Upload add_sub_sorl_v1_abs10_K1_10K

add_sub_sorl_v1_abs10_K1_10K/metrics.json

@@ -33,229 +33,229 @@
       1563
     ],
     "loss": [
-      4.547272682189941,
-      7.030660629272461,
-      3.9523563385009766,
-      3.047678232192993,
-      3.085705280303955,
-      2.618412494659424,
-      2.579326629638672,
-      1.4848873615264893,
-      -0.47891831398010254,
-      -5.560091495513916,
-      -4.520766735076904,
-      -2.9085822105407715,
-      -2.024510383605957,
-      -1.1992690563201904,
-      -0.9796783924102783,
-      -0.5373831987380981,
-      -0.5736587047576904,
-      -0.7218008041381836,
-      -0.23317885398864746,
-      -0.339227557182312,
-      -0.03128504753112793,
-      -0.006059408187866211,
-      0.16202318668365479,
-      0.23520565032958984,
-      0.015602707862854004,
-      0.4141479730606079,
-      0.1808193325996399,
-      0.15000015497207642,
-      0.025604248046875,
-      0.30849093198776245
+      9.340689659118652,
+      4.987771987915039,
+      3.1618072986602783,
+      2.856262445449829,
+      2.5268046855926514,
+      2.1091880798339844,
+      -0.44331324100494385,
+      -4.092588424682617,
+      -4.768612384796143,
+      -3.7999463081359863,
+      -1.6939505338668823,
+      -2.5403966903686523,
+      -2.0438146591186523,
+      -1.7041206359863281,
+      -1.5684587955474854,
+      -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
     ],
     "base_loss": [
-      8.916278839111328,
-      6.119595050811768,
-      3.673598527908325,
-      2.020040273666382,
-      1.944067120552063,
-      1.8892498016357422,
-      1.86012864112854,
-      1.8146944046020508,
-      1.869857907295227,
-      1.7928285598754883,
-      1.4957207441329956,
-      1.1758774518966675,
-      0.93010014295578,
-      0.6884778141975403,
-      0.6303120255470276,
-      0.5103054046630859,
-      0.4800644516944885,
-      0.431017130613327,
-      0.3630550801753998,
-      0.3632798194885254,
-      0.32823804020881653,
-      0.2981429398059845,
-      0.24132393300533295,
-      0.23764917254447937,
-      0.2843627333641052,
-      0.19818377494812012,
-      0.2125009298324585,
-      0.18415263295173645,
-      0.19887767732143402,
-      0.1734028160572052
+      7.390318870544434,
+      3.826241970062256,
+      1.9951651096343994,
+      1.8394466638565063,
+      1.8613300323486328,
+      1.835070252418518,
+      1.8304622173309326,
+      1.7413225173950195,
+      1.6582175493240356,
+      1.3251668214797974,
+      0.9685406684875488,
+      0.9228308796882629,
+      0.7192235589027405,
+      0.5874363780021667,
+      0.5402608513832092,
+      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
     ],
     "info_loss": [
-      -1.2568669319152832,
-      -0.31183719635009766,
-      -0.1900501251220703,
-      -0.09034490585327148,
-      -0.07560265064239502,
-      -0.11611974239349365,
-      -0.11614418029785156,
-      -0.2208923101425171,
-      -0.42284250259399414,
-      -0.9240456819534302,
-      -0.7907073497772217,
-      -0.5979017615318298,
-      -0.4831601679325104,
-      -0.3740352690219879,
-      -0.34062066674232483,
-      -0.2785865366458893,
-      -0.27203837037086487,
-      -0.27023085951805115,
-      -0.20767968893051147,
-      -0.21481892466545105,
-      -0.16902248561382294,
-      -0.1613115668296814,
-      -0.1327231526374817,
-      -0.12193454056978226,
-      -0.14406734704971313,
-      -0.09546159207820892,
-      -0.1119590476155281,
-      -0.11057599633932114,
-      -0.12191784381866455,
-      -0.09515686333179474
+      -0.4859275817871094,
+      -0.1129457950592041,
+      -0.07287442684173584,
+      -0.0863027572631836,
+      -0.12096810340881348,
+      -0.1600182056427002,
+      -0.41505467891693115,
+      -0.7713239789009094,
+      -0.8302392959594727,
+      -0.6981891989707947,
+      -0.4436832070350647,
+      -0.5114461183547974,
+      -0.42171621322631836,
+      -0.35522836446762085,
+      -0.32737311720848083,
+      -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
     ],
     "abs_loss": [
-      2.28905987739563,
-      2.068211555480957,
-      1.8956918716430664,
-      1.865671157836914,
-      1.8401455879211426,
-      1.8589462041854858,
-      1.8197526931762695,
-      1.8445491790771484,
-      1.8481858968734741,
-      1.8446826934814453,
-      1.8715795278549194,
-      1.8394945859909058,
-      1.7485918998718262,
-      1.6203765869140625,
-      1.4851690530776978,
-      1.3853811025619507,
-      1.1656250953674316,
-      1.0630995035171509,
-      0.8788065910339355,
-      0.8097158074378967,
-      0.7307209968566895,
-      0.6055489182472229,
-      0.6438609957695007,
-      0.5721376538276672,
-      0.5374541878700256,
-      0.5297055840492249,
-      0.45829495787620544,
-      0.428969144821167,
-      0.47001174092292786,
-      0.4526701867580414
+      2.2262747287750244,
+      1.8807449340820312,
+      1.833791732788086,
+      1.8724111318588257,
+      1.8476206064224243,
+      1.8559261560440063,
+      1.87335205078125,
+      1.8458162546157837,
+      1.813267707824707,
+      1.6594558954238892,
+      1.4701725244522095,
+      1.2008872032165527,
+      0.8842588067054749,
+      0.7724447250366211,
+      0.6772302985191345,
+      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
     ],
     "zipf_loss": [
-      7.970757007598877,
-      3.8226165771484375,
-      1.9896899461746216,
-      1.7445199489593506,
-      1.713650107383728,
-      1.704465627670288,
-      1.6986644268035889,
-      1.6946611404418945,
-      1.6948304176330566,
-      1.7030682563781738,
-      1.7034276723861694,
-      1.7106086015701294,
-      1.702131748199463,
-      1.690568208694458,
-      1.6476993560791016,
-      1.59963858127594,
-      1.550097942352295,
-      1.443180799484253,
-      1.392682433128357,
-      1.3647103309631348,
-      1.2576297521591187,
-      1.2483584880828857,
-      1.183544635772705,
-      1.159688115119934,
-      1.1181681156158447,
-      1.1176096200942993,
-      1.0420794486999512,
-      1.0287106037139893,
-      0.9989038109779358,
-      1.0413897037506104
+      6.587018966674805,
+      2.1029131412506104,
+      1.7120072841644287,
+      1.692602276802063,
+      1.6903936862945557,
+      1.6887073516845703,
+      1.6894363164901733,
+      1.6947468519210815,
+      1.694236397743225,
+      1.690833568572998,
+      1.6273235082626343,
+      1.531144618988037,
+      1.3656980991363525,
+      1.1834821701049805,
+      1.0972886085510254,
+      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
     ],
     "denoise_loss": [],
     "ortho_loss": [
-      0.5579876899719238,
-      0.252128541469574,
-      0.21211421489715576,
-      0.18933552503585815,
-      0.12901675701141357,
-      0.13211114704608917,
-      0.14735311269760132,
-      0.14610400795936584,
-      0.1557065099477768,
-      0.17582964897155762,
-      0.1939730942249298,
-      0.20376113057136536,
-      0.21460682153701782,
-      0.23509620130062103,
-      0.24628381431102753,
-      0.2590982913970947,
-      0.25852254033088684,
-      0.2667267620563507,
-      0.26971182227134705,
-      0.2820512354373932,
-      0.2821248769760132,
-      0.28232815861701965,
-      0.2819784879684448,
-      0.2866751551628113,
-      0.2907232642173767,
-      0.2921256422996521,
-      0.2953319251537323,
-      0.29460301995277405,
-      0.295316606760025,
-      0.2968752384185791
+      0.3918377161026001,
+      0.20118887722492218,
+      0.12065954506397247,
+      0.0842815414071083,
+      0.07851361483335495,
+      0.08886810392141342,
+      0.1118798479437828,
+      0.16310755908489227,
+      0.21554774045944214,
+      0.2462739199399948,
+      0.2753961980342865,
+      0.30527833104133606,
+      0.33816447854042053,
+      0.3525509536266327,
+      0.36425891518592834,
+      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
     ],
     "lr": [
-      1.9600000000000002e-05,
-      3.96e-05,
-      4e-05,
-      4e-05,
-      4e-05,
-      4e-05,
-      4e-05,
-      4e-05,
-      4e-05,
-      4e-05,
-      4e-05,
-      4e-05,
-      4e-05,
-      4e-05,
-      4e-05,
-      4e-05,
-      4e-05,
-      4e-05,
-      3.944897959183673e-05,
-      3.638775510204082e-05,
-      3.3326530612244897e-05,
-      2.983673469387755e-05,
-      2.6775510204081637e-05,
-      2.3714285714285713e-05,
-      2.022448979591837e-05,
-      1.7163265306122454e-05,
-      1.4102040816326535e-05,
-      1.0612244897959182e-05,
-      7.551020408163262e-06,
-      4.48979591836735e-06
+      3.9200000000000004e-05,
+      7.92e-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,
+      7.889795918367346e-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
     ],
     "emb_lr": [],
     "eval_step": [
@@ -283,7 +283,7 @@
       0.0
     ]
   },
-  "final_accuracy": 0.6116666666666667,
+  "final_accuracy": 0.8704166666666666,
   "sft_eval": {
     "config": {
       "ops": "add_sub",
@@ -294,21 +294,21 @@
     },
     "splits": {
       "add_S0": {
-        "full_accuracy": 0.89,
+        "full_accuracy": 0.9,
         "n_examples": 100,
         "per_subtask": {
           "SA": {
-            "accuracy": 0.9834710743801653,
+            "accuracy": 0.9818181818181818,
             "count": 605
           },
           "SS": {
-            "accuracy": 0.9473684210526315,
+            "accuracy": 1.0,
             "count": 95
           }
         }
       },
       "add_S1": {
-        "full_accuracy": 0.74,
+        "full_accuracy": 0.83,
         "n_examples": 100,
         "per_subtask": {
           "SA": {
@@ -316,7 +316,7 @@
             "count": 204
           },
           "SC": {
-            "accuracy": 0.9467455621301775,
+            "accuracy": 0.9940828402366864,
             "count": 169
           },
           "SS": {
@@ -324,91 +324,91 @@
             "count": 31
           },
           "UC": {
-            "accuracy": 0.9391891891891891,
+            "accuracy": 0.956081081081081,
             "count": 296
           }
         }
       },
       "add_S2": {
-        "full_accuracy": 0.5,
+        "full_accuracy": 0.54,
         "n_examples": 100,
         "per_subtask": {
           "SA": {
-            "accuracy": 0.9631901840490797,
+            "accuracy": 0.9815950920245399,
             "count": 163
           },
           "SC": {
-            "accuracy": 0.9230769230769231,
+            "accuracy": 0.9538461538461539,
             "count": 130
           },
           "SS": {
-            "accuracy": 0.8505747126436781,
+            "accuracy": 0.9770114942528736,
             "count": 87
           },
           "UC": {
-            "accuracy": 0.8325123152709359,
+            "accuracy": 0.812807881773399,
             "count": 203
           },
           "US": {
-            "accuracy": 0.9743589743589743,
+            "accuracy": 0.9401709401709402,
             "count": 117
           }
         }
       },
       "add_S3": {
-        "full_accuracy": 0.29,
+        "full_accuracy": 0.3,
         "n_examples": 100,
         "per_subtask": {
           "SA": {
-            "accuracy": 0.9917355371900827,
+            "accuracy": 0.9752066115702479,
             "count": 121
           },
           "SC": {
-            "accuracy": 0.9421487603305785,
+            "accuracy": 0.9752066115702479,
             "count": 121
           },
           "SS": {
-            "accuracy": 0.8979591836734694,
+            "accuracy": 0.9795918367346939,
             "count": 49
           },
           "UC": {
-            "accuracy": 0.7204301075268817,
+            "accuracy": 0.6666666666666666,
             "count": 186
           },
           "US": {
-            "accuracy": 0.8340807174887892,
+            "accuracy": 0.7668161434977578,
             "count": 223
           }
         }
       },
       "add_S4": {
-        "full_accuracy": 0.19,
+        "full_accuracy": 0.23,
         "n_examples": 100,
         "per_subtask": {
           "SA": {
-            "accuracy": 0.9711538461538461,
+            "accuracy": 0.9807692307692307,
             "count": 104
           },
           "SC": {
-            "accuracy": 0.9622641509433962,
+            "accuracy": 0.9811320754716981,
             "count": 106
           },
           "SS": {
-            "accuracy": 0.9565217391304348,
+            "accuracy": 1.0,
             "count": 23
           },
           "UC": {
-            "accuracy": 0.59375,
+            "accuracy": 0.58125,
             "count": 160
           },
           "US": {
-            "accuracy": 0.6156351791530945,
+            "accuracy": 0.5635179153094463,
             "count": 307
           }
         }
       },
       "add_S5": {
-        "full_accuracy": 0.19,
+        "full_accuracy": 0.12,
         "n_examples": 100,
         "per_subtask": {
           "SA": {
@@ -416,65 +416,65 @@
             "count": 100
           },
           "SC": {
-            "accuracy": 0.96,
+            "accuracy": 0.94,
             "count": 100
           },
           "UC": {
-            "accuracy": 0.31,
+            "accuracy": 0.3,
             "count": 100
           },
           "US": {
-            "accuracy": 0.4775,
+            "accuracy": 0.325,
             "count": 400
           }
         }
       },
       "add_S6": {
-        "full_accuracy": 0.53,
+        "full_accuracy": 0.34,
         "n_examples": 100,
         "per_subtask": {
           "SC": {
-            "accuracy": 0.99,
+            "accuracy": 1.0,
             "count": 100
           },
           "UC": {
-            "accuracy": 0.65,
+            "accuracy": 0.53,
             "count": 100
           },
           "US": {
-            "accuracy": 0.672,
+            "accuracy": 0.514,
             "count": 500
           }
         }
       },
       "add_random": {
-        "full_accuracy": 0.73,
+        "full_accuracy": 0.7,
         "n_examples": 200,
         "per_subtask": {
           "SA": {
-            "accuracy": 0.9776286353467561,
+            "accuracy": 0.9686800894854586,
             "count": 447
           },
           "SC": {
-            "accuracy": 0.975,
+            "accuracy": 0.971875,
             "count": 320
           },
           "SS": {
-            "accuracy": 0.9821428571428571,
+            "accuracy": 0.9642857142857143,
             "count": 56
           },
           "UC": {
-            "accuracy": 0.9338374291115312,
+            "accuracy": 0.9243856332703214,
             "count": 529
           },
           "US": {
-            "accuracy": 0.9583333333333334,
+            "accuracy": 0.8958333333333334,
             "count": 48
           }
         }
       },
       "add_C3": {
-        "full_accuracy": 0.42,
+        "full_accuracy": 0.48,
         "n_examples": 100,
         "per_subtask": {
           "SA": {
@@ -486,61 +486,61 @@
             "count": 100
           },
           "UC": {
-            "accuracy": 0.7253886010362695,
+            "accuracy": 0.7461139896373057,
             "count": 193
           },
           "US": {
-            "accuracy": 0.8504672897196262,
+            "accuracy": 0.7663551401869159,
             "count": 107
           }
         }
       },
       "add_C4": {
-        "full_accuracy": 0.43,
+        "full_accuracy": 0.38,
         "n_examples": 100,
         "per_subtask": {
           "SA": {
-            "accuracy": 1.0,
+            "accuracy": 0.985,
             "count": 200
           },
           "SC": {
-            "accuracy": 0.96,
+            "accuracy": 0.98,
             "count": 100
           },
           "UC": {
-            "accuracy": 0.76953125,
+            "accuracy": 0.7734375,
             "count": 256
           },
           "US": {
-            "accuracy": 0.8055555555555556,
+            "accuracy": 0.8194444444444444,
             "count": 144
           }
         }
       },
       "add_C5": {
-        "full_accuracy": 0.31,
+        "full_accuracy": 0.4,
         "n_examples": 100,
         "per_subtask": {
           "SA": {
-            "accuracy": 1.0,
+            "accuracy": 0.99,
             "count": 100
           },
           "SC": {
-            "accuracy": 0.98,
+            "accuracy": 1.0,
             "count": 100
           },
           "UC": {
-            "accuracy": 0.7745098039215687,
+            "accuracy": 0.7908496732026143,
             "count": 306
           },
           "US": {
-            "accuracy": 0.7783505154639175,
+            "accuracy": 0.845360824742268,
             "count": 194
           }
         }
       },
       "add_C6": {
-        "full_accuracy": 0.44,
+        "full_accuracy": 0.35,
         "n_examples": 100,
         "per_subtask": {
           "SC": {
@@ -548,39 +548,39 @@
             "count": 100
           },
           "UC": {
-            "accuracy": 0.8306010928961749,
+            "accuracy": 0.8005464480874317,
             "count": 366
           },
           "US": {
-            "accuracy": 0.8931623931623932,
+            "accuracy": 0.8290598290598291,
             "count": 234
           }
         }
       },
       "sub_M0": {
-        "full_accuracy": 0.87,
+        "full_accuracy": 0.89,
         "n_examples": 100,
         "per_subtask": {
           "MD": {
-            "accuracy": 0.9750415973377704,
+            "accuracy": 0.9833610648918469,
             "count": 601
           },
           "ME": {
-            "accuracy": 1.0,
+            "accuracy": 0.9797979797979798,
             "count": 99
           }
         }
       },
       "sub_M1": {
-        "full_accuracy": 0.82,
+        "full_accuracy": 0.81,
         "n_examples": 100,
         "per_subtask": {
           "MD": {
-            "accuracy": 0.989247311827957,
+            "accuracy": 0.992831541218638,
             "count": 279
           },
           "MB": {
-            "accuracy": 0.9586206896551724,
+            "accuracy": 0.993103448275862,
             "count": 145
           },
           "ME": {
@@ -588,47 +588,47 @@
             "count": 24
           },
           "UB": {
-            "accuracy": 0.9563492063492064,
+            "accuracy": 0.9285714285714286,
             "count": 252
           }
         }
       },
       "sub_M2": {
-        "full_accuracy": 0.47,
+        "full_accuracy": 0.42,
         "n_examples": 100,
         "per_subtask": {
           "MD": {
-            "accuracy": 0.9953051643192489,
+            "accuracy": 0.9859154929577465,
             "count": 213
           },
           "MB": {
-            "accuracy": 0.9380530973451328,
+            "accuracy": 0.9734513274336283,
             "count": 113
           },
           "ME": {
-            "accuracy": 1.0,
+            "accuracy": 0.9882352941176471,
             "count": 85
           },
           "UB": {
-            "accuracy": 0.7237569060773481,
+            "accuracy": 0.6740331491712708,
             "count": 181
           },
           "UD": {
-            "accuracy": 0.9444444444444444,
+            "accuracy": 0.9722222222222222,
             "count": 108
           }
         }
       },
       "sub_M3": {
-        "full_accuracy": 0.12,
+        "full_accuracy": 0.06,
         "n_examples": 100,
         "per_subtask": {
           "MD": {
-            "accuracy": 0.994413407821229,
+            "accuracy": 1.0,
             "count": 179
           },
           "MB": {
-            "accuracy": 0.9805825242718447,
+            "accuracy": 0.970873786407767,
             "count": 103
           },
           "ME": {
@@ -636,17 +636,17 @@
             "count": 56
           },
           "UB": {
-            "accuracy": 0.5302013422818792,
+            "accuracy": 0.3624161073825503,
             "count": 149
           },
           "UD": {
-            "accuracy": 0.7417840375586855,
+            "accuracy": 0.6384976525821596,
             "count": 213
           }
         }
       },
       "sub_M4": {
-        "full_accuracy": 0.0,
+        "full_accuracy": 0.02,
         "n_examples": 100,
         "per_subtask": {
           "MD": {
@@ -654,21 +654,21 @@
             "count": 200
           },
           "MB": {
-            "accuracy": 1.0,
+            "accuracy": 0.91,
             "count": 100
           },
           "UB": {
-            "accuracy": 0.36,
+            "accuracy": 0.38,
             "count": 100
           },
           "UD": {
-            "accuracy": 0.2966666666666667,
+            "accuracy": 0.29,
             "count": 300
           }
         }
       },
       "sub_M5": {
-        "full_accuracy": 0.01,
+        "full_accuracy": 0.0,
         "n_examples": 100,
         "per_subtask": {
           "MD": {
@@ -676,29 +676,29 @@
             "count": 100
           },
           "MB": {
-            "accuracy": 1.0,
+            "accuracy": 0.85,
             "count": 100
           },
           "UB": {
-            "accuracy": 0.3,
+            "accuracy": 0.25,
             "count": 100
           },
           "UD": {
-            "accuracy": 0.215,
+            "accuracy": 0.19,
             "count": 400
           }
         }
       },
       "sub_random": {
-        "full_accuracy": 0.765,
+        "full_accuracy": 0.71,
         "n_examples": 200,
         "per_subtask": {
           "MD": {
-            "accuracy": 0.9783333333333334,
+            "accuracy": 0.9816666666666667,
             "count": 600
           },
           "MB": {
-            "accuracy": 0.9700374531835206,
+            "accuracy": 0.9925093632958801,
             "count": 267
           },
           "ME": {
@@ -706,21 +706,21 @@
             "count": 53
           },
           "UB": {
-            "accuracy": 0.9384965831435079,
+            "accuracy": 0.8906605922551253,
             "count": 439
           },
           "UD": {
-            "accuracy": 0.9512195121951219,
+            "accuracy": 0.8536585365853658,
             "count": 41
           }
         }
       },
       "sub_B3": {
-        "full_accuracy": 0.49,
+        "full_accuracy": 0.33,
         "n_examples": 100,
         "per_subtask": {
           "MD": {
-            "accuracy": 0.9966666666666667,
+            "accuracy": 0.9866666666666667,
             "count": 300
           },
           "MB": {
@@ -728,17 +728,17 @@
             "count": 100
           },
           "UB": {
-            "accuracy": 0.766497461928934,
+            "accuracy": 0.6700507614213198,
             "count": 197
           },
           "UD": {
-            "accuracy": 0.7572815533980582,
+            "accuracy": 0.7378640776699029,
             "count": 103
           }
         }
       },
       "sub_B4": {
-        "full_accuracy": 0.4,
+        "full_accuracy": 0.21,
         "n_examples": 100,
         "per_subtask": {
           "MD": {
@@ -746,21 +746,21 @@
             "count": 200
           },
           "MB": {
-            "accuracy": 1.0,
+            "accuracy": 0.98,
             "count": 100
           },
           "UB": {
-            "accuracy": 0.7975708502024291,
+            "accuracy": 0.6761133603238867,
             "count": 247
           },
           "UD": {
-            "accuracy": 0.6928104575163399,
+            "accuracy": 0.6666666666666666,
             "count": 153
           }
         }
       },
       "sub_B5": {
-        "full_accuracy": 0.2,
+        "full_accuracy": 0.15,
         "n_examples": 100,
         "per_subtask": {
           "MD": {
@@ -768,22 +768,22 @@
             "count": 100
           },
           "MB": {
-            "accuracy": 1.0,
+            "accuracy": 0.97,
             "count": 100
           },
           "UB": {
-            "accuracy": 0.7348993288590604,
+            "accuracy": 0.6543624161073825,
             "count": 298
           },
           "UD": {
-            "accuracy": 0.6782178217821783,
+            "accuracy": 0.6089108910891089,
             "count": 202
           }
         }
       }
     },
     "summary": {
-      "overall_accuracy": 0.47041666666666665,
+      "overall_accuracy": 0.44083333333333335,
       "total_examples": 2400,
       "n_splits": 22
     }
@@ -798,11 +798,11 @@
     },
     "splits": {
       "add_S0": {
-        "full_accuracy": 0.97,
+        "full_accuracy": 1.0,
         "n_examples": 100,
         "per_subtask": {
           "SA": {
-            "accuracy": 0.9950413223140496,
+            "accuracy": 1.0,
             "count": 605
           },
           "SS": {
@@ -812,11 +812,11 @@
         }
       },
       "add_S1": {
-        "full_accuracy": 0.96,
+        "full_accuracy": 1.0,
         "n_examples": 100,
         "per_subtask": {
           "SA": {
-            "accuracy": 0.9901960784313726,
+            "accuracy": 1.0,
             "count": 204
           },
           "SC": {
@@ -824,33 +824,33 @@
             "count": 169
           },
           "SS": {
-            "accuracy": 0.967741935483871,
+            "accuracy": 1.0,
             "count": 31
           },
           "UC": {
-            "accuracy": 0.9932432432432432,
+            "accuracy": 1.0,
             "count": 296
           }
         }
       },
       "add_S2": {
-        "full_accuracy": 0.83,
+        "full_accuracy": 1.0,
         "n_examples": 100,
         "per_subtask": {
           "SA": {
-            "accuracy": 0.9815950920245399,
+            "accuracy": 1.0,
             "count": 163
           },
           "SC": {
-            "accuracy": 0.9769230769230769,
+            "accuracy": 1.0,
             "count": 130
           },
           "SS": {
-            "accuracy": 0.9885057471264368,
+            "accuracy": 1.0,
             "count": 87
           },
           "UC": {
-            "accuracy": 0.9408866995073891,
+            "accuracy": 1.0,
             "count": 203
           },
           "US": {
@@ -860,15 +860,15 @@
         }
       },
       "add_S3": {
-        "full_accuracy": 0.55,
+        "full_accuracy": 1.0,
         "n_examples": 100,
         "per_subtask": {
           "SA": {
-            "accuracy": 0.9917355371900827,
+            "accuracy": 1.0,
             "count": 121
           },
           "SC": {
-            "accuracy": 0.9917355371900827,
+            "accuracy": 1.0,
             "count": 121
           },
           "SS": {
@@ -876,17 +876,17 @@
             "count": 49
           },
           "UC": {
-            "accuracy": 0.7903225806451613,
+            "accuracy": 1.0,
             "count": 186
           },
           "US": {
-            "accuracy": 0.9461883408071748,
+            "accuracy": 1.0,
             "count": 223
           }
         }
       },
       "add_S4": {
-        "full_accuracy": 0.4,
+        "full_accuracy": 1.0,
         "n_examples": 100,
         "per_subtask": {
           "SA": {
@@ -902,17 +902,17 @@
             "count": 23
           },
           "UC": {
-            "accuracy": 0.7,
+            "accuracy": 1.0,
             "count": 160
           },
           "US": {
-            "accuracy": 0.8045602605863192,
+            "accuracy": 1.0,
             "count": 307
           }
         }
       },
       "add_S5": {
-        "full_accuracy": 0.27,
+        "full_accuracy": 0.58,
         "n_examples": 100,
         "per_subtask": {
           "SA": {
@@ -924,17 +924,17 @@
             "count": 100
           },
           "UC": {
-            "accuracy": 0.41,
+            "accuracy": 0.58,
             "count": 100
           },
           "US": {
-            "accuracy": 0.59,
+            "accuracy": 1.0,
             "count": 400
           }
         }
       },
       "add_S6": {
-        "full_accuracy": 0.32,
+        "full_accuracy": 0.88,
         "n_examples": 100,
         "per_subtask": {
           "SC": {
@@ -942,25 +942,25 @@
             "count": 100
           },
           "UC": {
-            "accuracy": 0.52,
+            "accuracy": 0.95,
             "count": 100
           },
           "US": {
-            "accuracy": 0.586,
+            "accuracy": 0.972,
             "count": 500
           }
         }
       },
       "add_random": {
-        "full_accuracy": 0.955,
+        "full_accuracy": 1.0,
         "n_examples": 200,
         "per_subtask": {
           "SA": {
-            "accuracy": 0.9977628635346756,
+            "accuracy": 1.0,
             "count": 447
           },
           "SC": {
-            "accuracy": 0.996875,
+            "accuracy": 1.0,
             "count": 320
           },
           "SS": {
@@ -968,7 +968,7 @@
             "count": 56
           },
           "UC": {
-            "accuracy": 0.9867674858223062,
+            "accuracy": 1.0,
             "count": 529
           },
           "US": {
@@ -978,7 +978,7 @@
         }
       },
       "add_C3": {
-        "full_accuracy": 0.67,
+        "full_accuracy": 0.99,
         "n_examples": 100,
         "per_subtask": {
           "SA": {
@@ -990,17 +990,17 @@
             "count": 100
           },
           "UC": {
-            "accuracy": 0.8341968911917098,
+            "accuracy": 0.9948186528497409,
             "count": 193
           },
           "US": {
-            "accuracy": 0.9065420560747663,
+            "accuracy": 1.0,
             "count": 107
           }
         }
       },
       "add_C4": {
-        "full_accuracy": 0.63,
+        "full_accuracy": 1.0,
         "n_examples": 100,
         "per_subtask": {
           "SA": {
@@ -1012,17 +1012,17 @@
             "count": 100
           },
           "UC": {
-            "accuracy": 0.86328125,
+            "accuracy": 1.0,
             "count": 256
           },
           "US": {
-            "accuracy": 0.9097222222222222,
+            "accuracy": 1.0,
             "count": 144
           }
         }
       },
       "add_C5": {
-        "full_accuracy": 0.57,
+        "full_accuracy": 0.97,
         "n_examples": 100,
         "per_subtask": {
           "SA": {
@@ -1034,17 +1034,17 @@
             "count": 100
           },
           "UC": {
-            "accuracy": 0.8725490196078431,
+            "accuracy": 0.9934640522875817,
             "count": 306
           },
           "US": {
-            "accuracy": 0.9020618556701031,
+            "accuracy": 0.9948453608247423,
             "count": 194
           }
         }
       },
       "add_C6": {
-        "full_accuracy": 0.66,
+        "full_accuracy": 0.98,
         "n_examples": 100,
         "per_subtask": {
           "SC": {
@@ -1052,31 +1052,31 @@
             "count": 100
           },
           "UC": {
-            "accuracy": 0.9098360655737705,
+            "accuracy": 0.994535519125683,
             "count": 366
           },
           "US": {
-            "accuracy": 0.9487179487179487,
+            "accuracy": 1.0,
             "count": 234
           }
         }
       },
       "sub_M0": {
-        "full_accuracy": 0.9,
+        "full_accuracy": 0.99,
         "n_examples": 100,
         "per_subtask": {
           "MD": {
-            "accuracy": 0.9850249584026622,
+            "accuracy": 0.9983361064891847,
             "count": 601
           },
           "ME": {
-            "accuracy": 0.98989898989899,
+            "accuracy": 1.0,
             "count": 99
           }
         }
       },
       "sub_M1": {
-        "full_accuracy": 0.99,
+        "full_accuracy": 1.0,
         "n_examples": 100,
         "per_subtask": {
           "MD": {
@@ -1084,7 +1084,7 @@
             "count": 279
           },
           "MB": {
-            "accuracy": 0.993103448275862,
+            "accuracy": 1.0,
             "count": 145
           },
           "ME": {
@@ -1098,15 +1098,15 @@
         }
       },
       "sub_M2": {
-        "full_accuracy": 0.6,
+        "full_accuracy": 0.95,
         "n_examples": 100,
         "per_subtask": {
           "MD": {
-            "accuracy": 0.9953051643192489,
+            "accuracy": 0.9906103286384976,
             "count": 213
           },
           "MB": {
-            "accuracy": 0.9734513274336283,
+            "accuracy": 1.0,
             "count": 113
           },
           "ME": {
@@ -1114,7 +1114,7 @@
             "count": 85
           },
           "UB": {
-            "accuracy": 0.7845303867403315,
+            "accuracy": 0.9834254143646409,
             "count": 181
           },
           "UD": {
@@ -1124,7 +1124,7 @@
         }
       },
       "sub_M3": {
-        "full_accuracy": 0.07,
+        "full_accuracy": 0.79,
         "n_examples": 100,
         "per_subtask": {
           "MD": {
@@ -1132,7 +1132,7 @@
             "count": 179
           },
           "MB": {
-            "accuracy": 0.9805825242718447,
+            "accuracy": 1.0,
             "count": 103
           },
           "ME": {
@@ -1140,17 +1140,17 @@
             "count": 56
           },
           "UB": {
-            "accuracy": 0.4563758389261745,
+            "accuracy": 0.87248322147651,
             "count": 149
           },
           "UD": {
-            "accuracy": 0.7699530516431925,
+            "accuracy": 0.9624413145539906,
             "count": 213
           }
         }
       },
       "sub_M4": {
-        "full_accuracy": 0.07,
+        "full_accuracy": 0.3,
         "n_examples": 100,
         "per_subtask": {
           "MD": {
@@ -1162,17 +1162,17 @@
             "count": 100
           },
           "UB": {
-            "accuracy": 0.38,
+            "accuracy": 0.36,
             "count": 100
           },
           "UD": {
-            "accuracy": 0.48,
+            "accuracy": 0.8133333333333334,
             "count": 300
           }
         }
       },
       "sub_M5": {
-        "full_accuracy": 0.02,
+        "full_accuracy": 0.04,
         "n_examples": 100,
         "per_subtask": {
           "MD": {
@@ -1184,25 +1184,25 @@
             "count": 100
           },
           "UB": {
-            "accuracy": 0.26,
+            "accuracy": 0.12,
             "count": 100
           },
           "UD": {
-            "accuracy": 0.38,
+            "accuracy": 0.6575,
             "count": 400
           }
         }
       },
       "sub_random": {
-        "full_accuracy": 0.89,
+        "full_accuracy": 0.985,
         "n_examples": 200,
         "per_subtask": {
           "MD": {
-            "accuracy": 0.9966666666666667,
+            "accuracy": 0.995,
             "count": 600
           },
           "MB": {
-            "accuracy": 0.9962546816479401,
+            "accuracy": 1.0,
             "count": 267
           },
           "ME": {
@@ -1210,7 +1210,7 @@
             "count": 53
           },
           "UB": {
-            "accuracy": 0.9567198177676538,
+            "accuracy": 1.0,
             "count": 439
           },
           "UD": {
@@ -1220,11 +1220,11 @@
         }
       },
       "sub_B3": {
-        "full_accuracy": 0.68,
+        "full_accuracy": 0.93,
         "n_examples": 100,
         "per_subtask": {
           "MD": {
-            "accuracy": 1.0,
+            "accuracy": 0.9966666666666667,
             "count": 300
           },
           "MB": {
@@ -1232,21 +1232,21 @@
             "count": 100
           },
           "UB": {
-            "accuracy": 0.8578680203045685,
+            "accuracy": 0.9695431472081218,
             "count": 197
           },
           "UD": {
-            "accuracy": 0.8932038834951457,
+            "accuracy": 1.0,
             "count": 103
           }
         }
       },
       "sub_B4": {
-        "full_accuracy": 0.51,
+        "full_accuracy": 0.82,
         "n_examples": 100,
         "per_subtask": {
           "MD": {
-            "accuracy": 1.0,
+            "accuracy": 0.995,
             "count": 200
           },
           "MB": {
@@ -1254,17 +1254,17 @@
             "count": 100
           },
           "UB": {
-            "accuracy": 0.8380566801619433,
+            "accuracy": 0.9352226720647774,
             "count": 247
           },
           "UD": {
-            "accuracy": 0.7647058823529411,
+            "accuracy": 0.9607843137254902,
             "count": 153
           }
         }
       },
       "sub_B5": {
-        "full_accuracy": 0.33,
+        "full_accuracy": 0.71,
         "n_examples": 100,
         "per_subtask": {
           "MD": {
@@ -1276,22 +1276,22 @@
             "count": 100
           },
           "UB": {
-            "accuracy": 0.8187919463087249,
+            "accuracy": 0.9026845637583892,
             "count": 298
           },
           "UD": {
-            "accuracy": 0.7524752475247525,
+            "accuracy": 0.9207920792079208,
             "count": 202
           }
         }
       }
     },
     "summary": {
-      "overall_accuracy": 0.6116666666666667,
+      "overall_accuracy": 0.8704166666666666,
       "total_examples": 2400,
       "n_splits": 22
     }
   },
-  "sorl_overall_accuracy": 0.6116666666666667,
-  "sft_overall_accuracy": 0.47041666666666665
+  "sorl_overall_accuracy": 0.8704166666666666,
+  "sft_overall_accuracy": 0.44083333333333335
 }

add_sub_sorl_v1_abs10_K1_10K/model.safetensors

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:b512b0970baed990d0164c674044de1463fd819596795fb6a137261185804ce5
+oid sha256:4dc104978754dca706830aa83c50e922f112f2961bfe046fb9acb75fc2f4cfb9
 size 650303660

add_sub_sorl_v1_abs10_K1_10K/train_config.json

@@ -17,7 +17,7 @@
   "target_vocab_util": 0.8,
   "min_abs_ppl": 0.0,
   "zipf_alpha": 1.0,
-  "lr": 4e-05,
+  "lr": 8e-05,
   "emb_lr_mult": 1.0,
   "weight_decay": 0.01,
   "warmup_steps": 100,
@@ -69,16 +69,16 @@
   "no_wandb": false,
   "n_params": 162499262,
   "run_name": "add_sub_sorl_v1_abs10_K1_10K",
-  "git_commit": "78d46f8665a87f4b44bd5894bd34f393f2dea51f",
-  "timestamp": "2026-04-12T08:59:10.916999+00:00",
+  "git_commit": "8d5ee5420119746ef4e2c87570eb250c9718f643",
+  "timestamp": "2026-04-12T23:50:30.170809+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": "pdbywwrx",
-  "wandb_url": "https://wandb.ai/nlp_and_interpretability/sorl-arithmetic/runs/pdbywwrx",
-  "final_accuracy": 0.6116666666666667,
-  "sft_accuracy": 0.47041666666666665,
+  "wandb_run_id": "yh5d4it1",
+  "wandb_url": "https://wandb.ai/nlp_and_interpretability/sorl-arithmetic/runs/yh5d4it1",
+  "final_accuracy": 0.8704166666666666,
+  "sft_accuracy": 0.44083333333333335,
   "eval_method": "ArithmeticEvaluator"
 }