-
Notifications
You must be signed in to change notification settings - Fork 3.3k
/
single_device.py
94 lines (76 loc) · 3.18 KB
/
single_device.py
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
# 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 typing import Any, Optional, Union
import torch
import pytorch_lightning as pl
from pytorch_lightning.plugins.io.checkpoint_plugin import CheckpointIO
from pytorch_lightning.plugins.precision import PrecisionPlugin
from pytorch_lightning.plugins.training_type.training_type_plugin import Strategy
from pytorch_lightning.utilities import _XLA_AVAILABLE
class SingleDeviceStrategy(Strategy):
"""Strategy that handles communication on a single device."""
def __init__(
self,
device: torch.device,
accelerator: Optional["pl.accelerators.accelerator.Accelerator"] = None,
checkpoint_io: Optional[CheckpointIO] = None,
precision_plugin: Optional[PrecisionPlugin] = None,
):
super().__init__(accelerator=accelerator, checkpoint_io=checkpoint_io, precision_plugin=precision_plugin)
self.device: torch.device = device
self.global_rank = 0
self.local_rank = 0
self.world_size = 1
@property
def on_tpu(self) -> bool:
return self.root_device.type == "xla" and _XLA_AVAILABLE
@property
def on_gpu(self) -> bool:
return self.root_device.type == "cuda" and torch.cuda.is_available()
def reduce(self, tensor: Union[Any, torch.Tensor], *args: Any, **kwargs: Any) -> Union[Any, torch.Tensor]:
"""Reduces a tensor from several distributed processes to one aggregated tensor. As this plugin only
operates with a single device, the reduction is simply the identity.
Args:
tensor: the tensor to sync and reduce
*args: ignored
**kwargs: ignored
Return:
the unmodified input as reduction is not needed for single process operation
"""
return tensor
def all_gather(self, tensor: torch.Tensor, group: Optional[Any] = None, sync_grads: bool = False) -> torch.Tensor:
"""Perform a all_gather on all processes."""
return tensor
@property
def root_device(self) -> torch.device:
return self.device
def model_to_device(self) -> None:
self.model.to(self.root_device)
def setup(self, trainer: "pl.Trainer") -> None:
self.model_to_device()
super().setup(trainer)
@property
def is_global_zero(self) -> bool:
return True
def barrier(self, *args, **kwargs) -> None:
pass
def broadcast(self, obj: object, src: int = 0) -> object:
return obj
def teardown(self) -> None:
super().teardown()
if self.on_gpu:
# GPU teardown
self.lightning_module.cpu()
# clean up memory
torch.cuda.empty_cache()