► Mainstream open-source models in the industry do not support Thai language well. In addition, the open-source model does not open datasets and cannot be optimized. As a result, the Thai Q&A effect is poor. Therefore, Thailand DTGO develop the native Thai foundation model.
► Foundation model technologies are frequently updated and iterated. If a self-built data center is used to train DTLM foundation models, the infrastructure construction period is long and the AI cluster O&M technology is difficult.
Solutions
► The Ascend AI cloud service provides enterprises with fast AI computing power services at their fingertips. It covers the Asia Pacific region through Hong Kong. The average network latency is as low as 27.6ms, ahead of other cloud service providers. The unique multi-level recovery mechanism and complete tool chain of Ascend AI Service can ensure continuous training for 40 consecutive days. The task recovery time is less than 10 minutes, providing the best computing power for the development and running of large models and AI applications.