The Julia community is small and has no large commercial backers. Projects such as TF/PyTorch require community support and a lot of investment which Julia just doesn't have. In fact, Julia isn't even trying at the moment to "compete" with TF/PyTorch [1, 2].

[1] https://discourse.julialang.org/t/state-of-machine-learning-... [2] https://news.ycombinator.com/item?id=29682507

I think this is the best answer, as someone who's been using the language for most of my work for a few years (but not in ML).

Flux.jl is probably the highest-profile relevant effort, but it's been (AFAIU) pretty much entirely volunteer-developed for the past 2+ years.

I've worked at 2 companies that would have liked to use Julia but it wasn't (and still isn't) product ready for anything involving high reliability or robustness.

Typical ML doesn't require an entirely differentiable language.

For applications that do need that (mainly "scientific machine learning") Julia has taken a considerable chunk of the market.

To be fair Pytorch is more convenient if you're not doing anything unusual. And Julias more flexible AD comes with some sharp edges.