Upload add_sub_baseline_10K_1L2H256d

add_sub_baseline_10K_1L2H256d/metrics.json

@@ -65,132 +65,132 @@
       3100
     ],
     "loss": [
-      11.231005668640137,
-      9.655884742736816,
-      9.104948997497559,
-      8.300870895385742,
-      7.815948486328125,
-      7.295914649963379,
-      6.637112140655518,
-      6.088160037994385,
-      5.606553554534912,
-      5.160516262054443,
-      4.693508625030518,
-      4.079601287841797,
-      3.7621495723724365,
-      3.3034651279449463,
-      2.915548324584961,
-      2.6438701152801514,
-      2.4529876708984375,
-      2.2501683235168457,
-      2.161332368850708,
-      2.0686194896698,
-      1.9981437921524048,
-      2.0475308895111084,
-      1.9610540866851807,
-      1.9816741943359375,
-      1.940521001815796,
-      1.947417974472046,
-      1.9467179775238037,
-      1.921958327293396,
-      1.8819931745529175,
-      1.8775389194488525,
-      1.8941129446029663,
-      1.8416680097579956,
-      1.8777066469192505,
-      1.8092600107192993,
-      1.8147611618041992,
-      1.8798166513442993,
-      1.7604089975357056,
-      1.8312472105026245,
-      1.8071128129959106,
-      1.80684494972229,
-      1.8678429126739502,
-      1.82756507396698,
-      1.7287473678588867,
-      1.8508025407791138,
-      1.8142287731170654,
-      1.7641321420669556,
-      1.8310877084732056,
-      1.7848055362701416,
-      1.7922300100326538,
-      1.8264408111572266,
-      1.8544548749923706,
-      1.8431966304779053,
-      1.841490626335144,
-      1.77731192111969,
-      1.8109773397445679,
-      1.8000199794769287,
-      1.801360845565796,
-      1.8505899906158447,
-      1.8166601657867432,
-      1.7919466495513916,
-      1.7338820695877075,
-      1.8010389804840088
+      11.318906784057617,
+      9.830679893493652,
+      8.917726516723633,
+      8.428927421569824,
+      7.825829029083252,
+      7.2453155517578125,
+      6.758902549743652,
+      6.228304386138916,
+      5.600834369659424,
+      5.050490379333496,
+      4.656107425689697,
+      4.113125801086426,
+      3.6875858306884766,
+      3.315659999847412,
+      2.9331486225128174,
+      2.6042838096618652,
+      2.5160679817199707,
+      2.234703779220581,
+      2.1570537090301514,
+      2.05960750579834,
+      2.041534662246704,
+      2.013371706008911,
+      1.9541598558425903,
+      1.84817636013031,
+      1.91841721534729,
+      1.9015415906906128,
+      1.9575022459030151,
+      1.8885481357574463,
+      1.8128883838653564,
+      1.9127439260482788,
+      1.9104502201080322,
+      1.8516839742660522,
+      1.7954167127609253,
+      1.8567707538604736,
+      1.8164469003677368,
+      1.8527641296386719,
+      1.84950852394104,
+      1.8511903285980225,
+      1.813899040222168,
+      1.8822053670883179,
+      1.790955662727356,
+      1.810065507888794,
+      1.860735535621643,
+      1.8452800512313843,
+      1.7577123641967773,
+      1.8937323093414307,
+      1.8409382104873657,
+      1.7796363830566406,
+      1.8138635158538818,
+      1.7971341609954834,
+      1.7884553670883179,
+      1.8205574750900269,
+      1.8187378644943237,
+      1.8169699907302856,
+      1.8283138275146484,
+      1.8101422786712646,
+      1.8362064361572266,
+      1.8084076642990112,
+      1.8131023645401,
+      1.784583330154419,
+      1.8322092294692993,
+      1.7847820520401
     ],
     "base_loss": [
-      11.231005668640137,
-      9.655884742736816,
-      9.104948997497559,
-      8.300870895385742,
-      7.815948486328125,
-      7.295914649963379,
-      6.637112140655518,
-      6.088160037994385,
-      5.606553554534912,
-      5.160516262054443,
-      4.693508625030518,
-      4.079601287841797,
-      3.7621495723724365,
-      3.3034651279449463,
-      2.915548324584961,
-      2.6438701152801514,
-      2.4529876708984375,
-      2.2501683235168457,
-      2.161332368850708,
-      2.0686194896698,
-      1.9981437921524048,
-      2.0475308895111084,
-      1.9610540866851807,
-      1.9816741943359375,
-      1.940521001815796,
-      1.947417974472046,
-      1.9467179775238037,
-      1.921958327293396,
-      1.8819931745529175,
-      1.8775389194488525,
-      1.8941129446029663,
-      1.8416680097579956,
-      1.8777066469192505,
-      1.8092600107192993,
-      1.8147611618041992,
-      1.8798166513442993,
-      1.7604089975357056,
-      1.8312472105026245,
-      1.8071128129959106,
-      1.80684494972229,
-      1.8678429126739502,
-      1.82756507396698,
-      1.7287473678588867,
-      1.8508025407791138,
-      1.8142287731170654,
-      1.7641321420669556,
-      1.8310877084732056,
-      1.7848055362701416,
-      1.7922300100326538,
-      1.8264408111572266,
-      1.8544548749923706,
-      1.8431966304779053,
-      1.841490626335144,
-      1.77731192111969,
-      1.8109773397445679,
-      1.8000199794769287,
-      1.801360845565796,
-      1.8505899906158447,
-      1.8166601657867432,
-      1.7919466495513916,
-      1.7338820695877075,
-      1.8010389804840088
+      11.318906784057617,
+      9.830679893493652,
+      8.917726516723633,
+      8.428927421569824,
+      7.825829029083252,
+      7.2453155517578125,
+      6.758902549743652,
+      6.228304386138916,
+      5.600834369659424,
+      5.050490379333496,
+      4.656107425689697,
+      4.113125801086426,
+      3.6875858306884766,
+      3.315659999847412,
+      2.9331486225128174,
+      2.6042838096618652,
+      2.5160679817199707,
+      2.234703779220581,
+      2.1570537090301514,
+      2.05960750579834,
+      2.041534662246704,
+      2.013371706008911,
+      1.9541598558425903,
+      1.84817636013031,
+      1.91841721534729,
+      1.9015415906906128,
+      1.9575022459030151,
+      1.8885481357574463,
+      1.8128883838653564,
+      1.9127439260482788,
+      1.9104502201080322,
+      1.8516839742660522,
+      1.7954167127609253,
+      1.8567707538604736,
+      1.8164469003677368,
+      1.8527641296386719,
+      1.84950852394104,
+      1.8511903285980225,
+      1.813899040222168,
+      1.8822053670883179,
+      1.790955662727356,
+      1.810065507888794,
+      1.860735535621643,
+      1.8452800512313843,
+      1.7577123641967773,
+      1.8937323093414307,
+      1.8409382104873657,
+      1.7796363830566406,
+      1.8138635158538818,
+      1.7971341609954834,
+      1.7884553670883179,
+      1.8205574750900269,
+      1.8187378644943237,
+      1.8169699907302856,
+      1.8283138275146484,
+      1.8101422786712646,
+      1.8362064361572266,
+      1.8084076642990112,
+      1.8131023645401,
+      1.784583330154419,
+      1.8322092294692993,
+      1.7847820520401
     ],
     "lr": [
       9.800000000000001e-06,
@@ -302,490 +302,490 @@
     ],
     "eval_accuracy": [
       0.0,
-      0.0011111111111111111,
-      0.0011111111111111111,
       0.0,
-      0.0011111111111111111,
-      0.005555555555555556,
-      0.01,
-      0.0033333333333333335,
-      0.0044444444444444444,
-      0.0011111111111111111,
-      0.0044444444444444444,
-      0.008888888888888889,
-      0.011111111111111112,
-      0.0011111111111111111,
-      0.005555555555555556,
-      0.0022222222222222222,
-      0.006666666666666667,
-      0.0022222222222222222,
-      0.0022222222222222222,
-      0.0033333333333333335
+      0.002105263157894737,
+      0.0,
+      0.0010526315789473684,
+      0.003157894736842105,
+      0.005263157894736842,
+      0.0010526315789473684,
+      0.003157894736842105,
+      0.00631578947368421,
+      0.005263157894736842,
+      0.00631578947368421,
+      0.003157894736842105,
+      0.007368421052631579,
+      0.0010526315789473684,
+      0.00631578947368421,
+      0.00631578947368421,
+      0.00631578947368421,
+      0.007368421052631579,
+      0.007368421052631579
     ]
   },
-  "final_accuracy": 0.0016666666666666668,
+  "final_accuracy": 0.0019230769230769232,
   "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": 0.0,
-        "digit_accuracy": 0.3457142857142857,
-        "n_examples": 50,
+        "digit_accuracy": 0.3142857142857143,
+        "n_examples": 100,
         "per_subtask": {
           "SA": {
-            "accuracy": 0.22372881355932203,
-            "count": 295
+            "accuracy": 0.2066115702479339,
+            "count": 605
           },
           "SS": {
             "accuracy": 1.0,
-            "count": 55
+            "count": 95
           }
         }
       },
       "add_S1": {
         "full_accuracy": 0.0,
-        "digit_accuracy": 0.24571428571428572,
-        "n_examples": 50,
+        "digit_accuracy": 0.2571428571428571,
+        "n_examples": 100,
         "per_subtask": {
           "SA": {
-            "accuracy": 0.29365079365079366,
-            "count": 126
+            "accuracy": 0.2647058823529412,
+            "count": 204
           },
           "SC": {
-            "accuracy": 0.08860759493670886,
-            "count": 79
+            "accuracy": 0.15976331360946747,
+            "count": 169
           },
           "SS": {
-            "accuracy": 0.9047619047619048,
-            "count": 21
+            "accuracy": 0.7419354838709677,
+            "count": 31
           },
           "UC": {
-            "accuracy": 0.18548387096774194,
-            "count": 124
+            "accuracy": 0.25675675675675674,
+            "count": 296
           }
         }
       },
       "add_S2": {
         "full_accuracy": 0.0,
-        "digit_accuracy": 0.37142857142857144,
-        "n_examples": 50,
+        "digit_accuracy": 0.36428571428571427,
+        "n_examples": 100,
         "per_subtask": {
           "SA": {
-            "accuracy": 0.4,
-            "count": 75
+            "accuracy": 0.38650306748466257,
+            "count": 163
           },
           "SC": {
-            "accuracy": 0.12903225806451613,
-            "count": 62
+            "accuracy": 0.13076923076923078,
+            "count": 130
           },
           "SS": {
-            "accuracy": 0.5128205128205128,
-            "count": 39
+            "accuracy": 0.4827586206896552,
+            "count": 87
           },
           "UC": {
-            "accuracy": 0.36936936936936937,
-            "count": 111
+            "accuracy": 0.3694581280788177,
+            "count": 203
           },
           "US": {
-            "accuracy": 0.49206349206349204,
-            "count": 63
+            "accuracy": 0.49572649572649574,
+            "count": 117
           }
         }
       },
       "add_S3": {
         "full_accuracy": 0.0,
-        "digit_accuracy": 0.37142857142857144,
-        "n_examples": 50,
+        "digit_accuracy": 0.3485714285714286,
+        "n_examples": 100,
         "per_subtask": {
           "SA": {
-            "accuracy": 0.5,
-            "count": 60
+            "accuracy": 0.4793388429752066,
+            "count": 121
           },
           "SC": {
-            "accuracy": 0.10526315789473684,
-            "count": 57
+            "accuracy": 0.049586776859504134,
+            "count": 121
           },
           "SS": {
-            "accuracy": 0.6842105263157895,
-            "count": 19
+            "accuracy": 0.7142857142857143,
+            "count": 49
           },
           "UC": {
-            "accuracy": 0.3557692307692308,
-            "count": 104
+            "accuracy": 0.3387096774193548,
+            "count": 186
           },
           "US": {
-            "accuracy": 0.4,
-            "count": 110
+            "accuracy": 0.36771300448430494,
+            "count": 223
           }
         }
       },
       "add_S4": {
         "full_accuracy": 0.0,
-        "digit_accuracy": 0.32,
-        "n_examples": 50,
+        "digit_accuracy": 0.35714285714285715,
+        "n_examples": 100,
         "per_subtask": {
           "SA": {
-            "accuracy": 0.4791666666666667,
-            "count": 48
+            "accuracy": 0.4519230769230769,
+            "count": 104
           },
           "SC": {
-            "accuracy": 0.057692307692307696,
-            "count": 52
+            "accuracy": 0.07547169811320754,
+            "count": 106
           },
           "SS": {
-            "accuracy": 0.8571428571428571,
-            "count": 7
+            "accuracy": 0.6521739130434783,
+            "count": 23
           },
           "UC": {
-            "accuracy": 0.3146067415730337,
-            "count": 89
+            "accuracy": 0.39375,
+            "count": 160
           },
           "US": {
-            "accuracy": 0.33766233766233766,
-            "count": 154
+            "accuracy": 0.3811074918566775,
+            "count": 307
           }
         }
       },
       "add_S5": {
         "full_accuracy": 0.0,
-        "digit_accuracy": 0.07714285714285714,
-        "n_examples": 50,
+        "digit_accuracy": 0.11285714285714285,
+        "n_examples": 100,
         "per_subtask": {
           "SA": {
-            "accuracy": 0.5,
-            "count": 50
+            "accuracy": 0.46,
+            "count": 100
           },
           "SC": {
-            "accuracy": 0.0,
-            "count": 50
+            "accuracy": 0.02,
+            "count": 100
           },
           "UC": {
-            "accuracy": 0.04,
-            "count": 50
+            "accuracy": 0.14,
+            "count": 100
           },
           "US": {
-            "accuracy": 0.0,
-            "count": 200
+            "accuracy": 0.0425,
+            "count": 400
           }
         }
       },
       "add_S6": {
-        "full_accuracy": 0.12,
-        "digit_accuracy": 0.3485714285714286,
-        "n_examples": 50,
+        "full_accuracy": 0.05,
+        "digit_accuracy": 0.56,
+        "n_examples": 100,
         "per_subtask": {
           "SC": {
-            "accuracy": 0.12,
-            "count": 50
+            "accuracy": 0.07,
+            "count": 100
           },
           "UC": {
-            "accuracy": 0.52,
-            "count": 50
+            "accuracy": 0.68,
+            "count": 100
           },
           "US": {
-            "accuracy": 0.36,
-            "count": 250
+            "accuracy": 0.634,
+            "count": 500
           }
         }
       },
       "add_random": {
         "full_accuracy": 0.0,
-        "digit_accuracy": 0.24357142857142858,
+        "digit_accuracy": 0.2507142857142857,
         "n_examples": 200,
         "per_subtask": {
           "SA": {
-            "accuracy": 0.2505800464037123,
-            "count": 431
+            "accuracy": 0.2684563758389262,
+            "count": 447
           },
           "SC": {
-            "accuracy": 0.10126582278481013,
-            "count": 316
+            "accuracy": 0.096875,
+            "count": 320
           },
           "SS": {
-            "accuracy": 0.8974358974358975,
-            "count": 39
+            "accuracy": 0.7142857142857143,
+            "count": 56
           },
           "UC": {
-            "accuracy": 0.2571428571428571,
-            "count": 560
+            "accuracy": 0.2646502835538752,
+            "count": 529
           },
           "US": {
-            "accuracy": 0.4074074074074074,
-            "count": 54
+            "accuracy": 0.4166666666666667,
+            "count": 48
           }
         }
       },
       "add_C1": {
         "full_accuracy": 0.0,
-        "digit_accuracy": 0.15714285714285714,
-        "n_examples": 50,
+        "digit_accuracy": 0.15142857142857144,
+        "n_examples": 100,
         "per_subtask": {
           "SA": {
-            "accuracy": 0.184,
-            "count": 250
+            "accuracy": 0.168,
+            "count": 500
           },
           "SC": {
-            "accuracy": 0.06,
-            "count": 50
+            "accuracy": 0.04,
+            "count": 100
           },
           "UC": {
-            "accuracy": 0.12,
-            "count": 50
+            "accuracy": 0.18,
+            "count": 100
           }
         }
       },
       "add_C2": {
         "full_accuracy": 0.0,
-        "digit_accuracy": 0.16285714285714287,
-        "n_examples": 50,
+        "digit_accuracy": 0.18142857142857144,
+        "n_examples": 100,
         "per_subtask": {
           "SA": {
-            "accuracy": 0.21,
-            "count": 200
+            "accuracy": 0.23,
+            "count": 400
           },
           "SC": {
             "accuracy": 0.02,
-            "count": 50
+            "count": 100
           },
           "UC": {
-            "accuracy": 0.13253012048192772,
-            "count": 83
+            "accuracy": 0.14743589743589744,
+            "count": 156
           },
           "US": {
-            "accuracy": 0.17647058823529413,
-            "count": 17
+            "accuracy": 0.22727272727272727,
+            "count": 44
           }
         }
       },
       "add_C3": {
         "full_accuracy": 0.0,
-        "digit_accuracy": 0.16857142857142857,
-        "n_examples": 50,
+        "digit_accuracy": 0.15571428571428572,
+        "n_examples": 100,
         "per_subtask": {
           "SA": {
-            "accuracy": 0.26,
-            "count": 150
+            "accuracy": 0.25333333333333335,
+            "count": 300
           },
           "SC": {
-            "accuracy": 0.04,
-            "count": 50
+            "accuracy": 0.02,
+            "count": 100
           },
           "UC": {
-            "accuracy": 0.13,
-            "count": 100
+            "accuracy": 0.12060301507537688,
+            "count": 199
           },
           "US": {
-            "accuracy": 0.1,
-            "count": 50
+            "accuracy": 0.06930693069306931,
+            "count": 101
           }
         }
       },
       "add_C4": {
         "full_accuracy": 0.0,
-        "digit_accuracy": 0.18857142857142858,
-        "n_examples": 50,
+        "digit_accuracy": 0.19,
+        "n_examples": 100,
         "per_subtask": {
           "SA": {
-            "accuracy": 0.39,
-            "count": 100
+            "accuracy": 0.36,
+            "count": 200
           },
           "SC": {
-            "accuracy": 0.0,
-            "count": 50
+            "accuracy": 0.05,
+            "count": 100
           },
           "UC": {
-            "accuracy": 0.09090909090909091,
-            "count": 132
+            "accuracy": 0.12121212121212122,
+            "count": 264
           },
           "US": {
-            "accuracy": 0.22058823529411764,
-            "count": 68
+            "accuracy": 0.17647058823529413,
+            "count": 136
           }
         }
       },
       "add_C5": {
         "full_accuracy": 0.0,
-        "digit_accuracy": 0.18857142857142858,
-        "n_examples": 50,
+        "digit_accuracy": 0.17714285714285713,
+        "n_examples": 100,
         "per_subtask": {
           "SA": {
-            "accuracy": 0.44,
-            "count": 50
+            "accuracy": 0.53,
+            "count": 100
           },
           "SC": {
-            "accuracy": 0.04,
-            "count": 50
+            "accuracy": 0.01,
+            "count": 100
           },
           "UC": {
-            "accuracy": 0.1506849315068493,
-            "count": 146
+            "accuracy": 0.12580645161290321,
+            "count": 310
           },
           "US": {
-            "accuracy": 0.19230769230769232,
-            "count": 104
+            "accuracy": 0.1631578947368421,
+            "count": 190
           }
         }
       },
       "add_C6": {
         "full_accuracy": 0.0,
-        "digit_accuracy": 0.33714285714285713,
-        "n_examples": 50,
+        "digit_accuracy": 0.37,
+        "n_examples": 100,
         "per_subtask": {
           "SC": {
-            "accuracy": 0.06,
-            "count": 50
+            "accuracy": 0.05,
+            "count": 100
           },
           "UC": {
-            "accuracy": 0.2962962962962963,
-            "count": 189
+            "accuracy": 0.2972972972972973,
+            "count": 370
           },
           "US": {
-            "accuracy": 0.5315315315315315,
-            "count": 111
+            "accuracy": 0.6260869565217392,
+            "count": 230
           }
         }
       },
       "sub_M0": {
         "full_accuracy": 0.0,
-        "digit_accuracy": 0.3057142857142857,
-        "n_examples": 50,
+        "digit_accuracy": 0.29285714285714287,
+        "n_examples": 100,
         "per_subtask": {
           "MD": {
-            "accuracy": 0.19801980198019803,
-            "count": 303
+            "accuracy": 0.1951219512195122,
+            "count": 615
           },
           "ME": {
             "accuracy": 1.0,
-            "count": 47
+            "count": 85
           }
         }
       },
       "sub_M1": {
         "full_accuracy": 0.0,
-        "digit_accuracy": 0.26285714285714284,
-        "n_examples": 50,
+        "digit_accuracy": 0.22428571428571428,
+        "n_examples": 100,
         "per_subtask": {
           "MD": {
-            "accuracy": 0.3900709219858156,
-            "count": 141
+            "accuracy": 0.3698630136986301,
+            "count": 292
           },
           "MB": {
             "accuracy": 0.0,
-            "count": 72
+            "count": 144
           },
           "ME": {
             "accuracy": 1.0,
-            "count": 18
+            "count": 25
           },
           "UB": {
-            "accuracy": 0.15966386554621848,
-            "count": 119
+            "accuracy": 0.100418410041841,
+            "count": 239
           }
         }
       },
       "sub_M2": {
         "full_accuracy": 0.0,
-        "digit_accuracy": 0.38285714285714284,
-        "n_examples": 50,
+        "digit_accuracy": 0.35428571428571426,
+        "n_examples": 100,
         "per_subtask": {
           "MD": {
-            "accuracy": 0.6428571428571429,
-            "count": 112
+            "accuracy": 0.6208530805687204,
+            "count": 211
           },
           "MB": {
             "accuracy": 0.0,
-            "count": 53
+            "count": 115
           },
           "ME": {
             "accuracy": 1.0,
-            "count": 47
+            "count": 85
           },
           "UB": {
-            "accuracy": 0.17647058823529413,
-            "count": 85
+            "accuracy": 0.17679558011049723,
+            "count": 181
           },
           "UD": {
             "accuracy": 0.0,
-            "count": 53
+            "count": 108
           }
         }
       },
       "sub_M3": {
         "full_accuracy": 0.0,
-        "digit_accuracy": 0.28,
-        "n_examples": 50,
+        "digit_accuracy": 0.3,
+        "n_examples": 100,
         "per_subtask": {
           "MD": {
-            "accuracy": 0.6494845360824743,
-            "count": 97
+            "accuracy": 0.7597765363128491,
+            "count": 179
           },
           "MB": {
             "accuracy": 0.0,
-            "count": 51
+            "count": 103
           },
           "ME": {
             "accuracy": 1.0,
-            "count": 27
+            "count": 56
           },
           "UB": {
-            "accuracy": 0.10810810810810811,
-            "count": 74
+            "accuracy": 0.12080536912751678,
+            "count": 149
           },
           "UD": {
             "accuracy": 0.0,
-            "count": 101
+            "count": 213
           }
         }
       },
       "sub_M4": {
         "full_accuracy": 0.0,
-        "digit_accuracy": 0.21142857142857144,
-        "n_examples": 50,
+        "digit_accuracy": 0.18571428571428572,
+        "n_examples": 100,
         "per_subtask": {
           "MD": {
             "accuracy": 0.5,
-            "count": 100
+            "count": 200
           },
           "MB": {
             "accuracy": 0.0,
-            "count": 50
+            "count": 100
           },
           "UB": {
-            "accuracy": 0.48,
-            "count": 50
+            "accuracy": 0.3,
+            "count": 100
           },
           "UD": {
             "accuracy": 0.0,
-            "count": 150
+            "count": 300
           }
         }
       },
       "sub_M5": {
         "full_accuracy": 0.0,
-        "digit_accuracy": 0.18285714285714286,
-        "n_examples": 50,
+        "digit_accuracy": 0.18714285714285714,
+        "n_examples": 100,
         "per_subtask": {
           "MD": {
             "accuracy": 1.0,
-            "count": 50
+            "count": 100
           },
           "MB": {
             "accuracy": 0.0,
-            "count": 50
+            "count": 100
           },
           "UB": {
-            "accuracy": 0.28,
-            "count": 50
+            "accuracy": 0.31,
+            "count": 100
           },
           "UD": {
             "accuracy": 0.0,
-            "count": 200
+            "count": 400
           }
         }
       },
@@ -795,101 +795,101 @@
         "n_examples": 200,
         "per_subtask": {
           "MD": {
-            "accuracy": 0.37719298245614036,
-            "count": 570
+            "accuracy": 0.3616666666666667,
+            "count": 600
           },
           "MB": {
             "accuracy": 0.0,
-            "count": 277
+            "count": 267
           },
           "ME": {
             "accuracy": 1.0,
             "count": 53
           },
           "UB": {
-            "accuracy": 0.11677282377919321,
-            "count": 471
+            "accuracy": 0.12072892938496584,
+            "count": 439
           },
           "UD": {
             "accuracy": 0.0,
-            "count": 29
+            "count": 41
           }
         }
       },
       "sub_B3": {
         "full_accuracy": 0.0,
-        "digit_accuracy": 0.18857142857142858,
-        "n_examples": 50,
+        "digit_accuracy": 0.19285714285714287,
+        "n_examples": 100,
         "per_subtask": {
           "MD": {
             "accuracy": 0.3333333333333333,
-            "count": 150
+            "count": 300
           },
           "MB": {
             "accuracy": 0.0,
-            "count": 50
+            "count": 100
           },
           "UB": {
-            "accuracy": 0.15841584158415842,
-            "count": 101
+            "accuracy": 0.17766497461928935,
+            "count": 197
           },
           "UD": {
             "accuracy": 0.0,
-            "count": 49
+            "count": 103
           }
         }
       },
       "sub_B4": {
         "full_accuracy": 0.0,
-        "digit_accuracy": 0.17714285714285713,
-        "n_examples": 50,
+        "digit_accuracy": 0.19428571428571428,
+        "n_examples": 100,
         "per_subtask": {
           "MD": {
             "accuracy": 0.5,
-            "count": 100
+            "count": 200
           },
           "MB": {
             "accuracy": 0.0,
-            "count": 50
+            "count": 100
           },
           "UB": {
-            "accuracy": 0.09917355371900827,
-            "count": 121
+            "accuracy": 0.145748987854251,
+            "count": 247
           },
           "UD": {
             "accuracy": 0.0,
-            "count": 79
+            "count": 153
           }
         }
       },
       "sub_B5": {
         "full_accuracy": 0.0,
-        "digit_accuracy": 0.18,
-        "n_examples": 50,
+        "digit_accuracy": 0.19,
+        "n_examples": 100,
         "per_subtask": {
           "MD": {
             "accuracy": 1.0,
-            "count": 50
+            "count": 100
           },
           "MB": {
             "accuracy": 0.0,
-            "count": 50
+            "count": 100
           },
           "UB": {
-            "accuracy": 0.08552631578947369,
-            "count": 152
+            "accuracy": 0.11073825503355705,
+            "count": 298
           },
           "UD": {
             "accuracy": 0.0,
-            "count": 98
+            "count": 202
           }
         }
       }
     },
     "summary": {
-      "overall_accuracy": 0.004,
-      "digit_accuracy": 0.24485714285714286,
-      "total_examples": 1500,
+      "overall_accuracy": 0.0019230769230769232,
+      "digit_accuracy": 0.25456043956043956,
+      "total_examples": 2600,
       "n_splits": 24
     }
   }

add_sub_baseline_10K_1L2H256d/model.safetensors

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:53a01202265b8cb5dcaaf9bceb2f9af7c042a0935ffc28b3f5d14570b87d9f0d
+oid sha256:aa05a1cd94e5d85c81af8afec9355ccfeab6c774365e3be321eeb593ab3c4c12
 size 315072674

add_sub_baseline_10K_1L2H256d/train_config.json

@@ -69,16 +69,20 @@
   "no_wandb": false,
   "n_params": 78691840,
   "run_name": "add_sub_baseline_10K_1L2H256d",
-  "git_commit": "17e935f460a7f9595b705c1d614101a6b0e520f7",
-  "timestamp": "2026-04-14T06:22:24.415740+00:00",
+  "git_commit": "1d5a160e16a5070d61b881494e832aa88149b15c",
+  "timestamp": "2026-04-15T05:21:44.700980+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": "2b4iz8fe",
-  "wandb_url": "https://wandb.ai/nlp_and_interpretability/sorl-arithmetic/runs/2b4iz8fe",
-  "final_accuracy": 0.0016666666666666668,
-  "sft_accuracy": 0.0016666666666666668,
+  "wandb_run_id": "s9ep067n",
+  "wandb_url": "https://wandb.ai/nlp_and_interpretability/sorl-arithmetic/runs/s9ep067n",
+  "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": 0.0019230769230769232,
+  "sft_accuracy": 0.0019230769230769232,
   "eval_method": "ArithmeticEvaluator"
 }