forjar 1.4.2

Rust-native Infrastructure as Code — bare-metal first, BLAKE3 state, provenance tracing
Documentation 
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
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405

//! Contract trait enforcement -- compiler verifies provable-contracts trait compliance.
//!
//! Generated via provable-contracts Section 23 trait enforcement.
//!
//! Each `impl` below uses reference scalar implementations. The compile-time
//! check proves the trait signatures are satisfiable. If the contract traits
//! ever change shape, this file fails to compile.
//!
//! Run with: `cargo test --test contract_traits`

use provable_contracts::traits::{
    ActivationKernelV1, AdamwKernelV1, AttentionKernelV1, CrossEntropyKernelV1, FlashAttentionV1,
    GqaKernelV1, LayernormKernelV1, MatmulKernelV1, RmsnormKernelV1, RopeKernelV1, SiluKernelV1,
    SoftmaxKernelV1, SwigluKernelV1,
};

/// Marker struct for reference scalar kernel implementations.
struct ReferenceKernels;

// ---------------------------------------------------------------------------
// SoftmaxKernelV1 -- numerically stable softmax
// ---------------------------------------------------------------------------
impl SoftmaxKernelV1 for ReferenceKernels {
    fn softmax(&self, x: &[f32]) -> Vec<f32> {
        if x.is_empty() {
            return vec![];
        }
        let max = x.iter().copied().fold(f32::NEG_INFINITY, f32::max);
        let exps: Vec<f32> = x.iter().map(|&xi| (xi - max).exp()).collect();
        let sum: f32 = exps.iter().sum();
        exps.iter().map(|&e| e / sum).collect()
    }
}

// ---------------------------------------------------------------------------
// ActivationKernelV1 -- gelu, relu, silu
// ---------------------------------------------------------------------------
impl ActivationKernelV1 for ReferenceKernels {
    fn gelu(&self, x: f32) -> Vec<f32> {
        let inner = (2.0_f32 / std::f32::consts::PI).sqrt() * (x + 0.044715 * x * x * x);
        vec![0.5 * x * (1.0 + inner.tanh())]
    }

    fn relu(&self, x: f32) -> Vec<f32> {
        vec![x.max(0.0)]
    }

    fn silu(&self, x: f32) -> Vec<f32> {
        vec![x / (1.0 + (-x).exp())]
    }
}

// ---------------------------------------------------------------------------
// SiluKernelV1 -- sigmoid and SiLU activation
// ---------------------------------------------------------------------------
impl SiluKernelV1 for ReferenceKernels {
    fn sigmoid(&self, x: &[f32]) -> Vec<f32> {
        x.iter().map(|&v| 1.0 / (1.0 + (-v).exp())).collect()
    }

    fn silu(&self, x: &[f32]) -> Vec<f32> {
        x.iter().map(|&v| v / (1.0 + (-v).exp())).collect()
    }
}

// ---------------------------------------------------------------------------
// RmsnormKernelV1 -- root-mean-square layer normalization
// ---------------------------------------------------------------------------
impl RmsnormKernelV1 for ReferenceKernels {
    fn rmsnorm(&self, x: &[f32]) -> Vec<f32> {
        let rms = (x.iter().map(|v| v * v).sum::<f32>() / x.len() as f32).sqrt();
        x.iter().map(|v| v / (rms + 1e-5)).collect()
    }
}

// ---------------------------------------------------------------------------
// LayernormKernelV1 -- layer normalization with affine transform
// ---------------------------------------------------------------------------
impl LayernormKernelV1 for ReferenceKernels {
    fn layernorm(&self, x: &[f32], gamma: &[f32]) -> Vec<f32> {
        let mean = x.iter().sum::<f32>() / x.len() as f32;
        let var = x.iter().map(|v| (v - mean).powi(2)).sum::<f32>() / x.len() as f32;
        let std = (var + 1e-5).sqrt();
        x.iter()
            .enumerate()
            .map(|(i, v)| ((v - mean) / std) * gamma.get(i).copied().unwrap_or(1.0))
            .collect()
    }

    fn statistics(&self, x: &[f32]) -> Vec<f32> {
        let mean = x.iter().sum::<f32>() / x.len() as f32;
        let var = x.iter().map(|v| (v - mean).powi(2)).sum::<f32>() / x.len() as f32;
        vec![mean, var]
    }
}

// ---------------------------------------------------------------------------
// CrossEntropyKernelV1 -- cross-entropy loss and log-softmax
// ---------------------------------------------------------------------------
impl CrossEntropyKernelV1 for ReferenceKernels {
    fn cross_entropy(&self, targets: &[f32], logits: &[f32]) -> Vec<f32> {
        let log_sm = CrossEntropyKernelV1::log_softmax(self, logits);
        let loss = -targets
            .iter()
            .zip(log_sm.iter())
            .map(|(t, l)| t * l)
            .sum::<f32>();
        vec![loss]
    }

    fn log_softmax(&self, x: &[f32]) -> Vec<f32> {
        let max = x.iter().copied().fold(f32::NEG_INFINITY, f32::max);
        let sum_exp = x.iter().map(|v| (v - max).exp()).sum::<f32>().ln();
        x.iter().map(|v| v - max - sum_exp).collect()
    }
}

// ---------------------------------------------------------------------------
// SwigluKernelV1 -- SiLU-gated linear unit
// ---------------------------------------------------------------------------
impl SwigluKernelV1 for ReferenceKernels {
    fn silu(&self, x: &[f32]) -> Vec<f32> {
        x.iter().map(|&v| v / (1.0 + (-v).exp())).collect()
    }

    fn swiglu(&self, x: &[f32], w: &[f32], v: &[f32], _b: &[f32], _c: &[f32]) -> Vec<f32> {
        let gate: Vec<f32> = x.iter().zip(w.iter()).map(|(xi, wi)| xi * wi).collect();
        let silu_gate: Vec<f32> = gate.iter().map(|&g| g / (1.0 + (-g).exp())).collect();
        let value: Vec<f32> = x.iter().zip(v.iter()).map(|(xi, vi)| xi * vi).collect();
        silu_gate
            .iter()
            .zip(value.iter())
            .map(|(s, val)| s * val)
            .collect()
    }
}

// ---------------------------------------------------------------------------
// Compile-time enforcement test
// ---------------------------------------------------------------------------
#[test]
fn contract_traits_compile() {
    let k = ReferenceKernels;

    // SoftmaxKernelV1: verify normalization invariant
    let out = k.softmax(&[1.0, 2.0, 3.0]);
    let sum: f32 = out.iter().sum();
    assert!((sum - 1.0).abs() < 1e-6, "softmax must sum to 1.0");

    // ActivationKernelV1: verify basic properties
    let gelu_zero = k.gelu(0.0);
    assert!(gelu_zero[0].abs() < 1e-6, "GELU(0) = 0");

    let relu_neg = k.relu(-1.0);
    assert_eq!(relu_neg[0], 0.0, "ReLU(-1) = 0");

    let relu_pos = k.relu(2.0);
    assert_eq!(relu_pos[0], 2.0, "ReLU(2) = 2");

    let silu_zero = ActivationKernelV1::silu(&k, 0.0);
    assert!(silu_zero[0].abs() < 1e-6, "SiLU(0) = 0");
}

#[test]
fn silu_kernel_v1_properties() {
    let k = ReferenceKernels;

    // sigmoid(0) = 0.5
    let sig = SiluKernelV1::sigmoid(&k, &[0.0]);
    assert!((sig[0] - 0.5).abs() < 1e-6, "sigmoid(0) = 0.5");

    // sigmoid output in (0, 1)
    let sig_wide = SiluKernelV1::sigmoid(&k, &[-10.0, 0.0, 10.0]);
    for &v in &sig_wide {
        assert!(v > 0.0 && v < 1.0, "sigmoid output must be in (0,1)");
    }

    // silu(0) = 0
    let silu_zero = SiluKernelV1::silu(&k, &[0.0]);
    assert!(silu_zero[0].abs() < 1e-6, "SiLU(0) = 0");
}

#[test]
fn rmsnorm_kernel_v1_properties() {
    let k = ReferenceKernels;

    let out = k.rmsnorm(&[3.0, 4.0]);
    // RMS of [3,4] = sqrt((9+16)/2) = sqrt(12.5) ≈ 3.5355
    // out ≈ [3/3.5355, 4/3.5355] ≈ [0.8485, 1.1314]
    assert!(out.len() == 2);
    let rms_out = (out.iter().map(|v| v * v).sum::<f32>() / out.len() as f32).sqrt();
    assert!(
        (rms_out - 1.0).abs() < 1e-3,
        "rmsnorm output should have ~unit RMS"
    );
}

#[test]
fn layernorm_kernel_v1_properties() {
    let k = ReferenceKernels;

    // With gamma = [1,1,1], layernorm should produce zero-mean, unit-variance output
    let out = k.layernorm(&[1.0, 2.0, 3.0], &[1.0, 1.0, 1.0]);
    let mean: f32 = out.iter().sum::<f32>() / out.len() as f32;
    assert!(mean.abs() < 1e-5, "layernorm output should be zero-mean");

    // statistics returns [mean, variance]
    let stats = k.statistics(&[2.0, 4.0, 6.0]);
    assert_eq!(stats.len(), 2);
    assert!((stats[0] - 4.0).abs() < 1e-6, "mean of [2,4,6] = 4");
    // var = ((2-4)^2 + (4-4)^2 + (6-4)^2) / 3 = 8/3 ≈ 2.6667
    assert!((stats[1] - 8.0 / 3.0).abs() < 1e-5, "var of [2,4,6] = 8/3");
}

#[test]
fn cross_entropy_kernel_v1_properties() {
    let k = ReferenceKernels;

    // log_softmax: output should be negative, logsumexp = 0
    let lsm = CrossEntropyKernelV1::log_softmax(&k, &[1.0, 2.0, 3.0]);
    assert_eq!(lsm.len(), 3);
    for &v in &lsm {
        assert!(v <= 0.0, "log_softmax values must be <= 0");
    }
    // exp(log_softmax) should sum to 1
    let sum_exp: f32 = lsm.iter().map(|v| v.exp()).sum();
    assert!(
        (sum_exp - 1.0).abs() < 1e-5,
        "exp(log_softmax) must sum to 1"
    );

    // cross_entropy with one-hot target
    let targets = vec![0.0, 0.0, 1.0];
    let logits = vec![1.0, 2.0, 3.0];
    let loss = k.cross_entropy(&targets, &logits);
    assert_eq!(loss.len(), 1);
    assert!(loss[0] > 0.0, "cross-entropy loss must be positive");
}

#[test]
fn swiglu_kernel_v1_properties() {
    let k = ReferenceKernels;

    // SwigluKernelV1::silu(0) = 0
    let silu_zero = SwigluKernelV1::silu(&k, &[0.0]);
    assert!(silu_zero[0].abs() < 1e-6, "SwigluKernelV1::silu(0) = 0");

    // swiglu with identity weights and zero biases
    let x = vec![1.0, 2.0];
    let w = vec![1.0, 1.0]; // gate weights
    let v = vec![1.0, 1.0]; // value weights
    let b = vec![0.0, 0.0]; // gate bias (unused in ref impl)
    let c = vec![0.0, 0.0]; // value bias (unused in ref impl)
    let out = k.swiglu(&x, &w, &v, &b, &c);
    assert_eq!(out.len(), 2);
    // swiglu(x, I, I, 0, 0) = silu(x) * x
    for (i, &xi) in x.iter().enumerate() {
        let expected = (xi / (1.0 + (-xi).exp())) * xi;
        assert!(
            (out[i] - expected).abs() < 1e-5,
            "swiglu with identity weights: element {i}"
        );
    }
}

// ---------------------------------------------------------------------------
// AttentionKernelV1 -- naive scaled dot-product attention
// ---------------------------------------------------------------------------
impl AttentionKernelV1 for ReferenceKernels {
    fn attention(&self, q: &[f32], k: &[f32], v: &[f32]) -> Vec<f32> {
        let n = (q.len() as f32).sqrt() as usize;
        if n == 0 {
            return vec![];
        }
        let d = q.len() / n;
        let mut out = vec![0.0f32; n * d];
        for i in 0..n {
            let mut scores = vec![0.0f32; n];
            for j in 0..n {
                for kk in 0..d {
                    scores[j] += q[i * d + kk] * k[j * d + kk];
                }
            }
            let scale = (d as f32).sqrt();
            let max = scores.iter().copied().fold(f32::NEG_INFINITY, f32::max);
            let exps: Vec<f32> = scores.iter().map(|s| ((s / scale) - max).exp()).collect();
            let sum: f32 = exps.iter().sum();
            for j in 0..n {
                for kk in 0..d {
                    out[i * d + kk] += (exps[j] / sum) * v[j * d + kk];
                }
            }
        }
        out
    }
}

// ---------------------------------------------------------------------------
// FlashAttentionV1 -- same math as attention (flash is an optimization)
// ---------------------------------------------------------------------------
impl FlashAttentionV1 for ReferenceKernels {
    fn flash_attention(&self, q: &[f32], k: &[f32], v: &[f32]) -> Vec<f32> {
        AttentionKernelV1::attention(self, q, k, v)
    }
}

// ---------------------------------------------------------------------------
// GqaKernelV1 -- grouped-query attention (equal heads = standard attention)
// ---------------------------------------------------------------------------
impl GqaKernelV1 for ReferenceKernels {
    fn gqa(&self, q: &[f32], k: &[f32], v: &[f32]) -> Vec<f32> {
        AttentionKernelV1::attention(self, q, k, v)
    }
}

// ---------------------------------------------------------------------------
// MatmulKernelV1 -- naive matrix multiplication (square matrices)
// ---------------------------------------------------------------------------
impl MatmulKernelV1 for ReferenceKernels {
    fn matmul(&self, a: &[f32], b: &[f32]) -> Vec<f32> {
        let n = (a.len() as f32).sqrt() as usize;
        if n == 0 {
            return vec![];
        }
        let mut c = vec![0.0f32; n * n];
        for i in 0..n {
            for j in 0..n {
                for k in 0..n {
                    c[i * n + j] += a[i * n + k] * b[k * n + j];
                }
            }
        }
        c
    }

    fn quantized_dot(&self, b: &[f32], s_b: f32) -> Vec<f32> {
        vec![b.iter().sum::<f32>() * s_b]
    }
}

// ---------------------------------------------------------------------------
// RopeKernelV1 -- rotary position embeddings
// ---------------------------------------------------------------------------
impl RopeKernelV1 for ReferenceKernels {
    fn rope(&self, x: &[f32], m: &[f32]) -> Vec<f32> {
        let pos = m.first().copied().unwrap_or(0.0);
        let mut out = x.to_vec();
        for i in (0..x.len()).step_by(2) {
            if i + 1 < x.len() {
                let theta = pos / 10000_f32.powf(i as f32 / x.len() as f32);
                let (sin_t, cos_t) = theta.sin_cos();
                out[i] = x[i] * cos_t - x[i + 1] * sin_t;
                out[i + 1] = x[i] * sin_t + x[i + 1] * cos_t;
            }
        }
        out
    }
}

// ---------------------------------------------------------------------------
// AdamwKernelV1 -- AdamW optimizer moments and weight update
// ---------------------------------------------------------------------------
impl AdamwKernelV1 for ReferenceKernels {
    fn adam_moments(&self, g_t: &[f32]) -> Vec<f32> {
        g_t.iter().map(|g| 0.9 * 0.0 + 0.1 * g).collect()
    }

    fn adam_variance(&self, g_t: &[f32]) -> Vec<f32> {
        g_t.iter().map(|g| 0.999 * 0.0 + 0.001 * g * g).collect()
    }

    fn bias_correction(&self, input: &[f32]) -> Vec<f32> {
        input.iter().map(|v| v / (1.0 - 0.9)).collect()
    }

    fn weight_update(&self, theta: &[f32]) -> Vec<f32> {
        theta.iter().map(|t| t - 0.001 * t).collect()
    }
}

#[test]
fn attention_output_size() {
    let k = ReferenceKernels;
    let out = AttentionKernelV1::attention(&k, &[1.0; 4], &[1.0; 4], &[1.0; 4]);
    assert_eq!(out.len(), 4);
}

#[test]
fn matmul_identity() {
    let k = ReferenceKernels;
    let identity = vec![1.0, 0.0, 0.0, 1.0]; // 2x2 identity
    let a = vec![1.0, 2.0, 3.0, 4.0];
    let out = MatmulKernelV1::matmul(&k, &a, &identity);
    assert!((out[0] - 1.0).abs() < 1e-5 && (out[3] - 4.0).abs() < 1e-5);
}

#[test]
fn rope_preserves_norm() {
    let k = ReferenceKernels;
    let x = vec![1.0, 0.0, 0.0, 1.0];
    let out = RopeKernelV1::rope(&k, &x, &[1.0]);
    let norm_in: f32 = x.iter().map(|v| v * v).sum::<f32>().sqrt();
    let norm_out: f32 = out.iter().map(|v| v * v).sum::<f32>().sqrt();
    assert!((norm_in - norm_out).abs() < 1e-5);
}