Last Updated: November 21, 2025
PyTorch
Deep learning framework by Meta
Core Concepts
| Item | Description |
|---|---|
Tensor
|
Multi-dimensional array |
Autograd
|
Automatic differentiation |
nn.Module
|
Base class for models |
Optimizer
|
Updates model parameters |
DataLoader
|
Batch data loading |
GPU Support
|
CUDA acceleration |
Basic Example
import torch
import torch.nn as nn
# Define model
class Net(nn.Module):
def __init__(self):
super().__init__()
self.fc1 = nn.Linear(784, 128)
self.fc2 = nn.Linear(128, 10)
def forward(self, x):
x = torch.relu(self.fc1(x))
return self.fc2(x)
model = Net()
criterion = nn.CrossEntropyLoss()
optimizer = torch.optim.Adam(model.parameters())
# Training step
outputs = model(inputs)
loss = criterion(outputs, labels)
loss.backward()
optimizer.step()Common Operations
| Item | Description |
|---|---|
torch.tensor()
|
Create tensor |
tensor.cuda()
|
Move to GPU |
model.train()
|
Set training mode |
model.eval()
|
Set evaluation mode |
torch.save()
|
Save model |
torch.load()
|
Load model |
Best Practices
- Use DataLoader for efficient batching
- Move data and model to same device (CPU/GPU)
- Use torch.no_grad() for inference
- Clear gradients with optimizer.zero_grad()
💡 Pro Tips
Quick Reference
PyTorch provides dynamic computation graphs