MosaicML

MosaicML is a company offering a generative AI platform for enterprises to build their own AI systems. MosaicML's technology is designed to make it easier to train and deploy machine learning models using the user's own data. This includes each layer of the model development process, including algorithms, systems, and hardware, to reduce the time and cost of training AI models.

MosaicML has released a series of open-source large language models (LLMs) called MPT (Mosaic pre-trained transformer). The MPT-7B series of models released in May 2023 was collectively downloaded more than 3M times in its first seven and a half weeks. In June 2023, MosaiscML released MPT-30B, a significantly more powerful series of models that includes fine-tuned variants MPT-30B-Instruct and MPT-30B-Chat for single-turn instructions and multi-turn conversations, respectively. MPT-30B models come with an 8k token context window at training time, support for even longer contexts via ALiBi, and efficient inference + training performance via FlashAttention. The MPT-30B family also has strong coding abilities due to its pretraining data mixture.

MosaicML was founded in 2021 by Naveen Rao (CEO), Hanlin Tang (CTO), Michael Carbin (founding advisor). Rao and Tang previously sold a start-up to Intel in 2016 and joined the company. After leaving Intel, Rao and Tang began working on MosaicML, a new start-up company aiming to make AI work more efficiently, using less compute power. On October 13, 2021, they came out of stealth mode, announcing $37 million in funding from DCVC, Lux Capital, Future Ventures, Playground Global, and other venture investors. Rao and Tang worked with Professor Micahel Carbin, a professor at MIT whose work includes making machine learning more efficient, and his doctoral candidate, Jonathan Frankle, who became the company's chief scientist. The group felt there was no single technology for reducing AI computing costs. Instead, a mosaic of different approaches was emerging. This inspired the name MosaicML. The company is headquartered in San Francisco, with additional offices in New York, Palo Alto, and San Diego.

On June 26, 2023, Databricks announced it had entered a definitive agreement to acquire MosaicML in a transaction valued at around $1.3 billion, inclusive of retention packages. The company plans to combine MosaicML's technology with its Databricks Lakehouse platform to offer customers a simple way to develop AI models while retaining control and ownership over their data. Databricks CEO and cofounder Ali Ghodsi said of the acquisition:

Every organization should be able to benefit from the AI revolution with more control over how their data is used. Databricks and MosaicML have an incredible opportunity to democratize AI and make the Lakehouse the best place to build generative AI and LLMs... Databricks and MosaicML’s shared vision, rooted in transparency and a history of open source contributions, will deliver value to our customers as they navigate the biggest computing revolution of our time.”

The MosaicML team is expected to join Databricks after the transaction closes, with the MosaicML platform becoming supported, scaled, and integrated over time to offer customers a unified platform for building, owning, and securing their generative AI models.

About the deal, MosaicML CEO and cofounder Naveen Rao stated:

At MosaicML, we believe in a world where everyone is empowered to build and train their own models, imbued with their own opinions and viewpoints — and joining forces with Databricks will help us make that belief a reality...We started MosaicML to solve the hard engineering and research problems necessary to make large scale training more accessible to everyone. With the recent generative AI wave, this mission has taken center stage. Together with Databricks, we will tip the scales in the favor of many — and we’ll do it as kindred spirits: researchers turned entrepreneurs sharing a similar mission. We look forward to continuing this journey together with the AI community.

The transaction is subject to customary closing conditions and regulatory clearances, and the companies will remain independent until they are complete. The deal is expected to close during Databricks’s second quarter, which ends on July 31, 2023.