Stablecoin giant Tether's new solution allows training LLMs on your smartphone
As part of its business diversification strategy, Tether, the issuer of the most popular namesake stablecoin, has launched a new framework that is said to allow training large language models on everyday devices.
Image by Diamond Visuals | Shutterstock
As part of its business diversification strategy, Tether, the issuer of the most popular namesake stablecoin, has launched a new framework that is said to allow training large language models on everyday devices.
Launched via Tether's arm, Tether Data, QVAC Fabric LLM is a large language model (LLM) inference runtime and fine-tuning framework. According to the company, it allows users to train and personalize large language models locally on devices they already own, including consumer GPUs, laptops, and smartphones, without the need for high-end cloud servers or specialized NVIDIA systems.
"It opens the door to personalized AI that can learn directly from users on their devices, preserving privacy and functioning even without an internet connection, and powering a new generation of highly resilient, anti-fragile, on-device AI applications," the company said, adding that this approach helps diversify the hardware available for AI development.
According to the company, the growing capabilities of LLMs have intensified the need for efficient fine-tuning methods to adapt these general-purpose models to specific tasks and domains. However, the prohibitive computational and memory costs of full fine-tuning make it impractical for most users.
"AI should not be something controlled only by large cloud platforms," Paolo Ardoino, CEO of Tether, said.
In the technical overview of the solution, the company said that it has validated its approach with two real-world applications: email style transfer and biomedical question answering.
QVAC Fabric LLM was released as open-source software under the Apache 2.0 license. To help developers, Tether Data has also publicly released multi-platform binaries, fine-tuned model adapters, and source code for new LLM extensions.
In any case, this is not the first attempt to democratize the AI industry. For example, as reported by Cybernews just a few days ago, Telegram launched its "decentralized confidential compute network" Cocoon, which aims to compete with Amazon and Microsoft in the AI compute market.
Unlock more exclusive Cybernews content on YouTube.