The Find X7 series brings the industry's first 7 billion parameter model for end-side applications
In 1831, Michael Faraday discovered electromagnetic induction, which created the electric generator. What really promoted the widespread use of electricity in human life was the invention of the tungsten light bulb by Thomas Edison nearly half a century later.
In 2022, ChatGPT was born, and the AI model is obvious to all to the subversive changes in human society, so where is the "tungsten light bulb" that can promote the popularization of AI models?
On January 8, 2024, the OPPO Find X7 series was officially released, in addition to the powerful hardware configuration, which can be called the god's mobile image, the Find X7 series is through high-precision 4bit quantization and other model compression, AI Boost inference engine, and hardware acceleration with in-depth cooperation with the chip platform, for the first time to bring a complete 7 billion parameter large model to the mobile phone side, completely changing the way the mobile phone side AI is used, and officially opening a new era of AI mobile phones.
It is no exaggeration to say that OPPO's 7 billion parameter model has entered a new stage in the development of mobile applications, making mobile phones a more intelligent platform, and users can enjoy more personalized and intelligent services through mobile phones.
What should a real AI model look like?
For users, AI mobile phones are both familiar and unfamiliar concepts, and mobile phone voice assistants represented by Apple's Siri have long been nothing new, but the functions they can achieve, the accuracy of speech recognition, and the depth of understanding of language are difficult to satisfy, and these artificial intelligence assistants are often ridiculed as "artificial intellectual disability".
By combining large models with mobile phones, we can break through the existing limitations and bottlenecks. With the help of mobile phones as the entrance to the Internet, we can better understand user needs and continuously optimize the application of large models through deep learning Xi and natural language processing technology. At the same time, it also enables the large model to better understand market needs and product defects, so as to make more informed decisions and provide personalized services to each user.
So, what are the characteristics of a truly smart AI phone and a large model that users can reach at their fingertips? It should have its own personality and intelligence, and be able to accurately understand the needs of users and realize self-learning Xi at the edge. Such an AI assistant can also obtain the user's information and behavior Xi in real time, have personality and memory, and share joys and sorrows with the user. On the premise of ensuring user privacy and security, it will bring faster response speed, stronger processing power, and higher generation quality, providing users with high-quality local AI experience and services.
However, as a leading smartphone manufacturer, OPPO has extremely high requirements for computing infrastructure, and there are also difficulties such as large model size, high computing power requirements, and uncontrollable security on landing mobile phones.
On the Find X7 series, OPPO has implemented a large language model with 7 billion parameters for the first time on the device side, and its intelligence capabilities have been comprehensively improved. The so-called device-side large model is a large model that is directly deployed on the user terminal, that is, the mobile phone. OPPO's self-trained, personalized and exclusive large model, AndesGPT, contains a variety of versions from one billion to hundreds of billions of parameters, rich in knowledge, unlimited memory, and good use of tools, and the technical architecture of device-cloud collaboration can support more diversified application scenarios, and of course, it can also expand more application scenarios. It is worth mentioning that the large model can directly use the computing power of the mobile phone processor to run, and in most scenarios it does not require network communication, and it can also be used in airplane mode, which is equivalent to the user getting the large model when buying a mobile phone, and the subsequent use cost is zero.
At present, with the introduction of AI large models into the field of mobile phones, the bottlenecks of existing mobile phone computing power and storage will gradually become prominent, and the promotion of large computing power chips has become a key step in cracking the landing of mobile phone side large models. In order to ensure the high-speed operation of the large model, OPPO used the exclusive tidal architecture technology for the first time. As a chip software and hardware integration technology stack, Tidal Architecture has built a set of chip-level performance solutions from scene identification, to computing power demand, cache matching, to computing power scheduling, to maximize the energy efficiency performance of flagship chips.
Mobile AI truly becomes your personal assistant
OPPO's core goal for the layout of large models is to build and optimize the interactive experience of smart terminals, so that users can get a smoother, more convenient and more user-friendly experience, so as to accelerate the upgrading of life. Behind this is not only the need for strong hardware support, but also the operation logic of innovative software systems and the design and function of innovative interactions. So, in the actual use scenario, what different experience does the OPPO Find X7 series end-side large model bring to users' lives?
1.AI large model voice summary: It can automatically summarize and extract key information in the call, and can also quickly find important content, making work and life easier. In terms of generation quality, compared with the 1 billion parameter model, AndesGPT's 7 billion parameter large language model brings a more complete and accurate record of call summarization with higher "IQ" comprehension ability.
2. Intelligent summary of articles: It can be read quickly, making it more convenient for users to obtain news information on a daily basis. Compared with other models on the same platform, OPPO AndesGPT's 7 billion parameter large language model can bring 20 times faster response in 200-word initial word generation, and 2.5 times faster speed for 2000-word initial word generation. The maximum number of abstract words can be 14,000, which is 3.5 times that of other models on the same platform.
3. AIGC elimination: Based on the self-trained AndesGPT generative vision model, it brings a new functional experience of subject elimination and real-world redrawing. Compared with other cloud models on the same platform, which can only support the recognition of the human body, AndesGPT not only supports the recognition and segmentation of more than 120 types of subjects, but also realizes hair-level segmentation and up to 6 multi-subject separations, as well as the filling and natural generation of super-large images, which greatly expands the scope and practicability of generative vision models, and in addition, the end-to-end application of AndesGPT generative vision models also brings 60% compared with other models on the same platform The shorter generation time, as well as the higher generation accuracy, help users better process photos.
Leverage the large model to win the high-end market
According to a report released by market research agency Canalys, in the third quarter of 2023, shipments in China's smartphone market fell flat for two consecutive quarters, down 5% year-on-year to 66.7 million units. China's mobile phone market is still declining as a whole but showing signs of recovery, and high-end phones above $600 are still strong in the Chinese market. At the same time, there is a general consensus in the industry that mobile phone manufacturers are committed to creating a new generation of intelligent ecological environment covering devices, cars and connected terminals, and the large model is the ecological base. Therefore, in the face of the dilemma of the development of the mobile phone industry, the AI model may be the best breakthrough direction, and the integration of AI and terminals is regarded as a new innovation anchor, which will drive the resonance of the industrial chain.
There is no doubt that only by grasping the cutting-edge and advanced technology can we have stronger competitiveness, so only by better innovation and reform can mobile phone brands have stronger competitiveness and better enter the global high-end market. In the future, the large model is no longer a marketing phrase or a simple function to apply a layer of AI skin, only by establishing a real and usable AI large model like OPPO and continuously optimizing it, making AI more intelligent, and closely combining AI with user experience and needs, is the future development direction of the industry, and more people will truly feel the convenience brought by AI to life.