Technology iterations develop evenly or even stagnate for a period of time, but will explode exponentially within a certain period of time.
The GPT 3.5 training behind ChatGPT is said to have cost several million dollars plus a few months, and the parameters are about 170 billion.
This is definitely too expensive for the vast majority of individuals or businesses.
However, Microsoft (MSFT) announced the open source Deep Speed Chat. From the announced training time and price, the last 175b, that is, a model with a scale of 175 billion parameters.
20 hours, more than 5,000 US dollars.
1. Brief introduction of Deep Speed Chat technology
Deep Speed Chat is developed based on Microsoft's Deep Speed deep learning optimization library. It has functions such as training and enhanced reasoning. It also uses RLHF (reinforced learning with artificial feedback mechanism) technology, which can increase the training speed by more than 15 times, but the cost is high. The magnitude is reduced. For example, a ChatGPT-like model with 13 billion parameters can be trained in only 1.25 hours.
To put it simply, through the "fool-like" operation provided by Deep Speed Chat, users can train ChatGPT-like large language models in the shortest time and at the most efficient cost, which marks the coming era of ChatGPT for everyone.
2. The era of ChatGPT for all people has come
The current large model is still general-purpose, and in the end, everyone will have their own GPT.
To make an inappropriate analogy, Microsoft's open source Deep Speed Chat is like the Internet back then. In order to increase the development speed and reduce the price of desktop computers, ordinary users can also own their own computers.
Only when there are more participating users, the entire industry ecology can grow and develop rapidly. Therefore, Microsoft's open source has played a vital role in promoting the development of the entire ChatGPT ecosystem, enabling everyone to have their own ChatGPT.
At present, ChatGPT is in the initial stage of development, and there are problems such as security risks, data privacy, and ethics. It is believed that with the increase of participating users, these problems will be effectively solved, thereby contributing to the development of the global economy.
3. Open source is not easy to come by
Open source software has the characteristics of openness, sharing, and freedom. It plays an increasingly important role in software development and is also an important part of the software supply chain. According to a Gartner survey, 99% of organizations use open source software in their IT systems.
So this is why it is said above that Microsoft's open source action will promote the development of the entire ChatGPT ecosystem!
Partners who are familiar with low-code may know that many low-code platforms do not adopt the source code delivery mechanism. This makes many developers feel dizzy, because there is no way to understand the underlying logic during development, and once special circumstances arise, it is difficult to solve them.
In terms of cost and benefit, source code delivery is definitely the best support for secondary development of software :
1. The source code delivery mechanism can get rid of the dependence of the original vendor:
Users no longer have to worry about secondary development in the later stage, including activities such as forms, processes, and interfaces;
2. Can independently develop some complex business logic:
Enterprises do not need to re-purchase when expanding business processes, greatly reducing costs and time.
3. Various business systems developed based on the low-code development platform are not limited, and software copyrights can be independently applied for:
Of course, even if there is no need for secondary development for the time being, from a psychological point of view, providing source code can also bring psychological security "comfort" to the enterprise, without being limited by people's worries.
As far as I know, the JNPF rapid development platform provides source code delivery. Relatively speaking, in the software market, it has high cost performance and excellent after-sales service, and it is on the rise. It supports mainstream databases and satisfies comprehensive capabilities such as rapid system development, flexible expansion, seamless integration, and high-performance applications; adopts the front-end and back-end separation mode, and front-end and back-end developers can work together to be responsible for different sections, which is trouble-free and convenient. If you are interested, you can try it.
Open source experience: https://www.jnpfsoft.com/?csdn
Based on the advanced engine-based software rapid development mode, it is equipped with visual function engines such as process engine, form engine, report engine, chart engine, interface engine, portal engine, and organizational user engine; over a hundred functional controls can be built by dragging and dropping, allowing Developers no longer reinvent the wheel and focus on more suitable business designs.
Four. Conclusion
This time Microsoft has generously made Deep Speed Chat open source, what kind of waves will domestic AI R&D manufacturers make, we will wait and see! Many researchers say: Research is dead. This is indeed not optimistic . In the end, the author still wants to call for domestic independent research and development to make something that truly belongs to us.