[Bug][Relax] x * t.item() fails in from_exported_program when x is float and t is an integer tensor #20939

Expected behavior

t.item() on an integer tensor gives a Python int, and x * t.item() with a float x gives a float tensor. It works in eager and torch.export, so from_exported_program should convert it.

Actual behavior

The conversion fails. The same happens for +, -, /, for t.item() * x and for a float16 x. A float t, or an integer x, works.

eager: [4.5, 6.0] torch.float32
TypeError: Binary operators must have the same datatype for both operands.  However, ... "relax.multiply" ... uses datatype T.float32 on the LHS (Type of R.Tensor((2,), dtype="float32")), and datatype T.int64 on the RHS (Type of T.int64).

Possibly _item turns an integer tensor into an int64 symbolic variable, and _binary_op passes it on without casting it to the dtype of the float tensor.

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 forward(self, x, t):
        return x * t.item()


x = torch.tensor([1.5, 2.0])
t = torch.tensor(3)
print("eager:", M()(x, t).tolist(), M()(x, t).dtype)

ep = torch.export.export(M(), (x, t))
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 (x, t)])
print("tvm:  ", out[0].numpy().tolist(), out[0].numpy().dtype)