>What does this say about Google's commitment to developer excellence?
Look inside the tensorflow code base for your answer.
I had the Kafkaesque experience of reporting a bug, being told there is no bug by a help desk employee, while the bug was throwing up errors in the official docs.
To top it off I got a message by one of the onshore team months later that they were going to solve it only for the person to be fired within a week.
I've mostly moved to jax for daily experiments. Hopefully the fact that codebase is small and modular will mean that when Google Googles all over the project there will be enough know how to maintain a community fork.
The user love/passion on the JAX team is super high. Interacting with them is whatever the opposite of Kafkaesque is.
Well, right up until they get suddenly laid off.
Hasn't Google given up on tensorflow for JAX now?
It looks like tensorflow is going down the slow legacy/sunset trajectory at this point.
Have they given up on tensorflow.js for the browser as well? Is there any replacement?
In the instances where I’ve needed to run inference in the browser I have used Onnx runtime web to run weights that were exported from PyTorch training. You have to convert your browser data to onnx tensors, but it’s not that bad.
I don’t do actual training in the browser though, so maybe someone else can answer that part.
Is Transformers.js a replacement?