🔩 Deeppy

Deeppy is a flexible deep learning framework built on PyTorch, designed to simplify training workflows while supporting powerful research capabilities. It embraces a modular approach by decoupling data, algorithms, and neural networks, making it easy to swap components, experiment with new ideas, and customize pipelines end-to-end.

✨ Key Features

• 🔧 Modular by Design – Swap networks, algorithms, and data pipelines easily
• 💡 Research-Oriented – Designed for flexibility and prototyping
• 📉 Integrated Plotting & Logging – Visualize your training instantly
• 🔍 XAI Tooling (Coming Soon) – Make black-box models more interpretable
• ⚡ Built with PyTorch – GPU, torch compile and AMP support and compatibility by default

🧰 Current Models and Networks

• 🕹️ Reinforcement Learning

  • ⚙️ DQN 📝 Tutorial

  • ⚙️ Double DQN - 📝 Tutorial

  • ⚙️ SAC - 📝 Tutorial

• 📢 Natural Language Processing

  • ⚙️ GPT 📝 Tutorial

• 🖼️ Computer Vision

  • ⚙️ SANE 📝 Tutorial

  • ⚙️ NeRF

• 🔍 Autoencoders

  • ⚙️ B-VAE

• 🔬 Basic Model

  • ⚙️ Model

    📝 Introduction Tutorial

    📝 Advanced Tutorial

⚡ Quickstart Demos

🧠 Train Your Own GPT Model

📝 Step 1: Load Your Dataset

with open("assets/shakespeare.txt", "r", encoding = "utf-8") as f:
    text = f.read()

encoding = tiktoken.encoding_for_model("gpt-2")
data = GPTText(text=text, tokenizer=encoding, context_size = context_size)

🧱 Step 2: Build the GPT-Model

GPT_params = {
    "optimizer_params":Optimizer_params,
    "vocab_size":vocab_size,
    "embed_dim":embed_dim,
    "num_heads":num_heads,
    "num_layers":num_layers,
    "context_size":context_size,
    "device":device,
    "criterion":nn.CrossEntropyLoss(ignore_index = -1),
}

model = GPT(GPT_params)

📊 Total Parameters: ~28.9M

🔁 Step 3: Train the Model

lf = LearnFrame(model,data)

for i in range(epoch):
    lf.optimize()
lf.plot(show_result=True, log=True)

✍️ Step 4: Generate Text

model.generate("KING RICHARD III: \n On this very beautiful day, let us")

KING RICHARD III:

On this very beautiful day, let us us hear

The way of the king.

DUKE OF YORK::

I will not be avoided'd with my heart.

DUKEKE VINCENTIO:

I thank you, good father.

LLUCIO:

I thank you, good my lord; I'll to your your daughter.

KING EDWARD IV:

Now, by the jealous queen

🎮 Train a RL Agent

🌍 Step 1: Set Up Environment

import deeppy as dp

env = gym.make("LunarLander-v1")
data = dp.EnvData(env, buffer_size=100000)

🧠 Step 2: Create Your Network

policy_network = {
    "layers" : [obs,128,128,act],
    "blocks" : [nn.Linear, nn.ReLU]
    "out_act" : nn.Softmax,
    "weight_init" : "uniform"
}

⚙️ Step 3: Choose a RL Algorithm

model = dp.SAC(sac_params) #Soft Actor Critic

🧪 Step 4: Train the Agent

lf = dp.LearningFrame(model, data)

for epoch in range(EPOCH):
	#Take one step in environment using the model
	lf.collect()
	#Train SAC one step
	lf.optimize()
#Automatic plotting 
lf.plot()

🎥 Step 5: Watch Your Agent

lf.get_anim()

💾 Step 6: Save / Load Your Model

lf.save(file_name)
lf.load(file_name)

📚 Explore More Tutorials

Looking to dive deeper? We've included hands-on examples covering everything from GPT training to reinforcement learning agents like LunarLander.

📂 Find them all in the tutorials/ folder.

🛠️ Setup

pip install -r requirements.txt

📖 Documentation

Documentation is in progress. Meanwhile, refer to the tutorials folder.

🧪 To-Do / Planned Features

RL

RL

Dueling DQN

PPO

MBPO (Model-Based Policy Optimization)

SafeMBPO

XAI Tools (Explainable AI)

Neuroevolution