Feature Request
Would be good to have a fully-implemented WandbLoggerSpec analogous to the tensorboard one. With a test suite that works without needing to access wandb.ai.
This would be based on Wandb.jl, I think.
However, since Wandb is based on PythonCall, and Python has its GIL, I worry if we would need to start a separate Python process in the same environment, launch Wandb inside that process, and use a Channel to ensure synchronous logging. Though at that point maybe it's easier to go through the Python API directly?
Feature Request
Would be good to have a fully-implemented WandbLoggerSpec analogous to the tensorboard one. With a test suite that works without needing to access wandb.ai.
This would be based on Wandb.jl, I think.
However, since Wandb is based on PythonCall, and Python has its GIL, I worry if we would need to start a separate Python process in the same environment, launch Wandb inside that process, and use a Channel to ensure synchronous logging. Though at that point maybe it's easier to go through the Python API directly?