Datasets
Training Data: The model was trained using FineWeb-Edu for English and FineWeb2 for Korean.
Validation Data: wikitext (English) and wikipedia (Korean) were used for evaluation and validation purposes.
Tokenizer
- The tokenizer is based on the GPT2 tokenizer architecture and has been further trained on the aforementioned English and Korean datasets to enhance its vocabulary and performance for bilingual tasks.
- Downloads last month
- 3