Unlocking the Power of Foundation Models on Google Cloud

Table of Contents

1. Introduction
2. What are Foundation Models?
3. Characteristics of Foundation Models
4. Accessing Foundation Models via APIs in the Vertex AI Model Garden
5. Text-Based Foundation Models
   • Natural Language Tasks
   • Dialogue Models
6. Code Completion and Generation
7. Image Generation
8. Media Models
9. Embeddings
10. Customizing Foundation Models
11. Conclusion

Introduction

Artificial intelligence has come a long way in recent years, and foundation models are the latest development in this field. These large AI models can be adapted to a wide range of tasks and can generate high-quality output. In this article, we will explore what foundation models are, their characteristics, and how they can be accessed via APIs in the Vertex AI Model Garden. We will also Delve into the different types of foundation models available, including text-based models, code completion and generation models, image generation models, media models, and embeddings.

What are Foundation Models?

Foundation models are large AI models that can be adapted to a wide range of tasks. They are typically trained on vast amounts of diverse data, allowing them to learn general Patterns and representations that can be applied across various domains and tasks. Unlike previous generations of AI models, foundation models are multitask rather than single task. One foundation model can perform a wide array of tasks out of the box, such as summarization, question and answering, and classification.

Characteristics of Foundation Models

Foundation models have several important characteristics that represent a change from the previous generation of AI models. They are multitask rather than single task, and they can perform a wide array of tasks out of the box. There are different modalities for data types, such as images, text, code, and more. Foundation models are typically trained on vast amounts of diverse data, allowing them to learn general patterns and representations that can be applied across various domains and tasks. With no or minimal training required, foundation models can work well out of the box and can be adapted for targeted use cases with very little example data.

Accessing Foundation Models via APIs in the Vertex AI Model Garden

Before now, foundation models were difficult to access, required specialized machine-learning skills, and compute resources to use in production. Now, developers can access these foundation models via APIs in the Vertex AI Model Garden. The Model Garden is where You go to discover both Google and external models. The garden includes models from Google Cloud, Google Research, and external sources for a variety of data formats, including chat and dialogue, co-generation and completion, images and embeddings.

Text-Based Foundation Models

Text-based foundation models allow you to perform natural language tasks with zero and few shot prompting. Our gallery of Prompts gets you started with tasks such as summarization, entity and information extraction, and idea generation, all in a place for you to experiment with your own prompts in the Studio.

Natural Language Tasks

Natural language tasks include summarization, entity and information extraction, and idea generation. These tasks can be performed with zero and few shot prompting, making them easy to use out of the box.

Dialogue Models

Dialogue models are text-based but have been specifically trained to work in conversation. Dialogue models allow you to have multi-turn conversations with preserved Context. This is useful for an in-browser or mobile assistant to help improve the customer experience for answering questions, summarization, and tuned for your custom domain.

Code Completion and Generation

Code completion and generation models can be used to make super-charged coding assistance. You can give a natural language prompt to describe a piece of code you like written, or use the completion model via the API, and an extension that could take a partial code snippet as input, and then suggest the output.

Image Generation

With image generation models, you can Create and edit images to your specifications. We also have models for working with media, such as classification, object detection, and more. And built into the product, we have content moderation for responsible AI safety practices.

Media Models

Media models are designed to work with different types of media, such as images, videos, and audio. These models can be used for classification, object detection, and more.

Embeddings

Embeddings are vector representations of data. For example, a text embedding is a numerical vector representation of a word or a phrase. This model allows you to extract semantic information for unstructured data. You can use this to power recommendation engines, ad-targeting systems, complex classification tasks, search, and more.

Customizing Foundation Models

By using and customizing the foundation models available in Vertex AI and the Model Garden, you can access a whole new level of features and functionality, possibilities to help make your apps more intuitive, personalized, and effective at meeting your users' needs.

Conclusion

Foundation models are a game-changer in the field of artificial intelligence. They are large AI models that can be adapted to a wide range of tasks and can generate high-quality output. With the ability to access these models via APIs in the Vertex AI Model Garden, developers can now take AdVantage of these powerful tools to create more intuitive, personalized, and effective apps.