LG AI Research open-sources three versions of its Exaone LLM

Home > Business > Industry

LG AI Research open-sources three versions of its Exaone LLM

Published: 09 Dec. 2024, 17:41

JIN EUN-SOO
[email protected]

Audio report: written by reporters, read by AI

LG AI Research President Bae Kyung-hoon [LG CORP.]

LG AI Research has open-sourced three versions of its large language model (LLM), Exaone 3.5, signaling confidence in its technology as it moves closer to monetization.

The research arm of LG Corp. announced Monday it had made Exaone 3.5's ultralight version for on-device use, ultralight model for general use cases and high-performance model for specialized applications available for developers to use.

Only the general-purpose version of its predecessor, Exaone 3.0, which was unveiled four months ago, was open source.

The updates allow the Exaone 3.5 to process texts as long as 100 pages at a time, including both input and output, according to the company.

The Exaone 3.5 has improved “retrieval-augmented generation,” which browses the web or uploaded document to provide more accurate answers. It has been deployed with multistep reasoning tactics to break down the user's question to offer more logical answers.

LG has been focusing on slimming down the model for cost reduction and efficiency. The heaviest version of Exaone 3.5 measures 32 billion parameters compared its first 300-billion-parameter version unveiled in 2021.

The latest version of Meta's Llama AI model, which is No. 1 in the open source LLM arena, is 70 billion parameters.

With the move to open source, Exaone intends to accelerate innovation through feedback from users and expedite the establishment of its AI ecosystem, the company said.

“The recent advancement of generative AI models has accelerated, and it is important to upgrade them quickly,’ said Bae Kyung-hoon, president of LG AI Research.

“We will speed up the pace of innovation and develop them into frontier models that represent Korea with the goal of artificial superintelligence that can be applied to real-world industries.”

LG AI Research expects Exaone to generate sales next year.

BY JIN EUN-SOO [[email protected]]