Hi I am actually trying to add a custom operator in TVM…I have registered it in onnx.py in tvm/nnvm/python/nnvm/frontend/onnx.py and also in relay tvm/python/tvm/relay/frontend/onnx.py…
I have registered its backend in TOPI…
However after compilation and building , when I try to invoke via: I am getting the above message…
Can someone help me figure out whats missing…?!
import nnvm
import tvm
import onnx
import numpy as np
onnx_graph = onnx.load(‘gru.onnx’)
sym, params = nnvm.frontend.from_onnx(onnx_graph)
HI @tqchen, @srkreddy1238 – Can you please help me with the above question…? What might I have missed during registering the op…?
I have made the changes in the following files to register the op…
I am invoking it via
sym, params = nnvm.frontend.from_onnx(onnx_graph)
tvm_dir=…/…/tvm
tvm_pydir=$tvm_dir/python/tvm
nnvm_dir=$tvm_dir/nnvm
nnvm_pydir=$nnvm_dir/python/nnvm
rsync -c nn_patch.h $nnvm_dir/include/nnvm/top/nn.h
rsync -c nn_patch.cc $nnvm_dir/src/top/nn/nn.cc
rsync -c onnx_patch.py $nnvm_pydir/frontend/onnx.py
rsync -c nn_patch.py $nnvm_pydir/top/nn.py
rsync -c elemwise_patch.py $tvm_dir/topi/python/topi/nn/elemwise.py
rsync -c onnx_relay_patch.py $tvm_pydir/relay/frontend/onnx.py