Feat: Add tensorboard support

[NanoCode012]

Should close #763

Untested, but should work.

[winglian]

based on the docs, perhaps we should convert this to None if it's empty? that way it should default back to the all case if not setup

report_to (str or List[str], optional, defaults to "all") — The list of integrations to report the 
results and logs to. Supported platforms are "azure_ml", "clearml", "codecarbon", 
"comet_ml", "dagshub", "flyte", "mlflow", "neptune", "tensorboard", and "wandb". Use 
"all" to report to all integrations installed, "none" for no integrations.

[ein-ich]

Ok I tested it and it works; but I realized it even works on main and without tensorboard: true. It seems it always created a tensorboard runs logdir in output_dir, and I just never noticed?

[winglian]

Ok I tested it and it works; but I realized it even works on main and without tensorboard: true. It seems it always created a tensorboard runs logdir in output_dir, and I just never noticed?

Maybe it never did that until you installed the tensorboard package?

[ein-ich]

Yes you're right, it only creates the logs when tensorboard is installed.

[NanoCode012]

Ok I tested it and it works; but I realized it even works on main and without tensorboard: true. It seems it always created a tensorboard runs logdir in output_dir, and I just never noticed?

Hmm, that's weird. It should've been report_to=None if not using wandb, which according to the docs would not have any integration. Maybe tensor board logs are just defaulted? In such a case, is this PR needed?

[ein-ich]

(from https://huggingface.co/docs/transformers/v4.34.1/en/main_classes/callback)

[winglian]

Yeah, I think the only thing we need is to add tensorboard to the requirements.txt or to document that people should install it if wanted. I'm leaning towards just installing it with requirements or maybe with an extras like pip install -e .[tensorboard]

[NanoCode012]

Closed in favor of adding tensorboard to requirements

[fakerybakery]

Hi, sorry I'm a bit confused, how do I use Tensorboard in Axolotl then?

[NanoCode012]

@fakerybakery , you can just train normally without setting wandb configs or setting it to disabled