Deepseek, the viral artificial intelligence society, has released a new series of multimodal artificial intelligence models which, in its opinion, can overperform from the Openai 3.
The models, available for download from the development platform to Hugging Face, are part of a new family of models that Deepseek calls Janus Pro. Their size varies from 1 billion to 7 billion parameters. The parameters match approximately to the ability to solve the problems of a model and the models with multiple parameters generally work better than those with less parameters.
Janus Pro is under the MIT license, which means that it can be used commercially without restrictions.
Janus Pro, which Deepseek describes as a “new author -scale framework”, can both analyze and create new images. According to the company, on two benchmark of evaluation AI, Geneval and DPG-Bench, the largest Janus Pro model, Janus Pro 7b, beats from the 3 as well as models such as Pixart-Alpha, Emu3-Gen and Stability to the stable diffusion XL.
Of course, some of these models are older and Janus Pro can analyze and generate only small images with a resolution of up to 384 x 384. But the performance of the Janus Pro family are impressive, considering the compact dimensions of the models.
“Janus Pro exceeds the previous unified model and corresponds or exceeds the performance of the specific models for activities,” writes Deepseek in a post on Hugging Face. “The simplicity, high flexibility and the effectiveness of Janus Pro make it an excellent candidate for the next generation unified multimodal models”.
Deepseek, a Chinese laboratory of artificial intelligence financed largely by the High-Flyer Capital Management quantitative trade company, broke into the public opinion this week after its app chatbot rose to the top of the rankings of the Apple App Store. The linguistic models of Deepseek, trained using efficient techniques from a computational point of view, have led many analysts – and technologists – of Wall Street to wonder if the United States can maintain their leadership in the race to artificial intelligence and if the demand for chips Ai will be supported.
