pyturk

pyturk is a minimal educational deep-learning framework inspired by micrograd and PyTorch. It is designed for clarity and teaching, not for production use.

Features

• Simple scalar autograd (Value)
• PyTorch-like nn modules: Module, Linear, Sequential, MLP
• Optimizers: SGD, Adam, RMSProp and LR schedulers
• Lightweight datasets and DataLoader for experiments
• Utilities: graph visualization, logging, and simple stats

Quick start

from pyturk import Value
import pyturk.nn as nn
import pyturk.optim as optim

model = nn.Sequential(
    nn.Linear(2, 16),
    nn.ReLU(),
    nn.Linear(16, 1),
)

optimizer = optim.SGD(model.parameters(), lr=1e-2)

# Single sample forward
x = [Value(1.0), Value(2.0)]
out = model(x)
loss = (out - Value(1.0)) ** 2

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

See docs/usage.md and docs/api.md for more examples and reference.