Rwanda Sugar Baby Wen Xin fully evolved, the pioneer went to the galaxy

作者:

分類:

Huaqiu PCB

Highly reliable multilayer boardRwandans EscortManufacturer

Huaqiu SMT

Highly reliable one-stop PCBA smart manufacturer

Huaqiu Mall

Self-operated spot electronic components Device Mall

Rwandans Sugardaddy

PCB LayoutRwandans Escort

High multi-layer, high-density product design

Steel mesh manufacturing

Focus on high-quality steel mesh manufacturing

BOM ordering

Specialized one-stop purchasing solution

Huaqiu DFM

One-click analysis of hidden design risks

Huaqiu certification

Certification testing is beyond doubt


The history of the information revolution in the past few decades tells us that the success of science and technology does not lie in the scale of the industry or the number of equipment, but in the number of devices. People lie in their imagination and creativity.

When teenagers develop amazing AI applications on campus; when office workers use AI to solve problems that have plagued the industry for a long time; when an ordinary person realizes his dream by becoming a developer, then we can There is no doubt that technology has won and AI has won.

Since 2023, the Chinese market has ushered in the “hundreds of models rushing to take the exam”of grand occasion. But when we put our perspective on specific AI developers, we will find that the difference that developers can feel is not big. Most large AI models cannot truly become capabilities and tools in the hands of developers. The ecosystem of AI developers around large models is difficult to consolidate. There is a relatively obvious separation between model and application development. There is a clear generation gap between technological evolution at the model level and the real needs of developers.

How to resolve this situation? How can big models really leave the hands of developers?

On April 16, the Create 2024 Baidu AI Developer Conference with the theme of “Creating the Future” was held at the Shenzhen International Convention and Exhibition Center. During the period, Baidu Chief Technology Officer Wang Haifeng delivered a speech with the theme of “Technological Foundation, Galaxy is Bright”, explaining the key technologies and latest progress of many cultural large-scale models such as agents, codes, and multi-models. Wang Haifeng said, “In the future, we will continue to devote ourselves to cultivating talents, so that the stars of talents can merge into a brilliant galaxy. In the intelligent era, everyone is a pioneer and everyone is an inventor. Let us work together to create a better future.” .

Unlike most large model projects, the emphasis is on parameters, accuracy and other indicators. Wenxin Big Model has chosen another path: to understand the needs and ideals of developers, and Wenxin will do whatever developers need.

Along this trajectory, Wen Xin evolved all the way, showing great technological explosive power. This not only established the scale effect of the Wenxin large-scale model, but also established its long-term development direction.

Wenxin’s future is very clear: by connecting the developers’ dreams together, a splendid galaxy of AI large-scale models can be formed.

Where is the galaxy? The developer’s dream is the galaxy

Today we can clearly say that the Wenxin large model is one of the large models with the fastest technological advancement and the most frequent systematic upgrades in China, and even in the world. Since the release of the knowledge-enhanced large language model Wenxin on March 16 this year, Baidu has continued to promote the upgrade and iteration of the Wenxin large model.

And every version upgrade shows the evolution of core technologies. Speed ​​has become a unique sight in the AI ​​industry.

For example, in Wenxin large model version 3.0, knowledge enhancement, search enhancement and dialogue enhancement technologies have been fully added; Wenxin version 3.5 RW Escorts This principle brings a comprehensive upgrade of the basic model, adding capabilities such as fine-tuning technical innovation, enhanced knowledge points, and enhanced logical reasoning; Wenxin 4.0 realizes large-scale parallelization of Wanka’s computing power Training, through joint tuning with the flying paddle platform, brings technical capabilities such as multi-dimensional data, multi-stage alignment, and reproducible training.

In the process of large-scale and high-efficiency upgrades, Wenxin’s large model capabilities have become more powerful, and its effects and performance have been comprehensively improved. At this time, many friends will be curious, how did Wen Xin discover Rwandans Sugardaddy So many technical upgrade directions? Why maintain such a high-intensity upgrade trend?

In Create 2024, we can find the answer.

From a developer’s perspective, AI large models are bringing about a cross-generational development singularity. The original application development gap was suddenly broken. AI has brought a huge number of new application types, a comprehensive lowering of development thresholds, and unprecedented development efficiency. But to truly realize the dream of AI application development, new technical foundations are needed as an anchor. The ability to truly turn AI large models into the hands of developers needs to solve three major problems today:

1. How to use AI large models as a basis to break past development gaps and turn those unprecedented development ideas into To be able to? How to achieve more intelligent and cutting-edge AI development?

2. How to take a further step to lower the development threshold and realize a low-code or even no-code development paradigm, thereby realizing the vision of the era in which everyone is an AI developer?

3. How to reduce development costs, improve development efficiency, and further improve the application development mechanism based on AI large models to make AI application development more maneuverable and implementable?

Breaking barriers, lowering barriers, and optimizing mechanisms are currently the three most urgent wishes of developers for AI large models. The latest upgrade of the Wenxin large model also just responds to the needs of developers.

Wang Haifeng said, “I hope that the little stars of talents can merge into a splendid galaxy.”

For Wenxin Big Model, only by responding to the needs of developers and understanding their wishes can these stars be lit.

For Wenxin Big Model, Rwanda Sugar Daddy The dream of AI developers is to pursue every step of the way of the galaxy.

Intelligent agents, crossing the development gap

First, let us look at how the Wenxin large model breaks the development gap and brings developers the most cutting-edge technological possibilities.

Today, the most hotly discussed technical direction in the global AI industry is intelligent agents. The intelligent agent mechanism with multiple capabilities such as thinking, analysis, and reasoning is widely regarded as the future development direction of AI technology. It is also an important pivot for the implementation of large models into native AI applications.

wKgaomYedvqADVmIAAA58IxRfIQ821.jpg

In October this year, Baidu introduced an agent mechanism into the Wenxin model. Wang Haifeng shared his understanding of fast thinking and slow thinking. At the just Create 2024 Baidu AI Developer Conference, intelligent agents were officially added to the technical architecture of Wenxin Large Model. Wang Haifeng said, “Intelligent agents are an important development direction and will bring about more application explosions. Intelligent agents are a further step to carry out thinking-enhanced training on the basic model, including supervised fine-tuning of the thinking process and action decision-making. Diligent learning of planning and enhanced learning of result reflection can lead to the acquisition of thinking models. The thinking model of an agent is like a human being, who can read instructions and learn how to use tools, and then can call tools to complete tasks.”

At the scene, Wang Haifeng demonstrated how to call tools on the Wenxin Big Model 4.0 tool version. Xiang Wen Xinyi asked, “I am going to the Greater Bay Area for a business trip for a week. I want to think about the weather changes so I can decide what clothes to bring. Please help me check the temperatures in Beijing and Shenzhen in the next week and tell me what clothes I should bring for the business trip. , and organized into tables.”

Faced with this series of complex questions, as well as the very specialized requirements for information processing and table generation, Wen Xin’s intelligent agent mechanism can think and plan like a human being, thereby breaking down the requirements into multiple As a subtask, first call the “Advanced Networking” tool to query weather information, then call the “Code Analyzer” to draw a temperature trend chart, and then choose appropriate clothing based on the weather conditions in the coming week. Finally, consider and confirm the results. And take the initiative to summarize it into a table.

Such technical capabilities have far exceeded the capability boundaries of previous AI technology and even large model technology. Developers can explore a more diverse and broader intelligent future through the new technology base.

Breaking the gap is the first gift that Wenxin 4.0 brings to developers.

Smart code, changes in development paradigm

We all know that the traditional application development paradigm is based on code writing in different programming languages. The biggest change that AI large models bring to the development paradigm is that the large models themselves can help generate code, and even achieve application generation without code.

In response to the widespread needs of AI developers and potential AI developers for low-threshold development, Baidu has developed code intelligence based on the natural language capabilities and coding capabilities of Wenxin large models Rwandans Sugardaddy and smart code assistant.

wKgaomYedvqAMRQWAABVOdyegdM168.jpg

Among them, code agents can realize code-free application development, allowing everyone to become an AI developer.

Code agents Rwandans Escort is technically based on the thinking model and code interpreter. First, the thinking model will understand the user needs, and after thinking, combine the instructions and instructions to complete the task. The relevant information is integrated into the prompts and output to the code interpreter. Then, the code interpreter interprets the user RW Escorts according to the prompts. It needs to be translated into code and executed to obtain the execution results or debugging information. Then, the thinking model will reflect and confirm the execution results of the code interpreter. If it is correct, the result will be returned to the user. If it is not correct, it will continue to iterate independently and replace it with new ones. Materials. Through such a logic mechanism, the agent can fully understand the user’s code generation needs and be responsible for the results, realizing “AI application development can be achieved by just speaking”

Intelligent code assistant. , which can help professional programmers write high-tool quality code more efficiently and more conveniently. On the basis of continuous improvement of model performance, Baidu has further built an intelligent code assistant with context enhancement, private domain knowledge enhancement, and RW Escorts capabilities such as seamless process integration

Currently, the overall adoption rate of the intelligent code assistant Comate has reached 46%. The proportion of new code generated has reached 27%, and Comate capability can seamlessly integrate code understanding, generation, optimization and other capabilities into all aspects of the R&D process, helping to improve the quality and efficiency of code development tools. a href=”https://rwanda-sugar.com/”>RW EscortsWith simple instructions, you can quicklyRwanda Sugar Quickly understand the structure of the entire code, and even the specific implementation logic of each module. You can also automatically generate new code that meets the requirements based on the current project code and third-party code, thus making the developer’s work easier and gaining more benefits. Support and assistance are more comprehensive.

Code assistance and code-free generation are the basis for the reform of development paradigm in the AI ​​era. Baidu has continued to work hard in these two aspects, giving AI developers the second important gift-super. Low-threshold development opportunities.

Multi-model, optimization of development mechanism

In addition to agent and code capabilities, Baidu also focuses on releasing multi-model technology for developers to actually develop AI applications.During the process, the call and selection of AI models are very complicated, and there are often problems that require multi-model matching based on different scenarios, and model switching based on the surrounding conditions of software and hardware. However, various compatibility difficulties and forced choices often arise during the multi-model selection process.

In order to achieve a balance between effect, efficiency and cost of large model applications, Baidu has launched multi-model technology.

First of all, it is a model that fully supports efficient and low-cost Rwanda Sugar. In order to achieve high-efficiency and low-cost model birth, Baidu has developed a training mechanism for large and small model collaboration, which can effectively carry out knowledge inheritance and efficiently give birth to high-quality small models with high tools. Small models can also be used to achieve comparative enhancement and assist in the training of large models. .

At the same time, Baidu has established a seed model matrix, a data improvement and enhancement mechanism, and a complete set of pre-training, fine-tuning and alignment, model compression, and inference deployment Rwandans SugardaddyEast-West chain. The efficient and low-cost model childbirth mechanism can help application development to be faster, lower cost, and better.

In terms of RW Escorts multi-model reasoning, Baidu has developed end-to-end multi-model reasoning technology based on feedback learning and built Intelligent routing models carry out end-to-end feedback learning and give full play to the ability of different models to handle different tasks to achieve a balance between effect, efficiency and cost.

Facing the future, multi-models have become an important trend in the implementation of large AI models, and Baidu once again takes the lead and uses it as the third gift to developers.

Progress all the way to the galaxy

Insight into the needs of developers and help develop their dreams. This simple logic constructs the progress coordinates of Wenxin’s large model, allowing it to continue running and move forward courageously.

In addition to agents, codes and multi-model technologies, Wenxin Large Model has also made continuous innovations in many aspects, including a data system based on model feedback closed loop, large model alignment technology based on self-reflective enhancement, and Multimodal techniques and more. Overall, the performance of Wenxin Large Model 4.0 has continued to improve. In the six months since its release, the overall performance has increased by 52.5%Rwanda SugarFruit.

wKgaomYedvuAGKOsAABa13aRE5Y720.jpg

The continuous rapid evolution of Wenxin large models benefits from Baidu’s full-stack layout in chips, frameworks, models and applications, especially The joint optimization of the Feipiao deep learning platform and Wenxin’s large model has achieved a weekly average training effectiveness of 98.8%. Compared with when Wenxinyiyan was released a year ago, the training efficiency has increased to 5.1 times and the inference rate has increased to 105 times.

In the continuous technological innovation, Wenxin large models have become the first choice for AI developers so far. Feipiao Wenxin ecosystem has gathered 12.95 million developers, served 244,000 enterprise work units, and created 895,000 models based on Feipiao and Wenxin. The cumulative user scale has reached 200 million, with average daily calls. It has reached 200 million, meeting the intelligent needs of thousands of industries.

wKgZomYedvyAGZ9IAABEWBFrz28672.jpg

In the key talent training link, Baidu proposed in 2020 that the goal of cultivating 5 million AI talents for the whole society in five years has been achieved ahead of schedule. It has become the only choice for AI developers to start their dream journey

Use the developers’ real needs as the guide, the developers’ broad expectations as the roadmap, and the developers’ great dreams as the future. This is how Wenxin progresses all the way, insisting on high-speed iteration and comprehensive evolution. Wenxin will continue to evolve. Today, on the day when countless AI developers realize their dreams, on the day when China’s science and technology fully awakens the power of new generations, on the day when AGI dawns, we meet across the galaxy

Review Editor Huang Yu


Baidu Wenxin Big Model 4.0 does not cost money. Baidu Wenxin Intelligent Platform does not cost money to open Wenxin Large Model 4.0 It is reported that from July 5, Baidu Wenxin Agent Platform (AgentBuilder) will open Wenxin Large Model 4.0 to develop Rwandans at no cost. When the Escort creates an agent on the Wenxin agent platform; it can be maneuvered Published on 07-06 17:12 •1088 views
Baidu released the intelligent code assistant “Wenxin Quick Code” At the WAVE SUMMIT conference held recently, Baidu has launched a new intelligent code assistant – “Wenxin Quick Code”. This powerful tool based on Baidu Wenxin large model combines massive programming data and is designed to create 's avatar Published on 07-01 11:Rwandans Escort20 •694 views
Baidu WAVE SUMMIT Deep Learning Developer Conference, Wenxin Large Model 4.0 Turbo Shock Released on June 28 , the attention of the technology community has focused on the Baidu WAVE SUMMIT Deep Learning Developer Conference. This event is not only about technical transportation Rwanda Sugar Daddy’s feast is a stage for Baidu to show off its latest achievements in the field of artificial intelligence. At this conference, Baidu Chief Technology Officer, Deep Learning Technology's avatar Published on 06-28 16:30 •389 views
OpenHarmony Developer Year Night Party 2024: Hongxin Gathers Strength and Wisdom to Lead the Future 2
Rwanda Sugar May 25, 2024, OpenAtomOpenHarmony (referred to as “OpenHarmony”) Committee With the theme of “Gathering Great Hearts and Leading the Future with Wisdom”, the OpenHarmony Developer Conference was held in the innovative city of Shenzhen 's avatar Published on 05-28 08:32 • 282 views
Baidu Chief Technology Officer Wang Haifeng explains the key technologies and latest progress of Wenxin Big Model. On April 16, the Create 2024 Baidu AI Developer Conference with the theme of “Creating the Future” was held at the Shenzhen International Convention and Exhibition The middle victory holds. Wang Haifeng, Chief Technology Officer of Baidu, delivered a speech titled “Technological Foundation, Galaxy is Bright”, interpreting multiple literatures such as agents, codes, and multi-models. Published on 04- 18 09:20 •551 views
Huawei Developer Conference will be held in Dongguan. Hongmeng Galaxy Edition will be officially launched in public beta in June. Huawei also revealed that in the fourth quarter of this year, the highly anticipatedThe commercial version of HarmonyOS NEXT Hongmeng Galaxy Edition will be officially launched on the market to meet with Rwanda Sugar Daddy consumers. 's avatar Issued on 04-12 14:58 •778 views
Rwanda Sugar Daddy The value behind learning Hongmeng? How to learn the open version of Galaxy? It is now 2024, and Huawei launched the Hongmeng Thousand Sails Ceremony on January 18. The Hongmeng Galaxy version was released and applications were open to developers. This conference mainly demonstrated that Hongmeng is already a full-stack self-developed base, and the programming language of the Hongmeng Galaxy version was changed to ArkTS/ Published on 02-22 20:55
How about the preview version of Hongmeng Developer? , native intelligence, native interconnection, 6 ultimate native experiences. After a preliminary browse of Hongmeng’s source code, I can basically conclude that it is a completely self-developed base, and Android is no longer visible. As a developer, if you want to develop in the direction of Hongmeng, you need to learn systematically. The knowledge system of Hongmeng is as follows. Published on 02-17 21:54
Hongmeng system has advantages and disadvantages. Can it be chosen as a developer? Galaxy version is once a pure-blood Hongmeng, but its development has some cycles. It is difficult to establish an ecosystem, and it will take time for major manufacturers to participate in Hongmeng native development. Hongmeng opens up a talent gap, because Hongmeng is a new system, and programmers start from scratch. Therefore, there are very few Hongmeng released on the market on 02-16 21:00
huawei announced that the developer preview of HarmonyOS NEXT Hongmeng Galaxy Edition is open to developers and is open to developers. huawei announced that the developer preview of HarmonyOS NEXT Hongmeng Galaxy Edition is open to developers. This means that Hongmeng Ecosystem has entered the second stage and will accelerate the application of Hongmeng in various industries. 's avatar Published on 01-29 16:42 •1251 views
HarmonyOS SDK helps developers create a whole new life cycle of Hongmeng native applications, deployment and maintenance . HarmonyOS SDK provides six major areas based on the functions and features required by developers: The application framework area provides basic facilities for mobile location application development including ArkTS, ArkUI, etc., assisted by published on 01-19 10:31
The scope exceeded 100 million! Wen Xinyiyan revealed its latest user scale, “Wenxinyiyan’s user scale exceeded 100 million.” December 28On the 10th, Wang Haifeng, chief technology officer of Baidu and director of the National Engineering Research Center for Deep Learning Technology and Application, presented at the 10th WAVE SUMMIT Deep Learning Development Announced at the conference. At the meeting, Wang Haifeng presented “Wen's avatar Published on 12-29 15:58 •485 views
Wen Xinyiyan’s user scale exceeded 100 million and its usage exceeded 4,000 At the 10th WAVE SUMMIT Deep Learning Developers Conference, Baidu Chief Technology Officer Wang Haifeng revealed a set of data. The user scale of Qian Wenxinyiyan has exceeded 100 million and the number of applications has exceeded 4,000. Since August 31, it has been allowed to open to the public to provide services. Since then, the article's avatar was published on 12-28 18:18 •972 views
[Developer Welfare MonthRW Escorts] Listen to the stories of the pioneers and enjoy holiday benefits. Pay tribute to the pioneers of Rwandans Sugardaddy in the name of 1024 Thank you to the developers who have used your keyboard and screen to build amazing digital worlds one after another. Happy holidays! Hear Your Story Behind the calm appearance of the developers, there is a passionate heart. In the previous article 's avatar was published on 10-24 15:45 • 267 views
The long-awaited open source Hongmeng developer phone is now available for purchase! ! The long-awaited open source Hongmeng developer mobile phone Rwanda Sugar is now available for purchase! ! “Open Source Harmony Developer Phone” is essentially a mobile phone-shaped development board for the majority of OpenHarmony developers. Published on 10-10 18:32


留言

發佈留言

發佈留言必須填寫的電子郵件地址不會公開。 必填欄位標示為 *