DGN Kernel

Discrete Gauge Network · WASM SIMD128

Channels

64
independent permutation tables

Context

1024
positions per channel

Layers

32
depth of the gauge network

SIMD

128-bit
WASM i8x16 swizzle

Architecture

The Discrete Gauge Network (DGN) is a permutation-based neural architecture that uses a static 64-byte bijective lookup table per channel. The forward pass computes output = table[ (state XOR gauge) & 0x3F ] across all positions simultaneously using SIMD vector registers.

Memory Layout

State Tensor:  LAYERS x CHANNELS x CONTEXT_SZ = 32 x 64 x 1024 = 2 MB
Gauge Tensor:  CHANNELS x CONTEXT_SZ           = 64 x 1024     = 64 KB
Perm Tables:   LAYERS x CHANNELS x 64          = 32 x 64 x 64  = 128 KB

API

GET /

Landing page

GET /status

Runtime health and configuration

POST /forward

Run layer forward pass

POST /gauge

Compute dyadic gauge field

POST /swap

Execute bijective swap

DGN Kernel v0.1.0