Boubou runtimes

Prebuilt runtimes that Boubou downloads on first use. Boubou's manifests pin each file by URL, size and SHA-256, and Boubou verifies it before use.

Path	Contents
qwen/linux-x64/<sha256>.tar.gz	Speech-to-speech runner for x86-64 Linux with an NVIDIA GPU: portable CPython 3.11, PyTorch 2.11 with NVIDIA's CUDA 13 libraries, llama.cpp b11960 and Hugging Face speech-to-speech 1.1.0. Built by packaging/qwen/build_runner.py.

Uploaded files are never replaced; a new runner gets a new path.

Third-party components keep their own licenses. Their license texts are inside each archive: under licenses/, in llama-b11960/LICENSE and in each Python package's .dist-info. NVIDIA's CUDA libraries are redistributed under the NVIDIA license included with them. Model weights are not included; Boubou downloads them from their own repositories.