How to use API in offline mode to access local log files?

I would like to use the wandb API at the end of the training process to output results, trainign arguments and history. Because the Run object in wandb_logger.experiment is missing some information, I need to use the API to load the Run object after training (See https://github.com/PyTorchLightning/pytorch-lightning/issues/3493)

If I am connected (wandb on), after training I can do:

    wandb_logger.experiment.finish() # I seem to need this sometimes
    run_path = wandb_logger.experiment.path
    read_access_run = wandb.Api().run(run_path)

And then use read_access_run.config, read_access_run.summary, read_access_run.history()

It I am offline (wandb off), however, as I will be often when testing code, I have the following error:

ValueError: Could not find run <Run yyy/xxx/16gi3ac0 (not found)>
wandb.errors.error.CommError: Could not find run <Run yyy/xxx/16gi3ac0 (not found)>

I would like to access things like training history even if I am offline. What shall I do?

[Tahlor]

I often run jobs on nodes that don’t have internet access, but other nodes that share the file system do have internet access. You could imagine the internet-less nodes produce output logs intermittently and a process on an internet node uploads these (or it’s uploaded upon completion, or some kind of supported local proxy server that relays the logs). Does any kind of functionality like this exist, or is it planned, or would it be difficult (for me) to hack? Thanks!

[bhattg]

Actually having such functionality will be great, and I agree with @Tahlor