Whereas working at Apple, Chris Lattner designed Swift to “totally leverage the ability of LLVM,” and “led a group for some time at Google to attempt to transfer Swift out of its Apple consolation zone, to turn out to be a substitute for Python in AI mannequin improvement.” That is in accordance with a weblog submit by Jeremy Howard, an advisor to Lattner’s Modular AI (which he co-founded in 2022 to construct a next-generation AI platform for builders).
“However sadly,” Howard writes, Swift “didn’t obtain the help it wanted from both Apple or from Google, and it was not in the end profitable.” And but…
[W]hilst at Google Chris did develop one other undertaking which grew to become vastly profitable: MLIR. MLIR is a substitute for LLVM’s intermediate illustration [or IR] for the trendy age of many-core computing and AI workloads. It is vital for totally leveraging the ability of {hardware} like GPUs, TPUs, and the vector models more and more being added to server-class CPUs.
So, if Swift was “syntax sugar for LLVM”, what’s “syntax sugar for MLIR”? The reply is: Mojo! Mojo is a model new language that is designed to take full benefit of MLIR. And in addition Mojo is Python.
Wait what?
OK let me clarify. Possibly it is higher to say Mojo is Python++. Will probably be (when full) a strict superset of the Python language. But it surely additionally has extra performance so we will write excessive efficiency code that takes benefit of recent accelerators…
Whereas Swift was a model new language packing every kind of cool options based mostly on newest analysis in programming language design, Mojo is, at its coronary heart, simply Python. This appears clever, not simply because Python is already nicely understood by thousands and thousands of coders, but in addition as a result of after many years of use its capabilities and limitations are actually nicely understood. Counting on the most recent programming language analysis is fairly cool, however its potentially-dangerous hypothesis since you by no means actually know the way issues will prove…
A key trick in Mojo is which you could choose in at any time to a quicker “mode” as a developer, by utilizing “fn” as an alternative of “def” to create your operate. On this mode, you must declare precisely what the kind of each variable is, and in consequence Mojo can create optimised machine code to implement your operate. Moreover, if you happen to use “struct” as an alternative of “class”, your attributes can be tightly packed into reminiscence, such that they will even be utilized in knowledge constructions with out chasing pointers round. These are the sorts of options that permit languages like C to be so quick, and now they’re accessible to Python programmers too — simply by studying a tiny bit of latest syntax…
I am unable to start to explain all of the little (and large!) concepts all through Mojo’s design and implementation — it is the results of Chris and his group’s many years of labor on compiler and language design and contains all of the tips and hard-won expertise from that point — however what I can describe is a tremendous consequence that I noticed with my very own eyes.
Mojo hasn’t been launched to the general public but, (apart from a web based “playground” with a waitlist the place they’re “rolling out entry slowly.”) However the weblog submit notes that making a programming language’s syntax is often complicated, error-prone, and controversial — an issue Mojo neatly avoids by “outsourcing” its syntax to an present language, “which additionally occurs to be the most generally used language in the present day.”
And “As a compiled language, Mojo’s deployment story is principally the identical as C,” the submit argues. [That is, “you can literally just make the compiled program available for direct download. It can be just 100k or so in size, and will launch and run quickly.”]
“Because of this Mojo is way more than a language for AI/ML functions. It is truly a model of Python that permits us to put in writing quick, small, easily-deployed functions that reap the benefits of all obtainable cores and accelerators!”
