MosaicML, a startup founded by former Intel executive Naveen Rao, is offering a cost-effective solution for AI training. The company uses software advancements and open-source models to make AI development more accessible to businesses of all sizes.
Instead of using APIs from machine learning startups or buying ready-made systems from cloud providers, MosaicML allows companies to partner with vendors who specialize in customizing private or open-source models. This helps reduce costs and protects valuable data.
MosaicML offers open-source large language models (LLMs) called MPT-7B, which have seven billion parameters. Unlike other LLMs used mainly for research, MPT-7B is suitable for commercial applications. It can quickly analyze large amounts of text, making it valuable for businesses.
Rao emphasizes the importance of making commercial use possible for the model and demonstrates its affordability. MosaicML offers to train a model for a mere $200,000, showcasing that quality AI development is no longer limited to exorbitant budgets.
Furthermore, MosaicML provides more powerful models than MPT-7B in-house, empowering businesses to create their own private models hosted on various cloud platforms or fine-tune open-source alternatives. With this approach, companies can safeguard their data, retain ownership of the model’s weights and intellectual property, and harness the rising value of their proprietary data.
The sources for this piece include an article in TheRegister.
