syntaxdot VS tensorflow

I am so happy about them releasing this. A few years ago I wrote a multi-task syntax annotator in Rust using Laurent Mazare's excellent tch-rs binding (it seems like he is also working on Candle):

https://github.com/tensordot/syntaxdot

However, the deployment story was always quite difficult. The PyTorch C++ API is not stable, so a particular version of tch-rs will only work with a particular PyTorch version. So, anyone wanting to use SyntaxDot always had to get exactly the right version of libtorch (and set some environment variables) to build the project.

The idea of making an abstraction over Torch and Rust ndarray (similar to Burn) crossed my mind several times, but there is only so much that I could do as a solo developer. So Candle would be a god-given if I was still working on this project.

Seeing Candle wants to make me port curated-transformers to Candle for fun:

https://github.com/explosion/curated-transformers

They are obviously not as good as in Python, but if you are willing to invest time, it's definitely doable. E.g. I made a multi-task transformer-based syntax annotator in Rust using the tch Torch binding:

https://github.com/tensordot/syntaxdot

In my current job, I do NLP with Python, Cython, and some C++. I don't think doing it in Rust was much more work. Once you are beyond the stage of implementing a small research project or toy model, most systems are going to contain a lot of custom, specialized code. You will have to do that work in any language.

What I like about PyTorch is that most of the functionality is actually available through the C++ API as well, which has 'beta API stability' as they call it. So, there are good bindings for some other languages as well. E.g., I have been using the Rust bindings in a larger project [1], and they have been awesome. A precursor to the project was implemented using Tensorflow, which was a world of pain.

Even things like mixed-precision training are fairly easy to do through the API.

[1] https://github.com/tensordot/syntaxdot

Huggingface fills the need for task based prediction when you have a GPU.

With model distillation, it should be possible to annotate hundreds of sentences per second on a single CPU with a library like Huggingface Transformers.

For instance, one of my distilled Dutch multi-task syntax models (UD POS, language-specific POS, lemmatization, morphology, dependency parsing) annotates 316 sentences per second with 4 threads on a Ryzen 3700X. This distilled model has virtually no loss in accuracy, compared to the finetuned XLM-RoBERTa base model.

I don't use Huggingface Transformers, but ported some of their implementations to Rust [1], but that should not make a big difference since all the heavy lifting happens in C++ in libtorch anyway.

tl;dr: it is not true that tranformers are only useful for GPU prediction. You can get high CPU prediction speeds with some tricks (distillation, length-based bucketing in batches, etc.).

[1] https://github.com/tensordot/syntaxdot/tree/main/syntaxdot-t...

# L2-normalize the encoding tensors image_encoding = tf.math.l2_normalize(image_encoding, axis=1) audio_encoding = tf.math.l2_normalize(audio_encoding, axis=1) # Find euclidean distance between image_encoding and audio_encoding # Essentially trying to detect if the face is saying the audio # Will return nan without the 1e-12 offset due to https://github.com/tensorflow/tensorflow/issues/12071 d = tf.norm((image_encoding - audio_encoding) + 1e-12, ord='euclidean', axis=1, keepdims=True) discriminator = keras.Model(inputs=[image_input, audio_input], outputs=[d], name="discriminator")

[3]: https://github.com/tensorflow/tensorflow/graphs/contributors

To get up to speed with TensorFlow, check their quickstart Support TensorFlow on GitHub ⭐

Explore here.

Part of the major userbase pull in GitHub revolves around hosting a considerable number of popular projects including Angular, React, Kubernetes, cpython, Ruby, tensorflow, and well even the software that powers this site Forem.

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.

and things like y = layers.ELU()(y) work as expected. I wanted to see a list of the available layers so I went to the Tensorflow GitHub repository and to the keras directory. There's a warning in that directory that says:

Exactly what language do you think TensorFlow is written in? :)