velar-sdk 0.3.8

Velar Python SDK - Deploy ML models to GPUs with one command

Velar Python SDK

Deploy ML models to GPUs with one command.

Installation

pip install velar-sdk

Authentication

velar login

This opens your browser, signs you in, and saves an API key to ~/.velar/token automatically.

Quick Start

import velar

app = velar.App("my-model")

image = velar.Image.from_registry("pytorch/pytorch:2.1.0-cuda12.1-cudnn8-runtime")
image = image.pip_install("transformers", "accelerate")

@app.function(gpu="A100", image=image)
def run_inference(prompt: str) -> str:
    from transformers import pipeline
    pipe = pipeline("text-generation", model="gpt2")
    return pipe(prompt)[0]["generated_text"]

@app.local_entrypoint()
def main():
    result = run_inference.remote("Hello, world!")
    print(result)

Deploy:

velar deploy app:app

Run locally (calls the remote GPU):

velar run app:app

GPU Types

Name	VRAM	Price/hr
L4	24 GB	$0.59
RTX4090	24 GB	$0.89
L40S	48 GB	$1.29
A100	80 GB	$2.09
H100	80 GB	$3.59
H100-SXM	80 GB	$4.04
H200	141 GB	$5.39

Prices are approximate. Use velar.VelarClient().get_gpu_prices() for live pricing.

Image Builder

image = (
    velar.Image.from_registry("python:3.11-slim")
    .pip_install("torch", "transformers")
    .run_commands("apt-get update && apt-get install -y ffmpeg")
    .env(HF_HOME="/tmp/hf")
)

CLI Reference

velar login              # Authenticate via browser
velar deploy app:app     # Deploy to GPU cloud
velar run app:app        # Run local entrypoint
velar status             # List deployments
velar balance            # Show credit balance
velar cancel <id>        # Cancel a deployment
velar token set <key>    # Set API key manually
velar whoami             # Show current user

Links

• velar.run
• Dashboard