apr-cli 0.32.0

CLI tool for APR model inspection, debugging, and operations
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


// ════════════════════════════════════════════════════════════════════
// Coverage tests for auto_diagnose (PMAT coverage gap)
// ════════════════════════════════════════════════════════════════════

#[cfg(test)]
mod diagnose_tests {
    use super::*;

    // Reuse the make_metrics helper from parity_spc_tests
    fn diagnose_metrics(
        cosine_similarity: f32,
        max_abs_diff: f32,
        cpu_argmax: u32,
        gpu_argmax: u32,
        cpu_nan: usize,
        gpu_nan: usize,
        position: usize,
    ) -> SpcMetrics {
        SpcMetrics {
            position,
            token_id: position as u32,
            cpu_argmax,
            gpu_argmax,
            _cpu_top_logit: 1.0,
            _gpu_top_logit: 1.0,
            max_abs_diff,
            _max_diff_idx: 0,
            mean_abs_diff: max_abs_diff * 0.5,
            rmse: max_abs_diff * 0.3,
            cosine_similarity,
            kl_divergence: if cosine_similarity > 0.999 { 0.001 } else { 5.0 },
            sigma_level: if cosine_similarity > 0.999 { 6.0 } else { 0.5 },
            cpu_nan,
            gpu_nan,
            out_of_spec_count: 0,
            vocab_size: 32000,
        }
    }

    // ── auto_diagnose: no failures ─────────────────────────────────────

    #[test]
    fn test_auto_diagnose_no_failures() {
        // All passing — should return early without diagnosis
        let metrics = vec![
            diagnose_metrics(1.0, 0.0, 42, 42, 0, 0, 0),
            diagnose_metrics(1.0, 0.0, 42, 42, 0, 0, 1),
            diagnose_metrics(1.0, 0.0, 42, 42, 0, 0, 2),
        ];
        // Should not panic, returns early
        auto_diagnose(&metrics, 3584, 28, 4);
    }

    #[test]
    fn test_auto_diagnose_empty_metrics() {
        auto_diagnose(&[], 3584, 28, 4);
    }

    // ── Pattern 1: Position 0 fails (core layer bug) ───────────────────

    #[test]
    fn test_auto_diagnose_pos0_catastrophic() {
        let metrics = vec![
            diagnose_metrics(0.5, 10.0, 42, 100, 0, 0, 0), // catastrophic at pos 0
            diagnose_metrics(0.5, 10.0, 42, 100, 0, 0, 1),
            diagnose_metrics(0.5, 10.0, 42, 100, 0, 0, 2),
        ];
        // Should output WHY 1, WHY 2 (catastrophic), WHY 3 (all fail + pos0)
        // + likely root causes + falsification tests
        auto_diagnose(&metrics, 3584, 28, 4);
    }

    // ── Pattern 2: Growing divergence (KV cache accumulation) ──────────

    #[test]
    fn test_auto_diagnose_growing_divergence() {
        let metrics = vec![
            diagnose_metrics(0.9999, 0.1, 42, 42, 0, 0, 0),  // pass
            diagnose_metrics(0.998, 0.5, 42, 42, 0, 0, 1),    // divergent
            diagnose_metrics(0.995, 1.5, 42, 43, 0, 0, 2),    // divergent, growing
            diagnose_metrics(0.990, 3.0, 42, 50, 0, 0, 3),    // more divergent
            diagnose_metrics(0.980, 6.0, 42, 100, 0, 0, 4),   // much more
            diagnose_metrics(0.950, 12.0, 42, 200, 0, 0, 5),  // catastrophic
        ];
        // Should detect growing pattern
        auto_diagnose(&metrics, 4096, 32, 8);
    }

    // ── Pattern 3: High cosine but wrong argmax ────────────────────────

    #[test]
    fn test_auto_diagnose_high_cos_wrong_argmax() {
        let metrics = vec![
            diagnose_metrics(0.9999, 0.1, 42, 42, 0, 0, 0),  // pass
            diagnose_metrics(0.995, 0.5, 42, 43, 0, 0, 1),    // divergent, wrong argmax
        ];
        // Should trigger Pattern 3 (high cosine wrong argmax)
        auto_diagnose(&metrics, 2048, 16, 4);
    }

    // ── Pattern 5: All positions fail uniformly ────────────────────────

    #[test]
    fn test_auto_diagnose_all_fail_uniformly() {
        let metrics = vec![
            diagnose_metrics(0.5, 10.0, 42, 100, 0, 0, 0),
            diagnose_metrics(0.5, 10.0, 42, 100, 0, 0, 1),
            diagnose_metrics(0.5, 10.0, 42, 100, 0, 0, 2),
            diagnose_metrics(0.5, 10.0, 42, 100, 0, 0, 3),
        ];
        // Pattern 5: all_fail + pos0_fails
        auto_diagnose(&metrics, 1536, 12, 4);
    }

    // ── NaN failures ───────────────────────────────────────────────────

    #[test]
    fn test_auto_diagnose_nan_failures() {
        let metrics = vec![
            diagnose_metrics(1.0, 0.0, 42, 42, 3, 0, 0), // NaN on CPU
            diagnose_metrics(1.0, 0.0, 42, 42, 0, 5, 1), // NaN on GPU
        ];
        auto_diagnose(&metrics, 3584, 28, 4);
    }

    // ── Edge case: zero kv_heads ───────────────────────────────────────

    #[test]
    fn test_auto_diagnose_zero_kv_heads() {
        let metrics = vec![
            diagnose_metrics(0.5, 10.0, 42, 100, 0, 0, 0),
        ];
        // Should not divide by zero
        auto_diagnose(&metrics, 3584, 28, 0);
    }

    // ── Edge case: zero num_heads ──────────────────────────────────────

    #[test]
    fn test_auto_diagnose_zero_num_heads() {
        let metrics = vec![
            diagnose_metrics(0.5, 10.0, 42, 100, 0, 0, 0),
        ];
        // Should not divide by zero
        auto_diagnose(&metrics, 3584, 0, 0);
    }

    // ── Mixed pass/fail ────────────────────────────────────────────────

    #[test]
    fn test_auto_diagnose_mixed_pass_fail() {
        let metrics = vec![
            diagnose_metrics(1.0, 0.0, 42, 42, 0, 0, 0),   // pass
            diagnose_metrics(1.0, 0.0, 42, 42, 0, 0, 1),   // pass
            diagnose_metrics(0.995, 0.5, 42, 43, 0, 0, 2),  // divergent
            diagnose_metrics(1.0, 0.0, 42, 42, 0, 0, 3),   // pass
        ];
        // Some fail, pos0 passes — should NOT trigger pattern 1 or 5
        auto_diagnose(&metrics, 4096, 32, 8);
    }

    // ── Verdict helper coverage ────────────────────────────────────────

    #[test]
    fn test_verdict_symbol_all_variants() {
        // Exercise all verdict symbol paths
        let _pass = Verdict::Pass.symbol();
        let _warn_argmax = Verdict::WarnArgmax.symbol();
        let _warn_oos = Verdict::WarnOutOfSpec.symbol();
        let _fail_div = Verdict::FailDivergent.symbol();
        let _fail_cat = Verdict::FailCatastrophic.symbol();
        let _fail_nan = Verdict::FailNan.symbol();
    }

    #[test]
    fn test_format_diff_ranges() {
        // Exercise all format_diff color paths
        let _tiny = format_diff(0.05);
        let _medium = format_diff(0.5);
        let _high = format_diff(2.5);
        let _very_high = format_diff(10.0);
    }

    #[test]
    fn test_format_cosine_ranges() {
        // Exercise all format_cosine color paths
        let _perfect = format_cosine(0.99999);
        let _good = format_cosine(0.9995);
        let _warning = format_cosine(0.995);
        let _bad = format_cosine(0.95);
        let _catastrophic = format_cosine(0.5);
    }

    // ── print_header / print_footer / print_row ────────────────────────

    #[test]
    fn test_print_header_no_panic() {
        print_header();
    }

    #[test]
    fn test_print_footer_no_panic() {
        print_footer();
    }

    #[test]
    fn test_print_row_pass() {
        let m = SpcMetrics {
            position: 0,
            token_id: 42,
            cpu_argmax: 100,
            gpu_argmax: 100,
            _cpu_top_logit: 5.0,
            _gpu_top_logit: 5.0,
            max_abs_diff: 0.001,
            _max_diff_idx: 0,
            mean_abs_diff: 0.0005,
            rmse: 0.0003,
            cosine_similarity: 0.99999,
            kl_divergence: 0.0001,
            sigma_level: 6.5,
            cpu_nan: 0,
            gpu_nan: 0,
            out_of_spec_count: 0,
            vocab_size: 32000,
        };
        print_row(&m);
    }

    #[test]
    fn test_print_row_fail() {
        let m = SpcMetrics {
            position: 5,
            token_id: 999,
            cpu_argmax: 100,
            gpu_argmax: 200,
            _cpu_top_logit: 5.0,
            _gpu_top_logit: -3.0,
            max_abs_diff: 10.0,
            _max_diff_idx: 42,
            mean_abs_diff: 5.0,
            rmse: 3.0,
            cosine_similarity: 0.5,
            kl_divergence: 5.0,
            sigma_level: 0.5,
            cpu_nan: 0,
            gpu_nan: 0,
            out_of_spec_count: 1000,
            vocab_size: 32000,
        };
        print_row(&m);
    }

    #[test]
    fn test_print_row_nan_verdict() {
        let m = SpcMetrics {
            position: 0,
            token_id: 1,
            cpu_argmax: 0,
            gpu_argmax: 0,
            _cpu_top_logit: 0.0,
            _gpu_top_logit: 0.0,
            max_abs_diff: 0.0,
            _max_diff_idx: 0,
            mean_abs_diff: 0.0,
            rmse: 0.0,
            cosine_similarity: 1.0,
            kl_divergence: 0.0,
            sigma_level: 99.0,
            cpu_nan: 5,
            gpu_nan: 0,
            out_of_spec_count: 0,
            vocab_size: 32000,
        };
        print_row(&m);
    }
}