lightning/tests/loops/test_loops.py

966 lines
36 KiB
Python

# Copyright The PyTorch Lightning team.
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
import os
from copy import deepcopy
from dataclasses import dataclass
from typing import Any, Dict, Iterator
from unittest import mock
from unittest.mock import ANY
import pytest
import torch
from torch.utils.data.dataloader import _MultiProcessingDataLoaderIter, DataLoader
from pytorch_lightning import LightningModule, Trainer
from pytorch_lightning.callbacks import Callback, ModelCheckpoint
from pytorch_lightning.loops import Loop, TrainingBatchLoop
from pytorch_lightning.trainer.progress import BaseProgress
from tests.helpers import BoringModel, RandomDataset
from tests.helpers.runif import RunIf
class NestedLoop(Loop):
def __init__(self):
super().__init__()
self.child_loop0 = None
self.child_loop1 = None
@property
def done(self) -> bool:
return False
def connect(self, child0, child1):
self.child_loop0 = child0
self.child_loop1 = child1
def reset(self) -> None:
pass
def advance(self, *args, **kwargs):
pass
@pytest.mark.parametrize("loop_name", ["fit_loop", "validate_loop", "test_loop", "predict_loop"])
def test_connect_loops_direct(loop_name):
"""Test Trainer referenes in loops on assignment."""
loop = NestedLoop()
with pytest.raises(RuntimeError, match="The loop is not attached to a Trainer"):
_ = loop.trainer
trainer = Trainer()
# trainer.loop_name = loop
setattr(trainer, loop_name, loop)
assert loop.trainer is trainer
def test_connect_loops_recursive():
"""Test Trainer references in a nested loop assigned to a Trainer."""
main_loop = NestedLoop()
child0 = NestedLoop()
child1 = NestedLoop()
main_loop.connect(child0, child1)
with pytest.raises(RuntimeError, match="The loop is not attached to a Trainer"):
_ = main_loop.trainer
with pytest.raises(RuntimeError, match="The loop is not attached to a Trainer"):
_ = main_loop.child_loop0.trainer
trainer = Trainer()
trainer.fit_loop = main_loop
assert child0.trainer is trainer
assert child1.trainer is trainer
def test_connect_subloops(tmpdir):
"""Test connecting individual subloops by calling `trainer.x.y.connect()`"""
model = BoringModel()
trainer = Trainer(default_root_dir=tmpdir, fast_dev_run=True)
epoch_loop = trainer.fit_loop.epoch_loop
new_batch_loop = TrainingBatchLoop()
epoch_loop.connect(batch_loop=new_batch_loop)
assert epoch_loop.batch_loop is new_batch_loop
with pytest.raises(RuntimeError, match="The loop is not attached to a Trainer"):
_ = new_batch_loop.trainer
trainer.fit(model)
assert new_batch_loop.trainer is trainer
class CustomException(Exception):
pass
def test_loop_restore():
class Simple(Loop):
def __init__(self, dataset: Iterator):
super().__init__()
self.iteration_count = 0
self.dataset = dataset
@property
def skip(self) -> bool:
return False
@property
def done(self) -> bool:
return self.iteration_count > len(self.dataset)
def reset(self) -> None:
self.iter_dataset = iter(self.dataset)
if self.restarting:
for _ in range(self.iteration_count):
next(self.iter_dataset)
self.iteration_count += 1
else:
self.outputs = []
def advance(self) -> None:
value = next(self.iter_dataset)
if self.iteration_count == 5:
raise CustomException
self.outputs.append(value)
def on_advance_end(self) -> None:
self.iteration_count += 1
def state_dict(self) -> Dict:
return {"iteration_count": self.iteration_count, "outputs": self.outputs}
def load_state_dict(self, state_dict: Dict) -> None:
self.iteration_count = state_dict["iteration_count"]
self.outputs = state_dict["outputs"]
trainer = Trainer()
data = range(10)
loop = Simple(data)
loop.trainer = trainer
try:
loop.run()
state_dict = {}
except CustomException:
state_dict = loop.state_dict()
loop = Simple(data)
loop.trainer = trainer
loop.load_state_dict(state_dict)
loop.restarting = True
loop.run()
assert not loop.restarting
assert loop.outputs == list(range(10))
def test_loop_hierarchy():
@dataclass
class SimpleProgress(BaseProgress):
increment: int = 0
class Simple(Loop):
def __init__(self, a):
super().__init__()
self.a = a
self.progress = SimpleProgress()
def advance(self, *args: Any, **kwargs: Any) -> None:
loop = getattr(self, "loop_child", None)
if not loop:
return
loop.run()
def on_advance_end(self):
self.progress.increment += 1
@property
def done(self) -> bool:
return self.progress.increment > 0
def reset(self) -> None:
...
def on_save_checkpoint(self) -> Dict:
return {"a": self.a}
def on_load_checkpoint(self, state_dict: Dict) -> None:
self.a = state_dict["a"]
loop_parent = Simple(1)
loop_child = Simple(2)
loop_parent.loop_child = loop_child
# check the trainer reference is propagated
loop_parent.trainer = Trainer()
assert loop_child.trainer is loop_parent.trainer
state_dict = loop_parent.state_dict()
assert state_dict == {
"state_dict": {"a": 1},
"progress": {"increment": 0},
"loop_child.state_dict": {"a": 2},
"loop_child.progress": {"increment": 0},
}
state_dict["loop_child.state_dict"]["a"] = 3
# check restarting after `load_state_dict`
loop_parent.load_state_dict(state_dict)
assert loop_parent.restarting
loop_parent.run()
# check the new state after `run`
state_dict = loop_parent.state_dict()
assert state_dict == {
"state_dict": {"a": 1},
"progress": {"increment": 1},
"loop_child.state_dict": {"a": 3},
"loop_child.progress": {"increment": 1},
}
loop_parent_copy = deepcopy(loop_parent)
assert loop_parent_copy.state_dict() == loop_parent.state_dict()
assert loop_parent_copy.on_save_checkpoint() == state_dict["state_dict"]
assert loop_parent_copy.loop_child.on_save_checkpoint() == state_dict["loop_child.state_dict"]
loop_parent = Simple(1)
loop_child = Simple(2)
loop_parent.loop_child = loop_child
loop_parent.load_state_dict(state_dict)
assert loop_parent.progress.increment == 1
assert loop_parent.loop_child.progress.increment == 1
del loop_parent.loop_child
state_dict = loop_parent.state_dict()
assert state_dict == {"state_dict": {"a": 1}, "progress": {"increment": 1}}
@mock.patch.dict(os.environ, {"PL_FAULT_TOLERANT_TRAINING": "1"})
@pytest.mark.parametrize("stop_epoch", (1, 2))
@pytest.mark.parametrize("stop_batch", (1, 2))
@pytest.mark.parametrize("n_dataloaders,stop_dataloader", [(2, 0), (2, 1), (3, 2)])
def test_loop_restart_progress_multiple_dataloaders(tmpdir, n_dataloaders, stop_dataloader, stop_epoch, stop_batch):
n_batches = 5
n_epochs = 3
class ValidationModel(BoringModel):
def __init__(self):
super().__init__()
def validation_step(self, batch, batch_idx, dataloader_idx):
if self.current_epoch == stop_epoch and batch_idx == stop_batch and dataloader_idx == stop_dataloader:
raise CustomException
return super().validation_step(batch, batch_idx)
def val_dataloader(self):
return [super(ValidationModel, self).val_dataloader() for _ in range(n_dataloaders)]
model = ValidationModel()
model.validation_epoch_end = None
trainer = Trainer(
default_root_dir=tmpdir,
max_epochs=n_epochs,
limit_train_batches=1,
limit_val_batches=n_batches,
num_sanity_val_steps=0,
)
# simulate a failure
with pytest.raises(CustomException):
trainer.fit(model)
ckpt_path = str(tmpdir / ".pl_auto_save.ckpt")
checkpoint = torch.load(ckpt_path)["loops"]["fit_loop"]
total_dataloader = stop_epoch * n_dataloaders + stop_dataloader
expected = {
"total": {"ready": total_dataloader + 1, "completed": total_dataloader},
"current": {"ready": stop_dataloader + 1, "completed": stop_dataloader},
}
assert checkpoint["epoch_loop.val_loop.dataloader_progress"] == expected
trainer.fit_loop.load_state_dict(checkpoint)
# `nbe_`: non-breaking epoch, as in, no exception will be raised. `be_`: breaking epoch
nbe_total_val_batch = stop_epoch * n_dataloaders * n_batches
be_total_val_batch = stop_dataloader * n_batches + stop_batch
total_val_batch = nbe_total_val_batch + be_total_val_batch
expected = {
"total": {
"ready": total_val_batch + 1,
"started": total_val_batch + 1,
"processed": total_val_batch,
"completed": total_val_batch,
},
"current": {
"ready": stop_batch + 1,
"started": stop_batch + 1,
"processed": stop_batch,
"completed": stop_batch,
},
"is_last_batch": False,
}
assert trainer.fit_loop.epoch_loop.val_loop.epoch_loop.batch_progress.state_dict() == expected
@mock.patch.dict(os.environ, {"PL_FAULT_TOLERANT_TRAINING": "1"})
@pytest.mark.parametrize("accumulate_grad_batches", (1, 2, 3))
@pytest.mark.parametrize("n_optimizers", (1, 3, 5))
@pytest.mark.parametrize("stop_epoch", (1, 2))
@pytest.mark.parametrize("stop_batch", (1, 2))
@pytest.mark.parametrize("stop_optimizer", (1, 2))
def test_loop_state_on_exception(accumulate_grad_batches, stop_epoch, stop_batch, stop_optimizer, n_optimizers, tmpdir):
stop_optimizer = stop_optimizer if stop_optimizer < n_optimizers else 0
n_epochs = 3
n_batches = 3
class TestModel(BoringModel):
def __init__(self):
super().__init__()
if n_optimizers > 1:
self.configure_optimizers = self.configure_optimizers_multiple
def training_step(self, batch, batch_idx, optimizer_idx=0):
if self.trainer.current_epoch == stop_epoch and batch_idx == stop_batch and optimizer_idx == stop_optimizer:
raise CustomException
return super().training_step(batch, batch_idx)
def configure_optimizers_multiple(self):
optimizers = [torch.optim.Adam(self.layer.parameters(), lr=0.1) for _ in range(n_optimizers)]
lr_scheduler_0 = torch.optim.lr_scheduler.StepLR(optimizers[0], step_size=1)
lr_scheduler_1 = torch.optim.lr_scheduler.StepLR(optimizers[1], step_size=1)
# no scheduler for optimizer_2
lr_schedulers = [lr_scheduler_0, {"scheduler": lr_scheduler_1, "interval": "step"}]
return optimizers, lr_schedulers
model = TestModel()
model.training_epoch_end = None
trainer = Trainer(
default_root_dir=tmpdir,
max_epochs=n_epochs,
limit_train_batches=n_batches,
limit_val_batches=0,
accumulate_grad_batches=accumulate_grad_batches,
enable_progress_bar=False,
logger=False,
enable_checkpointing=False,
)
# simulate a failure
with pytest.raises(CustomException):
trainer.fit(model)
ckpt_path = str(tmpdir / ".pl_auto_save.ckpt")
assert os.path.exists(ckpt_path)
checkpoint = torch.load(ckpt_path)
optim_progress = trainer.fit_loop.epoch_loop.batch_loop.optimizer_loop.optim_progress
sch_progress = trainer.fit_loop.epoch_loop.scheduler_progress
# `nbe_`: non-breaking epoch, as in, no exception will be raised. `be_`: breaking epoch
nbe_batches_completed = stop_epoch * n_batches
be_batches_completed = stop_batch
be_batches_ready = stop_batch + 1
# lightning applies leftover accumulated gradients when the epoch ends
has_leftover_accumulation_batches = n_batches % accumulate_grad_batches != 0
# number of batches that will call `optimizer.step()` during non-breaking and breaking epochs
nbe_stepping_batches = nbe_batches_completed // accumulate_grad_batches
be_stepping_batches = be_batches_completed // accumulate_grad_batches
nbe_total_opt_steps = (nbe_stepping_batches + has_leftover_accumulation_batches) * n_optimizers
does_last_be_batch_step = be_batches_ready % accumulate_grad_batches == 0 or has_leftover_accumulation_batches
be_total_opt_steps = be_stepping_batches * n_optimizers + does_last_be_batch_step * stop_optimizer
assert optim_progress.optimizer_steps == nbe_total_opt_steps + be_total_opt_steps
assert optim_progress.optimizer.step.current.completed == be_total_opt_steps
has_opt_stepped_in_be = stop_batch + 1 >= accumulate_grad_batches
nbe_total_zero_grad = (nbe_stepping_batches + has_leftover_accumulation_batches) * n_optimizers
does_last_be_batch_zero_grad = be_batches_completed % accumulate_grad_batches == 0
# `max` because the first batch always zero-grads
be_total_zero_grad = max(1, be_stepping_batches) * n_optimizers + stop_optimizer * does_last_be_batch_zero_grad
assert optim_progress.optimizer.zero_grad.total.completed == nbe_total_zero_grad + be_total_zero_grad
assert optim_progress.optimizer.zero_grad.current.completed == be_total_zero_grad
nbe_sch_steps = stop_epoch
be_sch_steps = 0 # the current epoch did not complete
if n_optimizers > 1:
# assumes that the scheduler config is unchanged
# `* 1` because there is only one step-level scheduler
nbe_sch_steps = stop_epoch + nbe_stepping_batches + has_leftover_accumulation_batches * 1
# `0 +` for the epoch-level scheduler
be_sch_steps = 0 + be_stepping_batches
assert sch_progress.total.completed == nbe_sch_steps + be_sch_steps
assert sch_progress.current.completed == be_sch_steps
expected = {
"state_dict": ANY,
"epoch_progress": {
"total": {
"ready": stop_epoch + 1,
"started": stop_epoch + 1,
"processed": stop_epoch,
"completed": stop_epoch,
},
"current": {
"ready": stop_epoch + 1,
"started": stop_epoch + 1,
"processed": stop_epoch,
"completed": stop_epoch,
},
},
"epoch_loop.state_dict": ANY,
"epoch_loop.batch_progress": {
"total": {
"ready": nbe_batches_completed + be_batches_completed + 1,
"started": nbe_batches_completed + be_batches_completed + 1,
"processed": nbe_batches_completed + be_batches_completed,
"completed": nbe_batches_completed + be_batches_completed,
},
"current": {
"ready": stop_batch + 1,
"started": stop_batch + 1,
"processed": stop_batch,
"completed": stop_batch,
},
"is_last_batch": False,
},
"epoch_loop.scheduler_progress": {
"total": {"ready": nbe_sch_steps + be_sch_steps, "completed": nbe_sch_steps + be_sch_steps},
"current": {"ready": be_sch_steps, "completed": be_sch_steps},
},
"epoch_loop.batch_loop.state_dict": ANY,
"epoch_loop.batch_loop.manual_loop.state_dict": ANY,
"epoch_loop.batch_loop.optimizer_loop.state_dict": {},
"epoch_loop.batch_loop.optimizer_loop.optim_progress": {
"optimizer_position": stop_optimizer,
"optimizer": {
"step": {
"total": {
"ready": nbe_total_opt_steps + be_total_opt_steps + has_opt_stepped_in_be,
"completed": nbe_total_opt_steps + be_total_opt_steps,
},
"current": {"ready": be_total_opt_steps + has_opt_stepped_in_be, "completed": be_total_opt_steps},
},
"zero_grad": {
"total": {
"ready": nbe_total_zero_grad + be_total_zero_grad,
"started": nbe_total_zero_grad + be_total_zero_grad,
"completed": nbe_total_zero_grad + be_total_zero_grad,
},
"current": {
"ready": be_total_zero_grad,
"started": be_total_zero_grad,
"completed": be_total_zero_grad,
},
},
},
},
"epoch_loop.val_loop.state_dict": ANY,
"epoch_loop.val_loop.dataloader_progress": ANY,
"epoch_loop.val_loop.epoch_loop.state_dict": ANY,
"epoch_loop.val_loop.epoch_loop.batch_progress": ANY,
"epoch_loop.val_loop._results": ANY,
"epoch_loop._results": ANY,
}
assert checkpoint["loops"]["fit_loop"] == expected
trainer.fit_loop.load_state_dict(checkpoint["loops"]["fit_loop"])
state_dict = trainer.fit_loop.state_dict()
# need to remove these elements for comparison; comparing with `fit_loop.state_dict()` would require the
# fit loop to have an iterator, which is only available during training
state_dict["epoch_loop.state_dict"]["dataloader_state_dict"] = ANY
checkpoint["loops"]["fit_loop"]["epoch_loop.state_dict"]["dataloader_state_dict"] = ANY
assert state_dict == checkpoint["loops"]["fit_loop"]
trainer.fit_loop.load_state_dict(checkpoint["loops"]["fit_loop"])
# test resetting manually, we expect all `ready` counters to be reset to `completed`
trainer.fit_loop.reset()
trainer.fit_loop.epoch_loop.reset()
trainer.fit_loop.epoch_loop.batch_loop.reset()
trainer.fit_loop.epoch_loop.batch_loop.optimizer_loop.reset()
trainer.fit_loop.epoch_loop.val_loop.reset()
trainer.fit_loop.epoch_loop.val_loop.epoch_loop.reset()
epoch_progress = trainer.fit_loop.epoch_progress
assert epoch_progress.current.ready == stop_epoch
assert epoch_progress.current.completed == stop_epoch
batch_progress = trainer.fit_loop.epoch_loop.batch_progress
assert batch_progress.current.ready == be_batches_completed
assert batch_progress.current.completed == be_batches_completed
optim_progress = trainer.fit_loop.epoch_loop.batch_loop.optimizer_loop.optim_progress
assert optim_progress.optimizer.step.current.ready == be_total_opt_steps
assert optim_progress.optimizer.step.current.completed == be_total_opt_steps
assert optim_progress.optimizer.zero_grad.current.ready == be_total_zero_grad
assert optim_progress.optimizer.zero_grad.current.completed == be_total_zero_grad
state_dict = trainer.fit_loop.state_dict()
assert state_dict != checkpoint["loops"]["fit_loop"]
assert state_dict["epoch_progress"]["total"]["started"] == stop_epoch + 1
assert state_dict["epoch_progress"]["current"]["started"] == stop_epoch
@mock.patch.dict(os.environ, {"PL_FAULT_TOLERANT_TRAINING": "1"})
@pytest.mark.parametrize("n_optimizers", (1, 3, 5))
def test_loop_state_on_complete_run(n_optimizers, tmpdir):
n_epochs = 3
n_batches = 3
accumulate_grad_batches = 1
class TestModel(BoringModel):
def __init__(self):
super().__init__()
if n_optimizers > 1:
self.configure_optimizers = self.configure_optimizers_multiple
def training_step(self, batch, batch_idx, optimizer_idx=0):
return super().training_step(batch, batch_idx)
def configure_optimizers_multiple(self):
optimizers = [torch.optim.Adam(self.layer.parameters(), lr=0.1) for _ in range(n_optimizers)]
lr_scheduler_0 = torch.optim.lr_scheduler.StepLR(optimizers[0], step_size=1)
lr_scheduler_1 = torch.optim.lr_scheduler.StepLR(optimizers[1], step_size=1)
# no scheduler for optimizer_2
lr_schedulers = [lr_scheduler_0, {"scheduler": lr_scheduler_1, "interval": "step"}]
return optimizers, lr_schedulers
def train_dataloader(self):
# override to test the `is_last_batch` value
return DataLoader(RandomDataset(32, n_batches))
model = TestModel()
model.training_epoch_end = None
trainer = Trainer(
default_root_dir=tmpdir,
max_epochs=n_epochs,
limit_val_batches=0,
accumulate_grad_batches=accumulate_grad_batches,
enable_progress_bar=False,
logger=False,
)
trainer.fit(model)
assert trainer.num_training_batches == n_batches
ckpt_path = trainer.checkpoint_callback.best_model_path
assert os.path.exists(ckpt_path)
checkpoint = torch.load(ckpt_path)
n_sch_steps_total = n_epochs
n_sch_steps_current = 1
if n_optimizers > 1:
n_sch_steps_total = n_epochs + n_epochs * n_batches
n_sch_steps_current = n_batches + 1
expected = {
"state_dict": ANY,
"epoch_progress": {
"total": {
"ready": n_epochs,
"started": n_epochs,
"processed": n_epochs,
# TODO: the following "-1" offset will be fixed by
# https://github.com/PyTorchLightning/pytorch-lightning/pull/8578
"completed": n_epochs - 1,
},
"current": {
"ready": n_epochs,
"started": n_epochs,
"processed": n_epochs,
# TODO: the following "-1" offset will be fixed by
# https://github.com/PyTorchLightning/pytorch-lightning/pull/8578
"completed": n_epochs - 1,
},
},
"epoch_loop.state_dict": ANY,
"epoch_loop.batch_progress": {
"total": {
"ready": n_epochs * n_batches,
"started": n_epochs * n_batches,
"processed": n_epochs * n_batches,
"completed": n_epochs * n_batches,
},
"current": {
"ready": n_batches,
"started": n_batches,
"processed": n_batches,
"completed": n_batches,
},
"is_last_batch": True,
},
"epoch_loop.scheduler_progress": {
"total": {"ready": n_sch_steps_total, "completed": n_sch_steps_total},
"current": {"ready": n_sch_steps_current, "completed": n_sch_steps_current},
},
"epoch_loop.batch_loop.state_dict": ANY,
"epoch_loop.batch_loop.manual_loop.state_dict": ANY,
"epoch_loop.batch_loop.optimizer_loop.state_dict": {},
"epoch_loop.batch_loop.optimizer_loop.optim_progress": {
"optimizer_position": n_optimizers,
"optimizer": {
"step": {
"total": {
"ready": n_epochs * n_batches * n_optimizers,
"completed": n_epochs * n_batches * n_optimizers,
},
"current": {
"ready": n_batches * n_optimizers,
"completed": n_batches * n_optimizers,
},
},
"zero_grad": {
"total": {
"ready": n_epochs * n_batches * n_optimizers,
"started": n_epochs * n_batches * n_optimizers,
"completed": n_epochs * n_batches * n_optimizers,
},
"current": {
"ready": n_batches * n_optimizers,
"started": n_batches * n_optimizers,
"completed": n_batches * n_optimizers,
},
},
},
},
"epoch_loop.val_loop.state_dict": ANY,
"epoch_loop.val_loop.dataloader_progress": ANY,
"epoch_loop.val_loop.epoch_loop.state_dict": ANY,
"epoch_loop.val_loop.epoch_loop.batch_progress": ANY,
"epoch_loop.val_loop._results": ANY,
"epoch_loop._results": ANY,
}
assert checkpoint["loops"]["fit_loop"] == expected
@mock.patch.dict(os.environ, {"PL_FAULT_TOLERANT_TRAINING": "1"})
def test_fit_loop_reset(tmpdir):
"""Test that the reset logic in fit- and epoch loop is aware of whether the loop is restarting from a completed
loop or from a mid-epoch checkpoint."""
# generate checkpoints at end of epoch and mid-epoch
model = BoringModel()
checkpoint_callback = ModelCheckpoint(
dirpath=tmpdir,
every_n_train_steps=2,
save_top_k=-1,
)
trainer = Trainer(
default_root_dir=tmpdir,
limit_train_batches=4,
num_sanity_val_steps=0,
max_epochs=2,
callbacks=[checkpoint_callback],
logger=False,
enable_model_summary=False,
)
trainer.fit(model)
# reset state loaded from a checkpoint from mid-epoch
mid_epoch_ckpt = torch.load(str(tmpdir / "epoch=0-step=1.ckpt"))
fit_loop = trainer.fit_loop
epoch_loop = fit_loop.epoch_loop
optimizer_loop = epoch_loop.batch_loop.optimizer_loop
assert not fit_loop.restarting
assert not epoch_loop.restarting
assert not optimizer_loop.restarting
# we load exactly what was saved - no reset yet
fit_loop.load_state_dict(mid_epoch_ckpt["loops"]["fit_loop"])
# resetting from a mid-of-epoch checkpoint SHOULD NOT reset the current counters to 0
fit_loop.reset()
epoch_loop.reset()
optimizer_loop.reset()
assert fit_loop.restarting
assert fit_loop.epoch_progress.total.ready == 1
assert fit_loop.epoch_progress.total.completed == 0 # the checkpoint was saved mid epoch
assert fit_loop.epoch_progress.current.ready == 0
assert fit_loop.epoch_progress.current.completed == 0
assert epoch_loop.restarting
assert epoch_loop.batch_progress.total.ready == 2
assert epoch_loop.batch_progress.total.processed == 2
assert epoch_loop.batch_progress.total.completed == 1 # the checkpoint was saved on train_batch_end
assert epoch_loop.batch_progress.current.ready == 1 # currents get set to the completed value
assert epoch_loop.batch_progress.current.processed == 1
assert epoch_loop.batch_progress.current.completed == 1
assert optimizer_loop.restarting
assert optimizer_loop.optim_progress.optimizer_position == 1
# reset state loaded from a checkpoint from the end of an epoch
end_of_epoch_ckpt = torch.load(str(tmpdir / "epoch=0-step=3.ckpt"))
fit_loop = trainer.fit_loop
epoch_loop = fit_loop.epoch_loop
fit_loop.restarting = False
epoch_loop.restarting = False
optimizer_loop.restarting = False
# we load exactly what was saved - no reset yet
fit_loop.load_state_dict(end_of_epoch_ckpt["loops"]["fit_loop"])
# resetting from a end-of-epoch checkpoint SHOULD reset the current counters to 0
fit_loop.reset()
epoch_loop.reset()
optimizer_loop.reset()
assert fit_loop.restarting
assert fit_loop.epoch_progress.total.ready == 1
assert fit_loop.epoch_progress.total.completed == 0 # the checkpoint saves before the epoch completes
assert fit_loop.epoch_progress.current.ready == 0
assert fit_loop.epoch_progress.current.completed == 0
assert epoch_loop.restarting
assert epoch_loop.batch_progress.total.ready == 4
assert epoch_loop.batch_progress.total.processed == 4
assert epoch_loop.batch_progress.total.completed == 3 # the checkpoint was saved on train_batch_end
assert epoch_loop.batch_progress.current.ready == 3 # currents get set to the completed value
assert epoch_loop.batch_progress.current.processed == 3
assert epoch_loop.batch_progress.current.completed == 3
assert optimizer_loop.optim_progress.optimizer_position == 1
@mock.patch.dict(os.environ, {"PL_FAULT_TOLERANT_TRAINING": "1"})
@pytest.mark.parametrize(
["train_datasets", "val_datasets"],
[([RandomDataset], [RandomDataset]), ([RandomDataset], [RandomDataset, RandomDataset])],
)
@pytest.mark.parametrize("val_check_interval", [0.5, 1.0])
def test_fit_can_fail_during_validation(train_datasets, val_datasets, val_check_interval, tmpdir):
size, n_batches = 2, 4
stop_batch = 1
n_val_dataloaders = len(val_datasets)
stop_dataloader = n_val_dataloaders - 1
class TestModel(LightningModule):
def __init__(self, should_fail):
super().__init__()
self.layer = torch.nn.Linear(size, 2)
self.should_fail = should_fail
def step(self, batch):
return sum(self.layer(b).sum() for b in batch)
def training_step(self, batch, batch_idx):
return self.step(batch)
def validation_step(self, batch, batch_idx, dataloader_idx=0):
if self.should_fail and dataloader_idx == stop_dataloader and batch_idx == stop_batch:
raise CustomException
return self.step(batch)
def configure_optimizers(self):
return torch.optim.SGD(self.layer.parameters(), lr=0.1)
def train_dataloader(self):
return [DataLoader(cls(size, n_batches)) for cls in train_datasets]
def val_dataloader(self):
return [DataLoader(cls(size, n_batches)) for cls in val_datasets]
model = TestModel(False)
trainer = Trainer(
default_root_dir=tmpdir,
max_epochs=1,
val_check_interval=val_check_interval,
num_sanity_val_steps=0,
enable_progress_bar=False,
)
trainer.fit(model)
ckpt_path = os.path.join(tmpdir, ".pl_auto_save.ckpt")
assert not os.path.exists(ckpt_path), "Shouldn't have failed"
state_dict = trainer.fit_loop.state_dict()
expected_global_step = trainer.global_step
assert state_dict["epoch_loop.batch_progress"] == {
"total": {"ready": n_batches, "started": n_batches, "processed": n_batches, "completed": n_batches},
"current": {"ready": n_batches, "started": n_batches, "processed": n_batches, "completed": n_batches},
"is_last_batch": True,
}
val_per_epoch = int(1 // val_check_interval)
assert state_dict["epoch_loop.val_loop.dataloader_progress"] == {
"total": {"ready": n_val_dataloaders * val_per_epoch, "completed": n_val_dataloaders * val_per_epoch},
"current": {"ready": n_val_dataloaders, "completed": n_val_dataloaders},
}
assert state_dict["epoch_loop.val_loop.epoch_loop.batch_progress"] == {
"total": {
"ready": n_val_dataloaders * val_per_epoch * n_batches,
"started": n_val_dataloaders * val_per_epoch * n_batches,
"processed": n_val_dataloaders * val_per_epoch * n_batches,
"completed": n_val_dataloaders * val_per_epoch * n_batches,
},
"current": {"ready": n_batches, "completed": n_batches, "started": n_batches, "processed": n_batches},
"is_last_batch": True,
}
model = TestModel(True)
trainer = Trainer(
default_root_dir=tmpdir,
max_epochs=1,
val_check_interval=val_check_interval,
num_sanity_val_steps=0,
enable_progress_bar=False,
)
with pytest.raises(CustomException):
# will stop during validation
trainer.fit(model)
assert os.path.exists(ckpt_path)
checkpoint = torch.load(ckpt_path)["loops"]["fit_loop"]
per_val_train_batches = int(n_batches * val_check_interval)
assert checkpoint["epoch_loop.batch_progress"] == {
"total": {
"ready": per_val_train_batches,
"started": per_val_train_batches,
"processed": per_val_train_batches,
"completed": per_val_train_batches,
},
"current": {
"ready": per_val_train_batches,
"started": per_val_train_batches,
"processed": per_val_train_batches,
"completed": per_val_train_batches,
},
"is_last_batch": val_check_interval == 1,
}
val_batch_progress = "epoch_loop.val_loop.epoch_loop.batch_progress"
# "nb_": non-breaking
nb_total_val_batch = stop_dataloader * n_batches
assert checkpoint[val_batch_progress] == {
"total": {
"ready": nb_total_val_batch + stop_batch + 1,
"started": nb_total_val_batch + stop_batch + 1,
"processed": nb_total_val_batch + stop_batch,
"completed": nb_total_val_batch + stop_batch,
},
"current": {
"ready": stop_batch + 1,
"started": stop_batch + 1,
"processed": stop_batch,
"completed": stop_batch,
},
"is_last_batch": False,
}
model = TestModel(False)
trainer = Trainer(
default_root_dir=tmpdir,
max_epochs=1,
val_check_interval=val_check_interval,
num_sanity_val_steps=0,
enable_progress_bar=False,
)
trainer.fit(model, ckpt_path=ckpt_path)
# TODO: -1 because there's a bug where global step is off by one on reload
assert trainer.global_step - 1 == expected_global_step
state_dict_after_restart = trainer.fit_loop.state_dict()
# should get the same values as in the run that did not fail
# totals are increased by 1 (the failed batch which never completed)
expected = state_dict.copy()
# TODO: `is_last_batch` is not correct on reload, the next line should not be necessary
expected["epoch_loop.batch_progress"]["is_last_batch"] = val_check_interval == 1.0
assert state_dict_after_restart["epoch_loop.batch_progress"] == expected["epoch_loop.batch_progress"]
val_dl_progress = "epoch_loop.val_loop.dataloader_progress"
expected[val_dl_progress]["total"]["ready"] += 1
assert state_dict_after_restart[val_dl_progress] == expected[val_dl_progress]
expected[val_batch_progress]["total"]["ready"] += 1
expected[val_batch_progress]["total"]["started"] += 1
assert state_dict_after_restart[val_batch_progress] == expected[val_batch_progress]
@RunIf(min_torch="1.8.0")
@pytest.mark.parametrize("should_fail", [False, True])
@pytest.mark.parametrize("persistent_workers", [pytest.param(False, marks=RunIf(slow=True)), True])
def test_workers_are_shutdown(tmpdir, should_fail, persistent_workers):
# `num_workers == 1` uses `_MultiProcessingDataLoaderIter`
# `persistent_workers` makes sure `self._iterator` gets set on the `DataLoader` instance
class _TestMultiProcessingDataLoaderIter(_MultiProcessingDataLoaderIter):
def __init__(self, *args, dataloader, **kwargs):
super().__init__(*args, **kwargs)
self.dataloader = dataloader
def _shutdown_workers(self):
self.dataloader.count_shutdown_workers += 1
super()._shutdown_workers()
class TestDataLoader(DataLoader):
def __init__(self, *args, **kwargs):
super().__init__(*args, **kwargs)
self.count_shutdown_workers = 0
def _get_iterator(self):
if self.num_workers == 0:
return super()._get_iterator()
else:
self.check_worker_number_rationality()
return _TestMultiProcessingDataLoaderIter(self, dataloader=self)
train_dataloader = TestDataLoader(RandomDataset(32, 64), num_workers=1, persistent_workers=persistent_workers)
val_dataloader = TestDataLoader(RandomDataset(32, 64), num_workers=1, persistent_workers=persistent_workers)
class TestCallback(Callback):
def on_train_epoch_end(self, trainer, *_):
if trainer.current_epoch == 1:
raise CustomException
max_epochs = 3
model = BoringModel()
trainer = Trainer(
default_root_dir=tmpdir,
limit_train_batches=2,
limit_val_batches=2,
max_epochs=max_epochs,
callbacks=TestCallback() if should_fail else None,
)
if should_fail:
with pytest.raises(CustomException):
trainer.fit(model, train_dataloader, val_dataloader)
else:
trainer.fit(model, train_dataloader, val_dataloader)
assert train_dataloader.count_shutdown_workers == 2 if should_fail else (2 if persistent_workers else max_epochs)
# on sanity checking end, the workers are being deleted too.
assert val_dataloader.count_shutdown_workers == 2 if persistent_workers else (3 if should_fail else max_epochs + 1)
assert train_dataloader._iterator is None
assert val_dataloader._iterator is None