GPT-2-style LLM built from scratch in C/CUDA with hand-written backprop, BPE tokenizer, FlashAttention, pretraining, and SFT.
Last updated Jul 7, 2026
79
Stars
9
Forks
0
Issues
+1
Stars/day
Attention Score
63
Topics
Language breakdown
Cuda 73.7%
C 23.2%
Shell 2.7%
Makefile 0.4%
โธ Files
click to expand
README
No README available.
๐ More in this category