←back to thread

ML needs a new programming language – Interview with Chris Lattner

(signalsandthreads.com)
311 points melodyogonna | 4 comments | 05 Sep 25 11:33 UTC | HN request time: 0.02s | source
Show context
MontyCarloHall ◴[05 Sep 25 14:19 UTC] No.45138920[source]▶
The reason why Python dominates is that modern ML applications don't exist in a vacuum. They aren't the standalone C/FORTRAN/MATLAB scripts of yore that load in some simple, homogeneous data, crunch some numbers, and spit out a single result. Rather, they are complex applications with functionality extending far beyond the number crunching, which requires a robust preexisting software ecosystem.

For example, a modern ML application might need an ETL pipeline to load and harmonize data of various types (text, images, video, etc., all in different formats) from various sources (local filesystem, cloud storage, HTTP, etc.) The actual computation then must leverage many different high-level functionalities, e.g. signal/image processing, optimization, statistics, etc. All of this computation might be too big for one machine, and so the application must dispatch jobs to a compute cluster or cloud. Finally, the end results might require sophisticated visualization and organization, with a GUI and database.

There is no single language with a rich enough ecosystem that can provide literally all of the aforementioned functionality besides Python. Python's numerical computing libraries (NumPy/PyTorch/JAX etc.) all call out to C/C++/FORTRAN under the hood and are thus extremely high-performance, and for functionality they don't implement, Python's C/C++ FFIs (e.g. Python.h, NumPy C integration, PyTorch/Boost C++ integration) are not perfect, but are good enough that implementing the performance-critical portions of code in C/C++ is much easier compared to re-implementing entire ecosystems of packages in another language like Julia.

replies(8): >>45139364 #>>45140601 #>>45141802 #>>45143317 #>>45144664 #>>45146179 #>>45146608 #>>45146905 #
Hizonner ◴[05 Sep 25 15:00 UTC] No.45139364[source]▶
This guy is worried about GPU kernels, which are never, ever written in Python. As you point out, Python is a glue language for ML.

> There is no single language with a rich enough ecosystem that can provide literally all of the aforementioned functionality besides Python.

That may be true, but some of us are still bitter that all that grew up around an at-least-averagely-annoying language rather than something nicer.

replies(5): >>45139454 #>>45140625 #>>45141909 #>>45142782 #>>45147478 #
1. MontyCarloHall ◴[05 Sep 25 15:08 UTC] No.45139454[source]▶
>This guy is worried about GPU kernels

Then the title should be "why GPU kernel programming needs a new programming language." I can get behind that; I've written CUDA C and it was not fun (though this was over a decade ago and things may have since improved, not to mention that the code I wrote then could today be replaced by a couple lines of PyTorch). That said, GPU kernel programming is fairly niche: for the vast majority of ML applications, the high-level API functions in PyTorch/TensorFlow/JAX/etc. provide optimal GPU performance. It's pretty rare that one would need to implement custom kernels.

>which are never, ever written in Python.

Not true! Triton is a Python API for writing kernels, which are JIT compiled.

replies(3): >>45140116 #>>45143238 #>>45147017 #
2. catgary ◴[05 Sep 25 16:04 UTC] No.45140116[source]▶
I agree with you that writing kernels isn’t necessarily the most important thing for most ML devs. I think an MLIR-first workflow with robust support for the StableHLO and LinAlg dialects is the best path forward for ML/array programming, so on one hand I do applaud what Mojo is doing.

But I’m much more interested in how MLIR opens the door to “JAX in <x>”. I think Julia is moving in that direction with Reactant.jl, and I think there’s a Rust project doing something similar (I think burn.dev may be using ONNX has an even higher-level IR). In my ideal world, I would be able to write an ML model and training loop in some highly verified language and call it from Python/Rust for training.

3. pama ◴[05 Sep 25 20:27 UTC] No.45143238[source]▶
Although they are incredibly useful, the ML applications, or MLops, or ML Devops, or whatever other production-related application tech stack terminology may come to mind, are providing critical scaffolding infrastructure and glue but are not strictly what comes to mind when you only use the term “Machine Learning”. The key to machine learning is the massively parallel ability to efficiently train large neural network models, and the key to using the benefits of these trained models is the ability to rapidly evaluate them. Yes you need to get data for the training and you need complex infrastructure for the applications but the conferences, papers, studies on machine learning dont refer to these (other than in passing), in part because they are solvable, and in part because they are largely orthogonal to ML. Another way to think about it is that the nvidia price is what goes to infinity in this recent run and not the database or disk drive providers. So if someone finds a way to make the core ML part better with a programming language solution that is certainly very welcome and the title is appropriate. (The fact that GPU programming is considered niche in the current state of ML is a strong argument for keeping the title as is.)
4. seanmcdirmid ◴[06 Sep 25 06:08 UTC] No.45147017[source]▶
There are a lot of tricks for writing GPU code in a high level language and using some sort of meta programming to make it work out (I think Conal Elliott first did this in Haskell, where he also does symbolic differentiation in the same paper!).