Run with local code

Begin by creating an account at app.valohai.com.

Install the tools

Install the valohai-cli tool on your local machine (or wherever you write your scripts) by running:

pipx install valohai-cli

# Then use your credentials to log in
vh login

Create a project

First, create a directory on your local machine:

mkdir hello-valohai

Navigate to the newly created folder. Then, create a new Valohai project and link local commands to it. This setup informs Valohai which project to use when running a job from your machine:

vh project create --name hello-valohai

Create a hello.py

Create a file named train.py and insert a simple yolov8 training example:

import shutil
from ultralytics import YOLO
import json

# Load a pretrained model (recommended for training)
model = YOLO("yolov8n.pt")

# Train the model
model.train(data="coco128.yaml", epochs=1, verbose=False)

# Export the model to ONNX format
path = model.export(format="onnx")

Next, modify the code to copy the generated model to the Valohai outputs directory, and create an alias so that you can easily reference the file later using the alias “latest-model”:

import shutil
from ultralytics import YOLO
import json

# Load a pretrained model (recommended for training)
model = YOLO("yolov8n.pt")

# Train the model
model.train(data="coco128.yaml", epochs=1, verbose=False)
path = model.export(format="onnx")

# Copy the exported model to the Valohai outputs directory
shutil.copy(path, '/valohai/outputs/')

file_metadata = {
    "valohai.alias": "latest-model"
}

# Attach the metadata to the file, enabling easy retrieval by the alias
with open("/valohai/outputs/best.onnx.metadata.json", "w") as outfile:
    f.write(json.dumps(file_metadata))

Create a valohai.yaml

Finally, create a valohai.yaml configuration file to describe the job for Valohai. This configuration includes a step name, the command to run the script, the environment configuration, and a base environment image that contains Python and YOLOv8 libraries, sourced from a Docker image on hub.docker.com.

Update the environment field with the GPU machine you want to run the job on. You can see a list of available GPU machines by running vh environments --gpu. Use the slug name provided in the output.

- step:
    name: yolo
    image: docker.io/ultralytics/ultralytics:8.0.180-python
    command: python train.py
    environment: aws-eu-west-1-p3-2xlarge

Run from the command line

When you’re ready, run your training on a remote cloud or on-premises machine by sending the job to the Valohai scheduler:

vh execution run yolo --adhoc --open-browser

A new browser tab will open, showing:

• Detailed execution information, including a snapcode of the code used for the job.
• All standard logs.
• Once the job completes, the trained model will be available in the outputs tab. It will be versioned and stored in the configured data store.