Hello1024/shared-tensor
View on GitHub ↗ ★ 35 C
A distributed, shared tensor with high performance approximate updates for machine learning
4 mentions on Hacker News by 1 person: 0 stories, 4 comments · first 2020-08-18 · latest 2025-11-04
20082026
-
…I think this is a solvable problem, and I started solving it a while ago with decent results: https://github.com/Hello1024/shared-tensor When someone gets this working well, I could totally see a distributed AI being tasked with expanding…
comment by londons_explore · 2025-11-04 · in This week in 1988, Robert Morris unleashed his eponymous worm -
…This project quantizes to 1 bit per weight and it works so well I didn't even make it configurable. https://github.com/Hello1024/shared-tensor
comment by londons_explore · 2024-10-12 · in INTELLECT–1: Launching the First Decentralized Training of a 10B Parameter Model -
…would be needed to ensure nodes aren't deliberately sending bad data to hurt the group effort. [1]: https://github.com/Hello1024/shared-tensor
comment by londons_explore · 2022-05-02 · in DALL-E 2 open source implementation -
I built this a few years ago: https://github.com/Hello1024/shared-tensor It does updates to weights based on 1 bit precision updates each iteration. It would be fairly trivial to go to less than 1 bit precision too - simply set some…
comment by londons_explore · 2020-08-18 · in Karpathy's MinGPT