1 通过relay.create_executor来执行Relay表达式

def test_any_concat():
    """测试concatenate算子"""
    x = relay.var('x', shape=(relay.Any(), 2), dtype="float32")
    y = relay.var('y', shape=(1, 2), dtype="float32")
    z = relay.op.concatenate([x, y], axis=0)
    # mod = relay.module.Module()
    # mod["main"] = relay.Function([x, y], z)
    f = relay.Function([x, y], z)
    mod = relay.module.Module.from_expr(f)
    x_np = np.random.uniform(size=(3, 2)).astype('float32')
    y_np = np.random.uniform(size=(1, 2)).astype('float32')
    ref = np.concatenate([x_np, y_np], axis=0)
    print("\n----------------------------\nexpected result:")
    print(ref)
    for kind in ["debug"]:
        ex = relay.create_executor(kind, mod=mod, ctx=tvm.cpu(), target="llvm") #创建执行者
        result = ex.evaluate()(x_np, y_np) #执行表达式，返回结果
        print("\n----------------------------\nreal result:")
        print(result.asnumpy())
        tvm.testing.assert_allclose(result.asnumpy(), ref) #比较结果

2 采用build + run获得运行结果

def tvm_run(input_val, params, inputs, level=2):
    """A hack for getting the value of an expression by evaluating a
    portion of the relay graph. This is often needed for functions that
    whose output shape depends on the value of a tensor.
    """
    import tvm
    from tvm.relay import expr as _expr
    from tvm.relay import analysis
    from tvm.contrib import graph_runtime
    # Check that all free variables have associated parameters.
    # assert all(var.name_hint in params.keys() for var in analysis.free_vars(
    #     input_val)), "All inputs to infer must be available in params."
    func = _expr.Function(analysis.free_vars(input_val), input_val)
    print("\n------------------------------------------------------------")
    print(func.astext(show_meta_data=False))
    # target = tvm.target.vacc()
    # ctx = tvm.vacc(0)
    target = "llvm"
    ctx = tvm.cpu(0)
    with tvm.relay.build_config(opt_level=level):
        graph, lib, params = tvm.relay.build(
            func, target=target, params=params)
    # print("\n------------------------------------------------------------")
    # print(graph)
    print("\n------------------------------------------------------------")
    print(params)
    m = graph_runtime.create(graph, lib, ctx)
    if inputs is not None:
        m.set_input(**inputs)
    m.set_input(**params)
    m.run()
    out = m.get_output(0)
    print("\n------------------------------------------------------------")
    print(out)
    return out