LG launches Korea's first open-source AI model, EXAONE 3.0

LG AI Research has officially unveiled South Korea's first open-source artificial intelligence model, EXAONE 3.0. This event was a significant step for South Korea in the global AI industry dominated by tech giants from the US, China, and the Middle East.

Here's What We Know

EXAONE 3.0 supports only two languages - English and Korean - and is based on the Transformer architecture with parameters of 7.8 billion and 8 trillion training tokens.

LG AI Research stated that this model was created for research purposes and is already available for widespread use.

It demonstrates a world-class level of English language proficiency, outperforming other models such as Llama 3.0 in real-world use cases. EXAONE 3.0 also showed strong results in mathematical computation and coding, reducing inference time by 56% and memory usage by 35%.

The model was trained on 60 million professional data cases related to patents, codes, mathematics, and chemistry.

LG AI Research plans to increase the amount of training data to 100 million by the end of the year. They are also working on making the model use less power and take up less space. Thanks to this, they managed to reduce the size of the model by 97% compared to the previous version of EXAONE 1.0, while even improving its performance.

Source: Gizmochina