realizar 0.8.4

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
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253

impl OwnedQuantizedModel {

    /// Forward pass with per-operation BrickProfiler instrumentation
    ///
    /// Identical to `forward()` but wraps each operation with
    /// `profiler.start()`/`profiler.stop()` using BrickId-aligned names.
    /// Zero overhead on production path (use `forward()` instead).
    ///
    /// Returns (logits, ProfileReport) with real per-operation timing.
    pub fn forward_profiled(
        &self,
        token_ids: &[u32],
        profiler: &mut BrickProfiler,
    ) -> Result<Vec<f32>> {
        let hidden_dim = self.config.hidden_dim;
        let _ = self.config.intermediate_dim;

        // 1. Token embedding
        profiler.start("Embedding");
        let mut hidden = self.embed(token_ids);
        // GH-278: Add learned position embeddings for absolute encoding (GPT-2, BERT, whisper)
        if self.config.constraints.uses_absolute_positions() {
            if let Some(ref pos_emb) = self.position_embedding {
                for (s, _) in token_ids.iter().enumerate() {
                    let pos_start = s * hidden_dim;
                    let pos_end = pos_start + hidden_dim;
                    if pos_end <= pos_emb.len() {
                        let h_start = s * hidden_dim;
                        for i in 0..hidden_dim {
                            hidden[h_start + i] += pos_emb[pos_start + i];
                        }
                    }
                }
            }
        }
        profiler.stop("Embedding");

        let use_rmsnorm = self.config.constraints.uses_rmsnorm();

        // 2. Transformer layers with per-operation instrumentation
        for (layer_idx, layer) in self.layers.iter().enumerate() {
            profiler.set_current_layer(layer_idx);

            // 2a. Attention norm
            profiler.start("RmsNorm");
            let normed = if use_rmsnorm {
                ops::rms_norm(&hidden, &layer.attn_norm_weight, self.config.eps)
            } else {
                ops::layer_norm(
                    &hidden,
                    &layer.attn_norm_weight,
                    layer.attn_norm_bias.as_deref(),
                    self.config.eps,
                )
            };
            profiler.stop("RmsNorm");

            // 2b. QKV projection
            profiler.start("QkvProjection");
            let qkv_dim = layer.qkv_weight.out_dim();
            let q_dim = layer.qkv_weight.q_dim_for_config(
                self.config.num_heads,
                self.config.num_kv_heads,
                self.config.hidden_dim,
                self.config.head_dim(),
            );
            let k_dim = layer.qkv_weight.k_dim_for_config(
                self.config.num_heads,
                self.config.num_kv_heads,
                self.config.hidden_dim,
                self.config.head_dim(),
            );
            let v_dim = layer.qkv_weight.v_dim_for_config(
                self.config.num_heads,
                self.config.num_kv_heads,
                self.config.hidden_dim,
                self.config.head_dim(),
            );
            let mut qkv = self.qkv_matmul(&normed, &layer.qkv_weight)?;
            if let Some(ref bias) = layer.qkv_bias {
                ops::add_bias(&mut qkv, bias);
            }
            profiler.stop("QkvProjection");

            // 2c. RoPE
            profiler.start("RopeEmbedding");
            let seq_len = token_ids.len();
            let mut q_all = Vec::with_capacity(seq_len * q_dim);
            let mut k_all = Vec::with_capacity(seq_len * k_dim);
            let mut v_all = Vec::with_capacity(seq_len * v_dim);

            for s in 0..seq_len {
                let qkv_start = s * qkv_dim;
                let mut q = qkv[qkv_start..qkv_start + q_dim].to_vec();
                let mut k = qkv[qkv_start + q_dim..qkv_start + q_dim + k_dim].to_vec();
                let v = &qkv[qkv_start + q_dim + k_dim..qkv_start + q_dim + k_dim + v_dim];

                // GH-278: Skip RoPE for models with learned position embeddings (GPT-2)
                if self.config.constraints.uses_rope() {
                    self.apply_rope(&mut q, s, self.config.num_heads);
                    self.apply_rope(&mut k, s, self.config.num_kv_heads);
                }

                q_all.extend_from_slice(&q);
                k_all.extend_from_slice(&k);
                v_all.extend_from_slice(v);
            }
            profiler.stop("RopeEmbedding");

            // 2d. Attention score + softmax + output
            profiler.start("AttentionScore");
            let attn_out = self.causal_attention(&q_all, &k_all, &v_all, seq_len);
            profiler.stop("AttentionScore");

            // 2e. Output projection
            profiler.start("OutputProjection");
            let mut attn_output = self.fused_matmul(&attn_out, &layer.attn_output_weight)?;
            if let Some(ref bias) = layer.attn_output_bias {
                ops::add_bias(&mut attn_output, bias);
            }
            profiler.stop("OutputProjection");

            // Residual connection
            for i in 0..hidden.len() {
                hidden[i] += attn_output[i];
            }

            // 2f. FFN norm
            profiler.start("RmsNorm");
            let ffn_input = if let Some(ref ffn_norm) = layer.ffn_norm_weight {
                if use_rmsnorm {
                    ops::rms_norm(&hidden, ffn_norm, self.config.eps)
                } else {
                    ops::layer_norm(
                        &hidden,
                        ffn_norm,
                        layer.ffn_norm_bias.as_deref(),
                        self.config.eps,
                    )
                }
            } else {
                hidden.clone()
            };
            profiler.stop("RmsNorm");

            // 2g. FFN (contract-driven activation selection, GH-278)
            let ffn_activated = if self.config.constraints.has_gate_ffn() {
                // SwiGLU path
                if let Some(ref gate_weight) = layer.ffn_gate_weight {
                    // Separate gate/up weights (LLaMA, Qwen2, Mistral)
                    profiler.start("UpProjection");
                    let mut ffn_up = self.fused_matmul(&ffn_input, &layer.ffn_up_weight)?;
                    if let Some(ref bias) = layer.ffn_up_bias {
                        ops::add_bias(&mut ffn_up, bias);
                    }
                    profiler.stop("UpProjection");

                    profiler.start("GateProjection");
                    let mut ffn_gate = self.fused_matmul(&ffn_input, gate_weight)?;
                    if let Some(ref bias) = layer.ffn_gate_bias {
                        ops::add_bias(&mut ffn_gate, bias);
                    }
                    profiler.stop("GateProjection");

                    profiler.start("Activation");
                    ops::silu(&mut ffn_gate);
                    for i in 0..ffn_gate.len() {
                        ffn_gate[i] *= ffn_up[i];
                    }
                    profiler.stop("Activation");

                    ffn_gate
                } else {
                    // GH-306: Fused gate_up weight (Phi-3.5) — single matmul, split in half
                    profiler.start("FusedGateUpProjection");
                    let fused = self.fused_matmul(&ffn_input, &layer.ffn_up_weight)?;
                    let half = fused.len() / 2;
                    let mut ffn_gate = fused[..half].to_vec();
                    let mut ffn_up = fused[half..].to_vec();
                    if let Some(ref bias) = layer.ffn_up_bias {
                        let bias_half = bias.len() / 2;
                        ops::add_bias(&mut ffn_gate, &bias[..bias_half]);
                        ops::add_bias(&mut ffn_up, &bias[bias_half..]);
                    }
                    profiler.stop("FusedGateUpProjection");

                    profiler.start("Activation");
                    ops::silu(&mut ffn_gate);
                    for i in 0..ffn_gate.len() {
                        ffn_gate[i] *= ffn_up[i];
                    }
                    profiler.stop("Activation");

                    ffn_gate
                }
            } else {
                // GELU path
                profiler.start("UpProjection");
                let mut ffn_hidden = self.fused_matmul(&ffn_input, &layer.ffn_up_weight)?;
                if let Some(ref bias) = layer.ffn_up_bias {
                    ops::add_bias(&mut ffn_hidden, bias);
                }
                profiler.stop("UpProjection");

                profiler.start("Activation");
                ops::gelu(&mut ffn_hidden);
                profiler.stop("Activation");

                ffn_hidden
            };

            // 2h. Down projection
            profiler.start("DownProjection");
            let mut ffn_output = self.fused_matmul(&ffn_activated, &layer.ffn_down_weight)?;
            if let Some(ref bias) = layer.ffn_down_bias {
                ops::add_bias(&mut ffn_output, bias);
            }
            profiler.stop("DownProjection");

            // Residual connection
            for i in 0..hidden.len() {
                hidden[i] += ffn_output[i];
            }
        }

        // 3. Final layer norm
        profiler.start("RmsNorm");
        let normed = if use_rmsnorm {
            ops::rms_norm(&hidden, &self.output_norm_weight, self.config.eps)
        } else {
            ops::layer_norm(
                &hidden,
                &self.output_norm_weight,
                self.output_norm_bias.as_deref(),
                self.config.eps,
            )
        };
        profiler.stop("RmsNorm");

        // 4. LM head projection (only last token)
        profiler.start("LmHead");
        let seq_len = token_ids.len();
        let last_hidden_start = (seq_len - 1) * hidden_dim;
        let last_hidden = &normed[last_hidden_start..last_hidden_start + hidden_dim];
        let mut logits = self.fused_matmul(last_hidden, &self.lm_head_weight)?;
        if let Some(ref bias) = self.lm_head_bias {
            ops::add_bias(&mut logits, bias);
        }
        profiler.stop("LmHead");

        Ok(logits)
    }
}