Building IRModules with tuple inputs

Hello,

Is there a way you can build relay IRModules that have tuple as their input? I am trying to run this scratch example but i keep running into an error (Check failed: node->Type() == kGraphOpNode (1 vs. 2) : )

import tvm
from tvm import relay
​
ctx = tvm.cpu(0)
​
input_type = relay.TensorType(shape=[1, 20, 20, 3])
input_tuple_type = relay.TupleType([input_type, input_type])
tup = relay.var('tuple_input', type_annotation=input_tuple_type)
concat = relay.op._make.concatenate(tup, 1)
​
f = relay.Function(relay.analysis.free_vars(concat), concat)
mod = tvm.IRModule.from_expr(f)
​
​
with tvm.transform.PassContext(opt_level=3):
    lib = tvm.relay.build(mod, target='llvm', target_host='llvm', params={})

I’m not sure if graph runtime supports tuple inputs, but VM definitely does. I remember this test has a tuple input.

Ok. Thanks for the info.