hr0nix
@hr0nix
🦾 Now onto something new 💻 Ex @nebiusai, @Yandex, @MSFTResearch, @CHARMTherapeutx 🧠 Interested in (M|D|R)L, AGI, rev. Bayes 🤤 Opinions stupid but my own
I'm shocked that most pytorch libraries that do random stuff (e.g. torchvision transforms) don't take entropy source as an input. How the hell do you guys build reproducible deterministic data pipelines? Fix global random seed and hope for the best?
JAX is good I don’t get why so many people still use torch, especially for small scale stuff. Stop torchuring yourself folks
Yesterday was my last day at @nebiusai. It was an amazing journey, but now a new adventure awaits! More to share soon.
f1 is the best movie about formula 1 ever made and a great movie overall
A throwback from the old life. So many great weekends spent flying with friends.
As promised, here is the tech report huggingface.co/papers/2505.20…
We will also drop a tech report soon, where we will describe our automated issue mining pipeline in more detail and share some observations on eval discrepancies between swe-rebench and swe-bench verified. ⬇️