JustVugg/nanoeuler
GPT-2-style LLM built from scratch in C/CUDA with hand-written backprop, BPE tokenizer, FlashAttention, pretraining, and SFT.
55 best points · 29 comments · 2 Hacker News posts · 127 GitHub stars · View on GitHub
Hacker News discussions
- Show HN: NanoEuler – GPT-2 scale model in pure C/CUDA from scratch (55 points, 26 comments, by vforno, 2026-06-28)
- Show HN: NanoEuler – GPT-2 scale model in pure C/CUDA from scratch (6 points, 3 comments, by vforno, 2026-06-19)