Write a custom CUDA kernel to optimize `EIS-3`.

Formula: f(x) = x / (1 + delta * exp(-theta * x))

This is a two-parameter generalization of Swish/SiLU.

Problem Analysis:
1. Memory Bound: This is an element-wise activation. Performance is limited by memory bandwidth.
2. Operator Chaining: The PyTorch implementation creates intermediate tensors for exp and arithmetic ops.

Optimization Strategy: Fused Element-wise Kernel with Vectorization

1. One-Thread-per-Element: Map each element to a CUDA thread.

2. Vectorized Loads (float4): Use `float4` to process 128 bits per memory transaction.

3. Fused Fast Math:
   - For each element `x`:
     `denom = 1.0f + delta * __expf(-theta * x)`
     `result = x / denom`
   - All computations are fused in registers.

4. One-Pass: Fuse all steps into a single read-compute-write kernel.
  
Here's an example to show you the syntax of inline embedding custom CUDA operators in torch: The example given architecture is:   
  
```python
import torch
import torch.nn as nn

BATCH_SIZE = 4096
HIDDEN_DIM = 4096
SHAPE = (BATCH_SIZE, HIDDEN_DIM)

DELTA_VAL = 0.68
THETA_VAL = 1.7

class EIS3(nn.Module):
    """
    EIS-3 Activation.
    EIS- A FAMILY OF ACTIVATION FUNCTIONS COMBINING EXPONENTIAL, ISRU, AND SOFTPLUS
    https://arxiv.org/pdf/2009.13501

    Formula: f(x) = x / (1 + delta * exp(-theta * x))
    """
    def __init__(self, delta=0.68, theta=1.7):
        super(EIS3, self).__init__()
        self.delta = delta
        self.theta = theta

    def forward(self, x: torch.Tensor) -> torch.Tensor:
        denominator = 1.0 + self.delta * torch.exp(-self.theta * x)
        return x / denominator

class Model(nn.Module):
    def __init__(self, delta=0.68, theta=1.7):
        super(Model, self).__init__()
        self.act = EIS3(delta, theta)
    
    def forward(self, x):
        return self.act(x)

def get_inputs():
    input_tensor = torch.randn(SHAPE, dtype=torch.float32) * 5.0
    return [input_tensor.contiguous()]

def get_init_inputs():
    return [DELTA_VAL, THETA_VAL]