Train your own ChatGPT on Apple Silicon — MLX port of nanochat

Train Llama 3 models from scratch. Any scale, any personality. By Arianna Method.

A minimal, hackable Vision-Language Model built on Karpathy’s nanochat — add image understanding and multimodal chat for under $200 in compute.

A legible and hackable port of Nanochat in JAX/Flax

The best GPT that $100-$125 worth of pre-training and finetuning can buy

Ascend NPU fork of nanochat for LLM training with torch_npu/HCCL (experimental)

nanochat's inference engine re-vibed in C++ with GGML.

JAX/Flax NNX port of Karpathy's nanochat, optimized for multi-host Cloud TPU pods.

llama.cpp fork adding nanochat architecture support, so Karpathy's nanochat checkpoints run as GGUF. See NANOCHAT.md

I built this repo to prove to my granny that I can implement GPT.

I see now that the path I choose through the maze makes me what I am.

Fork of karpathy/autoresearch: autonomous AI-agent experiments for single-GPU nanochat/LLM training research.

Rust-first continuation of nanochat-rs, benchmarked against karpathy/nanochat for faster, reproducible LLM training, evaluation, and inference.

Implementing Karpathy's nanochat in rust.

Controlled benchmark of memory mechanisms for transformers, built on nanochat

AI agents running research on single-GPU nanochat training automatically

A real LLM chat app that runs natively on Windows XP / Pentium 4, packaged as a single portable EXE with a custom C+SSE2 inference engine and Win32 GUI.

This script uses Modal to run Karpathy's nanochat training speedrun on cloud GPUs (8× H100s).

AI agents running research on single-GPU nanochat training automatically

Run nanochat training efficiently on Huawei Ascend NPUs with minimal code changes, supporting tokenizer, pretraining, and evaluation workflows.