bigscience-workshop / petals

🌸 Run LLMs at home, BitTorrent-style. Fine-tuning and inference up to 10x faster than offloading
https://petals.dev
MIT License
8.89k stars 490 forks source link

WIP GCP TPU XLA Build #567

Closed jmikedupont2 closed 2 months ago

jmikedupont2 commented 3 months ago

This is the first in a series of prs for getting petals running on v3 tpu on gcp using guix.