Cargo Features
[dependencies]
llama-cpp-2 = { version = "0.1.146", default-features = false, features = ["cuda", "cuda-no-vmm", "metal", "dynamic-link", "vulkan", "openmp", "rocm", "sampler", "android-shared-stdcxx", "android-static-stdcxx", "mtmd", "system-ggml", "system-ggml-static", "llguidance", "dynamic-backends"] }
- default = android-shared-stdcxx, openmp
-
These default features are set whenever
llama-cpp-2is added withoutsomewhere in the dependency tree.default-features = false - cuda cuda-no-vmm?
-
Enables cuda of llama-cpp-sys-2
- cuda-no-vmm = cuda
-
Enables cuda-no-vmm of llama-cpp-sys-2
llama-cpp-sys-2:
Disables the need to dynamically link against libcuda.so / cuda.dll
- metal
-
Enables metal of llama-cpp-sys-2
- dynamic-link
-
Enables dynamic-link of llama-cpp-sys-2
- vulkan
-
Enables vulkan of llama-cpp-sys-2
- openmp default
-
Enables openmp of llama-cpp-sys-2
- rocm
-
Enables rocm of llama-cpp-sys-2
- sampler
-
Only has an impact on Android.
Enables shared-stdcxx of llama-cpp-sys-2
- android-static-stdcxx
-
Enables static-stdcxx of llama-cpp-sys-2
- mtmd
-
Enables mtmd of llama-cpp-sys-2
Affects
llama-cpp-2::mtmd… - system-ggml
-
Enables system-ggml of llama-cpp-sys-2
- system-ggml-static
-
Enables system-ggml-static of llama-cpp-sys-2
- llguidance
-
Enables llguidance and toktrie
- dynamic-backends
-
Enables dynamic-backends of llama-cpp-sys-2
Affects
llama_backend::BACKENDS_DIR,llama_backend::load_backends_from_path,llama_backend::load_backends…