EXAONE 3.0 - An open-source AI model launched by LG, designed specifically for English and Korean.

What is EXAONE 3.0?

EXAONE 3.0 is an open-source AI model from LG AI Research, featuring 780 million parameters and designed specifically for English and Korean. EXAONE 3.0 performs exceptionally well in both English and Korean language tests, ranking first in both real-world use cases and mathematical coding. Compared to its predecessor, EXAONE 3.0 offers significant optimizations in inference speed, memory usage, and operational costs. It has currently been trained on 60 million professional datasets and is planned to expand to 100 million datasets by the end of the year. The model is available on the Hugging Face platform.

EXAONE 3.0 Features

• Bilingual supportDesigned specifically for English and Korean, it can handle natural language processing tasks in both languages.
• high performanceIt excels in multiple tests in English and Korean, including real-world use cases and mathematical coding skills.
• open sourceThe model code and training data are made public, making them easy for researchers and developers to use and conduct further research.
• Optimize efficiencyCompared to its predecessor, the inference time is reduced by 56%, memory usage is reduced by 35%, and operating costs are reduced by 72%.
• Professional trainingIt has been trained on 60 million data cases in professional fields such as patents, code, mathematics and chemistry.

The technical principles of EXAONE 3.0

• Decoder-only Transformer architectureThis architecture is a variant of the Transformer model that does not include an encoder part, only a decoder. The model is more direct and faster in generating text because the decoder can directly generate the output sequence.
• Large-scale parametersEXAONE 3.0 has 780 million parameters, and the model is able to capture more complex language patterns and relationships, thereby improving its ability to understand and generate text.
• Bilingual modelIt is trained for English and Korean and can handle natural language understanding (NLU) and natural language generation (NLG) tasks in these two languages.
• Large amount of training dataThe model used 8TB of training data, which included a wide range of language materials, helping to improve the model's generalization ability and accuracy.

EXAONE 3.0 project address

Application scenarios of EXAONE 3.0

• Language translationIt supports translation tasks between English and Korean, facilitating cross-language communication and information transfer.
• Text generationIt can be used to generate creative writing, news articles, social media content, etc.
• Question and Answer SystemBuild an intelligent question-and-answer system to provide users with accurate and rapid information feedback.
• Text SummaryAutomatically generates summaries of documents or web pages to help users quickly grasp the main content.