inconsistent behavior of nn.functional.interpolate api between oneflow and torch

[clackhan]

pytorch version: 1.12.1+cu102
oneflow:

version: 0.8.1+cu112.git.c0811b327a
git_commit: c0811b327a
cmake_build_type: Debug
rdma: False
mlir: False

pytorch 正常运行

import torch
import numpy as np

mode="bicubic"
x = torch.Tensor(8, 32, 64)
x = x.to("cuda")

window = 16
t = x.shape[2]
x = x[:, None]


np_center = np.random.randint(window, t - window, (1,))[0]
np_warped = np.random.randint(np_center - window, np_center + window, (1,))[0] + 1

center = torch.tensor(np_center)
warped = torch.tensor(np_warped)

left = torch.nn.functional.interpolate(
    x[:, :, :center], (warped, x.shape[3]), mode=mode, align_corners=False
)

oneflow 报错：

import  oneflow as torch
import numpy as np

mode="bicubic"
x = torch.Tensor(8, 32, 64)
x = x.to("cuda")

window = 16
t = x.shape[2]
x = x[:, None]


np_center = np.random.randint(window, t - window, (1,))[0]
np_warped = np.random.randint(np_center - window, np_center + window, (1,))[0] + 1

center = torch.tensor(np_center)
warped = torch.tensor(np_warped)

left = torch.nn.functional.interpolate(
    x[:, :, :center], (warped, x.shape[3]), mode=mode, align_corners=False
)

报错信息

Traceback (most recent call last):
  File "test1.py", line 19, in <module>
    left = torch.nn.functional.interpolate(
  File "/home/hanbinbin/oneflow/python/oneflow/nn/modules/interpolate.py", line 309, in interpolate
    return Interpolate(
  File "/home/hanbinbin/oneflow/python/oneflow/nn/module.py", line 146, in __call__
    res = self.forward(*args, **kwargs)
  File "/home/hanbinbin/oneflow/python/oneflow/nn/modules/interpolate.py", line 193, in forward
    return flow._C.upsample_bicubic_2d(
TypeError: upsample_bicubic_2d(): argument 'height_scale' must be double, not <class 'oneflow.Tensor'>

[BBuf]

fixed by #9118