MosaicML

MosaicML (now Databricks Mosaic Research) was an AI infrastructure company that built open-source tools and a cloud platform to make training, fine-tuning, and deploying large language models dramatically more efficient and cost-effective. Its flagship open-source projects — Composer (a PyTorch training library with algorithmic speedups), LLM Foundry (end-to-end LLM training and fine-tuning), and StreamingDataset (cloud-native data streaming for ML) — achieved up to 8x training speedups over baseline approaches. MosaicML also offered Mosaic AI Training (MCLI), a proprietary orchestration platform for large-scale GPU cluster management. Founded in 2021 and headquartered in San Francisco, the company was acquired by Databricks for $1.3 billion in July 2023 and its technology became the foundation of Databricks' generative AI portfolio.