GECKO: Generative Language Model for English, Code and Korean
This work provides a research baseline and practical insights for Korean LLM research, addressing the need for bilingual models in Korean and English, though it is incremental as it builds on existing architectures and focuses on domain-specific improvements.
The authors tackled the problem of creating a bilingual large language model optimized for Korean and English, along with programming languages, by pretraining GECKO on a balanced, high-quality corpus using the LLaMA architecture, resulting in great efficiency in token generation for both languages and strong performance on Korean benchmarks like KMMLU, with modest results in English and code despite a smaller vocabulary and fewer trained tokens.
We introduce GECKO, a bilingual large language model (LLM) optimized for Korean and English, along with programming languages. GECKO is pretrained on the balanced, high-quality corpus of Korean and English employing LLaMA architecture. In this report, we share the experiences of several efforts to build a better data pipeline for the corpus and to train our model. GECKO shows great efficiency in token generations for both Korean and English, despite its small size of vocabulary. We measure the performance on the representative benchmarks in terms of Korean, English and Code, and it exhibits great performance on KMMLU (Korean MMLU) and modest performance in English and Code, even with its smaller number of trained tokens compared to English-focused LLMs. GECKO is available to the open-source community under a permissive license. We hope our work offers a research baseline and practical insights for Korean LLM research. The model can be found at: https://huggingface.co/kifai/GECKO-7B