AMD is working hard on the opensource AMDGPU Linux drivers, they have been working well for me. Unfortunately, AMD doesn't have parity with Nvidia's CUDA, there are nascent efforts to compete with ROCm/OpenCL, but CUDA is quite far ahead for ML applications. Sadly, upstream TF relies on CUDA and isn't showing any signs of supporting AMD GPUs.
I'm hoping the "No data center usage" clause Nvidia recently sprung in its license will encourage more people to develop for, or expend more effort on porting ML libraries and tools to AMD GPUs.
Also, AMD graphics card support for graphics applications is generally excellent, especially in the last few months with newer cards. AMD have released a fully open-source driver chain called AMDGPU, the only caveat being that it uses an open-source LLVM-based shader compiler toolchain that's somewhat slower on some workloads (and somewhat faster on others!) than whatever proprietary thing is in their blob drivers, which are called AMDGPU PRO.
Now on graphics cards support for Machine Learning applications AMD falls short of Nvidia, but that's a different processor.