npc-backends

Compute backends for the language and speech models served in npc-models, one folder per backend, each shipped whole with the licence of what it carries as <name>.LICENSE. They are not models: a backend is the library that makes a model run on a particular kind of hardware. A game drops a folder's files beside the libraries of the extension that loads the model, and the extension picks the backend up at start; without it, the extension runs on whatever backend it already carries.

hf download shuko-tuto/npc-backends --local-dir backends --include "ggml_cuda/*"

Folder	What it holds	Source	Licence
ggml_cuda/	The CUDA backend of llama.cpp tag b10786, ggml-cuda.dll, with the three CUDA runtime libraries it imports (cudart64_13, cublas64_13, cublasLt64_13), 628 MB, NVIDIA only. Beside the libraries of tag b10786 it runs a language model and a speech codec on the card; the Vulkan backend of the same tag cannot run the codec. Windows x86_64	llama.cpp's release archive of tag b10786; NVIDIA's CUDA redistributable	MIT for ggml, NVIDIA's runtime terms for the rest, both in ggml_cuda.LICENSE

The backend is tied to the tag of the libraries it sits beside: one from another tag loads and may run slower, or not at all, with no check at load. Keep the two in step.