I think this work well for static visualizations. If I'll want a more dynamic one I guess I will use tensorboard, and save some static plot to ClearML.

Thanks!

In the end of the day these integration are based on Callback classes. For example, this is the wandb Calback:
class WandbCallback(TrainerCallback): """ A :class:~transformers.TrainerCallbackthat sends the logs toWeight and Biases < >`__.
"""

def __init__(self):
    has_wandb = is_wandb_available()
    assert has_wandb, "WandbCallback requires wandb to be installed. Run `pip install wandb`."
    if has_wandb:
        import wandb

        self._...

Mainly logging. Huggingface's trainer has a "report_to" argument that is supported by tensorboard, wandb, comet, etc. This means that during training all of the metrics are automatically logged to the specified platform (which is very convenient). Is there anyone who has made something similar for clearml?