experiments/STT-tensorflow
1
0
Fork 0
Code Issues Pull Requests Projects Releases Wiki Activity

Link to the bug where we change the inter_op pool size in in-process cluster.

Browse Source 
PiperOrigin-RevId: 341889256
Change-Id: I8cdeadb775cf20c73d50fb6b0a47a50b5a78b330
This commit is contained in:
Yuefeng Zhou 2020-11-11 12:41:15 -08:00 committed by TensorFlower Gardener
parent f473a447dd
commit 6a83a28990
1 changed files with 2 additions and 1 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

3
tensorflow/python/distribute/multi_worker_test_base.py
View File

@ -164,7 +164,8 @@ def create_in_process_cluster(num_workers,
worker_config = config_pb2.ConfigProto() worker_config = config_pb2.ConfigProto()
worker_config.gpu_options.per_process_gpu_memory_fraction = gpu_mem_frac worker_config.gpu_options.per_process_gpu_memory_fraction = gpu_mem_frac
# The cluster may hang if workers don't have enough inter_op threads. # The cluster may hang if workers don't have enough inter_op threads. See
# b/172296720 for more details.
if multiprocessing.cpu_count() < 4: if multiprocessing.cpu_count() < 4:
worker_config.inter_op_parallelism_threads = 4 worker_config.inter_op_parallelism_threads = 4