AdaFace IR-18 face embedding on the AMD NPU
An IRON export of minchul/cvlface_adaface_ir18_webface4m for AMD Ryzen AI NPUs: the compiled NPU kernels (.xclbin + instruction streams) and the packed weights the taconite-adaface Rust runtime replays, over XRT or directly over the amdxdna driver.
The kernels are compiled for NPU2 (AIE2P: Strix Point, Strix Halo, Krackan) and will not load on NPU1 (Phoenix, Hawk Point).
Download
hf download brishen/iron-adaface-ir18-npu2 --local-dir adaface-ir18
# or, from an IRON checkout:
python scripts/hf_models.py download adaface-ir18 --repo brishen/iron-adaface-ir18-npu2 --out adaface-ir18
Usage
[dependencies]
taconite-adaface = "0.1"
# or, with no XRT at all (straight to the amdxdna driver):
# taconite-adaface = { version = "0.1", default-features = false, features = ["direct"] }
let mut emb = taconite_adaface::IrEmbedder::load("adaface-ir18")?;
// an aligned 112x112 face, CHW f32, ((rgb / 255) - 0.5) / 0.5
let embedding = emb.embed(&chw)?; // 512-d; compare faces by cosine similarity
See taconite-adaface (docs, source) for the full API. The bundle is exported by IRON's iron/applications/adaface_ir18.
Provenance
- Upstream model:
minchul/cvlface_adaface_ir18_webface4m(license:other; its terms apply to these weights) - IRON commit:
204c4ac - Uploaded: 2026-09-23
- Files: 35, 46.9 MB