https://github.com/saturncloud/weights-and-biases

https://github.com/saturncloud/weights-and-biases

Last synced: about 1 month ago
JSON representation

• Host: GitHub
• URL: https://github.com/saturncloud/weights-and-biases
• Owner: saturncloud
• Created: 2021-04-13T15:06:53.000Z (about 5 years ago)
• Default Branch: main
• Last Pushed: 2021-05-10T18:53:24.000Z (about 5 years ago)
• Last Synced: 2024-04-16T02:03:26.386Z (about 2 years ago)
• Language: Jupyter Notebook
• Homepage: https://saturncloud.io/docs/examples/python/weights-and-biases/qs-wandb/
• Size: 32.2 KB
• Stars: 1
• Watchers: 2
• Forks: 1
• Open Issues: 0
• Metadata Files:
  • Readme: README.md

Awesome Lists containing this project

README

          
# Use Weights and Biases on Saturn Cloud

## Welcome!

This project provides examples showing how to use  Weights and Biases in Saturn Cloud, to monitor 

the training and performance of your machine learning models. We have included single-node 

training of an image classifier, as well as a multi-node example that lets you see how nicely 

Weights and Biases can integrate with distributed training on Dask clusters. 

## [Train image classifier on single node (PyTorch)](train-pytorch-singlenode.ipynb)

In this notebook, you'll train an image classifier with PyTorch and use Weights and Biases to

 monitor model performance.

## [Train image classifier on cluster (PyTorch)](train-pytorch-cluster.ipynb)

This notebook expands upon the image classifier training task, using a Dask cluster to accelerate

 the same task. Weights and Biases is still able to easily and clearly provide model performance monitoring.

***

## How to Use

To run these examples on Saturn Cloud, please follow these instructions.

1. **Create Account**: If you haven't already, [create a Saturn Cloud account and sign in](https://www.saturncloud.io/docs/getting-started/start_in_ten/). 

2. **Add Weights and Biases Credential**: Open the Credentials tab in the left side menu, and [add your Weights and Biases user token as an Environment Variable](https://www.saturncloud.io/docs/getting-started/credentials/). Name it `WANDB_LOGIN`. (This is the [same token you would use if you logged in to Weights and Biases](https://docs.wandb.ai/ref/cli/wandb-login) at the command line.)

3. **[Create a custom project](https://www.saturncloud.io/docs/getting-started/start_project/#create-a-custom-project)**: The specifications we recommend are:

     * T4 GPU instances

     * The `saturncloud/saturn-pytorch:2021.02.22` image

     * Keep default disk space and "Shutoff After" settings

     * Add the following to your Advanced Settings > Start Script box.

       ```

       pip install wandb dask-pytorch-ddp

       wandb login --relogin $WANDB_LOGIN

       ```

4. **Optional: Create a Cluster**: If you plan to run the cluster based example, [create a cluster in the project](https://www.saturncloud.io/docs/getting-started/create_cluster_ui/) as well. Requesting at least 3 workers is recommended. 

5. **Start Resources**: Start the Jupyter Instance. Clone this repository inside your Jupyter instance at terminal, as you would in a local environment.

***

## Next steps

Thanks for trying out these examples! To learn more about how Saturn Cloud works, check out our  Documentation, blog, or join an  upcoming event. 

To learn more about Weights and Biases, visit  their website and documentation.

If you have any questions or suggestions for example projects, reach out to us at support@saturncloud.io or open an issue on the [examples Github repo](https://github.com/saturncloud/examples).