Mollel/hazina-fiscal-finetune

Hazina Fiscal Embedding Fine-tune (Swahili–English) A self-supervised contrastive training set for fine-tuning sentence-embedding / retriever models on Tanzanian fiscal & public-finance text. Built from the same 10 Government of Tanzania documents as the hazina-fiscal-retrieval evaluation benchmark, but in (anchor, positive, hard-negatives) form ready for MultipleNegativesRankingLoss. Target base model: intfloat/multilingual-e5-base (works with any multilingual encoder; E5… See the full description on the dataset page: https://huggingface.co/datasets/Mollel/hazina-fiscal-finetune.