Upload add_sub_baseline_10K_1L3H510d

add_sub_baseline_10K_1L3H510d/metrics.json

@@ -65,132 +65,132 @@
       3100
     ],
     "loss": [
-      7.182804584503174,
-      4.20333194732666,
-      2.0677404403686523,
-      1.829144835472107,
-      1.8410581350326538,
-      1.820699691772461,
-      1.7079013586044312,
-      1.655383825302124,
-      1.6272556781768799,
-      1.56789231300354,
-      1.2935447692871094,
-      0.8198140263557434,
-      0.7896751761436462,
-      0.7217035293579102,
-      0.6449242830276489,
-      0.6155810356140137,
-      0.6273579001426697,
-      0.5691636800765991,
-      0.5195116400718689,
-      0.5188413262367249,
-      0.4571366012096405,
-      0.46536850929260254,
-      0.45766669511795044,
-      0.4275817275047302,
-      0.40487247705459595,
-      0.4464508593082428,
-      0.39323002099990845,
-      0.39669889211654663,
-      0.34110480546951294,
-      0.3717530369758606,
-      0.3865724503993988,
-      0.36133044958114624,
-      0.3357429802417755,
-      0.3121393620967865,
-      0.3670870363712311,
-      0.2978810667991638,
-      0.3120233118534088,
-      0.34596773982048035,
-      0.29005879163742065,
-      0.34079042077064514,
-      0.29599788784980774,
-      0.30136188864707947,
-      0.2756097614765167,
-      0.3011065125465393,
-      0.30379655957221985,
-      0.2766466736793518,
-      0.2690426707267761,
-      0.3143273890018463,
-      0.26247408986091614,
-      0.27483493089675903,
-      0.2900548279285431,
-      0.2740040123462677,
-      0.2515677511692047,
-      0.2737451195716858,
-      0.246206596493721,
-      0.2460457980632782,
-      0.2329309731721878,
-      0.25465384125709534,
-      0.2698599696159363,
-      0.2485286146402359,
-      0.24830453097820282,
-      0.2897026538848877
+      7.394570350646973,
+      4.051821708679199,
+      2.021190881729126,
+      1.8420897722244263,
+      1.9143022298812866,
+      1.7933646440505981,
+      1.7442551851272583,
+      1.6997934579849243,
+      1.6858255863189697,
+      1.603048324584961,
+      1.1169459819793701,
+      0.9371060729026794,
+      0.7447897791862488,
+      0.6269596815109253,
+      0.6037188768386841,
+      0.5636410713195801,
+      0.5894449949264526,
+      0.5685290694236755,
+      0.5134449601173401,
+      0.49189919233322144,
+      0.4935498833656311,
+      0.45771291851997375,
+      0.4145750105381012,
+      0.409036785364151,
+      0.445596843957901,
+      0.39751124382019043,
+      0.38312414288520813,
+      0.3672215938568115,
+      0.3891475200653076,
+      0.364010751247406,
+      0.34130558371543884,
+      0.3457995057106018,
+      0.3009479343891144,
+      0.33354875445365906,
+      0.31102973222732544,
+      0.2982536852359772,
+      0.3054274022579193,
+      0.28427326679229736,
+      0.266096830368042,
+      0.29840412735939026,
+      0.29664090275764465,
+      0.2837351858615875,
+      0.2988941967487335,
+      0.3007279932498932,
+      0.26320937275886536,
+      0.2581765055656433,
+      0.28365617990493774,
+      0.2829630970954895,
+      0.2507253885269165,
+      0.2723925709724426,
+      0.24912698566913605,
+      0.24788996577262878,
+      0.26047465205192566,
+      0.2179279625415802,
+      0.2571420967578888,
+      0.26797670125961304,
+      0.24822230637073517,
+      0.23186878859996796,
+      0.27942460775375366,
+      0.21844741702079773,
+      0.24789755046367645,
+      0.2550739645957947
     ],
     "base_loss": [
-      7.182804584503174,
-      4.20333194732666,
-      2.0677404403686523,
-      1.829144835472107,
-      1.8410581350326538,
-      1.820699691772461,
-      1.7079013586044312,
-      1.655383825302124,
-      1.6272556781768799,
-      1.56789231300354,
-      1.2935447692871094,
-      0.8198140263557434,
-      0.7896751761436462,
-      0.7217035293579102,
-      0.6449242830276489,
-      0.6155810356140137,
-      0.6273579001426697,
-      0.5691636800765991,
-      0.5195116400718689,
-      0.5188413262367249,
-      0.4571366012096405,
-      0.46536850929260254,
-      0.45766669511795044,
-      0.4275817275047302,
-      0.40487247705459595,
-      0.4464508593082428,
-      0.39323002099990845,
-      0.39669889211654663,
-      0.34110480546951294,
-      0.3717530369758606,
-      0.3865724503993988,
-      0.36133044958114624,
-      0.3357429802417755,
-      0.3121393620967865,
-      0.3670870363712311,
-      0.2978810667991638,
-      0.3120233118534088,
-      0.34596773982048035,
-      0.29005879163742065,
-      0.34079042077064514,
-      0.29599788784980774,
-      0.30136188864707947,
-      0.2756097614765167,
-      0.3011065125465393,
-      0.30379655957221985,
-      0.2766466736793518,
-      0.2690426707267761,
-      0.3143273890018463,
-      0.26247408986091614,
-      0.27483493089675903,
-      0.2900548279285431,
-      0.2740040123462677,
-      0.2515677511692047,
-      0.2737451195716858,
-      0.246206596493721,
-      0.2460457980632782,
-      0.2329309731721878,
-      0.25465384125709534,
-      0.2698599696159363,
-      0.2485286146402359,
-      0.24830453097820282,
-      0.2897026538848877
+      7.394570350646973,
+      4.051821708679199,
+      2.021190881729126,
+      1.8420897722244263,
+      1.9143022298812866,
+      1.7933646440505981,
+      1.7442551851272583,
+      1.6997934579849243,
+      1.6858255863189697,
+      1.603048324584961,
+      1.1169459819793701,
+      0.9371060729026794,
+      0.7447897791862488,
+      0.6269596815109253,
+      0.6037188768386841,
+      0.5636410713195801,
+      0.5894449949264526,
+      0.5685290694236755,
+      0.5134449601173401,
+      0.49189919233322144,
+      0.4935498833656311,
+      0.45771291851997375,
+      0.4145750105381012,
+      0.409036785364151,
+      0.445596843957901,
+      0.39751124382019043,
+      0.38312414288520813,
+      0.3672215938568115,
+      0.3891475200653076,
+      0.364010751247406,
+      0.34130558371543884,
+      0.3457995057106018,
+      0.3009479343891144,
+      0.33354875445365906,
+      0.31102973222732544,
+      0.2982536852359772,
+      0.3054274022579193,
+      0.28427326679229736,
+      0.266096830368042,
+      0.29840412735939026,
+      0.29664090275764465,
+      0.2837351858615875,
+      0.2988941967487335,
+      0.3007279932498932,
+      0.26320937275886536,
+      0.2581765055656433,
+      0.28365617990493774,
+      0.2829630970954895,
+      0.2507253885269165,
+      0.2723925709724426,
+      0.24912698566913605,
+      0.24788996577262878,
+      0.26047465205192566,
+      0.2179279625415802,
+      0.2571420967578888,
+      0.26797670125961304,
+      0.24822230637073517,
+      0.23186878859996796,
+      0.27942460775375366,
+      0.21844741702079773,
+      0.24789755046367645,
+      0.2550739645957947
     ],
     "lr": [
       3.9200000000000004e-05,
@@ -301,595 +301,595 @@
       20
     ],
     "eval_accuracy": [
-      0.0,
-      0.0044444444444444444,
-      0.006666666666666667,
-      0.01888888888888889,
-      0.057777777777777775,
-      0.09444444444444444,
-      0.11888888888888889,
-      0.20222222222222222,
-      0.24555555555555555,
-      0.2611111111111111,
-      0.32666666666666666,
-      0.3011111111111111,
-      0.33555555555555555,
-      0.39222222222222225,
-      0.41888888888888887,
-      0.3888888888888889,
-      0.4111111111111111,
-      0.39666666666666667,
-      0.41,
-      0.42444444444444446
+      0.01368421052631579,
+      0.002105263157894737,
+      0.005263157894736842,
+      0.0431578947368421,
+      0.06315789473684211,
+      0.09473684210526316,
+      0.17578947368421052,
+      0.1968421052631579,
+      0.2831578947368421,
+      0.27052631578947367,
+      0.3105263157894737,
+      0.33789473684210525,
+      0.39473684210526316,
+      0.3368421052631579,
+      0.3831578947368421,
+      0.4073684210526316,
+      0.41789473684210526,
+      0.42736842105263156,
+      0.41789473684210526,
+      0.41578947368421054
     ]
   },
-  "final_accuracy": 0.33625,
+  "final_accuracy": 0.35846153846153844,
   "sft_eval": {
     "config": {
       "ops": "add_sub",
       "K": null,
       "mode": "sft",
       "n_digits": 6,
-      "n_per_split": 50
+      "n_per_split": 100
     },
     "splits": {
       "add_S0": {
-        "full_accuracy": 0.56,
-        "digit_accuracy": 0.92,
-        "n_examples": 50,
+        "full_accuracy": 0.66,
+        "digit_accuracy": 0.94,
+        "n_examples": 100,
         "per_subtask": {
           "SA": {
-            "accuracy": 0.9220338983050848,
-            "count": 295
+            "accuracy": 0.9388429752066115,
+            "count": 605
           },
           "SS": {
-            "accuracy": 0.9090909090909091,
-            "count": 55
+            "accuracy": 0.9473684210526315,
+            "count": 95
           }
         }
       },
       "add_S1": {
-        "full_accuracy": 0.56,
-        "digit_accuracy": 0.9285714285714286,
-        "n_examples": 50,
+        "full_accuracy": 0.53,
+        "digit_accuracy": 0.9185714285714286,
+        "n_examples": 100,
         "per_subtask": {
           "SA": {
-            "accuracy": 0.9365079365079365,
-            "count": 126
+            "accuracy": 0.9509803921568627,
+            "count": 204
           },
           "SC": {
-            "accuracy": 0.9746835443037974,
-            "count": 79
+            "accuracy": 0.9585798816568047,
+            "count": 169
           },
           "SS": {
-            "accuracy": 0.9523809523809523,
-            "count": 21
+            "accuracy": 0.9354838709677419,
+            "count": 31
           },
           "UC": {
-            "accuracy": 0.8870967741935484,
-            "count": 124
+            "accuracy": 0.8716216216216216,
+            "count": 296
           }
         }
       },
       "add_S2": {
-        "full_accuracy": 0.32,
-        "digit_accuracy": 0.8714285714285714,
-        "n_examples": 50,
+        "full_accuracy": 0.28,
+        "digit_accuracy": 0.8585714285714285,
+        "n_examples": 100,
         "per_subtask": {
           "SA": {
-            "accuracy": 0.9733333333333334,
-            "count": 75
+            "accuracy": 0.9447852760736196,
+            "count": 163
           },
           "SC": {
-            "accuracy": 0.8870967741935484,
-            "count": 62
+            "accuracy": 0.8615384615384616,
+            "count": 130
           },
           "SS": {
-            "accuracy": 0.9743589743589743,
-            "count": 39
+            "accuracy": 0.9080459770114943,
+            "count": 87
           },
           "UC": {
-            "accuracy": 0.7387387387387387,
-            "count": 111
+            "accuracy": 0.7241379310344828,
+            "count": 203
           },
           "US": {
-            "accuracy": 0.9047619047619048,
-            "count": 63
+            "accuracy": 0.9316239316239316,
+            "count": 117
           }
         }
       },
       "add_S3": {
-        "full_accuracy": 0.26,
-        "digit_accuracy": 0.7771428571428571,
-        "n_examples": 50,
+        "full_accuracy": 0.23,
+        "digit_accuracy": 0.78,
+        "n_examples": 100,
         "per_subtask": {
           "SA": {
-            "accuracy": 0.9666666666666667,
-            "count": 60
+            "accuracy": 0.9834710743801653,
+            "count": 121
           },
           "SC": {
-            "accuracy": 0.9649122807017544,
-            "count": 57
+            "accuracy": 0.9256198347107438,
+            "count": 121
           },
           "SS": {
-            "accuracy": 0.9473684210526315,
-            "count": 19
+            "accuracy": 0.8979591836734694,
+            "count": 49
           },
           "UC": {
-            "accuracy": 0.6153846153846154,
-            "count": 104
+            "accuracy": 0.6129032258064516,
+            "count": 186
           },
           "US": {
-            "accuracy": 0.7,
-            "count": 110
+            "accuracy": 0.7040358744394619,
+            "count": 223
           }
         }
       },
       "add_S4": {
-        "full_accuracy": 0.28,
-        "digit_accuracy": 0.7,
-        "n_examples": 50,
+        "full_accuracy": 0.23,
+        "digit_accuracy": 0.7242857142857143,
+        "n_examples": 100,
         "per_subtask": {
           "SA": {
-            "accuracy": 1.0,
-            "count": 48
+            "accuracy": 0.9903846153846154,
+            "count": 104
           },
           "SC": {
-            "accuracy": 0.9807692307692307,
-            "count": 52
+            "accuracy": 0.8867924528301887,
+            "count": 106
           },
           "SS": {
-            "accuracy": 1.0,
-            "count": 7
+            "accuracy": 0.9565217391304348,
+            "count": 23
           },
           "UC": {
-            "accuracy": 0.651685393258427,
-            "count": 89
+            "accuracy": 0.69375,
+            "count": 160
           },
           "US": {
-            "accuracy": 0.525974025974026,
-            "count": 154
+            "accuracy": 0.5765472312703583,
+            "count": 307
           }
         }
       },
       "add_S5": {
-        "full_accuracy": 0.26,
-        "digit_accuracy": 0.6142857142857143,
-        "n_examples": 50,
+        "full_accuracy": 0.17,
+        "digit_accuracy": 0.5471428571428572,
+        "n_examples": 100,
         "per_subtask": {
           "SA": {
             "accuracy": 1.0,
-            "count": 50
+            "count": 100
           },
           "SC": {
-            "accuracy": 0.98,
-            "count": 50
+            "accuracy": 0.96,
+            "count": 100
           },
           "UC": {
-            "accuracy": 0.44,
-            "count": 50
+            "accuracy": 0.39,
+            "count": 100
           },
           "US": {
-            "accuracy": 0.47,
-            "count": 200
+            "accuracy": 0.37,
+            "count": 400
           }
         }
       },
       "add_S6": {
-        "full_accuracy": 0.3,
-        "digit_accuracy": 0.4685714285714286,
-        "n_examples": 50,
+        "full_accuracy": 0.39,
+        "digit_accuracy": 0.5628571428571428,
+        "n_examples": 100,
         "per_subtask": {
           "SC": {
             "accuracy": 1.0,
-            "count": 50
+            "count": 100
           },
           "UC": {
-            "accuracy": 0.36,
-            "count": 50
+            "accuracy": 0.5,
+            "count": 100
           },
           "US": {
-            "accuracy": 0.384,
-            "count": 250
+            "accuracy": 0.488,
+            "count": 500
           }
         }
       },
       "add_random": {
-        "full_accuracy": 0.54,
-        "digit_accuracy": 0.9214285714285714,
+        "full_accuracy": 0.595,
+        "digit_accuracy": 0.925,
         "n_examples": 200,
         "per_subtask": {
           "SA": {
-            "accuracy": 0.9675174013921114,
-            "count": 431
+            "accuracy": 0.9574944071588367,
+            "count": 447
           },
           "SC": {
-            "accuracy": 0.9715189873417721,
-            "count": 316
+            "accuracy": 0.946875,
+            "count": 320
           },
           "SS": {
-            "accuracy": 0.9487179487179487,
-            "count": 39
+            "accuracy": 0.9464285714285714,
+            "count": 56
           },
           "UC": {
-            "accuracy": 0.8589285714285714,
-            "count": 560
+            "accuracy": 0.8865784499054821,
+            "count": 529
           },
           "US": {
-            "accuracy": 0.8888888888888888,
-            "count": 54
+            "accuracy": 0.875,
+            "count": 48
           }
         }
       },
       "add_C1": {
-        "full_accuracy": 0.56,
-        "digit_accuracy": 0.9342857142857143,
-        "n_examples": 50,
+        "full_accuracy": 0.73,
+        "digit_accuracy": 0.9514285714285714,
+        "n_examples": 100,
         "per_subtask": {
           "SA": {
-            "accuracy": 0.952,
-            "count": 250
+            "accuracy": 0.968,
+            "count": 500
           },
           "SC": {
-            "accuracy": 1.0,
-            "count": 50
+            "accuracy": 0.97,
+            "count": 100
           },
           "UC": {
-            "accuracy": 0.78,
-            "count": 50
+            "accuracy": 0.85,
+            "count": 100
           }
         }
       },
       "add_C2": {
-        "full_accuracy": 0.44,
-        "digit_accuracy": 0.9028571428571428,
-        "n_examples": 50,
+        "full_accuracy": 0.56,
+        "digit_accuracy": 0.9257142857142857,
+        "n_examples": 100,
         "per_subtask": {
           "SA": {
-            "accuracy": 0.945,
-            "count": 200
+            "accuracy": 0.9675,
+            "count": 400
           },
           "SC": {
-            "accuracy": 0.96,
-            "count": 50
+            "accuracy": 0.97,
+            "count": 100
           },
           "UC": {
-            "accuracy": 0.7590361445783133,
-            "count": 83
+            "accuracy": 0.8012820512820513,
+            "count": 156
           },
           "US": {
-            "accuracy": 0.9411764705882353,
-            "count": 17
+            "accuracy": 0.8863636363636364,
+            "count": 44
           }
         }
       },
       "add_C3": {
-        "full_accuracy": 0.32,
-        "digit_accuracy": 0.8571428571428571,
-        "n_examples": 50,
+        "full_accuracy": 0.28,
+        "digit_accuracy": 0.8442857142857143,
+        "n_examples": 100,
         "per_subtask": {
           "SA": {
-            "accuracy": 0.9466666666666667,
-            "count": 150
+            "accuracy": 0.9666666666666667,
+            "count": 300
           },
           "SC": {
-            "accuracy": 0.94,
-            "count": 50
+            "accuracy": 0.96,
+            "count": 100
           },
           "UC": {
-            "accuracy": 0.76,
-            "count": 100
+            "accuracy": 0.6582914572864321,
+            "count": 199
           },
           "US": {
-            "accuracy": 0.7,
-            "count": 50
+            "accuracy": 0.7326732673267327,
+            "count": 101
           }
         }
       },
       "add_C4": {
         "full_accuracy": 0.34,
-        "digit_accuracy": 0.8371428571428572,
-        "n_examples": 50,
+        "digit_accuracy": 0.8485714285714285,
+        "n_examples": 100,
         "per_subtask": {
           "SA": {
-            "accuracy": 0.99,
-            "count": 100
+            "accuracy": 0.995,
+            "count": 200
           },
           "SC": {
             "accuracy": 0.96,
-            "count": 50
+            "count": 100
           },
           "UC": {
-            "accuracy": 0.7348484848484849,
-            "count": 132
+            "accuracy": 0.7462121212121212,
+            "count": 264
           },
           "US": {
-            "accuracy": 0.7205882352941176,
-            "count": 68
+            "accuracy": 0.75,
+            "count": 136
           }
         }
       },
       "add_C5": {
         "full_accuracy": 0.28,
-        "digit_accuracy": 0.7828571428571428,
-        "n_examples": 50,
+        "digit_accuracy": 0.8271428571428572,
+        "n_examples": 100,
         "per_subtask": {
           "SA": {
             "accuracy": 1.0,
-            "count": 50
+            "count": 100
           },
           "SC": {
-            "accuracy": 1.0,
-            "count": 50
+            "accuracy": 0.99,
+            "count": 100
           },
           "UC": {
-            "accuracy": 0.7123287671232876,
-            "count": 146
+            "accuracy": 0.7709677419354839,
+            "count": 310
           },
           "US": {
-            "accuracy": 0.6730769230769231,
-            "count": 104
+            "accuracy": 0.7421052631578947,
+            "count": 190
           }
         }
       },
       "add_C6": {
-        "full_accuracy": 0.12,
-        "digit_accuracy": 0.7628571428571429,
-        "n_examples": 50,
+        "full_accuracy": 0.26,
+        "digit_accuracy": 0.7914285714285715,
+        "n_examples": 100,
         "per_subtask": {
           "SC": {
             "accuracy": 1.0,
-            "count": 50
+            "count": 100
           },
           "UC": {
-            "accuracy": 0.7037037037037037,
-            "count": 189
+            "accuracy": 0.7702702702702703,
+            "count": 370
           },
           "US": {
-            "accuracy": 0.7567567567567568,
-            "count": 111
+            "accuracy": 0.7347826086956522,
+            "count": 230
           }
         }
       },
       "sub_M0": {
-        "full_accuracy": 0.74,
-        "digit_accuracy": 0.96,
-        "n_examples": 50,
+        "full_accuracy": 0.71,
+        "digit_accuracy": 0.9514285714285714,
+        "n_examples": 100,
         "per_subtask": {
           "MD": {
-            "accuracy": 0.9537953795379538,
-            "count": 303
+            "accuracy": 0.9495934959349593,
+            "count": 615
           },
           "ME": {
-            "accuracy": 1.0,
-            "count": 47
+            "accuracy": 0.9647058823529412,
+            "count": 85
           }
         }
       },
       "sub_M1": {
-        "full_accuracy": 0.44,
-        "digit_accuracy": 0.8971428571428571,
-        "n_examples": 50,
+        "full_accuracy": 0.5,
+        "digit_accuracy": 0.9171428571428571,
+        "n_examples": 100,
         "per_subtask": {
           "MD": {
-            "accuracy": 0.9716312056737588,
-            "count": 141
+            "accuracy": 0.952054794520548,
+            "count": 292
           },
           "MB": {
-            "accuracy": 0.9305555555555556,
-            "count": 72
+            "accuracy": 0.9722222222222222,
+            "count": 144
           },
           "ME": {
-            "accuracy": 0.9444444444444444,
-            "count": 18
+            "accuracy": 0.96,
+            "count": 25
           },
           "UB": {
-            "accuracy": 0.7815126050420168,
-            "count": 119
+            "accuracy": 0.8368200836820083,
+            "count": 239
           }
         }
       },
       "sub_M2": {
-        "full_accuracy": 0.24,
-        "digit_accuracy": 0.8228571428571428,
-        "n_examples": 50,
+        "full_accuracy": 0.26,
+        "digit_accuracy": 0.85,
+        "n_examples": 100,
         "per_subtask": {
           "MD": {
-            "accuracy": 0.9196428571428571,
-            "count": 112
+            "accuracy": 0.966824644549763,
+            "count": 211
           },
           "MB": {
-            "accuracy": 0.9056603773584906,
-            "count": 53
+            "accuracy": 0.9652173913043478,
+            "count": 115
           },
           "ME": {
-            "accuracy": 0.9574468085106383,
-            "count": 47
+            "accuracy": 0.9529411764705882,
+            "count": 85
           },
           "UB": {
-            "accuracy": 0.5764705882352941,
-            "count": 85
+            "accuracy": 0.6187845303867403,
+            "count": 181
           },
           "UD": {
-            "accuracy": 0.8113207547169812,
-            "count": 53
+            "accuracy": 0.8055555555555556,
+            "count": 108
           }
         }
       },
       "sub_M3": {
-        "full_accuracy": 0.1,
-        "digit_accuracy": 0.7628571428571429,
-        "n_examples": 50,
+        "full_accuracy": 0.14,
+        "digit_accuracy": 0.7714285714285715,
+        "n_examples": 100,
         "per_subtask": {
           "MD": {
-            "accuracy": 0.979381443298969,
-            "count": 97
+            "accuracy": 0.9888268156424581,
+            "count": 179
           },
           "MB": {
-            "accuracy": 0.9411764705882353,
-            "count": 51
+            "accuracy": 0.9320388349514563,
+            "count": 103
           },
           "ME": {
             "accuracy": 1.0,
-            "count": 27
+            "count": 56
           },
           "UB": {
-            "accuracy": 0.581081081081081,
-            "count": 74
+            "accuracy": 0.5436241610738255,
+            "count": 149
           },
           "UD": {
-            "accuracy": 0.5346534653465347,
-            "count": 101
+            "accuracy": 0.6103286384976526,
+            "count": 213
           }
         }
       },
       "sub_M4": {
-        "full_accuracy": 0.04,
-        "digit_accuracy": 0.6,
-        "n_examples": 50,
+        "full_accuracy": 0.09,
+        "digit_accuracy": 0.6271428571428571,
+        "n_examples": 100,
         "per_subtask": {
           "MD": {
-            "accuracy": 0.97,
-            "count": 100
+            "accuracy": 0.995,
+            "count": 200
           },
           "MB": {
-            "accuracy": 1.0,
-            "count": 50
+            "accuracy": 0.98,
+            "count": 100
           },
           "UB": {
-            "accuracy": 0.46,
-            "count": 50
+            "accuracy": 0.44,
+            "count": 100
           },
           "UD": {
-            "accuracy": 0.26666666666666666,
-            "count": 150
+            "accuracy": 0.32666666666666666,
+            "count": 300
           }
         }
       },
       "sub_M5": {
-        "full_accuracy": 0.08,
-        "digit_accuracy": 0.5314285714285715,
-        "n_examples": 50,
+        "full_accuracy": 0.04,
+        "digit_accuracy": 0.48714285714285716,
+        "n_examples": 100,
         "per_subtask": {
           "MD": {
             "accuracy": 1.0,
-            "count": 50
+            "count": 100
           },
           "MB": {
-            "accuracy": 0.98,
-            "count": 50
+            "accuracy": 1.0,
+            "count": 100
           },
           "UB": {
-            "accuracy": 0.46,
-            "count": 50
+            "accuracy": 0.48,
+            "count": 100
           },
           "UD": {
-            "accuracy": 0.32,
-            "count": 200
+            "accuracy": 0.2325,
+            "count": 400
           }
         }
       },
       "sub_random": {
-        "full_accuracy": 0.515,
-        "digit_accuracy": 0.9107142857142857,
+        "full_accuracy": 0.505,
+        "digit_accuracy": 0.9121428571428571,
         "n_examples": 200,
         "per_subtask": {
           "MD": {
-            "accuracy": 0.9719298245614035,
-            "count": 570
+            "accuracy": 0.9466666666666667,
+            "count": 600
           },
           "MB": {
-            "accuracy": 0.9422382671480144,
-            "count": 277
+            "accuracy": 0.9438202247191011,
+            "count": 267
           },
           "ME": {
             "accuracy": 0.9433962264150944,
             "count": 53
           },
           "UB": {
-            "accuracy": 0.8131634819532909,
-            "count": 471
+            "accuracy": 0.8473804100227791,
+            "count": 439
           },
           "UD": {
-            "accuracy": 0.9310344827586207,
-            "count": 29
+            "accuracy": 0.8536585365853658,
+            "count": 41
           }
         }
       },
       "sub_B3": {
-        "full_accuracy": 0.26,
-        "digit_accuracy": 0.8142857142857143,
-        "n_examples": 50,
+        "full_accuracy": 0.23,
+        "digit_accuracy": 0.8114285714285714,
+        "n_examples": 100,
         "per_subtask": {
           "MD": {
-            "accuracy": 0.96,
-            "count": 150
+            "accuracy": 0.9733333333333334,
+            "count": 300
           },
           "MB": {
-            "accuracy": 1.0,
-            "count": 50
+            "accuracy": 0.95,
+            "count": 100
           },
           "UB": {
-            "accuracy": 0.6435643564356436,
-            "count": 101
+            "accuracy": 0.6192893401015228,
+            "count": 197
           },
           "UD": {
-            "accuracy": 0.5306122448979592,
-            "count": 49
+            "accuracy": 0.5728155339805825,
+            "count": 103
           }
         }
       },
       "sub_B4": {
-        "full_accuracy": 0.22,
-        "digit_accuracy": 0.7514285714285714,
-        "n_examples": 50,
+        "full_accuracy": 0.08,
+        "digit_accuracy": 0.7328571428571429,
+        "n_examples": 100,
         "per_subtask": {
           "MD": {
-            "accuracy": 0.98,
-            "count": 100
+            "accuracy": 0.975,
+            "count": 200
           },
           "MB": {
-            "accuracy": 0.98,
-            "count": 50
+            "accuracy": 0.99,
+            "count": 100
           },
           "UB": {
-            "accuracy": 0.6611570247933884,
-            "count": 121
+            "accuracy": 0.5506072874493927,
+            "count": 247
           },
           "UD": {
-            "accuracy": 0.45569620253164556,
-            "count": 79
+            "accuracy": 0.5424836601307189,
+            "count": 153
           }
         }
       },
       "sub_B5": {
-        "full_accuracy": 0.18,
-        "digit_accuracy": 0.6971428571428572,
-        "n_examples": 50,
+        "full_accuracy": 0.13,
+        "digit_accuracy": 0.6985714285714286,
+        "n_examples": 100,
         "per_subtask": {
           "MD": {
             "accuracy": 1.0,
-            "count": 50
+            "count": 100
           },
           "MB": {
             "accuracy": 1.0,
-            "count": 50
+            "count": 100
           },
           "UB": {
-            "accuracy": 0.6052631578947368,
-            "count": 152
+            "accuracy": 0.6308724832214765,
+            "count": 298
           },
           "UD": {
-            "accuracy": 0.5306122448979592,
-            "count": 98
+            "accuracy": 0.5,
+            "count": 202
           }
         }
       }
     },
     "summary": {
-      "overall_accuracy": 0.37066666666666664,
-      "digit_accuracy": 0.8174285714285714,
-      "total_examples": 1500,
+      "overall_accuracy": 0.35846153846153844,
+      "digit_accuracy": 0.8091758241758241,
+      "total_examples": 2600,
       "n_splits": 24
     }
   }

add_sub_baseline_10K_1L3H510d/model.safetensors

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:aeb08c950f588dd85bf054c11d3921a45bb77e8a8be21e8944ddd67270255396
+oid sha256:b5bbfcb52eff2217ebf7270baf0f88acdff3defcd30ae5cc7aa22315d0713c18
 size 634642298

add_sub_baseline_10K_1L3H510d/train_config.json

@@ -69,16 +69,20 @@
   "no_wandb": false,
   "n_params": 158584246,
   "run_name": "add_sub_baseline_10K_1L3H510d",
-  "git_commit": "17e935f460a7f9595b705c1d614101a6b0e520f7",
-  "timestamp": "2026-04-14T05:05:40.288836+00:00",
+  "git_commit": "1d5a160e16a5070d61b881494e832aa88149b15c",
+  "timestamp": "2026-04-15T06:51:51.171053+00:00",
   "tokenizer": "Qwen/Qwen3-0.6B",
   "dataset_repo": "thoughtworks/arithmetic-sorl-data",
   "dataset_config": "add_sub_6digit",
+  "train_dataset": "fixed_train/train_10K_seed42.pt",
   "model_repo": "thoughtworks/arithmetic-sorl",
   "trainer_version": "sft",
-  "wandb_run_id": "nneomu0d",
-  "wandb_url": "https://wandb.ai/nlp_and_interpretability/sorl-arithmetic/runs/nneomu0d",
-  "final_accuracy": 0.33625,
-  "sft_accuracy": 0.33625,
+  "wandb_run_id": "ncfggziw",
+  "wandb_url": "https://wandb.ai/nlp_and_interpretability/sorl-arithmetic/runs/ncfggziw",
+  "eval_final_dataset": "eval_sets/eval_add_sub_6d_N100_seed42.json",
+  "eval_epoch_dataset": "eval_sets/eval_add_sub_6d_N25_seed42.json",
+  "eval_hf_repo": "thoughtworks/arithmetic-sorl-data",
+  "final_accuracy": 0.35846153846153844,
+  "sft_accuracy": 0.35846153846153844,
   "eval_method": "ArithmeticEvaluator"
 }