Source code for pytorch_lightning.strategies.hpu_parallel

# 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.
import logging
import os
from typing import Any, Callable, Dict, List, Optional, Union

import torch.distributed
from torch.nn import Module
from torch.optim.optimizer import Optimizer

import pytorch_lightning as pl
from lightning_fabric.plugins import CheckpointIO, ClusterEnvironment
from lightning_fabric.utilities.distributed import group as _group
from pytorch_lightning.accelerators.hpu import _HPU_AVAILABLE
from pytorch_lightning.overrides import LightningDistributedModule
from pytorch_lightning.plugins.io.hpu_plugin import HPUCheckpointIO
from pytorch_lightning.plugins.io.wrapper import _WrappingCheckpointIO
from pytorch_lightning.plugins.precision import PrecisionPlugin
from pytorch_lightning.strategies.ddp import DDPStrategy
from pytorch_lightning.utilities.exceptions import MisconfigurationException
from pytorch_lightning.utilities.imports import _TORCH_LESSER_EQUAL_1_10_2
from pytorch_lightning.utilities.types import STEP_OUTPUT

if _HPU_AVAILABLE:
    import habana_frameworks.torch.core as htcore
    import habana_frameworks.torch.distributed.hccl  # noqa: F401

log = logging.getLogger(__name__)


[docs]class HPUParallelStrategy(DDPStrategy):
    """Strategy for distributed training on multiple HPU devices."""

    strategy_name = "hpu_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,
        ddp_comm_state: Optional[object] = None,
        ddp_comm_hook: Optional[Callable] = None,
        ddp_comm_wrapper: Optional[Callable] = None,
        model_averaging_period: Optional[int] = None,
        process_group_backend: Optional[str] = "hccl",
        **kwargs: Any,
    ) -> None:

        if not _HPU_AVAILABLE:
            raise MisconfigurationException("`HPUParallelStrategy` requires HPU devices to run")

        super().__init__(
            accelerator=accelerator,
            parallel_devices=parallel_devices,
            cluster_environment=cluster_environment,
            checkpoint_io=checkpoint_io,
            precision_plugin=precision_plugin,
            ddp_comm_state=ddp_comm_state,
            ddp_comm_hook=ddp_comm_hook,
            ddp_comm_wrapper=ddp_comm_wrapper,
            model_averaging_period=model_averaging_period,
            process_group_backend=process_group_backend,
            **kwargs,
        )

    @property
    def checkpoint_io(self) -> CheckpointIO:
        if self._checkpoint_io is None:
            self._checkpoint_io = HPUCheckpointIO()
        elif isinstance(self._checkpoint_io, _WrappingCheckpointIO):
            self._checkpoint_io.checkpoint_io = HPUCheckpointIO()

        return self._checkpoint_io

    @checkpoint_io.setter
    def checkpoint_io(self, io: Optional[CheckpointIO]) -> None:
        self._checkpoint_io = io

[docs]    def setup_environment(self) -> None:

        os.environ["ID"] = str(self.local_rank)
        if self._process_group_backend == "hccl":
            # this env is used in overrides to check the backend initiated
            os.environ["HCCL_DISTRIBUTED_BACKEND"] = str(1)
        super().setup_environment()

    def determine_ddp_device_ids(self) -> None:
        return None

    def _pre_configure_ddp(self) -> None:
        # if unset, default `find_unused_parameters` `True`
        # Many models require setting this parameter to True, as there are corner cases
        # when not all parameter backward hooks are fired by the autograd engine even if require_grad is set to True.
        # This flag does come with a performance hit, so it is suggested to disable in cases where it is possible.
        self._ddp_kwargs["find_unused_parameters"] = self._ddp_kwargs.get("find_unused_parameters", True)

        self._static_graph = False
        static_graph = self._ddp_kwargs.get("static_graph")
        if static_graph:
            # when _set_static_graph() is called find_unused_parameters does not have any significance.
            # Resetting the value of find_unused_parameters to False which is the default value to DDP
            self._ddp_kwargs["find_unused_parameters"] = False
            self._static_graph = True
        if static_graph is not None:
            # DDP does not accept static_graph as a parameter, hence removing it from the list
            del self._ddp_kwargs["static_graph"]

    def configure_ddp(self) -> None:
        # DDP does not accept static graph as param with torch < 1.11
        if _TORCH_LESSER_EQUAL_1_10_2:
            log.detail(f"{self.__class__.__name__}: configuring DistributedDataParallel")
            self._pre_configure_ddp()
            self.model = self._setup_model(LightningDistributedModule(self.model))  # type: ignore
            if self.root_device.type == "hpu" and self._static_graph:
                self._model._set_static_graph()  # type: ignore
            self._register_ddp_hooks()
        else:
            super().configure_ddp()

[docs]    def broadcast(self, obj: object, src: int = 0) -> object:  # type: ignore
        obj = [obj]
        if self.global_rank != src:
            obj = [None]

        _hpu_broadcast_object_list(obj, src, group=_group.WORLD)
        return obj[0]

    def on_after_backward(self) -> None:
        # Break lazy accumulation of graph after fwd+bwd
        htcore.mark_step()

[docs]    def optimizer_step(
        self,
        optimizer: Optimizer,
        opt_idx: int,
        closure: Callable[[], Any],
        model: Optional[Union["pl.LightningModule", Module]] = None,
        **kwargs: Any,
    ) -> Any:
        optimizer_output = super().optimizer_step(optimizer, opt_idx, closure, model, **kwargs)
        # Break lazy accumulation of graph after optimizer
        htcore.mark_step()
        return optimizer_output

    def validation_step_end(self, step_output: STEP_OUTPUT) -> STEP_OUTPUT:
        # Break lazy accumulation of graph after every step
        htcore.mark_step()
        return step_output

    def test_step_end(self, step_output: STEP_OUTPUT) -> STEP_OUTPUT:
        # Break lazy accumulation of graph after every step
        htcore.mark_step()
        return step_output

    @classmethod
    def register_strategies(cls, strategy_registry: Dict) -> None:
        strategy_registry.register(
            cls.strategy_name,
            cls,
            description=f"{cls.__class__.__name__}",
        )

[docs]    def teardown(self) -> None:
        super().teardown()
        # Was set to local rank
        os.environ.pop("ID", None)
        os.environ.pop("HCCL_DISTRIBUTED_BACKEND", None)


# The code underneath is taken from PyTorch `torch/distributed/distributed_c10d.py`
# the distributed backend and tensor type updates for habana backend is done here before broadcast
def _hpu_broadcast_object_list(object_list, src=0, group=None, device=None):  # type: ignore
    from torch.distributed import _rank_not_in_group, Backend, broadcast, get_backend, get_rank
    from torch.distributed.distributed_c10d import _object_to_tensor, _tensor_to_object

    if _rank_not_in_group(group):
        return

    my_rank = get_rank()
    # Serialize object_list elements to tensors on src rank.
    if my_rank == src:
        tensor_list, size_list = zip(*[_object_to_tensor(obj, device) for obj in object_list])
        object_sizes_tensor = torch.cat(size_list)
    else:
        object_sizes_tensor = torch.empty(len(object_list), dtype=torch.long)

    # Current device selection.
    # To preserve backwards compatibility, ``device`` is default to ``None``
    # in which case we run current logic of device selection, i.e.
    # ``current_device`` is CUDA if backend is NCCL otherwise CPU device. In the
    # case it is not ``None`` we move the size and object tensors to be
    # broadcasted to this device.
    group_backend = get_backend(group)
    is_nccl_backend = group_backend == Backend.NCCL
    is_hpu_backend = os.environ.get("HCCL_DISTRIBUTED_BACKEND") == "1"
    if device is not None:
        if is_nccl_backend and device.type != "cuda":
            raise ValueError("device type must be cuda for nccl backend")
        current_device = device
    else:
        current_device = torch.device("cpu")
        if is_nccl_backend:
            # See note about using torch.cuda.current_device() here in
            # docstring. We cannot simply use my_rank since rank == device is
            # not necessarily true.
            current_device = torch.device("cuda", torch.cuda.current_device())
    if is_nccl_backend:
        object_sizes_tensor = object_sizes_tensor.to(current_device)

    elif is_hpu_backend:
        current_device = torch.device("hpu")
        # Workaround: HPU doesn't not support long tensors for collectives
        if (object_sizes_tensor.type() == "torch.LongTensor") or (object_sizes_tensor.type() == "torch.hpu.LongTensor"):
            object_sizes_tensor = object_sizes_tensor.int()
        else:
            print("unhandled hpu object_sizes_tensor type :: ", object_sizes_tensor.type())
        object_sizes_tensor = object_sizes_tensor.to(current_device)

    # Broadcast object sizes
    broadcast(object_sizes_tensor, src=src, group=group)

    # Concatenate and broadcast serialized object tensors
    if my_rank == src:
        object_tensor = torch.cat(tensor_list)
    else:
        object_tensor = torch.empty(
            torch.sum(object_sizes_tensor).int().item(),
            dtype=torch.uint8,
        )

    if is_nccl_backend or is_hpu_backend:
        object_tensor = object_tensor.to(current_device)

    broadcast(object_tensor, src=src, group=group)
    # Deserialize objects using their stored sizes.
    offset = 0
    if my_rank != src:
        for i, obj_size in enumerate(object_sizes_tensor):
            obj_view = object_tensor[offset : offset + obj_size]
            obj_view = obj_view.type(torch.uint8)
            if obj_view.device != torch.device("cpu"):
                obj_view = obj_view.cpu()
            offset += obj_size
            object_list[i] = _tensor_to_object(obj_view, obj_size)