nanochat
Build your won ChatGPT for $100 on a single GPU
nanochat is a full-stack implementation of an LLM like ChatGPT in a single, clean, minimal, hackable, dependency-lite codebase. Run tokenization, pretraining, finetuning, evaluation, inference, and web UI on a single 8XH100 node.
nanochat is Andrej Karpathy's capstone project for LLM101n by Eureka Labs. It's a full-stack LLM implementation in ~1000 lines of clean, hackable code (Python, Rust, HTML, Shell). You can run the entire pipeline—tokenization, pretraining, finetuning, evaluation, inference, and web UI—on a single 8XH100 node for under $1000. It achieves competitive performance (~4e19 FLOPs, $100 tier model) without massive frameworks. Its purpose is to make LLMs accessible and understandable for developers who want to learn by doing!
nanochat is Andrej Karpathy's capstone project for LLM101n by Eureka Labs.
It's a full-stack LLM implementation in ~1000 lines of clean, hackable code (Python, Rust, HTML, Shell).
You can run the entire pipeline—tokenization, pretraining, finetuning, evaluation, inference, and web UI—on a single 8XH100 node for under $1000.
It achieves competitive performance (~4e19 FLOPs, $100 tier model) without massive frameworks.
Its purpose is to make LLMs accessible and understandable for developers who want to learn by doing!
I recommend this solution because it offers a unique opportunity to run a full-fledged ChatGPT-like assistant for just $100 on a single GPU.nanochat is a minimalist yet powerful framework that includes everything from tokenization and pretraining to inference and a web interface.
I recommend this solution because it offers a unique opportunity to run a full-fledged ChatGPT-like assistant for just $100 on a single GPU.
nanochat is a minimalist yet powerful framework that includes everything from tokenization and pretraining to inference and a web interface.
Amazing concept—democratizing LLMs at this price point is a founder’s dream!!
nanochat is Andrej Karpathy's capstone project for LLM101n by Eureka Labs.
It's a full-stack LLM implementation in ~1000 lines of clean, hackable code (Python, Rust, HTML, Shell).
You can run the entire pipeline—tokenization, pretraining, finetuning, evaluation, inference, and web UI—on a single 8XH100 node for under $1000.
It achieves competitive performance (~4e19 FLOPs, $100 tier model) without massive frameworks.
Its purpose is to make LLMs accessible and understandable for developers who want to learn by doing!