realizar 0.8.5

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


// =============================================================================
// Tests for SIMD Q4 Transformer (PMAT-802: T-COV-95)
// =============================================================================

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

    // -------------------------------------------------------------------------
    // QuantizedAprTensorQ4 Tests
    // -------------------------------------------------------------------------

    #[test]
    fn test_quantized_tensor_q4_new() {
        let data = vec![0u8; 36]; // 2 blocks * 18 bytes
        let tensor = QuantizedAprTensorQ4::new(data.clone(), 32, 2);
        assert_eq!(tensor.data.len(), 36);
        assert_eq!(tensor.in_dim, 32);
        assert_eq!(tensor.out_dim, 2);
    }

    #[test]
    fn test_quantized_tensor_q4_zeros_small() {
        let tensor = QuantizedAprTensorQ4::zeros(32, 1); // 32 elements = 1 block
        assert_eq!(tensor.in_dim, 32);
        assert_eq!(tensor.out_dim, 1);
        // 1 block = 18 bytes
        assert_eq!(tensor.data.len(), 18);
        assert!(tensor.data.iter().all(|&b| b == 0));
    }

    #[test]
    fn test_quantized_tensor_q4_zeros_multiple_blocks() {
        let tensor = QuantizedAprTensorQ4::zeros(64, 2); // 128 elements = 4 blocks
        // 4 blocks = 72 bytes
        assert_eq!(tensor.data.len(), 72);
    }

    #[test]
    fn test_quantized_tensor_q4_zeros_partial_block() {
        let tensor = QuantizedAprTensorQ4::zeros(33, 1); // 33 elements = 2 blocks (rounds up)
        // 2 blocks = 36 bytes
        assert_eq!(tensor.data.len(), 36);
    }

    #[test]
    fn test_quantized_tensor_q4_expected_bytes() {
        // 32 elements = 1 block = 18 bytes
        assert_eq!(QuantizedAprTensorQ4::expected_bytes(32), 18);
        // 33 elements = 2 blocks = 36 bytes
        assert_eq!(QuantizedAprTensorQ4::expected_bytes(33), 36);
        // 64 elements = 2 blocks = 36 bytes
        assert_eq!(QuantizedAprTensorQ4::expected_bytes(64), 36);
        // 256 elements = 8 blocks = 144 bytes
        assert_eq!(QuantizedAprTensorQ4::expected_bytes(256), 144);
    }

    #[test]
    fn test_quantized_tensor_q4_expected_bytes_zero() {
        // 0 elements = 0 blocks = 0 bytes
        assert_eq!(QuantizedAprTensorQ4::expected_bytes(0), 0);
    }

    // -------------------------------------------------------------------------
    // AprInferenceScratch Tests
    // -------------------------------------------------------------------------

    fn make_test_config() -> AprTransformerConfig {
        AprTransformerConfig {
            architecture: "apr".to_string(),
            hidden_dim: 64,
            num_layers: 2,
            num_heads: 4,
            num_kv_heads: 2,
            vocab_size: 100,
            intermediate_dim: 128,
            context_length: 512,
            rope_theta: 10000.0,
            eps: 1e-5,
            eos_token_id: None,
        ..Default::default()
        }
    }

    #[test]
    fn test_inference_scratch_from_config() {
        let config = make_test_config();
        let scratch = AprInferenceScratch::from_config(&config);

        // Check buffer sizes match config
        assert_eq!(scratch.hidden.len(), 64); // hidden_dim
        assert_eq!(scratch.normed.len(), 64); // hidden_dim
        assert_eq!(scratch.qkv_out.len(), 192); // hidden_dim * 3
        assert_eq!(scratch.q.len(), 64); // hidden_dim
        assert_eq!(scratch.k.len(), 64); // hidden_dim
        assert_eq!(scratch.v.len(), 64); // hidden_dim
        assert_eq!(scratch.attn_out.len(), 64); // hidden_dim
        assert_eq!(scratch.ffn_input.len(), 64); // hidden_dim
        assert_eq!(scratch.ffn_up.len(), 128); // intermediate_dim
        assert_eq!(scratch.ffn_gate.len(), 128); // intermediate_dim
        assert_eq!(scratch.ffn_out.len(), 64); // hidden_dim
    }

    #[test]
    fn test_inference_scratch_initialized_to_zero() {
        let config = make_test_config();
        let scratch = AprInferenceScratch::from_config(&config);

        // All buffers should be zero-initialized
        assert!(scratch.hidden.iter().all(|&x| x == 0.0));
        assert!(scratch.normed.iter().all(|&x| x == 0.0));
        assert!(scratch.qkv_out.iter().all(|&x| x == 0.0));
        assert!(scratch.attn_out.iter().all(|&x| x == 0.0));
        assert!(scratch.ffn_up.iter().all(|&x| x == 0.0));
    }

    #[test]
    fn test_inference_scratch_clear() {
        let config = make_test_config();
        let mut scratch = AprInferenceScratch::from_config(&config);

        // Modify buffers
        scratch.hidden[0] = 1.0;
        scratch.normed[0] = 2.0;
        scratch.attn_out[0] = 3.0;
        scratch.ffn_up[0] = 4.0;

        // Clear and verify
        scratch.clear();
        assert!(scratch.hidden.iter().all(|&x| x == 0.0));
        assert!(scratch.normed.iter().all(|&x| x == 0.0));
        assert!(scratch.attn_out.iter().all(|&x| x == 0.0));
        assert!(scratch.ffn_up.iter().all(|&x| x == 0.0));
    }

    #[test]
    fn test_inference_scratch_large_config() {
        let config = AprTransformerConfig {
            architecture: "apr".to_string(),
            hidden_dim: 4096,
            num_layers: 32,
            num_heads: 32,
            num_kv_heads: 8,
            vocab_size: 32000,
            intermediate_dim: 11008,
            context_length: 4096,
            rope_theta: 10000.0,
            eps: 1e-5,
            eos_token_id: None,
        ..Default::default()
        };
        let scratch = AprInferenceScratch::from_config(&config);

        assert_eq!(scratch.hidden.len(), 4096);
        assert_eq!(scratch.ffn_up.len(), 11008);
        assert_eq!(scratch.ffn_gate.len(), 11008);
    }
}