[pilot-probe] runs/pilot-probe/artifacts/pilot/pilot_results.json
Browse files
runs/pilot-probe/artifacts/pilot/pilot_results.json
ADDED
|
@@ -0,0 +1,378 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"run_id": "pilot-probe",
|
| 3 |
+
"created_at": "2026-08-07T05:11:20Z",
|
| 4 |
+
"scale": "PILOT -- real measurements, underpowered, 1B-scale models",
|
| 5 |
+
"conditions": [
|
| 6 |
+
{
|
| 7 |
+
"condition": "olmo2_1b:base",
|
| 8 |
+
"stage": "base",
|
| 9 |
+
"stage_index": 0,
|
| 10 |
+
"n_cases": 47,
|
| 11 |
+
"coverage": {
|
| 12 |
+
"recall@1": 0.0,
|
| 13 |
+
"recall@5": 0.0,
|
| 14 |
+
"soft_coverage": 0.0
|
| 15 |
+
},
|
| 16 |
+
"mean_tau": null,
|
| 17 |
+
"premature_rate": 0.0,
|
| 18 |
+
"mean_vendi_final": 0.0,
|
| 19 |
+
"mean_rho": -2.3889260797097086e-16,
|
| 20 |
+
"mean_tokens": 384.0
|
| 21 |
+
},
|
| 22 |
+
{
|
| 23 |
+
"condition": "olmo2_1b:sft",
|
| 24 |
+
"stage": "sft",
|
| 25 |
+
"stage_index": 1,
|
| 26 |
+
"n_cases": 47,
|
| 27 |
+
"coverage": {
|
| 28 |
+
"recall@1": 0.0,
|
| 29 |
+
"recall@5": 0.0,
|
| 30 |
+
"soft_coverage": 0.0
|
| 31 |
+
},
|
| 32 |
+
"mean_tau": null,
|
| 33 |
+
"premature_rate": 0.0,
|
| 34 |
+
"mean_vendi_final": 0.0,
|
| 35 |
+
"mean_rho": -3.4094726818497813e-16,
|
| 36 |
+
"mean_tokens": 140.40425531914894
|
| 37 |
+
},
|
| 38 |
+
{
|
| 39 |
+
"condition": "olmo2_1b:dpo",
|
| 40 |
+
"stage": "dpo",
|
| 41 |
+
"stage_index": 2,
|
| 42 |
+
"n_cases": 47,
|
| 43 |
+
"coverage": {
|
| 44 |
+
"recall@1": 0.0,
|
| 45 |
+
"recall@5": 0.0,
|
| 46 |
+
"soft_coverage": 0.0425531914893617
|
| 47 |
+
},
|
| 48 |
+
"mean_tau": 0.25,
|
| 49 |
+
"premature_rate": 0.02127659574468085,
|
| 50 |
+
"mean_vendi_final": 0.0425531914893617,
|
| 51 |
+
"mean_rho": -3.398491157786818e-16,
|
| 52 |
+
"mean_tokens": 384.0
|
| 53 |
+
},
|
| 54 |
+
{
|
| 55 |
+
"condition": "olmo2_1b:instruct",
|
| 56 |
+
"stage": "instruct",
|
| 57 |
+
"stage_index": 3,
|
| 58 |
+
"n_cases": 47,
|
| 59 |
+
"coverage": {
|
| 60 |
+
"recall@1": 0.0,
|
| 61 |
+
"recall@5": 0.0,
|
| 62 |
+
"soft_coverage": 0.0
|
| 63 |
+
},
|
| 64 |
+
"mean_tau": null,
|
| 65 |
+
"premature_rate": 0.0,
|
| 66 |
+
"mean_vendi_final": 0.06062518051803314,
|
| 67 |
+
"mean_rho": -5.003956006969916e-16,
|
| 68 |
+
"mean_tokens": 382.1914893617021
|
| 69 |
+
},
|
| 70 |
+
{
|
| 71 |
+
"condition": "olmo2_1b:rlvr",
|
| 72 |
+
"stage": "rlvr",
|
| 73 |
+
"stage_index": 4,
|
| 74 |
+
"n_cases": 47,
|
| 75 |
+
"coverage": {
|
| 76 |
+
"recall@1": 0.0,
|
| 77 |
+
"recall@5": 0.0,
|
| 78 |
+
"soft_coverage": 0.0
|
| 79 |
+
},
|
| 80 |
+
"mean_tau": null,
|
| 81 |
+
"premature_rate": 0.0,
|
| 82 |
+
"mean_vendi_final": 0.02127659574468085,
|
| 83 |
+
"mean_rho": -6.366737668598568e-16,
|
| 84 |
+
"mean_tokens": 380.0851063829787
|
| 85 |
+
}
|
| 86 |
+
],
|
| 87 |
+
"v4_redundancy": {
|
| 88 |
+
"n": 235,
|
| 89 |
+
"phi": null,
|
| 90 |
+
"verdict": "correctness is constant",
|
| 91 |
+
"flag_rate": 0.00425531914893617,
|
| 92 |
+
"accuracy": 0.0,
|
| 93 |
+
"passed": null
|
| 94 |
+
},
|
| 95 |
+
"v1_manipulation": {
|
| 96 |
+
"orderings": {
|
| 97 |
+
"early": {
|
| 98 |
+
"n": 44,
|
| 99 |
+
"n_decidable": 44,
|
| 100 |
+
"premature_rate": 0.0,
|
| 101 |
+
"mean_tau": NaN
|
| 102 |
+
},
|
| 103 |
+
"late": {
|
| 104 |
+
"n": 44,
|
| 105 |
+
"n_decidable": 44,
|
| 106 |
+
"premature_rate": 0.0,
|
| 107 |
+
"mean_tau": NaN
|
| 108 |
+
},
|
| 109 |
+
"shuffled": {
|
| 110 |
+
"n": 44,
|
| 111 |
+
"n_decidable": 44,
|
| 112 |
+
"premature_rate": 0.022727272727272728,
|
| 113 |
+
"mean_tau": NaN
|
| 114 |
+
}
|
| 115 |
+
},
|
| 116 |
+
"delta_late_minus_early": 0.0,
|
| 117 |
+
"direction_correct": false,
|
| 118 |
+
"n_paired": 22,
|
| 119 |
+
"paired_mean_difference": 0.0,
|
| 120 |
+
"paired_se": 0.0,
|
| 121 |
+
"paired_ci95": [
|
| 122 |
+
0.0,
|
| 123 |
+
0.0
|
| 124 |
+
],
|
| 125 |
+
"ci_excludes_zero": false,
|
| 126 |
+
"discordant_late_only": 0,
|
| 127 |
+
"discordant_early_only": 0,
|
| 128 |
+
"passed": false,
|
| 129 |
+
"verdict": "NOT ESTABLISHED at pilot scale -- direction WRONG, CI includes zero"
|
| 130 |
+
},
|
| 131 |
+
"e1_prevalence": {
|
| 132 |
+
"n": 235,
|
| 133 |
+
"outcome": "recall_at_5",
|
| 134 |
+
"model": {
|
| 135 |
+
"formula": "recall_at_5 ~ stage_index * _prev_c",
|
| 136 |
+
"backend": "statsmodels_mixedlm",
|
| 137 |
+
"converged": false,
|
| 138 |
+
"primary_term": "stage_index:_prev_c",
|
| 139 |
+
"estimate": -0.0,
|
| 140 |
+
"ci": [
|
| 141 |
+
NaN,
|
| 142 |
+
NaN
|
| 143 |
+
],
|
| 144 |
+
"p_value": NaN,
|
| 145 |
+
"crosses_zero": "False"
|
| 146 |
+
},
|
| 147 |
+
"n_head": 150,
|
| 148 |
+
"n_tail": 85,
|
| 149 |
+
"by_stage": {
|
| 150 |
+
"base": {
|
| 151 |
+
"stage_index": 0,
|
| 152 |
+
"n": 47,
|
| 153 |
+
"overall": 0.0,
|
| 154 |
+
"head": 0.0,
|
| 155 |
+
"tail": 0.0,
|
| 156 |
+
"n_head": 30,
|
| 157 |
+
"n_tail": 17,
|
| 158 |
+
"mean_tau": null,
|
| 159 |
+
"premature_rate": 0.0
|
| 160 |
+
},
|
| 161 |
+
"sft": {
|
| 162 |
+
"stage_index": 1,
|
| 163 |
+
"n": 47,
|
| 164 |
+
"overall": 0.0,
|
| 165 |
+
"head": 0.0,
|
| 166 |
+
"tail": 0.0,
|
| 167 |
+
"n_head": 30,
|
| 168 |
+
"n_tail": 17,
|
| 169 |
+
"mean_tau": null,
|
| 170 |
+
"premature_rate": 0.0
|
| 171 |
+
},
|
| 172 |
+
"dpo": {
|
| 173 |
+
"stage_index": 2,
|
| 174 |
+
"n": 47,
|
| 175 |
+
"overall": 0.0,
|
| 176 |
+
"head": 0.0,
|
| 177 |
+
"tail": 0.0,
|
| 178 |
+
"n_head": 30,
|
| 179 |
+
"n_tail": 17,
|
| 180 |
+
"mean_tau": 0.25,
|
| 181 |
+
"premature_rate": 0.02127659574468085
|
| 182 |
+
},
|
| 183 |
+
"instruct": {
|
| 184 |
+
"stage_index": 3,
|
| 185 |
+
"n": 47,
|
| 186 |
+
"overall": 0.0,
|
| 187 |
+
"head": 0.0,
|
| 188 |
+
"tail": 0.0,
|
| 189 |
+
"n_head": 30,
|
| 190 |
+
"n_tail": 17,
|
| 191 |
+
"mean_tau": null,
|
| 192 |
+
"premature_rate": 0.0
|
| 193 |
+
},
|
| 194 |
+
"rlvr": {
|
| 195 |
+
"stage_index": 4,
|
| 196 |
+
"n": 47,
|
| 197 |
+
"overall": 0.0,
|
| 198 |
+
"head": 0.0,
|
| 199 |
+
"tail": 0.0,
|
| 200 |
+
"n_head": 30,
|
| 201 |
+
"n_tail": 17,
|
| 202 |
+
"mean_tau": null,
|
| 203 |
+
"premature_rate": 0.0
|
| 204 |
+
}
|
| 205 |
+
},
|
| 206 |
+
"delta_head": 0.0,
|
| 207 |
+
"delta_tail": 0.0,
|
| 208 |
+
"head_tail_divergence": 0.0,
|
| 209 |
+
"law_direction": "inconsistent with H2 -- supports the 'RLVR extends the boundary' position"
|
| 210 |
+
},
|
| 211 |
+
"case_store": {
|
| 212 |
+
"n_loaded": 180,
|
| 213 |
+
"n_usable": 92,
|
| 214 |
+
"n_selected": 47,
|
| 215 |
+
"by_stratum": {
|
| 216 |
+
"uncommon": 9,
|
| 217 |
+
"ultra_rare": 10,
|
| 218 |
+
"rare": 23,
|
| 219 |
+
"common": 5
|
| 220 |
+
},
|
| 221 |
+
"report": {
|
| 222 |
+
"n_cases": 180,
|
| 223 |
+
"by_corpus": {
|
| 224 |
+
"rarebench": 60,
|
| 225 |
+
"rarearena": 60,
|
| 226 |
+
"ddxplus": 60
|
| 227 |
+
},
|
| 228 |
+
"by_stratum": {
|
| 229 |
+
"unknown": 88,
|
| 230 |
+
"rare": 68,
|
| 231 |
+
"common": 5,
|
| 232 |
+
"uncommon": 9,
|
| 233 |
+
"ultra_rare": 10
|
| 234 |
+
},
|
| 235 |
+
"n_gold_linked": 152,
|
| 236 |
+
"n_prevalence_known": 92,
|
| 237 |
+
"n_with_differential": 60,
|
| 238 |
+
"n_with_hpo": 60,
|
| 239 |
+
"n_with_fstar": 60,
|
| 240 |
+
"n_dolma_queried": 180,
|
| 241 |
+
"linker_stats": {
|
| 242 |
+
"n": 630,
|
| 243 |
+
"linked": 408,
|
| 244 |
+
"abstain_rate": 0.3523809523809524,
|
| 245 |
+
"ambiguous_rate": 0.0,
|
| 246 |
+
"tier_abstain": 222,
|
| 247 |
+
"tier_exact_synonym": 86,
|
| 248 |
+
"tier_exact_label": 270,
|
| 249 |
+
"tier_normalised": 52
|
| 250 |
+
},
|
| 251 |
+
"warnings": [],
|
| 252 |
+
"gold_link_rate": 0.8444444444444444,
|
| 253 |
+
"prevalence_known_rate": 0.5111111111111111
|
| 254 |
+
},
|
| 255 |
+
"kernels": {
|
| 256 |
+
"n_vocab": 37,
|
| 257 |
+
"bc_offdiag_min": 1.5443506154326965e-05,
|
| 258 |
+
"bc_offdiag_mean": 0.3860080295489061,
|
| 259 |
+
"tv_offdiag_min": 0.9982421697485498,
|
| 260 |
+
"tv_offdiag_mean": 0.9996610369991826,
|
| 261 |
+
"bc_audit": {
|
| 262 |
+
"name": "k_disc_bhattacharyya",
|
| 263 |
+
"n_diseases": 37,
|
| 264 |
+
"min_eigenvalue": -1.732733864965153e-15,
|
| 265 |
+
"psd_projected": false,
|
| 266 |
+
"projection_mass": 0.0,
|
| 267 |
+
"mean_offdiag": 0.3860080295489061,
|
| 268 |
+
"diag_is_unit": true,
|
| 269 |
+
"variant": "bhattacharyya",
|
| 270 |
+
"n_features": 4000,
|
| 271 |
+
"n_without_profile": 0,
|
| 272 |
+
"psd_by_construction": true,
|
| 273 |
+
"smoothing": 0.01
|
| 274 |
+
},
|
| 275 |
+
"tv_audit": {
|
| 276 |
+
"name": "k_disc_tv",
|
| 277 |
+
"n_diseases": 37,
|
| 278 |
+
"min_eigenvalue": -3.9098019020324435e-15,
|
| 279 |
+
"psd_projected": false,
|
| 280 |
+
"projection_mass": 0.0,
|
| 281 |
+
"mean_offdiag": 0.9996610369991826,
|
| 282 |
+
"diag_is_unit": true,
|
| 283 |
+
"variant": "tv",
|
| 284 |
+
"n_features": 4000,
|
| 285 |
+
"n_without_profile": 0,
|
| 286 |
+
"psd_by_construction": false,
|
| 287 |
+
"smoothing": 0.01
|
| 288 |
+
}
|
| 289 |
+
},
|
| 290 |
+
"ontology_version": "963c60bbe2eaa862",
|
| 291 |
+
"seconds": 43.696001441
|
| 292 |
+
},
|
| 293 |
+
"kernels": {
|
| 294 |
+
"n_vocab": 37,
|
| 295 |
+
"bc_offdiag_min": 1.5443506154326965e-05,
|
| 296 |
+
"bc_offdiag_mean": 0.3860080295489061,
|
| 297 |
+
"tv_offdiag_min": 0.9982421697485498,
|
| 298 |
+
"tv_offdiag_mean": 0.9996610369991826,
|
| 299 |
+
"bc_audit": {
|
| 300 |
+
"name": "k_disc_bhattacharyya",
|
| 301 |
+
"n_diseases": 37,
|
| 302 |
+
"min_eigenvalue": -1.732733864965153e-15,
|
| 303 |
+
"psd_projected": false,
|
| 304 |
+
"projection_mass": 0.0,
|
| 305 |
+
"mean_offdiag": 0.3860080295489061,
|
| 306 |
+
"diag_is_unit": true,
|
| 307 |
+
"variant": "bhattacharyya",
|
| 308 |
+
"n_features": 4000,
|
| 309 |
+
"n_without_profile": 0,
|
| 310 |
+
"psd_by_construction": true,
|
| 311 |
+
"smoothing": 0.01
|
| 312 |
+
},
|
| 313 |
+
"tv_audit": {
|
| 314 |
+
"name": "k_disc_tv",
|
| 315 |
+
"n_diseases": 37,
|
| 316 |
+
"min_eigenvalue": -3.9098019020324435e-15,
|
| 317 |
+
"psd_projected": false,
|
| 318 |
+
"projection_mass": 0.0,
|
| 319 |
+
"mean_offdiag": 0.9996610369991826,
|
| 320 |
+
"diag_is_unit": true,
|
| 321 |
+
"variant": "tv",
|
| 322 |
+
"n_features": 4000,
|
| 323 |
+
"n_without_profile": 0,
|
| 324 |
+
"psd_by_construction": false,
|
| 325 |
+
"smoothing": 0.01
|
| 326 |
+
}
|
| 327 |
+
},
|
| 328 |
+
"spend": {
|
| 329 |
+
"ceiling_usd": 15.0,
|
| 330 |
+
"spent_usd": 0.6871,
|
| 331 |
+
"remaining_usd": 14.3129,
|
| 332 |
+
"entries": [
|
| 333 |
+
{
|
| 334 |
+
"stage": "S0 case store",
|
| 335 |
+
"gpu": "cpu",
|
| 336 |
+
"seconds": 48.0,
|
| 337 |
+
"usd": 0.0025,
|
| 338 |
+
"at": "2026-08-07T04:33:54Z"
|
| 339 |
+
},
|
| 340 |
+
{
|
| 341 |
+
"stage": "gen olmo2_1b:base",
|
| 342 |
+
"gpu": "A10G",
|
| 343 |
+
"seconds": 751.0,
|
| 344 |
+
"usd": 0.2295,
|
| 345 |
+
"at": "2026-08-07T04:46:25Z"
|
| 346 |
+
},
|
| 347 |
+
{
|
| 348 |
+
"stage": "gen olmo2_1b:sft",
|
| 349 |
+
"gpu": "A10G",
|
| 350 |
+
"seconds": 134.4,
|
| 351 |
+
"usd": 0.0411,
|
| 352 |
+
"at": "2026-08-07T04:48:40Z"
|
| 353 |
+
},
|
| 354 |
+
{
|
| 355 |
+
"stage": "gen olmo2_1b:dpo",
|
| 356 |
+
"gpu": "A10G",
|
| 357 |
+
"seconds": 315.5,
|
| 358 |
+
"usd": 0.0964,
|
| 359 |
+
"at": "2026-08-07T04:53:55Z"
|
| 360 |
+
},
|
| 361 |
+
{
|
| 362 |
+
"stage": "gen olmo2_1b:instruct",
|
| 363 |
+
"gpu": "A10G",
|
| 364 |
+
"seconds": 306.5,
|
| 365 |
+
"usd": 0.0937,
|
| 366 |
+
"at": "2026-08-07T04:59:02Z"
|
| 367 |
+
},
|
| 368 |
+
{
|
| 369 |
+
"stage": "gen olmo2_1b:rlvr",
|
| 370 |
+
"gpu": "A10G",
|
| 371 |
+
"seconds": 732.7,
|
| 372 |
+
"usd": 0.2239,
|
| 373 |
+
"at": "2026-08-07T05:11:15Z"
|
| 374 |
+
}
|
| 375 |
+
]
|
| 376 |
+
},
|
| 377 |
+
"warnings": []
|
| 378 |
+
}
|