Mosaic ML
Rigorous science. Real impact.
IvaraX Analysis
Databricks Mosaic Research represents the AI research arm of Databricks, focusing on developing efficient, open-source large language models and deep learning tools. The team combines academic-quality research with enterprise-ready AI solutions, offering organizations the ability to train custom models on proprietary data at competitive costs.
Key Strengths
- +Strong open-source contributions with commercially usable licenses
- +Cost-efficient model training demonstrated by $20 BERT pretraining capability
- +Advanced optimization techniques including FP8 inference and sparse mixture-of-experts architecture
- +Full-stack AI capabilities from data management through model deployment
- +Established enterprise credibility through high-profile partnerships like Shutterstock
Ideal For
- →Enterprises looking to train custom LLMs on proprietary data
- →Organizations seeking open-source AI solutions with commercial licensing
- →Companies already using or considering the Databricks ecosystem
- →Data-intensive businesses requiring efficient large-scale model training
Things to Consider
- !As part of Databricks, the full solution benefits may require adoption of the broader Databricks platform
- !Enterprise-grade features and scale may exceed the needs of smaller organizations or simpler use cases
- !Organizations should evaluate whether open-source model customization aligns with their AI strategy versus using pre-built commercial APIs
About Mosaic ML
Mosaic ML, now operating as Databricks Mosaic Research, is a pioneering artificial intelligence research organization that has become an integral part of Databricks following its acquisition. The team specializes in developing and training large language models, diffusion models, and advanced deep learning tools, with a strong commitment to open-source research and commercially usable AI solutions. The organization has made significant contributions to the AI landscape, including the development of DBRX, an open-source large language model that set new benchmarks for quality at the time of its release. Their portfolio also includes the MPT family of models, Mosaic BERT, Mosaic Diffusion, and various efficiency-focused tools like Composer and LLM Foundry. Through their integration with Databricks' Data Intelligence Platform, they offer enterprise-grade solutions for training, fine-tuning, and deploying AI models at scale, combining cutting-edge research with practical business applications.
Why Choose Mosaic ML
- Proven track record of developing high-quality open-source LLMs including DBRX and the MPT model family
- Highly efficient deep learning stack optimized for training, fine-tuning, and deploying large models at scale
- Integration with Databricks platform provides unified data management, governance, and AI capabilities
- Commitment to open-source tools including Composer, LLM Foundry, and StreamingDataset
- Demonstrated enterprise partnerships, including co-developing Shutterstock ImageAI for commercial applications
Services
AI/ML Model TrainingLarge Language Model DevelopmentDeep LearningModel Fine-tuningGenerative AIText-to-Image Generation
Technologies
DBRXMPTBERTMosaic DiffusionComposerLLM FoundryStreamingDatasetPyTorchCUDAFP8H100
Tech Stack(detected from website)
PyTorchCUDAFP8 (H100 GPU optimization)Hugging Face integrationGitHubDatabricks PlatformMulti-Cloud (AWS, Azure, GCP)
Industries Served
Notable Clients
Shutterstock
Categories
Company Info
- Headquarters
- San Francisco, CA, US
