NInferEZ Engine 1.0.0

NInferEZ Engine 1.0.0

Independent Windows distribution of iamwavecut/ninfer-all, with complete upstream attribution and third-party notices.

• Separate Windows x64 packages for sm86, sm89 and sm120a.
• Approved engine artwork, machine-readable version/capabilities contract and host-only artifact inspector.
• App-local CUDA, media, networking and Visual C++ dependencies; no developer toolchain required to run.
• NInfer v3 artifacts with the architectures, encodings and embedded components supported by the selected build. The format version alone does not guarantee every model is supported.
• Manager 1.0.0 downloads and verifies the appropriate package; the engine can also run independently.

The 1.0.0 host programs are compiled against the existing inference libraries and CUDA kernels from core build 69c75330. This release changes product identity and distribution, not inference precision, kernel routes or benchmark settings. Previously recorded RTX 5090 measurements are linked from the model cards; they are not new benchmark runs on this release.

All packages are unsigned. sm86 and sm89 remain community-preview targets without local hardware qualification. RTX 5090 operation was tested in the preceding build; other compatible Blackwell cards remain unqualified. A successful compilation is not a guarantee that a model fits every card's VRAM.

Downloads include SHA-256 checksums, a package manifest, SPDX file inventory and dependency licenses. Model weights are not bundled.

Manager · Converter · OrcaRouter models · Swift model

FFmpeg corresponding source

FFmpeg 8.1.2 shared libraries use LGPL 2.1 or later. Download the complete corresponding source, including the exact patched source, port patches, original verified tarball and build records. DLL hashes and patch hashes were checked against the actual distributed libraries. This covers all three Windows GPU packages; their executable and DLL bytes remain unchanged.

Source/build details · Licenses. Version remains 1.0.0.