Closed azazhu closed 1 year ago
After you compile the source code (in C++), you should either install the python package via
python3 python/setup.py install
or set environment variable PYTHONPATH
:
export PYTHONPATH=${YOUR_SPARSETIR_DIR}/python:${PYTHONPATH}
btw, you shouldn't build tvm from tvm repo, SparseTIR is a fork of TVM and is self-contained.
export PYTHONPATH can work. "python3 python/setup.py install" fails with following error:
root@84eaf74fddb8:/workspace2/sparsetir# python3 python/setup.py install
WARNING:root:git describe: fatal: detected dubious ownership in repository at '/workspace2/sparsetir'
To add an exception for this directory, call:
git config --global --add safe.directory /workspace2/sparsetir
, use 0.9.dev0
Traceback (most recent call last):
File "python/setup.py", line 251, in <module>
ext_modules=config_cython(),
File "python/setup.py", line 133, in config_cython
for fn in os.listdir(path):
FileNotFoundError: [Errno 2] No such file or directory: 'tvm/_ffi/_cython'
I think it is due to using a wrong tvm version. BTW, where is self-contained tvm in sparseTIR? is it in docker image or submodules?
I think it is due to using a wrong tvm version.
No, it's not.
Let me explain it again, you shouldn't install TVM for SparseTIR. If you did, please pip uninstall
it to avoid duplication.
SparseTIR is basically a modified version of TVM with sparse support, the tvm
in your error message No module named 'tvm.sparse'
refers to the SparseTIR itself instead of standard TVM in upstream repo.
It works now. Thx Zihao. This issue can be closed.
Did I use a wrong TVM version?
Expected behavior
Run examples/spmm/bench_spmm_naive.py successfully
Actual behavior
Met following error: [01:56:18] /opt/dgl/src/runtime/tensordispatch.cc:43: TensorDispatcher: dlopen failed: /venv/apache-tvm-py3.7/lib/python3.7/site-packages/dgl/tensoradapter/pytorch/libtensoradapter_pytorch_1.12.0.so: cannot open shared object file: No such file or directory Using backend: pytorch Traceback (most recent call last): File "examples/spmm/bench_spmm_naive.py", line 27, in
from tvm.sparse import FormatRewriteRule, lower_sparse_buffer, lower_sparse_iter
ModuleNotFoundError: No module named 'tvm.sparse'
Environment
Ubuntu 18.04.6 LTS TVM: 0.11.dev0 LLVM version 14.0.0
Steps to reproduce