"Concat" node has no "topi" key when auto tuning onnx model to tvm

Encounter a “KeyError:‘topi_op’” problem when running tvm auto tune. It occurs in _iterate_layout_transform() when running DPTuner.benchmark_layout_trasform.

As you see, in_node_entry[964] has no key “topi_op”, this node is the second(last) "concat"node in the model pic below.

Actually the question is why this “concat” node(in_node_entry[964]) has multipe inputs(two slice inputs and add), but not become a valid node with "topi_op"and “workloads” ?

1 Like

The model pic is here:

1 Like