Highlights:

  • IBM provides pre-trained models, both proprietary and from the Hugging Face community, supporting various NLP tasks.
  • Foundation models like Llama train on vast unlabeled data for versatile fine-tuning.

IBM Corp. recently announced that the watsonx.ai studio enterprise artificial intelligence development platform, which it announced in May, will host Meta Platforms Inc.’s Llama 2, a 70-billion-parameter large language model. Select clients will gain early access.

With watsonx.ai, users can train generative AI models and other neural networks without starting model creation from scratch. The platform contains a library of prepackaged AI models as well as datasets for training them.

In February, Meta stunned the AI world by making Llama open source. Llama and similar foundational models usually undergo training on extensive unlabeled datasets, enabling fine-tuning for various tasks. Llama 2, formally known as Large Language Model Meta AI 2-chat, was made open source by Meta recently to allow businesses to customize their own AI models with proprietary data and tools.

Collaboration Is Ongoing

IBM and Meta have previously collaborated on several open-source projects, including the PyTorch machine learning framework and the Presto query engine used in watsonx.data, a purpose-fit data store built on an open lakehouse architecture.

IBM’s approach involves offering a combination of third-party and proprietary AI models. Presently, it provides pretrained models from both its collection and the Hugging Face Inc. community. These models cater to various natural language processing tasks, including question-answering, text classification and extraction, and content generation and summarization. The company said it would soon release an AI tuning studio, fact sheets for watsonx.ai models, and its generative AI models.

In response to enterprise concerns about data security and privacy, IBM announced that users running the Llama-2 model through the prompt lab in watsonx.ai could use a guardrails function to automatically remove harmful language from input and output text. Meta also provides documentation for the development of responsible AI models.

IBM is strongly pushing to become the preferred technology provider for enterprise AI development. A recent Upwork Global Inc. survey shows 73% of CEOs believe their companies embrace generative AI. IBM claims that its 21,000 data, AI, and automation consultants have completed over 40,000 enterprise client engagements, and the Center of Excellence for Generative AI, which is part of its consulting division, employs over 1,000 consultants.