Recently, another dark horse in the field of programming suddenly burst into the eyes of developers and was officially opened for download.
Its name is Mojo , and I believe many friends have seen it recently.
Mojo is a programming language for AI developers, with a syntax similar to Python.
According to the description on Mojo’s official website, it combines the ease of use of Python with the high performance of C language, unlocking the programmability of AI hardware and the scalability of AI models.
Mojo seems to be quite capable, but which company launched it?
After taking a look, I discovered that Mojo is a new programming language launched by the artificial intelligence company Modular.
Modular is a very young AI startup founded in 2022 by Chris Lattner and Tim Davis.
When it comes to these two founders, I believe some students also know about them. They are top experts in the industry. Among them, Chris Lattner is also known as the “Father of LLVM” and “Father of Swift”. He has led the construction of AI and core systems at many well-known technology giants such as Apple, Google, and Tesla.
Modular’s vision is very ambitious, and its goal is to reshape AI infrastructure from the bottom up.
Last year, Modular AI received US$30 million in financing. Just a few days ago, Modular once again announced that it had successfully raised US$100 million, which is a remarkable achievement for a start-up company that has just been born.
In addition, you can see from the investor list on the company’s official website that many well-known investment institutions in the AI field are participating.
The programming language Mojo has several obvious features.
1. First is the performance aspect.
Mojo takes full advantage of the features and capabilities of the hardware, including multi-cores, vector units and accelerator units, as well as advanced compilers and heterogeneous runtime mechanisms, to achieve performance comparable to C++ and CUDA without adding complexity.
In terms of parallelization, Mojo uses MLIR to enable Mojo developers to take full advantage of vectors, threads, and AI hardware units.
2. The second aspect is interoperability.
As we all know, today, the Python ecosystem is extremely prosperous, with countless functions, libraries, frameworks, models, tools, etc.
Mojo provides access to the entire Python ecosystem. For example, using Mojo, you can seamlessly integrate and mix libraries like Numpy and Matplotlib in your code.
3. The third aspect is scalability.
Scalability is also an advantage of Mojo. Mojo can upgrade existing operations in the user model so that developers can easily extend the user’s model using pre-processing, post-processing, custom replacement and other operations.
Mojo was initially released in early May this year, and has formed a basic scale and ecosystem within a few months of its launch.
Not long ago, Modular’s official website announced that Mojo is officially open for download , starting with Linux systems, and will gradually add support for Mac and Windows in subsequent iterative versions.
This also means developers can experiment and write their own Mojo code through the Mojo SDK.
Shortly after Mojo was officially announced as available for download, an open source author named Aydyn Tairov made a breakthrough attempt using Mojo.
This author has previously transplanted the popular llama2.c project implemented in pure C on GitHub to llama2.py based on Python.
This time Aydyn Tairov transplanted llama2.py to llama2.mojo. The result was very unexpected. After the transplant, the performance increased by nearly 250 times.
Even so, the author still thinks there’s some room for improvement.
Seeing Mojo’s performance, many netizens said that Python has encountered a strong opponent this time. Is it possible that Mojo will even replace Python in the future?
In response, company CEO Chris Lattner responded directly:
Mojo does not pose a threat to Python. On the contrary, it will help Python developers become more powerful. It’s not Python that needs to be worried, but C++.
At the end of the article, relevant pages are also attached. Interested friends can try it out.
- Mojo address: www.modular.com/mojo
- Mojo documentation: docs.modular.com/mojo
As for how this programming language will develop in the next AI era, we can wait and see.