Summary: 30 instances, 29 unique

Text	Count
# TODO If init_tree is enabled, during the first pass we might want to	1
# TODO It may be interesting to get a random bucket among the acquirable	1
# TODO We could check that, for all i, we have either tensor[i] < 0 or	1
# TODO Change this function to use the same cost model as the LockServer	1
* the left- and right-hand side of relation type i. (TODO: remove these	1
# TODO Check that the batch size is a multiple of the batch negative number	1
# FIXME join distributed workers (not really necessary)	1
FIXME: torchbiggraph.rpc should be fixed to not require torch.serialization,	1
FIXME: torch.distributed.recv should not require you to provide the	1
# TODO have two permanent storages on GPU and move stuff in and out	1
# FIXME: how to properly copy bytes to ByteTensor?	1
# FIXME: there's a slight danger here, say that a multi-machine job fails	1
# FIXME Workaround for https://github.com/pytorch/pytorch/issues/15870	1
# TODO Remove noqa when flake8 will understand kw_only added in attrs-18.2.0.	1
# TODO Remove noqa when flake8 will understand kw_only added in attrs-18.2.0.	1
# FIXME This adapts from the pre-D14024710 format; remove eventually.	2
# FIXME Add the rank to the name of each process.	1
# FIXME should we only delay if iteration_idx == 0?	1
# FIXME: this is a terrible API	1
# FIXME Workaround for https://github.com/pytorch/pytorch/issues/15223.	1
# FIXME: we've got to get rid of this two-pass nonsense for dynamically sized	1
# TODO Check that all partitioned entity types have the same number of	1
# FIXME: This is not ideal. Perhaps we should pass in the config	1
// TODO check type and sizes first;	1
# FIXME: is it efficient to torch.save into a buf? It's going to have to copy	1
# FIXME Is PyTorch's sort stable? Won't this risk messing up the random shuffle?	1
# TODO make this a non-inplace operation	1
# FIXME This order assumes higher affinity on the left-hand side, as it's	1
# FIXME: it's not really safe to do partial batches if num_batch_negs != 0	1