Show HN: Biopetals – Run biology tuned Llama, BitTorrent-style

A developer created Biopetals, a modified version of the Petals library that enables distributed, BitTorrent-style inference for biology-tuned Llama models across a network of computers. The project was inspired by a two-hour wait for an AI protein folding server submission, aiming to accelerate biological model inference using consumer GPUs or Google Colab. The system requires a minimum number of participants on the network to function, with a setup Colab notebook provided for testing.

About a month ago, I heard about petals. Petals is basically a library that lets you run LLMs by loading the weights onto a network of computers that are all running petals. Then it splits up inference allowing you to run larger LLMs with consumer GPUs or just a Google colab. Earlier I was submitting a sequence to an AI protein folding server and had to wait almost 2 hours. This gave me the idea of trying to modify petals for biology models. Rewriting petals completely for different architectures genome language models, protein language models would be difficult, so I decided to start off by just rewriting for a biology tuned llama. If u wanna try it out, theres a fully setup Google colab linked in the readme, it might not actually run though since it needs a certain amount of people on the network. Comments URL: https://news.ycombinator.com/item?id=48296186 https://news.ycombinator.com/item?id=48296186 Points: 1 Comments: 0