📦 Latent-Based Continual Learning with Dual-Layered Distillation and a Streamlined U- Net for Efficient Text-to-Image Generation

A PyTorch-based framework for Latent-Based Continual Learning with Dual-Layered Distillation and a Streamlined U- Net for Efficient Text-to-Image Generation — a compact, efficient diffusion model that reuses weights across timesteps, distilled for faster sampling without quality loss.

---

📌 Table of Contents

• Overview
• Key Features
• Installation
• Usage
  • Training
  • Sampling
• Configuration
• Examples
• Benchmark & Results
• Citation
• License

---

📈 Overview

WSDD aims to accelerate diffusion sampling by:

• Sharing neural network weights across multiple timesteps.
• Distilling a full-scale teacher model into a lightweight student.
• Striking a balance between sampling speed and image fidelity.

The result is a model that runs significantly faster (e.g., 4–8× speedup) with perceptually similar output to Full Diffusion.

---

✨ Key Features

• 🔁 Weight Sharing across distillation steps
• 🎯 Step Reduction via progressive distillation
• 🛠️ Fully compatible with Hugging Face diffusers pipelines
• 🔋 Supports CUDA/FP16 inference
• 🧠 Extensible modular architecture

---

🛠️ Installation

git clone https://github.com/naimurborno/WSDD-Weight-Shared-Distilled-Diffusion.git
cd WSDD-Weight-Shared-Distilled-Diffusion

# Install framework and dependencies
pip install -r requirements.txt
python train.py \
  --teacher_model "sd-full" \
  --student_scales 32 64 96 128 \
  --sigmas 1.0 0.9 0.8 0.6 0.0 \
  --batch_size 32 \
  --steps_per_epoch 1000 \
  --epochs 10 \
  --output_dir "checkpoints/wsdd"
from wsdd import WSDDPipeline

pipe = WSDDPipeline.from_pretrained("checkpoints/wsdd/latest.ckpt", device="cuda")
img = pipe(
    prompt="a serene mountain lake at sunrise",
    scales=[32,64,96,128],
    sigmas=[1.0,0.9,0.8,0.6,0.0],
    guidance_scale=7.5
).images[0]

img.save("output.png")