Hacker News Reader

Some people care to deliver ML-enhanced services without peeking over the shoulder of their users for every keypress. Client inference performance matters.

> I thought tensorflow was mostly for training models

You need tensorflow to actually use the models trained with tensorflow.

Why? You could export your weights and everything into any other framework, no?

Probably, but looking at that chain of comments I think the emphasis was that you need _something_ like tensorflow to do inference client-side using pre-trained models, not that you need _tensorflow_ specifically to operate on tensorflow models.

Tensorflow-JS is that framework that you export to from Tensorflow. Tensorflow-JS is not Tensorflow, it's just the same brand.

Wouldn't you need the topology as well?

I was curious of iPhone performance - looks like latest one will outperform my MacBook again...