[tune] Tensorboard

[nrudakov]

Description

Ray[tune] is using tensorboardX as a dependency. TensorboardX requires protobuf<=3.20.1
The latest version of onnx (1.13.0) requires protobuf>=3.20.2
Therefore, it is impossible to use ray[tune] in the same environment with latest onnx.

Use case

Suggestion is to switch from tensorboardX to tensorboard to be able to use onnx and ray[tune] in the same environment.

[amogkam]

Thanks for raising this @nrudakov! Do you get an error when trying to use onnx with ray tune? If so, could you share the stack trace?

[nrudakov]

@amogkam
I get an error message from pip

When first installing ray[tune] and then onnx:
ERROR: pip's dependency resolver does not currently take into account all the packages that are installed. This behaviour is the source of the following dependency conflicts. tensorboardx 2.5.1 requires protobuf<=3.20.1,>=3.8.0, but you have protobuf 3.20.3 which is incompatible.

And when first installing onnx then ray[tune]
ERROR: pip's dependency resolver does not currently take into account all the packages that are installed. This behaviour is the source of the following dependency conflicts. onnx 1.13.0 requires protobuf<4,>=3.20.2, but you have protobuf 3.20.1 which is incompatible.

To avoid possible runtime errors I have to use older version of onnx (1.12.0).

[amogkam]

Got it yeah, pip will complain, but there might not be any actual errors during usage. Your script may still work despite this incompatibility.

Sometimes these requirement pins are not as strict in practice.

[nrudakov]

Pytorch-Lightning recently switched from tensorboard to tensorboardx Lightning-AI/pytorch-lightning#15728
So probably it makes sense to stay on tensorboardx. I'm closing this issue.