Matt Painter 6e7dc9c236 Fixes resuming checkpoints rerunning last epoch (#866) ...

* Properly restore current epoch and global step on resume

* Add test

* Move increment to saving rather than loading

* Fix other tests that refer to current epoch

* Formatting

* Add warning for mid-epoch resuming

* Formatting

* Fix warning check for accumulated batches

* Add variable to init

* Formatting

* Add check for 0 training steps

* Make check more readable

2020-02-21 20:27:19 -05:00

..

__init__.py

update Docs [links & formatting] (#769)

2020-02-09 17:39:10 -05:00

auto_mix_precision.py

Enable TPU support (#868)

2020-02-17 16:01:20 -05:00

callback_config.py

Refactor callbacks (#776)

2020-02-16 00:03:05 -05:00

data_loading.py

Enable TPU support (#868)

2020-02-17 16:01:20 -05:00

distrib_data_parallel.py

Enable TPU support (#868)

2020-02-17 16:01:20 -05:00

distrib_parts.py

fix tpu transfer bug 2

2020-02-17 17:47:16 -05:00

evaluation_loop.py

Enable TPU support (#868)

2020-02-17 16:01:20 -05:00

ignored_warnings.py

Dim 0 warning (#256)

2019-09-26 13:20:54 -04:00

logging.py

Allow user to specify 'step' key while logging metrics (#808)

2020-02-15 23:35:23 -05:00

model_hooks.py

rename trainer modules, drop _mixin (#571)

2019-12-04 11:39:14 -05:00

trainer.py

new way of passing dataloaders (#759)

2020-02-19 06:00:08 -05:00

training_io.py

Fixes resuming checkpoints rerunning last epoch (#866)

2020-02-21 20:27:19 -05:00

training_loop.py

remove deprecated args to learning rate step function (#890)

2020-02-19 06:37:35 -05:00

training_tricks.py

Refactor callbacks (#776)

2020-02-16 00:03:05 -05:00