Model Summary:
This model is a Sentence Transformer based on Omartificial-Intelligence-Space/Arabic-Triplet-Matryoshka-V2, fine-tuned for semantic textual similarity and information retrieval tasks. It maps sentences to dense vector representations for tasks like search, clustering, and text classification.
Dataset:
- The dataset used for training is derived from Egyptian law books.
- It consists of synthetic data generated using a Large Language Model (LLM).
- The dataset contains 20,252 samples, formatted as question-answer pairs.
Key Features:
- Vector Representation: 768-dimensional embeddings.
- Training Loss: MatryoshkaLoss & MultipleNegativesRankingLoss.
- Evaluation Metrics: Cosine similarity-based metrics (Accuracy, Precision, Recall, NDCG).
🏆 Leaderboard Performance
The Muffakir_Embedding model has achieved notable rankings on the Arabic RAG Leaderboard, securing:
🥇 1th place in the Islamic Dataset
These results underscore the model's effectiveness in both retrieving relevant information and accurately ranking it within Arabic Retrieval-Augmented Generation (RAG) systems.
This model is optimized for legal document retrieval and other NLP applications in Arabic.