tysam-code

Pinned Repositories

fileshare
A misc space to hold (smallish) files to share with others online
1 2 00
hlb-CIFAR10
Train to 94% on CIFAR-10 in <6.3 seconds on a single A100. Or ~95.79% in ~110 seconds (or less!)
Language:Python1.3k 19 376
hlb-gpt
Minimalistic, extremely fast, and hackable researcher's toolbench for GPT models in 307 lines of code. Reaches <3.8 validation loss on wikitext-103 on a single A100 in <100 seconds. Scales to larger models with one parameter change (feature currently in alpha).
Language:Python343 7 528
modded-nanogpt
NanoGPT (124M) in 5 minutes
Language:Python10

tysam-code's Repositories

tysam-code/hlb-CIFAR10
Train to 94% on CIFAR-10 in <6.3 seconds on a single A100. Or ~95.79% in ~110 seconds (or less!)
Language:Python1.3k 19 376
tysam-code/hlb-gpt
Minimalistic, extremely fast, and hackable researcher's toolbench for GPT models in 307 lines of code. Reaches <3.8 validation loss on wikitext-103 on a single A100 in <100 seconds. Scales to larger models with one parameter change (feature currently in alpha).
Language:Python343 7 528
tysam-code/fileshare
A misc space to hold (smallish) files to share with others online
1 2 00
tysam-code/modded-nanogpt
NanoGPT (124M) in 5 minutes
Language:Python10
tysam-code/HeavyBall
Efficient optimizers
Language:Python00