llama-cpp is a project to run models locally on your computer. But downloading models is a bit of a pain. This package is here to help you with that.
It finds the largest model you can run on your computer, and download it for you. It leverage the excelent TheBloke's HuggingFace models to do so.
- find the largest model i can run
- dychotomy + download + validate ?
- download the largest model i can run for a specific computer
- improve the search capability
- validate a downloaded model can run
- download the model
- use node-llama-cpp chat on it
Download "zephyr-7B-alpha-GGUF" model with quantisation "Q6_K"
npx llamacpp-download download zephyr-7B-alpha-GGUF -q Q6_K
List all the models available
llamacpp-download.js list
Search for a model: e.g. all models names which includes "mistral" and "7b"
llamacpp-download.js search mistral 7b
This package is published on npmjs.com here, nevertheless most people don't need to install it. because it is meant to be used with npx.
npm install llamacpp-download
You need to have wget installed on your system.
Here this is a reminder for myself on how to publish this package.
npm login
npm run release
Internally it use np to publish the package. It will ask you some questions (e.g. version number), will do a bunch of checks and then it will publish it. See this article for more details.