IBM unveiled plans to enhance its AI platform, watsonx, with new generative AI foundation models and features. These new features will be available for developers at TechXchange, IBM's premier technical learning event, set for Sept. 11-14 in Las Vegas, according to a Sept. 7 news release.
"As demonstrated by the ongoing rollout of the watsonx platform within just a few months since launch, we are here to support clients through the entire AI lifecycle" Dinesh Nirmal, senior vice president of products, IBM Software, said in the release. "As a transformation partner, IBM is collaborating with clients to help them scale AI in a trustworthy way – from helping to institute foundational elements of their data strategies to tuning models for their specific business uses cases to helping them govern models beyond that."
These additions include watsonx.governance's technical preview, generative AI data services for watsonx.data and the integration of watsonx.ai foundation models into select software and infrastructure products, allowing developers to access these capabilities, the release reported.
IBM's new generative AI models for watsonx.ai include the Granite series models, slated for release later this month. These models utilize the "Decoder" architecture to support enterprise natural language processing tasks such as summarization, content generation and insight extraction, the release said.
Additionally, IBM will provide a detailed list of data sources and processing steps used to create the training data for the Granite series models, with availability planned for the third quarter of 2023, according to the release.
IBM is also introducing third-party models, including Meta's Llama 2-chat 70 billion parameter model and the StarCoder large language model for code generation in watsonx.ai on IBM Cloud. The company implemented a rigorous training process for its foundation models, emphasizing principles of trust and transparency, the release reported.
This process covers data collection and culminates in control points for responsible model deployment, governance, risk assessment, privacy, bias mitigation and compliance, the release said.
IBM's plans for the watsonx platform extend to watsonx.ai, watsonx.data and watsonx.governance, according to the release.
Watsonx.ai will see the introduction of the Tuning Studio, offering prompt tuning for adapting foundation models to unique downstream tasks. A synthetic data generator is also being launched to help create artificial tabular data sets for AI model training, the release reported.
Watsonx.data will receive generative AI capabilities to enhance data discovery, augmentation, visualization and refinement through a self-service experience powered by natural language interaction. A vector database capability will be integrated to support retrieval augmented generation use cases, the release said.
Watsonx.governance is launching a tech preview focused on model risk governance for generative AI. This feature automates the collection and documentation of foundation model details and model risk governance capabilities, according to the release.
