Weights & Biases for Autonomous Vehicles

Experiment tracking, model governance, collaboration

We use W&B for all our large-scale, serious experiments. The scalability, the customizability, live monitoring, not frozen reports of the conclusion — these features vastly improved our productivity.

Adrien Gaidon
Machine Learning Lead
Toyota Research Institute


Visualize point clouds

See the predictions and ground truth of your model in our interactive interface. You can explore the different predictions that your model makes over time, or compare across different model versions.


Our live dashboard allows you to collect results across members of your team and display them in one place. You can also build reports to share results across teams in your organization.

Massively scalable

We work with some of the largest machine learning teams in the world, and our product is built to scale to millions of experiments. We natively support distributed training.

Benchmark Model Performance

Decide on criteria, visualize performance, and customize queries to compare your model variants and focus on the right ones.

System of Record

Track everything in one place. W&B automatically logs every input into your training for regulatory compliance and reproducibility.


Use the Keras callback to automatically save all the metrics and the loss values tracked in model.fit.

  1. import wandb
  2. from wandb.keras import WandbCallback
  3. wandb.init(config={"hyper": "parameter"})
  4. # Magic
  5. model.fit(X_train, y_train,  validation_data=(X_test, y_test),
  6. callbacks=[WandbCallback()])


W&B provides first class support for PyTorch. To automatically log gradients and store the network topology, you can call watch and pass in your PyTorch model.

  1. import wandb
  2. wandb.init(config=args)
  3. # Magic
  4. wandb.watch(model)
  5. model.train()
  6. for batch_idx, (data, target) in enumerate(train_loader):
  7. output = model(data)
  8. loss = F.nll_loss(output, target)
  9. loss.backward()
  10. optimizer.step()
  11. if batch_idx % args.log_interval == 0:
  12. wandb.log({"loss": loss})


If you're already using TensorBoard, it's easy to integrate with wandb.

  1. import tensorflow as tf
  2. import wandb
  3. wandb.init(config=tf.flags.FLAGS, sync_tensorboard=True)

3D Point Cloud Bounding Boxes

Compare how different models perform on 3D object detection problems. In our interactive visualizations, you can explore point cloud scenes and compare ground truth and prediction bounding boxes.

2D Semantic Segmentation

Dynamically visualize the results of image segmentation models.

Never lose track of another model



Experiment tracking for machine learning models.

Create a free account
timer icon
Add just a few lines of code to your script to get the power of W&B logging.
We're framework agnostic.
dashboard icon
Easily log metrics from your script to visualize results in real time as your model trains.
search glass icon
Peek under the hood and see what your model is producing at each time step.


Scalable, customizable hyperparameter tuning.

Create free account

Parameter importance

Visualize which hyperparameters affect the metrics you care about. W&B comes with default visualizations that make it easy to get started without writing custom code to compare experiments.

Bayesian optimization

Use our transparent implementations of popular algorithms, or customize your own logic for sweeps.

Early stopping

We implemented the Hyperband algorithm to save GPU hours with customizable early stopping. This feature keeps the most promising, best performing runs running and kills off the bottom runs. Agents are then freed up to try new hyperparameter combinations.

Massive scale

Our sweeps can handle massive scale, and we support early stopping so you can quickly try thousands of hyperparameter combinations without wasting GPU hours.

eye / transparent icon
We cite all the algorithms that we’re using. We give you total information about what’s happening as the sweep is running, so you have complete control.
two gears icon
Quick setup
It’s easy to get started. We’ve dealt with the edge cases, so you don’t have to worry about concurrent runs and crashing runs.
thunder icon
Our sweeps are infinitely customizable. You can pick your own distribution for inputs, specify logic, and use early stopping.


Effortless pipeline tracking and production model management.

Create free account
pipeline icon
Pipeline management
Track every input and and output of your model, including data preparation, model dependencies, and evaluation.
dataset icon
Dataset versioning
Securely record every version of your dataset with just a line of code. See the lineage of production models.
two gears icon
Simple integration
Add just a few lines to your scripts and we’ll build a dependency graph for you, locally or in the cloud.