JAKARTA - Indosat Ooredoo Hutchison (Indosat or IOH) announced that it has handled a memorandum of understanding (MoU) with Tech Mahindra, in developing Garuda, an Indonesian Large Language Model (LLM).

Built on the principles of LLM original Tech Mahindra 'Project Indus', a basic model designed to communicate in various Indian languages and dialects, this Garuda NGO is intended to preserve Indonesian and its entire dialects.

As part of this partnership, Tech Mahindra will leverage its technological expertise to collect and summarize data in Indonesian, which will then be trained first and then released as a conversation model for Indosat.

Later, this model is expected to be able to present personalized customer experience by analyzing sentiment and understanding behavior patterns, thus resulting in flexible learning based on previous interactions.

"Using the latest technology in preserving Indonesian language with its various dialects, we support linguistic diversity and increase accessibility and inclusion in the digital era," saidtengan Sinha, President Director and CEO of Indosat Ooredoo Hutchison in Mobile World Congress (MWC) 2024.

menjelaskan, Garuda akan dikembangkan dengan 16 miliar token bahasa Indonesia asli, yang akan menghasilkan 1.2 miliar parameter untuk membentuk pemahaman model terhadap bahasa Indonesia tersebut.

The beta version of the Garuda model will be released for testing by Indosat and Indonesian language speakers. After that, both parties will continue to improve their model using the RLHF (Reinforcement Learning from Human Feedback) technique to ensure reliability in the conversation.

It is estimated that the LLM market will reach 40.8 billion US dollars by 2029. In this context, the emergence of LLMs such as Garuda and Indus can facilitate online communication for individuals and companies in their dialects and local languages, creating new opportunities in the digital world,' concluded Harshvendra Soin, President - Asia Pacific and Japan Business, Tech Mahindra.


The English, Chinese, Japanese, Arabic, and French versions are automatically generated by the AI. So there may still be inaccuracies in translating, please always see Indonesian as our main language. (system supported by DigitalSiber.id)