IBM has inked an settlement with AI Singapore (AISG) to check the latter’s Southeast Asian massive language mannequin (LLM) and make it obtainable for builders to construct personalized synthetic intelligence (AI) purposes.
Below the partnership, IBM will check the Southeast Asian Languages in One Community (SEA-LION) mannequin utilizing Massive Blue’s AI expertise and knowledge platform, Watsonx, and work with AISG to fine-tune the LLM. The purpose is to assist organizations select appropriate AI fashions for his or her enterprise necessities, IBM and AISG mentioned in a joint assertion on Tuesday.
IBM will even make SEA-LION obtainable in its AI use case library, dubbed Digital Self-Serve Co-Create Expertise (DSCE), enabling builders and knowledge scientists to construct localized generative AI (GenAI) purposes.
An open-source LLM developed by AISG, SEA-LION is designed to be smaller, extra versatile, and sooner than different LLMs, based on AISG. Its present iteration runs on two base fashions: a 3-billion-parameter mannequin and a 7-billion-parameter mannequin. The LLM’s coaching knowledge consists of 981 billion language tokens, which AISG defines as fragments of phrases created from breaking down textual content throughout the tokenization course of. These fragments embrace 623 billion English tokens, 128 billion Southeast Asia tokens, and 91 billion Chinese language tokens.
With SEA-LION, Singapore goals to drive the event of LLMs that higher replicate Southeast Asia’s societal combine and exhibit stronger contextual understanding of the area’s cultures and languages.
The partnership goals to push ahead a “custom-made basis mannequin” for Southeast Asia and made by Southeast Asians, based on Leslie Teo, AISG’s senior director of AI merchandise. The 2 organizations will even look to construct use instances, gasoline SEA-LION’s adoption, and assist organizations “scale AI safely and responsibly,” Teo mentioned.
The collaboration encompasses efforts to include AI governance into SEA-LION, so companies can higher navigate compliance, danger administration, and mannequin lifecycle administration, whilst authorities rules on AI proceed to evolve.
“[IBM] believes additional progress of GenAI will carry higher efficiency in smaller language fashions, with customers given the chance to personalize fashions primarily based on their enterprise and business necessities,” Catherine Lian, IBM Asean’s common supervisor and expertise chief, mentioned in an announcement.
“Nobody mannequin is a one-size-fits-all for companies, and organizations have to be empowered with a selection to make use of their fashions primarily based on their wants,” Lian mentioned. “[The] SEA-LION LLM is an enormous step ahead in creating an open AI system and addressing the Asean language challenges that corporations and governments face when working with AI.”
In March, AISG additionally introduced a partnership with Google to reinforce datasets used to coach, fine-tune, and assess AI fashions in languages particular to Southeast Asia. Known as Challenge Southeast Asian Languages in One Community Information, the initiative goals to “enhance cultural context consciousness” in LLMs constructed for the area.
Initially, the venture will give attention to Indonesian, Thai, Tamil, Filipino, and Burmese — languages for which AISG and Google will develop translocalization and translation fashions. They will even construct instruments to assist scale translocalization capabilities, share finest practices for tuning datasets, and publish pre-training guides for Southeast Asian languages.