"Tiangong 3.0" has 400 billion parameters, surpassing grok-1 with 314 billion parameters, and is the world's largest open source moe model. "Tiangong 3.0" has breakthrough performance improvements in the fields of semantic understanding, logical reasoning, versatility, generalization, uncertainty knowledge, and learning capabilities, and its mathematics/reasoning/coding/cultural and creative capabilities have increased by more than 30%.
(Tiangong 3.0 model parameters surpass grok-1, becoming the world's largest open source moe expert hybrid model)
's powerful model technical strength gives "Tiangong 3.0" super performance. In many authoritative multi-modal evaluation results such as mmbench, "Tiangong 3.0" surpassed gpt-4v and took the lead in the world.
(Tiangong 3.0 multi-modal performance surpasses gpt-4v, leading the world)
At the same time, the "Tiangong skymusic" music model under "Tiangong 3.0" is also open to the whole society for public testing today. "Tiangong SkyMusic" is China's first music sota model, and it is the first time that China's self-developed large model technology has led the world in the AIGC field.
(The comprehensive performance of Tiangong SkyMusic surpasses that of Suno v3, and it has obtained the music large model SOTA, leading the world)
Tiangong SkyMusic: China’s first music AIGC SOTA model
Previously, large models have made breakthroughs in many technical fields such as text and images. Bring about comprehensive changes in the industry. However, in the field of AI music generation, the world has been waiting for a product to start the "music chatgpt moment".
This is because a lot of research in the AI music industry has been focused on the technical route of symbolic music generation, and most of them can only achieve the generation of background music (bgm) without voice. The quality, effect, and aesthetics of the music are far behind. Far from reaching usable levels, the industry has been slow to explode.
("Tiangong skymusic" self-developed ai music large model technical architecture)
Different from the mainstream path in the industry, "Tiangong skymusic" adopts the self-developed large model music audio generation technology route. This route directly uses large model technology to achieve integrated end-to-end music generation of instruments, vocals, melody, volume, and notes. It is extremely technically difficult. Only a handful of top players in the world, including Kunlun Wanwei, participate. In the horizontal evaluation of
with the top overseas AI music model suno v3, "Tiangong skymusic" significantly leads its opponents in areas such as vocal & BGM sound quality, vocal naturalness, and pronunciation intelligibility, with a comprehensive score of 6.65 points Surpass suno v3 and become the global AI music sota model.
In addition, "Tiangong skymusic" also has original reference music generation and dialect song generation capabilities.
reference music generation: Users can upload their own reference music, or select existing reference music in the "Tiangong skymusic" database to generate songs with similar styles and vocals, further lowering the threshold for using large music models and allowing unfamiliar people to Users with music theory knowledge can also play it easily.
dialect song generation: The music generated by "Tiangong skymusic" not only performs well in areas such as naturalness of human voices and intelligibility of vocalizations, but also supports many dialects such as Cantonese, Chengdu dialect, and Beijing dialect, allowing users to achieve their goals more freely. Musical expression and spread of dialect culture.
"Tiangong SkyMusic" is China's first publicly available AI music generation model, and it is the first time that China's self-developed large model technology has led the world in the AIGC field.
Currently, in the field of large text models, openai has attracted global attention; but in subdivided fields such as ai search and ai music generation, Chinese players are moving forward bravely and continue to obtain top sota in the subdivided fields through self-developed technology. Performance, jointly build China's large model industry and create an independent and controllable large model industry ecology.
Tiangong 3.0: 400 billion parameters, the world's largest open source moe large model
Based on the leading position of the previous generation "Tiangong 2.0" moe large model, "Tiangong 3.0" has achieved a comprehensive performance upgrade and adopted 400 billion parameters. The moe hybrid expert model architecture is currently the open source moe model with the largest model parameters and the strongest performance in the world.
"Tiangong 3.0" has comprehensively upgraded its logical reasoning capabilities, semantic understanding capabilities, ability to respond to complex needs, and content creation capabilities, and has added multiple rounds of search and comprehensive tool calls, chart drawing, research mode, enhancement mode, and image modification and expansion. Multiple AI capabilities such as graphics bring users a new AI experience.
multiple rounds of search and comprehensive tool calling: "Tiangong 3.0" has conducted special training on the model's ability to independently plan, call, combine external tools and integrate information, so that it can independently generate and call code, complete industrial research, product horizontal evaluation, information analysis, picture generation, chart drawing and other complex user needs.
At the same time, "Tiangong 3.0" can break down user tasks into subdivided links through its powerful semantic understanding capabilities, judge in real time whether it is necessary to connect to the Internet or call tools, conduct single or multiple rounds of network searches and tool calls, and complete tasks including multiple tasks. Round search, hot information analysis, image generation and other complex user needs.
chart drawing: "Tiangong 3.0" comprehensively improves the logical reasoning ability and the user's natural language query understanding ability, enabling it to more accurately judge user needs, independently generate and call codes, and perform real-time content analysis and chart construction based on text requirements. , bringing users more intuitive and efficient comparison results.
(query: Which one is more fun, Beijing, Shanghai, or Chongqing?)
Multiple rounds of searches, comprehensive tool calls, chart drawing, etc. are the unique large-scale model comprehensive capabilities of "Tiangong 3.0", which connects "Tiangong 3.0" from the bottom "'s underlying capabilities such as AI search, AI dialogue, AI code generation, AI picture recognition, and AI image generation are directly triggered by semantic recognition capabilities, bringing users a more convenient and efficient AI experience and becoming a true AI productivity tool.
In addition, "Tiangong 3.0" also adds many AI capabilities such as research mode, enhancement mode, image modification and image expansion.
research mode: In the research mode, "Tiangong 3.0" can extend relevant issues around a simple instruction from the user, and automatically generate research outlines, maps, practice summaries, and mind maps to help users quickly and clearly grasp the core content. , to complete users’ complex research needs.
(query: the heyday of Kangxi and Qianlong)
enhanced mode: In the enhanced mode, "Tiangong 3.0" can disassemble, refine, and question, understand and complete the user's complex query, making it natural It has stronger performance in semantic understanding, performs better in the face of uncertain knowledge, and can meet user needs more accurately and efficiently.
(query: 2024 Spring Festival movies; "Tiangong 3.0" understands and asks user needs)
changed the image and expanded the image: "Tiangong 3.0" has achieved a comprehensive breakthrough in multi-modal performance, surpassing gpt-4v, and ranking first in the world . With the support of a powerful technical base, the AI drawing capabilities of "Tiangong 3.0" have added new functions such as image size expansion, image orientation adjustment, mat drawing generation, mat drawing evolution, and mat drawing expansion.
