novita-gpus

Python SDK for Novita serverless GPU endpoints.

Installation

pip install novita-gpus

Quick Start (Client)

import novita_gpus

novita_gpus.api_key = "your-api-key"

endpoint = novita_gpus.Endpoint("your-endpoint-id")
job = endpoint.run({"prompt": "hello world"})

print(job.status())
output = job.output(timeout=60)
print(output)

Quick Start (Worker)

import novita_gpus

def handler(job):
    job_input = job["input"]
    # ... process ...
    return {"result": "done"}

novita_gpus.start({"handler": handler})

Client API

• Endpoint(endpoint_id) -- create an endpoint client
• Endpoint.run(input) -- async submit a task, returns a Job immediately
• Endpoint.health() -- check endpoint health
• Endpoint.purge_queue() -- cancel all pending tasks
• Job.status() -- get current task status
• Job.output(timeout=0) -- get task output, optionally poll until complete
• Job.cancel() -- cancel a running task

Worker API

• novita_gpus.start(config) -- start the worker loop (blocks until SIGTERM)
• novita_gpus.progress_update(job, data) -- send in-progress status update
• novita_gpus.upload_image(job_id, path) -- upload image to S3, returns presigned URL
• novita_gpus.upload_file(name, path) -- upload file to S3
• novita_gpus.upload_bytes(name, data) -- upload bytes to S3

Configuration

Set credentials via environment variables:

export NOVITA_API_KEY="your-api-key"
export NOVITA_ENDPOINT_BASE_URL="https://async-public.serverless.novita.ai/v1"  # default

Or set them directly:

novita_gpus.api_key = "your-api-key"
novita_gpus.endpoint_url_base = "https://your-custom-url/v1"