File size: 3,587 Bytes
9c6594c |
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 95 96 97 98 99 100 101 |
# Copyright The Lightning AI 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 AbstractContextManager, nullcontext
from typing import TYPE_CHECKING, Any, Literal
import torch
from lightning_utilities.core.apply_func import apply_to_collection
from torch import Tensor
from torch.nn import Module
from typing_extensions import get_args, override
from lightning.fabric.plugins.precision.precision import Precision
from lightning.fabric.plugins.precision.utils import _convert_fp_tensor, _DtypeContextManager
from lightning.fabric.utilities.types import Steppable
if TYPE_CHECKING:
from deepspeed import DeepSpeedEngine
_PRECISION_INPUT = Literal["32-true", "16-true", "bf16-true", "16-mixed", "bf16-mixed"]
class DeepSpeedPrecision(Precision):
"""Precision plugin for DeepSpeed integration.
Args:
precision: Full precision (32-true), half precision (16-true, bf16-true) or
mixed precision (16-mixed, bf16-mixed).
Raises:
ValueError:
If unsupported ``precision`` is provided.
"""
def __init__(self, precision: _PRECISION_INPUT) -> None:
supported_precision = get_args(_PRECISION_INPUT)
if precision not in supported_precision:
raise ValueError(
f"`precision={precision!r})` is not supported in DeepSpeed."
f" `precision` must be one of: {supported_precision}."
)
self.precision = precision
precision_to_type = {
"bf16-mixed": torch.bfloat16,
"16-mixed": torch.float16,
"bf16-true": torch.bfloat16,
"16-true": torch.float16,
"32-true": torch.float32,
}
self._desired_dtype = precision_to_type[self.precision]
@override
def convert_module(self, module: Module) -> Module:
if "true" in self.precision:
return module.to(dtype=self._desired_dtype)
return module
@override
def tensor_init_context(self) -> AbstractContextManager:
if "true" not in self.precision:
return nullcontext()
return _DtypeContextManager(self._desired_dtype)
@override
def module_init_context(self) -> AbstractContextManager:
return self.tensor_init_context()
@override
def convert_input(self, data: Any) -> Any:
return apply_to_collection(data, function=_convert_fp_tensor, dtype=Tensor, dst_type=self._desired_dtype)
@override
def convert_output(self, data: Any) -> Any:
return apply_to_collection(data, function=_convert_fp_tensor, dtype=Tensor, dst_type=torch.get_default_dtype())
@override
def backward(self, tensor: Tensor, model: "DeepSpeedEngine", *args: Any, **kwargs: Any) -> None:
"""Performs back-propagation using DeepSpeed's engine."""
model.backward(tensor, *args, **kwargs)
@override
def optimizer_step(
self,
optimizer: Steppable,
**kwargs: Any,
) -> Any:
# DeepSpeed handles the optimizer step internally
return optimizer.step(**kwargs)
|