tessl/pypi-torch

Deep learning framework providing tensor computation with GPU acceleration and dynamic neural networks with automatic differentiation

Overview

Eval results

Files

Neural Networks

Name: tessl/pypi-torch
Author: tessl

Complete neural network building blocks including layers, activation functions, loss functions, and containers for building deep learning models. The torch.nn module provides high-level abstractions for neural network construction.

Capabilities

Base Classes

Core classes that form the foundation of all neural network components.

class Module:
    """Base class for all neural network modules."""
    def __init__(self): ...
    def forward(self, *input): 
        """Define forward computation."""
    def parameters(self, recurse=True):
        """Return iterator over module parameters."""
    def named_parameters(self, prefix='', recurse=True):
        """Return iterator over (name, parameter) pairs."""
    def modules(self):
        """Return iterator over all modules."""
    def named_modules(self, memo=None, prefix=''):
        """Return iterator over (name, module) pairs."""
    def children(self):
        """Return iterator over immediate children modules."""
    def named_children(self):
        """Return iterator over (name, child) pairs."""
    def train(self, mode=True):
        """Set module in training mode."""
    def eval(self):
        """Set module in evaluation mode."""
    def zero_grad(self, set_to_none=False):
        """Set gradients to zero."""
    def to(self, *args, **kwargs):
        """Move module to device/dtype."""
    def cuda(self, device=None):
        """Move module to CUDA device."""
    def cpu(self):
        """Move module to CPU."""
    def state_dict(self, destination=None, prefix='', keep_vars=False):
        """Return dictionary of module state."""
    def load_state_dict(self, state_dict, strict=True):
        """Load parameters and buffers."""

class Parameter(Tensor):
    """Trainable parameter tensor."""
    def __init__(self, data=None, requires_grad=True): ...

class UninitializedParameter(Parameter):
    """Parameter that is not yet initialized."""
    def __init__(self, requires_grad=True): ...

Linear Layers

Dense layers that perform linear transformations.

class Linear(Module):
    """Linear transformation: y = xA^T + b."""
    def __init__(self, in_features: int, out_features: int, bias: bool = True, device=None, dtype=None): ...
    def forward(self, input: Tensor) -> Tensor: ...

class Bilinear(Module):  
    """Bilinear transformation: y = x1^T A x2 + b."""
    def __init__(self, in1_features: int, in2_features: int, out_features: int, bias: bool = True): ...
    def forward(self, input1: Tensor, input2: Tensor) -> Tensor: ...

class LazyLinear(Module):
    """Linear layer with lazy weight initialization."""
    def __init__(self, out_features: int, bias: bool = True): ...
    def forward(self, input: Tensor) -> Tensor: ...

class Identity(Module):
    """Identity transformation."""
    def __init__(self, *args, **kwargs): ...
    def forward(self, input: Tensor) -> Tensor: ...

Convolution Layers

Convolutional layers for spatial feature extraction.

class Conv1d(Module):
    """1D convolution layer."""
    def __init__(self, in_channels: int, out_channels: int, kernel_size, stride=1, 
                 padding=0, dilation=1, groups=1, bias=True, padding_mode='zeros'): ...
    def forward(self, input: Tensor) -> Tensor: ...

class Conv2d(Module):
    """2D convolution layer."""  
    def __init__(self, in_channels: int, out_channels: int, kernel_size, stride=1,
                 padding=0, dilation=1, groups=1, bias=True, padding_mode='zeros'): ...
    def forward(self, input: Tensor) -> Tensor: ...

class Conv3d(Module):
    """3D convolution layer."""
    def __init__(self, in_channels: int, out_channels: int, kernel_size, stride=1,
                 padding=0, dilation=1, groups=1, bias=True, padding_mode='zeros'): ...
    def forward(self, input: Tensor) -> Tensor: ...

class ConvTranspose1d(Module):
    """1D transposed convolution layer."""
    def __init__(self, in_channels: int, out_channels: int, kernel_size, stride=1,
                 padding=0, output_padding=0, groups=1, bias=True, dilation=1, padding_mode='zeros'): ...
    def forward(self, input: Tensor, output_size=None) -> Tensor: ...

class ConvTranspose2d(Module):
    """2D transposed convolution layer."""
    def __init__(self, in_channels: int, out_channels: int, kernel_size, stride=1,
                 padding=0, output_padding=0, groups=1, bias=True, dilation=1, padding_mode='zeros'): ...
    def forward(self, input: Tensor, output_size=None) -> Tensor: ...

class ConvTranspose3d(Module):
    """3D transposed convolution layer."""
    def __init__(self, in_channels: int, out_channels: int, kernel_size, stride=1,
                 padding=0, output_padding=0, groups=1, bias=True, dilation=1, padding_mode='zeros'): ...
    def forward(self, input: Tensor, output_size=None) -> Tensor: ...

Activation Functions

Non-linear activation functions for introducing non-linearity.

class ReLU(Module):
    """Rectified Linear Unit: max(0, x)."""
    def __init__(self, inplace: bool = False): ...
    def forward(self, input: Tensor) -> Tensor: ...

class ReLU6(Module):
    """ReLU clamped to maximum value of 6."""
    def __init__(self, inplace: bool = False): ...
    def forward(self, input: Tensor) -> Tensor: ...

class LeakyReLU(Module):
    """Leaky ReLU: max(negative_slope * x, x)."""
    def __init__(self, negative_slope: float = 0.01, inplace: bool = False): ...
    def forward(self, input: Tensor) -> Tensor: ...

class PReLU(Module):
    """Parametric ReLU with learnable negative slope."""
    def __init__(self, num_parameters: int = 1, init: float = 0.25): ... 
    def forward(self, input: Tensor) -> Tensor: ...

class ELU(Module):
    """Exponential Linear Unit."""
    def __init__(self, alpha: float = 1.0, inplace: bool = False): ...
    def forward(self, input: Tensor) -> Tensor: ...

class SELU(Module):
    """Scaled Exponential Linear Unit."""
    def __init__(self, inplace: bool = False): ...
    def forward(self, input: Tensor) -> Tensor: ...

class GELU(Module):
    """Gaussian Error Linear Unit."""
    def __init__(self, approximate: str = 'none'): ...
    def forward(self, input: Tensor) -> Tensor: ...

class SiLU(Module):
    """Sigmoid Linear Unit (Swish): x * sigmoid(x)."""
    def __init__(self, inplace: bool = False): ...
    def forward(self, input: Tensor) -> Tensor: ...

class Mish(Module):
    """Mish activation: x * tanh(softplus(x))."""
    def __init__(self, inplace: bool = False): ...
    def forward(self, input: Tensor) -> Tensor: ...

class Sigmoid(Module):
    """Sigmoid activation: 1 / (1 + exp(-x))."""
    def __init__(self): ...
    def forward(self, input: Tensor) -> Tensor: ...

class Tanh(Module):
    """Hyperbolic tangent activation."""
    def __init__(self): ...
    def forward(self, input: Tensor) -> Tensor: ...

class Softmax(Module):
    """Softmax activation along specified dimension."""
    def __init__(self, dim=None): ...
    def forward(self, input: Tensor) -> Tensor: ...

class LogSoftmax(Module):
    """Log-Softmax activation."""
    def __init__(self, dim=None): ...
    def forward(self, input: Tensor) -> Tensor: ...

Normalization Layers

Normalization techniques for training stability and performance.

class BatchNorm1d(Module):
    """Batch normalization for 2D or 3D inputs."""
    def __init__(self, num_features, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True): ...
    def forward(self, input: Tensor) -> Tensor: ...

class BatchNorm2d(Module):
    """Batch normalization for 4D inputs."""
    def __init__(self, num_features, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True): ...
    def forward(self, input: Tensor) -> Tensor: ...

class BatchNorm3d(Module):
    """Batch normalization for 5D inputs."""  
    def __init__(self, num_features, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True): ...
    def forward(self, input: Tensor) -> Tensor: ...

class LayerNorm(Module):
    """Layer normalization."""
    def __init__(self, normalized_shape, eps=1e-05, elementwise_affine=True, bias=True): ...
    def forward(self, input: Tensor) -> Tensor: ...

class GroupNorm(Module):
    """Group normalization."""
    def __init__(self, num_groups: int, num_channels: int, eps=1e-05, affine=True): ...
    def forward(self, input: Tensor) -> Tensor: ...

class InstanceNorm1d(Module):
    """Instance normalization for 3D inputs."""
    def __init__(self, num_features, eps=1e-05, momentum=0.1, affine=False, track_running_stats=False): ...
    def forward(self, input: Tensor) -> Tensor: ...

class InstanceNorm2d(Module):
    """Instance normalization for 4D inputs."""
    def __init__(self, num_features, eps=1e-05, momentum=0.1, affine=False, track_running_stats=False): ...
    def forward(self, input: Tensor) -> Tensor: ...

class InstanceNorm3d(Module):
    """Instance normalization for 5D inputs."""  
    def __init__(self, num_features, eps=1e-05, momentum=0.1, affine=False, track_running_stats=False): ...
    def forward(self, input: Tensor) -> Tensor: ...

Pooling Layers

Pooling operations for spatial dimension reduction.

class MaxPool1d(Module):
    """1D max pooling."""
    def __init__(self, kernel_size, stride=None, padding=0, dilation=1, return_indices=False, ceil_mode=False): ...
    def forward(self, input: Tensor) -> Tensor: ...

class MaxPool2d(Module):
    """2D max pooling."""
    def __init__(self, kernel_size, stride=None, padding=0, dilation=1, return_indices=False, ceil_mode=False): ...
    def forward(self, input: Tensor) -> Tensor: ...

class MaxPool3d(Module):
    """3D max pooling."""
    def __init__(self, kernel_size, stride=None, padding=0, dilation=1, return_indices=False, ceil_mode=False): ...
    def forward(self, input: Tensor) -> Tensor: ...

class AvgPool1d(Module):
    """1D average pooling."""
    def __init__(self, kernel_size, stride=None, padding=0, ceil_mode=False, count_include_pad=True): ...
    def forward(self, input: Tensor) -> Tensor: ...

class AvgPool2d(Module):
    """2D average pooling."""
    def __init__(self, kernel_size, stride=None, padding=0, ceil_mode=False, count_include_pad=True, divisor_override=None): ...
    def forward(self, input: Tensor) -> Tensor: ...

class AvgPool3d(Module):
    """3D average pooling."""
    def __init__(self, kernel_size, stride=None, padding=0, ceil_mode=False, count_include_pad=True, divisor_override=None): ...
    def forward(self, input: Tensor) -> Tensor: ...

class AdaptiveMaxPool1d(Module):
    """1D adaptive max pooling."""
    def __init__(self, output_size, return_indices=False): ...
    def forward(self, input: Tensor) -> Tensor: ...

class AdaptiveMaxPool2d(Module):
    """2D adaptive max pooling."""
    def __init__(self, output_size, return_indices=False): ...
    def forward(self, input: Tensor) -> Tensor: ...

class AdaptiveAvgPool1d(Module):
    """1D adaptive average pooling."""
    def __init__(self, output_size): ...
    def forward(self, input: Tensor) -> Tensor: ...

class AdaptiveAvgPool2d(Module):
    """2D adaptive average pooling."""
    def __init__(self, output_size): ...
    def forward(self, input: Tensor) -> Tensor: ...

Loss Functions

Loss functions for training neural networks.

class MSELoss(Module):
    """Mean Squared Error loss."""
    def __init__(self, size_average=None, reduce=None, reduction='mean'): ...
    def forward(self, input: Tensor, target: Tensor) -> Tensor: ...

class L1Loss(Module):
    """Mean Absolute Error loss."""
    def __init__(self, size_average=None, reduce=None, reduction='mean'): ...
    def forward(self, input: Tensor, target: Tensor) -> Tensor: ...

class CrossEntropyLoss(Module):
    """Cross entropy loss for classification."""
    def __init__(self, weight=None, size_average=None, ignore_index=-100, reduce=None, reduction='mean', label_smoothing=0.0): ...
    def forward(self, input: Tensor, target: Tensor) -> Tensor: ...

class NLLLoss(Module):
    """Negative log likelihood loss."""
    def __init__(self, weight=None, size_average=None, ignore_index=-100, reduce=None, reduction='mean'): ...
    def forward(self, input: Tensor, target: Tensor) -> Tensor: ...

class BCELoss(Module):
    """Binary cross entropy loss."""
    def __init__(self, weight=None, size_average=None, reduce=None, reduction='mean'): ...
    def forward(self, input: Tensor, target: Tensor) -> Tensor: ...

class BCEWithLogitsLoss(Module):
    """Binary cross entropy with sigmoid."""
    def __init__(self, weight=None, size_average=None, reduce=None, reduction='mean', pos_weight=None): ...
    def forward(self, input: Tensor, target: Tensor) -> Tensor: ...

class KLDivLoss(Module):
    """Kullback-Leibler divergence loss."""
    def __init__(self, size_average=None, reduce=None, reduction='mean', log_target=False): ...
    def forward(self, input: Tensor, target: Tensor) -> Tensor: ...

class SmoothL1Loss(Module):
    """Smooth L1 loss (Huber loss)."""
    def __init__(self, size_average=None, reduce=None, reduction='mean', beta=1.0): ...
    def forward(self, input: Tensor, target: Tensor) -> Tensor: ...

class HuberLoss(Module):
    """Huber loss."""
    def __init__(self, reduction='mean', delta=1.0): ...
    def forward(self, input: Tensor, target: Tensor) -> Tensor: ...

Recurrent Neural Networks

RNN, LSTM, and GRU layers for sequential data processing.

class RNN(Module):
    """Multi-layer RNN with tanh or ReLU non-linearity."""
    def __init__(self, input_size, hidden_size, num_layers=1, nonlinearity='tanh', bias=True,
                 batch_first=False, dropout=0.0, bidirectional=False): ...
    def forward(self, input, h_0=None) -> Tuple[Tensor, Tensor]: ...

class LSTM(Module):
    """Multi-layer Long Short-Term Memory network."""
    def __init__(self, input_size, hidden_size, num_layers=1, bias=True, batch_first=False,
                 dropout=0.0, bidirectional=False, proj_size=0): ...
    def forward(self, input, hx=None) -> Tuple[Tensor, Tuple[Tensor, Tensor]]: ...

class GRU(Module):
    """Multi-layer Gated Recurrent Unit network."""
    def __init__(self, input_size, hidden_size, num_layers=1, bias=True, batch_first=False,
                 dropout=0.0, bidirectional=False): ...
    def forward(self, input, h_0=None) -> Tuple[Tensor, Tensor]: ...

class RNNCell(Module):
    """RNN cell."""
    def __init__(self, input_size, hidden_size, bias=True, nonlinearity='tanh'): ...
    def forward(self, input, hidden) -> Tensor: ...

class LSTMCell(Module):
    """LSTM cell."""
    def __init__(self, input_size, hidden_size, bias=True): ...
    def forward(self, input, hx=None) -> Tuple[Tensor, Tensor]: ...

class GRUCell(Module):
    """GRU cell."""
    def __init__(self, input_size, hidden_size, bias=True): ...
    def forward(self, input, hidden) -> Tensor: ...

Transformer Components

Transformer architecture components for attention-based models.

class Transformer(Module):
    """Complete transformer model."""
    def __init__(self, d_model=512, nhead=8, num_encoder_layers=6, num_decoder_layers=6,
                 dim_feedforward=2048, dropout=0.1, activation='relu', custom_encoder=None, custom_decoder=None): ...
    def forward(self, src, tgt, src_mask=None, tgt_mask=None, memory_mask=None) -> Tensor: ...

class TransformerEncoder(Module):
    """Stack of transformer encoder layers."""
    def __init__(self, encoder_layer, num_layers, norm=None): ...
    def forward(self, src, mask=None, src_key_padding_mask=None) -> Tensor: ...

class TransformerEncoderLayer(Module):
    """Single transformer encoder layer."""
    def __init__(self, d_model, nhead, dim_feedforward=2048, dropout=0.1, activation='relu', batch_first=False): ...
    def forward(self, src, src_mask=None, src_key_padding_mask=None) -> Tensor: ...

class TransformerDecoder(Module):
    """Stack of transformer decoder layers."""
    def __init__(self, decoder_layer, num_layers, norm=None): ...
    def forward(self, tgt, memory, tgt_mask=None, memory_mask=None) -> Tensor: ...

class TransformerDecoderLayer(Module):
    """Single transformer decoder layer."""
    def __init__(self, d_model, nhead, dim_feedforward=2048, dropout=0.1, activation='relu', batch_first=False): ...
    def forward(self, tgt, memory, tgt_mask=None, memory_mask=None) -> Tensor: ...

class MultiheadAttention(Module):
    """Multi-head attention mechanism."""
    def __init__(self, embed_dim, num_heads, dropout=0.0, bias=True, add_bias_kv=False, add_zero_attn=False, kdim=None, vdim=None, batch_first=False): ...
    def forward(self, query, key, value, key_padding_mask=None, need_weights=True, attn_mask=None) -> Tuple[Tensor, Tensor]: ...

Container Classes

Containers for organizing and combining multiple modules.

class Sequential(Module):
    """Sequential container of modules."""
    def __init__(self, *args): ...
    def forward(self, input): ...

class ModuleList(Module):
    """List container for modules."""
    def __init__(self, modules=None): ...
    def append(self, module): ...
    def extend(self, modules): ...
    def insert(self, index, module): ...

class ModuleDict(Module):
    """Dictionary container for modules."""
    def __init__(self, modules=None): ...
    def __getitem__(self, key): ...
    def __setitem__(self, key, module): ...
    def keys(self): ...
    def items(self): ...
    def values(self): ...

class ParameterList(Module):
    """List container for parameters."""
    def __init__(self, parameters=None): ...
    def append(self, parameter): ...
    def extend(self, parameters): ...

class ParameterDict(Module):
    """Dictionary container for parameters."""
    def __init__(self, parameters=None): ...
    def __getitem__(self, key): ...
    def __setitem__(self, key, parameter): ...

Dropout and Regularization

Regularization techniques to prevent overfitting.

class Dropout(Module):
    """Randomly zeros elements with probability p."""
    def __init__(self, p=0.5, inplace=False): ...
    def forward(self, input: Tensor) -> Tensor: ...

class Dropout1d(Module):
    """1D channel-wise dropout."""
    def __init__(self, p=0.5, inplace=False): ...
    def forward(self, input: Tensor) -> Tensor: ...

class Dropout2d(Module):
    """2D channel-wise dropout.""" 
    def __init__(self, p=0.5, inplace=False): ...
    def forward(self, input: Tensor) -> Tensor: ...

class Dropout3d(Module):
    """3D channel-wise dropout."""
    def __init__(self, p=0.5, inplace=False): ...
    def forward(self, input: Tensor) -> Tensor: ...

class AlphaDropout(Module):
    """Alpha dropout for SELU networks."""
    def __init__(self, p=0.5, inplace=False): ...
    def forward(self, input: Tensor) -> Tensor: ...

Embedding Layers

Embedding layers for discrete inputs like words or tokens.

class Embedding(Module):
    """Lookup table for embeddings."""
    def __init__(self, num_embeddings: int, embedding_dim: int, padding_idx=None, max_norm=None,
                 norm_type=2.0, scale_grad_by_freq=False, sparse=False): ...
    def forward(self, input: Tensor) -> Tensor: ...

class EmbeddingBag(Module):
    """Embedding bag for variable length sequences."""
    def __init__(self, num_embeddings: int, embedding_dim: int, max_norm=None, norm_type=2.0,
                 scale_grad_by_freq=False, mode='mean', sparse=False, include_last_offset=False): ...
    def forward(self, input: Tensor, offsets=None, per_sample_weights=None) -> Tensor: ...

Usage Examples

Simple Neural Network

import torch
import torch.nn as nn
import torch.optim as optim

class SimpleNet(nn.Module):
    def __init__(self, input_size, hidden_size, output_size):
        super(SimpleNet, self).__init__()
        self.layers = nn.Sequential(
            nn.Linear(input_size, hidden_size),
            nn.ReLU(),
            nn.Linear(hidden_size, hidden_size),
            nn.ReLU(),
            nn.Linear(hidden_size, output_size)
        )
    
    def forward(self, x):
        return self.layers(x)

# Initialize model
model = SimpleNet(784, 128, 10)
criterion = nn.CrossEntropyLoss()
optimizer = optim.Adam(model.parameters(), lr=0.001)

# Example forward pass
x = torch.randn(32, 784)  # Batch of 32 samples
y = torch.randint(0, 10, (32,))  # Labels

output = model(x)
loss = criterion(output, y)

# Backward pass
optimizer.zero_grad()
loss.backward()
optimizer.step()

print(f"Loss: {loss.item()}")

Convolutional Neural Network

import torch
import torch.nn as nn

class CNN(nn.Module):
    def __init__(self, num_classes=10):
        super(CNN, self).__init__()
        self.features = nn.Sequential(
            nn.Conv2d(3, 32, kernel_size=3, padding=1),
            nn.ReLU(inplace=True),
            nn.MaxPool2d(kernel_size=2),
            nn.Conv2d(32, 64, kernel_size=3, padding=1),
            nn.ReLU(inplace=True),
            nn.MaxPool2d(kernel_size=2),
            nn.Conv2d(64, 128, kernel_size=3, padding=1),
            nn.ReLU(inplace=True),
            nn.AdaptiveAvgPool2d((1, 1))
        )
        self.classifier = nn.Sequential(
            nn.Dropout(0.5),
            nn.Linear(128, num_classes)
        )
    
    def forward(self, x):
        x = self.features(x)
        x = torch.flatten(x, 1)
        x = self.classifier(x)
        return x

# Initialize model
model = CNN(num_classes=10)

# Example forward pass
x = torch.randn(8, 3, 32, 32)  # Batch of images
output = model(x)
print(f"Output shape: {output.shape}")

LSTM for Sequence Processing

import torch
import torch.nn as nn

class LSTMModel(nn.Module):
    def __init__(self, vocab_size, embed_size, hidden_size, num_layers, num_classes):
        super(LSTMModel, self).__init__()
        self.embedding = nn.Embedding(vocab_size, embed_size)
        self.lstm = nn.LSTM(embed_size, hidden_size, num_layers, batch_first=True)
        self.fc = nn.Linear(hidden_size, num_classes)
        self.dropout = nn.Dropout(0.5)
    
    def forward(self, x):
        embedded = self.embedding(x)
        lstm_out, (hidden, cell) = self.lstm(embedded)
        # Use the last output
        output = self.fc(self.dropout(lstm_out[:, -1, :]))
        return output

# Initialize model
model = LSTMModel(vocab_size=10000, embed_size=128, hidden_size=256, num_layers=2, num_classes=5)

# Example forward pass
x = torch.randint(0, 10000, (16, 50))  # Batch of sequences
output = model(x)
print(f"Output shape: {output.shape}")

Install with Tessl CLI