Leadmatic, a startup based in Indianapolis working on AI technology, has released an exciting product called tinyChat. This open-source language model is much more compact than larger models like ChatGPT, at 0.1% their size. The goal of tinyChat is to provide cost and compute efficiency, handling smaller tasks that can be offloaded from larger models like ChatGPT.
Kishan Bhoopalam, the company’s founder and CEO, had this to say on the release of tinyChat: “What our findings show is that a model that is 0.1% the size of ChatGPT can exhibit similar qualities. This means that in the near future many of the use cases of ChatGPT can be run on local devices, resulting in decentralization of the technology. While ChatGPT is great, it can be prohibitively expensive to use. We think that a small open-source model can open up new possibilities.”
In terms of performance, tinyChat has been shown to either outperform or maintain a similar performance to other open-source models depending on the task. It has been developed by fine-tuning Google’s Flan-T5-large model, using the open-source dataset from Databricks and Microsoft’s LoRA training method. Applications of the model include potential use in mobile and embedded devices and enabling developers to use it for smaller tasks while reserving larger models for bigger challenges.
About Leadmatic
Leadmatic is an AI-based company helping B2B companies get more meetings with their prospective customers. Through the use of machine learning and natural language processing, Leadmatic’s platform can help to identify key decision-makers within an organisation and mine the web for intent signals. From the gathered data the platform’s team of copywriters can then craft personalized messages for each contact, resulting in more meetings and sales.
