In recent years, artificial intelligence has rapidly advanced, offering businesses and developers an expanding range of tools to build intelligent systems. One of the most powerful platforms in this space is Vertex AI, Google Cloud’s end-to-end machine learning platform. A key question that often arises is: Which model does Vertex AI use? The answer to this question reveals the depth and versatility of Vertex AI, as it supports a variety of models ranging from pre-trained machine learning models to sophisticated large language models (LLMs).
Vertex AI is not tied to a single model. Instead, it provides a flexible infrastructure that supports multiple types of AI models. Depending on your use case, the platform allows you to either use a pre-built model, import a model trained elsewhere, or train your own model from scratch on the platform.
Types of Models Used in Vertex AI
There are several categories of models that can be utilized within Vertex AI:
- Custom-trained Models: Trained on your own data using TensorFlow, PyTorch, Scikit-learn, or XGBoost models. These can be brought into Vertex AI via containers or by using Vertex AI Training for fully-managed training jobs.
- AutoML Models: Uses Google’s proprietary AutoML technology to automatically generate models based on structured or unstructured data. This is ideal for users without deep machine learning expertise.
- Pre-trained Google Models: Vertex AI provides access to high-performance models that Google has already trained, including those for image analysis, natural language processing, and tabular data.
- Foundation Models and LLMs: Recently, Vertex AI has integrated large language models such as PaLM 2 and Gemini, which support generative AI tasks like summarization, translation, and code generation.
Google’s PaLM 2 and Gemini Models
As of 2023 and continuing into 2024, Vertex AI prominently leverages PaLM 2 (Pathways Language Model) and Gemini, Google’s next-generation multimodal AI system. These models are a significant step forward in natural language understanding and generative AI, and they are available in Vertex AI through the Generative AI Studio.
PaLM 2 supports over 100 languages and is optimized for reasoning, multi-turn dialogue, and code generation. Gemini improves on this foundation by adding features such as multimodal processing — enabling it to understand and generate content across text, images, audio, and more.
Developers can easily interact with these large models using APIs provided through Vertex AI. The interface includes prompt engineering tools, model tuning workflows, and output evaluation metrics that allow businesses to tailor the LLMs for specific domains and applications.
Integration with Open Source and OpenAI Models
While Vertex AI offers its own suite of proprietary Google models, it is also compatible with other open-source frameworks. Users can bring in models trained outside Google Cloud, and Vertex AI supports popular ML frameworks such as Hugging Face Transformers. Recently, there is increasing support for models like LLaMA, Falcon, and others in the open-source LLM community.
Notably, Vertex AI does not natively include OpenAI’s GPT models as part of its own offerings. However, users can choose to interconnect with those services through APIs outside the Google Cloud ecosystem if their project requires it.
How Vertex AI Chooses the Right Model for You
One of the advantages of Vertex AI lies in its intelligent design and tight integration with Google Cloud. When using the AutoML functionality, the platform evaluates the dataset and recommends or automatically trains a model suited to the problem domain. For more complex needs, custom training or leveraging foundation models is strongly encouraged.
Here’s how Vertex AI helps users select the most appropriate model:
- Problem Type Selection: Classification, regression, and forecasting needs might suggest a tabular model, whereas NLP problems may benefit from PaLM 2 or Gemini.
- Level of Expertise: For users with limited ML experience, AutoML offers a powerful out-of-the-box solution. Advanced users can configure custom containers for specialized needs.
- Data Structure: Structured data points to tools like Vertex AI Tabular, while image, audio, or text data favors foundation and vision models.
In summary, Vertex AI uses a diverse range of models, with the latest foundation models like PaLM 2 and Gemini setting a new standard for cloud-based AI. This flexibility allows developers and enterprises to harness the full power of machine learning and generative AI with a trusted Google infrastructure.
As artificial intelligence continues to evolve, platforms like Vertex AI are poised to remain at the forefront, empowering innovation across every industry sector.
