Leadmatic introduces tinyChat, an open-source language model with a compact size, designed to complement larger models like ChatGPT. With a focus on increasing accessibility and cost-efficiency, tinyChat aims to foster innovation and expand the reach of generative AI across industries.
Leadmatic, an Indianapolis-based AI startup, has announced the launch of tinyChat, a large language model (LLM) that is less than 1% the size of ChatGPT (GPT 3.5). tinyChat is capable of performing a variety of natural language processing tasks, including summarization, text generation, question and answering, and text classification.
tinyChat is an open-source project, released under the Apache 2.0 license. This new model aims to complement larger LLMs like ChatGPT by serving as a co-agent, handling smaller, repetitive tasks which could result in efficiencies and cost savings.
Read More: HCLTech Delivers Another Year Of Stellar Growth With Robust Deal Pipeline
The development of tinyChat was motivated by the belief that smaller language models can play a significant role in generative AI. Smaller models like tinyChat can provide compute efficiencies and cost savings by offloading repetitive tasks from larger models. Additionally, researching smaller architectures and novel datasets could lead to more efficient artificial intelligence systems in comparison to the large language models of today.
tinyChat was developed by fine-tuning Google’s Flan-T5-large model using the open-source databricks-dolly-15k dataset from Databricks and Microsoft’s Low-Rank Adaptation of Large Language Models (LoRA) training method. In terms of performance, tinyChat either outperforms or is comparable to other open-source models depending on the task.
Read More: SalesTechStar Interview with Joe Massa, Chief Revenue Officer at Rootstock Software
Potential applications for tinyChat include mobile and embedded devices where compute resources are limited, and enabling developers to use tinyChat for smaller tasks while reserving larger LLMs for more complex tasks.
Kishan Bhoopalam, Founder and CEO of Leadmatic, said, “This finding is significant because what it shows is that a model that is 0.1% the size of chatGPT can exhibit similar qualities to chatGPT which could mean in the near future many of the use cases of chatGPT could be run on local devices resulting in decentralization of the technology.” Bhoopalam added, “While chatGPT is great, at scale it can be prohibitively expensive for companies to use. We think a small open-source model will enable all kinds of new possibilities. It’s still early days and tinyChat is still in its research phase, but we hope other companies will benefit from it as well.”