How Databricks’ Mosaic AI Produced One Of The Fastest Open-Source LLMs Available

Last year, Databricks acquired MosaicML, renamed Mosaic AI, a Black co-founded generative AI (genAI) startup know for its cutting-edge large language (LLMs) models.

Building on both companies vision of making genAI accessible for all organizations, Databricks’ Mosaic team has since developed DBRX, one of the fastest and highest-performing LLMs available.

Democratizing deep learning

Co-founded by Jonathan Frankle PhD ’23 and MIT Associate Professor Michael Carbin, MosaicML started with the mission to make AI models more accessible. 

The startup developed a platform where users can train, improve, and monitor open-source models with their own data. They also created their own open-source models using Nvidia GPUs, broadening access to deep learning for more organizations.

Since the release of OpenAI’s ChatGPT-3.5, interest in deep learning has surged. MosaicML capitalized on this by providing tools that simplify the use of deep learning for organizations.

Their techniques increased image classification speed fourfold and led to the release of an open-source LLM and a library of training methods in 2023.

Databricks’ Mosaic Releases DBRX

In July 2023, Databricks acquired MosaicML for $1.3 billion, and rebranded it Mosaic AI.

Since the acquisition, the combined strengths of Databricks and MosaicML have resulted in DBRX. This model has gained a reputation for being one of the fastest open-source LLMs available. 

Built using Databricks’ tools, DBRX allows enterprises to achieve similar performance with their own models, accelerating the impact of generative AI.

DBRX has set new benchmarks in tasks like reading comprehension, general knowledge questions, and logic puzzles, and is particularly beneficial for large enterprises.

Databricks is launching five new Mosaic AI tools at its conference: Mosaic AI Agent Framework, Mosaic AI Agent Evaluation, Mosaic AI Tools Catalog, Mosaic AI Model Training and Mosaic AI Gateway, according to TechCrunch.

Feature Image Credit: MIT News