API 参考#
- 进程组
ErrorSwallowingProcessGroupWrapper
FakeProcessGroupWrapper
ManagedProcessGroup
ProcessGroup
ProcessGroupBaby
ProcessGroupBabyGloo
ProcessGroupBabyNCCL
ProcessGroupDummy
ProcessGroupGloo
ProcessGroupNCCL
ProcessGroupWrapper
create_store_client()
- 管理器
ExceptionWithTraceback
管理器
WorldSizeMode
get_timeout()
- 优化器
OptimizerWrapper
- 分布式数据并行
DistributedDataParallel
PureDistributedDataParallel
- LocalSGD
DiLoCo
LocalSGD
extract_local_tensor()
- 数据
DistributedSampler
- 检查点
CheckpointTransport
HTTPTransport
- 参数服务器
ParameterServer
- 协调(低级 API)
LighthouseClient
LighthouseServer
ManagerClient
ManagerServer
Quorum
QuorumMember