May 23rd.NetEaseYesterday, it was announced that it was decided that the "sub" large model 4.0 core two engines --"Multimodal Model"and"Speech Synthesis（TTS) ModelsOfficially global in fullOpen SourceI don't know. Developers can download, deploy and redevelop on this basis free of charge。
THIS OPEN-SOURCE "SUB" MULTI-MODULAR MODEL (27B PARAMETER SIZE) IS ORIENTED TOWARDS THE EDUCATIONAL SCENE, SUPPORTING THE MATHEMATICAL CAPABILITY OF VISUAL INPUT AND ACHIEVING THE TOP INDUSTRY LEVEL (SOTA)。
- In the size model of the same parameter, handles a chartHard visual math.
- It's a Chinese-language problemACCURACY RATE 81.41 TP3T.
IN ADDITION, THE NEW MODEL USES A FINE-TUNED THINKING CHAIN RE-ENGINEERING PROGRAMME. THE LENGTH OF THE THOUGHT CHAIN OUTPUT HAS BEEN REDUCED BY IN-DEPTH OPTIMIZATION BY BRINGING TOGETHER LARGE-SCALE, HIGH-QUALITY, STREAMLINED SAMPLES OF REASONING. THIS MEANS ANSWERING THE SAME QUESTIONIt's output Token, less, shorter, faster.
The immediate effects for developers and enterprises doing actual business are:Decline in reasoning costs.
In addition, cyber-friendly teams target students in the countryReal job, test and question sceneDepth optimization has been made to enable it to address the complex pains encountered in real learning。
AND OPEN SOURCE VOICE SYNTHESIS (TTS) MODEL SUPPORTSTranslingual sound and sexual migration cloningIf you upload a Chinese audio, you can clone the voice of the speaker and speak fluently English, Korean, Vietnamese... without a Chinese accent. And emotions can be precise in moving cloning -- if you say one word in anger, the synthetic foreign language is also angry。
- 3 seconds: Upload any audio material so that the system can complete the original copy of zero samples within 3 seconds。
- 97%: MORE THAN 97% IN A CLONING MISSION AND 85% IN A CLONED SOUND SIMILAR TO THE ORIGINAL。
- 14 languages: CE, Japan, Korea, Germany, France, West, Indonesia, Italy, Thailand, Portugal, Russia, Malay, Vietnamese, etc。
1AI WITH THE FOLLOWING TWO-PART OPEN SOURCE ADDRESSES:
- Multimodel model: https://huggingface.co/netase-youudao/Confucius4
- TTS Model: https://github.com/netase-youudao/Confucus4-TTS