Hypnos v0.4.0

Prebuilt archives for v0.4.0. Unpack, then follow README.txt inside
(download models with scripts/fetch-models, run hypnos, open http://127.0.0.1:9999).

Archive	Needs
linux-x64-cuda	glibc 2.39+, NVIDIA driver 570+, CUDA 12 runtime (cudart, cuBLAS), libgomp1
windows-x64-cuda	NVIDIA driver 570+, plus windows-x64-cuda12-runtime.zip unpacked beside hypnos.exe (or the CUDA Toolkit 12.x)
macos-arm64-metal	Apple Silicon, macOS 14+

Desktop app (a chat client with live server stats; it connects to a running Hypnos):
hypnos-desktop-*-macos-arm64.dmg, -windows-x64.msi / .setup.exe,
-linux-x64.deb / .AppImage. The macOS app and server are signed and notarized. The Windows installers are not signed: choose More info > Run anyway.

Document conversion on the GPU (docling with device = "gpu", off by default) additionally
needs driver 580+, the CUDA 13 runtime and cuDNN 9 for CUDA 13; see README.txt in the archive.

Docker (CUDA): docker pull ghcr.io/advancedlogic/hypnos:0.4.0

The binaries link espeak-ng statically, so each archive is distributed under the GNU GPL
v3 or later (text included); Hypnos's own code is MIT OR Apache-2.0. Corresponding source:
https://github.com/advancedlogic/hypnos/tree/v0.4.0