2. Hardware Benchmark Suite & Real-time GFLOPS Profiler
Profiles real-time WebGPU compute throughput (GFLOPS) against CPU NumPy and executes browser-native Python validation test suites.
Matrix Multiplication (Matmul) Throughput Benchmark
[ Idle ]Python Test Suites
Select a test script to execute directly in WebGPU:
test_autograd.py
Reverse-mode autodiff & DAG gradient checks
test_matrix_ops.py
Tiled GEMM matmul & transpose kernels
test_optimizers.py
In-place SGD & Adam optimizer kernels
test_broadcasting.py
N-D coordinate broadcasting strides
test_security.py
Prototype pollution & tensor memory lock
Console Logs
--- AMEVA-Forge Benchmark Environment Ready ---
Select a benchmark dimension or Python test suite to begin.
AMEVA-Forge Python Matmul Benchmark Script
[ PyTorch Drop-in API ]import forge as torch
import numpy as np
import time
# 1. NumPy CPU Matmul
A_np = np.random.randn(N, N).astype(np.float32)
B_np = np.random.randn(N, N).astype(np.float32)
C_cpu = A_np @ B_np
# 2. AMEVA-Forge WebGPU Matmul (Tiled WGSL GEMM)
A_gpu = torch.tensor(A_np, device="gpu")
B_gpu = torch.tensor(B_np, device="gpu")
C_gpu = A_gpu @ B_gpu
await C_gpu.realize() # Realizes GPU texture & buffers