llm.c GitHub repository · inference-research

Fields

id

sources/niche-llmc-repo

URL

https://github.com/karpathy/llm.c

Publisher

Andrej Karpathy

Kind

repo

Accessed

2026-10-03

notes

llm.c README. GPT-2/GPT-3 pretraining in plain C/CUDA, 'no need for 245MB of PyTorch or 107MB of cPython', with a ~1,000-line fp32 CPU reference in one file and a parallel PyTorch reference (train_gpt2.py, a tweaked nanoGPT). README explicitly says current focus is PRETRAINING; there is no documented inference/serving path. API metadata 2026-10-03: created 2024-04-08, last push 2025-06-26, no GitHub releases at all, MIT, 164 open issues, 31093 stars.

Notes

llm.c README. GPT-2/GPT-3 pretraining in plain C/CUDA, 'no need for 245MB of PyTorch or 107MB of cPython', with a ~1,000-line fp32 CPU reference in one file and a parallel PyTorch reference (train_gpt2.py, a tweaked nanoGPT). README explicitly says current focus is PRETRAINING; there is no documented inference/serving path. API metadata 2026-10-03: created 2024-04-08, last push 2025-06-26, no GitHub releases at all, MIT, 164 open issues, 31093 stars.

Cross-references

This record references no other records.