LG’s new open source artificial intelligence model: Exaone 3.0

Focusing on artificial intelligence research LG AI Researchopen source artificial intelligence model Exaone 3.0He announced. 7.8 billion parameters It is stated that the model is successful in both Korean and English language tasks.

It is seen that Llama, Gemma, QWEN, Phi and Mistral perform better compared to close-parameter models, especially in real-world usage scenarios, mathematics and coding categories in the English language. In Korean, only real-world usage scenarios and superior performance are mentioned in the general category.

According to the information shared by LG; Exaone 3.0 improves in inference time compared to the previous model 56 percentin memory usage 35 percent and operational costs 72 percent provided a decrease. Model, patents, codes, mathematics and chemistry 60 million professional data was trained on. However, it is planned to expand to 100 million cases in various areas by the end of the year. On the other hand, it is worth noting that the model cannot currently outperform other open source models in various criteria.

The research team plans to accelerate artificial intelligence research with this model. It is also intended to contribute to the creation of a robust artificial intelligence ecosystem in Korea. When we look at the field of artificial intelligence, we see that open source models produced by US-based companies are generally common. However, this situation has started to change recently. Alibaba’s Qwen and United Arab Emirates’ Falcon are the best examples in this regard. For example, Qwen, which received a major update in June, reaches more than 90 thousand corporate customers. Moreover, in terms of performance, the model surpasses Meta’s Llama 3.1 and Microsoft’s Phi-3 on platforms such as Hugging Face.

This move by LG is an indication that the company intends to establish a presence beyond consumer electronics. By making Exaone 3.0 open source, LG is both demonstrating its technological prowess and focusing on a new revenue stream in cloud computing and artificial intelligence services.