aprender-serve 0.50.0

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


    // =========================================================================
    // GH-219 Coverage Gap: transpose_weight, concat_qkv_transposed,
    // has_tensor_with_fallback_generic, get_tensor_with_fallback_generic
    // =========================================================================

    // -------------------------------------------------------------------------
    // transpose_weight (PMAT-095: no-op pass-through)
    // -------------------------------------------------------------------------

    #[test]
    fn test_transpose_weight_identity_gh219() {
        let weight = vec![1.0, 2.0, 3.0, 4.0, 5.0, 6.0];
        let result = SafetensorsToAprConverter::transpose_weight(&weight, 2, 3);
        assert_eq!(result, weight);
    }

    #[test]
    fn test_transpose_weight_empty_gh219() {
        let result = SafetensorsToAprConverter::transpose_weight(&[], 0, 0);
        assert!(result.is_empty());
    }

    #[test]
    fn test_transpose_weight_single_element_gh219() {
        let weight = vec![42.0];
        let result = SafetensorsToAprConverter::transpose_weight(&weight, 1, 1);
        assert_eq!(result, vec![42.0]);
    }

    #[test]
    fn test_transpose_weight_preserves_data_gh219() {
        // Verify that it copies, not moves
        let weight: Vec<f32> = (0..100).map(|i| i as f32 * 0.1).collect();
        let result = SafetensorsToAprConverter::transpose_weight(&weight, 10, 10);
        assert_eq!(result.len(), 100);
        assert!((result[50] - 5.0).abs() < 1e-6);
    }

    // -------------------------------------------------------------------------
    // transpose_weight_owned (PMAT-787: zero-copy owned move; bit-identical to
    // the borrowing `transpose_weight`, but without the redundant allocation)
    // -------------------------------------------------------------------------

    #[test]
    fn test_transpose_weight_owned_identity_pmat787() {
        let weight = vec![1.0, 2.0, 3.0, 4.0, 5.0, 6.0];
        let result = SafetensorsToAprConverter::transpose_weight_owned(weight.clone());
        assert_eq!(result, weight);
    }

    #[test]
    fn test_transpose_weight_owned_empty_pmat787() {
        let result = SafetensorsToAprConverter::transpose_weight_owned(Vec::new());
        assert!(result.is_empty());
    }

    #[test]
    fn test_transpose_weight_owned_matches_borrowing_variant_pmat787() {
        // The owned move MUST produce byte-identical output to the borrowing
        // clone variant — this is what guarantees inference output is unchanged.
        let weight: Vec<f32> = (0..256).map(|i| (i as f32) * 0.125 - 7.0).collect();
        let via_borrow = SafetensorsToAprConverter::transpose_weight(&weight, 16, 16);
        let via_move = SafetensorsToAprConverter::transpose_weight_owned(weight);
        assert_eq!(via_borrow, via_move);
    }

    // -------------------------------------------------------------------------
    // concat_qkv_transposed
    // -------------------------------------------------------------------------

    #[test]
    fn test_concat_qkv_transposed_basic_gh219() {
        let q = vec![1.0, 2.0];
        let k = vec![3.0, 4.0];
        let v = vec![5.0, 6.0];
        let result = SafetensorsToAprConverter::concat_qkv_transposed(&q, &k, &v, 2, 2);
        assert_eq!(result, vec![1.0, 2.0, 3.0, 4.0, 5.0, 6.0]);
    }

    #[test]
    fn test_concat_qkv_transposed_different_sizes_gh219() {
        // Q: hidden_dim x hidden_dim (4 elements for 2x2)
        let q = vec![1.0, 2.0, 3.0, 4.0];
        // K: kv_dim x hidden_dim (2 elements for 1x2)
        let k = vec![5.0, 6.0];
        // V: kv_dim x hidden_dim (2 elements for 1x2)
        let v = vec![7.0, 8.0];
        let result = SafetensorsToAprConverter::concat_qkv_transposed(&q, &k, &v, 2, 1);
        assert_eq!(result.len(), 8);
        assert_eq!(result, vec![1.0, 2.0, 3.0, 4.0, 5.0, 6.0, 7.0, 8.0]);
    }

    #[test]
    fn test_concat_qkv_transposed_empty_gh219() {
        let result =
            SafetensorsToAprConverter::concat_qkv_transposed(&[], &[], &[], 0, 0);
        assert!(result.is_empty());
    }

    #[test]
    fn test_concat_qkv_transposed_capacity_gh219() {
        let q = vec![0.0; 100];
        let k = vec![0.0; 50];
        let v = vec![0.0; 50];
        let result = SafetensorsToAprConverter::concat_qkv_transposed(&q, &k, &v, 10, 5);
        assert_eq!(result.len(), 200);
    }