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

Function _AllReduceBlobs

caffe2/python/data_parallel_model.py:1237–1255  ·  view source on GitHub ↗
(blob_names, devices, model, net, rendezvous, use_nccl,
                    max_concurrent_distributed_ops)

Source from the content-addressed store, hash-verified

1235
1236
1237def _AllReduceBlobs(blob_names, devices, model, net, rendezvous, use_nccl,
1238 max_concurrent_distributed_ops):
1239 if rendezvous is None or rendezvous['num_shards'] <= 1:
1240 _AllReduceBlobsSingleHost(
1241 blob_names,
1242 devices,
1243 model,
1244 net,
1245 use_nccl
1246 )
1247 else:
1248 _AllReduceBlobsDistributed(
1249 blob_names,
1250 devices,
1251 model,
1252 net,
1253 rendezvous,
1254 max_concurrent_distributed_ops,
1255 )
1256
1257
1258def _PruneParametersForSharing(model):

Callers 2

ParallelizeFunction · 0.85
Parallelize_BMUFFunction · 0.85

Calls 2

Tested by

no test coverage detected

Used in the wild real call sites across dependent graphs

searching dependent graphs…