Thanks for sending an issue! Please fill in the following template to help quickly solve your problem.

一、问题描述 （必填）

脚本如下：

import torch
import numpy as np
import torch_npu  # 导入NPU适配库，必须放在torch之后[reference:2][reference:3]

torch.manual_seed(42)
np.random.seed(42)

device = "npu" if torch.npu.is_available() else "cpu"
print(f"使用设备: {device}")

input_cpu = torch.tensor(
    np.random.uniform(-100, 100, size=(1, 1, 1, 1)), 
    dtype=torch.float16
)

input_tensor = input_cpu.npu() if device == "npu" else input_cpu

k = 1
dim = 0
keepdim = False  # torch.kthvalue默认keepdim=False

values, indices = torch.kthvalue(input_tensor, k=k, dim=dim, keepdim=keepdim)
values_cpu = values.cpu()
indices_cpu = indices.cpu()

print("Input tensor (on NPU):")
print(input_tensor)
print(f"\nShape: {input_tensor.shape}, dtype: {input_tensor.dtype}, device: {input_tensor.device}")
print(f"\ntorch.kthvalue(input, k={k}, dim={dim})")
print("Values:", values_cpu)
print("Indices:", indices_cpu)
print("Values shape:", values_cpu.shape)
print("Indices shape:", indices_cpu.shape)

报错信息为
Input tensor's format[NCHW] should be same with output's format[NCL]. Params check failed.

二、环境信息 （可选）

昇腾硬件型号：950PR
CANN版本：最新周构建包

三、重现步骤 （可选）

见第一章

四、预期结果 （可选）

💡 备注（选填）