Repository navigation
CPU llama.cpp and authenticated whisper.cpp sidecars for Linux and Windows, plus a Linux CUDA SM120a llama.cpp profile. Bundles include pinned-source provenance and license records.
The CUDA profile passed automated native acceptance on an RTX 5080 with Qwen3.5-9B Q8_0, a 32,768-token context, all 34 offloadable layers on the GPU, authenticated inference, unauthorized-request rejection and owned-process cleanup. Its acceptance receipt is included in the profile. This verifies the native boundary; application quality, latency and packaged-service acceptance are separate work.
Publication verifies CPU bundle startup on Linux and Windows and attaches all nine bundle, checksum and provenance assets before making the release immutable.