v0.1.0

infer-core v0.1.0

Native library (desktop — ONNX Runtime)

• infer-core-windows-x86_64.zip — lib/infer_core.dll + lib/infer_core.dll.lib
• infer-core-windows-aarch64.zip — lib/infer_core.dll + lib/infer_core.dll.lib

Native library (Android — MNN, no ORT bundled)

• infer-core-android-arm64-v8a.zip — jniLibs/arm64-v8a/ (libinfer_core.so + MNN runtime .so)
• infer-core-android-x86_64.zip — jniLibs/x86_64/*.so (emulator)

Android OCR/embed use MNN model packs below.

Model packs — ONNX (desktop)

• ocr.paddle.ppocr6-tiny.onnx.fp32 — default chat OCR (~8 MB)
• ocr.paddle.ppocr6-small.onnx.fp32 — optional upgrade
• ocr.paddle.ppocr6-medium.onnx.fp32 — optional upgrade
• embed.mobileclip2-s0.onnx.fp32 — icon embedding（桌面 fp32，不做 ONNX 权重量化）
• icons.bundled.v1.mobileclip2-s0.int8 / .fp32 — bundled icon index

Model packs — MNN (mobile / Android)

• ocr.paddle.ppocr6-{tiny,small,medium}.mnn.fp32
• embed.mobileclip2-s0.mnn.{int8,fp32}

Icon index (icons.bundled.v1.*) is shared across desktop and mobile — not duplicated for MNN.

SHA256 checksums: SHA256SUMS.txt

Model packs cache (model-packs-latest)

Reusable model packs for release workflow.