Updating NanoGPT with a "speed run" version #181
Description
Karpathy was only soft-patching NanoGPT for the last year, and the community in Modded-NanoGPT pretty much added a lot of SOTA to accelerate trainings forward in case NanoGPT-Lite (as a testing standard) needs a little speed boost https://github.com/KellerJordan/modded-nanogpt
P.S. Thinking if a collab might be useful as well? KellerJordan/modded-nanogpt#76