Artificial intelligence (AI) is now quite popular, especially in the field of text generation. This includes big language models like ChatGPT. A recent survey of about 1,000 industrial firms found that 67.2% of them consider the adoption of large language models (LLMs) to be a top priority to achieve by early 2024.
However, obstacles hinder progress. Based on the same poll, the absence of customization and flexibility, together with the incapacity to retain corporate expertise and intellectual property, have been major obstacles preventing several organizations from using LLMs in their operations.
Varun Vummadi and Esha Manideep Dinne were prompted to contemplate the nature of a potential resolution to the predicament of corporate LLM adoption. Seeking a solution, they established Giga ML, a firm that develops a platform enabling organizations to implement LLMs on-site, therefore reducing expenses and safeguarding privacy.
“Enterprises encounter significant challenges in terms of data privacy and customization when implementing LLMs to address issues,” Vummadi said in an email interview with Eltrys. “Giga ML tackles both of these challenges.”
Giga ML provides a dedicated collection of Language and Learning Models (LLMs) known as the “X1 series.” These models are specifically designed for activities such as code generation and responding to frequently asked customer inquiries, such as queries about order delivery time. The firm claims that the models, constructed on top of Meta’s Llama 2, surpass widely used LLMs on certain benchmarks, namely the MT-Bench test set for dialogs. However, it is difficult to determine the qualitative comparison of X1. This journalist attempted to use Giga ML’s online demonstration but encountered technical difficulties. (The application consistently experienced a timeout error regardless of the input prompt provided.)
While Giga ML’s models may possess some advantages, it is uncertain if they can really have a significant impact on the vast realm of open-source, offline LLMs.
During my conversation with Vummadi, I learned that Giga ML’s primary focus is not on creating the most high-performing language model models (LLMs) available but rather on developing tools that enable organizations to locally optimize LLMs without depending on external resources and platforms.
“Giga ML aims to assist enterprises in securely and effectively implementing LLMs on their own infrastructure located on-premises or in a virtual private cloud,” said Vummadi. “Giga ML streamlines the training, fine-tuning, and execution of LLMs by managing these tasks through a user-friendly API, effectively eliminating any related difficulties.”
Vummadi highlighted the privacy benefits of doing modeling offline, which are likely to be compelling for some firms.
Predibase, a low-code artificial intelligence development platform, discovered that less than 25% of businesses feel at ease using commercial language and machine learning models (LLMs) due to apprehensions around the disclosure of sensitive or private information to vendors. Approximately 77% of survey participants said that they do not use or have no intention of using commercial LLMs beyond prototypes in production. This decision is mostly influenced by concerns over privacy, cost, and the absence of customization options.
“IT managers at the executive level appreciate the value of Giga ML’s offerings due to the secure deployment of LLMs on-premise, the ability to customize models to their specific needs, and the fast inference, which guarantees data compliance and optimal efficiency,” said Vummadi.
Giga ML, a company that has secured around $3.74 million in venture capital investment from Nexus Venture Partners, Y Combinator, Liquid 2 Ventures, 8vdx, and other investors, intends to expand its staff and intensify its research and development efforts in the near future. According to Vummadi, a part of the funds is being allocated to assist Giga ML’s client base, which presently consists of undisclosed “enterprise” organizations in the financial and healthcare sectors.
