Relay.build_module.create_executor retruns an array full of 'nan's

1

Hello everyone. I’m new to TVM and I’m trying to compile an ONNX model replicating the code in the tutorial. I’m running it on colab:

target = "llvm"

input_name = "1"
shape_dict = {input_name: x.shape}
mod, params = relay.frontend.from_onnx(onnx_model, shape_dict)

with tvm.transform.PassContext(opt_level=1):
    executor = relay.build_module.create_executor(
        "graph", mod, tvm.cpu(0), target
    ).evaluate()

...100%, 0.02 MB, 819 KB/s, 0 seconds passed/usr/local/lib/python3.7/dist-packages/tvm/relay/frontend/onnx.py:1869: UserWarning: Mismatched attribute type in ' : kernel_shape'

==> Context: Bad node spec for node. Name:  OpType: Conv
  warnings.warn(str(e))
Cannot find config for target=llvm, workload=('conv2d_NCHWc.x86', ('TENSOR', (1, 32, 32, 32), 'float32'), ('TENSOR', (9, 32, 3, 3), 'float32'), (1, 1), (1, 1, 1, 1), (1, 1), 'NCHW', 'NCHW', 'float32'). A fallback configuration is used, which may bring great performance regression.
Cannot find config for target=llvm, workload=('conv2d_NCHWc.x86', ('TENSOR', (1, 64, 32, 32), 'float32'), ('TENSOR', (32, 64, 3, 3), 'float32'), (1, 1), (1, 1, 1, 1), (1, 1), 'NCHW', 'NCHW', 'float32'). A fallback configuration is used, which may bring great performance regression.

Cannot find config for target=llvm, workload=('conv2d_NCHWc.x86', ('TENSOR', (1, 64, 32, 32), 'float32'), ('TENSOR', (64, 64, 3, 3), 'float32'), (1, 1), (1, 1, 1, 1), (1, 1), 'NCHW', 'NCHW', 'float32'). A fallback configuration is used, which may bring great performance regression.
Cannot find config for target=llvm, workload=('conv2d_NCHWc.x86', ('TENSOR', (1, 1, 32, 32), 'float32'), ('TENSOR', (64, 1, 5, 5), 'float32'), (1, 1), (2, 2, 2, 2), (1, 1), 'NCHW', 'NCHW', 'float32'). A fallback configuration is used, which may bring great performance regression.

dtype = "float32"
tvm_output = np.array(executor(tvm.nd.array(x.astype(dtype))))
tvm_output

The tvm_output is a Matrix of nans:

array(<tvm.nd.NDArray shape=(1, 1, 672, 672), cpu(0)>
    array([[[[nan, nan, nan, ..., nan, nan, nan],
             [nan, nan, nan, ..., nan, nan, nan],
             [nan, nan, nan, ..., nan, nan, nan],
             ...,
             [nan, nan, nan, ..., nan, nan, nan],
             [nan, nan, nan, ..., nan, nan, nan],
             [nan, nan, nan, ..., nan, nan, nan]]]], dtype=float32),
          dtype=object)

And the shape is zero:

tvm_output.shape

()

Thank you in advance for sharing your knowledge.