yarn-sdk 0.1.0

Python SDK for Yarn — sovereign GPU compute for Australian researchers

Yarn SDK

Python SDK for Yarn -- sovereign GPU compute for Australian researchers.

Install

pip install yarn-sdk

Quick Start

import yarn

client = yarn.Client(
    api_key="yarn_...",  # from Account Portal -> API Keys
    base_url="https://api.yarn.prosodylabs.com.au",
)

# Submit a training job (single file)
job = client.jobs.submit(
    "my-experiment",
    script="train.py",
    gpu_count=1,
    pip_packages=["torch", "wandb"],
)

# Submit a multi-file project
job = client.jobs.submit(
    "kairos-phase1",
    directory="./my_project/",  # bundles all .py, .yaml, .json, etc.
    pip_packages=["torch", "wandb"],
)

# Wait for results
result = client.jobs.wait(job["job_id"])

# Stream logs
for line in client.jobs.stream_logs(job["job_id"]):
    print(line)

Interactive Sessions

with client.sessions.session("my-session", gpu_count=1) as session:
    print(session.ray_address)  # Connect Ray client
    print(session.dashboard_url)

Features

• Jobs: Submit training code, poll status, stream logs, retrieve results
• Sessions: Interactive GPU sessions with Ray
• Notebooks: JupyterLab workloads
• Storage: Upload/download datasets and checkpoints
• Billing: Check credit balance, view usage
• Secrets: Manage API keys and tokens for training jobs
• Chat: SSE streaming inference

Authentication

Get an API key from the Account Portal at account.yarn.prosodylabs.com.au -> API Keys.

client = yarn.Client(api_key="yarn_...")

License

MIT