The Department of Intellectual and Cultural Affairs at the Al-Abbas's (p) holy shrine has launched the artificial intelligence model "Mizan" to enhance research technologies in Arabic.
Mr. Ali al-Jayashi, director of the Haqeebat al-Mo'min Application of the Department, said:"this model is the first open source Arabic model specialized in rearranging search results for long texts in Arabic, and constitutes a qualitative leap in digital search technologies."
He added that "the model was developed by the Haqeebat al-Mo'min Application team, based on the transformer architecture that allows it to process long texts with high efficiency up to 8192 characters at a time, and the model, despite its light size of only 149 million parameters, proved its superiority over many open source models with larger sizes".
"Performance tests have shown that the Mizan model has surpassed many global models, surpassing the (bg-rerank-v2-m3) model with a volume of 568 million parameters, and the (jina-reranker-v2-base-multilingual) model with a volume of 278 million parameters.
"The model has proven its ability to provide accurate results with a response speed of only 0.1 seconds, with a very low memory consumption of no more than 1 GB, which makes it three times more efficient compared to competing models," Al-Jayashi said, pointing out that "it is characterized by the ability to analyze and arrange texts with high accuracy, which makes it ideal for use in search engines, digital archiving systems, artificial intelligence applications for conversation, electronic educational platforms, and specialized information retrieval systems."
The director of the Center stressed that "the launch of the (Mizan) model comes within the framework of the ongoing efforts of the Al-Abbas's (p) holy shrine to enhance the presence of the Arabic language in the field of artificial intelligence, and the development of search technologies to provide more accurate and efficient Arabic content".
Those interested can try the model or benefit from it in their projects by visiting the following links:
- Link to the form on the Huggingface (https://huggingface.co/ALJIACHI/Mizan-Rerank-v1)
- Model experience link
(https://huggingface.co/spaces/ALJIACHI/Mizan-Rerank-v1-Demo)