Last updated: 2026-09-07

Inference Runtime

ZINC_RT ISA T-CPU Module

All API Sections

Pure Zig T-CPU opcode implementations.

The modules exported here are clarity-first reference kernels used by the CPU ring and by future cross-tier validation tests.

6 exports shown

constant

rms_norm

#
pub const rms_norm = @import("rms_norm.zig")

Scalar RMS normalization with learned per-channel weight, used as the CPU oracle for layer norm.

src/zinc_rt/isa/cpu_zig/mod.zig:7

constant

swiglu

#
pub const swiglu = @import("swiglu.zig")

Scalar SwiGLU activation (`silu(gate) * up`) used by dense and MoE MLP paths.

src/zinc_rt/isa/cpu_zig/mod.zig:9

constant

argmax

#
pub const argmax = @import("argmax.zig")

Deterministic argmax over a logits vector; picks the lowest index on ties.

src/zinc_rt/isa/cpu_zig/mod.zig:11

constant

dequant

#
pub const dequant = @import("dequant.zig")

Shared GGML row dequantization and quantized row dot-product helpers consumed by other CPU ops.

src/zinc_rt/isa/cpu_zig/mod.zig:13

constant

embed

#
pub const embed = @import("embed.zig")

Token embedding lookup: dequantize one row of a GGUF embedding matrix into f32 hidden state.

src/zinc_rt/isa/cpu_zig/mod.zig:15

constant

lm_head

#
pub const lm_head = @import("lm_head.zig")

LM head projection: multiply a GGUF output matrix by the final hidden state to produce vocab logits.

src/zinc_rt/isa/cpu_zig/mod.zig:17