Upload add_sub_sorl_v1_abs10_K1_10K

add_sub_sorl_v1_abs10_K1_10K/metrics.json

@@ -30,7 +30,37 @@
       1406,
       1463,
       1513,
-      1563
+      1563,
+      1620,
+      1670,
+      1720,
+      1777,
+      1827,
+      1877,
+      1934,
+      1984,
+      2034,
+      2091,
+      2141,
+      2191,
+      2248,
+      2298,
+      2348,
+      2405,
+      2455,
+      2505,
+      2562,
+      2612,
+      2662,
+      2719,
+      2769,
+      2819,
+      2876,
+      2926,
+      2976,
+      3033,
+      3083,
+      3133
     ],
     "loss": [
       9.340689659118652,
@@ -51,18 +81,48 @@
       -1.9102389812469482,
       -1.880913257598877,
       -1.9313758611679077,
-      -1.818078875541687,
-      -1.5362316370010376,
-      -2.0820319652557373,
-      -1.3230443000793457,
-      -1.1353564262390137,
-      -1.5558422803878784,
-      -1.8902345895767212,
-      -1.0403810739517212,
-      -1.1732807159423828,
-      -0.8931325078010559,
-      -1.0903886556625366,
-      -0.5779882073402405
+      -1.8197805881500244,
+      -1.7383019924163818,
+      -2.4177093505859375,
+      -1.3264975547790527,
+      -1.2404942512512207,
+      -1.5992217063903809,
+      -1.9741289615631104,
+      -1.0616717338562012,
+      -1.5642304420471191,
+      -1.3654898405075073,
+      -1.3151346445083618,
+      -0.7584792971611023,
+      -1.3367366790771484,
+      -1.2703979015350342,
+      -1.0298550128936768,
+      -1.6327556371688843,
+      -1.734535813331604,
+      -1.002307653427124,
+      -0.9018746018409729,
+      -2.051492214202881,
+      -0.6267629861831665,
+      -0.9340898990631104,
+      -0.7012732028961182,
+      -0.9501388669013977,
+      -0.9147610664367676,
+      -0.5015538930892944,
+      -1.2906075716018677,
+      -0.9741727113723755,
+      -0.7373249530792236,
+      -0.8238300085067749,
+      -0.48194777965545654,
+      -0.7015870213508606,
+      -0.4147356152534485,
+      -0.649371325969696,
+      -0.7445440292358398,
+      -0.7792887687683105,
+      -0.3351507782936096,
+      -0.4870011508464813,
+      -0.2853846549987793,
+      -0.4475783109664917,
+      -0.42194119095802307,
+      -0.4129505455493927
     ],
     "base_loss": [
       7.390318870544434,
@@ -83,18 +143,48 @@
       0.5179902911186218,
       0.47875410318374634,
       0.4619639217853546,
-      0.3892979025840759,
-      0.4061528742313385,
-      0.4215702414512634,
-      0.30625566840171814,
-      0.26082319021224976,
-      0.30791884660720825,
-      0.3336397111415863,
-      0.22246377170085907,
-      0.2154165357351303,
-      0.1796569526195526,
-      0.1971123218536377,
-      0.13742247223854065
+      0.39008235931396484,
+      0.43836668133735657,
+      0.48746830224990845,
+      0.2996414601802826,
+      0.26576855778694153,
+      0.3262496590614319,
+      0.33699464797973633,
+      0.24052418768405914,
+      0.2867306172847748,
+      0.23591551184654236,
+      0.233742356300354,
+      0.15799184143543243,
+      0.2535308301448822,
+      0.22049279510974884,
+      0.1870928704738617,
+      0.2405395358800888,
+      0.25070902705192566,
+      0.16667716205120087,
+      0.15809106826782227,
+      0.31137070059776306,
+      0.11368274688720703,
+      0.14004917442798615,
+      0.13582302629947662,
+      0.14286470413208008,
+      0.13543015718460083,
+      0.08498942852020264,
+      0.18833109736442566,
+      0.13422875106334686,
+      0.10775727778673172,
+      0.12300104647874832,
+      0.07167976349592209,
+      0.11587230116128922,
+      0.0645364299416542,
+      0.09650751203298569,
+      0.09376537799835205,
+      0.09722624719142914,
+      0.045721422880887985,
+      0.06196039915084839,
+      0.05605100467801094,
+      0.06696300953626633,
+      0.053617555648088455,
+      0.052670858800411224
     ],
     "info_loss": [
       -0.4859275817871094,
@@ -115,18 +205,48 @@
       -0.3521292805671692,
       -0.33850210905075073,
       -0.33436745405197144,
-      -0.3095327913761139,
-      -0.2765413522720337,
-      -0.32649245858192444,
-      -0.24035842716693878,
-      -0.20812112092971802,
-      -0.2552123963832855,
-      -0.28583821654319763,
-      -0.19351282715797424,
-      -0.19369123876094818,
-      -0.16221696138381958,
-      -0.18230006098747253,
-      -0.1274920552968979
+      -0.3098219931125641,
+      -0.3004074692726135,
+      -0.3650834560394287,
+      -0.2377147525548935,
+      -0.21807271242141724,
+      -0.2605586349964142,
+      -0.2963896691799164,
+      -0.19201940298080444,
+      -0.2403343915939331,
+      -0.2125036120414734,
+      -0.2005012035369873,
+      -0.14176680147647858,
+      -0.20001220703125,
+      -0.18811163306236267,
+      -0.16207079589366913,
+      -0.22838185727596283,
+      -0.23002460598945618,
+      -0.1475900262594223,
+      -0.1343834400177002,
+      -0.26728555560112,
+      -0.10240201652050018,
+      -0.1339702010154724,
+      -0.11256629973649979,
+      -0.13482710719108582,
+      -0.12813962996006012,
+      -0.08094710111618042,
+      -0.1647377759218216,
+      -0.13163283467292786,
+      -0.1023663803935051,
+      -0.11182866990566254,
+      -0.06797388941049576,
+      -0.09571453928947449,
+      -0.06312839686870575,
+      -0.0817064419388771,
+      -0.09321703016757965,
+      -0.09572642296552658,
+      -0.04448504000902176,
+      -0.06147210672497749,
+      -0.040348708629608154,
+      -0.05947120487689972,
+      -0.05293218418955803,
+      -0.05189599096775055
     ],
     "abs_loss": [
       2.2262747287750244,
@@ -147,18 +267,48 @@
       0.6123369336128235,
       0.49044322967529297,
       0.4312553107738495,
-      0.38406622409820557,
-      0.3418884575366974,
-      0.2913786768913269,
-      0.3654420077800751,
-      0.28334739804267883,
-      0.25913718342781067,
-      0.2655659019947052,
-      0.2588687837123871,
-      0.21410413086414337,
-      0.21956706047058105,
-      0.2182735651731491,
-      0.23050324618816376
+      0.38433346152305603,
+      0.3575459420681,
+      0.2791888415813446,
+      0.3738294541835785,
+      0.2938316762447357,
+      0.2500404715538025,
+      0.24418945610523224,
+      0.2506295442581177,
+      0.1918429583311081,
+      0.21762271225452423,
+      0.19322265684604645,
+      0.174521803855896,
+      0.1660517454147339,
+      0.1862434595823288,
+      0.14573685824871063,
+      0.13860176503658295,
+      0.13301999866962433,
+      0.12413140386343002,
+      0.0981183648109436,
+      0.1176028847694397,
+      0.12784387171268463,
+      0.08845017105340958,
+      0.10072199255228043,
+      0.08419985324144363,
+      0.10068891197443008,
+      0.08726515620946884,
+      0.08775246143341064,
+      0.06938570737838745,
+      0.07202962785959244,
+      0.07672454416751862,
+      0.06882075220346451,
+      0.0632840245962143,
+      0.06827760487794876,
+      0.05087992176413536,
+      0.04930369183421135,
+      0.05453367158770561,
+      0.04361484572291374,
+      0.05198704078793526,
+      0.042146146297454834,
+      0.0432199127972126,
+      0.042553484439849854,
+      0.054908428341150284
     ],
     "zipf_loss": [
       6.587018966674805,
@@ -179,18 +329,48 @@
       1.031829595565796,
       0.9763095378875732,
       0.9072092771530151,
-      0.8495444059371948,
-      0.7888401746749878,
-      0.7321844100952148,
-      0.7377402782440186,
-      0.6566966772079468,
-      0.6624492406845093,
-      0.6079515218734741,
-      0.6463965177536011,
-      0.5268046855926514,
-      0.5274234414100647,
-      0.5136723518371582,
-      0.5364596247673035
+      0.8499235510826111,
+      0.7916512489318848,
+      0.7177382707595825,
+      0.7136257290840149,
+      0.6450810432434082,
+      0.655110776424408,
+      0.6283543109893799,
+      0.5929351449012756,
+      0.5331985950469971,
+      0.5018686056137085,
+      0.43681272864341736,
+      0.48374468088150024,
+      0.39324942231178284,
+      0.37160131335258484,
+      0.3891863524913788,
+      0.3966630697250366,
+      0.3016990125179291,
+      0.2945023775100708,
+      0.2740568518638611,
+      0.29823243618011475,
+      0.27079012989997864,
+      0.2567179501056671,
+      0.278494656085968,
+      0.24684756994247437,
+      0.22113612294197083,
+      0.21420112252235413,
+      0.1596638262271881,
+      0.2009882926940918,
+      0.17137852311134338,
+      0.1637832671403885,
+      0.1192292720079422,
+      0.13335762917995453,
+      0.14518415927886963,
+      0.06609760969877243,
+      0.08893045783042908,
+      0.07529586553573608,
+      0.05961669981479645,
+      0.06056078150868416,
+      0.05783679336309433,
+      0.07584869861602783,
+      0.0495077446103096,
+      0.04784762114286423
     ],
     "denoise_loss": [],
     "ortho_loss": [
@@ -212,18 +392,48 @@
       0.3624407649040222,
       0.36967605352401733,
       0.3742343485355377,
-      0.39829763770103455,
-      0.39412182569503784,
-      0.3936292827129364,
-      0.3937382996082306,
-      0.3880367875099182,
-      0.39024752378463745,
-      0.3874780535697937,
-      0.39575129747390747,
-      0.3984874486923218,
-      0.39981427788734436,
-      0.40026330947875977,
-      0.4032067358493805
+      0.3982963562011719,
+      0.3951874375343323,
+      0.3906378149986267,
+      0.3858691453933716,
+      0.390056312084198,
+      0.3822699785232544,
+      0.3838372230529785,
+      0.3822362422943115,
+      0.375870019197464,
+      0.38709157705307007,
+      0.3905230164527893,
+      0.39558085799217224,
+      0.3952634334564209,
+      0.39085277915000916,
+      0.39921504259109497,
+      0.39666083455085754,
+      0.3942829966545105,
+      0.3991093337535858,
+      0.39491263031959534,
+      0.3742454946041107,
+      0.3708445429801941,
+      0.37462660670280457,
+      0.3777347207069397,
+      0.36790767312049866,
+      0.36087462306022644,
+      0.3528926968574524,
+      0.3517568111419678,
+      0.3517394959926605,
+      0.3531302809715271,
+      0.35252636671066284,
+      0.35542812943458557,
+      0.35618269443511963,
+      0.3573307693004608,
+      0.3604697287082672,
+      0.36059707403182983,
+      0.36255255341529846,
+      0.362486869096756,
+      0.3620503544807434,
+      0.363882839679718,
+      0.36408552527427673,
+      0.3636517822742462,
+      0.3643637001514435
     ],
     "lr": [
       3.9200000000000004e-05,
@@ -244,18 +454,48 @@
       8e-05,
       8e-05,
       8e-05,
-      7.889795918367346e-05,
-      7.277551020408164e-05,
-      6.665306122448979e-05,
-      5.96734693877551e-05,
-      5.3551020408163274e-05,
-      4.7428571428571427e-05,
-      4.044897959183674e-05,
-      3.432653061224491e-05,
-      2.820408163265307e-05,
-      2.1224489795918364e-05,
-      1.5102040816326524e-05,
-      8.9795918367347e-06
+      8e-05,
+      8e-05,
+      8e-05,
+      8e-05,
+      8e-05,
+      8e-05,
+      8e-05,
+      8e-05,
+      8e-05,
+      8e-05,
+      8e-05,
+      8e-05,
+      8e-05,
+      8e-05,
+      8e-05,
+      8e-05,
+      8e-05,
+      8e-05,
+      7.946710526315791e-05,
+      7.650657894736843e-05,
+      7.354605263157895e-05,
+      7.017105263157896e-05,
+      6.721052631578948e-05,
+      6.425e-05,
+      6.0875e-05,
+      5.791447368421054e-05,
+      5.495394736842105e-05,
+      5.157894736842105e-05,
+      4.861842105263157e-05,
+      4.565789473684212e-05,
+      4.2282894736842104e-05,
+      3.9322368421052625e-05,
+      3.636184210526315e-05,
+      3.2986842105263165e-05,
+      3.0026315789473686e-05,
+      2.7065789473684206e-05,
+      2.3690789473684223e-05,
+      2.0730263157894743e-05,
+      1.7769736842105264e-05,
+      1.4394736842105275e-05,
+      1.1434210526315796e-05,
+      8.473684210526318e-06
     ],
     "emb_lr": [],
     "eval_step": [
@@ -268,7 +508,17 @@
       1092,
       1199,
       1356,
-      1513
+      1513,
+      1670,
+      1827,
+      1984,
+      2141,
+      2298,
+      2455,
+      2612,
+      2769,
+      2926,
+      3083
     ],
     "eval_accuracy": [
       0.01,
@@ -280,10 +530,20 @@
       0.0,
       0.0,
       0.0,
+      0.0,
+      0.0,
+      0.0,
+      0.0,
+      0.0,
+      0.0,
+      0.0,
+      0.0,
+      0.0,
+      0.0,
       0.0
     ]
   },
-  "final_accuracy": 0.8704166666666666,
+  "final_accuracy": 0.9470833333333334,
   "sft_eval": {
     "config": {
       "ops": "add_sub",
@@ -294,29 +554,29 @@
     },
     "splits": {
       "add_S0": {
-        "full_accuracy": 0.9,
+        "full_accuracy": 0.94,
         "n_examples": 100,
         "per_subtask": {
           "SA": {
-            "accuracy": 0.9818181818181818,
+            "accuracy": 0.9917355371900827,
             "count": 605
           },
           "SS": {
-            "accuracy": 1.0,
+            "accuracy": 0.9894736842105263,
             "count": 95
           }
         }
       },
       "add_S1": {
-        "full_accuracy": 0.83,
+        "full_accuracy": 0.99,
         "n_examples": 100,
         "per_subtask": {
           "SA": {
-            "accuracy": 0.9803921568627451,
+            "accuracy": 0.9950980392156863,
             "count": 204
           },
           "SC": {
-            "accuracy": 0.9940828402366864,
+            "accuracy": 1.0,
             "count": 169
           },
           "SS": {
@@ -324,47 +584,47 @@
             "count": 31
           },
           "UC": {
-            "accuracy": 0.956081081081081,
+            "accuracy": 1.0,
             "count": 296
           }
         }
       },
       "add_S2": {
-        "full_accuracy": 0.54,
+        "full_accuracy": 0.91,
         "n_examples": 100,
         "per_subtask": {
           "SA": {
-            "accuracy": 0.9815950920245399,
+            "accuracy": 1.0,
             "count": 163
           },
           "SC": {
-            "accuracy": 0.9538461538461539,
+            "accuracy": 0.9615384615384616,
             "count": 130
           },
           "SS": {
-            "accuracy": 0.9770114942528736,
+            "accuracy": 0.9885057471264368,
             "count": 87
           },
           "UC": {
-            "accuracy": 0.812807881773399,
+            "accuracy": 0.9852216748768473,
             "count": 203
           },
           "US": {
-            "accuracy": 0.9401709401709402,
+            "accuracy": 1.0,
             "count": 117
           }
         }
       },
       "add_S3": {
-        "full_accuracy": 0.3,
+        "full_accuracy": 0.67,
         "n_examples": 100,
         "per_subtask": {
           "SA": {
-            "accuracy": 0.9752066115702479,
+            "accuracy": 0.9917355371900827,
             "count": 121
           },
           "SC": {
-            "accuracy": 0.9752066115702479,
+            "accuracy": 1.0,
             "count": 121
           },
           "SS": {
@@ -372,25 +632,25 @@
             "count": 49
           },
           "UC": {
-            "accuracy": 0.6666666666666666,
+            "accuracy": 0.8333333333333334,
             "count": 186
           },
           "US": {
-            "accuracy": 0.7668161434977578,
+            "accuracy": 0.9910313901345291,
             "count": 223
           }
         }
       },
       "add_S4": {
-        "full_accuracy": 0.23,
+        "full_accuracy": 0.58,
         "n_examples": 100,
         "per_subtask": {
           "SA": {
-            "accuracy": 0.9807692307692307,
+            "accuracy": 1.0,
             "count": 104
           },
           "SC": {
-            "accuracy": 0.9811320754716981,
+            "accuracy": 1.0,
             "count": 106
           },
           "SS": {
@@ -398,17 +658,17 @@
             "count": 23
           },
           "UC": {
-            "accuracy": 0.58125,
+            "accuracy": 0.7875,
             "count": 160
           },
           "US": {
-            "accuracy": 0.5635179153094463,
+            "accuracy": 0.9218241042345277,
             "count": 307
           }
         }
       },
       "add_S5": {
-        "full_accuracy": 0.12,
+        "full_accuracy": 0.43,
         "n_examples": 100,
         "per_subtask": {
           "SA": {
@@ -416,21 +676,21 @@
             "count": 100
           },
           "SC": {
-            "accuracy": 0.94,
+            "accuracy": 1.0,
             "count": 100
           },
           "UC": {
-            "accuracy": 0.3,
+            "accuracy": 0.5,
             "count": 100
           },
           "US": {
-            "accuracy": 0.325,
+            "accuracy": 0.775,
             "count": 400
           }
         }
       },
       "add_S6": {
-        "full_accuracy": 0.34,
+        "full_accuracy": 0.85,
         "n_examples": 100,
         "per_subtask": {
           "SC": {
@@ -438,43 +698,43 @@
             "count": 100
           },
           "UC": {
-            "accuracy": 0.53,
+            "accuracy": 0.85,
             "count": 100
           },
           "US": {
-            "accuracy": 0.514,
+            "accuracy": 0.91,
             "count": 500
           }
         }
       },
       "add_random": {
-        "full_accuracy": 0.7,
+        "full_accuracy": 0.975,
         "n_examples": 200,
         "per_subtask": {
           "SA": {
-            "accuracy": 0.9686800894854586,
+            "accuracy": 0.9977628635346756,
             "count": 447
           },
           "SC": {
-            "accuracy": 0.971875,
+            "accuracy": 0.996875,
             "count": 320
           },
           "SS": {
-            "accuracy": 0.9642857142857143,
+            "accuracy": 0.9821428571428571,
             "count": 56
           },
           "UC": {
-            "accuracy": 0.9243856332703214,
+            "accuracy": 0.994328922495274,
             "count": 529
           },
           "US": {
-            "accuracy": 0.8958333333333334,
+            "accuracy": 1.0,
             "count": 48
           }
         }
       },
       "add_C3": {
-        "full_accuracy": 0.48,
+        "full_accuracy": 0.84,
         "n_examples": 100,
         "per_subtask": {
           "SA": {
@@ -482,47 +742,47 @@
             "count": 300
           },
           "SC": {
-            "accuracy": 0.99,
+            "accuracy": 1.0,
             "count": 100
           },
           "UC": {
-            "accuracy": 0.7461139896373057,
+            "accuracy": 0.9222797927461139,
             "count": 193
           },
           "US": {
-            "accuracy": 0.7663551401869159,
+            "accuracy": 0.9906542056074766,
             "count": 107
           }
         }
       },
       "add_C4": {
-        "full_accuracy": 0.38,
+        "full_accuracy": 0.83,
         "n_examples": 100,
         "per_subtask": {
           "SA": {
-            "accuracy": 0.985,
+            "accuracy": 1.0,
             "count": 200
           },
           "SC": {
-            "accuracy": 0.98,
+            "accuracy": 1.0,
             "count": 100
           },
           "UC": {
-            "accuracy": 0.7734375,
+            "accuracy": 0.9375,
             "count": 256
           },
           "US": {
-            "accuracy": 0.8194444444444444,
+            "accuracy": 0.9722222222222222,
             "count": 144
           }
         }
       },
       "add_C5": {
-        "full_accuracy": 0.4,
+        "full_accuracy": 0.76,
         "n_examples": 100,
         "per_subtask": {
           "SA": {
-            "accuracy": 0.99,
+            "accuracy": 1.0,
             "count": 100
           },
           "SC": {
@@ -530,17 +790,17 @@
             "count": 100
           },
           "UC": {
-            "accuracy": 0.7908496732026143,
+            "accuracy": 0.9215686274509803,
             "count": 306
           },
           "US": {
-            "accuracy": 0.845360824742268,
+            "accuracy": 0.9329896907216495,
             "count": 194
           }
         }
       },
       "add_C6": {
-        "full_accuracy": 0.35,
+        "full_accuracy": 0.81,
         "n_examples": 100,
         "per_subtask": {
           "SC": {
@@ -548,39 +808,39 @@
             "count": 100
           },
           "UC": {
-            "accuracy": 0.8005464480874317,
+            "accuracy": 0.9590163934426229,
             "count": 366
           },
           "US": {
-            "accuracy": 0.8290598290598291,
+            "accuracy": 0.9786324786324786,
             "count": 234
           }
         }
       },
       "sub_M0": {
-        "full_accuracy": 0.89,
+        "full_accuracy": 0.98,
         "n_examples": 100,
         "per_subtask": {
           "MD": {
-            "accuracy": 0.9833610648918469,
+            "accuracy": 0.9966722129783694,
             "count": 601
           },
           "ME": {
-            "accuracy": 0.9797979797979798,
+            "accuracy": 1.0,
             "count": 99
           }
         }
       },
       "sub_M1": {
-        "full_accuracy": 0.81,
+        "full_accuracy": 0.98,
         "n_examples": 100,
         "per_subtask": {
           "MD": {
-            "accuracy": 0.992831541218638,
+            "accuracy": 0.996415770609319,
             "count": 279
           },
           "MB": {
-            "accuracy": 0.993103448275862,
+            "accuracy": 1.0,
             "count": 145
           },
           "ME": {
@@ -588,39 +848,39 @@
             "count": 24
           },
           "UB": {
-            "accuracy": 0.9285714285714286,
+            "accuracy": 0.996031746031746,
             "count": 252
           }
         }
       },
       "sub_M2": {
-        "full_accuracy": 0.42,
+        "full_accuracy": 0.95,
         "n_examples": 100,
         "per_subtask": {
           "MD": {
-            "accuracy": 0.9859154929577465,
+            "accuracy": 1.0,
             "count": 213
           },
           "MB": {
-            "accuracy": 0.9734513274336283,
+            "accuracy": 1.0,
             "count": 113
           },
           "ME": {
-            "accuracy": 0.9882352941176471,
+            "accuracy": 1.0,
             "count": 85
           },
           "UB": {
-            "accuracy": 0.6740331491712708,
+            "accuracy": 0.9723756906077348,
             "count": 181
           },
           "UD": {
-            "accuracy": 0.9722222222222222,
+            "accuracy": 1.0,
             "count": 108
           }
         }
       },
       "sub_M3": {
-        "full_accuracy": 0.06,
+        "full_accuracy": 0.19,
         "n_examples": 100,
         "per_subtask": {
           "MD": {
@@ -628,7 +888,7 @@
             "count": 179
           },
           "MB": {
-            "accuracy": 0.970873786407767,
+            "accuracy": 1.0,
             "count": 103
           },
           "ME": {
@@ -636,17 +896,17 @@
             "count": 56
           },
           "UB": {
-            "accuracy": 0.3624161073825503,
+            "accuracy": 0.48322147651006714,
             "count": 149
           },
           "UD": {
-            "accuracy": 0.6384976525821596,
+            "accuracy": 0.9530516431924883,
             "count": 213
           }
         }
       },
       "sub_M4": {
-        "full_accuracy": 0.02,
+        "full_accuracy": 0.01,
         "n_examples": 100,
         "per_subtask": {
           "MD": {
@@ -654,15 +914,15 @@
             "count": 200
           },
           "MB": {
-            "accuracy": 0.91,
+            "accuracy": 1.0,
             "count": 100
           },
           "UB": {
-            "accuracy": 0.38,
+            "accuracy": 0.13,
             "count": 100
           },
           "UD": {
-            "accuracy": 0.29,
+            "accuracy": 0.5666666666666667,
             "count": 300
           }
         }
@@ -676,29 +936,29 @@
             "count": 100
           },
           "MB": {
-            "accuracy": 0.85,
+            "accuracy": 1.0,
             "count": 100
           },
           "UB": {
-            "accuracy": 0.25,
+            "accuracy": 0.09,
             "count": 100
           },
           "UD": {
-            "accuracy": 0.19,
+            "accuracy": 0.405,
             "count": 400
           }
         }
       },
       "sub_random": {
-        "full_accuracy": 0.71,
+        "full_accuracy": 0.97,
         "n_examples": 200,
         "per_subtask": {
           "MD": {
-            "accuracy": 0.9816666666666667,
+            "accuracy": 1.0,
             "count": 600
           },
           "MB": {
-            "accuracy": 0.9925093632958801,
+            "accuracy": 1.0,
             "count": 267
           },
           "ME": {
@@ -706,61 +966,61 @@
             "count": 53
           },
           "UB": {
-            "accuracy": 0.8906605922551253,
+            "accuracy": 0.9863325740318907,
             "count": 439
           },
           "UD": {
-            "accuracy": 0.8536585365853658,
+            "accuracy": 1.0,
             "count": 41
           }
         }
       },
       "sub_B3": {
-        "full_accuracy": 0.33,
+        "full_accuracy": 0.74,
         "n_examples": 100,
         "per_subtask": {
           "MD": {
-            "accuracy": 0.9866666666666667,
+            "accuracy": 0.9966666666666667,
             "count": 300
           },
           "MB": {
-            "accuracy": 0.98,
+            "accuracy": 1.0,
             "count": 100
           },
           "UB": {
-            "accuracy": 0.6700507614213198,
+            "accuracy": 0.8730964467005076,
             "count": 197
           },
           "UD": {
-            "accuracy": 0.7378640776699029,
+            "accuracy": 1.0,
             "count": 103
           }
         }
       },
       "sub_B4": {
-        "full_accuracy": 0.21,
+        "full_accuracy": 0.62,
         "n_examples": 100,
         "per_subtask": {
           "MD": {
-            "accuracy": 0.995,
+            "accuracy": 1.0,
             "count": 200
           },
           "MB": {
-            "accuracy": 0.98,
+            "accuracy": 1.0,
             "count": 100
           },
           "UB": {
-            "accuracy": 0.6761133603238867,
+            "accuracy": 0.854251012145749,
             "count": 247
           },
           "UD": {
-            "accuracy": 0.6666666666666666,
+            "accuracy": 0.8888888888888888,
             "count": 153
           }
         }
       },
       "sub_B5": {
-        "full_accuracy": 0.15,
+        "full_accuracy": 0.46,
         "n_examples": 100,
         "per_subtask": {
           "MD": {
@@ -768,22 +1028,22 @@
             "count": 100
           },
           "MB": {
-            "accuracy": 0.97,
+            "accuracy": 1.0,
             "count": 100
           },
           "UB": {
-            "accuracy": 0.6543624161073825,
+            "accuracy": 0.8288590604026845,
             "count": 298
           },
           "UD": {
-            "accuracy": 0.6089108910891089,
+            "accuracy": 0.8811881188118812,
             "count": 202
           }
         }
       }
     },
     "summary": {
-      "overall_accuracy": 0.44083333333333335,
+      "overall_accuracy": 0.7258333333333333,
       "total_examples": 2400,
       "n_splits": 22
     }
@@ -912,7 +1172,7 @@
         }
       },
       "add_S5": {
-        "full_accuracy": 0.58,
+        "full_accuracy": 0.6,
         "n_examples": 100,
         "per_subtask": {
           "SA": {
@@ -924,7 +1184,7 @@
             "count": 100
           },
           "UC": {
-            "accuracy": 0.58,
+            "accuracy": 0.6,
             "count": 100
           },
           "US": {
@@ -934,7 +1194,7 @@
         }
       },
       "add_S6": {
-        "full_accuracy": 0.88,
+        "full_accuracy": 1.0,
         "n_examples": 100,
         "per_subtask": {
           "SC": {
@@ -942,11 +1202,11 @@
             "count": 100
           },
           "UC": {
-            "accuracy": 0.95,
+            "accuracy": 1.0,
             "count": 100
           },
           "US": {
-            "accuracy": 0.972,
+            "accuracy": 1.0,
             "count": 500
           }
         }
@@ -1022,7 +1282,7 @@
         }
       },
       "add_C5": {
-        "full_accuracy": 0.97,
+        "full_accuracy": 0.98,
         "n_examples": 100,
         "per_subtask": {
           "SA": {
@@ -1038,13 +1298,13 @@
             "count": 306
           },
           "US": {
-            "accuracy": 0.9948453608247423,
+            "accuracy": 1.0,
             "count": 194
           }
         }
       },
       "add_C6": {
-        "full_accuracy": 0.98,
+        "full_accuracy": 1.0,
         "n_examples": 100,
         "per_subtask": {
           "SC": {
@@ -1052,7 +1312,7 @@
             "count": 100
           },
           "UC": {
-            "accuracy": 0.994535519125683,
+            "accuracy": 1.0,
             "count": 366
           },
           "US": {
@@ -1062,11 +1322,11 @@
         }
       },
       "sub_M0": {
-        "full_accuracy": 0.99,
+        "full_accuracy": 1.0,
         "n_examples": 100,
         "per_subtask": {
           "MD": {
-            "accuracy": 0.9983361064891847,
+            "accuracy": 1.0,
             "count": 601
           },
           "ME": {
@@ -1098,11 +1358,11 @@
         }
       },
       "sub_M2": {
-        "full_accuracy": 0.95,
+        "full_accuracy": 1.0,
         "n_examples": 100,
         "per_subtask": {
           "MD": {
-            "accuracy": 0.9906103286384976,
+            "accuracy": 1.0,
             "count": 213
           },
           "MB": {
@@ -1114,7 +1374,7 @@
             "count": 85
           },
           "UB": {
-            "accuracy": 0.9834254143646409,
+            "accuracy": 1.0,
             "count": 181
           },
           "UD": {
@@ -1124,7 +1384,7 @@
         }
       },
       "sub_M3": {
-        "full_accuracy": 0.79,
+        "full_accuracy": 1.0,
         "n_examples": 100,
         "per_subtask": {
           "MD": {
@@ -1140,17 +1400,17 @@
             "count": 56
           },
           "UB": {
-            "accuracy": 0.87248322147651,
+            "accuracy": 1.0,
             "count": 149
           },
           "UD": {
-            "accuracy": 0.9624413145539906,
+            "accuracy": 1.0,
             "count": 213
           }
         }
       },
       "sub_M4": {
-        "full_accuracy": 0.3,
+        "full_accuracy": 0.98,
         "n_examples": 100,
         "per_subtask": {
           "MD": {
@@ -1162,17 +1422,17 @@
             "count": 100
           },
           "UB": {
-            "accuracy": 0.36,
+            "accuracy": 1.0,
             "count": 100
           },
           "UD": {
-            "accuracy": 0.8133333333333334,
+            "accuracy": 0.9933333333333333,
             "count": 300
           }
         }
       },
       "sub_M5": {
-        "full_accuracy": 0.04,
+        "full_accuracy": 0.22,
         "n_examples": 100,
         "per_subtask": {
           "MD": {
@@ -1184,21 +1444,21 @@
             "count": 100
           },
           "UB": {
-            "accuracy": 0.12,
+            "accuracy": 0.22,
             "count": 100
           },
           "UD": {
-            "accuracy": 0.6575,
+            "accuracy": 0.9975,
             "count": 400
           }
         }
       },
       "sub_random": {
-        "full_accuracy": 0.985,
+        "full_accuracy": 1.0,
         "n_examples": 200,
         "per_subtask": {
           "MD": {
-            "accuracy": 0.995,
+            "accuracy": 1.0,
             "count": 600
           },
           "MB": {
@@ -1220,11 +1480,11 @@
         }
       },
       "sub_B3": {
-        "full_accuracy": 0.93,
+        "full_accuracy": 1.0,
         "n_examples": 100,
         "per_subtask": {
           "MD": {
-            "accuracy": 0.9966666666666667,
+            "accuracy": 1.0,
             "count": 300
           },
           "MB": {
@@ -1232,7 +1492,7 @@
             "count": 100
           },
           "UB": {
-            "accuracy": 0.9695431472081218,
+            "accuracy": 1.0,
             "count": 197
           },
           "UD": {
@@ -1242,11 +1502,11 @@
         }
       },
       "sub_B4": {
-        "full_accuracy": 0.82,
+        "full_accuracy": 1.0,
         "n_examples": 100,
         "per_subtask": {
           "MD": {
-            "accuracy": 0.995,
+            "accuracy": 1.0,
             "count": 200
           },
           "MB": {
@@ -1254,17 +1514,17 @@
             "count": 100
           },
           "UB": {
-            "accuracy": 0.9352226720647774,
+            "accuracy": 1.0,
             "count": 247
           },
           "UD": {
-            "accuracy": 0.9607843137254902,
+            "accuracy": 1.0,
             "count": 153
           }
         }
       },
       "sub_B5": {
-        "full_accuracy": 0.71,
+        "full_accuracy": 0.96,
         "n_examples": 100,
         "per_subtask": {
           "MD": {
@@ -1276,22 +1536,22 @@
             "count": 100
           },
           "UB": {
-            "accuracy": 0.9026845637583892,
+            "accuracy": 0.9865771812080537,
             "count": 298
           },
           "UD": {
-            "accuracy": 0.9207920792079208,
+            "accuracy": 1.0,
             "count": 202
           }
         }
       }
     },
     "summary": {
-      "overall_accuracy": 0.8704166666666666,
+      "overall_accuracy": 0.9470833333333334,
       "total_examples": 2400,
       "n_splits": 22
     }
   },
-  "sorl_overall_accuracy": 0.8704166666666666,
-  "sft_overall_accuracy": 0.44083333333333335
+  "sorl_overall_accuracy": 0.9470833333333334,
+  "sft_overall_accuracy": 0.7258333333333333
 }

add_sub_sorl_v1_abs10_K1_10K/model.safetensors

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:4dc104978754dca706830aa83c50e922f112f2961bfe046fb9acb75fc2f4cfb9
+oid sha256:82f05b0b5afcd6ea5dc437d2363128885fcfd080a9b715032b42b76a2c61eb2d
 size 650303660

add_sub_sorl_v1_abs10_K1_10K/train_config.json

@@ -30,7 +30,7 @@
   "vq_abs_pretrain_target_vectors": 20000,
   "batch_size": 64,
   "gradient_accumulation_steps": 1,
-  "num_epochs": 10,
+  "num_epochs": 20,
   "emb_warmup_steps": 0,
   "log_every": 50,
   "eval_every": 156,
@@ -69,16 +69,16 @@
   "no_wandb": false,
   "n_params": 162499262,
   "run_name": "add_sub_sorl_v1_abs10_K1_10K",
-  "git_commit": "8d5ee5420119746ef4e2c87570eb250c9718f643",
-  "timestamp": "2026-04-12T23:50:30.170809+00:00",
+  "git_commit": "57deaa28d9c21e39ddac5ef448d6e1be992fba91",
+  "timestamp": "2026-04-13T10:55:46.960375+00:00",
   "tokenizer": "Qwen/Qwen3-0.6B",
   "dataset_repo": "thoughtworks/arithmetic-sorl-data",
   "dataset_config": "add_sub_6digit",
   "model_repo": "thoughtworks/arithmetic-sorl",
   "trainer_version": "v1",
-  "wandb_run_id": "yh5d4it1",
-  "wandb_url": "https://wandb.ai/nlp_and_interpretability/sorl-arithmetic/runs/yh5d4it1",
-  "final_accuracy": 0.8704166666666666,
-  "sft_accuracy": 0.44083333333333335,
+  "wandb_run_id": "0r6snjo4",
+  "wandb_url": "https://wandb.ai/nlp_and_interpretability/sorl-arithmetic/runs/0r6snjo4",
+  "final_accuracy": 0.9470833333333334,
+  "sft_accuracy": 0.7258333333333333,
   "eval_method": "ArithmeticEvaluator"
 }