[package]
name = "local_inference"
authors.workspace = true
edition = "2024"
publish.workspace = true
license.workspace = true

[features]
default = ["cpu"]
cpu = []
metal = ["candle-core/metal", "candle-nn/metal", "candle-transformers/metal"]
cuda = ["candle-core/cuda", "candle-nn/cuda", "candle-transformers/cuda"]

[dependencies]
anyhow.workspace = true
async-trait.workspace = true
candle-core = "0.9.2"
candle-nn = "0.9.2"
candle-transformers = "0.9.2"
directories.workspace = true
hf-hub = { version = "0.4", features = ["tokio"] }
log.workspace = true
serde.workspace = true
serde_json.workspace = true
tokenizers = "0.21.4"
tokio = { workspace = true, features = ["fs", "sync", "rt", "macros", "rt-multi-thread"] }

[dev-dependencies]
env_logger = "0.10"
instant = { workspace = true }