Highlights:

  • One of the scenarios the company focuses on with its platform involves hosting a combination of expert applications known as MoE applications.
  • Gradient provides integration with LangChain, an open-source tool that simplifies incorporating large language models into corporate applications.

Startup Gradient has raised USD 10 million in initial funding to increase its cloud platform’s uptake, allowing developers to host personalized variations of open-source language models.

Mango Capital, The New Normal Fund, Tokyo Black, Secure Octane, and Global Founders Capital also invested in the round, which Wing VC led.

Numerous open-source, pretrained neural networks are available in the market, so software development teams don’t have to code everything from scratch when developing artificial intelligence applications. However, building and maintaining the infrastructure that will support their open-source AI models is still the responsibility of the developers. Additionally, many businesses modify their neural networks for particular tasks, which adds to the workload.

Gradient is working to make the procedure simpler. The business, incorporated as Preemo Inc., offers a cloud platform created especially to house open-source large language models. It promises to automate a large portion of the manual work required for managing AI infrastructure.

More than a half-dozen open-source language models are pre-installed on Gradient’s platform. They range in complexity from a few hundred million to tens of billions of parameters. While smaller models execute more quickly, models with more parameters can perform complex computing tasks with high accuracy.

Some neural networks are tailored for particular use cases in Gradient’s cloud. The business offers a prepackaged version of Code Llama, a language model created by Meta Platforms Inc. for use in code generation tasks. Additionally, Gradient provides models that are tailored for use in the legal, financial, and healthcare sectors.

Development teams can further alter the models by training the company’s models on their own datasets. To manage the process, Gradient offers a command-line interface tool. The business claims it only takes a few lines of code to train one of its large language models on a unique dataset.

Additionally, Gradient promises to simplify some additional steps in the process of developing AI.

The business offers an API for encoding data into embeddings, the mathematical structures AI models use to store their knowledge. Language models can use a piece of data that has been transformed into an embedding to produce answers. Additionally, Gradient integrates LangChain, an open-source tool that simplifies incorporating large language models into business applications.

The company’s platform is designed to cater to various use cases, including hosting a blend of expert applications known as ‘MoE’ applications. These AI applications don’t rely on a single language model but leverage multiple specialized neural networks. An MoE service assigns each task it receives to the neural network best suited for its execution.

Twenty businesses are said to use Gradient’s platform. The business will expand its cloud platform with the help of its recently closed seed round to support the growth of its customer base. By the end of the year, Gradient also intends to add eight new employees.