Loris Bennett <[email protected]> writes: > Hi, > > With EB 4.4.2, My attempt to build > > Keras-2.3.1-fosscuda-2019b-Python-3.7.4.eb > > is failing on the dependency > > TensorFlow-2.1.0-fosscuda-2019b-Python-3.7.4 > > with > > ERROR: > /trinity/shared/easybuild/build/TensorFlow/2.1.0/fosscuda-2019b-Python-3.7.4/TensorFlow/tensorflow-2.1.0/tensorflow/core/platform/BUILD:313:1: > C++ compilation of rule '//tensorflow/core/platform:numbers' failed (Exit 1) > tensorflow/core/platform/numbers.cc:28:10: fatal error: > double-conversion/double-conversion.h: No such file or directory > > but earlier in the log I have > > == 2021-09-28 14:19:59,377 modules.py:617 INFO Checking whether > double-conversion/3.1.4-GCCcore-8.3.0 exists... > == 2021-09-28 14:19:59,377 modules.py:622 INFO Module > double-conversion/3.1.4-GCCcore-8.3.0 exists (found in list of available > modules) > == 2021-09-28 14:19:59,377 modules.py:645 INFO Result for existence check > of double-conversion/3.1.4-GCCcore-8.3.0 module: True > > Does this ring any bells for anyone?
I tried building TensorFlow directly via eb TensorFlow-2.1.0-fosscuda-2019b-Python-3.7.4.eb --robot --cuda-compute-capabilities=6.1,7.5 but got the error == 2021-09-28 15:32:01,661 filetools.py:1635 INFO Adjusting permissions recursively for /trinity/shared/easybuild/build/TensorFlow/2.1.0/fosscuda-2019b-Python-3.7.4 == 2021-09-28 15:32:18,381 build_log.py:169 ERROR EasyBuild crashed with an error (at easybuild/software/EasyBuild/4.4.2/lib/python2.7/site-packages/easybuild/base/exceptions.py:124 in __init__): Failed to chmod/chown several paths: ['/trinity/shared/easybuild/build/TensorFlow/2.1.0/fosscuda-2019b-Python-3.7.4/tmpmA5Ye_-bazel-tf/9b6ce34a74d722f472939a0388a2a554/server/.nfs00000000a98ef9d700000018'] (last error: [Errno 2] No such file or directory: '/trinity/shared/easybuild/build/TensorFlow/2.1.0/fosscuda-2019b-Python-3.7.4/tmpmA5Ye_-bazel-tf/9b6ce34a74d722f472939a0388a2a554/server/.nfs00000000a98ef9d700000018') (at easybuild/software/EasyBuild/4.4.2/lib/python2.7/site-packages/easybuild/tools/filetools.py:1707 in adjust_permissions) This looks like a problem with the NFS share '/trinity/shared', which is mounted rw on the GPU node I am compiling on and which contains all the EB stuff. With eb TensorFlow-2.1.0-fosscuda-2019b-Python-3.7.4.eb --robot --cuda-compute-capabilities=6.1,7.5 --buildpath=/dev/shm --tmpdir=/scratch/eb-build I get back to the original error. Cheers, Loris -- Dr. Loris Bennett (Hr./Mr.) ZEDAT, Freie Universität Berlin Email [email protected]
