IBM announced today that it plans to host Meta’s open source large language model (LLM) Llama 2-chat on its AI and data platform, watsonx.
Watsonx, which came out in May, sought to offer enterprises the full technology stack, including a studio, data store and governance toolkit, so that they can train, tune and deploy AI models, including foundation models and machine learning capabilities, all in one place and to run across any cloud environment.
Llama 2 will be hosted in the studio, watsonx.ai, which currently offers models from IBM and the Hugging Face community. These models are pre-trained to support a range of natural language processing (NLP) tasks including question answering, content generation and summarization, text classification and extraction.
IBM says that hosting Llama 2 builds on its strategy to offer both third-party and its own AI models, while advancing its ongoing collaboration with Meta on open innovation for AI such as the PyTorch machine learning framework and the Presto query engine used in watsonx.data.
Meta touted its open approach when it released Llama 2 in July in collaboration with Microsoft. The 70 billion parameter model, free for research and commercial use, leverages publicly available instruction datasets and over one million human annotations. Meta also deemed Llama 2 to be safer, as opening access enables a generation of developers and researchers to stress test it, improve the model and fix vulnerabilities.
IBM also says it is committed to keeping trust and security principles at the forefront. For instance, when users run the Llama 2 model through the prompt lab in watsonx.ai, they can toggle on the AI guardrails function to help automatically remove harmful language from the input prompt text, as well as from the output generated by the model. Meta also provides an account of the fine-tuning methodology used in their large language models.
IBM boasts of its 21,000 data, AI and automation consultants, in addition to its Centre of Excellence for Generative AI that comprises more than 1,000 consultants with specialized generative AI expertise. These consultants, the company says, will be responsible for working with clients to tune and operationalize models for targeted use cases.
AI builders can explore watsonx.ai today via the free trial. Early access to Llama 2 is available to select clients and partners.