“Artificial Intelligence Research Center”, LLM has released eight billion parameters from the group of language models “Dorna” as open source.
According to Iran digital economy annotation, it was March 2023 when this center unveiled the large language model group (LLM) of Dorna, which was developed in three capacities.
Now Part has announced that it has upgraded one of the 8 billion parameter models belonging to Dorna language models group and made it available to the country’s artificial intelligence ecosystem in an open source form.
Part has announced that it intends to compensate for the “weak performance of artificial intelligence services in Farsi” and the large language model group Dorna has also been promoted to emphasize the Persian language in the world of artificial intelligence, to provide the opportunity to create new artificial intelligence tools for businesses, personal developers, Specialists and researchers should be provided according to their needs and goals.
Part has claimed that this LLM is among the most prominent language models in the world and has no competitors in the indicators of the Persian language. Part also said that the overall score of “Dorna” is better compared to models like Meta Llama 3, especially in native language interactions and Iranian and Farsi cultural backgrounds. According to Part, more details about the comparison of these two language models can be seen on the Hugging Face website.
Part has also announced that for the first time in Iran, it will host this large language model on Hugging Face servers so that users can test this LLM on this site for free without the need for infrastructure or hardware.
Part has announced that this LLM has the capability of summarizing texts, coding, answering long questions, recognizing emotions and creating new content based on entered commands, and the ability to answer general, historical, mathematical and has etc.
Part has said that he plans to use other larger Dorna models in his commercial products in the coming months.
No Comment! Be the first one.