on Hacker News
The Flash Attention 2 path was easier to finish after the algorithm existed in plain JavaScript first. The forward pass used one workgroup per batch, head, and query tile. It walked keys and values in blocks while keeping the online-softmax state in registers.
2 points, 1 comments on Hacker News
Ipswich Town host Arsenal in the Carabao Cup third round on Tuesday, with Gary O’Neil’s side looking to cause an upset against the reigning Premier League champions under the Portman Road lights. V...
Reading host Brentford in the Carabao Cup third round on Tuesday, with the League One side looking to spring an upset against Premier League opposition at the Select Car Leasing Stadium. Leam Richards...
The strength of a randomly generated password is one number — its entropy , in bits — and it comes from a single line of arithmetic. I computed it across every length and character-set choice using the exact character sets our password generator ships, then turned each into a crack time. One th...