Summary: 33 instances, 27 unique

Text	Count
# FIXME: Models will use context "with torch.no_grad():", so the lifetime of no_grad will end after the eval().	1
# TODO: This is a _HORRIBLE_ patch related to #208	2
# TODO: deduplicate with `torchbenchmark.util.model.no_grad`	1
# TODO: perform subsamping	1
# TODO: currently only support 1 GPU device	1
TODO: this is a little HACK now, put batch_size here now.	1
# TODO (@mttk): Populate classs with default values of special symbols	1
# TODO: this should also be done with the ProgressMeter	1
# TODO: implement state passing for lstms	1
# FIXME: Must incorporate this "torch.is_grad_enabled()" inside of actual eval() func.	1
# TODO: Modify and update the model to apply metadata changes by the user.	1
# TODO: Translate bpe encoded files	1
# TODO: in torch 1.0, torch.mean() support dim list	1
# TODO: make it with torch instead of numpy	1
# TODO: expand to batch operation.	1
# TODO - a lot of this was copied from pytorch/jit/scripts/log_extract.py,	1
# assert 'weight' not in benchmarks[benchmark], "TODO implement manual benchmark weights"	2
# TODO: Batch translation	1
# TODO - currently load_model assumes cuda	1
# TODO: Use nn.LayerNorm to impl cLN to speed up	1
assert 'weight' not in category_spec, "TODO implement manual category weights"	2
assert benchmarks[benchmark] is None, "TODO handle benchmark as dict of config specs"	2
assert 'weight' not in benchmarks, "TODO implement manual task weights"	2
# TODO: when P = 3 here works fine, but when P = 2 maybe need to pad?	1
assert 'weight' not in tasks, "TODO implement manual domain weights"	2
# TODO: Try different terminate conditions.	1
# TODO: Also update the `freq`, although it is not likely to be used.	1