Host a Machine Learning Model Demo for Free With Streamlit and the Hugging Face Hub, Step by Step

Save the model

For scikit-learn, joblib.dump(model, 'model.joblib'). For PyTorch, save the state dict and keep the model class in a file. GitHub rejects files over 100 MB, so larger weights go on the Hugging Face Hub (step 7).

Write the Streamlit app

Create app.py:

import streamlit as st, joblib

@st.cache_resource
def load():
    return joblib.load('model.joblib')

model = load()
st.title('Spam classifier')
text = st.text_area('Message')
if st.button('Predict') and text:
    proba = model.predict_proba([text])[0]
    st.write('Prediction:', model.predict([text])[0])
    st.write('Confidence:', round(float(max(proba)), 2))

The cache keeps the model in memory instead of reloading it on every click.

List dependencies

Create requirements.txt with streamlit, scikit-learn, joblib and anything else you import, with versions pinned to match your machine.

Run it locally

pip install -r requirements.txt
streamlit run app.py

Open the local address and test a few inputs, including an empty one.

Push to GitHub

Create a repository with app.py, requirements.txt, the model file (if small) and a README. Commit and push.

Deploy on Streamlit Community Cloud

Sign in at Streamlit Community Cloud with your GitHub account, choose Create app, pick the repository, branch and app.py, and deploy. Streamlit's documentation says Community Cloud is free and connects to public or private repositories. Check its limits page before you rely on it.

Keep large weights on the Hugging Face Hub

Upload big model files to a model repository on the Hugging Face Hub and download them when the app starts with huggingface_hub. Store any tokens in the app's secrets settings, never in the code.

Check the logs

If the app fails to start, open its logs. The cause is almost always a missing package, a version mismatch or a wrong file path.

Write the model card

In the README, explain what the model does, the data, the metric on a held-out set and known failure cases. Add a few example inputs people can paste.

Link it and keep it honest

Add the URL to your project README, portfolio and resume bullet. Show confidence, and state the limits; reviewers trust a demo that admits what it cannot do.