[package]
name = "llama_cpp"
version = "0.3.2"
description = "High-level bindings to llama.cpp with a focus on just being really, really easy to use"
edition = "2021"
authors = ["Dakota Thompson <me@scriptis.net>", "Pedro Valente <pedro.amaral.valente@gmail.com>"]
repository = "https://github.com/edgenai/llama_cpp-rs"
license = "MIT OR Apache-2.0"
readme = "../../README.md"
publish = true

[dependencies]
derive_more = "0.99.17"
futures = { workspace = true }
llama_cpp_sys = { version = "^0.3.2", path = "../llama_cpp_sys", default-features = false }
num_cpus = "1.16.0"
thiserror = "1.0.57"
tokio = { workspace = true, features = ["sync", "rt", "rt-multi-thread"] }
tracing = "0.1.40"

[features]
default = ["compat", "native"]
compat = ["llama_cpp_sys/compat"] # this feature modifies the symbols exposed by the generated libraries to avoid conflicts
native = ["llama_cpp_sys/native", "avx", "avx2", "fma", "f16c", "accel"]
avx = ["llama_cpp_sys/avx"]
avx2 = ["llama_cpp_sys/avx2"]
avx512 = ["llama_cpp_sys/avx512"]
avx512_vmbi = ["llama_cpp_sys/avx512_vmbi"]
avx512_vnni = ["llama_cpp_sys/avx512_vnni"]
fma = ["llama_cpp_sys/fma"]
f16c = ["llama_cpp_sys/f16c"] # implied when compiled using MSVC with avx2/avx512
accel = ["llama_cpp_sys/accel"] # Accelerate framework
mpi = ["llama_cpp_sys/mpi"]
cuda = ["llama_cpp_sys/cuda"]
cuda_f16 = ["llama_cpp_sys/cuda_f16", "cuda"]
cuda_dmmv = ["llama_cpp_sys/cuda_dmmv", "cuda"] # use dmmv instead of mmvq CUDA kernels
cuda_mmq = ["llama_cpp_sys/cuda_mmq", "cuda"] # use mmq kernels instead of cuBLAS
metal = ["llama_cpp_sys/metal"]
blas = ["llama_cpp_sys/blas"]
hipblas = ["llama_cpp_sys/hipblas"]
clblast = ["llama_cpp_sys/clblast"]
vulkan = ["llama_cpp_sys/vulkan"]
sys_verbosity = [] # increase sys crate tracing log verbosity level

[lib]
doctest = false