def test_amp_gpu_ddp_slurm_managed(tmpdir):
"""Make sure DDP + AMP work."""
# simulate setting slurm flags
tutils.set_random_master_port()
os.environ['SLURM_LOCALID'] = str(0)
model = EvalModelTemplate(tutils.get_default_hparams())
# exp file to get meta
logger = tutils.get_default_logger(tmpdir)
# exp file to get weights
checkpoint = tutils.init_checkpoint_callback(logger)
# fit model
trainer = Trainer(
max_epochs=1,
gpus=[0],
distributed_backend='ddp',
precision=16,
checkpoint_callback=checkpoint,
logger=logger,
)
trainer.is_slurm_managing_tasks = True
result = trainer.fit(model)
# correct result and ok accuracy
assert result == 1, 'amp + ddp model failed to complete'
# test root model address
assert trainer.resolve_root_node_address('abc') == 'abc'
assert trainer.resolve_root_node_address('abc[23]') == 'abc23'
assert trainer.resolve_root_node_address('abc[23-24]') == 'abc23'
assert trainer.resolve_root_node_address(
'abc[23-24, 45-40, 40]') == 'abc23'
def test_amp_gpu_ddp_slurm_managed(tmpdir):
"""Make sure DDP + AMP work."""
if not tutils.can_run_gpu_test():
return
tutils.reset_seed()
# simulate setting slurm flags
tutils.set_random_master_port()
os.environ['SLURM_LOCALID'] = str(0)
hparams = tutils.get_hparams()
model = LightningTestModel(hparams)
trainer_options = dict(
show_progress_bar=True,
max_epochs=1,
gpus=[0],
distributed_backend='ddp',
precision=16
)
# exp file to get meta
logger = tutils.get_test_tube_logger(tmpdir, False)
# exp file to get weights
checkpoint = tutils.init_checkpoint_callback(logger)
# add these to the trainer options
trainer_options['checkpoint_callback'] = checkpoint
trainer_options['logger'] = logger
# fit model
trainer = Trainer(**trainer_options)
trainer.is_slurm_managing_tasks = True
result = trainer.fit(model)
# correct result and ok accuracy
assert result == 1, 'amp + ddp model failed to complete'
# test root model address
assert trainer.resolve_root_node_address('abc') == 'abc'
assert trainer.resolve_root_node_address('abc[23]') == 'abc23'
assert trainer.resolve_root_node_address('abc[23-24]') == 'abc23'
assert trainer.resolve_root_node_address('abc[23-24, 45-40, 40]') == 'abc23'
def test_amp_gpu_ddp_slurm_managed(tmpdir):
"""Make sure DDP + AMP work."""
# simulate setting slurm flags
tutils.set_random_master_port()
model = BoringModel()
# exp file to get meta
logger = tutils.get_default_logger(tmpdir)
# exp file to get weights
checkpoint = tutils.init_checkpoint_callback(logger)
# fit model
trainer = Trainer(
default_root_dir=tmpdir,
max_epochs=1,
gpus=[0],
accelerator='ddp_spawn',
precision=16,
callbacks=[checkpoint],
logger=logger,
)
trainer.is_slurm_managing_tasks = True
trainer.fit(model)
# correct result and ok accuracy
assert trainer.state == TrainerState.FINISHED, 'amp + ddp model failed to complete'
# test root model address
assert trainer.slurm_connector.resolve_root_node_address('abc') == 'abc'
assert trainer.slurm_connector.resolve_root_node_address(
'abc[23]') == 'abc23'
assert trainer.slurm_connector.resolve_root_node_address(
'abc[23-24]') == 'abc23'
assert trainer.slurm_connector.resolve_root_node_address(
'abc[23-24, 45-40, 40]') == 'abc23'
def test_dp_resume(tmpdir):
"""Make sure DP continues training correctly."""
hparams = EvalModelTemplate.get_default_hparams()
model = EvalModelTemplate(**hparams)
trainer_options = dict(
max_epochs=1,
gpus=2,
distributed_backend='dp',
)
# get logger
logger = tutils.get_default_logger(tmpdir)
# exp file to get weights
# logger file to get weights
checkpoint = tutils.init_checkpoint_callback(logger)
# add these to the trainer options
trainer_options['logger'] = logger
trainer_options['checkpoint_callback'] = checkpoint
# fit model
trainer = Trainer(**trainer_options)
trainer.is_slurm_managing_tasks = True
result = trainer.fit(model)
# track epoch before saving. Increment since we finished the current epoch, don't want to rerun
real_global_epoch = trainer.current_epoch + 1
# correct result and ok accuracy
assert result == 1, 'amp + dp model failed to complete'
# ---------------------------
# HPC LOAD/SAVE
# ---------------------------
# save
trainer.hpc_save(tmpdir, logger)
# init new trainer
new_logger = tutils.get_default_logger(tmpdir, version=logger.version)
trainer_options['logger'] = new_logger
trainer_options['checkpoint_callback'] = ModelCheckpoint(tmpdir)
trainer_options['limit_train_batches'] = 0.5
trainer_options['limit_val_batches'] = 0.2
trainer_options['max_epochs'] = 1
new_trainer = Trainer(**trainer_options)
# set the epoch start hook so we can predict before the model does the full training
def assert_good_acc():
assert new_trainer.current_epoch == real_global_epoch and new_trainer.current_epoch > 0
# if model and state loaded correctly, predictions will be good even though we
# haven't trained with the new loaded model
dp_model = new_trainer.model
dp_model.eval()
dataloader = trainer.train_dataloader
tutils.run_prediction(dataloader, dp_model, dp=True)
# new model
model = EvalModelTemplate(**hparams)
model.on_train_start = assert_good_acc
# fit new model which should load hpc weights
new_trainer.fit(model)
# test freeze on gpu
model.freeze()
model.unfreeze()
def test_dp_resume(tmpdir):
"""Make sure DP continues training correctly."""
model = BoringModel()
trainer_options = dict(max_epochs=1,
gpus=2,
accelerator='dp',
default_root_dir=tmpdir)
# get logger
logger = tutils.get_default_logger(tmpdir)
# exp file to get weights
# logger file to get weights
checkpoint = tutils.init_checkpoint_callback(logger)
# add these to the trainer options
trainer_options['logger'] = logger
trainer_options['callbacks'] = [checkpoint]
# fit model
trainer = Trainer(**trainer_options)
trainer.is_slurm_managing_tasks = True
trainer.fit(model)
# track epoch before saving. Increment since we finished the current epoch, don't want to rerun
real_global_epoch = trainer.current_epoch + 1
# correct result and ok accuracy
assert trainer.state == TrainerState.FINISHED, f"Training failed with {trainer.state}"
# ---------------------------
# HPC LOAD/SAVE
# ---------------------------
# save
trainer.checkpoint_connector.hpc_save(tmpdir, logger)
# init new trainer
new_logger = tutils.get_default_logger(tmpdir, version=logger.version)
trainer_options['logger'] = new_logger
trainer_options['callbacks'] = [ModelCheckpoint(dirpath=tmpdir)]
trainer_options['limit_train_batches'] = 0.5
trainer_options['limit_val_batches'] = 0.2
trainer_options['max_epochs'] = 1
new_trainer = Trainer(**trainer_options)
class CustomModel(BoringModel):
def __init__(self):
super().__init__()
self.on_train_start_called = False
# set the epoch start hook so we can predict before the model does the full training
def on_train_start(self):
assert self.trainer.current_epoch == real_global_epoch and self.trainer.current_epoch > 0
# if model and state loaded correctly, predictions will be good even though we
# haven't trained with the new loaded model
dp_model = new_trainer.model
dp_model.eval()
dp_model.module.module.running_stage = RunningStage.EVALUATING
dataloader = self.train_dataloader()
tpipes.run_prediction(self.trainer.lightning_module, dataloader)
self.on_train_start_called = True
# new model
model = CustomModel()
# fit new model which should load hpc weights
new_trainer.fit(model)
assert model.on_train_start_called
# test freeze on gpu
model.freeze()
model.unfreeze()
def test_amp_gpu_ddp_slurm_managed():
"""
Make sure DDP + AMP work
:return:
"""
if not can_run_gpu_test():
return
# simulate setting slurm flags
os.environ['MASTER_PORT'] = str(np.random.randint(12000, 19000, 1)[0])
os.environ['SLURM_LOCALID'] = str(0)
hparams = get_hparams()
model = LightningTestModel(hparams)
trainer_options = dict(show_progress_bar=True,
max_nb_epochs=1,
gpus=[0],
distributed_backend='ddp',
use_amp=True)
save_dir = init_save_dir()
# exp file to get meta
exp = get_exp(False)
exp.argparse(hparams)
exp.save()
# exp file to get weights
checkpoint = ModelCheckpoint(save_dir)
# add these to the trainer options
trainer_options['checkpoint_callback'] = checkpoint
trainer_options['experiment'] = exp
# fit model
trainer = Trainer(**trainer_options)
trainer.is_slurm_managing_tasks = True
result = trainer.fit(model)
# correct result and ok accuracy
assert result == 1, 'amp + ddp model failed to complete'
# test root model address
assert trainer.resolve_root_node_address('abc') == 'abc'
assert trainer.resolve_root_node_address('abc[23]') == 'abc23'
assert trainer.resolve_root_node_address('abc[23-24]') == 'abc23'
assert trainer.resolve_root_node_address(
'abc[23-24, 45-40, 40]') == 'abc23'
# test model loading with a map_location
map_location = 'cuda:1'
pretrained_model = load_model(exp, save_dir, True, map_location)
# test model preds
run_prediction(model.test_dataloader, pretrained_model)
if trainer.use_ddp:
# on hpc this would work fine... but need to hack it for the purpose of the test
trainer.model = pretrained_model
trainer.optimizers, trainer.lr_schedulers = pretrained_model.configure_optimizers(
)
# test HPC loading / saving
trainer.hpc_save(save_dir, exp)
trainer.hpc_load(save_dir, on_gpu=True)
# test freeze on gpu
model.freeze()
model.unfreeze()
clear_save_dir()
def test_dp_resume():
"""
Make sure DP continues training correctly
:return:
"""
if not tutils.can_run_gpu_test():
return
tutils.reset_seed()
hparams = tutils.get_hparams()
model = LightningTestModel(hparams)
trainer_options = dict(
show_progress_bar=True,
max_nb_epochs=2,
gpus=2,
distributed_backend='dp',
)
save_dir = tutils.init_save_dir()
# get logger
logger = tutils.get_test_tube_logger(debug=False)
# exp file to get weights
# logger file to get weights
checkpoint = tutils.init_checkpoint_callback(logger)
# add these to the trainer options
trainer_options['logger'] = logger
trainer_options['checkpoint_callback'] = checkpoint
# fit model
trainer = Trainer(**trainer_options)
trainer.is_slurm_managing_tasks = True
result = trainer.fit(model)
# track epoch before saving
real_global_epoch = trainer.current_epoch
# correct result and ok accuracy
assert result == 1, 'amp + dp model failed to complete'
# ---------------------------
# HPC LOAD/SAVE
# ---------------------------
# save
trainer.hpc_save(save_dir, logger)
# init new trainer
new_logger = tutils.get_test_tube_logger(version=logger.version)
trainer_options['logger'] = new_logger
trainer_options['checkpoint_callback'] = ModelCheckpoint(save_dir)
trainer_options['train_percent_check'] = 0.2
trainer_options['val_percent_check'] = 0.2
trainer_options['max_nb_epochs'] = 1
new_trainer = Trainer(**trainer_options)
# set the epoch start hook so we can predict before the model does the full training
def assert_good_acc():
assert new_trainer.current_epoch == real_global_epoch and new_trainer.current_epoch > 0
# if model and state loaded correctly, predictions will be good even though we
# haven't trained with the new loaded model
dp_model = new_trainer.model
dp_model.eval()
dataloader = trainer.get_train_dataloader()
tutils.run_prediction(dataloader, dp_model, dp=True)
# new model
model = LightningTestModel(hparams)
model.on_sanity_check_start = assert_good_acc
# fit new model which should load hpc weights
new_trainer.fit(model)
# test freeze on gpu
model.freeze()
model.unfreeze()
tutils.clear_save_dir()
def main():
"""
Make sure DDP + AMP continue training correctly
:return:
"""
hparams = get_hparams()
model = LightningTestModel(hparams)
trainer_options = dict(
show_progress_bar=True,
max_nb_epochs=4,
gpus=2,
distributed_backend='dp',
)
save_dir = init_save_dir()
# exp file to get meta
exp = get_exp(False)
exp.argparse(hparams)
exp.save()
# exp file to get weights
checkpoint = ModelCheckpoint(save_dir)
# add these to the trainer options
trainer_options['experiment'] = exp
trainer_options['checkpoint_callback'] = checkpoint
# fit model
trainer = Trainer(**trainer_options)
trainer.is_slurm_managing_tasks = True
result = trainer.fit(model)
# track epoch before saving
real_global_epoch = trainer.current_epoch
# correct result and ok accuracy
assert result == 1, 'amp + dp model failed to complete'
# ---------------------------
# HPC LOAD/SAVE
# ---------------------------
# save
trainer.hpc_save(save_dir, exp)
# init new trainer
new_exp = get_exp(False, version=exp.version)
trainer_options['experiment'] = new_exp
trainer_options['checkpoint_callback'] = ModelCheckpoint(save_dir)
trainer_options['train_percent_check'] = 0.2
trainer_options['val_percent_check'] = 0.2
trainer_options['max_nb_epochs'] = 1
new_trainer = Trainer(**trainer_options)
# set the epoch start hook so we can predict before the model does the full training
def assert_good_acc():
assert trainer.current_epoch == real_global_epoch and trainer.current_epoch > 0
# if model and state loaded correctly, predictions will be good even though we
# haven't trained with the new loaded model
dp_model = new_trainer.model
dp_model.eval()
_ = [run_prediction(dataloader, dp_model, dp=True) for dataloader in trainer.val_dataloader]
# new model
model = LightningTestModel(hparams)
model.on_sanity_check_start = assert_good_acc
# fit new model which should load hpc weights
new_trainer.fit(model)
# test freeze on gpu
model.freeze()
model.unfreeze()
clear_save_dir()
def test_amp_gpu_ddp_slurm_managed():
"""
Make sure DDP + AMP work
:return:
"""
if not can_run_gpu_test():
return
reset_seed()
# simulate setting slurm flags
set_random_master_port()
os.environ['SLURM_LOCALID'] = str(0)
hparams = get_hparams()
model = LightningTestModel(hparams)
trainer_options = dict(show_progress_bar=True,
max_nb_epochs=1,
gpus=[0],
distributed_backend='ddp',
use_amp=True)
save_dir = init_save_dir()
# exp file to get meta
logger = get_test_tube_logger(False)
# exp file to get weights
checkpoint = init_checkpoint_callback(logger)
# add these to the trainer options
trainer_options['checkpoint_callback'] = checkpoint
trainer_options['logger'] = logger
# fit model
trainer = Trainer(**trainer_options)
trainer.is_slurm_managing_tasks = True
result = trainer.fit(model)
# correct result and ok accuracy
assert result == 1, 'amp + ddp model failed to complete'
# test root model address
assert trainer.resolve_root_node_address('abc') == 'abc'
assert trainer.resolve_root_node_address('abc[23]') == 'abc23'
assert trainer.resolve_root_node_address('abc[23-24]') == 'abc23'
assert trainer.resolve_root_node_address(
'abc[23-24, 45-40, 40]') == 'abc23'
# test model loading with a map_location
pretrained_model = load_model(logger.experiment,
trainer.checkpoint_callback.filepath)
# test model preds
[
run_prediction(dataloader, pretrained_model)
for dataloader in trainer.get_test_dataloaders()
]
if trainer.use_ddp:
# on hpc this would work fine... but need to hack it for the purpose of the test
trainer.model = pretrained_model
trainer.optimizers, trainer.lr_schedulers = pretrained_model.configure_optimizers(
)
# test HPC loading / saving
trainer.hpc_save(save_dir, logger)
trainer.hpc_load(save_dir, on_gpu=True)
# test freeze on gpu
model.freeze()
model.unfreeze()
clear_save_dir()