Closed antoajayraj closed 7 years ago
That shouldn't happen. Have you modified the fb.resnet.torch code? Are you running a recent version of Torch?
No.. I had not modified the fb.resnet.torch code. and I am using the recent version of Torch
@antoajayraj We are using a power8 ( ppc64le ) and we are having the same problem. Are you using the same architecture?
@pedropgusmao Yup.. i was using ppc64le. Now I got it resolved by adding the current directory to LUA_PATH:
$ export LUA_PATH="$LUA_PATH;./?.lua" $ cd multi-resnet/ $ th main.lua -netType multi-resnet -depth 200 -k 5 -batchSize 64 -nGPU 2 -nThreads 4 -dataset cifar10 -nEpochs 200
Thanks !!
I see a similar problem as reported in #168 th main.lua -depth 50 -batchSize 256 -nGPU 4 -nThreads 8 -shareGradInput true -data /ImageNet2012/ => Creating model from file: models/resnet.lua | ResNet-50 ImageNet nil warning: could not load nccl, falling back to default communication => Generating list of images | finding all validation images | finding all training images | saving list of images to /ml/fb.resnet.torch/gen/imagenet.t7 FATAL THREAD PANIC: (read) /oasd/torch/share/lua/5.1/torch/File.lua:343: unknown Torch class
FATAL THREAD PANIC: (read) /oasd/torch/share/lua/5.1/torch/File.lua:343: unknown Torch class
Segmentation fault
Am I missing something basic here ? Any help ?