cortiq-engine 0.5.21

Portable inference runtime for the CMF model format, with no ML framework underneath: runs on CPU, and on GPU (Vulkan / Metal / DX12) with the `gpu` feature; tokenizer, chat templates and dynamic per-skill weight overlay.
Documentation 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44

[package]
name = "cortiq-engine"
version.workspace = true
edition.workspace = true
description = "Portable inference runtime for the CMF model format, with no ML framework underneath: runs on CPU, and on GPU (Vulkan / Metal / DX12) with the `gpu` feature; tokenizer, chat templates and dynamic per-skill weight overlay."
license.workspace = true
authors.workspace = true
repository.workspace = true
homepage.workspace = true
readme.workspace = true
keywords.workspace = true
categories.workspace = true
rust-version.workspace = true

[dependencies]
cortiq-core = { path = "../cortiq-core", version = "0.5.21" }
serde = { workspace = true }
serde_json = { workspace = true }
tokio = { workspace = true }
tracing = { workspace = true }
anyhow = { workspace = true }
thiserror = { workspace = true }
fancy-regex = { workspace = true }
unicode-normalization = { workspace = true }
minijinja = { workspace = true }
minijinja-contrib = { workspace = true }
base64 = { workspace = true }

# Cross-platform GPU backend (C1): wgpu -> Vulkan/DX12/Metal (NVIDIA,
# Radeon, Intel, Apple). Optional: a plain CPU build does not pull it in.
wgpu = { version = "30", optional = true }
pollster = { version = "0.4", optional = true }
bytemuck = { version = "1", optional = true, features = ["derive"] }

[features]
# `--features gpu` enables the wgpu path; test on a discrete card.
gpu = ["dep:wgpu", "dep:pollster", "dep:bytemuck"]

# GPU path (D5): Metal on Apple Silicon; unified memory + no-copy buffers.
[target.'cfg(target_os = "macos")'.dependencies]
metal = "0.29"

[target.'cfg(target_os = "android")'.dependencies]
libc = "0.2"