Quick Start

Get up and running with Dalva in 5 minutes.

1. Install Dalva

# uv
uv add dalva

# pip
pip install dalva

2. Start the Server

dalva server start

The server will automatically find an available port and display the URL.

Important

The Dalva server must be running before you can log experiments. The Python SDK communicates with the server via HTTP.

3. Log Your First Experiment

Create a new file my_experiment.py:

import dalva

# Initialize a run with nested config
run = dalva.init(
    project="quickstart",
    name="first-experiment",
    config={
        "optimizer": {"name": "adam", "lr": 0.001},
        "batch_size": 32,
        "epochs": 100,
    },
    server_url="http://localhost:8000"
)

# Log metrics during training — nested dicts are flattened with '/' separator
for step in range(100):
    loss = 1.0 / (step + 1)
    accuracy = min(0.95, step / 100)

    run.log({"train": {"loss": loss, "accuracy": accuracy}}, step=step)
    # Equivalent to: run.log({"train/loss": loss, "train/accuracy": accuracy}, step=step)

# Retrieve metrics:
run.get("train/loss", step=0)   # {"key": "train/loss", "value": 1.0, "step": 0}
run.get("train/accuracy")       # latest step value

# Retrieve config:
run.get_config("optimizer/lr")  # {"key": "optimizer/lr", "value": 0.001}

# To overwrite a value, remove it first then re-log:
run.remove("train/loss", step=0)
run.log({"train": {"loss": 0.9}}, step=0)

run.finish()
print("Experiment logged successfully!")

Run the experiment:

python my_experiment.py

4. View Results

Open your browser to http://localhost:8000 to see your experiment.

5. More Examples

See examples/nested_metrics_and_config.py for a complete walkthrough covering nested metrics, get/remove/relog patterns, and nested config.

See examples/fork_run.py for forking runs, and examples/fork_run_full.py for a comprehensive test of all fork features.