realizar 0.8.4

Pure Rust ML inference engine built from scratch - model serving for GGUF and safetensors
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
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165


impl WarmupExecutor {
    /// Create a new warm-up executor
    #[must_use]
    pub fn new(config: WarmupConfig) -> Self {
        Self { config }
    }

    /// Get configuration
    #[must_use]
    pub fn config(&self) -> &WarmupConfig {
        &self.config
    }

    /// Simulate warm-up (for testing without actual model)
    ///
    /// This runs the warm-up process with test inference delays.
    #[must_use]
    pub fn simulate_warmup(&self) -> WarmupResult {
        let start = Instant::now();
        let mut latencies = Vec::with_capacity(self.config.warmup_iterations);

        // Simulate decreasing latency as model warms up
        for i in 0..self.config.warmup_iterations {
            // First iteration is "cold" (slower)
            let base_latency_us = if i == 0 { 1000 } else { 100 };
            let jitter = (i * 10) as u64;
            let latency = Duration::from_micros(base_latency_us - jitter.min(50));
            latencies.push(latency);
        }

        WarmupResult::success(self.config.warmup_iterations, start.elapsed(), &latencies)
    }

    /// Check if timeout has been exceeded
    #[allow(dead_code)]
    fn check_timeout(&self, start: Instant, iterations: usize) -> Option<WarmupResult> {
        if start.elapsed() > self.config.timeout {
            Some(WarmupResult::timed_out(iterations, start.elapsed()))
        } else {
            None
        }
    }
}

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

// ============================================================================
// WARM-006: Pre-load Configuration
// ============================================================================

/// Configuration for model pre-loading
#[derive(Debug, Clone, Serialize, Deserialize)]
pub struct PreloadConfig {
    /// Models to pre-load on startup
    pub models: Vec<PreloadModelConfig>,
    /// Load models in parallel
    pub parallel_loading: bool,
    /// Maximum concurrent model loads
    pub max_concurrent: usize,
    /// Fail startup if any model fails to load
    pub fail_fast: bool,
}

impl Default for PreloadConfig {
    fn default() -> Self {
        Self {
            models: Vec::new(),
            parallel_loading: true,
            max_concurrent: 4,
            fail_fast: false,
        }
    }
}

impl PreloadConfig {
    /// Create new pre-load configuration
    #[must_use]
    pub fn new() -> Self {
        Self::default()
    }

    /// Add a model to pre-load
    #[must_use]
    pub fn with_model(mut self, model: PreloadModelConfig) -> Self {
        self.models.push(model);
        self
    }

    /// Set parallel loading
    #[must_use]
    pub fn with_parallel_loading(mut self, parallel: bool) -> Self {
        self.parallel_loading = parallel;
        self
    }

    /// Set maximum concurrent loads
    #[must_use]
    pub fn with_max_concurrent(mut self, max: usize) -> Self {
        self.max_concurrent = max.max(1);
        self
    }

    /// Set fail-fast behavior
    #[must_use]
    pub fn with_fail_fast(mut self, fail_fast: bool) -> Self {
        self.fail_fast = fail_fast;
        self
    }
}

/// Configuration for a single model to pre-load
#[derive(Debug, Clone, Serialize, Deserialize)]
pub struct PreloadModelConfig {
    /// Model identifier
    pub model_id: String,
    /// Model URI (pacha://, file://, hf://)
    pub uri: String,
    /// Priority (lower = load first)
    pub priority: u32,
    /// Run warm-up after loading
    pub warmup: bool,
    /// Warm-up configuration
    pub warmup_config: Option<WarmupConfig>,
}

impl PreloadModelConfig {
    /// Create new model pre-load config
    #[must_use]
    pub fn new(model_id: impl Into<String>, uri: impl Into<String>) -> Self {
        Self {
            model_id: model_id.into(),
            uri: uri.into(),
            priority: 100,
            warmup: true,
            warmup_config: None,
        }
    }

    /// Set priority
    #[must_use]
    pub fn with_priority(mut self, priority: u32) -> Self {
        self.priority = priority;
        self
    }

    /// Enable/disable warm-up
    #[must_use]
    pub fn with_warmup(mut self, warmup: bool) -> Self {
        self.warmup = warmup;
        self
    }

    /// Set warm-up configuration
    #[must_use]
    pub fn with_warmup_config(mut self, config: WarmupConfig) -> Self {
        self.warmup_config = Some(config);
        self.warmup = true;
        self
    }
}