support relative path for best_model_checkpoint

[HaebinShin]

What does this PR do?

Fixes #7431

When I give a relative path to output_dir, it raises some error at line 1222.

transformers/src/transformers/trainer.py

Lines 1205 to 1227 in 8348105

	def _sorted_checkpoints(self, checkpoint_prefix=PREFIX_CHECKPOINT_DIR, use_mtime=False) -> List[str]:
	ordering_and_checkpoint_path = []
	glob_checkpoints = [str(x) for x in Path(self.args.output_dir).glob(f"{checkpoint_prefix}-*")]
	for path in glob_checkpoints:
	if use_mtime:
	ordering_and_checkpoint_path.append((os.path.getmtime(path), path))
	else:
	regex_match = re.match(f".*{checkpoint_prefix}-([0-9]+)", path)
	if regex_match and regex_match.groups():
	ordering_and_checkpoint_path.append((int(regex_match.groups()[0]), path))
	checkpoints_sorted = sorted(ordering_and_checkpoint_path)
	checkpoints_sorted = [checkpoint[1] for checkpoint in checkpoints_sorted]
	# Make sure we don't delete the best model.
	if self.state.best_model_checkpoint is not None:
	best_model_index = checkpoints_sorted.index(self.state.best_model_checkpoint)
	checkpoints_sorted[best_model_index], checkpoints_sorted[-1] = (
	checkpoints_sorted[-1],
	checkpoints_sorted[best_model_index],
	)
	return checkpoints_sorted

If I give './path1/path2' to output_dir, checkpoints_sorted becomes 'path1/path2' by Path library on line 1208.
But self.state.best_model_checkpoint is still './path1/path2'.
So it will raise error as below.

Traceback (most recent call last):
  File "../finetune.py", line 369, in <module>
    main(**vars(args))
  File "../finetune.py", line 315, in main
    device)
  File "../finetune.py", line 120, in train
    trainer.train(optim_pretrained_path)
  File "/usr/local/lib/python3.6/dist-packages/transformers/trainer.py", line 803, in train
    self._maybe_log_save_evaluate(tr_loss, model, trial, epoch)
  File "/usr/local/lib/python3.6/dist-packages/transformers/trainer.py", line 860, in _maybe_log_save_evaluate
    self._save_checkpoint(model, trial, metrics=metrics)
  File "/usr/local/lib/python3.6/dist-packages/transformers/trainer.py", line 918, in _save_checkpoint
    self._rotate_checkpoints(use_mtime=True)
  File "/usr/local/lib/python3.6/dist-packages/transformers/trainer.py", line 1235, in _rotate_checkpoints
    checkpoints_sorted = self._sorted_checkpoints(use_mtime=use_mtime)
  File "/usr/local/lib/python3.6/dist-packages/transformers/trainer.py", line 1223, in _sorted_checkpoints
    best_model_index = checkpoints_sorted.index(self.state.best_model_checkpoint)
ValueError: './results/use_pretrained_test/checkpoint-1162' is not in list

So I resolve this error by using Path for self.state.best_model_checkpoint.
Any other idea is also welcome. please check it

[sgugger]

Ah yes, since there is a Path above, one might be necessary here too. Thanks for fixing!
Failure seems spurious but just in case, waiting for the test to pass to merge.