Getting started
Quickstart
Install the client, authenticate, and pull your first checkpoint. Five minutes end to end on a machine with a single GPU.
Install
The client ships as a single wheel with optional extras for training and quantisation.
Install
pip install "osai[torch]" osai login # paste a token from Settings → Access tokens
Pull a model
Repositories are content addressed. A pull resolves the revision, verifies checksums and caches weights under ~/.cache/osai.
Pull a model
osai pull mistralai/Mistral-7B-Instruct-v0.3 --revision main
Generate
Any repository with a config.json loads through the auto classes; a chat template ships in the tokenizer.
Generate
from osai import pipeline
pipe = pipeline("text-generation", "mistralai/Mistral-7B-Instruct-v0.3")
print(pipe("Summarise rectified flow in two sentences.")[0]["text"])Where next
Read the Inference API page for hosted endpoints, or Datasets for streaming corpora without local storage.