watsonx.ai

IBM announces the general availability of watsonx.ai version 2.1 (December 11, 2024). This software update includes several new features and foundation models as well as feature enhancements.  Watsonx.ai is an enterprise-grade AI development studio that runs on an open and trusted hybrid cloud infrastructure and helps developers operationalize and scale the development of AI applications by bringing together traditional machine learning and generative AI capabilities. Watsonx.ai gives the developer choice when selecting and customizing models as well as choice to deploy when, where and how they want.

Below are the highlighted features in this update that will help improve retrieval augmented generation (RAG) use cases, introduce multi-model capabilities and a collection of new foundation models to accelerate the value of AI across your organizations. If some features seem familiar, it's because they were released in IBM Cloud in advance of this software update. Full details of this release can be found in our “What’s New” overview page.

Highlights of watsonx.ai v2.1

Accelerate RAG Pattern Development with AutoAI

Introducing AutoAI for Retrieval-Augmented Generation (RAG) development. This innovative tool streamlines the evaluation and selection of RAG configuration parameters, such as LLMs, embedding models, and chunking, to optimize your AI use cases. By automating the evaluation process to find the best performing set of configurations, the AutoAI RAG feature dramatically reduces the time and resources required to develop and deploy an optimized, production-quality RAG solution. Another bonus is that you can perform rapid re-evaluation when something changes, for example when a new model version becomes available or when production evaluation results signal a change in the quality of responses. Now you can quickly re-evaluate and make informed decisions about what changes should be incorporated into your solution. This feature has an intuitive user interface, you can also leverage the API to perform the same functions. Learn more in documentation.

Boost Answer Retrieval Precision with Text Rerank API

The Text Rerank API enhances search and retrieval tasks by reordering document passages (from most-to-least likely to answer) based on their similarity to a specified query, adding precision to answer retrieval workflows such as those in retrieval-augmented generation (RAG) use cases. This feature is powered by the ms-marco-minilm-l-12-v2 model that has also been added to the platform in this update. This feature is accessible through the watsonx.ai REST API. Learn more in documentation.

Extract Text from Complex Documents with Ease

The new text extraction API enables developers to extract content from complex document structures, such as images, diagrams, and tables, and transforming them into simpler formats such as markdown or JSON making it possible to programmatically incorporate this data into AI solutions. With this new API, developers can easily incorporate relevant information into model input, enhancing the overall quality of the model's output. This process is particularly useful for retrieval-augmented generation (RAG) tasks, where accurate contextual information is crucial for generating factual and up-to-date output. Learn more in documentation. 

Ground Model Responses with Contextual Information

In the Prompt Lab, you can now add contextual information to foundation model prompts enabling the model to generate more factual and up-to-date responses in retrieval-augmented generation (RAG) use cases. This feature (a.k.a. “Chat with Documents”) allows you to quickly upload and automatically "vectorize" documents which will then be used to retrieve relevant grounding information to be added to model prompts. This new feature in the Prompt Lab greatly accelerates the prototyping of RAG use cases. Learn more in documentation. 

Build Conversational AI

The new Chat API implements methods for interacting with foundation models in a conversational way, such as the way users currently interact with a foundation model via the chat mode tab in the Prompt Lap. Developers can identify different message types, such as a system prompt, user inputs, and foundation model outputs, including user-specific follow-up questions and answers. By leveraging the Chat API, developers can also build agent-driven applications where the foundation model controls the interaction flow with the user. This functionality, also known as tool calling or function calling, enables the creation of more dynamic and adaptive conversational interfaces. Learn more about adding chat functions and building agents in documentation. 

Unlock Multi-Modal Capabilities with Vision Models

We're excited to introduce multi-modal capabilities in watsonx.ai that enable the inference of vision models, opening up a range of computer vision use cases, including image classification, object detection, transcription, and more. To support this feature, we've added two new models: llama-3-2-11b-vision-instruct, and llama-3-2-90b-vision-instruct. We’ve also added llama-guard-3-11b-vision which can help with detecting harmful image content. With these advancements, you can now engage in image-to-text interactions either through the Prompt Lab (chat mode tab) or the Chat API, by simply adding an image to your prompt. Learn more in documentation.

Leverage New Foundation Models

Besides the four new models listed above, a host of additional foundation models are now available to provide choice and flexibility for your generative AI use cases: 

See the full list of supported foundation models here.

To get started with watsonx.ai, visit the product page and start a free trial, then visit the Developer Hub to leverage quickstarts and examples designed to help you jumpstart your development of AI based solutions.

#watsonx.ai
#AutoAI
#GenerativeAI

#PromptLab