container2wasm VS Tribuo

The existence of vscode.dev always makes me wonder why Microsoft never released an iOS version of VSCode to get more users into its ecosystem. Sure, it's almost as locked down as the web environment, but there's a lot of space in that "almost" - you could do all sorts of things like let users run their code, or complex extensions, in containers in a web view using https://github.com/ktock/container2wasm or similar.

Unfortunately cosmopolitan wouldn't work for dockerc. Cosmopolitan works as long as you only use it but container runtimes require additional features. Also containers contain arbitrary executables so not sure how that would work either...

As for WASM, this is already possible using container2wasm[0] and wasmer[1]'s ability to generate static binaries.

[0]: https://github.com/ktock/container2wasm

[1]: https://wasmer.io/

Really impressed by the depth and breadth of this project, well done!

A particularly interesting part is the socket layer inside the browser. Other people solving this problem have previously used a proxy to a server that does the real socket implementation. This means you can't have a "browser-only" solution.

The author has solved this (for HTTP/S only) by proxying HTTP requests and then re-creating them as fetch requests (details here: https://github.com/ktock/container2wasm/tree/main/examples/n...). I'm very interested in using this approach for my own project Runno (https://runno.dev).

It seems to have been picked up by Eclipse and there is also Oracle Labs' Tribuo and Deep Java Library. All seem active, but I don't know much about any of them. I agree it's probably best to follow the community and use a more popular tool like PyTorch.

We built model & data provenance into our open source ML library, though it's admittedly not the W3C PROV standard. There were a few gaps in it until we built an automated reproducibility system on top of it, but now it's pretty solid for all the algorithms we implement. Unfortunately some of the things we wrap (notably TensorFlow) aren't reproducible enough due to some unfixed bugs. There's an overview of the provenance system in this reprise of the JavaOne talk I gave here https://www.youtube.com/watch?v=GXOMjq2OS_c. The library is on GitHub - https://github.com/oracle/tribuo.

Although it may be a bit more work, you can do both machine learning and AI in Java. If you are doing deep learning, you can use DeepJavaLibrary (I do work on this one at Amazon). If you are looking for other ML algorithms, I have seen Smile, Tribuo, or some around Spark.

We've been developing Tribuo on Github for two years now, MS are very actively developing ONNX Runtime (and the Java layer is fairly thin and wrapped over the same C API they use for node.js and C#), and things like XGBoost and LibSVM have been around for many years and the Java bits are developed in tree with the rest of the code so updated along with it. Amazon have a team of people working on DJL, though you'd have to ask them what their plans are.

FWIW, Oracle actually did release a Java ML library - https://github.com/oracle/tribuo.

Tribuo (tribuo.org, github.com/oracle/tribuo). ONNX export support is there for 2 models at the moment in main, there's a PR for factorization machines which supports ONNX export, and we plan to add another couple of models and maybe ensembles before the upcoming release. Plus I need to write a tutorial on how it all works, but you can check the tests in the meantime.

You can do ML & data science in Java (full disclosure: I help run TensorFlow-Java, I maintain ONNX Runtime's Java interface, and I'm the lead developer on Oracle Labs' Java ML library Tribuo, so I'm pretty biased). It tends not to be as favoured in research, though I've published academic ML papers which used Java implementations. People do deploy ML models quite a bit in Java in industry.

It might be worth having a look at the ONNX Runtime Java API in addition to TF-Java, it'll let you deploy the rest of the HuggingFace pytorch models that don't have TF equivalents. I built the Java API a few years ago, and it's now a supported part of the ONNX Runtime project. We use it in Tribuo to provide one of our text feature embedding classes (BERTFeatureExtractor).

The IJava notebook kernel works pretty well for data science on top of Java. We use it in Tribuo to write all our tutorials, and if you've got the jar file in the right folder everything is runnable. For example, this is our intro classification tutorial - https://github.com/oracle/tribuo/blob/main/tutorials/irises-tribuo-v4.ipynb.