[Bug][Relax] Arithmetic on an unbatched AdaptiveAvgPool2d output fails in from_exported_program: "Please convert 2.0 with const first" #20943

Expected behavior

AdaptiveAvgPool2d accepts an unbatched (C, H, W) input, and pool(x) * 2.0 works in eager and torch.export, so from_exported_program should convert it.

Actual behavior

The conversion fails. torch.maximum or torch.minimum of two such outputs fails with maximum(): argument 'input' (position 1) must be Tensor, not Call. AdaptiveAvgPool3d with an unbatched (C, D, H, W) input fails the same way. A batched input, or an output size of (1, 1), works.

TypeError: Please convert 2.0 with `const` first

Possibly _adaptive_avg_pool2d and _adaptive_avg_pool3d return the squeeze for an unbatched input without emitting it, so _binary_op gets a Call instead of a Var.

Environment

• TVM: main @ 2bf0532 (0.28.dev0), built from source, target llvm
• PyTorch: 2.12.0+cu130
• Python: 3.12, Ubuntu x86_64

Steps to reproduce

import torch
import tvm
import tvm_tirx
from tvm import relax
from tvm.relax.frontend.torch import from_exported_program


class M(torch.nn.Module):
    def __init__(self):
        super().__init__()
        self.pool = torch.nn.AdaptiveAvgPool2d((2, 2))

    def forward(self, x):
        return self.pool(x) * 2.0


x = torch.rand(2, 4, 4)  # no batch dimension
args = (x,)
print("eager:", M()(*args).flatten().tolist())

ep = torch.export.export(M(), args)
mod = from_exported_program(ep)
ex = tvm.compile(mod, target="llvm")
vm = relax.VirtualMachine(ex, tvm_tirx.cpu())
out = vm["main"](*[tvm_tirx.runtime.tensor(a.numpy()) for a in args])
print("tvm:  ", out[0].numpy().flatten().tolist())