From 5896a5f11f62554c4b4af8b0fb40bb5bc443a19d Mon Sep 17 00:00:00 2001 From: giangph00 Date: Sat, 14 Mar 2026 17:02:47 +0700 Subject: [PATCH 1/3] Add how nanoGPT is different to minGPT in README.md --- README.md | 8 ++++++++ 1 file changed, 8 insertions(+) diff --git a/README.md b/README.md index 9debd17..dc98a81 100644 --- a/README.md +++ b/README.md @@ -1,3 +1,11 @@ +# nanoGPT + +In short: nanoGPT is the more practical, performance-oriented successor to the education-focused minGPT. + +What's different? +- minGPT was designed with a focus on education +- nanoGPT "prioritizes teeth over education." In other words, it shifts from being purely a teaching tool to something that's still simple but can actually reproduce real benchmarks. The code is still plain and readable: train.py is a ~300-line boilerplate training loop and model.py a ~300-line GPT model definition, which can optionally load the GPT-2 weights from OpenAI. + # minGPT From a4dc188816a81aff77e8645a862f4e21beed85ec Mon Sep 17 00:00:00 2001 From: giangph00 Date: Sat, 14 Mar 2026 17:06:59 +0700 Subject: [PATCH 2/3] Add a long version explaining how nanoGPT is different --- README.md | 11 +++++++++++ 1 file changed, 11 insertions(+) diff --git a/README.md b/README.md index 9debd17..d91ccef 100644 --- a/README.md +++ b/README.md @@ -1,3 +1,14 @@ +# nanoGPT + +Both minGPT and nanoGPT were created by Andrej Karpathy. Here's the key difference: + +- minGPT was designed with a focus on education. The minGPT library is three files: mingpt/model.py (the Transformer model definition), mingpt/bpe.py (a Byte Pair Encoder), and mingpt/trainer.py (PyTorch boilerplate training code). It's meant to be easy to read and learn from. + +- nanoGPT is Karpathy's rewrite that "prioritizes teeth over education." In other words, it shifts from being purely a teaching tool to something that's still simple but can actually reproduce real benchmarks. The code is plain and readable: train.py is a ~300-line boilerplate training loop and model.py a ~300-line GPT model definition, which can optionally load the GPT-2 weights from OpenAI. + +As of Jan 2023, minGPT is in a semi-archived state, and Karpathy points people to nanoGPT for more recent developments. + +So in short: nanoGPT is the more practical, performance-oriented successor to the education-focused minGPT. # minGPT From bc396dcd9891b359e74383e921be05c9a9d92a9c Mon Sep 17 00:00:00 2001 From: giangph00 Date: Sat, 14 Mar 2026 18:23:40 +0700 Subject: [PATCH 3/3] Update README.md --- README.md | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) diff --git a/README.md b/README.md index aded93f..4ac1174 100644 --- a/README.md +++ b/README.md @@ -1,6 +1,6 @@ # nanoGPT -Both minGPT and nanoGPT were created by Andrej Karpathy. In short: nanoGPT is the more practical, performance-oriented successor to the education-focused minGPT. +Both minGPT and nanoGPT were created by Andrej Karpathy. In short, nanoGPT is the more practical, performance-oriented successor to the education-focused minGPT. What's different? - minGPT was designed with a focus on education