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

_cpp_source = """
torch::Tensor fused_scale_leaky_cuda(
    torch::Tensor x, double multiplier, double negative_slope);
"""

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

__device__ __forceinline__ float activate(
    float x, float multiplier, float negative_slope) {
    float scaled = x * multiplier;
    return scaled > 0.0f ? scaled : scaled * negative_slope;
}

__global__ void scale_leaky_float4(
    const float* __restrict__ x,
    float* __restrict__ out,
    int64_t n,
    float multiplier,
    float negative_slope) {
    const int64_t i =
        static_cast<int64_t>(blockIdx.x) * blockDim.x + threadIdx.x;
    const int64_t offset = i * 4;

    if (offset + 3 < n) {
        float4 v = reinterpret_cast<const float4*>(x)[i];
        v.x = activate(v.x, multiplier, negative_slope);
        v.y = activate(v.y, multiplier, negative_slope);
        v.z = activate(v.z, multiplier, negative_slope);
        v.w = activate(v.w, multiplier, negative_slope);
        reinterpret_cast<float4*>(out)[i] = v;
    } else {
        for (int64_t j = offset; j < n; ++j) {
            out[j] = activate(x[j], multiplier, negative_slope);
        }
    }
}

template <typename scalar_t>
__global__ void scale_leaky_generic(
    const scalar_t* __restrict__ x,
    scalar_t* __restrict__ out,
    int64_t n,
    double multiplier,
    double negative_slope) {
    using acc_t = at::acc_type<scalar_t, true>;
    const int64_t i =
        static_cast<int64_t>(blockIdx.x) * blockDim.x + threadIdx.x;
    if (i < n) {
        // Preserve the intermediate rounding of the original multiplication.
        scalar_t scaled = static_cast<scalar_t>(
            static_cast<acc_t>(x[i]) * static_cast<acc_t>(multiplier));
        acc_t value = static_cast<acc_t>(scaled);
        out[i] = static_cast<scalar_t>(
            value > acc_t(0)
                ? value
                : value * static_cast<acc_t>(negative_slope));
    }
}

torch::Tensor fused_scale_leaky_cuda(
    torch::Tensor x, double multiplier, double negative_slope) {
    TORCH_CHECK(x.is_cuda(), "Expected a CUDA tensor");
    TORCH_CHECK(x.is_contiguous(), "Expected a contiguous tensor");

    c10::cuda::CUDAGuard guard(x.device());
    auto out = torch::empty_like(x);
    const int64_t n = x.numel();
    if (n == 0) {
        return out;
    }

    constexpr int threads = 256;
    cudaStream_t stream = at::cuda::getCurrentCUDAStream();

    if (x.scalar_type() == at::kFloat) {
        const int64_t vectors = (n + 3) / 4;
        const int blocks = static_cast<int>((vectors + threads - 1) / threads);
        scale_leaky_float4<<<blocks, threads, 0, stream>>>(
            x.data_ptr<float>(), out.data_ptr<float>(), n,
            static_cast<float>(multiplier),
            static_cast<float>(negative_slope));
    } else {
        const int blocks = static_cast<int>((n + threads - 1) / threads);
        AT_DISPATCH_FLOATING_TYPES_AND2(
            at::ScalarType::Half,
            at::ScalarType::BFloat16,
            x.scalar_type(),
            "fused_scale_leaky",
            [&] {
                scale_leaky_generic<scalar_t><<<blocks, threads, 0, stream>>>(
                    x.data_ptr<scalar_t>(), out.data_ptr<scalar_t>(),
                    n, multiplier, negative_slope);
            });
    }

    C10_CUDA_KERNEL_LAUNCH_CHECK();
    return out;
}
"""

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


class _FusedScaleLeaky(torch.autograd.Function):
    @staticmethod
    def forward(ctx, x, multiplier, negative_slope):
        ctx.multiplier = multiplier
        ctx.negative_slope = negative_slope
        if ctx.needs_input_grad[0]:
            ctx.save_for_backward(x)
        return _fused_ops.fused_scale_leaky_cuda(
            x.contiguous(), multiplier, negative_slope
        )

    @staticmethod
    def backward(ctx, grad_output):
        (x,) = ctx.saved_tensors
        scaled = x * ctx.multiplier
        grad = torch.where(
            scaled > 0,
            grad_output,
            grad_output * ctx.negative_slope,
        )
        return grad * ctx.multiplier, None, None


class ModelNew(nn.Module):
    def __init__(self, in_features, out_features, multiplier, negative_slope):
        super().__init__()
        self.gemm = nn.Linear(in_features, out_features)
        self.multiplier = multiplier
        self.leaky_relu = nn.LeakyReLU(negative_slope)

    def forward(self, x):
        x = self.gemm(x)
        if not x.is_cuda:
            return self.leaky_relu(x * self.multiplier)
        return _FusedScaleLeaky.apply(
            x, self.multiplier, self.leaky_relu.negative_slope
        )
```