Source code for pytorch_lightning.strategies.sharded
# 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 contextlib import contextmanager
from typing import Dict, Generator, List, Optional, Tuple, Union
from torch import Tensor
from torch.nn import Module
from torch.optim import Optimizer
import pytorch_lightning as pl
from pytorch_lightning.core.optimizer import LightningOptimizer
from pytorch_lightning.overrides.fairscale import _FAIRSCALE_AVAILABLE
from pytorch_lightning.strategies.ddp import DDPStrategy
from pytorch_lightning.trainer.states import TrainerFn
from pytorch_lightning.utilities.enums import PrecisionType
from pytorch_lightning.utilities.exceptions import MisconfigurationException
from pytorch_lightning.utilities.imports import _FAIRSCALE_OSS_FP16_BROADCAST_AVAILABLE
from pytorch_lightning.utilities.optimizer import optimizers_to_device
from pytorch_lightning.utilities.rank_zero import rank_zero_only
if _FAIRSCALE_AVAILABLE:
from fairscale.nn.data_parallel.sharded_ddp import ShardedDataParallel
from fairscale.optim import OSS
from pytorch_lightning.overrides.fairscale import LightningShardedDataParallel, unwrap_lightning_module_sharded
else:
OSS = ShardedDataParallel = object
[docs]class DDPShardedStrategy(DDPStrategy):
"""Optimizer and gradient sharded training provided by FairScale."""
strategy_name = "ddp_sharded"
_REDUCE_BUFFER_SIZE_DEFAULT: int = 2**23 # 8M
[docs] def setup(self, trainer: "pl.Trainer") -> None:
# share ddp pids to all processes
self._rank_0_will_call_children_scripts = self.broadcast(self._rank_0_will_call_children_scripts)
if self._should_run_deadlock_detection():
self._share_information_to_prevent_deadlock()
self.accelerator.setup(trainer)
# move the model to the correct device
self.model_to_device()
# skip wrapping the model if we are not fitting as no gradients need to be exchanged
trainer_fn = trainer.state.fn
if trainer_fn == TrainerFn.FITTING:
if self._layer_sync:
self.model = self._layer_sync.apply(self.model)
self.setup_precision_plugin()
if trainer_fn == TrainerFn.FITTING:
self.configure_ddp()
def configure_ddp(self) -> None:
self._set_ddp_kwargs()
self.setup_optimizers(self.model.trainer)
self.model, self.optimizers = self._setup_model_and_optimizers(
model=LightningShardedDataParallel(self.model),
optimizers=self.optimizers,
)
optimizers_to_device(self.optimizers, self.root_device)
def _set_ddp_kwargs(self) -> None:
if "reduce_buffer_size" not in self._ddp_kwargs:
# For multi-node training, enabling bucketing will improve performance.
self._ddp_kwargs["reduce_buffer_size"] = self._REDUCE_BUFFER_SIZE_DEFAULT if self.num_nodes > 1 else 0
def _setup_model_and_optimizers(self, model: Module, optimizers: List[Optimizer]) -> Tuple[Module, List[Optimizer]]:
"""Wraps the model and optimizers with fairscale components.
Return:
The model wrapped into a :class:`~fairscale.nn.data_parallel.ShardedDataParallel` module
and a list of optimizer wrapped in :class:~`fairscale.optim.OSS`.
"""
optimizers = self._wrap_optimizers(optimizers)
model = ShardedDataParallel(model, sharded_optimizer=optimizers, **self._ddp_kwargs)
return model, optimizers
def _wrap_optimizers(self, optimizers: List[Optimizer]) -> List["OSS"]:
if self.model is not None and self.model.trainer.state.fn != TrainerFn.FITTING:
return optimizers
return self._reinit_optimizers_with_oss(optimizers)
def _reinit_optimizers_with_oss(self, optimizers: List[Union[Optimizer, LightningOptimizer]]) -> List["OSS"]:
for x, optimizer in enumerate(optimizers):
if isinstance(optimizer, LightningOptimizer):
optimizer = optimizer._optimizer
if not isinstance(optimizer, OSS):
optim_class = type(optimizer)
zero_optimizer = OSS(params=optimizer.param_groups, optim=optim_class, **optimizer.defaults)
if _FAIRSCALE_OSS_FP16_BROADCAST_AVAILABLE:
is_fp16 = self.precision_plugin.precision in (PrecisionType.MIXED, PrecisionType.HALF)
# For multi-node training, compressing the model shards in fp16 before broadcasting
# improves performance. When using PyTorch AMP, it will not degrade
# the model performance.
zero_optimizer.broadcast_fp16 = is_fp16 and self.num_nodes > 1
optimizers[x] = zero_optimizer
del optimizer
return optimizers
[docs] def optimizer_state(self, optimizer: "OSS") -> Optional[dict]:
if isinstance(optimizer, LightningOptimizer):
optimizer = optimizer._optimizer
optimizer.consolidate_state_dict()
return self._optim_state_dict(optimizer)
@rank_zero_only
def _optim_state_dict(self, optimizer):
"""
Retrieves state dict only on rank 0, which contains the entire optimizer state after calling
:meth:`consolidate_state_dict`.
"""
return optimizer.state_dict()
@property
def lightning_module(self) -> Optional["pl.LightningModule"]:
if not _FAIRSCALE_AVAILABLE: # pragma: no cover
raise MisconfigurationException(
"`DDPShardedStrategy` requires `fairscale` to be installed."
" Install it by running `pip install fairscale`."
)
return unwrap_lightning_module_sharded(self.model) if self.model is not None else None
[docs] @contextmanager
def block_backward_sync(self) -> Generator:
"""Blocks syncing gradients behaviour on backwards pass.
This is useful for skipping sync when accumulating gradients, reducing communication overhead
Returns: context manager with sync behaviour off
"""
if isinstance(self.model, ShardedDataParallel):
with self.model.no_sync():
yield None
else:
yield None
def post_training_step(self):
pass
@classmethod
def register_strategies(cls, strategy_registry: Dict) -> None:
strategy_registry.register(
"ddp_sharded_find_unused_parameters_false",
cls,
description="DDP Sharded Strategy with `find_unused_parameters` as False",
find_unused_parameters=False,
)
strategy_registry.register(
cls.strategy_name,
cls,
description=f"{cls.__class__.__name__}",
)
