1 Dataset Exploration | Build an LLM from Scratch with Pytorch | Ep.1

Meet the data behind Jabarti. We explore the jabarti-llm-dataset on Hugging Face. The pretrain subset has about  bilingual Wikipedia records, mixing general Wikipedia with a curated Egyptian-history collection. The finetune subset has Arabic and English question–answer pairs. We look at the columns, the Arabic/English mix, how the eval split holds out whole articles to avoid leakage, and the data-quality issues that shape the cleaning and tokenizer choices in the next section. dataset : https://huggingface.co/datasets/bakri...