```python
import torch
import torch.nn as nn
from torch.utils.cpp_extension import load_inline

_cpp_source = r"""
#include <torch/extension.h>
torch::Tensor subtract_tanh_cuda_(
    torch::Tensor x, torch::Tensor bias);
"""

_cuda_source = r"""
#include <torch/extension.h>
#include <c10/cuda/CUDAGuard.h>
#include <c10/cuda/CUDAStream.h>
#include <c10/cuda/CUDAException.h>
#include <cuda_runtime.h>

__global__ void subtract_tanh_kernel(
    float* __restrict__ x,
    const float* __restrict__ bias,
    const int64_t size,
    const int64_t spatial,
    const int64_t channels)
{
    const int64_t thread =
        static_cast<int64_t>(blockIdx.x) * blockDim.x + threadIdx.x;
    const int64_t step =
        static_cast<int64_t>(blockDim.x) * gridDim.x;

    float4* vectors = reinterpret_cast<float4*>(x);
    const int64_t vector_count = size / 4;

    for (int64_t v = thread; v < vector_count; v += step) {
        const int64_t i = v * 4;
        float4 value = vectors[v];

        const int64_t channel = (i / spatial) % channels;
        const float b0 = bias[channel];
        float b1 = b0;
        float b2 = b0;
        float b3 = b0;

        if (spatial < 4 || i % spatial > spatial - 4) {
            b1 = bias[((i + 1) / spatial) % channels];
            b2 = bias[((i + 2) / spatial) % channels];
            b3 = bias[((i + 3) / spatial) % channels];
        }

        value.x = tanhf(value.x - b0);
        value.y = tanhf(value.y - b1);
        value.z = tanhf(value.z - b2);
        value.w = tanhf(value.w - b3);
        vectors[v] = value;
    }

    const int64_t tail = vector_count * 4 + thread;
    if (tail < size) {
        x[tail] = tanhf(x[tail] - bias[(tail / spatial) % channels]);
    }
}

torch::Tensor subtract_tanh_cuda_(
    torch::Tensor x, torch::Tensor bias)
{
    TORCH_CHECK(x.is_cuda() && bias.is_cuda(), "Expected CUDA tensors");
    TORCH_CHECK(x.device() == bias.device(), "Device mismatch");
    TORCH_CHECK(
        x.scalar_type() == at::kFloat &&
        bias.scalar_type() == at::kFloat,
        "Expected float32 tensors");
    TORCH_CHECK(x.is_contiguous() && bias.is_contiguous(),
                "Expected contiguous tensors");
    TORCH_CHECK(x.dim() == 4 && bias.numel() == x.size(1),
                "Expected NCHW input and channel bias");

    const c10::cuda::CUDAGuard guard(x.device());
    const int64_t size = x.numel();
    if (size == 0) {
        return x;
    }

    constexpr int threads = 256;
    const int64_t work = (size + 3) / 4;
    const int64_t required_blocks = (work + threads - 1) / threads;
    const int blocks = static_cast<int>(
        required_blocks < 4096 ? required_blocks : 4096);

    subtract_tanh_kernel<<<
        blocks, threads, 0,
        c10::cuda::getCurrentCUDAStream(x.get_device()).stream()
    >>>(
        x.data_ptr<float>(),
        bias.data_ptr<float>(),
        size,
        x.size(2) * x.size(3),
        x.size(1));

    C10_CUDA_KERNEL_LAUNCH_CHECK();
    return x;
}
"""

_fused_ops = load_inline(
    name="convtranspose_subtract_tanh_fused_v1",
    cpp_sources=_cpp_source,
    cuda_sources=_cuda_source,
    functions=["subtract_tanh_cuda_"],
    extra_cflags=["-O3"],
    extra_cuda_cflags=["-O3"],
    verbose=False,
)


class _SubtractTanh(torch.autograd.Function):
    @staticmethod
    def forward(ctx, x, bias):
        result = _fused_ops.subtract_tanh_cuda_(x, bias)
        ctx.mark_dirty(x)
        ctx.bias_shape = bias.shape
        ctx.save_for_backward(result)
        return result

    @staticmethod
    def backward(ctx, grad_output):
        (result,) = ctx.saved_tensors
        grad = grad_output * (1 - result.square())
        grad_bias = -grad.sum(dim=(0, 2, 3)).reshape(ctx.bias_shape)
        return grad, grad_bias


class ModelNew(nn.Module):
    def __init__(
        self,
        in_channels,
        out_channels,
        kernel_size,
        bias_shape,
        stride=2,
        padding=1,
        output_padding=1,
    ):
        super().__init__()
        self.conv_transpose = nn.ConvTranspose2d(
            in_channels,
            out_channels,
            kernel_size,
            stride=stride,
            padding=padding,
            output_padding=output_padding,
        )
        self.bias = nn.Parameter(torch.randn(bias_shape))

    def forward(self, x):
        x = self.conv_transpose(x)
        channel_bias = tuple(self.bias.shape) in (
            (x.shape[1], 1, 1),
            (1, x.shape[1], 1, 1),
        )

        if (
            x.is_cuda
            and self.bias.device == x.device
            and x.dtype == torch.float32
            and self.bias.dtype == torch.float32
            and x.is_contiguous()
            and self.bias.is_contiguous()
            and channel_bias
        ):
            if torch.is_grad_enabled() and (
                x.requires_grad or self.bias.requires_grad
            ):
                return _SubtractTanh.apply(x, self.bias)
            return _fused_ops.subtract_tanh_cuda_(x, self.bias)

        return torch.tanh(x - self.bias)
```