Startup Gradient has raised $10 million in seed funding to grow the adoption of its cloud platform, which developers can use to host customized versions of open-source language models.

TechCrunch reported the investment today in the company. Wing VC led the round while Mango Capital, Tokyo Black, The New Normal Fund, Secure Octane and Global Founders Capital participated as well. 

Software teams building artificial intelligence applications don’t have to code everything from scratch because there are many open-source, pretrained neural networks on the market. However, developers still have to set up and maintain the infrastructure on which their open-source AI models will run. Furthermore, many companies customize the neural networks they adopt for specific tasks, which involves additional work.

Gradient is working to simplify the process. The company, which is incorporated as Preemo Inc., provides a cloud platform specifically designed to host open-source large language models. It promises to automate much of the manual work involved in AI infrastructure management.

Gradient’s platform includes more than a half-dozen open-source language models out of the box. They range in complexity from a few hundred million parameters to tens of billions. Models with high parameter counts can perform complicated computing tasks with a high degree of accuracy, while smaller models run faster.

Some of the neural networks in Gradient’s cloud are optimized for specific use cases. The company provides a prepackaged version of Code Llama, a language model from Meta Platforms Inc. that is built for code generation tasks. Gradient also offers models optimized for use cases specific to the healthcare, financial and legal sectors.

Development teams can further customize the company’s models by training them on their own datasets. Gradient provides a command line interface tool for managing the process. According to the company, training one of its large language models on a custom dataset requires as little as a few lines of code.

Gradient also promises to ease certain other aspects of the AI development workflow. 

The company provides an application programming interface for encoding data into embeddings, the mathematical structures in which AI models keep their knowledge. After a piece of data is turned into an embedding, it can be used by language models to generate answers. Gradient also offers an integration with LangChain, an open-source tool that eases the task of integrating large language models into enterprise applications.

One of the use cases the company targets with its platform is hosting mixture of experts, or MoE, applications. Those are AI applications that are powered not by a single language model but rather several specialized neural networks. A MoE service relegates each task that it receives to the neural network best equipped to carry it out. 

Gradient’s platform is reportedly used by about 20 enterprises. To support the continued growth of its customer base, the company will use its newly closed seed round to scale up its cloud platform. Gradient also plans to hire eight new employees by year’s end. 

Image: Unsplash