DDP communication hook is a generic interface to control how to communicate gradients across workers by overriding the vanilla allreduce in DistributedDataParallel. A few built-in communication hooks are provided, and users can easily apply any of these hooks to optimize communication. Besides, the hook interface can also support user-defined communication strategies for more advanced use cases.
To use a communication hook, the user just needs to let the DDP model register the hook before the training loop as below.
:func:`torch.nn.parallel.DistributedDataParallel.register_comm_hook`
Communication hook provides a flexible way to allreduce gradients. Therefore, it mainly operates on the gradients on each replica before allreduce, which are bucketized to increase the overlap between communication and computation. Particularly, :class:`torch.distributed.GradBucket` represents a bucket of gradient tensors to be allreduced.
.. autoclass:: torch.distributed.GradBucket
.. autofunction:: torch.distributed.GradBucket.index
.. autofunction:: torch.distributed.GradBucket.buffer
.. autofunction:: torch.distributed.GradBucket.gradients
.. autofunction:: torch.distributed.GradBucket.is_last
.. autofunction:: torch.distributed.GradBucket.set_buffer
.. autofunction:: torch.distributed.GradBucket.parameters
Default communication hooks are simple stateless hooks, so the input state
in register_comm_hook is either a process group or None.
The input bucket is a :class:`torch.distributed.GradBucket` object.
.. currentmodule:: torch.distributed.algorithms.ddp_comm_hooks.default_hooks
.. autofunction:: allreduce_hook
.. autofunction:: fp16_compress_hook
.. autofunction:: bf16_compress_hook
Additionally, a communication hook wraper is provided to support :meth:`~fp16_compress_hook` or :meth:`~bf16_compress_hook` as a wrapper, which can be combined with other communication hooks.
.. autofunction:: fp16_compress_wrapper
.. autofunction:: bf16_compress_wrapper
PowerSGD (Vogels et al., NeurIPS 2019) is a gradient compression algorithm, which can provide very high compression rates and accelerate bandwidth-bound distributed training. This algorithm needs to maintain both some hyperparameters and the internal state. Therefore, PowerSGD communication hook is a stateful hook, and the user needs to provide a state object defined as below.
.. currentmodule:: torch.distributed.algorithms.ddp_comm_hooks.powerSGD_hook
.. autoclass:: PowerSGDState
Warning
PowerSGD typically requires extra memory of the same size as the model's gradients to enable error feedback, which can compensate for biased compressed communication and improve accuracy.
Warning
PowerSGD hooks may conflict with Apex automatic mixed precision package. Please use PyTorch native automatic mixed precision package instead.
.. autofunction:: powerSGD_hook
.. autofunction:: batched_powerSGD_hook
As the name implies, debugging communication hooks are only used for debugging and performance optimization purpose.
.. currentmodule:: torch.distributed.algorithms.ddp_comm_hooks.debugging_hooks
Warning
Debugging communication hooks do not necessarily output the correct results.
.. autofunction:: noop_hook
Many thanks to PowerSGD paper author Thijs Vogels for the code review on PowerSGD communication hook, as well as the comparison experiments, which show that the performance of PowerSGD communication hook is on par with the implementation in the original paper.