pytorch

mirror of https://github.com/pytorch/pytorch.git synced 2025-10-20 21:14:14 +08:00

Files

Sam Gross c1fa449763 Break reference cycle in load_state_dict (#20397 )

Summary:
load_state_dict includes a recursive inner function `load` that captures
Tensors through the close-over variable `state_dict`. Because it's
recursive, it also captures itself leading to a reference cycle.

This breaks the reference cycle so that any Tensors in state_dict can be
collected immediately instead of waiting until the next GC cycle.

Alternatively, we could have passed `state_dict` and `metadata` as
arguments to load to prevent capture of Tensors. (That would still
result in cyclic garbage, but not any cyclic garbage of Tensors).

See:
https://github.com/pytorch/pytorch/issues/20199#issuecomment-491089004
Pull Request resolved: https://github.com/pytorch/pytorch/pull/20397

Differential Revision: D15414834

Pulled By: colesbury

fbshipit-source-id: 4c2275a08b2d8043deb3779db28be03bda15872d

2019-05-20 11:46:00 -07:00

_functions

Port adaptive_avg_pool3d to ATen (#19898 )

2019-05-13 11:29:22 -07:00

backends

Move RNN implementations to C++ (#10481 )

2018-08-15 13:25:41 -07:00

modules

Break reference cycle in load_state_dict (#20397 )

2019-05-20 11:46:00 -07:00

parallel

Emphasize all DDP forward() outputs must participate in computing loss (#20586 )