"Tiangong Big Model 3.0" was officially released on April 17th - a 400 billion parameter MoE super model that is simultaneously open source and has performance exceeding Grok1.0

On April 17, 2023, Kunlun Wanwei released its self-developed double-hundred-billion-level large language model "Tiangong 1.0", officially paving the way for the rise of domestic large-scale models.

On the upcoming April 17, 2024, on the first anniversary of the "Tiangong" model, Kunlun Wanwei announced that "Tiangong 3.0" has officially launched public beta!

"Tiangong3.0" adopts 4100 billion level parameters The MoE hybrid expert model will be open sourced at the same time. It is one of the MoE models with the largest model parameters and the strongest performance in the world. Compared with the previous generation "Tiangong2.0"MoE large model, "Tiangong 3.0" has amazing performance improvements in areas such as model semantic understanding, logical reasoning, versatility, generalization, uncertainty knowledge, and learning capabilities. Its model technical knowledge capabilities have improved by more than 20%, Mathematics/Reasoning/Code/ The cultural and creative ability has improved by more than 30%.

At the same time, "Tiangong3.0" has added search enhancements, research mode, calling code and drawing charts, and multiple calls to the Internet Search and other capabilities, and specifically trained the Agent ability of the model, so that "Tiangong3.0" can independently complete planning, Call and combine external tools and information to accurately and efficiently complete various complex needs such as industrial analysis and product comparison, bringing a new disruptive artificial intelligence experience.

At the same time, "Tianchao 3.0" is also the world's first multi-modal "Super Model" (Super Model), which integrates AI search, AI writing, AI long text reading, AI dialogue, AI Speech synthesis, AI image generation, AI comic creation, AI image recognition, AI music generation, AI code writing, AI form generation and many other capabilities are "super applications" in the era of large models.

Among them, "Tiangong3.0"AIMusic generation large modelSkyMusic will also open an invitation test to the public on 4#2 (tomorrow).

Four major innovations Disruption and upgrade

The MoE hybrid expert model is the global The foundation model technology path with the most advanced technology and the most powerful performance. Compared with other models, the MoE large model has stronger ability to handle complex tasks, faster model response, higher training and inference efficiency, and scalability. Stronger.

Based on the previous generation "Tianqiao 2.0" MoE large model, "Tianqiao 3.0" has achieved a comprehensive performance upgrade, including a 400 billion-level parameter MoE hybrid expert model architecture. It is one of the MoE models with the largest model parameters and the strongest performance in the world.

The model capabilities of "Tiangong 3.0" are improved in the following four aspects:

1. Stronger logical reasoning ability: smarter

The improvement of logical reasoning capabilities is crucial for large models to solve complex problems. The mathematics and reasoning capabilities of "Tiangong 3.0" have both improved by more than 30%. The powerful logical reasoning capabilities enable it to process information more accurately and efficiently in practical applications. . For example, in the "Tiangong 3.0" AI search research model, the model can extend relevant questions around a simple instruction from the user, and determine in real time whether the information in this paragraph needs to be searched online. This can enable detailed analysis of an industry, for example. Dismantling analysis, summarizing relevant events, dismantling industrial chain maps and other complex functions, and finally displaying it in the form of a structured or mind map to make the model more "smart".

2. Stronger semantic understanding: understand you better

"Tiangong 3.0" can better understand and process the complex semantic information in the user's natural language Query , including metaphors, polysemy, etc. For example, in the enhanced search of "Tiangong 3.0" AI search, the model can disassemble and refine the user's complex Query, and perform questioning, information understanding and completion, making it more powerful in terms of natural semantic understanding. It performs better when faced with uncertain knowledge and can meet user needs more accurately and efficiently.

3. Special Agent training, stronger ability to cope with complex needs: more versatile

In the era of large models, AI Agent has become the mainstream implementation direction of large model technology. "Tiangong 3.0" has conducted special training on the model's ability to independently plan, call, and combine external tools and information, allowing it to independently generate and call code, including industrial research, product reviews, information analysis, picture generation, and chart drawing. and a variety of complex user needs, and become an all-round expert with professional knowledge and capabilities in multiple fields. Use strong semantic understanding and logical reasoning capabilities to deeply understand user needs, break down tasks into subdivided links, and send them to different Use the optimal model to process and maximize model performance. At the same time, for B-side users, "Tiangong 3.0" has also been comprehensively upgraded in areas such as knowledge base capabilities, the ability to call any tool, and the ability to trace complex role instructions. Enterprise users can build exclusive knowledge bases and agents by uploading knowledge documents, and Realize practical capabilities such as automatically calling formulation tools and completing complex instructions to follow Agent construction.

4. Comprehensive upgrade of content creation capabilities: versatile

Content creation capabilities have always been the strength of the "Tiangong" series of large models. In the previous generation of "Tiangong 2.0" On the basis of the large model, "Tiangong 3.0" has undergone a comprehensive content creation capability upgrade. It can not only achieve powerful content creation capabilities such as AI music generation, AI voice, AI dialogue, and AI two-dimensional comic generation, but also Through special Agent training, the ability to generate images in real time based on text requirements during conversations, real-time content analysis and chart construction based on text requirements has been achieved, making the agent truly capable of searching, writing, reading, chatting, listening, speaking, and drawing. , a super model that can see and sing, bringing a new subversive AI experience upgrade.

The world's first "Super Model"

"Tiangong 3.0" is a This large artificial intelligence model integrates many cutting-edge technologies such as natural language processing, computer vision, multi-modality, AI search, and AI agents. It is also the world's first multi-modal "super model".

The concept of "Super Model" was born from "Super App". In the Internet era, a super application is an application that integrates multiple services. Users can enjoy multiple functions such as communication, payment, shopping, social networking, and travel on one platform. These services can connect and interact with each other to maximize satisfaction. comprehensive user experience.

Super model is the inevitable development direction in the era of large models, and it is also the strategy that Kunlun Wanwei Tiangong series of large models has always adhered to. Fang Han, chairman and CEO of Kunlun Wanwei, said that "super models" are inevitable for the development of the era of large models. In the future, there will be more than one "super model" in the industry, and Kunlun Wanwei will continue to work hard in this direction. Continue to provide users with smarter, more efficient and more reliable artificial intelligence services.

If you want to know more about the new AI function upgrade of "Tiangong3.0", please continue Follow the official account of Kunlun Wanwei Group and lock in 4月17#" TiangongAI Assistant”App, enjoy the shock of super modelAIExperience.

All in AGI and AIGC

will be released since April 17, 2023. After the billion-level large language model "Tiangong", Kunlun Wanwei created a series of disruptive cutting-edge AI products around the self-developed "Tiangong" series of large models:

In August 2023, Kunlun Wanwei Tiangong AI Search, the first domestic AI search product, was launched; in September, Kunlun Wanwei launched the multi-modal large model Skywork-MM, which ranked first in the comprehensive score in the multi-modal large language model evaluation MME; in October, Kunlun Wanwei Weiyuan Kaiyuan's tens of billions-level large language model Tiangong Skywork-13B series; on December 1, Kunlun Wanwei released Tiangong SkyAgents, the leading domestic AI Agent development platform; in February 2024, the Tiangong base large model will be ushered in Tiangong 2.0, the largest version update since its launch, has become the first domestic AI application equipped with MoE architecture and open to all C-side users for free with hundreds of billions of parameter large language model AI applications.

Currently, Kunlun Wanwei Tiangong series of large models have made remarkable achievements in technology, products, cooperation, social recognition, awards and honors, etc., and have won recognition from all walks of life. Based on the Tiangong series of large models, Kunlun Wanwei has built an AI business matrix including AI large models, AI search, AI music, AI social networking, AI animation, AI games, etc. It is the company with the strongest model technology and engineering capabilities and the most comprehensive layout in China. One of the artificial intelligence companies.

On the first anniversary of April 17, 2024, the shocking release of Kunlun Wanwei’s “Tiangong 3.0” not only achieved a major breakthrough in AI technology, but also profoundly affected the development direction of the AI ​​large model industry, leading The AI ​​industry reaches a new milestone. Driven by the "All in AGI and AIGC" strategy, Kunlun Wanwei has always been committed to the innovation and development of AI technology, constantly lowering the threshold for users to learn and use AI, continuing to promote the AI ​​business to new heights, and improving the quality of many AI products. User experience, work with users to explore the unknown world and create a better future.

The above is the detailed content of "Tiangong Big Model 3.0" was officially released on April 17th - a 400 billion parameter MoE super model that is simultaneously open source and has performance exceeding Grok1.0. For more information, please follow other related articles on the PHP Chinese website!