Unlocking the power of AI: Deploying open and third-party models on Google Cloud's Vertex AI

Prashanth Subrahmanyam, Head of Cloud Developer Advocacy at Google Cloud India, led an insightful workshop on Discover and Deploy Open and Third-Party Models on Vertex AI at YourStory’s inaugural tech summit, DevSparks 2024.

This exclusive event, focused on developers, engineers, and technologists, provided a platform for attendees to collaborate, learn, and explore emerging technologies.

In his workshop, Prashanth emphasised the explosion of the Generative AI (GenAI) ecosystem since late 2022, highlighting the importance of understanding and utilising the myriad of models now available. The workshop delved into various types of models, including open models, hosted models, third-party models, and proprietary models, all of which can be accessed through Google Cloud’s Vertex AI.

Key highlights from the workshop:

Understanding the model ecosystem

Prashanth began by discussing the rapid evolution of GenAI models and the challenges of keeping up with new developments. He stressed the importance of model discovery and selection, given the vast array of models available, each with unique capabilities and use cases.

Introduction to Model Garden

The session introduced participants to Model Garden on Google Cloud, a comprehensive repository where over 130 models from Google, third parties, and open-source communities are available. Model Garden helps developers explore and select the right models for their applications, ensuring they meet specific performance, cost, and compliance requirements.

Leveraging Vertex AI

Prashanth explained how Vertex AI on Google Cloud offers the infrastructure necessary for deploying these models. He highlighted the following features of Vertex AI:

Best-in-class infrastructure: Access to powerful GPUs and TPUs for deploying various models.

Model builder: Tools for fine-tuning models to meet specific business needs, including capabilities for tuning, distilling, and running notebooks.

Agent builder: Enables the creation of applications on top of models, with functionalities like Retrieval-Augmented Generation (RAG) and function calling for real-time data access.

Selecting the right model

Prashanth provided guidance on how to choose the right model based on factors such as:

Open vs. hosted models: Weighing the flexibility of open models against the managed, enterprise-ready nature of hosted models.

Model size and performance: Balancing the power and cost of large models against the efficiency of smaller models.

Model source and ethics: Ensuring models are responsibly trained and maintained, with considerations for data ethics and compliance.

The workshop also included practical advice and examples, such as:

Integration with Hugging Face: How Google Cloud's partnership with Hugging Face allows seamless deployment of models from Hugging Face on Google Cloud.

Use case scenarios: Evaluating whether a use case requires a high-powered model or if a smaller, more efficient model would suffice.

Networking and future of tech: In addition to the technical deep dive, the workshop provided opportunities for participants to network with peers and industry experts, fostering discussions on real-world applications and the future of technology.

Prashanth Subrahmanyam’s workshop at DevSparks 2024 offered a comprehensive overview of how developers can discover and deploy open and third-party models using Google Cloud's Vertex AI. The session underscored the importance of selecting the right models, leveraging the right infrastructure, and understanding the ethical implications of AI.

You can view the full session in the below detailed video,