Enterprise AI projects struggle at two critical points: initiation and deployment. Initiation involves time-consuming and surprisingly complex technical steps. And when a model is finally ready for deployment, industry surveys tell us that most AI models never make it to production.
Domino already enables enterprises to scale models to production quickly. It lets data scientists deploy models as APIs or dashboard apps with minimal IT involvement. With Domino's Fall '23 release, we are now tackling the challenge of starting your AI project.
Starting a new AI project involves several challenges. First, you need to figure out which tools and frameworks can address your use case. You need to identify the proper hardware infrastructure. You need to select code packages and then avoid or resolve version conflicts. Then, you must also ensure the new project follows the enterprise's coding and security procedures. This list is likely longer at your company, and you must follow it for every new project.
Domino's AI Hub will help you jumpstart your next AI project.
AI Hub: An Overview
AI Hub is a repository of curated, templated projects from Domino and its AI ecosystem partners, with leaders including NVIDIA, AWS, and Deci.ai contributing. The template collection covers a growing array of domains and use cases. They help data science teams avoid reinventing the wheel. You only need to pick a template, and Domino will generate a project around it. Each template includes:
- A state-of-the-art model
- A verified Domino environment definition
- Sample source code
- Sample data
- Infrastructure recommendation
Templates save you time and give you repeatable, validated starting points to use as project foundations. What's left is to adapt the project to your unique requirements, intellectual property, and data. See AI Hub in action below.
Build Your AI Center of Excellence
AI Hub enables enterprises to use templates available in the hub. They can customize them to their requirements and even create new project templates to add to their local Domino instance’s collection. These templates will be visible only to members of their organization. You can share source code, data samples, and best practices in your corporate project templates. You can also incorporate required corporate coding and security standards.
Users can contribute to their private hubs seamlessly in their Domino environment by following these steps:
- Specify the Domino environment the project runs on
- Collect the project source code
- Point at a Git repo for the template's code
- Provide a template overview document and usage instructions
You retain complete control over your templates: Template access, permissions, and security settings are as robust as the ones implemented for any project in Domino.
Curated AI Ecosystem Innovation
The hub gives data scientists the latest and greatest open-source and commercial AI innovations. Besides accelerating development, they can also see the optimal use cases for the model at the center of each project.
The projects include both traditional and generative AI models. The models come from Domino and leading partners like AWS, Hugging Face, OpenAI, Meta, Fiddler, Deci, Artefact, KSM Technology Partners., and more.
Here's a sneak peek at projects making their way into AI Hub:
- Enterprise Document Q&A with RAG and OpenAI's API
- Text Summarization Fine-Tuning: Fine-tune the Falcon LLM to offer dialog summaries. The template uses Hugging Face Trainer to improve domain-specific task performance. The template even includes apps you can deploy as model UI in Domino.
- Chatbots and Content Creation Assistants: DeciLM is a 6-billion parameter LLM offering 15x Llama 2 7B's throughput while maintaining output quality.
- Text Generation Fine-Tuning: Fine-tune the Falcon LLM on a conversational dataset and use prompts to generate outputs.
- Prompt Tuning for Multiple Tasks with NVIDIA NeMo: Prompt-tuning a single LLM for multiple tasks using NVIDIA's NeMo™ framework. Analyze sentiment, intent, and slot classification tasks. The template even shows you how to add a Q&A task.
- Object Detection: Deci's YOLO-NAS offers exceptional real-time detection, 50% higher throughput, and better accuracy, even on older NVIDIA T4 GPUs.
- Product Feedback and Response Summary: Summarize product feedback and generate email text for customer service use cases. The template works with Amazon Titan Bedrock and Anthropic foundation models and uses LangChain RAG.
- Code Generation: DeciCoder is a fast and lightweight code-generation LLM. It delivers high accuracy even on 'budget' GPUs.
- LLM Safety and Performance Evaluation: Assess LLM robustness, accuracy, and safety using Fiddler Auditor. The template will also help you perform red-team tests in pre-production on your LLMs.
- Defect Detection: Detect anomalies and perform other image processing tasks for quality control and classification.
- Image Generation: DeciDiffusion is a text-to-image model with 3x the speed of Stable Diffusion v1.5 with the same image quality and using 40% fewer iterations.
Yet getting started faster with your project is only a part of the story. Besides offering self-serve tools and the ability to collaborate, reuse, and deploy models, Domino also provides teams with its Model Sentry framework.
Model Sentry helps enterprises follow responsible AI practices. It builds on Domino's reproducibility engine, adds automated model cards, and tracks production models in its model registry. It helps validate models with staging functionality and facilitates model review and approval. Once in production, Domino monitors model performance to ensure they remain valid.
Future-Proof Your AI Initiatives
As Domino continues to expand its partner and solution ecosystem, so will the AI Hub. The hub is poised to become an indispensable resource for enterprises looking to fast-track their AI projects and ensure they leverage the latest innovations for their initiatives. As enterprises around the globe race to operationalize generative AI models responsibly and quickly, the AI Hub emerges as a catalyst, orchestrating a conducive ecosystem for AI deployment at scale.
Our Fall 2023 release is a stride towards eliminating the barriers to AI operationalization. Explore the Domino AI Hub and discover how it can accelerate your AI journey.