MCPcopy Index your code
hub / github.com/pytorch/pytorch / _SyncAllParams

Function _SyncAllParams

caffe2/python/data_parallel_model.py:1104–1124  ·  view source on GitHub ↗
(
    devices,
    model,
    init_net,
    net,
    rendezvous,
    unique_param_names,
    max_concurrent_distributed_ops=4
)

Source from the content-addressed store, hash-verified

1102
1103
1104def _SyncAllParams(
1105 devices,
1106 model,
1107 init_net,
1108 net,
1109 rendezvous,
1110 unique_param_names,
1111 max_concurrent_distributed_ops=4
1112):
1113 if rendezvous is None or rendezvous['num_shards'] <= 1:
1114 _SyncAllParamsSingleHost(devices, model, net, unique_param_names)
1115 else:
1116 _SyncAllParamsDistributed(
1117 devices,
1118 model,
1119 init_net,
1120 net,
1121 rendezvous,
1122 unique_param_names,
1123 max_concurrent_distributed_ops
1124 )
1125
1126
1127def AddBlobSync(model, blobs, net=None):

Callers 4

ParallelizeFunction · 0.85
Parallelize_BMUFFunction · 0.85
FinalizeAfterCheckpointFunction · 0.85
AddBlobSyncFunction · 0.85

Calls 2

_SyncAllParamsSingleHostFunction · 0.85

Tested by

no test coverage detected

Used in the wild real call sites across dependent graphs

searching dependent graphs…