dstack is an open-source platform to build and share data and ML applications within hours
Installation
Installing and running dstack
is very easy:
pip install --index-url https://test.pypi.org/simple/ --upgrade --no-cache-dir --extra-index-url=https://pypi.org/simple/ dstack==0.6.1.dev3
dstack server start
If you run it for the first time, it may take a while. Once it's done, you'll see the following output:
To access the application, open this URL in the browser: http://localhost:8080/auth/verify?user=dstack&code=xxxxxxxx-xxxx-xxxx-xxxx-xxxxxxxxxxxx&next=/
The default profile in "~/.dstack/config.yaml" is already configured. You are welcome to push your applications using Python package.
To access dstack
, click the URL provided in the output. If you try to access dstack
without using this URL, it will require you to sign up using a username and a password.
If you open the URL, you'll see the following interface:
You'll be logged as the dstack
user. The page you'll see is Applications
. It shows you all published applications which you have access to. The sidebar on the left lets you open other pages: ML Models
, Settings
, Documentation
, and Chat
.
Minimal Application
Here's an elementary example of using dstack
. The application takes real-time stock exchange data from Yahoo Finance for the FAANG companies and renders it for a selected symbol. Here's the Python code that you have to run to make such an application:
from datetime import datetime, timedelta
import dstack.controls as ctrl
import dstack as ds
import plotly.graph_objects as go
import pandas_datareader.data as web
def output_handler(self: ctrl.Output, symbols: ctrl.ComboBox):
start = datetime.today() - timedelta(days=30)
end = datetime.today()
df = web.DataReader(symbols.value(), 'yahoo', start, end)
fig = go.Figure(
data=[go.Candlestick(x=df.index, open=df['Open'], high=df['High'], low=df['Low'], close=df['Close'])])
self.data = fig
app = ds.app(controls=[ctrl.ComboBox(items=["FB", "AMZN", "AAPL", "NFLX", "GOOG"])],
outputs=[ctrl.Output(handler=output_handler)])
result = ds.push("minimal_app", app)
print(result.url)
If you run it and click the provided URL, you'll see the application:
To learn about how this application works and to see other examples, please check out the Tutorials documentation page.
To learn in more detail about what applications consist of and how to use all their features, check out the Concepts documentation page.
ML Models
dstack
decouples the development of applications from the development of ML models by offering an ML registry
. This way, one can develop ML models, push them to the registry, and then later pull these models from applications.
dstack
's ML Registry
supports Tensorflow
, PyTorch
, or Scikit-Learn
models.
Here's a very simple example of how to push a model to dstack
:
from sklearn import datasets
from sklearn import svm
import dstack as ds
digits = datasets.load_digits()
clf = svm.SVC(gamma=0.001, C=100.)
clf.fit(digits.data[:-1], digits.target[:-1])
url = ds.push("clf_app", clf)
print(url)
Now, if you click the URL, it will open the following page:
Here you can see the snippet of how to pull the model from an application or from anywhere else:
import dstack as ds
model = ds.pull('/dstack/clf_app')
To learn how to build an application that uses a simple ML model, check out the corresponding tutorial.
Feedback
Do you have any feedback either minor or critical? Please, file an issue in our GitHub repo or write to us on our Discord Channel.
Have you tried dstack
? Please share your feedback with us using this form!
Documentation
For more details on the API and code samples, check out the docs.
Contribution
The instructions on how to build dstack from sources can be found here.
License
dstack
is an open-source library licensed under the Apache 2.0 license