Typhoon model adds Thai language to Samba-1

Posted by Keith Parker on July 10, 2024

The Samba-1 model from SambaNova offers enterprises an unmatched combination of data security and privacy, access control, and the ability to address multiple use cases with the fastest inference performance. Because Samba-1 is based on a Composition of Experts (CoE) architecture, it offers all these benefits in a way that cannot be found in any other generative AI model.

In addition to all these benefits, one of the most compelling features of the CoE architecture is the flexibility it offers to add in new models. The capacity to add models enables enterprises to add new capabilities as their needs change and to always have the latest open-source models for the best possible accuracy and highest performance.

Due to that capability SCBX, which is part of the SCBX financial technology business group, leveraged Samba-1 to make Typhoon-1.5X available. Typhoon is a series of Thai Large Language Models (LLMs) available in 8B and 70B sizes. Offering significant enhancements to Thai NLP and AI, it performs comparably to GPT-4. Built on a Llama 3 model, Typhoon features enhanced training and alignment techniques, including improved human alignment in instruct models. This model addresses the language gap prevalent in models predominately trained in English, rectifying the scarcity of resources for the Thai language.

Typhoon is available in two models, a pre-trained model and an instruction-tuned model. The pre-trained model  is focused on Thai language proficiency, including vocabulary, context, cultural nuances, and global knowledge. This is an open-source model that has been made freely available to promote competitiveness in Thailand’s AI industry. The instruction-tuned  model, which is accessible through an API service, builds upon this foundation to enable tasks such as translations, summaries, and question answering.

With the inclusion of Typhoon, Samba-1 is now able to deliver generative AI capabilities in the Thai language. This is in addition to the wide range of languages, including low resource languages, that Samba-1 was already capable of supporting.

Learn more about the many languages that Samba-1 supports with high performance and accuracy.

Topics: business, Blog