Interactive neural architecture playgrounds. Hover to preview, click to explore.
Multi-head self-attention driving modern generative AI.
Introduced
2017
Layers
6-12+
Params
100M+
Fully connected layers with backpropagation and activation flows.
3
Weights
35
Biases
8
Convolutional filters scanning for spatial hierarchies.
1989
Kernel
3×3
Stride
1 / 2
Sequential memory through hidden state recurrence and gates.
1986
Steps
4
Hidden
128
Adversarial training where G and D compete, improving together.
2014
Loss
BCE
Latent
100
Iterative denoising from pure noise to structured output.
10
Grid
8×8
Noise
Gaussian