tungstenkit 0.0.1a4

Tungstenkit is an open-source tool for building and using versatile and standardized ML model containers, Tungsten models.

Tungstenkit

Tungstenkit is an open-source tool for building and using versatile and standardized ML model containers, Tungsten models. The key features of Tungsten models are:

• Easy: Require only a few lines of Python code.
• Versatile: Support multiple usages:
  • RESTful API server
  • GUI application
  • Serverless function
  • CLI application (coming soon)
  • Python function (coming soon)
• Abstracted: User-defined JSON input/output.
• Standardized: Support advanced workflows.
• Scalable: Support adaptive batching and clustering (coming soon).

Learn More

• Documentation
• Getting Started
• Installation

Take the tour

Build a Tungsten model

Building a Tungsten model is easy. All you have to do is write a simple tungsten_model.py like below:

from typing import List

import torch
from tungstenkit import io, model


class Input(io.BaseIO):
    prompt: str


class Output(io.BaseIO):
    image: io.Image


@model.config(
    gpu=True,
    python_packages=["torch", "torchvision"],
    batch_size=4,
    description="Text to image"
)
class Model(model.TungstenModel[Input, Output]):
    def setup(self):
        weights = torch.load("./weights.pth")
        self.model = load_torch_model(weights)

    def predict(self, inputs: List[Input]) -> List[Output]:
        input_tensor = preprocess(inputs)
        output_tensor = self.model(input_tensor)
        outputs = postprocess(output_tensor)
        return outputs

Now, you can start a build process with the following command:

$ tungsten build

✅ Successfully built tungsten model: 'text-to-image:latest'

Run it as a RESTful API server

You can start a prediction with a REST API call.

Start a server:

$ docker run -p 3000:3000 --gpus all text-to-image:latest

INFO:     Setting up the model
INFO:     Getting inputs from the input queue
INFO:     Starting the prediction service
INFO:     Started server process [1]
INFO:     Waiting for application startup.
INFO:     Application startup complete.
INFO:     Uvicorn running on http://0.0.0.0:3000 (Press CTRL+C to quit)

Send a prediction request with a JSON payload:

$ curl -X 'POST' 'http://localhost:3000/predict' \
  -H 'accept: application/json' \
  -H 'Content-Type: application/json' \
  -d '[{"prompt": "a professional photograph of an astronaut riding a horse"}]'

{
    "status": "success",
    "outputs": [{"image": "data:image/png;base64,..."}],
    "error_message": null
}

Run it as a GUI application

If you need a more user-friendly way to make predictions, start a GUI app with the following command:

$ tungsten demo text-to-image:latest -p 8080

INFO:     Uvicorn running on http://localhost:8080 (Press CTRL+C to quit)

Run it as a serverless function

We support remote, serverless executions via a Tungsten server.

Push a model:

$ tungsten push exampleuser/exampleproject -n text-to-image:latest

✅ Successfully pushed to 'https://server.tungsten-ai.com'

Now, you can start a remote prediction in the Tungsten server:

Prerequisites

• Python 3.7+
• Docker
• (Optional) nvidia-docker for running GPU models locally. It enables using GPUs in Docker containers. However, you can build and push GPU models without it.

Installation

pip install tungstenkit