Shortcuts

Source code for pytorch_lightning.strategies.parallel

# 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 abc import ABC, abstractmethod
from contextlib import contextmanager
from typing import Any, Dict, Generator, List, Optional

import torch
from torch import Tensor

import pytorch_lightning as pl
from lightning_lite.plugins import CheckpointIO, ClusterEnvironment
from lightning_lite.utilities.distributed import _all_gather_ddp_if_available, ReduceOp
from pytorch_lightning.plugins import LayerSync
from pytorch_lightning.plugins.precision import PrecisionPlugin
from pytorch_lightning.strategies.strategy import Strategy


[docs]class ParallelStrategy(Strategy, ABC): """Plugin for training with multiple processes in parallel.""" def __init__( self, accelerator: Optional["pl.accelerators.Accelerator"] = None, parallel_devices: Optional[List[torch.device]] = None, cluster_environment: Optional[ClusterEnvironment] = None, checkpoint_io: Optional[CheckpointIO] = None, precision_plugin: Optional[PrecisionPlugin] = None, ): super().__init__(accelerator=accelerator, checkpoint_io=checkpoint_io, precision_plugin=precision_plugin) self.parallel_devices = parallel_devices self.cluster_environment: Optional[ClusterEnvironment] = cluster_environment self._layer_sync: Optional[LayerSync] = None @property @abstractmethod def root_device(self) -> torch.device: """Return the root device.""" @property def global_rank(self) -> int: return self.cluster_environment.global_rank() if self.cluster_environment is not None else 0 @property def local_rank(self) -> int: return self.cluster_environment.local_rank() if self.cluster_environment is not None else 0 @property def node_rank(self) -> int: return self.cluster_environment.node_rank() if self.cluster_environment is not None else 0 @property def world_size(self) -> int: return self.cluster_environment.world_size() if self.cluster_environment is not None else 1 @property def is_global_zero(self) -> bool: return self.global_rank == 0 @property def parallel_devices(self) -> Optional[List[torch.device]]: return self._parallel_devices @parallel_devices.setter def parallel_devices(self, parallel_devices: Optional[List[torch.device]]) -> None: self._parallel_devices = parallel_devices @property def distributed_sampler_kwargs(self) -> Dict[str, Any]: return dict( num_replicas=len(self.parallel_devices) if self.parallel_devices is not None else 0, rank=self.global_rank, )
[docs] def reconciliate_processes(self, trace: str) -> None: """Function to re-conciliate processes on failure."""
[docs] def all_gather(self, tensor: Tensor, group: Optional[Any] = None, sync_grads: bool = False) -> Tensor: """Perform a all_gather on all processes.""" return _all_gather_ddp_if_available(tensor, group=group, sync_grads=sync_grads)
[docs] def reduce_boolean_decision(self, decision: bool, all: bool = True) -> bool: """Reduces a boolean decision over distributed processes. By default is analagous to ``all`` from the standard library, returning ``True`` only if all input decisions evaluate to ``True``. If ``all`` is set to ``False``, it behaves like ``any`` instead. Args: decision: A single input decision. all: Whether to logically emulate ``all`` or ``any``. Defaults to True. Returns: bool: The reduced boolean decision. """ decision = torch.tensor(int(decision), device=self.root_device) decision = self.reduce(decision, reduce_op=ReduceOp.SUM) decision = bool(decision == self.world_size) if all else bool(decision) return decision
[docs] @contextmanager def block_backward_sync(self) -> Generator: """Blocks ddp sync 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, pl.utilities.types.DistributedDataParallel): with self.model.no_sync(): yield None else: yield None
[docs] def teardown(self) -> None: assert self.cluster_environment is not None self.cluster_environment.teardown() super().teardown()

© Copyright Copyright (c) 2018-2022, Lightning AI et al... Revision f4fcad36.

Built with Sphinx using a theme provided by Read the Docs.
Read the Docs v: latest
Versions
latest
stable
1.8.3post1
1.8.3.post0
1.8.3
1.8.2
1.8.1
1.8.0.post1
1.8.0
1.7.7
1.7.6
1.7.5
1.7.4
1.7.3
1.7.2
1.7.1
1.7.0
1.6.5
1.6.4
1.6.3
1.6.2
1.6.1
1.6.0
1.5.10
1.5.9
1.5.8
1.5.7
1.5.6
1.5.5
1.5.4
1.5.3
1.5.2
1.5.1
1.5.0
1.4.9
1.4.8
1.4.7
1.4.6
1.4.5
1.4.4
1.4.3
1.4.2
1.4.1
1.4.0
1.3.8
1.3.7
1.3.6
1.3.5
1.3.4
1.3.3
1.3.2
1.3.1
1.3.0
1.2.10
1.2.8
1.2.7
1.2.6
1.2.5
1.2.4
1.2.3
1.2.2
1.2.1
1.2.0
1.1.8
1.1.7
1.1.6
1.1.5
1.1.4
1.1.3
1.1.2
1.1.1
1.1.0
1.0.8
1.0.7
1.0.6
1.0.5
1.0.4
1.0.3
1.0.2
1.0.1
1.0.0
0.10.0
0.9.0
0.8.5
0.8.4
0.8.3
0.8.2
0.8.1
0.8.0
0.7.6
0.7.5
0.7.4
0.7.3
0.7.2
0.7.1
0.7.0
0.6.0
0.5.3
0.4.9
Downloads
html
On Read the Docs
Project Home
Builds

Free document hosting provided by Read the Docs.