[package]

name = "xinfer"

version = "0.14.4"

edition = "2021"

default-run = "xinfer"

description = "xInfer — a minimal, high-performance large language model (LLM) inference engine in Rust."

repository = "https://github.com/guoqingbao/xinfer"

homepage = "https://github.com/guoqingbao/xinfer"

readme = "ReadMe.md"

keywords = ["llm", "inference", "rust", "candle", "xinfer"]

categories = ["algorithms", "hardware-support", "science"]

license = "MIT"

[dependencies]

candle-core = { git = "https://github.com/guoqingbao/candle.git", version = "0.8.3", rev = "b60efcf" }

candle-nn = { git = "https://github.com/guoqingbao/candle.git", version = "0.8.3", rev = "b60efcf" }

serde = { version = "1.0.190", features = ["serde_derive"] }

tokenizers = {version = "0.21.2", features = ["http"] }

hf-hub = "0.4.1"

anyhow = "1.0.75"

itertools = "0.13.0"

akin = "0.4.0"

indicatif = "0.17.11"

serde_json = "1.0.108"

llguidance = { version = "1.7", default-features = false, features = ["lark", "referencing", "jsonschema_validation"] }

toktrie_hf_tokenizers = "1.7"

toktrie = "1.4"

half = { version = "2.5.0", features = ["num-traits", "use-intrinsics", "rand_distr"] }

tokio = { version = "1.38.0", features = ["sync"] }

tracing = "0.1.40"

tracing-subscriber = { version = "0.3.18", features = ["env-filter"] }

either = { version = "1.13.0", features = ["serde"] }

minijinja = { version = "2.10.2", features = ["builtins", "json"] }

minijinja-contrib = { version = "2.10.2", features = ["pycompat"] }

lazy_static = {version = "1.4.0"}

interprocess = "2.2.2"

serde-big-array = "0.5.1"

bincode = { version = "1.3.1" }

twox-hash = "2.1.1"

rand = "0.9.0"

rayon="1.10.0"

clap = { version = "4.4.7", features = ["derive"] }

thiserror = "1.0.58"

schemars = "0.8"

ahash = "0.8.11"

reedline = "0.40.0"

pyo3 = { version = "0.25.1", features = ["extension-module", "abi3-py38"], optional = true }

parking_lot = "0.12.4"

attention-rs = { git = "https://github.com/guoqingbao/attention.rs.git", version="0.6.8", rev = "f3a7b69" }

once_cell = "1.21.3"

tqdm = "0.8.0"

futures = "0.3.31"

crossbeam = "0.8.4"

ctrlc = "3.4.7"

dirs = "5.0.1"

base64 = "0.22.1"

metal = { version = "0.27.0", features = ["mps"], optional = true }

uuid = { version = "1.5.0", features = ["v4"] }

axum = { version = "0.8.9", features = ["tokio"] }

utoipa = { version = "4.2", features = ["axum_extras"] }

colored = { version = "3.0.0" }

tower-http = { version = "0.6.6", features = ["cors"] }

rustchatui = "0.3.2"

sysinfo = "0.38.3"

image = { version = "0.25.6", default-features = false, features = ['bmp', 'gif', 'jpeg', 'png', 'tiff', 'webp'] }

reqwest = { version = "0.12.24", features = ["blocking", "json", "rustls-tls"]}

bytemuck = "1.24.0"

regex = "1.12.2"

local-ip-address = "0.6.5"

url = "2.5.7"

tool-parser = "1.3.1"

openai-protocol = "1.8.1"

nvtx = { version = "1.3", optional = true }

[lib]

name = "xinfer"

path = "src/lib.rs"

crate-type = ["rlib", "cdylib"]

[features]

cuda = ["candle-core/cuda", "candle-nn/cuda", "attention-rs/cuda", "graph"]

nccl = ["candle-core/nccl"]

graph = ["flash", "attention-rs/graph", "candle-core/graph"]

cutlass = ["attention-rs/cutlass"]

flash = ["attention-rs/flash"]

flashattn = ["attention-rs/flashattn"]

flashinfer = ["attention-rs/flashinfer"]

metal = ["candle-core/metal", "candle-nn/metal", "attention-rs/metal", "attention-rs/metal-flash", "dep:metal"]

python = ["pyo3"]

nvtx = ["dep:nvtx"]

[target.'cfg(target_os = "linux")'.dependencies]

openssl-sys = { version = "0.9", features = ["vendored"] }

# CUDA differential probes live with the integration tests, but are exposed

# as explicit Cargo examples so the release probe runner can build them all in

# one compilation phase.

[[example]]

name = "candle_precision_probe"

path = "tests/probes/candle_precision_probe.rs"

[[example]]

name = "attention_misc_precision_probe"

path = "tests/probes/attention_misc_precision_probe.rs"

[[example]]

name = "flashinfer_precision_probe"

path = "tests/probes/flashinfer_precision_probe.rs"

[[example]]

name = "fp8_precision_probe"

path = "tests/probes/fp8_precision_probe.rs"

[[example]]

name = "gdn_precision_probe"

path = "tests/probes/gdn_precision_probe.rs"

[[example]]

name = "nvfp4_precision_probe"

path = "tests/probes/nvfp4_precision_probe.rs"

[package.metadata.deb]

name = "xinfer"

maintainer = "xInfer Team"

section = "science"

priority = "optional"

assets = [

["target/release/xinfer", "usr/local/bin/", "755"],

["target/release/lib/*", "usr/local/lib/xinfer/", "644"],

]