TianGong LM

heavenly workmanshipLarge ModelIt is Kunlun Wanwei's self-developed double-kilobillion large language model, and is also the first double-kilobillion large language model in China that benchmarks ChatGPT.

1. Model Background and Positioning::
   • Tiangong Da Model is independently developed by Kunlun World Wide Company, aiming to provide users with high-quality copywriting, knowledge quiz, code programming, logical deduction, mathematical reasoning and other services.
   • As China's first model to benchmark ChatGPT, the Tiangong Grand Model demonstrates the ability to match ChatGPT in several aspects.
2. Model version and release time::
   • Since April 17, 2023, the Skyworks model has gone through several iterations and upgrades.
   • On April 17, 2023, Tiangong 1.0 was released, followed by the launch of Tiangong AI Assistant App and Tiangong AI Search.
   • On February 6, 2024, "Tiangong 2.0" and the new version of "Tiangong AI Intelligent Assistant" APP were released, which further improved the model performance and user experience.
   • On April 17, 2024, the "Skyworks 3.0" base model and its "Skyworks SkyMusic" music model were officially launched for public beta testing, showing more advanced and powerful functions.
3. Model Features and Benefits::
   • The TGU model adopts the advanced MoE (Mixture of Experts) architecture, which supports more than 150,000 Chinese character context windows and is capable of handling more complex and rich text information.
   • Tiangong version 3.0 demonstrates significant performance improvements in several areas, including model semantic understanding, logical reasoning, generalization, and generalization.
   • Tiangong Big Model is a multimodal "super model" that integrates AI search, AI writing, AI long text reading, AI dialog, and other capabilities to provide users with a one-stop AI service experience.
4. Model Application and Effectiveness::
   • Tiangong Da Model has been widely used in many fields such as copywriting, knowledge quiz, code programming, etc., and has shown good application results.
   • In the model evaluation, the TGU model performed well on the SuperCLUE benchmark, further narrowing the gap between the Chinese closed-source model and GPT3.5.
   • The TIU model has achieved excellent results on several review datasets, especially surpassing GPT-3.5 and LLaMA2-70B on the inference review set GSM8K.
5. Model development and outlook::
   • Tiangong Da Model continues to iterate and upgrade, introducing new technologies and features to meet the growing needs of users.
   • Kunlun will continue to invest in R&D resources to promote the application and innovation of the Tiangong Big Model in more fields.
   • With the continuous advancement of technology and the expansion of application scenarios, the Skyworks model is expected to become one of the important cornerstones in the field of artificial intelligence in the future.