Indonesian operator Indosat Ooredoo Hutchison and India’s Tech Mahindra, an information technology services and consulting company, have announced the signing of a memorandum of understanding (MoU) at MWC 2024 to build Garuda, a large language model (LLM) to preserve Bahasa Indonesia, the official and national language of Indonesia and its dialects.
Garuda will be built on the principles of Tech Mahindra's indigenous LLM Project Indus, a foundational model designed to converse in a multitude of Indic languages and dialects.
The partners say that the unique features of Garuda will create multiple use cases and enable Indosat's customers to leverage various applications, including customer support, experience, and content creation across industries such as healthcare, e-commerce, rural education, banking and finance, agriculture, and telecom.
The model will offer personalised customer experiences by analysing sentiment and understanding behavioural patterns, resulting in flexible learning based on customers' previous interactions. As part of this partnership, Tech Mahindra will leverage its technology expertise to gather and curate data in the Indonesia language, which will be pre-trained and released as a conversational model for Indosat.
Garuda will be developed with 16 billion original Bahasa tokens, providing 1.2 billion parameters to shape the model's understanding of the Bahasa language. These parameters will influence how the model processes input and formulates output. A beta version of the Garuda model will be released for testing by Indosat and Bahasa Indonesia speakers. The model will be further improved using RLHF (reinforcement learning from human feedback) techniques to ensure its robustness for conversation.
