srdjan/zlora

A small Zig inference POC that demonstrates LoRA adapters and steering vectors as runtime “sidecar” modules for a Llama-2-style Transformer. Both can be hot-swapped during inference without reloading base weights, and they compose naturally since they operate at different points in the computation.

语言：Zig · GitHub 星标：6 · Fork：1

话题：ai · inference-engine · lora · zig · ziglang

在 GitHub 上查看原仓库