aprender-serve 0.34.0

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

//! Design by Contract example for realizar.
//!
//! Demonstrates `ValidatedModelConfig` as a Poka-Yoke newtype gate (PMAT-235):
//! invalid model configs are rejected at construction time, before any tensor
//! allocation or inference computation can occur.
//!
//! Run: `cargo run --example design_by_contract`

use realizar::gguf::{
    Activation, ArchConstraints, GGUFConfig, MlpType, NormType, PositionalEncoding,
    ValidatedModelConfig, WeightLayout,
};

fn main() {
    // -----------------------------------------------------------------------
    // 1. Valid config: LLaMA-7B passes all 11 structural invariants
    // -----------------------------------------------------------------------
    let llama_config = GGUFConfig {
        architecture: "llama".to_string(),
        constraints: ArchConstraints::from_architecture("llama"),
        hidden_dim: 4096,
        num_layers: 32,
        num_heads: 32,
        num_kv_heads: 8, // 4:1 GQA ratio
        vocab_size: 32_000,
        intermediate_dim: 11_008,
        context_length: 4096,
        rope_theta: 10_000.0,
        eps: 1e-5,
        rope_type: 0, // NORM (adjacent pairs)
        explicit_head_dim: None,
        bos_token_id: Some(128_000),
        eos_token_id: Some(128_001),
    };

    let validated = ValidatedModelConfig::validate(llama_config)
        .expect("LLaMA-7B config should pass validation");
    println!(
        "[PASS] LLaMA-7B: hidden={}, heads={}, kv_heads={}, head_dim={}",
        validated.hidden_dim(),
        validated.num_heads(),
        validated.num_kv_heads(),
        validated.head_dim()
    );

    // -----------------------------------------------------------------------
    // 2. Invalid config: hidden_dim=0 is rejected immediately
    // -----------------------------------------------------------------------
    let bad_config = GGUFConfig {
        architecture: "llama".to_string(),
        constraints: ArchConstraints::from_architecture("llama"),
        hidden_dim: 0, // <-- violates SI-001
        num_layers: 32,
        num_heads: 32,
        num_kv_heads: 8,
        vocab_size: 32_000,
        intermediate_dim: 11_008,
        context_length: 4096,
        rope_theta: 10_000.0,
        eps: 1e-5,
        rope_type: 0,
        explicit_head_dim: None,
        bos_token_id: None,
        eos_token_id: None,
    };

    let err =
        ValidatedModelConfig::validate(bad_config).expect_err("hidden_dim=0 must be rejected");
    println!("[REJECTED] hidden_dim=0: {err}");

    // -----------------------------------------------------------------------
    // 3. Invalid config: num_heads does not divide hidden_dim
    // -----------------------------------------------------------------------
    let misaligned = GGUFConfig {
        architecture: "llama".to_string(),
        constraints: ArchConstraints::from_architecture("llama"),
        hidden_dim: 4096,
        num_layers: 32,
        num_heads: 33, // <-- 4096 % 33 != 0
        num_kv_heads: 33,
        vocab_size: 32_000,
        intermediate_dim: 11_008,
        context_length: 4096,
        rope_theta: 10_000.0,
        eps: 1e-5,
        rope_type: 0,
        explicit_head_dim: None,
        bos_token_id: None,
        eos_token_id: None,
    };

    let err = ValidatedModelConfig::validate(misaligned)
        .expect_err("hidden_dim % num_heads != 0 must be rejected");
    println!("[REJECTED] head_dim misaligned: {err}");

    // -----------------------------------------------------------------------
    // 4. ArchConstraints: contract-driven architecture behavior
    // -----------------------------------------------------------------------
    println!("\n--- ArchConstraints comparison ---");

    let qwen2 = ArchConstraints::from_architecture("qwen2");
    println!(
        "Qwen2:  norm={:?}, activation={:?}, pos={:?}, mlp={:?}, layout={:?}, bias={}",
        qwen2.norm_type,
        qwen2.activation,
        qwen2.positional_encoding,
        qwen2.mlp_type,
        qwen2.weight_layout,
        qwen2.has_bias
    );

    let gpt2 = ArchConstraints::from_architecture("gpt2");
    println!(
        "GPT-2:  norm={:?}, activation={:?}, pos={:?}, mlp={:?}, layout={:?}, bias={}",
        gpt2.norm_type,
        gpt2.activation,
        gpt2.positional_encoding,
        gpt2.mlp_type,
        gpt2.weight_layout,
        gpt2.has_bias
    );

    // Verify contract differences
    assert_eq!(qwen2.norm_type, NormType::RmsNorm);
    assert_eq!(gpt2.norm_type, NormType::LayerNorm);
    assert_eq!(qwen2.activation, Activation::Silu);
    assert_eq!(gpt2.activation, Activation::Gelu);
    assert_eq!(qwen2.positional_encoding, PositionalEncoding::Rope);
    assert_eq!(gpt2.positional_encoding, PositionalEncoding::Absolute);
    assert_eq!(qwen2.mlp_type, MlpType::SwiGlu);
    assert_eq!(gpt2.mlp_type, MlpType::GeluMlp);
    assert_eq!(gpt2.weight_layout, WeightLayout::Conv1D);
    assert!(gpt2.needs_transpose());
    assert!(!qwen2.needs_transpose());

    // -----------------------------------------------------------------------
    // 5. Metadata bounds: OOM prevention
    // -----------------------------------------------------------------------
    let oom_config = GGUFConfig {
        architecture: "llama".to_string(),
        constraints: ArchConstraints::from_architecture("llama"),
        hidden_dim: 100_000, // <-- exceeds 65536 max
        num_layers: 32,
        num_heads: 1,
        num_kv_heads: 1,
        vocab_size: 32_000,
        intermediate_dim: 11_008,
        context_length: 4096,
        rope_theta: 10_000.0,
        eps: 1e-5,
        rope_type: 0,
        explicit_head_dim: None,
        bos_token_id: None,
        eos_token_id: None,
    };

    let err = ValidatedModelConfig::validate(oom_config)
        .expect_err("hidden_dim=100000 exceeds 65536 max");
    println!("\n[REJECTED] OOM prevention: {err}");

    println!("\nAll Design by Contract checks passed.");
}