中文
Libraries & frameworks

nanochat

karpathy/nanochat

A minimal, hackable LLM training harness covering tokenization, pretraining, finetuning, evaluation, and inference, designed to train GPT-2-class models affordably on a single GPU node.