SLIMA Kashif is a new open-source AI model designed specifically for Arabic

SILMA Kashif 2B Instruct v1.0 is a new bilingual AI model specifically designed for Arabic and English retrieval-augmented generation (RAG) tasks, with a primary focus on question answering and secondary capabilities in entity extraction.

Core capabilities and architecture: The model is built on Google Gemma’s foundation and operates within the 3-9 billion parameter range, featuring a 12,000-token context window for processing large amounts of text.

• The model excels at answering questions in both Arabic and English languages
• It processes both short snippets and lengthy passages effectively
• The system can provide both concise and detailed responses based on context
• Entity extraction capabilities allow it to identify and pull key information from text

Technical performance and benchmarks: SILMA Kashif demonstrates strong performance across multiple evaluation metrics and datasets.

• The model achieved an overall benchmark score of 0.3478 in comprehensive testing
• Evaluation included diverse datasets like FinQA, TatQA, MS MARCO, and others
• Testing covered both Arabic and English language capabilities
• Performance metrics included exact match, ROUGE1, BLEU, and BERTScore

Implementation requirements: The model offers flexibility in deployment while maintaining specific hardware recommendations for optimal performance.

• Recommended hardware includes GPUs with 24GB memory (like NVIDIA RTX 4090)
• Can operate on GPUs with 8GB memory with some performance impact
• 4-bit quantization option available with minimal performance loss (2.6% drop)
• Implementation requires simple setup through the Transformers library

Key limitations and constraints: Despite its strong capabilities, the model has several notable limitations.

• Complex numerical and financial reasoning tasks present challenges
• Performance is limited to text-based question answering
• The model may struggle with tasks outside its specialized focus
• Parameter size constrains certain advanced reasoning capabilities

Looking ahead: Arabic NLP innovation: SILMA Kashif represents an important step forward for Arabic natural language processing, offering specialized capabilities while acknowledging current technological constraints. Its open-source nature and strong performance in targeted applications suggest it could serve as a foundation for future developments in multilingual AI systems, particularly in the Middle East region.