This quickstart guide shows you how to build and deploy your first model, using Baseten’s Truss framework.

​
Prerequisites

To use Truss, install a recent Truss version and ensure pydantic is v2: 
pip install --upgrade truss 'pydantic>=2.0.0'
To deploy Truss remotely, you also need a Baseten account. It is handy to export your API key to the current shell session or permanently in your .bashrc:
~/.bashrc
export BASETEN_API_KEY="nPh8..."

​
Initialize your model

Truss is a tool that helps you package your model code and configuration, and ship it to Baseten for deployment, testing, and scaling. To create your first model, you can use the truss init command. 
$ truss init hello-world
? 📦 Name this model: HelloWorld
Truss HelloWorld was created in ~/hello-world
This will create a new directory called hello-world with the following files:
  • config.yaml - A configuration file for your model.
  • model/model.py - A Python file that contains your model code
  • packages/ - A folder to hold any dependencies your model needs
  • data/ - A folder to hold any data your model needs
For this example, we’ll focus on the config.yaml file and the model.py file.

​
config.yaml

The config.yaml file is used to configure dependencies, resources, and other settings for your model. Let’s take a look at the contents:
config.yaml
build_commands: []
environment_variables: {}
external_package_dirs: []
model_metadata: {}
model_name: HelloWorld
python_version: py311
requirements: []
resources:
  accelerator: null
  cpu: '1'
  memory: 2Gi
  use_gpu: false
secrets: {}
system_packages: []
Some key fields to note:
  • requirements: This is a list of pip packages that will be installed when your model is deployed.
  • resources: This is where you can specify the resources your model will use.
  • secrets: This is where you can specify any secrets your model will need, such as HuggingFace API keys.
See the Configuration page for more information on the config.yaml file.

​
model.py

Next, let’s take a look at the model.py file. 
class Model:
    def __init__(self, **kwargs):
        pass

    def load(self):
        pass

    def predict(self, model_input):
        return model_input
In Truss models, we expect users to provide a Python class with the following methods:
  • __init__: This is the constructor.
  • load: This is called at model startup, and should include any setup logic, such as weight downloading or initialization
  • predict: This is the method that is called during inference.

​
Deploy your model

To deploy your model, you can use the truss push command. 
$ truss push
This will deploy your model to Baseten.

​
Invoke your model

After deploying your model, you can invoke it with the invocation URL provided: 
$ curl -X POST https://model-{model-id}.api.baseten.co/development/predict \
  -H "Authorization: Api-Key $BASETEN_API_KEY" \
  -d '"some text"'
"some text"

​
A Real Example

To show a slightly more complex example, let’s deploy a text classification model from HuggingFace! In this example, we’ll use the transformers library to load a pre-trained model, from HuggingFace, and use it to classify the given text.

​
config.yaml

To deploy this model, we need to add a few more dependencies to our config.yaml file.
config.yaml
requirements:
  - transformers
  - torch

​
model.py

Next, let’s change our model.py file to use the transformers library to load the model, and then use it to predict the sentiment of a given text.
model.py
from transformers import pipeline

class Model:
    def __init__(self, **kwargs):
        pass

    def load(self):
        self._model = pipeline("text-classification")

    def predict(self, model_input):
        return self._model(model_input)

​
Running inference

Similarly to our previous example, we can deploy this model using truss push 
$ truss push
And then invoke it using the invocation URL on Baseten. 
$ curl -X POST https://model-{model-id}.api.baseten.co/development/predict \
  -H "Authorization: Api-Key $BASETEN_API_KEY" \
  -d '{"text": "some text"}'

​
Next steps

Now that you’ve deployed your first model, you can learn more about more options for configuring your model, and implementing your model.