3. WebGPU Real-time Training Demo & Interactive Decision Boundary
Live neural network training directly on your GPU via WebGPU. Watch the loss curve converge and the 2D decision boundary evolve frame-by-frame in real-time.
1. Select Dataset Preset & Hyperparameters
2. Live 2D Decision Boundary & Loss Curve
[ Ready ]2D Neural Decision Field
Real-time Loss Convergence
3. WebGPU Execution Log Console
System idle. Ready to train.
[gpuCore.ts] WebGPU Context initialized.
[Pyodide Bridge] Autograd and in-place tensor buffers registered.
Click [Run Live WebGPU Training] to start.
AMEVA-Forge Python Autograd & In-Place Optimizer Loop (Running on WebGPU)
[ loss.backward() & optim.Adam ]import forge as torch
import forge.nn as nn
import forge.optim as optim
# 1. Multi-Layer Perceptron on WebGPU
model = nn.Sequential(
nn.Linear(2, 32),
nn.ReLU(),
nn.Linear(32, 16),
nn.ReLU(),
nn.Linear(16, 1)
).to("gpu")
opt = optim.Adam(model.parameters(), lr=0.04)
# 2. In-Browser Reverse-Mode Autograd Training Loop
for epoch in range(60):
opt.zero_grad()
pred = model(X_gpu)
loss = torch.sum((pred - Y_gpu) ** 2) * 0.5
loss.backward() # WebGPU GPU Kernel Gradient Backpropagation!
opt.step() # WebGPU In-Place VRAM Parameter Update!