AI is multifarious, multiform, multi-vector, multi-workflow and both multi-model & multi-modal in its form, use cases and underlying architecture. Because of this core truism – and the fact that AI is only ever as smart as the data resources and models that it has access to and is able to ingest – the best AI systems may be the ones that are the most diverse, multi-faceted and intricately engineered.
Describing itself as a ‘data and AI’ company, Databricks thinks it can fabricate a more intricate (and perhaps beautifully irregular) approach to Machine Learning (ML) and Artificial Intelligence (AI) through the combination of its own platform and what is its most recent acquisition.
Having now completed its acquisition of MosaicML, a generative AI platform known for its MPT large language models (LLMs), Databricks says that it has a shared vision of making generative AI accessible for all organisations, enabling them to build, own and secure generative AI models, including LLMs with their proprietary data.
Mosaic Pretrained Transformer
TECHNICAL NOTE: MPT denotes Mosaic Pretrained Transformer and May this year saw MosaicML release a new model series called MPT (MosaicML Pretrained Transformer) to address the limitations of the previous models and provide a commercially-usable, open-source model that matches (and – in many ways – surpasses) previous ML models (such as LLaMA-7B).
“Through this acquisition, we will be able to provide our 10,000 customers the ability to use their own data on the Databricks Lakehouse to train and deploy bespoke LLMs that offer a differentiated experience for their end users. Combining MosiacML with the Lakehouse provides a secure and best-in-class generative AI platform,” said Ali Ghodsi, co-founder and CEO of Databricks. “We are excited to welcome the MosaicML team of top AI researchers and engineers to the Databricks family and look forward to collaborating with them to build the future of data and AI.”
MosaicML’s product is said to enable organizations to efficiently build and deploy large AI models on their own data and business processes. In the coming months, the Databricks and MosaicML teams will work to make training custom models even more seamless for Databricks Lakehouse customers.
Open source innovation
“Together with Databricks, we will empower enterprise customers to more securely and cost-effectively build LLMs and generative AI with their own data,” said Naveen Rao, co-founder and CEO at MosaicML. “Our team is excited to join the Databricks family and continue innovating for customers and the open source community.”
The Databricks Lakehouse Platform, combined with MosaicML’s technology, will offer customers a generative AI training platform while allowing them to retain control, security and ownership over their valuable data.
Free image use: Wikipedia.