tensorflow

Recently, I was working on various open source projects. Mostly I was working on issues labelled as "good first issues". Before I start working on a issue I usually take a look at the previous attempts made by other people. Usually I can find a couple of PRs or comments if the issue is older than one year. Looking at the work history sometimes it becomes obvious why previous attempts were not successful: someone completely misunderstood the task and went the wrong direction, someone can barely code, someone is just pinging the contributors instead of doing the actual work[1]. What was surprising to me was that the maintainers almost always found time to answer every question, review ridiculously bad and incorrect PRs and spend days communicating with the contributor (including me) to give feedback about the work, while they themselves cou

You guys using convoluted git commands when a single line of subversion works: svn checkout https://github.com/tensorflow/tensorflow/trunk/tensorflow/ex...

The project looks fairly active, based on the commit history: https://github.com/tensorflow/tensorflow/commits/master/

Yes, production code is C++, but the way the models are built is Python code, colab, save via Orbax/TF SavedModel, etc and then serve in C++. All (most, to hedge it) development is done via Python. I don't directly develop the models, I played with some for fuzzing [1] and I'm working on security for them [2]. And, before joining GOSST, I was leading the OSS DevInfra team in TF. I still have the most number of commits made by a human [3] even after 2 days of leaving the team, though I see the next person only needs 4 more :) [1]: https://security.googleblog.com/2023/08/ai-powered-fuzzing-b... [2]: https://security.googleblog.com/2023/10/increasing-transpare... [3]: https://github.com/tensorflow/tensorflow/graphs/contributors

Oh interesting, I thought this was going to be a .gitignore file we could use in all projects :D Do you have any thoughts on rather than listing all the file extensions you don't want to track: - https://github.com/tensorflow/tensorflow/blob/master/.gitign... - https://github.com/redis/redis/blob/unstable/.gitignore - https://github.com/elastic/elasticsearch/blob/main/.gitignor... You'd list the file extensions you want to track? Here is an example for a PHP project: * !/ !.php !.md !.xml !.gitignore !composer.json Or Clojure: * !/ !.clj !*.md target/ !.gitignore If you have exceptions on these rules, you can always force-add the files or add the extension to .gitignore, to configure it not being ignored.

I think just native https://pkg.go.dev/github.com/tensorflow/tensorflow/tensorfl... but tfgo looks interesting. Actually the docs around this weren’t great. Took the train-in-python & inference-in-go approach. And only for versions greater than tf2

Right but that's not an inherent GPU determinism issue. It's a software issue. https://github.com/tensorflow/tensorflow/issues/3103#issueco... is correct that it's not necessary, it's a choice. Your line of reasoning appears to be "GPUs are inherently non-deterministic don't be quick to judge someone's code" which as far as I can tell is dead wrong. Admittedly there are some cases and instructions that may result in non-determinism but they are inherently necessary. The author should thinking carefully before introducing non-determinism. There are many scenarios where it is irrelevant, but ultimately the issue we are discussing here isn't the GPU's fault.

This is indeed the bread-and-butter, but there is use of all sorts of standard linear algebra algorithms. You can check various xla-related (accelerated linear algebra) folders in tensorflow or torch folders in pytorch to see the list of what is used [1],[2] [1] https://github.com/tensorflow/tensorflow/tree/8d9b35f442045b... [2] https://github.com/pytorch/pytorch/blob/6e3e3dd477e0fb9768ee...

you can't store a function as a tensor. the tensors are the inputs/outputs that flow along the edges of the graph. TF stores things directly as a graph: https://github.com/tensorflow/tensorflow/blob/master/tensorf... while Jax represents computations in an intermediate representation: https://jax.readthedocs.io/en/latest/jaxpr.html

I never understood why people have extensive lists of file extensions in their .gitignore files, for example `.DS_Store, .so, .dll, .pyc, .pyd, *.bak, ...`: - https://github.com/tensorflow/tensorflow/blob/a7691d53e5d3a4... - https://github.com/pytorch/pytorch/blob/02179827cb7cbc805277... I find it much easier to list the file extensions I wish to track: - https://github.com/nikonyrh/docker-scripts/blob/4aa3dc38ab59... - https://github.com/nikonyrh/rubikstimer-clj/blob/3e77b9ad88f...

If you mean StableHLO, then it has an MLIR dialect: https://github.com/openxla/stablehlo/blob/main/stablehlo/dia.... In the StableHLO spec, we are talking about this in more abstract terms - "StableHLO opset" - to be able to unambiguously reason about the semantics of StableHLO programs. However, in practice the StableHLO dialect is the primary implementation of the opset at the moment. I wrote "primary implementation" because e.g. there is also ongoing work on adding StableHLO support to the TFLite flatbuffer schema: https://github.com/tensorflow/tensorflow/blob/master/tensorf.... Having an abstract notion of the StableHLO opset enables us to have a source of truth that all the implementations correspond to.