Launch of an enhanced Arabic AI model tailored for businesses in the Middle East and North Africa by Cohere
Cohere, a cloud-agnostic AI platform for enterprises, has announced the release of its latest AI model, Command R7B Arabic. The company, founded by Google Brain alumni and a co-author of the Transformer research paper, is backed by leading investors including DTCP, Index Ventures, NVIDIA, Oracle, Salesforce Ventures, and Tiger Global, with a valuation of $5.5 billion in its 2024 funding round.
The new model is designed specifically for Arabic and English, with advanced language capabilities tailored for Middle Eastern enterprises. Command R7B Arabic boasts a lightweight, state-of-the-art design, making it suitable for deployment efficiency and scalable Arabic NLP tasks.
Key Features of Command R7B Arabic
Command R7B Arabic employs an optimized transformer with three layers of sliding window attention and ROPE (Rotary Positional Embeddings) for efficient local context modeling and relative positional encoding. A fourth layer uses global attention without positional embeddings, enabling unrestricted token interaction across the entire 128,000-token context window.
The model is primarily trained for Arabic and English but includes samples from other languages, supporting multilingual tasks. It is specially fine-tuned for Retrieval Augmented Generation (RAG) tasks in Arabic and English, improving retrieval-based text generation.
Command R7B Arabic is versatile, offering both conversational mode (for interactive dialogues with Markdown and LaTeX support, follow-up questions, and introductory statements) and instruct mode (for concise, task-focused outputs such as summarization, translation, categorization).
Post-pretraining supervised fine-tuning (SFT) and preference training ensure the model's outputs are helpful and safe.
Applications and Availability
Command R7B Arabic can perform document summarization, question answering on company materials, and automate repetitive work. It is compact in size, enabling businesses to easily scale Arabic language AI applications to production environments.
The model is available on the Cohere platform, HuggingFace, and Ollama, and can be served on low-end GPUs, MacBooks, or CPUs with a 128k context length. On-premises deployment options are offered for enterprise customers.
Making a Difference
Cohere is committed to ensuring its AI technology serves as many people, organizations, and markets as possible. Command R7B Arabic is intended for use in the Middle East and North Africa, and Dubai Police has already announced plans to use artificial intelligence and data analytics to prevent crimes.
By providing the model weights, Cohere also gives access to the research community, fostering collaboration and innovation in AI research.
Command R7B Arabic outperforms other leading models in its class across key enterprise tasks, making it a strong choice for advanced and scalable Arabic NLP tasks.
- Advancements in technology, such as Command R7B Arabic, are designed for specific regions and languages, catering to the unique needs of Middle Eastern enterprises.
- Command R7B Arabic, a cutting-edge AI model, is integrated with artificial intelligence and data analytics systems, with Dubai Police already planning to use it for crime prevention in the Middle East and North Africa.
