llama-cpp-sys-2 0.1.157

Low Level Bindings to llama.cpp
Documentation 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97

[package]
name = "llama-cpp-sys-2"
description = "Low Level Bindings to llama.cpp"
version = "0.1.157"
edition = "2021"
license = "MIT OR Apache-2.0"
repository = "https://github.com/utilityai/llama-cpp-rs"
links = "llama"

include = [
    # Our files
    "README.md",
    "wrapper.h",
    "wrapper_common.h",
    "wrapper_common.cpp",
    "wrapper_utils.h",
    "wrapper_mtmd.h",
    "build.rs",
    "/src",

    # Everything needed for building GGML (CMake, include and src)
    #
    # `llama.cpp` already only includes the most important stuff
    # from https://github.com/ggml-org/ggml.
    "/llama.cpp/ggml",

    # lLama.cpp's own sources.
    "/llama.cpp/include",
    "/llama.cpp/common",
    "/llama.cpp/src",

    # mtmd sources. Contains tests files that we don't want, so those
    # are filtered out.
    "/llama.cpp/tools/mtmd/**/*.h",
    "/llama.cpp/tools/mtmd/**/*.cpp",
    "/llama.cpp/tools/mtmd/**/CMakeLists.txt",

    # Vendored sources.
    "/llama.cpp/vendor",

    # Various build files.
    "/llama.cpp/convert_hf_to_gguf.py", # Yes, it's required
    "/llama.cpp/pocs",
    "/llama.cpp/CMakeLists.txt",
    "/llama.cpp/cmake",
    "/llama.cpp/tools/CMakeLists.txt",

    # License.
    "/llama.cpp/LICENSE",
]

# See more keys and their definitions at https://doc.rust-lang.org/cargo/reference/manifest.html

[dependencies]

[build-dependencies]
bindgen = { workspace = true }
cc = { workspace = true, features = ["parallel"] }
cmake = "0.1"
find_cuda_helper = "0.2.0"
glob = "0.3.3"
walkdir = "2"

[features]
default = ["common"]
# Build the JSON-schema-to-grammar helper backed by llama.cpp's `common/` static
# library. Disabling this drops `libcommon.a` (~14 MB) and the
# `wrapper_common.cpp` bindings from the link line, which matters for
# size-constrained downstream binaries that only need the core inference API.
# Enabled by default to preserve backwards compatibility.
common = []
cuda = []
# Disables the need to dynamically link against libcuda.so / cuda.dll
cuda-no-vmm = ["cuda"]
metal = []
dynamic-link = []
vulkan = []
# OpenCL backend (GGML_OPENCL=ON) — the Qualcomm-supported path for Adreno GPUs.
# The build needs OpenCL headers + a libOpenCL.so to link; on a cross-compile
# (e.g. Android) point CMake at them via the OPENCL_INCLUDE_DIR / OPENCL_LIBRARY
# env vars (build.rs forwards them), since FindOpenCL can't locate an SDK.
opencl = []
# Build the BLAS backend with Intel MKL (x86_64-only; requires MKLROOT set).
mkl = []
openmp = []
# Link GNU OpenMP (`libgomp`) statically on GNU targets. This is useful for
# downstreams that ship self-contained binaries while keeping OpenMP enabled.
static-openmp = ["openmp"]
rocm = []
# Link the C++ standard library statically where the target toolchain supports
# it. Currently used by Android and Linux GNU/musl targets.
static-stdcxx = []
shared-stdcxx = []
system-ggml = []
system-ggml-static = ["system-ggml"]
mtmd = []
dynamic-backends = ["dynamic-link"]