Demos

Interactive notebooks for exploring ThirdAI's BOLT library.

[Website] · [Docs] · [Report Bug] · [We're Hiring]

Table of Contents

1. Welcome
2. Quickstart
   • Downloading a License
   • Installation
3. Usage
4. License
5. Contact

👋 Welcome

ThirdAI's BOLT library is a deep-learning framework that leverages sparsity to enable training and deploying very large scale deep learning models on any CPU. This demo repo will help get you familiar with BOLT's Universal Deep Transformer (UDT) through interactive notebooks.

(back to top)

🚀 Quickstart

Step 1: Running a UDT Demo

All of our UDT capability demos are mirrored to Google Colab, so you can immediately run any of them by clicking the associated link:

• CensusIncomePrediction.ipynb shows how to build an income prediction model with ThirdAI's Universal Deep Transformer (UDT) model, our all-purpose classifier for tabular datasets.
  https://colab.research.google.com/github/ThirdAILabs/Demos/blob/main/classification/CensusIncomePrediction.ipynb
• ClickThroughPrediction.ipynb shows how you can use UDT to achieve SOTA AUC on Click Through Prediction.
  https://githubtocolab.com/ThirdAILabs/Demos/blob/main/classification/ClickThroughPrediction.ipynb
• EmbeddingsAndColdStart.ipynb takes care of your most NLP, search, and recommendations needs on unstructured raw text. Learn with simple commands how to train large neural models on raw text to perform search, recommendations, and generate entity emebeddings as well as embeddings for any text. Yes, all (training, inference, and retraining) on simple CPUs.
  https://githubtocolab.com/ThirdAILabs/Demos/blob/main/embeddings/EmbeddingsAndColdStart.ipynb
• IntentClassification.ipynb will show you how to get near SOTA accuracy on most text classification via a plug and play classifier at any given budget (everything autotuned).
  https://githubtocolab.com/ThirdAILabs/Demos/blob/main/classification/IntentClassification.ipynb
• GraphNodeClassification.ipynb will show you how to build the fastest graph neural network beat the SOTA accuracy on Graph Node Classification via a plug and play classifier at any given budget (everything autotuned).
  https://colab.research.google.com/github/ThirdAILabs/Demos/blob/main/graph_neural_networks/GraphNodeClassification.ipynb
• FraudDetection.ipynb will show you how easy to build a fraud detection model with UDT.
  https://githubtocolab.com/ThirdAILabs/Demos/blob/main/classification/FraudDetection.ipynb
• PersonalizedMovieRecommendations.ipynb will show you how to build personalization model for movie recommendation. UDT can be used to build any kind of personlization and recomnedation models with ease and deliver SOTA results.
  https://githubtocolab.com/ThirdAILabs/Demos/blob/main/personlization_and_recommendation/PersonalizedMovieRecommendations.ipynb
• QueryReformulation.ipynb shows how to build a query reformulation model with UDT, providing an easy and faster (less than 1 ms) solution for query reformulation.
  https://colab.research.google.com/github/ThirdAILabs/Demos/blob/main/QueryReformulation.ipynb
• SentimentAnalysis.ipynb will take you through the process of creating a network to use during sparse training and sparse inference with the goal of predicting positive/negative sentiment.
  https://githubtocolab.com/ThirdAILabs/Demos/blob/main/classification/SentimentAnalysis.ipynb
• TrainingDistributedUDT.ipynb shows how you can use ThirdAI's UDT in distributed setting using Ray cluster. For this demo, we are using clinc-small for training and evaluation.
  https://github.com/ThirdAILabs/Demos/blob/main/distributed/TrainingDistributedUDT.ipynb

You can also clone this repo and run any of these demo notebooks on any CPU (ARM, AMD, Intel), and even desktops and laptops

Step 2: Integrating UDT with an Existing Platform

We also have demos explaining how to integrate UDT with different platforms you may already be comfortable with:

• integrations/DeployThirdaiwithDatabricks.ipynb will show how to use thirdai for inference on Databricks with UDT.

(back to top)

Step 3: Trying UDT on Your Own Dataset

Each of these notebooks has an API key that will only work on the dataset in the demo. If you want to try out ThirdAI on your own dataset, simply register for a free license here. We look forward to hearing from you!

(back to top)

📄 License

Please refer to LICENSE.txt for more information on usage terms.

(back to top)

🎙 Contact

ThirdAILabs - @ThirdAILab - [email protected]

Project Link: https://github.com/ThirdAILabs/Demos

(back to top)