SCB 10X Hosts "Typhoon 2 Unveiled: Advancing AI Research in Thailand" Forum to Advance Thai AI Research with the Latest Thai Language Model

"Typhoon 2" has been designed to respond to a wider range of uses. It is available in five sizes (1B, 3B, 7B, 8B, and 70B) with outstanding key features including:

• Increased efficiency and accuracy of Thai language processing, assessed by standards such as ThaiExam and M3Exam, which show superior performance compared to other open-source models.
• Improved instruction-following performance compared to Typhoon 1.0 and Typhoon 1.5/ 1.5X, assessed by IFEval-TH and MT-Bench standards.
• Extended context length from 8,192 to 128,000 tokens, allowing the model to handle more complex data and generate longer-form text.
• Enhanced Function Calling capabilities delivering best-in-class performance.
• Support for mobile devices with a smaller model for simple tasks such as summarizing data and translating languages.
• Safety Classifier Model (Preview) adjusted for the Thai context, which helps assess the appropriateness of input words and provides warnings if inappropriate content is detected.

In addition, 2 Multimodal models have been developed that supports the processing of text, audio, and images, opening up new opportunities for use in multiple contexts and enhancing the utility of "Typhoon 2". There are two primary models:

1. Typhoon2-Audio (Research Preview)
   • End-to-end architecture: Supports both text and audio input, with the ability to generate text and audio outputs simultaneously.
   • Improved voice analysis: It can understand content and distinguish emotions in audio.
   • Enhanced efficiency in processing complex commands: Supports interactions requiring precision and multi-step operations.
   • Text-to-Speech function: Converts text into Thai audio with superior quality compared to other open-source models.
2. Typhoon2-Vision (Research Preview)
   • Enhances image processing and understanding: Supports detailed image analysis and comprehensive content understanding.
   • Built-in OCR (Optical Character Recognition) capability: Converts text from images or documents into digital format, suitable for data management in various tasks.

Disclaimer: Both Typhoon2-Audio and Typhoon2-Vision models can be tested and applied in various applications. However, they are still in the development and experimental research phase, which may not be suitable for fully commercial use.

"Typhoon 2" is supported by leading partners in Thailand and abroad, including VISTEC, Mahidol University, the Artificial Intelligence Association of Thailand (AIAT), AI Singapore (AISG), Stanford Human-Centered Artificial Intelligence (Stanford HAI), Together AI, SEA AI Lab, InnovestX, and Sambanova Systems. These collaborations aim to drive the application of large Thai language models across various industries.

Those interested in joining "Typhoon 2 Unveiled: Advancing AI Research in Thailand" can register at https://lu.ma/droty4z9 or follow information about "Typhoon 2" and try out the latest version of the large Thai language model at https://opentyphoon.ai/.

• ข่าวo:editor
• ข่าวo:loc
• ข่าวen
• ข่าวo:it