Quickstart – Valohai Docs

Run the following commands in your local development machine.

pipx install valohai-cli

# Login with your credentials
vh login

# Creat a directory in your local machine
mkdir hello-valohai
cd hello-valohai

# Create a project
vh project create --name hello-valohai

Then add a new Python file hello.py and place inside it:

import valohai
import json

my_data = {
    'myinput': 'https://valohaidemo.blob.core.windows.net/mnist/preprocessed_mnist.npz'
}

# Define parameters in a dictionary
my_parameters = {
    'iterations': 5,
}

# Prepare a Valohai step called "train"
# that'll use a public PyTorch image
valohai.prepare(
    step="train",
    image="tensorflow/tensorflow:2.6.1-gpu",
    default_inputs=my_data,
    default_parameters=my_parameters
)

# Get the path to the downloaded file
downloaded_file_path = valohai.inputs('myinput').path()
print(f"Downloaded file available at: {downloaded_file_path}")

# Access the parameter defined at runtime
num_iterations = valohai.parameters('iterations').value

for i in range(num_iterations):
    # Print random metrics
    with valohai.metadata.logger() as logger:
        logger.log("iteration", num_iterations)
        logger.log("accuracy", random.uniform(60.0, 95.0))

Finally launch your job in Valohai — but let’s change the default parameter value

vh execution run train --iterations=10 --adhoc --open

That’s it! :tada: