spikingjelly.activation\_based.distributed package =========================================================== 本子包提供基于 ``torch.distributed``、``DTensor``、tensor parallel 与 FSDP2 的实验性分布式训练工具，面向 ``spikingjelly.activation_based`` 的多步 SNN。 ---- This package provides experimental distributed-training helpers for multi-step SNNs in ``spikingjelly.activation_based`` based on ``torch.distributed``, ``DTensor``, tensor parallelism, and FSDP2. Distributed Helpers +++++++++++++++++++++++++++++ .. list-table:: * - :class:`SNNDistributedConfig ` - High-level configuration for DTensor-ready SNN distribution. * - :class:`SNNDistributedAnalysis ` - Capability analysis for stateful modules and tensor-parallel candidates. * - :func:`ensure_distributed_initialized ` - Initialize ``torch.distributed`` when needed. * - :func:`build_device_mesh ` - Build a ``DeviceMesh`` for tensor/data parallelism. * - :func:`configure_snn_distributed ` - The main low-level entry for DTensor-ready SNN distribution. * - :func:`configure_cifar10dvs_vgg_distributed ` - Convenience helper for ``CIFAR10DVSVGG`` with DP / TP. * - :func:`configure_cifar10dvs_vgg_fsdp2 ` - Convenience helper for ``CIFAR10DVSVGG`` with FSDP2 / FSDP2+TP. * - :func:`materialize_dtensor_output ` - Convert a ``DTensor`` output back to a regular tensor when needed. .. automodule:: spikingjelly.activation_based.distributed :members: :undoc-members: :show-inheritance: