A lightweight and modular deep learning framework

---

🚀 Features

• 🔁 Automatic Differentiation (autograd engine built from scratch)
• 🧱 Modular Layers (build models layer by layer)
• 🎯 Optimizers (SGD, Adam, RMSProp etc.)
• 📊 Metrics & Losses (MSE, CrossEntropy, etc.)
• 🧪 Custom Training Loops with flexibility
• 🧵 Numpy-based backend (no heavy dependencies)

---

🧠 Quick Start

from deepgrad import Tensor, Linear, MSELoss, SGD

# Dummy training example
x = Tensor([[1.0], [2.0]])
y = Tensor([[2.0], [4.0]])

model = Linear(1, 1)
loss_fn = MSELoss()
optimizer = SGD(model.parameters(), lr=0.01)

for epoch in range(100):
    pred = model(x)
    loss = loss_fn(pred, y)

    optimizer.zero_grad()
    loss.backward()
    optimizer.step()

    print(f"Epoch {epoch}, Loss: {loss.data:.4f}")