FStar VS onnx

I don't think something that specific exists. There are a very large number of formal methods tools, each with different specialties / domains.

For verification with proof assistants, [Software Foundations](https://softwarefoundations.cis.upenn.edu/) and [Concrete Semantics](http://concrete-semantics.org/) are both solid.

For verification via model checking, you can check out [Learn TLA+](https://learntla.com/), and the more theoretical [Specifying Systems](https://lamport.azurewebsites.net/tla/book-02-08-08.pdf).

For more theory, check out [Formal Reasoning About Programs](http://adam.chlipala.net/frap/).

And for general projects look at [F*](https://www.fstar-lang.org/) and [Dafny](https://dafny.org/).

Don't get me wrong, there are times when Microsoft got it right the first time that was technically far superior to their competitors. Windows IOCP was theoretically capable of doing C10K as far back in 1994-95 when there wasn't any hardware support yet and UNIX world was bickering over how to do asynchronous I/O. Years later POSIX came up with select which was a shoddy little shit in comparison. Linux caved in finally only as recently as 2019 and implemented io_uring. Microsoft research has contributed some very interesting things to computer science like Z3 SAT solver and in collaboration with INRIA made languages like F* and Low* for formal specification and verification. But all this dwarfs in comparison to all the harm they did.

Most of the proof assistants out there: Lean, Coq, Dafny, Isabelle, F*, Idris 2, and Agda. And the main concepts are dependent types, Homotopy Type Theory AKA HoTT, and Category Theory. Warning: HoTT and Category Theory are really dense, you're going to really need to research them.

F* is a dependently typed language that can be transpiled to idiomatic C via the KReMLin compiler. It’s very ML-ish to write and you can leave out some proofs. It also has the benefit of being used to write a formally verified TLS implementation that’s in wide use throughout industry.

As I said, dependent types attempt to solve this problem. F* is a language where you can express complex logic as a type. The catch is, these types are checked by an SMT solver. If the solver can satisfy the type checking, then great, and you move on. If it can’t, you have no idea why, and either have to guess or manually write the proof anyway. Contrast this with Standard ML which has a proof of the soundness of its type system.

So kind of like F*? https://www.fstar-lang.org/

F*

Nevermind that academia has come up with far safer ways to do a few things but social norms & inertia prevent their wider adoption (well okay, it also has a barrier to entry in the education required to use it but I don't think someone with the knowledge to meaningfully contribute to an OS kernel can be considered uneducated nor unable to learn).

Maybe that's a little off topic here, but if you like fixing bugs, i suspect you might also enjoy showing that there are no bugs at all. Check out languages like F* https://www.fstar-lang.org/ It's a proof-oriented programming language. You can use it to write code that has no bugs at all. And you once you're done, can convert F* to C or other languages.

Once your model has been trained and validated using Anomalib, the next step is to prepare it for real-time implementation. This is where ONNX (Open Neural Network Exchange) or OpenVINO (Open Visual Inference and Neural network Optimization) comes into play.

ONNX is an open format from the Linux Foundation to represent machine learning models. It is becoming extensively adopted by the Machine Learning community and is compatible with most of the machine learning frameworks like PyTorch, TensorFlow, etc. Converting a model between any of those formats and ONNX is really simple and can be done in most cases with a single command.

ONNX[0], model-as-protosbufs, continuing to gain adoption will hopefully solve this issue.

[0] https://github.com/onnx/onnx

Running inference for large models on CPU is not a new problem and fortunately there has been great development in many different optimization frameworks for speeding up matrix and tensor computations on CPU. We explored multiple optimization frameworks and methods to improve latency, namely TorchScript, BetterTransformer and ONNX.

ONNX is a format for representing machine learning models in a portable way. Additionally, ONNX models can be easily optimized and thus become smaller and faster.

I would say onnx.ai [0] provides more information about ONNX for those who aren’t working with ML/DL.

[0] https://onnx.ai

It's not clear why you thing this sub is appropriate for some third party system with a Python interface. Why don't you try their discussion group: https://github.com/onnx/onnx/discussions

This kind of indirection through standardisation is pretty common to make compatibility between different kinds of software components easier. Some other good examples are the LSP project from Microsoft and ONNX to represent machine learning models. The first provides a standard so that IDEs don't have to re-invent the weel for every programming language. The latter decouples training frameworks from inference frameworks. Going back to WSGI, you can find a pretty extensive rationale for the WSGI standard here if interested.