lightning/pytorch_lightning/loops/optimization/manual_loop.py

171 lines
7.0 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.
from dataclasses import dataclass, field
from typing import Any, Dict, Optional
from torch import Tensor
from pytorch_lightning.loops import Loop
from pytorch_lightning.loops.optimization.closure import OutputResult
from pytorch_lightning.loops.utilities import (
_build_training_step_kwargs,
_check_training_step_output,
_extract_hiddens,
check_finite_loss,
)
from pytorch_lightning.utilities.apply_func import apply_to_collection
from pytorch_lightning.utilities.memory import recursive_detach
from pytorch_lightning.utilities.types import STEP_OUTPUT
from pytorch_lightning.utilities.warnings import rank_zero_deprecation
@dataclass
class ManualResult(OutputResult):
"""A container to hold the result returned by the ``ManualLoop``.
It is created from the output of :meth:`~pytorch_lightning.core.lightning.LightningModule.training_step`.
Attributes:
closure_loss: The loss with a graph attached.
loss: A detached copy of the closure loss.
extra: Any keys other than the loss returned.
"""
closure_loss: Optional[Tensor]
loss: Optional[Tensor] = field(init=False, default=None)
extra: Dict[str, Tensor] = field(default_factory=dict)
def __post_init__(self) -> None:
# TODO: remove with the deprecation removal in v1.6
self._check_extra_detach_deprecation(self.extra)
self.extra = recursive_detach(self.extra)
self._clone_loss()
def _clone_loss(self) -> None:
if self.closure_loss is not None:
# the loss will get scaled for amp. avoid any modifications to it
self.loss = self.closure_loss.detach().clone()
@classmethod
def from_training_step_output(
cls, training_step_output: Optional[STEP_OUTPUT], normalize: int = 1
) -> "ManualResult":
closure_loss, extra = None, {}
if isinstance(training_step_output, dict):
# this should not modify the `training_step_output`, as the user could be using it after `training_step_end`
closure_loss = training_step_output.get("loss")
extra = {k: v for k, v in training_step_output.items() if k not in ("loss", "hiddens")}
elif isinstance(training_step_output, Tensor):
closure_loss = training_step_output
if closure_loss is not None:
# accumulate the loss. If ``accumulate_grad_batches == 1``, no effect
closure_loss /= normalize
return cls(closure_loss, extra=extra)
@staticmethod
def _check_extra_detach_deprecation(extra: Dict[str, Any]) -> None:
def check_fn(v: Tensor) -> Tensor:
if v.grad_fn is not None:
rank_zero_deprecation(
f"One of the returned values {set(extra.keys())} has a `grad_fn`. We will detach it automatically"
" but this behaviour will change in v1.6. Please detach it manually:"
" `return {'loss': ..., 'something': something.detach()}`"
)
return v
apply_to_collection(extra, Tensor, check_fn)
def drop_closure_loss(self) -> "ManualResult":
"""Return itself without the closure loss which could have a `grad_fn`"""
self.closure_loss = None
return self
class ManualOptimization(Loop):
"""A special loop implementing what is known in Lightning as Manual Optimization where the optimization happens
entirely in the :meth:`~pytorch_lightning.core.lightning.LightningModule.training_step` and therefore the user
is responsible for back-propagating gradients and making calls to the optimizers.
This loop is a trivial case because it performs only a single iteration (calling directly into the module's
:meth:`~pytorch_lightning.core.lightning.LightningModule.training_step`) and passing through the output(s).
"""
def __init__(self) -> None:
super().__init__()
self._done: bool = False
self._hiddens: Optional[Any] = None
self._output: Optional[ManualResult] = None
@property
def done(self) -> bool:
return self._done
def reset(self) -> None:
self._done = False
def advance(self, batch: Any, batch_idx: int) -> None: # type: ignore[override]
"""Performs the training step for manual optimization.
Args:
batch: the current tbptt split of the current batch
batch_idx: the index of the current batch
"""
assert self.trainer is not None
lightning_module = self.trainer.lightning_module
with self.trainer.profiler.profile("model_forward"):
step_kwargs = _build_training_step_kwargs(
lightning_module, self.trainer.optimizers, batch, batch_idx, opt_idx=None, hiddens=self._hiddens
)
# manually capture logged metrics
lightning_module._current_fx_name = "training_step"
with self.trainer.profiler.profile("training_step"):
training_step_output = self.trainer.accelerator.training_step(step_kwargs)
self.trainer.accelerator.post_training_step()
del step_kwargs
training_step_output = self.trainer.call_hook("training_step_end", training_step_output)
_check_training_step_output(lightning_module, training_step_output)
self._hiddens = _extract_hiddens(training_step_output, lightning_module.truncated_bptt_steps)
result = ManualResult.from_training_step_output(training_step_output, self.trainer.accumulate_grad_batches)
if self.trainer.terminate_on_nan:
check_finite_loss(result.closure_loss)
if self.trainer.move_metrics_to_cpu:
# hiddens and the training step output are not moved as they are not considered "metrics"
# the user might need them on the correct device for an operation in `training_epoch_end`
assert self.trainer._results is not None
self.trainer._results.cpu()
self._done = True
self._output = result
def on_run_end(self) -> Optional[ManualResult]:
"""Returns the result of this loop, i.e., the post-processed outputs from the training step."""
output, self._output = self._output, None # free memory
# #9052 added support for raising `StopIteration` in the `training_step`. If that happens, then `advance`
# doesn't finish and `self._output` stays as `None`. If #9415 happens then this would always return a result
return output