SILMA Kashif v1.0: A Specialized Model for RAG Tasks
Introducing SILMA Kashif 2B Instruct v1.0, the newest addition to the SILMA Kashif Family of models, specifically designed for Retrieval-Augmented Generation tasks.
Kashif excels in answering questions based on contextual pieces in both Arabic and English, making it a versatile tool for users. Additionally, the model is capable of performing Entity Extraction tasks.
SILMA Kashif 2B v1.0 has been identified as the top-performing small/open model (within the 3-9 billion parameter range) based on a special benchmark we have created to evaluate the model: SILMA RAGQA Benchmark
The model is built on the robust foundational models of Google Gemma, has 12k context length and is free to use under the Gemma license
Try it out now: https://huggingface.co./silma-ai/SILMA-Kashif-2B-Instruct-v1.0
Model Skills and Capabilities:
- Mastery in answering questions - based on context- in Arabic and English
- Adaptability in handling short and long contexts
- Efficiency in providing concise and detailed answers
- Proficiency in tackling complex numerical questions
- Ability to answer questions based on tabular data
- Skilled in addressing multi-hop questions by utilizing information from multiple paragraphs
- Capability to reject inaccurate answers and provide a more precise response
- Versatility in answering questions across various domains such as finance, medical, legal, etc.
- Competence in understanding and addressing ambiguous contexts
- Proficiency in entity extraction from text
- Versatility in dealing with diverse and complex prompts
Note: SILMA Kashif is a specialized model that should ONLY be utilized in RAG setups.