The idea came from the friction we kept seeing when trying to run models locally. Setting up models often involves downloading weights, configuring runtimes, and figuring out which models are actually good for a given task.
With llmpm the goal is to make models installable like packages:
llmpm install llama3
llmpm run llama3
We’ve also been working on the ability to package models with applications, so projects can declare model dependencies and reproduce the same setup easily.
Alongside the CLI, we’ve been experimenting with a model ranking and benchmarking tool on the website to help developers compare models across benchmarks and choose the right one before installing.
Check out rankings at: https://llmpm.co/rankings