shimmy 2.1.1

Lightweight Ollama-compatible inference server with native SafeTensors support. No Python dependencies, cross-platform WebGPU acceleration via Airframe.
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
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

use anyhow::Result;
use async_trait::async_trait;

use super::{InferenceEngine, LoadedModel, ModelSpec};

pub struct InferenceEngineAdapter {
    #[cfg(feature = "airframe")]
    airframe_engine: super::airframe::AirframeEngine,
    safetensors_engine: super::safetensors_native::SafeTensorsEngine,
}

impl Default for InferenceEngineAdapter {
    fn default() -> Self {
        Self::new()
    }
}

impl InferenceEngineAdapter {
    pub fn new() -> Self {
        Self {
            #[cfg(feature = "airframe")]
            airframe_engine: super::airframe::AirframeEngine::new(),
            safetensors_engine: super::safetensors_native::SafeTensorsEngine::new(),
        }
    }

    /// Backend arg retained for CLI compatibility; ignored in v2 (airframe auto-detects GPU)
    pub fn new_with_backend(_gpu_backend: Option<&str>) -> Self {
        Self::new()
    }

    fn select_backend(&self, spec: &ModelSpec) -> BackendChoice {
        let path_str = spec.base_path.to_string_lossy();

        if let Some(ext) = spec.base_path.extension().and_then(|s| s.to_str()) {
            match ext {
                "safetensors" => return BackendChoice::SafeTensors,
                "gguf" => {
                    #[cfg(feature = "airframe")]
                    return BackendChoice::Airframe;
                    #[cfg(not(feature = "airframe"))]
                    return BackendChoice::SafeTensors;
                }
                _ => {}
            }
        }

        // Ollama blob files (extensionless GGUF)
        if path_str.contains("ollama") && path_str.contains("blobs") {
            #[cfg(feature = "airframe")]
            return BackendChoice::Airframe;
        }

        // Name/path heuristic for GGUF
        if path_str.contains(".gguf") {
            #[cfg(feature = "airframe")]
            return BackendChoice::Airframe;
        }

        BackendChoice::SafeTensors
    }
}

#[derive(Debug, Clone, PartialEq)]
enum BackendChoice {
    #[cfg(feature = "airframe")]
    Airframe,
    SafeTensors,
}

#[async_trait]
impl InferenceEngine for InferenceEngineAdapter {
    async fn load(&self, spec: &ModelSpec) -> Result<Box<dyn LoadedModel>> {
        match self.select_backend(spec) {
            BackendChoice::SafeTensors => self.safetensors_engine.load(spec).await,
            #[cfg(feature = "airframe")]
            BackendChoice::Airframe => self.airframe_engine.load(spec).await,
        }
    }
}

#[cfg(test)]
mod tests {
    use super::*;
    use std::path::PathBuf;

    fn spec(name: &str, path: &str) -> ModelSpec {
        ModelSpec {
            name: name.to_string(),
            base_path: PathBuf::from(path),
            lora_path: None,
            template: None,
            ctx_len: 2048,
            n_threads: None,
        }
    }

    #[test]
    fn test_safetensors_routes_to_safetensors() {
        let a = InferenceEngineAdapter::new();
        assert_eq!(
            a.select_backend(&spec("m", "model.safetensors")),
            BackendChoice::SafeTensors
        );
    }

    #[test]
    #[cfg(feature = "airframe")]
    fn test_gguf_routes_to_airframe() {
        let a = InferenceEngineAdapter::new();
        assert_eq!(
            a.select_backend(&spec("m", "model.gguf")),
            BackendChoice::Airframe
        );
    }

    #[test]
    fn test_safetensors_priority() {
        let a = InferenceEngineAdapter::new();
        assert_eq!(
            a.select_backend(&spec("m", "/models/org/pytorch_model.safetensors")),
            BackendChoice::SafeTensors
        );
    }
}