How to Continue Epoch Count from Last Checkpoint in TensorBoard Logs? #2873

If training is interrupted and then resumed, it will continue from the last epoch number.
However, when recording training results with TensorBoard, the newly recorded epochs start from 0 again, as shown in the image below:

In the examples from https://github.com/speechbrain/speechbrain/tree/develop/recipes/Voicebank/enhance (and others behave similarly), for instance, in MetricGAN’s train.py:

if stage == sb.Stage.VALID:
    if self.hparams.use_tensorboard:
        valid_stats = {
            "SI-SNR": -stage_loss,
            "pesq": 5 * self.pesq_metric.summarize("average") - 0.5,
            "stoi": -self.stoi_metric.summarize("average"),
        }
        self.hparams.tensorboard_train_logger.log_stats(valid_stats)
    self.hparams.train_logger.log_stats(
        {"Epoch": epoch},
        train_stats={"loss": self.train_loss},
        valid_stats=stats,
    )
    self.checkpointer.save_and_keep_only(
        meta=stats, max_keys=[self.hparams.target_metric]
    )

Here, {"Epoch": epoch} starts from 0. Is there a specific reason or intent behind this design choice?

Using "relative" or "wall" in TensorBoard can partially address this, but it doesn’t allow for side-by-side comparison with other runs that have the same epoch count.

Is there a way to modify this so that the epoch numbering continues from the last epoch number instead of resetting to 0?