src/datatuner/lm/evaluate.py (5 lines):
	- line 53: )  # TODO: Only works for batch size 1 currently
	- line 239: # TODO: fix the hardcoded newline character
	- line 264: # TODO: ignore if prev_beam is not complete
	- line 338: # TODO: the reranker model compatible with the batched beam search
	- line 550: # TODO: validate this strategy in languages without a space separating words


src/datatuner/lm/train.py (1 line):
	- line 180: # TODO: fix issue in mlflow trying to delete the experiment multiple times


src/datatuner/lm/data_loader.py (1 line):
	- line 34: # TODO: change this to be the max of the current tokenizer by name, not min of all maxes