Cargo.toml

[workspace]
members = [
    "core",
    "example",
    "vllm",
    "tops",
    "common",
    "entrypoints",
    "tensor-ext",
    # "triton-kernels",
]
resolver = "2"

[workspace.package]
version = "0.1.0"
edition = "2021"

[patch.crates-io]
candle = { package = "candle-core", git = "https://github.com/yinqiwen/candle", branch = "main" }

[workspace.dependencies]
clap = { version = "4.1.4", features = ["derive", "env"] }
tracing = "0.1"
tracing-subscriber = "0.3"
tracing-appender = "0.2"
tokio = { version = "1.0", features = ["full"] }
serde = "1.0"
anyhow = "1.0"
metrics = "0.22.1"
metrics-util = "0.16.2"

candle = { package = "candle-core", git = "https://github.com/yinqiwen/candle.git", branch = "main" }
candle-kernels = { git = "https://github.com/yinqiwen/candle.git", branch = "main" }
candle-nn = { git = "https://github.com/yinqiwen/candle.git", branch = "main" }
candle-flash-attn-v1 = { git = "https://github.com/yinqiwen/candle-flash-attn-v1.git", branch = "dev", package = "candle-flash-attn-v1" }
# candle-cublaslt = { git = "https://github.com/huggingface/candle-cublaslt" }
# candle-cublaslt = { path = "/data/dev/rust/candle-cublaslt" }
half = "2.4.0"
libc = "0.2"
minijinja = { version = "1.0.12", features = ["default", "loader"] }
uuid = { version = "1.7.0", features = ["v4", "fast-rng", "macro-diagnostics"] }
async-stream = "0.3.5"
serde_json = "1.0"
futures-core = "0.3"
cudarc = { version = "0.10.0", features = ["f16", "nccl", "cublaslt"] }
strum = "0.26"
strum_macros = "0.26"