IBM Embeds Meta’s Llama 2 Large Language Model in New Watsonx Generative AI Platform
IBM will augment its new watsonx generative AI studio with Meta’s open-source Llama 2 large language model. The 70-billion parameter LLM will be available in the watsonx.ai studio, with early access for a limited number of clients. Llama 2 is joining a stable of LLMs supported on watsonx, including ones built in-house as well as pre-trained models from Hugging Face designed for Natural Language Processing and conversation.
Llama 2 is part of IBM’s bigger plans for incorporating models, tuning tools, and governance features into the watsonx platform. Those deploying Meta’s LLM can choose to enable guardrails that will automatically filter harmful language from prompts and outputs. Meta has also shared details on its rigorous training methodology. Adding Llama 2 brings extra juice to watsonx, which arrived earlier this year and is already a Wimbledon tennis championship commentator.
Llama 2 upgrades the power (and capitalization choices) of the version released in February. It is trained on a database that Meta claims is 40% larger as well, and the context window, how much the AI remembers within a single conversation, has doubled to two trillion tokens. The model is available with no license fee to businesses through a deal with “preferred partner” Microsoft, but Meta has been quickly augmenting its distributor list through deals with Qualcomm, Alibaba, and now IBM. The open-source design is part of Meta’s stated vision for the technology. Should there be any stumbling blocks, IBM’s Center of Excellence for Generative AI holds more than 1,000 experts on specialized forms of generative AI who are available to help.
“These experts can work with clients to help tune and operationalize models for targeted use cases aligned to their specific business requirements,” IBM explained in a blog post. “IBM, like Meta, is a supporter of open innovation. There is value in engaging a robust and diverse community of AI builders and researchers to test, share feedback and collaborate on these technologies to drive further innovation. We are excited to see what these innovators will build with Llama 2 and other models on the watsonx platform.”