The changes are quite local and limited in nature -- viz., checking for
some indicator environment variables. We check for (SLURM_LOCALID,
NODE_RANK, GROUP_RANK) in order. If multiple are found set, a warning is
logged.
This patch also fixes a minor bug with comparing the `WORLD_SIZE`
environment variable. This can be a string type.
* Join Horovod workers at the end of trainer.fit() to prevent race conditions following training
* flake8
* flake8
Co-authored-by: Jirka <jirka.borovec@seznam.cz>
* update prog. bar metrics on train epoch end
* changelog
* wip test
* more thorough testing
* comments
* update docs
* move test
Co-authored-by: Jirka <jirka.borovec@seznam.cz>
* diable val and test shuffling
* diable val and test shuffling
* diable val and test shuffling
* diable val and test shuffling
* log
* condition
* shuffle
* refactor
Co-authored-by: J. Borovec <jirka.borovec@seznam.cz>
* squash and rebase
sanity check hooks
sanity check callback hook finish
moved core progress bar functionality into callback
wip
remove duplicate merge
clean up
imports
docs
sanity check progress bar main
sanity
move callback calls
init progrss bar callback
configuration and docs
changelog
rate decorator
pass process_position
disable on rank > 0
position index
is_enabled
remove decorator
refactor init tqdm bars
callback method ordering
cannot reset when disabled
sequence -> list
default values
fix has no attr _time()
move on_val_end to proper place
fix the pickle issue
update warning
properties
check for None
remove old comment
switch order
pull out non-tqdm functionality into base class
documentation for the base class
docs
fix refresh rate issue in validation
restrict type hint of trainer arg
more docs
update trainer docs
rst docs
fix lines too long
fix test
add missing type hints
fix typo
move docstring to __init__ solves doctest failures
remove doctest :(( can't fix the pickle error
fix example
simplify by saving trainer reference
fix docs errors
move docstring
initial value
multiple val checks per epoch
simpler handling of inf dataset sizes
update inf docs
renamed training_tqdm_dict
rename get_tqdm_dict
rename occurences of tqdm
update changelog
fix doctest
fix formatting errors
added callback tests
progress bar on off test
more tests for progress bar
weird test fix?
add ignored property
disable default progress bar in LR finder
change enable/disable behavior
trying doctest in CI again
undo doctest pickle error
undo doctest pickle error :((
remove progress_bar_callback Trainer arg and fix tests
restore progress bar after auto lr find
update docs
fix rebase
fix wrong negation
* fix fast dev run total
* more thorough testing
* remove old args
* fix merge
* fix merge
* separate tests
* type hint total batches
* reduce if
Co-Authored-By: Jirka Borovec <Borda@users.noreply.github.com>
* is_disabled
Co-Authored-By: Jirka Borovec <Borda@users.noreply.github.com>
* is_enabled
Co-Authored-By: Jirka Borovec <Borda@users.noreply.github.com>
* rename enabled/disabled
* move deprecated api
* remove duplicated test from merge
* fix rename is_disabled
* newline
* test also testprogress for fast dev run
Co-authored-by: J. Borovec <jirka.borovec@seznam.cz>
Co-authored-by: Jirka Borovec <Borda@users.noreply.github.com>
* Attempt to fix#1468
* Remove the if statement, it doesn't actually make any difference
* Update docs
* Correct warnings I caused in the last commit
* Add to changelog
* Actually add to changelog
* Clarify documentation and examples
* Update CHANGELOG.md
Co-authored-by: Jirka Borovec <Borda@users.noreply.github.com>
* improved docs for core
update links
add references to hooks lifecycle
wip
continue with __init__.py
improve docs for memory.py
improve docs for saving.py
simpler links
fix formatting
* move hooks lifecycle to top of file
* fix doctest import problem
* add missing hook in lifecycle
* Add tests for distributed backend config
* Refactor set_distributed_mode
* Use gloo backend on cpu
* Use 127.0.0.1 instead of 127.0.0.2
Not totally clear on why this is necessary, but it seemt to work
* Update LightningDDP so that it works with CPU
* Add ddp_cpu backend and num_processes Trainer arg
* PEP8
* Fix test skipping. Inequalities are hard :/
* Skip ddp_cpu test on Windows
* Make a few more cases fall back to ddp_cpu
* New function name
* Flake8
* Don't test distributed on MacOS with torch < 1.3
Support for distributed in MacOS was added in Torch 1.3.0
* Add ddp_cpu and num_processes to docs
* Parametrize trainer config tests
* Tweak warning
Co-Authored-By: Jirka Borovec <Borda@users.noreply.github.com>
* Remove redundant test
* Replace pass branches with comments
* Add missing warnings import
* save_path -> root_dir
* Use new rank_zero_warn
* Whitespace
* Apply suggestions from code review
* formatting
Co-authored-by: Jirka Borovec <Borda@users.noreply.github.com>
Co-authored-by: J. Borovec <jirka.borovec@seznam.cz>
* Make training_epoch_end behave like validation_epoch_end + minor fixes in docstrings.
* Minor fixes (Borda's comments).
* Detach tensors in batch_output (to avoid possible memory leak) + doc fix.
Co-authored-by: Jean-Baptiste SCHIRATTI <jean-baptisteschiratti@MacBook-Pro-de-Jean-Baptiste.local>
* fix RunningMean
* changelog
* fix none
* Update supporters.py
just needed to multiply by zero for init
* Revert "Update supporters.py"
This reverts commit 7e0da6c6
* fix NaN
* formatting
Co-authored-by: William Falcon <waf2107@columbia.edu>
* Update docs so users know the desired manner of configuring learning rate schedulers.
* update list
* as note
Co-authored-by: Jirka Borovec <Borda@users.noreply.github.com>
* check if hparams_type exists in checkpoint dictionary for backward compatibility
* concisely maintain backward compatibility for hparams type
* Bug fix in checkpoint loading (#1132)
Peter Yu