Installation problem with TVM/NNVM

Questions
1

Hi,

We have installed ONNX, NNVM and TVM and when we do the following we get a error while importing nnvm. Can you please let me know what the issue might be. We have build TVM without CUDA support and using LLVM 5.0

/workspace# python Python 3.6.6 |Anaconda, Inc.| (default, Jun 28 2018, 17:14:51) [GCC 7.2.0] on linux Type “help”, “copyright”, “credits” or “license” for more information.

import onnx import nnvm Traceback (most recent call last): File “”, line 1, in File “/workspace/tvm/nnvm/python/nnvm/init.py”, line 10, in from . import frontend File “/workspace/tvm/nnvm/python/nnvm/frontend/init.py”, line 4, in from .onnx import from_onnx File “/workspace/tvm/nnvm/python/nnvm/frontend/onnx.py”, line 8, in from ..compiler import graph_util File “/workspace/tvm/nnvm/python/nnvm/compiler/init.py”, line 19, in from .. import top as _top File “/workspace/tvm/nnvm/python/nnvm/top/init.py”, line 6, in from . import tensor File “/workspace/tvm/nnvm/python/nnvm/top/tensor.py”, line 6, in import topi File “/workspace/tvm/topi/python/topi/init.py”, line 26, in from . import cuda File “/workspace/tvm/topi/python/topi/cuda/init.py”, line 5, in from . import conv2d, depthwise_conv2d, conv2d_transpose_nchw File “/workspace/tvm/topi/python/topi/cuda/depthwise_conv2d.py”, line 11, in nn.depthwise_conv2d_nchw.fdefault) AttributeError: ‘function’ object has no attribute ‘fdefault’ import tvm quit()

root@b3fd5c689fcd:/workspace# echo $PATH /opt/conda/envs/pytorch-py3.6/bin:/opt/conda/envs/pytorch-py3.6/bin:/opt/conda/bin:/usr/local/cuda-9.0/bin:usr/local/bin:/usr/lib:/opt/conda/bin:/usr/local/nvidia/bin:/usr/local/cuda/bin:/usr/local/sbin:/usr/local/bin:/usr/sbin:/usr/bin:/sbin:/bin

root@b3fd5c689fcd:/workspace# echo $PYTHONPATH /opt/conda/lib/python3.6/site-packages:/workspace/tvm/nnvm/python:/workspace/tvm/topi/python:/workspace/np-compiler/tvm/python

root@b3fd5c689fcd:/workspace# echo $LD_LIBRARY_PATH /opt/conda/lib:/usr/local/cuda-9.0/lib64/:/usr/local/lib:/workspace/tvm/nnvm/build:/workspace/tvm/build:/usr/local/nvidia/lib:/usr/local/nvidia/lib64

2

fdefault is a recently added attribute. It means your topi package is up to date but tvm package is stale.

In you $PYTHONPATH, you have two tvm directories “/workspace/tvm/” and “/workspace/np-compiler/tvm/”. I think “”/workspace/np-compiler/tvm/" is stale, you should update them to a same version or use a single tvm.