sipp-rs 0.1.0

Unified Rust library for extensible Sipp inference
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
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131

//! Tests the `lifecycle::backend_policy` module in `sipp`.
//!
//! Covers lifecycle registry, storage, browser, service, and pairing behavior with temporary storage and pure fixtures instead of native runtime loading.

use super::*;
use crate::lifecycle::test_support::strings;
use crate::runtime::config::SplitMode;

fn caps(compiled: &[&str], available: &[&str]) -> BackendCapabilities {
    BackendCapabilities {
        compiled: strings(compiled),
        available: strings(available),
        gpu_offload_supported: compiled.iter().any(|value| *value != "cpu"),
    }
}

#[test]
fn cpu_forces_gpu_layers_zero() {
    let options = ModelLoadOptions {
        backend: BackendPreference::Cpu,
        stats: StatsMode::Off,
        ..ModelLoadOptions::default()
    };

    let plan =
        BackendPolicy::select_with_capabilities(&options, &caps(&[], &["cpu"])).expect("backend");

    assert_eq!(plan.selection.selected, "cpu");
    assert_eq!(plan.config.placement.gpu_layers, GpuLayerConfig::Count(0));
    assert_eq!(plan.config.placement.split_mode, SplitMode::Layer);
    assert_eq!(
        plan.config.context.flash_attention,
        FlashAttentionMode::Disabled
    );
    assert!(!plan.config.context.offload_kqv);
    assert!(!plan.config.context.op_offload);
    assert!(!plan.config.observability.runtime_metrics);
    assert!(!plan.config.observability.backend_profiling);
}

#[test]
fn cuda_requires_compiled_available_backend() {
    let options = ModelLoadOptions {
        backend: BackendPreference::Cuda,
        ..ModelLoadOptions::default()
    };

    let error = BackendPolicy::select_with_capabilities(&options, &caps(&[], &["cpu"]))
        .expect_err("missing cuda");

    assert!(matches!(error, ModelError::InvalidModelSource(_)));
}

#[test]
fn cuda_selects_full_offload_by_default() {
    let options = ModelLoadOptions {
        backend: BackendPreference::Cuda,
        stats: StatsMode::Profile,
        ..ModelLoadOptions::default()
    };

    let plan =
        BackendPolicy::select_with_capabilities(&options, &caps(&["cuda"], &["cuda", "cpu"]))
            .expect("cuda");

    assert_eq!(plan.selection.selected, "cuda");
    assert_eq!(plan.config.placement.gpu_layers, GpuLayerConfig::Auto);
    assert!(plan.config.observability.runtime_metrics);
    assert!(plan.config.observability.backend_profiling);
}

#[test]
fn webgpu_selects_full_offload_by_default() {
    let options = ModelLoadOptions {
        backend: BackendPreference::WebGpu,
        ..ModelLoadOptions::default()
    };

    let plan =
        BackendPolicy::select_with_capabilities(&options, &caps(&["webgpu"], &["cpu", "webgpu"]))
            .expect("webgpu");

    assert_eq!(plan.selection.selected, "webgpu");
    assert_eq!(plan.config.placement.gpu_layers, GpuLayerConfig::Auto);
    assert!(plan.selection.gpu_offload_expected);
}

#[test]
fn auto_prefers_accelerator_then_cpu() {
    let plan = BackendPolicy::select_with_capabilities(
        &ModelLoadOptions::default(),
        &caps(&["vulkan"], &["cpu", "vulkan"]),
    )
    .expect("auto");
    assert_eq!(plan.selection.selected, "vulkan");

    let plan =
        BackendPolicy::select_with_capabilities(&ModelLoadOptions::default(), &caps(&[], &[]))
            .expect("auto cpu");
    assert_eq!(plan.selection.selected, "cpu");
    assert_eq!(plan.config.placement.gpu_layers, GpuLayerConfig::Count(0));
    assert_eq!(plan.config.placement.split_mode, SplitMode::Layer);
    assert!(!plan.config.context.offload_kqv);
}

#[test]
fn select_with_capabilities_normalizes_backend_names() {
    let plan = BackendPolicy::select_with_capabilities(
        &ModelLoadOptions {
            backend: BackendPreference::Cuda,
            ..ModelLoadOptions::default()
        },
        &BackendCapabilities {
            compiled: strings(&["CUDA backend", "cuda"]),
            available: strings(&["NVIDIA CUDA", "CPU"]),
            gpu_offload_supported: true,
        },
    )
    .expect("cuda");

    assert_eq!(plan.selection.selected, "cuda");
    assert_eq!(plan.selection.available, vec!["cpu", "cuda"]);
}

#[test]
fn normalize_backend_names_drops_empty_and_deduplicates() {
    let names = strings(&[" CUDA backend ", "cuda", " ", "CPU"]);

    assert_eq!(normalize_backend_names(&names), vec!["cpu", "cuda"]);
    assert_eq!(normalize_backend_names_or_cpu(&[]), vec!["cpu"]);
}