Light

Gold Verified

Publisher: EverestQ Engineering · Category: Chat

Parameters

Context Window

16K

Default Size

2.1 GB

Downloads

310,000

Lightweight edge model engineered specifically for mobile devices, laptops, and resource-constrained micro-servers.

PULL MODEL

eq pull everestq/light

RUN INFERENCE

eq run everestq/light "Format input JSON into bullet list"

Attention Architecture

Utilizes Grouped-Query Attention (GQA) with 16-token fixed Paged KV Cache block allocation for zero-fragmentation memory residency.

ONNX & EQC Compatible

Compiles directly through the EQC toolchain into standalone .eqx binary packages with fused SwiGLU kernels.

Multi-Hardware Support

Auto-detects CUDA RTX/A100/H100, Apple Silicon Metal Performance Shaders, or AVX-512 CPU execution backends.