pytorch

PyTorch: Deep Learning Framework

Overview

PyTorch is an open-source machine learning framework that accelerates the path from research prototyping to production deployment. Apply this skill for building neural networks, tensor computations, automatic differentiation, GPU training, and custom deep learning models.

When to Use This Skill

This skill should be used when:

• Building custom neural network architectures
• Working with tensors and matrix operations
• Training deep learning models on GPU
• Implementing custom loss functions and optimizers
• Doing research prototyping with automatic differentiation
• Deploying models to production (TorchScript, ONNX)
• Working with distributed training
• Using pretrained models from torchvision, torchaudio

Quick Start

Basic Import and Setup

import torch
import torch.nn as nn
import torch.optim as optim
import torch.nn.functional as F

# Check GPU availability
device = torch.device('cuda' if torch.cuda.is_available() else 'cpu')
print(f"Using device: {device}")

# Version info
print(f"PyTorch version: {torch.__version__}")

Tensor Operations

# Create tensors
x = torch.randn(3, 4)  # Random normal
x = torch.zeros(3, 4)  # Zeros
x = torch.ones(3, 4)  # Ones
x = torch.tensor([1, 2, 3])  # From data

# Move to GPU
x = x.to(device)

# Basic operations
y = torch.randn(3, 4)
z = x + y  # Addition
z = torch.matmul(x, y.T)  # Matrix multiplication
z = x.mean()  # Reduction

# Reshape
x = x.view(-1)  # Flatten
x = x.reshape(2, 6)  # Reshape
x = x.unsqueeze(0)  # Add dimension

Autograd (Automatic Differentiation)

# Enable gradient tracking
x = torch.randn(3, 4, requires_grad=True)
y = torch.randn(3, 4, requires_grad=True)

# Forward pass
z = x * y
loss = z.sum()

# Backward pass
loss.backward()

# Access gradients
print(x.grad)  # Gradient of loss w.r.t. x
print(y.grad)  # Gradient of loss w.r.t. y

# Disable gradient tracking
with torch.no_grad():
    # Operations here won't track gradients
    pass