torsh-nn 0.1.2

Neural network modules for ToRSh with PyTorch-compatible API
Documentation 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53

# ToRSh Neural Network Layers

## Conv2d Implementation

The Conv2d layer has been implemented with the following features:

### Basic Functionality
- 2D convolution operation with configurable kernel size, stride, padding, dilation, and groups
- Bias support (optional)
- PyTorch-compatible API

### Current Implementation Details
- The forward pass creates a placeholder output tensor of the correct shape
- Full convolution algorithm (im2col or direct convolution) is pending implementation
- Supports various padding and stride configurations
- Properly calculates output dimensions based on input size and convolution parameters

### Usage Example
```rust
use torsh_nn::modules::Conv2d;
use torsh_nn::Module;

// Create a Conv2d layer
let conv = Conv2d::new(
    3,          // in_channels
    32,         // out_channels
    (3, 3),     // kernel_size
    None,       // stride (default: 1)
    Some((1, 1)), // padding
    None,       // dilation (default: 1)
    None,       // groups (default: 1)
    true,       // bias
);

// Forward pass
let input = randn(&[batch_size, 3, 32, 32]);
let output = conv.forward(&input)?;
```

### Test Coverage
- Basic convolution with different configurations
- Padding effects on output size
- Stride effects on output size
- Depthwise convolution (groups = in_channels)
- Parameter shapes and counts
- Training mode switching
- Input validation

### Future Improvements
- Implement efficient convolution algorithms (im2col, Winograd, FFT-based)
- Add CUDA/GPU backend support
- Optimize for different hardware architectures
- Add more specialized convolution variants (transposed, dilated, etc.)