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
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177


#[test]
fn test_cov001_weight_quant_type_detection() {
    // Test from_ggml_type
    assert!(matches!(
        WeightQuantType::from_ggml_type(12),
        Some(WeightQuantType::Q4K)
    ));
    assert!(matches!(
        WeightQuantType::from_ggml_type(13),
        Some(WeightQuantType::Q5K)
    ));
    assert!(matches!(
        WeightQuantType::from_ggml_type(14),
        Some(WeightQuantType::Q6K)
    ));
    assert!(matches!(
        WeightQuantType::from_ggml_type(8),
        Some(WeightQuantType::Q8_0)
    ));
    assert!(matches!(
        WeightQuantType::from_ggml_type(6),
        Some(WeightQuantType::Q5_0)
    ));
    assert!(matches!(
        WeightQuantType::from_ggml_type(2),
        Some(WeightQuantType::Q4_0)
    ));
    assert!(matches!(
        WeightQuantType::from_ggml_type(3),
        Some(WeightQuantType::Q4_1)
    ));
    assert!(WeightQuantType::from_ggml_type(999).is_none());

    // Test bytes_per_superblock
    assert_eq!(WeightQuantType::Q4K.bytes_per_superblock(), 144);
    assert_eq!(WeightQuantType::Q5K.bytes_per_superblock(), 176);
    assert_eq!(WeightQuantType::Q6K.bytes_per_superblock(), 210);

    // Test bytes_per_block
    assert_eq!(WeightQuantType::Q8_0.bytes_per_block(), 34);
    assert_eq!(WeightQuantType::Q4_0.bytes_per_block(), 18);

    // Test matches_size
    let q4k = WeightQuantType::Q4K;
    assert!(q4k.matches_size(144, 1, 256)); // 1 row, 256 cols = 1 superblock
}

#[test]
fn test_cov001_ptx_optimization_hints() {
    let max_throughput = PtxOptimizationHints::max_throughput();
    assert!(max_throughput.uses_vectorized_loads());
    assert_eq!(max_throughput.vector_width(), 4);
    assert_eq!(max_throughput.shared_mem_padding(), 1);

    let low_latency = PtxOptimizationHints::low_latency();
    assert!(!low_latency.uses_vectorized_loads());
    assert_eq!(low_latency.vector_width(), 1);
    assert_eq!(low_latency.shared_mem_padding(), 0);

    let balanced = PtxOptimizationHints::balanced();
    assert!(balanced.uses_vectorized_loads());
    assert_eq!(balanced.vector_width(), 2);
}

#[test]
fn test_cov001_ptx_optimizer() {
    let hints = PtxOptimizationHints::max_throughput();
    let optimizer = PtxOptimizer::new(hints);

    // Test summary generation
    let summary = optimizer.summary();
    assert!(summary.contains("PtxOptimizer"));

    // Test padded row calculation
    assert_eq!(optimizer.padded_shared_mem_row(32), 33);

    // Test register estimation
    let regs = optimizer.estimated_registers();
    assert!(regs > 0);

    // Test high register pressure detection
    let _high_pressure = optimizer.is_high_register_pressure();
}

#[test]
fn test_cov001_register_tiling() {
    let large = RegisterTiling::large();
    assert_eq!(large.registers_needed(), 64);

    let medium = RegisterTiling::medium();
    assert_eq!(medium.registers_needed(), 16);

    let small = RegisterTiling::small();
    assert_eq!(small.registers_needed(), 4);
}

#[test]
fn test_cov001_memory_pattern() {
    let scalar = MemoryPattern::Scalar;
    let vec2 = MemoryPattern::Vector2;
    let vec4 = MemoryPattern::Vector4;

    // Just ensure they can be compared
    assert_ne!(scalar, vec2);
    assert_ne!(vec2, vec4);
}

#[test]
fn test_cov001_bank_conflict_strategy() {
    let none = BankConflictStrategy::None;
    let padding = BankConflictStrategy::Padding;
    let xor = BankConflictStrategy::Xor;

    assert_ne!(none, padding);
    assert_ne!(padding, xor);
}

#[test]
fn test_cov001_presets_coverage() {
    // Test all preset functions
    let _llama_attn = presets::llama_attention(2048, 64);
    let _ffn = presets::ffn_gemm(1, 4096, 11008);
    let _q4k = presets::q4k_inference(1, 4096, 4096);
    let _q4k_ggml = presets::q4k_ggml_inference(1, 4096, 4096);
    let _rmsnorm = presets::rmsnorm(4096);
    let _mha = presets::multi_head_attention(2048, 64, 32);
    let _phi2_mha = presets::phi2_multi_head_attention(2048);
    let _tc_attn = presets::tensor_core_attention(2048, 64, 32);
    let _llama_tc = presets::llama_tensor_core_attention(2048);
}

#[test]
fn test_cov001_kernel_type_kernel_names() {
    let kernels = CudaKernels::new();

    // Test all kernel type names
    let types = [
        KernelType::GemmNaive { m: 1, n: 1, k: 1 },
        KernelType::GemmTiled {
            m: 1,
            n: 1,
            k: 1,
            tile_size: 32,
        },
        KernelType::Softmax { dim: 128 },
        KernelType::LayerNorm {
            hidden_size: 256,
            epsilon: 1e-5,
            affine: true,
        },
        KernelType::Attention {
            seq_len: 16,
            head_dim: 64,
            causal: true,
        },
        KernelType::QuantizedGemm {
            m: 1,
            n: 256,
            k: 256,
        },
        KernelType::QuantizedGemmGgml {
            m: 1,
            n: 256,
            k: 256,
        },
        KernelType::Q4KGemv { k: 256, n: 256 },
        KernelType::Q5KGemv { k: 256, n: 256 },
        KernelType::Q6KGemv { k: 256, n: 256 },
    ];

    for kt in types {
        let name = kernels.kernel_name(&kt);
        assert!(!name.is_empty(), "Kernel name should not be empty");
    }
}