Home › LLM Tools › nanoeuler

nanoeuler

by JustVugg · github.com/JustVugg/nanoeuler

GPT-2-style LLM built from scratch in C/CUDA with hand-written backprop, BPE tokenizer, FlashAttention, pretraining, and SFT.

Stars
★ 105
Forks
17
Language
Cuda
Category
LLM Tools
License
MIT
bpe-tokenizerbyte-pair-encodingccublascudadeep-learningflashattentionfrom-scratchgpt2language-modelllmmachine-learninggpt

View on GitHub ↗

Featured in