Distributed RPC Framework

Note

若API“是否支持”为“是”,“限制与说明”为“-”,说明此API和原生API支持度保持一致。

API名称 是否支持 限制与说明
torch.distributed.rpc.init_rpc NPU设备启用rpc时需要在init_rpc中进行特定的设置:backend绑定rpc.backend_registry.BackendType.NPU_TENSORPIPE;
options绑定NPUTensorPipeRpcBackendOptions,需要from torch_npu.distributed.rpc.options import NPUTensorPipeRpcBackendOptions并设置option选项,参数格式和原版TensorPipeRpcBackendOptions相同。
torch.distributed.rpc.rpc_sync -
torch.distributed.rpc.rpc_async -
torch.distributed.rpc.remote -
torch.distributed.rpc.get_worker_info -
torch.distributed.rpc.shutdown -
torch.distributed.rpc.WorkerInfo -
torch.distributed.rpc.functions.async_execution -
torch.distributed.rpc.BackendType -
torch.distributed.rpc.RpcBackendOptions -
torch.distributed.rpc.TensorPipeRpcBackendOptions 建议使用已适配的torch.distributed.rpc.NPUTensorPipeRpcBackendOptions
torch.distributed.rpc.TensorPipeRpcBackendOptions.set_device_map 建议使用已适配的torch.distributed.rpc.NPUTensorPipeRpcBackendOptions.set_device_map
torch.distributed.rpc.TensorPipeRpcBackendOptions.set_devices 建议使用已适配的torch.distributed.rpc.NPUTensorPipeRpcBackendOptions.set_devices
torch.distributed.rpc.PyRRef -
torch.distributed.nn.api.remote_module.RemoteModule -
torch.distributed.nn.api.remote_module.RemoteModule.get_module_rref -
torch.distributed.nn.api.remote_module.RemoteModule.remote_parameters -
torch.distributed.autograd.backward -
torch.distributed.autograd.context -
torch.distributed.autograd.get_gradients -