[package]
name = "inference-engine"
version = "0.1.0"
edition = "2021"

[dependencies]
candle-core = { git = "https://github.com/huggingface/candle.git" }
candle-nn = { git = "https://github.com/huggingface/candle.git" }
candle-transformers = { git = "https://github.com/huggingface/candle.git" }
candle-flash-attn = { version = "=0.9.1", optional = true }
candle-onnx = { version = "=0.9.1", optional = true }

serde = { version = "1.0.171", features = ["derive"] }
serde_json = "1.0.99"
symphonia = { version = "0.5.3", features = ["all"], optional = true }
tokenizers = { version = "0.21.0", default-features = false, features = ["onig", "http"] }
cpal = { version = "0.15.2", optional = true }
pdf2image = { version = "0.1.2" , optional = true}
anyhow = "1.0.98"
clap= { version = "4.2.4", features = ["derive"] }
tracing = "0.1.37"
tracing-chrome = "0.7.1"
tracing-subscriber = { version = "0.3.7", features = ["env-filter"] }
axum = { version = "0.8.4", features = ["json"] }
tower = "0.5.2"
tower-http = { version = "0.6.6", features = ["cors"] }
tokio = { version = "1.43.0", features = ["full"] }
tokio-stream = { version = "0.1.16", features = ["sync"] }
either = { version = "1.9.0", features = ["serde"] }
utoipa = { version = "4.2.0", features = ["axum_extras"] }
uuid = { version = "1.7.0", features = ["v4"] }
reborrow = "0.5.5"
futures-util = "0.3.31"
gemma-runner = { path = "../gemma-runner" }
llama-runner = { path = "../llama-runner" }

[target.'cfg(target_os = "macos")'.dependencies]
candle-core = { git = "https://github.com/huggingface/candle.git", features = ["metal"] }
candle-nn = { git = "https://github.com/huggingface/candle.git", features = ["metal"] }
candle-transformers = { git = "https://github.com/huggingface/candle.git", features = ["metal"] }


[dev-dependencies]
anyhow = { version = "1", features = ["backtrace"] }
byteorder = { version = "1.4.3" }
clap = { version = "4.2.4", features = ["derive"] }
imageproc = { version = "0.24.0", default-features = false }
memmap2 = { version = "0.9.3", features = ["stable_deref_trait"] }
rand = { version = "0.9.0" }
ab_glyph = { version = "0.2.23" }
tracing = { version = "0.1.37" }
tracing-chrome = { version = "0.7.1" }
tracing-subscriber = { version = "0.3.7" }
# Necessary to disambiguate with tokio in wasm examples which are 1.28.1
tokio = "1.43.0"

[build-dependencies]
anyhow = { version = "1", features = ["backtrace"] }
bindgen_cuda = { version = "0.1.1", optional = true }

[features]
bin = []



[package.metadata.compose]
image = "ghcr.io/geoffsee/inference-engine:latest"
port = 8080


# generates kubernetes manifests
[package.metadata.kube]
image = "ghcr.io/geoffsee/inference-service:latest"
replicas = 1
port = 8080