401 KB single binary. DSpark 572 tok/s (5.90x spec-decode). 73+ models. NPU + GPU + CPU. Zero Python. MIT.
vulkan zig image-generation mit-license video-generation model-agnostic cplusplus-23 audio-generation local-llm open-source-ai stable-audio amd-strix-halo npu-inference one-binary-to-rule-them-all fused-engine xdna-2 zero-python zaya amd-native
-
Updated
Jul 8, 2026 - C++