SwanLab Quick Start Guide

Edit Me

I've been running a lot of experiments lately and things were getting messy fast — results scattered everywhere, and I was seriously considering just tracking everything in Excel.

Then I remembered a teammate from group 9444 mentioning SwanLab for experiment tracking and visualization. So I put together this quick start guide — it'll be part of my own experiment pipeline going forward.

If you have a better system for managing research experiments, feel free to share it.

Anyway, let's get into SwanLab. Here's the quick start workflow for experiment tracking with SwanLab.

1. Create an Account & Get Your API Key

First, open the SwanLab website: https://swanlab.cn/ If you don't have an account yet, register on the site first.

Once you're logged in, create a project. You'll see a Quick Start guide — just follow it step by step and your API key will be right there. I've copied the key steps here for convenience.

2. Install the SwanLab Library

In any environment with Python 3, install the SwanLab client library via pip:

pip install swanlab

3. Log In to SwanLab

Run the following command in your terminal:

swanlab login

You'll see a prompt like this:

swanlab: Logging into swanlab cloud.
swanlab: You can find your API key at: https://swanlab.cn/settings
swanlab: Paste an API key from your profile and hit enter, or press 'CTRL-C' to quit:

Paste the API key you copied from your profile settings page and you're in. (docs.swanlab.cn)

4. Submit an Experiment

import swanlab
import random

# Initialize a new SwanLab run to track this script
swanlab.init(
  # Set the project this run belongs to
  project="mteb-ailastatue",
  workspace="mira",
  # Track hyperparameters and run metadata
  config={
    "learning_rate": 0.02,
    "architecture": "CNN",
    "dataset": "CIFAR-100",
    "epochs": 10
  }
)

# Simulate training
epochs = 10
offset = random.random() / 5
for epoch in range(2, epochs):
  acc = 1 - 2 ** -epoch - random.random() / epoch - offset
  loss = 2 ** -epoch + random.random() / epoch + offset

  # Log training metrics to SwanLab
  swanlab.log({"acc": acc, "loss": loss})

# [Optional] Mark the run as finished — required in notebook environments
swanlab.finish()