Generate an AI model with random weights in 4 seconds. No training, pure BS.

pip install -r requirements.txt
python app.py

Generate completely random model weights using any HuggingFace model ID. Downloads only the config.json (a few KB - no weights), then creates a deterministic random model from your seed string.

Design your own model architecture from scratch with the Config Builder. Randomize the entire structure (layers, hidden size, attention heads, etc.) using a seed string - no HuggingFace download required.

Control how randomness spreads across your model's layers:

• Canvas Drawing - Draw a curve to set weight intensity per layer. High points = more random, low points = less random
• Math Equation - Use formulas like sin(layer * 0.8) * 0.02 + 0.01 to define per-layer behavior
• Random Normal - Standard random noise (the baseline)
• Random Walk - Layers drift together, weights gradually shift up or down

The canvas lets you paint the weight distribution directly instead of guessing parameters.

• TurboQuant by back2matching - quantization support