Torch Model Viewer

Visualize PyTorch weights in 1D, 2D & 3D — instantly!

A 2-file, zero-install web dashboard to explore .pth / .pt models.
No training, no setup — just drop in your model and see every layer in colorful, interactive 3D.

---

Features

Feature	Description
3D Point Cloud	Red = low values, Blue = high values, Size scales with magnitude
2D Heatmap	Full-resolution layer view (auto-scaled)
1D Line Chart	Raw weight values (sampled)
Histogram	Distribution of values per layer
Smart Indicators	Overfit, dead weights, extreme values
Dark Mode UI	Clean, collapsible panels

---

Quick Start

# 1. Clone & enter
git clone https://github.com/jdwebprogrammer/torch-model-viewer.git
cd torch-model-dashboard

# 2. Install (once)
pip install flask torch pillow numpy

# 3. Run!
python app.py

Open http://127.0.0.1:5000 → Upload your .pth or .pt file → Explore!

---

How It Works

• Backend (app.py): Loads model → extracts weights → computes stats, heatmaps, 3D layouts
• Frontend (templates/index.html): Three.js + Chart.js → interactive 3D orbit + responsive UI
• Smart Sampling: Up to 10,000 points per layer → smooth performance even on huge models

---

Why Use It?

• Spot dead neurons (all red, tiny)
• Find exploding gradients (huge blue points)
• Compare layer patterns across models
• Debug quantization, pruning, or training bugs

---

Project Structure

torch-model-dashboard/
├── app.py              Flask + Torch backend
├── templates/
│   └── index.html      Three.js + Chart.js frontend
└── README.md

---

License

MIT — Free to use, modify, and share.

---

> Made with **love** for the PyTorch community  
> *“See your weights. Understand your model.”*

---