nanoeuler
JustVugg/nanoeuler
GPT-2-style LLM built from scratch in C/CUDA with hand-written backprop, BPE tokenizer, FlashAttention, pretraining, and SFT.
121stars
Forks
23
Open issues
0
Watchers
121
Size
14.4 MB
CudaMIT License
bpe-tokenizerbyte-pair-encodingccublascudadeep-learningflashattentionfrom-scratchgpt2language-modelllmmachine-learningneural-networknlpopenmptrainingtrasformer
Created: Jun 15, 2026
Updated: Sep 19, 2026
Last push: Jun 18, 2026