2024-03-01 15:23:52
- IBM offers an optimized version of Mixtral-8x7B that has shown potential to cut latency by up to 75%;
- Adds to IBM’s growing catalog of third-party and open source models to give customers choice and flexibility;
- The latest open source model available on the Watsonx AI and data platform with enterprise-ready AI studio, data storage, and governance capabilities.
São Paulo, February 29, 2024 – IBM (NYSE: IBM) today announces the availability of the open source LLM Mixtral-8x7B, powered by Mistral AI, on its Watsonx data and AI platform as it continues to expand capabilities to help customers innovate with IBM’s own base models and those from a variety of open source providers.
IBM offers an optimized version of Mixtral-8x7b which, in internal tests, was able to increase transfer rate by 50% when compared to the normal model, that is, it increased the amount of data that can be processed in a given period. This might potentially reduce latency by 35-75% depending on batch size – speeding up time to insights. This is possible through a process called quantization, which reduces model size and memory requirements for LLMs and, in turn, can speed up processing to help reduce costs and power consumption.
The addition of Mixtral-8x7B expands IBM’s open, multi-model strategy to meet customers where they are and give them choice and flexibility to scale enterprise AI solutions across their businesses. Through decades of AI research and development, open collaboration with Meta and Hugging Face, and partnerships with model leaders, IBM is expanding its catalog of watsonxfazerai models and bringing new features, languages, and modalities.
IBM’s enterprise-ready foundational model choices and its Watsonx AI and data platform can empower clients to use generative AI to gain new insights and efficiencies, and create new business models based on trust principles. IBM enables customers to select the right model for the right use cases and price and performance goals for targeted business domains such as finance.
Mixtral-8x7B was built using a combination of Sparse modeling, an innovative technique that finds and uses only the most essential pieces of data to create the most efficient models, and the Mixture-of-Experts technique, which combines different models that specialize in solve different parts of a problem. The Mixtral-8x7B model is widely known for its ability to quickly process and analyze large amounts of data to provide context-relevant insights.
“Customers are asking for choice and flexibility to implement models that best fit their unique use cases and business requirements,” said Kareem Yusuf, Ph.D, Senior Vice President of Product and Growth Management at IBM Software. “By offering the Mixtral-8x7B and other models on Watsonx, we’re not just giving them choices in how they deploy AI – we’re empowering a robust ecosystem of AI builders and business leaders with tools and technologies to drive innovation across diverse industries. and domains.”
This week, IBM also announced the availability of ELYZA-japanese-Llama-2-7b, an open source Japanese LLM template from ELYZA Corporation, on watsonx. IBM also offers open source Meta templates Llama-2-13B-chat and Llama-2-70B-chat and other third-party templates in watsonx, with more to come in the coming months.
1709332402
#IBM #announces #availability #open #source #Mistral #model #Watsonx
