🔥The 9th Round of Easy Loan, Earn $40 Reward is in progress❗️ ⏰ Promotion Period: January 15th - Feburary 15th, 2025 👉 Register now and check more details at gate.io/campaigns/358

NCCL sending the loss value from the last pipeline parallel stage back to rank 0 so the user can print it

wellp and there goes the austria doesn’t have school shootings streak. lots of things we can learn from the US, but this ain’t one of them

Statement by Foreign Minister Meinl-Reisinger Beate Meinl-Reisinger on the school shooting in #Graz

It's easy to blame graphics as it's visible, but that's not the reason for modern software slowness. The reason is CPU cache misses, mutex waits and stupid slow code everywhere that nobody knows exists, because they never run a debugger or profiler.

Panos Karabelas Thank you for the inspiration

primeintellect sending a tensor through PCI-E to host memory, through ethernet/TCP/IP across a continent to another node's host memory, and back down through PCI-E again

can't even trust battle tested string diffing libraries to not break smh

python3 beep_on_crash.py --pid 1362 gn everyone

wow, no crash :)

"You can break AI down into 5 tiers." - George Hotz 🌑 "Data centers - tier 1, fabs - tier 2, Nvidia/AMD - tier 3, OpenAI/Anthropic - tier 4, and completely worthless things like Cursor and Windsurf, which are tier 5." "OpenAI and Anthropic will eat all the value from the

It is intuitively obvious that reasoning in continuous embedding space is dramatically more powerful than reasoning in discrete token space. This paper from Yuandong Tian and team show that it is the case theoretically.

streaming backward decomposition of non-atemporal models is possible. don't let the residuals scare you

Extremely hyped about what we are building rn at prime we are accelerating like never on all front