51CTO homepage
AI. x Community
Blog
school
Quality class
Live training camp
Enterprise training
Hongmeng Developer Community
WOT Technical Conference
AIGC Innovation China Tour
IT Certificate
Public number matrix
Mobile terminal
Register/Login
Short video Free courses Course ranking Live class Soft test school
All courses Manufacturer certification IT technology 2024 Soft test PMP project management Soft test information
Online learning
article resources Q&A Classroom special column live broadcast
51CTO
Hongmeng Developer Community
51CTO technology stack
51CTO official micro blog
51CTO School
51CTO Blog
CTO Training Camp
Hongmeng developer community subscription number
51CTO question bank applet
51CTO School APP
51CTO School Enterprise APP
Hongmeng developer community video number
Account Settings sign out
recommend Tencent Yuanbao comprehensive evaluation! The era of domestic AI "watching drama" starts! Read "Celebrate the Past Years" Fan Xian and reveal the secret of Altman Palace Battle! Get through Tencent's ecological "hope of the whole village"!
original
AI tencent Celebrate the remaining years
Is it more appropriate to deploy the large model "partition" in the cloud and edge?
translation Deploying large language models (LLMs) partitions of generative AI on edge and center/cloud infrastructure is the next frontier of AI deployment. This hybrid approach improves performance and response speed, and optimizes resource usage and security. However, most enterprises and even technology providers are afraid of this architecture, believing that it is too complex, too expensive, and too slow to build and deploy.
Discussion on Safety Management of Life Cycle of Medical Devices
original This article will deeply discuss various excellent practices of safety management from the life cycle of medical devices and the roles and responsibilities involved.
Develop arXiv paper engine robot program: based on RAG+LangChain+Chainlit+ChromaDB
translation This article will introduce how to use RAG technology, LangChain, Chainlit Copilot components and Literal AI observability features to jointly develop a semantic paper search engine program.
7B？ 13B？ 175B？ Interpreting the parameters of the large model
original In engineering practice, it is meaningful to understand the parameters of the large model. Parameters play a decisive role in the large model. They define the behavior, performance, implementation cost and resource requirements of the large model. To understand the parameters of a large model in engineering is to grasp the relationship between the complexity, performance and capability of the model.
Old Huang decrypts the third-generation GPU at one go! Shattering Moore's Law, Building AI Empire, and Mass Production of Blackwell to Solve ChatGPT's Global Power Consumption Problem
Reprint Just now, Old Huang showed another high-profile show to the world: Blackwell, which has been mass produced, will cut the training energy consumption of 1.8 trillion parameter GPT-4 to 1/350 in 8 years; Nvidia's amazing product iteration directly breaks Moore's Law; Blackwell's roadmap for the next three generations has also been released.
[Workplace Acceleration Package]: How can IT technicians create excellent resumes? Unlock interview skills and precise career planning
school [Original -- no plagiarism, 2 months of careful polishing] Master resume writing skills, easily respond to interviews, and share the experience of architects with an annual salary of one million

live broadcast

more
playback
playback

Express Lane

topic of conversation
#AIGC #ChatGPT #Internet of Things #Cloud computing #Move #Windows #Linux #Security #Database #Open source #Operation and maintenance #Artificial intelligence
special
Enterprise Dynamics Hongmeng Community video translation Community Editor association Volcanic engine IBM original HarmonyOS Thematic summary Journalism
Selected promotion

Quality class

Live training camp Huawei certification network security Red Hat Linux OCP database Oracle 2024 Soft test PMP project management
recommend development artificial intelligence cloud computing big data network security system data base Internet of Things Mobile Development Blockchain Digital transformation
Big personnel adjustment of byte game; A team from Stanford has built a large model of China, which has been deleted from the database; Yu Chengdong of Huawei: not counting the price; Samsung Electronics employees strike for the first time; Embrace and expose security loopholes original
Edgate is a british software company that provides database development tools and solutions. The email said that Redgate would not make major changes to the DB Engines website for the time being, and would continue to let it operate independently - including the database ranking report, advertising display and other businesses as usual.

2024-06-03 13:21:37

Byte game Model Delete database
Nvidia's new research: the context length is heavily marked with false signs, and there are few 32K qualified
The new benchmark, named RULER, contains 13 tasks in four categories: retrieval, multi hop tracking, aggregation and question answering. RULER defines the "effective context length", that is, the maximum length at which the model can maintain the same performance as the Llama-7B baseline at 4K.

2024-06-03 14:24:00

Yingweida Model
Stanford exploded that Llama3-V copied domestic open source projects, and the author quickly deleted the database
Stanford Llama3-V vs Tsinghua MiniCPM-Llama3-V-2.5

2024-06-03 14:19:00

AI train
Load balancing and fault tolerance: the application of cluster mode in distributed systems
Cluster mode is a widely used architecture mode in distributed systems. It provides services through multiple nodes, achieving high availability, load balancing and scalability.

2024-06-03 14:17:00

Cluster mode Distributed system
Ten use cases of generative AI in enterprises
In the dynamic AI world, generative AI develops as a transformative constraint, reshapes the enterprise, and reclassifies the possibility of development.

2024-06-03 13:33:58

artificial intelligence Generative artificial intelligence
Beyond Devin! Yao Ban led the team, they set a new world record for big model programming
Benchmark itself is just a figure. With the launch of GPT4-o, the test results of SWEBench are expected to exceed 30% soon, and it is optimistic that it will exceed 50% next year.

2024-06-03 12:03:56

Model programming
I found the secret of writing concise code in Python!
Writing concise code is critical to the success of a software project, but it requires rigorous attitude and continuous practice. As data scientists, we tend to focus more on running code in Jupyter Notebooks, finding good models and obtaining ideal indicators, while ignoring the cleanliness of the code. However, writing concise code is also a compulsory course for data scientists, because it can ensure that models are put into production faster.

2024-06-03 11:43:55

Python code Jupyter
Write concise and efficient Python formatted output code through f-string
The f-string introduced in Python 3.6 is one of the most commonly used features of Python. It allows us to write cleaner, more efficient and easier to maintain code. Today, we will introduce some techniques to use it in detail from simple to deep.

2024-06-03 11:36:06

Python f-string
ICML2024 high score! Magic changes attention so that small models can play twice as large models
DCMHA removes the fixed binding of MHA's attention head's search and selection circuit and transformation circuit, allowing them to dynamically combine according to input, and fundamentally improving the expression ability of the model. It can be approximately understood that each layer originally had a fixed number of H attention heads. Now, with almost the same amount of parameters and computing power, as many as HxH attention heads can be dynamically combined as needed.

2024-06-03 10:56:53

DCMHA Small model artificial intelligence
OpenAI gave them the original version of GPT-4: study how far they can go without fine-tuning the words
It has always been the research direction of the industry to avoid fine-tuning alignment, so that the newly released pre training model can not only "complete the text", but only learn to talk with users and follow instructions from prompt words. The current SOTA method Urial comes from Allen Research Institute. It can achieve good results by using system prompts+a few style examples.

2024-06-03 10:54:04

OpenAI GPT-4 artificial intelligence
45 JavaScript super tips that every developer should know
JavaScript is a language with multiple application scenarios and powerful functions, which is crucial to modern Web development. The following are some super skills that will help you become a more efficient and effective JavaScript developer. Each skill has detailed explanations and examples.

2024-06-03 10:35:41

JavaScript Web development
Can QR code play like this? Create a 3D dynamic particle 2D code!
According to the implementation effect of particle cloud mentioned above, this paper expands how to implement the particle effect of a two-dimensional code; With the modeling tool Blender, you can create a model of our QR code and import it into three. js; It is suitable for the place where you need to show the effect of QR code on the web page.

2024-06-03 10:33:08

QR code Modeling tools Webpage
Python decorators generalize public and private attributes
The Python decorator provides programmers with flexible tools to dynamically modify the behavior of functions or classes, including controlling the access and setting behavior of public and private properties in a class. This paper discusses the application of decorators in this field.

2024-06-03 10:26:47

Python Decorator Object Programming
Evolution of Vipshop's microservice architecture
The top level aggregate API service is oriented to different business lines and provides a unified interface to facilitate front-end application calls. This design ensures the unity and consistency of user operations.

2024-06-03 10:19:05

Vipshop Microservice framework
Three "weird" slow queries of Redis caused by system and hardware failures
This article briefly analyzes the weird and slow query problems encountered in the use of Redis, including hardware failure related problems, system parameter error optimization and other problems. Of course, there may be other weird situations in the actual use and operation and maintenance process, which need specific analysis.

2024-06-03 10:10:01

Redis Hardware optimization
Count the common methods of adding elements to vector class and vector class vector, and the common methods of deleting element objects from vector class vector
The Vector class is used to realize the function of dynamic array, and its four construction methods are introduced. The common methods for adding elements to vector of Vector class are addElement (Object obj), adding an element at the end of the collection, adding its toString() return value regardless of its type, and insetElementAt (Object obj, int index), inserting obj elements at the specified index.

2024-06-03 10:07:22

Vector class element vector
So this is generative AI!
With the development of the Internet, the data scale expands rapidly, providing massive training data for artificial intelligence algorithms. However, due to the limited hardware foundation, the development is not rapid at this time.

2024-06-03 09:52:08

AI internet data
A record breaking competitor of Neuralink, 4096 electrodes were implanted into human brain in a minimally invasive way! Brain computer interface is expected to enter the "5G" era
Neuralink is not the only company that has made breakthroughs in the field of brain computer interface. Precision Neuroscience announced that it has successfully placed 4096 electrodes in the human brain using a different technology path than Neurolink, breaking the record of 2048 electrodes last year.

2024-06-03 09:13:18

Brain computer interface technology
The programmers in this place are too idle. They have written three popular operating systems in the world!
Andreas developed Quiet OS to escape drug addiction. Ville M. Turjanmaa wrote MenuetOS in assembly because of its high assembly performance. He found that the performance of C/C++written components in Linux can be improved by 10%~40% when rewritten by assembly, so he began to rewrite them.

2024-06-03 09:04:30

programmer OS C++
Confused team code style? Spotless will take care of it!
Spotless is such a tool. It can not only format the code automatically, but also ensure that the code always conforms to the predetermined standards.

2024-06-03 08:55:27

team code tool
Living brain cells are made into 16 core chips and can be programmed with Python. Nine institutions carry out experiments and 36 universities queue up
The power consumption of this biological processor is 1 million times lower than that of traditional digital processors.

2024-06-03 08:55:00

chip programming Cells
Spring Boot performance is too poor? Try these moves!
SupplyAsync is used to provide return values. CompleteFuture also has an asynchronous call method runAsync (Runnable runnable) that does not need return values. Generally, we use this method more often when optimizing the Controller.

2024-06-03 08:52:40

Spring asynchronous optimization
What benefits can 5G enable intelligent manufacturing bring?
5G's high-speed and reliable connection has brought revolutionary solutions to the manufacturing industry. It enables real-time data analysis, promotes flexible automation processes, and creates a more intelligent and efficient factory.

2024-06-03 08:52:29

5G
Decrypting Claude 3 brain for the first time! 25 year old Anthropic Chief of Staff predicted that he would be eliminated by AI within 3 years
Anthropic's 25 year old chief of staff said that his work would be replaced by AI in the next three years because he felt deeply AGI. In a recent article, she predicted the types of work to be eliminated in the future. Is it true that Claude 3 model has begun to show AGI?

2024-06-03 08:40:00

AI AGI work
Using Springboot 3. x to Realize the Recognition and Processing of Answering and Making Calls in the Examination System
We can effectively identify and deal with the examinees' phone calls in the examination system, and improve the fairness and effectiveness of the examination. If there is further demand, the solution can be further optimized and improved by combining machine learning and big data analysis technology. ​

2024-06-03 08:38:41

examination system distinguish
Serverless Computing Optimization Guide: Best Practices for Improving Enterprise Performance
From following the principle of single responsibility and adopting an event driven architecture to implementing effective monitoring and error handling strategies, a clear serverless approach is critical to building highly available, resilient and cost-effective scalable applications.

2024-06-03 08:37:26

Serverless computing cloud computing
accident! Old Huang revealed the next generation of GPU chip, and the new speech put Moore's Law on the ground
Nvidia has no competitors, but chooses to put Moore's Law on the ground for friction

2024-06-03 08:35:00

Yingweida chip
UrbanGPT, the first big model of smart city, is fully open source | HKU&Baidu
The joint team of HKU and Baidu has proposed a new spatiotemporal large-scale language model UbanGPT, drawing on the idea of large-scale language model.

2024-06-03 08:30:00

AI Language model
SpringBoot and local database store and retrieve face data
This article explains in detail how to use SpringBoot and local databases (such as MySQL) to store and retrieve face data. Through the building of the SpringBoot project, database configuration, face data storage and retrieval API implementation, combined with data security and privacy protection policies, we have provided a complete set of solutions for everyone. I hope you can learn from it and apply it to actual projects. ​

2024-06-03 08:29:20

Spring data base Face data
Several Methods of Implementing Delay Operation in Android Development
If you are processing an animation and want to perform some operations after the animation, you can use the listener of the Animator class to achieve an effect similar to delay.

2024-06-03 08:26:34

Android development monitor
An inventory of the progress of the open source model: how good are the latest Mixtral, Llama 3, Phi-3, and OpenELM?
In a blog published in May, he reviewed and analyzed four major new models released in April: Mixtral, Meta AI's Llama 3, Microsoft's Phi-3, and Apple's OpenELM.

2024-06-03 08:25:00

Model Open Source
Project highlight: analyze and solve the problem of stuck applet page switching
Page switching refers to switching from one page to another, repeatedly checking your own code, and not finding that you have written any very time-consuming operations, so you are very confused.

2024-06-03 08:22:33

WeChat applet Page switching Delete Positioning Method
Ilya, who doesn't want French fries, and Nvidia, who doesn't give GPUs, Hinton's latest interview: thousands of roads, safety first
One year after leaving Google, Hinton, the "Godfather of Artificial Intelligence", was interviewed. The semi retired Hinton told us about the wisdom accumulated from time and the days he spent with his apprentice Ilya.

2024-06-03 08:15:00

Google AI
Pinduoduo Interview: How does Netty solve the problem of package sticking?
Packet sticking refers to the phenomenon that multiple small data packets sent by the sender in succession are received by the receiver at one time in network communication. This may be because the underlying transport layer protocol (such as TCP) will combine multiple small packets into a large data block for transmission, resulting in the receiver receiving multiple packets at one time when receiving data, causing adhesion.

2024-06-03 08:09:46

Sticking problem Netty data transmission
Quickly learn an algorithm, xLSTM
The traditional LSTM model solves the problem of gradient disappearance and gradient explosion in long sequence data of standard RNN (Recurrent Neural Network) by introducing forgetting gates, input gates and output gates.

2024-06-03 08:09:39

xLSTM algorithm Model
Caching strategy of Internet giants: a secret weapon against ultra-high concurrency, which has been open source!
Caching is not only used to store hot data and improve the read performance of hot data, but also a powerful tool against high concurrency and large traffic in business systems.

2024-06-03 08:01:20

cache strategy High concurrency
Six front-line AI engineers summarize the explosion! One year's experience of big model application is open, netizen: full process high energy
This long article on dry goods became a hot topic in the developer community.

2024-06-03 08:00:00

AI development
What is LLM's "mother tongue"?
Is English used as the internal language of multilingual LLM trained on English based corpus? In this regard, researchers from EPFL conducted a series of experiments on the Llama 2 family.

2024-06-03 07:57:32

LLM Llama 2 token
Yann LeCun: ViT is slow and inefficient. Real time image processing depends on convolution
Recently, Yann LeCun, the Turing Prize winner and Meta's chief scientist who has been active on social media, also joined the discussion of the debate between ViT and CNN.

2024-06-03 07:55:00

computer vision AI
Can ChatGPT really remember your words? DeepMind and Open Source Boss Reveal the Mystery of LLM Memory
Does LLM have memory ability? Yes, neither. Although ChatGPT seems to be able to remember what you said before when chatting, in fact, models can't remember anything when reasoning, and their memory in training is not as simple as we think.

2024-06-03 07:45:00

LLM Model
AlphaFold 3 is not satisfied! The server is hacked, and scientists all over the world compete to crack it
The emergence of AlphaFold3 shocked the entire academic community again. However, Google DeepMind's "non open source" has caused dissatisfaction in the academic community. The AlphaFold server has been attacked by hackers, and the open source project has begun to work.

2024-06-03 07:30:00

AlphaFold3 The server
The multi-mode large model is not flexible enough. Google DeepMind innovative architecture Zipper: separate training and "compression"
Multimodal pre training method has strong performance advantages (for example, one model can understand multiple modes natively), but it also has disadvantages. For example, it is impossible to solve the problem of how to add a new mode after pre training, and it is also lack of flexibility, because adding another mode requires training a new model from scratch, and conducting a super parameter search to obtain the best training data mixing ratio between modes.

2024-06-03 06:49:53

Multimodal large model Zipper encoder
Timers in. NET: Types, Uses and Sample Codes
According to the application scenarios and requirements NET provides a variety of timers for developers to choose from. This article will introduce NET, and provide corresponding sample code.

2024-06-03 00:00:20

.NET timer
15 design patterns of Python microservice architecture
Today, we will unveil the mystery of these ten design patterns. You will learn how to use these patterns to solve common problems in microservices, so as to build a more flexible and reliable system.

2024-06-03 00:00:10

Microservice Python
In just a few years, why has Vite become so popular?
The Vite team is also constantly optimizing and improving Vite. These improvements start with vite node (the engine that provides power for Vitest and Nuxt Dev SSR), and have been extended to a comprehensive review of the Vite API.

2024-06-03 00:00:03

Vite SSR API
Five excellent ones NET open source free Redis client component library
Section 5 NET open source and free Redis client component libraries have their own characteristics. Developers can select suitable libraries for integration according to their project needs. These libraries provide rich functions and options, making the interaction with Redis more simple and efficient.

2024-06-03 00:00:02

Open Source Redis client
Encapsulate a tool with 30 lines of code to solve the multi concurrency problem of Promise
We need to make some changes to enable Promise.all to support concurrency control when executing promises. However, we should not change Promise.all, but fetchFn one by one.

2024-06-03 00:00:01

code tool promises
Get Things Interview: Will the table be locked when creating an index?
Before MySQL 5.6, tables were locked when creating indexes. However, after MySQL 5.6.7, because of the new Online DDL technology, indexes can be added together with the DML data operation languages INSERT, UPDATE, DELETE, SELECT and other commands. ​

2024-06-03 00:00:01

Indexes MySQL technology
Tips for significantly reducing the disk space occupied by OH code
A few tips to significantly reduce the disk space occupied by OH code Foreword With the evolution and update of the version, the amount of OpenHarmony code is increasing, which consumes disk space very much. With v4.1 Release version as

2024-05-31 20:02:49

OpenHarmony
[Open Source Hongmeng] Download the source code of OpenHarmony 4.1 Release
[toc] (Table of Contents) This article describes how to download the source code of OpenHarmony operating system version 4.1 Release. This method can also be used to download the latest development version of OpenHarmony

2024-04-27 23:28:39

OpenHarmony git python
Load more
24 hours Read the leaderboard
Challenges and methods of data processing and interpretation of vibrating wire acquisition instrument in geotechnical engineering monitoring
Challenges and methods of data processing and interpretation of vibrating string acquisition instrument in geotechnical engineering monitoring Geotechnical engineering monitoring is an important link to ensure engineering safety, and vibrating string acquisition instrument is a commonly used equipment in geotechnical engineering monitoring
Get Things Interview: Will the table be locked when creating an index?
Explore behind the enterprise and have no worries about information inquiry
four Tips for significantly reducing the disk space occupied by OH code
five Stanford exploded that Llama3-V copied domestic open source projects, and the author quickly deleted the database
six Ten use cases of generative AI in enterprises
seven Using Springboot 3. x to Realize the Recognition and Processing of Answering and Making Calls in the Examination System
eight What is LLM's "mother tongue"?
nine Project highlight: analyze and solve the problem of stuck applet page switching
ten Confused team code style? Spotless will take care of it!

One week selection

more

Hongmeng Selection

more

Selected Blogs

more
Enterprise blog
Meituan Technical Team Asynchronous community Baidu Geek said Zabbix China
business
Scanning
Online Service