leezu commented on issue #18765: URL: https://github.com/apache/incubator-mxnet/issues/18765#issuecomment-662047865
Simpler reproducible example for latest master: ``` import mxnet as mx from mxnet import gluon from mxnet.gluon import nn a = mx.nd.random.uniform(shape=(1,3,224,224)) backbone = gluon.model_zoo.vision.resnet18_v1() backbone.initialize() backbone.hybridize() backbone(a) # Alternative: # backbone.reset_ctx(mx.gpu(0)) # b = backbone(a.as_in_context(mx.gpu(0))) # print([x.shape for x in b]) sym_file, params_file = backbone.export('/tmp/model') f = gluon.SymbolBlock.imports(sym_file, 'data', params_file) f.reset_ctx(mx.gpu(0)) b = f(a.as_in_context(mx.gpu(0))) print([x.shape for x in b]) ``` It fails with ``` [18:59:34] ../src/storage/storage.cc:198: Using Pooled (Naive) StorageManager for CPU /home/ubuntu/src/mxnet-master/python/mxnet/gluon/block.py:1723: UserWarning: Cannot decide type for the following arguments. Consider providing them as input: data: None input_sym_arg_type = in_param.infer_type()[0] [18:59:37] ../src/storage/storage.cc:198: Using Pooled (Naive) StorageManager for GPU [(1000,)] [18:59:38] ../src/operator/nn/./cudnn/./cudnn_algoreg-inl.h:97: Running performance tests to find the best convolution algorithm, this can take a while... (set the environment variable MXNET_CUDNN_AUTOTUNE_DEFAULT to 0 to disable) Segmentation fault: 11 zsh: abort (core dumped) python3 symbolblockbug.py ``` However, the following will work: ``` import mxnet as mx from mxnet import gluon from mxnet.gluon import nn a = mx.nd.random.uniform(shape=(1,3,224,224)) backbone = gluon.model_zoo.vision.resnet18_v1() backbone.initialize() backbone.hybridize() # backbone(a) # Alternative: backbone.reset_ctx(mx.gpu(0)) b = backbone(a.as_in_context(mx.gpu(0))) print([x.shape for x in b]) sym_file, params_file = backbone.export('/tmp/model') f = gluon.SymbolBlock.imports(sym_file, 'data', params_file) f.reset_ctx(mx.gpu(0)) b = f(a.as_in_context(mx.gpu(0))) print([x.shape for x in b]) ``` ---------------------------------------------------------------- This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. For queries about this service, please contact Infrastructure at: us...@infra.apache.org