Wistfare Python SDK

The official Python SDK for the Wistfare Serverless GPU Platform.

Installation

pip install wistfare

Quick Start

from wistfare import endpoint, Image

image = Image(
    python_version="3.12",
    python_packages=["torch", "transformers"],
    cuda_version="12.4",
)

@endpoint(name="predict", cpu=2, memory="16Gi", gpu="H100", image=image)
def predict(context, prompt: str = ""):
    model = context.on_start_result
    return model.generate(prompt)

@predict.on_start()
def load_model(context):
    from transformers import pipeline
    return pipeline("text-generation", model="meta-llama/Llama-3-8B", device="cuda")

wistfare deploy app.py:predict

Features

• Decorators: @endpoint, @function, @task_queue, @schedule
• GPU Support: RTX 4090, A100, H100 with per-second billing
• Image Builder: Define container environments with Image()
• CLI: wistfare deploy, wistfare serve, wistfare logs
• Client: Programmatic invocation via WistfareClient

Documentation

See the Getting Started Guide.

License

MIT