The AI Hub is a valuable resource for quickly deploying AI solutions and developing models to drive innovation within your organization. It is designed to speed up AI development for various business tasks and industries, providing tools, best practices, and resources for projects ranging from traditional predictive models to cutting-edge Generative AI.
These templates focus on efficiency and accessibility, empowering organizations to create, share, and customize AI solutions tailored to their specific needs while gaining valuable insights into Domino’s features.
This document provides a summary of the AI Hub templates available in Domino:
You can find this project on GitHub.
This template guides you through building a chatbot by fine-tuning Meta’s state-of-the-art Llama3 model using Supervised Fine Tuning (SFT).
The model undergoes domain-specific adaptive fine-tuning, enhancing its focus and alignment by training on the guanaco-llama2-1K dataset. You can interact with the resulting model via a Streamlit application or an API.
For a more detailed exploration of the project, visit the Domino website.
You can find this project on GitHub.
This project trains a predictive model on a Supervisory Control and Data Acquisition (SCADA) dataset collected from a physical wind turbine. SCADA systems are used to control, monitor, and analyze industrial devices and processes.
Our repository provides a step-by-step notebook for training a machine learning model using an Extra Trees Regressor and the freely available SCADA dataset from Kaggle.
To explore the project in more detail, visit the Domino website.
You can find this project on GitHub.
This template provides a guide on leveraging the open-source LangChain framework, OpenAI’s language models, and either Facebook AI Similarity Search (FAISS) or Pinecone to build an interactive search engine capable of performing Q&A on information not included in OpenAI’s training data.
It utilizes a Retrieval-Augmented Generation (RAG) framework integrated within a Streamlit web application.
Prerequisites:
-
OPENAI_API_KEY
, required -
PINECONE_API_KEY
, optional
To explore the project in more detail, visit the Domino website.
You can find this project on GitHub.
This template demonstrates how to fine-tune DistilBERT — a lighter, faster variant of HuggingFace’s BERT — using the Amazon Polarity dataset and interacting with a model that determines sentiment on product reviews.
With this template code, your team can learn to import libraries, check NVIDIA GPU acceleration availability, load DistilBERT, read a CSV dataset, prepare training, test, and validation subsets, fine-tune a model, and output an F1 metric.
For a more detailed exploration of the project, visit the Domino website.
You can find this project on GitHub.
This template uses different inference frameworks to generate text output from Falcon-7b, a fine-tuned Large Language Model (LLM). Additionally, this template guides the deployment of the fine-tuned LLM as a Domino endpoint and a Streamlit app in Domino.
This project utilizes a ctranslate2 model, which offers optimized implementations for various hardware, including CPUs and GPUs, which makes it faster and more resource-efficient than many other inference engines.
For a more detailed exploration of the project, visit the Domino website.
You can find this project on GitHub.
This reference project shows how to fine-tune the Falcon-40b parameter Large Language Model (LLM) on a dataset to summarize conversations using the Hugging face Trainer. Falcon-40B can create a wide range of contextually accurate content to generate high-quality natural language outputs like blogs, emails, and text translations.
In this project, we will use the model’s 4-bit and 8-bit quantized version and train a LoRA adapter.
For a more detailed exploration of the project, visit the Domino website.
You can find this project on GitHub.
In this template, we generate an email response to a customer who has provided negative feedback on services received from a customer support engineer.
The three notebooks demonstrate how to provide feedback using three different techniques:
-
First, the Amazon (AWS) Titan, a large language model, and Bedrock API utilize a zero-shot prompt without context as an instruction for the model.
-
Second, Anthropic’s Claude text model uses the Langchain framework integration with Bedrock and uses a zero-shot prompt without context.
-
The third notebook provides additional context to the prompt, including the original customer email from LangChain.
Prerequisites:
-
AWS_DEFAULT_REGION
, required -
AWS_PROFILE
, required -
BEDROCK_ASSUME_ROLE
, required
For a more detailed exploration of the project, visit the Domino website.
You can find this project on GitHub.
In this project, we apply anomaly detection for quality control and defect detection.
We fit a PaDIM model against the MVTecAD dataset using Anomalib, a comprehensive deep-learning library designed to serve as a hub for state-of-the-art anomaly detection algorithms. The Domino platform can expose the persisted model as a Domino endpoint. The score function accepts an image path as an argument and returns a boolean prediction (anomalous or not) and a confidence score of the prediction.
To learn more, visit the Domino website.
You can find this project on GitHub.
Managing and analyzing large volumes of clinical information efficiently and securely is critical for innovation and regulatory compliance. While GenAI is promising, traditional information management methods often fall short, leading to inefficiencies, increased costs, and potential risks to patient safety.
Enter BioRAG (RAG - Retrieval Augmented Generation), an advanced AI solution in Domino’s AI Hub, designed to transform how clinical data is handled and utilized. This template utilizes LangChain, Qdrant, a high-performance vector database, and Azure Blob Storage for document storage to allow users to interact with their documents via a streamlit application.
Prerequisites:
-
QDRANT_URL
,QDRANT_KEY
, required -
AZURE_EMBEDDINGS_DEPLOYMENT_NAME
,AZURE_EMBEDDINGS_API_KEY
,AZURE_EMBEDDINGS_MODEL_NAME
,AZURE_EMBEDDINGS_ENDPOINT
, required -
OPENAI_API_VERSION
,OPEN_AI_TYPE
, required -
AZURE_CHAT_ENDPOINT
,AZURE_CHAT_DEPLOYMENT_NAME
,AZURE_CHAT_API_KEY
,AZURE_CHAT_MODEL
,AZURE_BLOB_CONTAINER_NAME
,AZURE_BLOB_CONNECTION_STRING
, required
To learn more, visit the Domino website.
You can find this project on GitHub.
This template presents three unsupervised algorithms to detect anomalies in CPU utilization streaming data using the Numenta Anomaly Benchmark dataset. This open-source dataset comprises over 50 labeled real-world and artificial time series data files, plus a novel scoring mechanism designed for real-time applications.
The three algorithms presented by the templates are moving average, exponential moving average, and isolation forest.
For a more detailed exploration of the project, visit the Domino website.
You can find this project on GitHub.
This project demonstrates a sample AI training script using Domino Flows. A sample data set is provided so users can run a flow by executing a provided command and utilizing the Flyte console.
The sample flow contains two tasks - one for data preparation and one for model training. Each task ultimately triggers a Domino Job and returns the outputs.
Prerequisites:
-
Domino version 5.11.0.
For a more detailed video explanation of the project, visit the Domino website.
You can find this project on GitHub.
Track and control AI infrastructure spending and save without manually tagging all infrastructure assets or reconciling cloud bills. Obtain critical insights into computing and storage spending by users, projects, organizations, computing clusters, and any dimension through clear and accurate cost breakdowns within the Domino platform.
Prerequisites:
-
Admin permissions in the Domino platform.
-
Storage Configuration Information (i.e., bucket, endpoint, region, etc.).
To learn more, visit the Domino website.
Note
| Disclaimer - Domino Reference Projects are starter kits built by Domino researchers. They are not officially supported by Domino. Once loaded, they are yours to use or modify as you see fit. We hope they will be a beneficial tool on your journey! |
You might also find the following Domino products and topics helpful: