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
178


impl CudaKernels {

    /// Get kernel name for the specified type
    #[must_use]
    pub fn kernel_name(&self, kernel_type: &KernelType) -> &'static str {
        Self::gemm_kernel_name(kernel_type)
            .or_else(|| Self::gemv_kernel_name(kernel_type))
            .or_else(|| Self::q4k_gemv_kernel_name(kernel_type))
            .or_else(|| Self::attention_kernel_name(kernel_type))
            .or_else(|| Self::norm_rope_kernel_name(kernel_type))
            .or_else(|| Self::activation_misc_kernel_name(kernel_type))
            .unwrap_or("unknown")
    }

    /// GEMM kernel names (matrix-matrix multiplication)
    fn gemm_kernel_name(kernel_type: &KernelType) -> Option<&'static str> {
        let name = match kernel_type {
            KernelType::GemmNaive { .. } => "gemm_naive",
            KernelType::GemmTiled { .. }
            | KernelType::GemmOptimized { .. }
            | KernelType::GemmBiasActivation { .. } => "gemm_tiled",
            KernelType::GemmTensorCore { .. } => "gemm_tensor_core",
            KernelType::GemmFp16TensorCore { .. } => "gemm_wmma_fp16",
            KernelType::QuantizedGemm { .. } => "q4k_gemm_fused",
            KernelType::QuantizedGemmGgml { .. } => "q4k_gemm_ggml",
            KernelType::QuantizedGemmGgmlTiled { .. } => "q4k_gemm_ggml_tiled",
            KernelType::Q5KQuantizedGemm { .. } => "q5k_gemm_ggml",
            KernelType::Q6KQuantizedGemm { .. } => "q6k_gemm_ggml",
            KernelType::FusedQ4Q8Dot { .. } => "q4k_gemm_ggml",
            KernelType::TensorCoreQ4KGemm { .. } => "tensor_core_q4k_gemm",
            KernelType::MultiWarpTensorCoreQ4KGemm { .. } => "mw_tensor_core_q4k_gemm",
            KernelType::InterleavedWmmaQ4KGemm { .. } => "interleaved_wmma_q4k_gemm",
            KernelType::W4a16WmmaQ4KGemm { .. } => "w4a16_wmma_q4k_gemm",
            KernelType::Dp4aQ4KGemm { .. } => "dp4a_q4k_gemm",
            _ => return None,
        };
        Some(name)
    }

    /// Standard GEMV kernel names (matrix-vector, non-Q4K)
    fn gemv_kernel_name(kernel_type: &KernelType) -> Option<&'static str> {
        let name = match kernel_type {
            KernelType::Gemv { .. } => "gemv_warp_reduce",
            KernelType::CoalescedGemv { .. } => "gemv_coalesced",
            KernelType::Q5KGemv { .. } => "q5k_gemv_warp_reduce",
            KernelType::Q6KGemv { .. } => "q6k_gemv_warp_reduce",
            KernelType::CoalescedQ6KGemv { .. } => "coalesced_q6k_gemv",
            KernelType::BatchedQ6KGemv { .. } => "batched_q6k_gemv_warp_reduce",
            KernelType::MwvQ6KGemv { .. } => "mwv_q6k_gemv",
            KernelType::Dp4aQ6KGemv { .. } => "dp4a_q6k_gemv",
            KernelType::HwDp4aQ6KGemv { .. } => "hw_dp4a_q6k_gemv",
            KernelType::Fp16Q4KGemv { .. } => "fp16_q4k_gemv",
            KernelType::Q8_0Gemv { .. } => "q8_0_gemv_warp_reduce",
            KernelType::Q5_0Gemv { .. } => "q5_0_gemv_warp_reduce",
            KernelType::Q4_0Gemv { .. } => "q4_0_gemv_warp_reduce",
            KernelType::Q4_1Gemv { .. } => "q4_1_gemv_warp_reduce",
            _ => return None,
        };
        Some(name)
    }

    /// Q4K GEMV kernel names (quantized 4-bit matrix-vector variants)
    fn q4k_gemv_kernel_name(kernel_type: &KernelType) -> Option<&'static str> {
        let name = match kernel_type {
            KernelType::Q4KGemv { .. } => "q4k_gemv_warp_reduce",
            KernelType::TiledQ4KGemv { .. } => "tiled_q4k_gemv",
            KernelType::ChunkedTiledQ4KGemv { .. } => "chunked_tiled_q4k_gemv",
            KernelType::CoalescedQ4KGemv { .. } => "coalesced_q4k_gemv",
            KernelType::WideQ4KGemv { .. } => "wide_q4k_gemv",
            KernelType::VectorizedQ4KGemv { .. } => "vectorized_q4k_gemv",
            KernelType::MwvQ4KGemv { .. } => "mwv_q4k_gemv",
            KernelType::MwvDp4aQ4KGemv { .. } => "mwv_dp4a_q4k_gemv",
            KernelType::HwDp4aQ4KGemv { .. } => "hw_dp4a_q4k_gemv",
            KernelType::Dp4aQ4KGemv { .. } => "dp4a_q4k_gemv",
            KernelType::Dp4aSIMDQ4KGemv { .. } => "dp4a_simd_q4k_gemv",
            KernelType::TrueDp4aQ4KGemv { .. } => "true_dp4a_q4k_gemv",
            KernelType::BatchedQ4KGemv { .. }
            | KernelType::MultiWarpBatchedQ4KGemv { .. } => "batched_q4k_gemv_warp_reduce",
            KernelType::BatchedHwDp4aQ4KGemv { .. } => "batched_hw_dp4a_q4k_gemv",
            KernelType::FusedFp32Q4KGemv { .. } => "fused_fp32_q4k_gemv",
            KernelType::InlineQ8Dp4aQ4KGemv { .. } => "inline_q8_dp4a_q4k_gemv",
            KernelType::FusedKVHwDp4aQ4KGemv { .. } => "fused_qkv_hw_dp4a_q4k_gemv",
            KernelType::Q4KDequant { .. } => "q4k_dequant_to_f32",
            KernelType::Q4KDequantFp16 { .. } => "q4k_dequant_to_f16",
            KernelType::Q6KDequant { .. } => "q6k_dequant_to_f32",
            _ => return None,
        };
        Some(name)
    }

    /// Attention kernel names (flash attention, incremental, multi-warp)
    fn attention_kernel_name(kernel_type: &KernelType) -> Option<&'static str> {
        let name = match kernel_type {
            KernelType::Softmax { .. } => "softmax_warp_shuffle",
            KernelType::LayerNorm { .. } => "layernorm_warp_shuffle",
            KernelType::Attention { causal, .. }
            | KernelType::MultiHeadAttention { causal, .. } => {
                if *causal { "flash_attention_causal" } else { "flash_attention" }
            },
            KernelType::AttentionTensorCore { causal, .. } => {
                if *causal { "flash_attention_tensor_core_causal" } else { "flash_attention_tensor_core" }
            },
            KernelType::IncrementalAttention { indirect, .. } => {
                if *indirect { "incremental_attention_indirect" } else { "incremental_attention" }
            },
            KernelType::MultiWarpAttention { indirect, .. } => {
                if *indirect { "multi_warp_attention_indirect" } else { "multi_warp_attention" }
            },
            _ => return None,
        };
        Some(name)
    }

    /// Normalization and RoPE kernel names
    fn norm_rope_kernel_name(kernel_type: &KernelType) -> Option<&'static str> {
        let name = match kernel_type {
            KernelType::RmsNorm { .. } => "rmsnorm",
            KernelType::VectorizedRmsNorm { .. } => "rmsnorm_vectorized",
            KernelType::BatchedVectorizedRmsNorm { .. } => "batched_rmsnorm_vectorized",
            KernelType::BatchedFusedResidualRmsNorm { .. } => "batched_fused_residual_rmsnorm",
            KernelType::PreciseRmsNorm { .. } => "rmsnorm_precise",
            KernelType::PerHeadRmsNorm { .. } => "per_head_rmsnorm",
            KernelType::FusedResidualRmsNorm { .. } => "fused_residual_rmsnorm",
            KernelType::FusedRmsNormQ4KGemv { .. } => "fused_rmsnorm_q4k_gemv",
            KernelType::FusedRmsNormGateUpSwigluQ4K { .. } => "fused_rmsnorm_gate_up_swiglu_q4k",
            KernelType::Rope { .. } => "rope",
            KernelType::RopeIndirect { .. } => "rope_indirect",
            KernelType::RopeNeox { .. } => "rope_neox",
            KernelType::RopeNeoxIndirect { .. } => "rope_neox_indirect",
            KernelType::PreciseRopeNeoxIndirect { .. } => "rope_precise_indirect",
            KernelType::BatchedRope { .. } => "batched_rope",
            _ => return None,
        };
        Some(name)
    }

    /// Activation, fusion, and miscellaneous kernel names
    fn activation_misc_kernel_name(kernel_type: &KernelType) -> Option<&'static str> {
        let name = match kernel_type {
            KernelType::BiasActivation { .. } => "bias_activation",
            KernelType::Silu { .. } => "silu",
            KernelType::Gelu { .. } => "gelu",
            KernelType::ElementwiseMul { .. } => "elementwise_mul",
            KernelType::FusedSwiglu { .. } => "fused_swiglu",
            KernelType::FusedQKV { .. } => "fused_qkv_gemv",
            KernelType::FusedGateUp { .. } => "fused_gate_up_swiglu",
            KernelType::FusedGateUpQ4KGemv { .. } => "fused_gate_up_q4k_gemv",
            KernelType::FusedGateUpSwigluHwDp4aQ4KGemv { .. } => "fused_gate_up_swiglu_hw_dp4a_q4k_gemv",
            KernelType::ResidualAdd { .. } => "residual_add",
            KernelType::BatchedResidualAdd { .. } => "batched_residual_add",
            KernelType::BatchedSwiglu { .. } => "batched_swiglu",
            KernelType::KvCacheScatter { .. } => "kv_cache_scatter",
            KernelType::KvCacheScatterIndirect { .. } => "kv_cache_scatter_indirect",
            KernelType::Q8Quantize { .. } => "q8_quantize",
            KernelType::Q4KQ8Dot { .. } => "q4k_q8_dot",
            KernelType::PackedDp4aQ4KQ8 { .. } => "packed_dp4a_q4k_q8",
            KernelType::ArgMax { .. } => "argmax_block_reduce",
            KernelType::ArgMaxFinal { .. } => "argmax_final_reduce",
            KernelType::Q8Dequant { .. } => "q8_dequant",
            _ => return None,
        };
        Some(name)
    }

    /// Check if CUDA is likely available on this system
    ///
    /// Note: This is a heuristic check. Actual CUDA availability requires
    /// driver API initialization.
    #[must_use]
    pub fn cuda_likely_available() -> bool {
        // Check for NVIDIA GPU indicators
        std::path::Path::new("/dev/nvidia0").exists()
            || std::env::var("CUDA_VISIBLE_DEVICES").is_ok()
    }
}

include!("cuda.rs");