Impressive work! For comparison, an H100 can do roughly 100-200 tok/s of Muse 30B for raw inference single-thread, going up to low thousands of tok/s with a large number of threads - and I am sure that for the massively-multi-threaded case they can optimize the prover further.

更多来自 @VitalikButerin

1. #1

   My third (and last) post in my series on obfuscation (iO): local mixing https://t.co/Du1c052w1Z Local mixing is a very different philosophy from the other two, no prior background…

   vitalik.ethvitalik.eth @VitalikButerin · 2026-08-21
2. #2

   Bitcoiners deserve a lot of credit for pioneering many of these ideas (see Utreexo). But yes, this is what the current proposed Ethereum scaling strategy looks like in action. We…

   vitalik.ethvitalik.eth @VitalikButerin · 2026-08-16
3. #3

   Yabloko, Russia's only remaining openly anti-war political party, was just banned. https://t.co/mMHe58XYX1 Also recently, Nadezhdin (from a different party) had to flee to France:…

   vitalik.ethvitalik.eth @VitalikButerin · 2026-08-10
4. #4

   I updated my 2023 roadmap diagram to overlay where the items that were there sit in the current Strawmap ( https://t.co/9deLIQWG24 ). In general, a lot of overlap, but: * Some thi…

   vitalik.ethvitalik.eth @VitalikButerin · 2026-08-10

   
5. #5

   Minimax H3 is the first *open* video model I've seen that outperforms HunyuanVideo 1.5 - which is actually an impressive accomplishment for @TencentHunyuan to have held that thron…

   vitalik.ethvitalik.eth @VitalikButerin · 2026-08-07

   ▶
6. #6

   Very welcome recent news from Signal: they are working on letting you register an account without a phone number. https://t.co/W8GGKCbONq That said, an important counterpoint abou…

   vitalik.ethvitalik.eth @VitalikButerin · 2026-08-07

查看 @VitalikButerin 的全部帖子 →