yue2.cpp, the native port of YuE2 (built for Windows with CUDA)

The native port of YuE2 (yue2.cpp) built for Windows with CUDA, for Audante's own Windows receiver.

What this is

yue2-cpp-windows.zip holds the program (yue-server.exe), the tools around it (quantize.exe and the others), the port's own runtime (the ggml DLLs) and the licence texts. Unpacked, it is the engine of the model variant yue2-gguf-cuda-windows in Audante's catalog.

How it was built

From the source in Audante's own repository (Audante/yue2-cpp-src, the trees of the pinned commits of yue2.cpp and its ggml submodule), with scripts/build-yue2-cpp.ps1 from the Audante repository: MSVC with the Ninja generator, the CUDA toolkit assembled from NVIDIA's redistributable archives (scripts/build-yue-cuda-toolkit.ps1).

The build machine: Windows 11 mit NVIDIA RTX 3070 Ti (8 GB). The CUDA architectures: 86-real.

Changes

None to the source: the archive is the build of the pinned commits. The patch on the pinned commit (the UTF-8 code page on Windows) is part of that commit.

The CUDA runtime is not in this archive. cudart64_12.dll, cublas64_12.dll and cublasLt64_12.dll come from NVIDIA's redistributable archives and are fetched by the receiver when it installs the engine (the addresses and checksums are in redistrib_12.9.1.json in Audante/yue2-cpp-src). NVIDIA allows these files to travel with the program. Without them Windows ends the program at the start with STATUS_DLL_NOT_FOUND (0xC0000135) and no message.

The port's own runtime is in this archive: the ggml DLLs are part of Audante's build of the port (MIT), not a third-party component. Without them the program does not start either, with the same error.

Licence

MIT, as the port. See LICENSE for the text and NOTICE for the authors.

The weights the program reads are a separate matter and are not in this repository: they are CC BY-NC 4.0 and live in Audante's own folders.

Downloads last month

-

Downloads are not tracked for this model. How to track
Inference Providers NEW
This model isn't deployed by any Inference Provider. ๐Ÿ™‹ Ask for provider support