Upload add_sub_baseline_10K

add_sub_baseline_10K/metrics.json

@@ -65,196 +65,196 @@
       3100
     ],
     "loss": [
-      7.930843830108643,
-      5.878052711486816,
-      4.171169757843018,
-      2.4461560249328613,
-      1.9941270351409912,
-      1.885154366493225,
-      1.7704766988754272,
-      1.7200186252593994,
-      1.6930021047592163,
-      1.659714698791504,
-      1.6491907835006714,
-      1.440868616104126,
-      1.2277355194091797,
-      0.9296635389328003,
-      0.727922797203064,
-      0.6531419157981873,
-      0.6338068842887878,
-      0.5315085053443909,
-      0.471783846616745,
-      0.44315391778945923,
-      0.37063688039779663,
-      0.31710734963417053,
-      0.32954543828964233,
-      0.2792213559150696,
-      0.28969308733940125,
-      0.27333250641822815,
-      0.2600690424442291,
-      0.22665680944919586,
-      0.22485388815402985,
-      0.18426764011383057,
-      0.20698437094688416,
-      0.20130981504917145,
-      0.16024114191532135,
-      0.20614077150821686,
-      0.16621388494968414,
-      0.1674460619688034,
-      0.13519535958766937,
-      0.17641237378120422,
-      0.14274020493030548,
-      0.15076635777950287,
-      0.14852766692638397,
-      0.12220268696546555,
-      0.13161125779151917,
-      0.12244851887226105,
-      0.14852531254291534,
-      0.15561267733573914,
-      0.10794571787118912,
-      0.14725656807422638,
-      0.14284299314022064,
-      0.10321654379367828,
-      0.10347577184438705,
-      0.1123846247792244,
-      0.11255372315645218,
-      0.11303780227899551,
-      0.07373049110174179,
-      0.1012827679514885,
-      0.08089525252580643,
-      0.10804768651723862,
-      0.08694573491811752,
-      0.11614620685577393,
-      0.090724878013134,
-      0.11524895578622818
+      7.068239688873291,
+      4.189955234527588,
+      2.0407068729400635,
+      1.8267366886138916,
+      1.7952507734298706,
+      1.7645628452301025,
+      1.595243215560913,
+      1.4900548458099365,
+      1.0821096897125244,
+      0.7350229024887085,
+      0.586064875125885,
+      0.43922391533851624,
+      0.40184348821640015,
+      0.31044527888298035,
+      0.26080775260925293,
+      0.30810627341270447,
+      0.2941185534000397,
+      0.2382466048002243,
+      0.19741371273994446,
+      0.24084995687007904,
+      0.16109716892242432,
+      0.16342215240001678,
+      0.15979401767253876,
+      0.15030500292778015,
+      0.1553926020860672,
+      0.1616949737071991,
+      0.15501153469085693,
+      0.15730805695056915,
+      0.1374218612909317,
+      0.12141235917806625,
+      0.15107305347919464,
+      0.12318699806928635,
+      0.08949179947376251,
+      0.11493054777383804,
+      0.09638842195272446,
+      0.1008308082818985,
+      0.11183486878871918,
+      0.08597197383642197,
+      0.10642959177494049,
+      0.07882422208786011,
+      0.09636445343494415,
+      0.06600546836853027,
+      0.09191834181547165,
+      0.058375708758831024,
+      0.08556245267391205,
+      0.09554620832204819,
+      0.05065993592143059,
+      0.06083018705248833,
+      0.06561274826526642,
+      0.05150684341788292,
+      0.04865796118974686,
+      0.05840981379151344,
+      0.06329210847616196,
+      0.05198041722178459,
+      0.041428614407777786,
+      0.04235445708036423,
+      0.0411263071000576,
+      0.0459698922932148,
+      0.04036566987633705,
+      0.042875390499830246,
+      0.05301470309495926,
+      0.05982867628335953
     ],
     "base_loss": [
-      7.930843830108643,
-      5.878052711486816,
-      4.171169757843018,
-      2.4461560249328613,
-      1.9941270351409912,
-      1.885154366493225,
-      1.7704766988754272,
-      1.7200186252593994,
-      1.6930021047592163,
-      1.659714698791504,
-      1.6491907835006714,
-      1.440868616104126,
-      1.2277355194091797,
-      0.9296635389328003,
-      0.727922797203064,
-      0.6531419157981873,
-      0.6338068842887878,
-      0.5315085053443909,
-      0.471783846616745,
-      0.44315391778945923,
-      0.37063688039779663,
-      0.31710734963417053,
-      0.32954543828964233,
-      0.2792213559150696,
-      0.28969308733940125,
-      0.27333250641822815,
-      0.2600690424442291,
-      0.22665680944919586,
-      0.22485388815402985,
-      0.18426764011383057,
-      0.20698437094688416,
-      0.20130981504917145,
-      0.16024114191532135,
-      0.20614077150821686,
-      0.16621388494968414,
-      0.1674460619688034,
-      0.13519535958766937,
-      0.17641237378120422,
-      0.14274020493030548,
-      0.15076635777950287,
-      0.14852766692638397,
-      0.12220268696546555,
-      0.13161125779151917,
-      0.12244851887226105,
-      0.14852531254291534,
-      0.15561267733573914,
-      0.10794571787118912,
-      0.14725656807422638,
-      0.14284299314022064,
-      0.10321654379367828,
-      0.10347577184438705,
-      0.1123846247792244,
-      0.11255372315645218,
-      0.11303780227899551,
-      0.07373049110174179,
-      0.1012827679514885,
-      0.08089525252580643,
-      0.10804768651723862,
-      0.08694573491811752,
-      0.11614620685577393,
-      0.090724878013134,
-      0.11524895578622818
+      7.068239688873291,
+      4.189955234527588,
+      2.0407068729400635,
+      1.8267366886138916,
+      1.7952507734298706,
+      1.7645628452301025,
+      1.595243215560913,
+      1.4900548458099365,
+      1.0821096897125244,
+      0.7350229024887085,
+      0.586064875125885,
+      0.43922391533851624,
+      0.40184348821640015,
+      0.31044527888298035,
+      0.26080775260925293,
+      0.30810627341270447,
+      0.2941185534000397,
+      0.2382466048002243,
+      0.19741371273994446,
+      0.24084995687007904,
+      0.16109716892242432,
+      0.16342215240001678,
+      0.15979401767253876,
+      0.15030500292778015,
+      0.1553926020860672,
+      0.1616949737071991,
+      0.15501153469085693,
+      0.15730805695056915,
+      0.1374218612909317,
+      0.12141235917806625,
+      0.15107305347919464,
+      0.12318699806928635,
+      0.08949179947376251,
+      0.11493054777383804,
+      0.09638842195272446,
+      0.1008308082818985,
+      0.11183486878871918,
+      0.08597197383642197,
+      0.10642959177494049,
+      0.07882422208786011,
+      0.09636445343494415,
+      0.06600546836853027,
+      0.09191834181547165,
+      0.058375708758831024,
+      0.08556245267391205,
+      0.09554620832204819,
+      0.05065993592143059,
+      0.06083018705248833,
+      0.06561274826526642,
+      0.05150684341788292,
+      0.04865796118974686,
+      0.05840981379151344,
+      0.06329210847616196,
+      0.05198041722178459,
+      0.041428614407777786,
+      0.04235445708036423,
+      0.0411263071000576,
+      0.0459698922932148,
+      0.04036566987633705,
+      0.042875390499830246,
+      0.05301470309495926,
+      0.05982867628335953
     ],
     "lr": [
-      1.9600000000000002e-05,
-      3.96e-05,
-      3.9974363901222355e-05,
-      3.9895421087752256e-05,
-      3.976337160140893e-05,
-      3.957856792072718e-05,
-      3.9341503340549716e-05,
-      3.9052810655279495e-05,
-      3.8713260469767256e-05,
-      3.832375914234272e-05,
-      3.788534636548025e-05,
-      3.739919239055685e-05,
-      3.6866594904110466e-05,
-      3.62889755639367e-05,
-      3.566787620427007e-05,
-      3.5004954720179526e-05,
-      3.4301980642163606e-05,
-      3.3560830412758255e-05,
-      3.2783482377765215e-05,
-      3.197201150547096e-05,
-      3.112858384795204e-05,
-      3.0255450759251313e-05,
-      2.9354942885858323e-05,
-      2.842946394553513e-05,
-      2.7481484311093542e-05,
-      2.6513534416250682e-05,
-      2.5528198001164462e-05,
-      2.4528105215678678e-05,
-      2.3515925598687097e-05,
-      2.249436095235672e-05,
-      2.146613813023101e-05,
-      2.043400175846362e-05,
-      1.9400706909611906e-05,
-      1.8369011748545936e-05,
-      1.734167017010322e-05,
-      1.6321424448141397e-05,
-      1.5310997915610664e-05,
-      1.4313087695185074e-05,
-      1.3330357499856637e-05,
-      1.2365430522709616e-05,
-      1.1420882434854255e-05,
-      1.049923451021051e-05,
-      9.602946895493652e-06,
-      8.734412043366293e-06,
-      7.89594832628547e-06,
-      7.089793848091389e-06,
-      6.318100469856501e-06,
-      5.582928065941624e-06,
-      4.886239025591397e-06,
-      4.229893014745887e-06,
-      3.6156420120506306e-06,
-      3.045125632315402e-06,
-      2.5198667499047936e-06,
-      2.0412674337430526e-06,
-      1.610605204783806e-06,
-      1.2290296259345835e-06,
-      8.975592335386451e-07,
-      6.170788186048593e-07,
-      3.883370650428364e-07,
-      2.11944551207528e-07,
-      8.837212008778961e-08,
-      1.79496224893283e-08
+      3.9200000000000004e-05,
+      7.92e-05,
+      7.994872780244471e-05,
+      7.979084217550451e-05,
+      7.952674320281786e-05,
+      7.915713584145437e-05,
+      7.868300668109943e-05,
+      7.810562131055899e-05,
+      7.742652093953451e-05,
+      7.664751828468545e-05,
+      7.57706927309605e-05,
+      7.47983847811137e-05,
+      7.373318980822093e-05,
+      7.25779511278734e-05,
+      7.133575240854014e-05,
+      7.000990944035905e-05,
+      6.860396128432721e-05,
+      6.712166082551651e-05,
+      6.556696475553043e-05,
+      6.394402301094192e-05,
+      6.225716769590408e-05,
+      6.0510901518502626e-05,
+      5.8709885771716645e-05,
+      5.685892789107026e-05,
+      5.4962968622187084e-05,
+      5.3027068832501364e-05,
+      5.1056396002328924e-05,
+      4.9056210431357356e-05,
+      4.703185119737419e-05,
+      4.498872190471344e-05,
+      4.293227626046202e-05,
+      4.086800351692724e-05,
+      3.880141381922381e-05,
+      3.673802349709187e-05,
+      3.468334034020644e-05,
+      3.2642848896282794e-05,
+      3.062199583122133e-05,
+      2.862617539037015e-05,
+      2.6660714999713274e-05,
+      2.4730861045419232e-05,
+      2.284176486970851e-05,
+      2.099846902042102e-05,
+      1.9205893790987304e-05,
+      1.7468824086732586e-05,
+      1.579189665257094e-05,
+      1.4179587696182778e-05,
+      1.2636200939713001e-05,
+      1.1165856131883247e-05,
+      9.772478051182794e-06,
+      8.459786029491775e-06,
+      7.231284024101261e-06,
+      6.090251264630804e-06,
+      5.039733499809587e-06,
+      4.082534867486105e-06,
+      3.221210409567612e-06,
+      2.458059251869167e-06,
+      1.7951184670772902e-06,
+      1.2341576372097185e-06,
+      7.766741300856728e-07,
+      4.23889102415056e-07,
+      1.7674424017557922e-07,
+      3.58992449786566e-08
     ],
     "eval_step": [
       156,
@@ -301,29 +301,29 @@
       20
     ],
     "eval_accuracy": [
-      0.006666666666666667,
-      0.0022222222222222222,
-      0.01,
-      0.014444444444444444,
-      0.051111111111111114,
-      0.13444444444444445,
-      0.26555555555555554,
-      0.39,
-      0.49666666666666665,
-      0.56,
-      0.6044444444444445,
-      0.6322222222222222,
-      0.6344444444444445,
-      0.6344444444444445,
-      0.6866666666666666,
-      0.6888888888888889,
-      0.7033333333333334,
-      0.7033333333333334,
-      0.7144444444444444,
-      0.71
+      0.008888888888888889,
+      0.005555555555555556,
+      0.024444444444444446,
+      0.21333333333333335,
+      0.3522222222222222,
+      0.4311111111111111,
+      0.57,
+      0.5322222222222223,
+      0.6111111111111112,
+      0.6211111111111111,
+      0.6477777777777778,
+      0.7,
+      0.7322222222222222,
+      0.7133333333333334,
+      0.7388888888888889,
+      0.7644444444444445,
+      0.7766666666666666,
+      0.7833333333333333,
+      0.7866666666666666,
+      0.7855555555555556
     ]
   },
-  "final_accuracy": 0.6270833333333333,
+  "final_accuracy": 0.7241666666666666,
   "sft_eval": {
     "config": {
       "ops": "add_sub",
@@ -334,11 +334,11 @@
     },
     "splits": {
       "add_S0": {
-        "full_accuracy": 0.87,
+        "full_accuracy": 0.94,
         "n_examples": 100,
         "per_subtask": {
           "SA": {
-            "accuracy": 0.9785123966942149,
+            "accuracy": 0.9900826446280991,
             "count": 605
           },
           "SS": {
@@ -348,7 +348,7 @@
         }
       },
       "add_S1": {
-        "full_accuracy": 0.98,
+        "full_accuracy": 0.99,
         "n_examples": 100,
         "per_subtask": {
           "SA": {
@@ -364,13 +364,13 @@
             "count": 31
           },
           "UC": {
-            "accuracy": 0.9966216216216216,
+            "accuracy": 1.0,
             "count": 296
           }
         }
       },
       "add_S2": {
-        "full_accuracy": 0.7,
+        "full_accuracy": 0.87,
         "n_examples": 100,
         "per_subtask": {
           "SA": {
@@ -378,15 +378,15 @@
             "count": 163
           },
           "SC": {
-            "accuracy": 0.9692307692307692,
+            "accuracy": 1.0,
             "count": 130
           },
           "SS": {
-            "accuracy": 0.9655172413793104,
+            "accuracy": 0.9885057471264368,
             "count": 87
           },
           "UC": {
-            "accuracy": 0.8669950738916257,
+            "accuracy": 0.9408866995073891,
             "count": 203
           },
           "US": {
@@ -396,7 +396,7 @@
         }
       },
       "add_S3": {
-        "full_accuracy": 0.52,
+        "full_accuracy": 0.69,
         "n_examples": 100,
         "per_subtask": {
           "SA": {
@@ -412,17 +412,17 @@
             "count": 49
           },
           "UC": {
-            "accuracy": 0.7795698924731183,
+            "accuracy": 0.8387096774193549,
             "count": 186
           },
           "US": {
-            "accuracy": 0.8923766816143498,
+            "accuracy": 1.0,
             "count": 223
           }
         }
       },
       "add_S4": {
-        "full_accuracy": 0.58,
+        "full_accuracy": 0.57,
         "n_examples": 100,
         "per_subtask": {
           "SA": {
@@ -438,17 +438,17 @@
             "count": 23
           },
           "UC": {
-            "accuracy": 0.8125,
+            "accuracy": 0.74375,
             "count": 160
           },
           "US": {
-            "accuracy": 0.8273615635179153,
+            "accuracy": 0.9511400651465798,
             "count": 307
           }
         }
       },
       "add_S5": {
-        "full_accuracy": 0.29,
+        "full_accuracy": 0.32,
         "n_examples": 100,
         "per_subtask": {
           "SA": {
@@ -460,17 +460,17 @@
             "count": 100
           },
           "UC": {
-            "accuracy": 0.41,
+            "accuracy": 0.39,
             "count": 100
           },
           "US": {
-            "accuracy": 0.58,
+            "accuracy": 0.7075,
             "count": 400
           }
         }
       },
       "add_S6": {
-        "full_accuracy": 0.56,
+        "full_accuracy": 0.46,
         "n_examples": 100,
         "per_subtask": {
           "SC": {
@@ -478,25 +478,25 @@
             "count": 100
           },
           "UC": {
-            "accuracy": 0.64,
+            "accuracy": 0.55,
             "count": 100
           },
           "US": {
-            "accuracy": 0.688,
+            "accuracy": 0.718,
             "count": 500
           }
         }
       },
       "add_random": {
-        "full_accuracy": 0.89,
+        "full_accuracy": 0.945,
         "n_examples": 200,
         "per_subtask": {
           "SA": {
-            "accuracy": 0.9910514541387024,
+            "accuracy": 0.9932885906040269,
             "count": 447
           },
           "SC": {
-            "accuracy": 0.99375,
+            "accuracy": 1.0,
             "count": 320
           },
           "SS": {
@@ -504,17 +504,17 @@
             "count": 56
           },
           "UC": {
-            "accuracy": 0.9716446124763705,
+            "accuracy": 0.9848771266540642,
             "count": 529
           },
           "US": {
-            "accuracy": 0.9791666666666666,
+            "accuracy": 1.0,
             "count": 48
           }
         }
       },
       "add_C3": {
-        "full_accuracy": 0.76,
+        "full_accuracy": 0.83,
         "n_examples": 100,
         "per_subtask": {
           "SA": {
@@ -526,21 +526,21 @@
             "count": 100
           },
           "UC": {
-            "accuracy": 0.8808290155440415,
+            "accuracy": 0.9119170984455959,
             "count": 193
           },
           "US": {
-            "accuracy": 0.9532710280373832,
+            "accuracy": 1.0,
             "count": 107
           }
         }
       },
       "add_C4": {
-        "full_accuracy": 0.58,
+        "full_accuracy": 0.88,
         "n_examples": 100,
         "per_subtask": {
           "SA": {
-            "accuracy": 1.0,
+            "accuracy": 0.995,
             "count": 200
           },
           "SC": {
@@ -548,17 +548,17 @@
             "count": 100
           },
           "UC": {
-            "accuracy": 0.84375,
+            "accuracy": 0.95703125,
             "count": 256
           },
           "US": {
-            "accuracy": 0.9236111111111112,
+            "accuracy": 1.0,
             "count": 144
           }
         }
       },
       "add_C5": {
-        "full_accuracy": 0.6,
+        "full_accuracy": 0.8,
         "n_examples": 100,
         "per_subtask": {
           "SA": {
@@ -570,17 +570,17 @@
             "count": 100
           },
           "UC": {
-            "accuracy": 0.8823529411764706,
+            "accuracy": 0.934640522875817,
             "count": 306
           },
           "US": {
-            "accuracy": 0.865979381443299,
+            "accuracy": 0.9690721649484536,
             "count": 194
           }
         }
       },
       "add_C6": {
-        "full_accuracy": 0.52,
+        "full_accuracy": 0.78,
         "n_examples": 100,
         "per_subtask": {
           "SC": {
@@ -588,11 +588,11 @@
             "count": 100
           },
           "UC": {
-            "accuracy": 0.8797814207650273,
+            "accuracy": 0.9398907103825137,
             "count": 366
           },
           "US": {
-            "accuracy": 0.9102564102564102,
+            "accuracy": 0.9871794871794872,
             "count": 234
           }
         }
@@ -606,21 +606,21 @@
             "count": 601
           },
           "ME": {
-            "accuracy": 0.98989898989899,
+            "accuracy": 1.0,
             "count": 99
           }
         }
       },
       "sub_M1": {
-        "full_accuracy": 0.93,
+        "full_accuracy": 1.0,
         "n_examples": 100,
         "per_subtask": {
           "MD": {
-            "accuracy": 0.989247311827957,
+            "accuracy": 1.0,
             "count": 279
           },
           "MB": {
-            "accuracy": 0.9793103448275862,
+            "accuracy": 1.0,
             "count": 145
           },
           "ME": {
@@ -628,21 +628,21 @@
             "count": 24
           },
           "UB": {
-            "accuracy": 0.996031746031746,
+            "accuracy": 1.0,
             "count": 252
           }
         }
       },
       "sub_M2": {
-        "full_accuracy": 0.72,
+        "full_accuracy": 0.9,
         "n_examples": 100,
         "per_subtask": {
           "MD": {
-            "accuracy": 0.9859154929577465,
+            "accuracy": 1.0,
             "count": 213
           },
           "MB": {
-            "accuracy": 1.0,
+            "accuracy": 0.9911504424778761,
             "count": 113
           },
           "ME": {
@@ -650,7 +650,7 @@
             "count": 85
           },
           "UB": {
-            "accuracy": 0.861878453038674,
+            "accuracy": 0.9447513812154696,
             "count": 181
           },
           "UD": {
@@ -660,7 +660,7 @@
         }
       },
       "sub_M3": {
-        "full_accuracy": 0.26,
+        "full_accuracy": 0.47,
         "n_examples": 100,
         "per_subtask": {
           "MD": {
@@ -668,7 +668,7 @@
             "count": 179
           },
           "MB": {
-            "accuracy": 0.9805825242718447,
+            "accuracy": 0.9902912621359223,
             "count": 103
           },
           "ME": {
@@ -676,17 +676,17 @@
             "count": 56
           },
           "UB": {
-            "accuracy": 0.46308724832214765,
+            "accuracy": 0.6375838926174496,
             "count": 149
           },
           "UD": {
-            "accuracy": 0.9859154929577465,
+            "accuracy": 1.0,
             "count": 213
           }
         }
       },
       "sub_M4": {
-        "full_accuracy": 0.0,
+        "full_accuracy": 0.08,
         "n_examples": 100,
         "per_subtask": {
           "MD": {
@@ -698,17 +698,17 @@
             "count": 100
           },
           "UB": {
-            "accuracy": 0.26,
+            "accuracy": 0.3,
             "count": 100
           },
           "UD": {
-            "accuracy": 0.55,
+            "accuracy": 0.7433333333333333,
             "count": 300
           }
         }
       },
       "sub_M5": {
-        "full_accuracy": 0.0,
+        "full_accuracy": 0.02,
         "n_examples": 100,
         "per_subtask": {
           "MD": {
@@ -720,25 +720,25 @@
             "count": 100
           },
           "UB": {
-            "accuracy": 0.25,
+            "accuracy": 0.38,
             "count": 100
           },
           "UD": {
-            "accuracy": 0.3975,
+            "accuracy": 0.5575,
             "count": 400
           }
         }
       },
       "sub_random": {
-        "full_accuracy": 0.905,
+        "full_accuracy": 0.96,
         "n_examples": 200,
         "per_subtask": {
           "MD": {
-            "accuracy": 0.995,
+            "accuracy": 0.9983333333333333,
             "count": 600
           },
           "MB": {
-            "accuracy": 0.9775280898876404,
+            "accuracy": 0.9925093632958801,
             "count": 267
           },
           "ME": {
@@ -746,7 +746,7 @@
             "count": 53
           },
           "UB": {
-            "accuracy": 0.9749430523917996,
+            "accuracy": 0.9886104783599089,
             "count": 439
           },
           "UD": {
@@ -756,11 +756,11 @@
         }
       },
       "sub_B3": {
-        "full_accuracy": 0.67,
+        "full_accuracy": 0.75,
         "n_examples": 100,
         "per_subtask": {
           "MD": {
-            "accuracy": 0.98,
+            "accuracy": 1.0,
             "count": 300
           },
           "MB": {
@@ -768,7 +768,7 @@
             "count": 100
           },
           "UB": {
-            "accuracy": 0.8527918781725888,
+            "accuracy": 0.8730964467005076,
             "count": 197
           },
           "UD": {
@@ -778,11 +778,11 @@
         }
       },
       "sub_B4": {
-        "full_accuracy": 0.58,
+        "full_accuracy": 0.68,
         "n_examples": 100,
         "per_subtask": {
           "MD": {
-            "accuracy": 0.985,
+            "accuracy": 1.0,
             "count": 200
           },
           "MB": {
@@ -790,17 +790,17 @@
             "count": 100
           },
           "UB": {
-            "accuracy": 0.854251012145749,
+            "accuracy": 0.8866396761133604,
             "count": 247
           },
           "UD": {
-            "accuracy": 0.8758169934640523,
+            "accuracy": 0.934640522875817,
             "count": 153
           }
         }
       },
       "sub_B5": {
-        "full_accuracy": 0.42,
+        "full_accuracy": 0.59,
         "n_examples": 100,
         "per_subtask": {
           "MD": {
@@ -812,18 +812,18 @@
             "count": 100
           },
           "UB": {
-            "accuracy": 0.8154362416107382,
+            "accuracy": 0.87248322147651,
             "count": 298
           },
           "UD": {
-            "accuracy": 0.8564356435643564,
+            "accuracy": 0.9158415841584159,
             "count": 202
           }
         }
       }
     },
     "summary": {
-      "overall_accuracy": 0.6270833333333333,
+      "overall_accuracy": 0.7241666666666666,
       "total_examples": 2400,
       "n_splits": 22
     }

add_sub_baseline_10K/model.safetensors

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:83183edc6b1fa903ee491f2ac4fae76e8abb9d76035b9ce79aec6265c267c8ec
+oid sha256:99090d5772338ac21e3d890eee50dfa2204eeafe4c53123e495571b22267bec9
 size 650266922

add_sub_baseline_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": 162490082,
   "run_name": "add_sub_baseline_10K",
-  "git_commit": "78d46f8665a87f4b44bd5894bd34f393f2dea51f",
-  "timestamp": "2026-04-12T08:59:11.045620+00:00",
+  "git_commit": "8d5ee5420119746ef4e2c87570eb250c9718f643",
+  "timestamp": "2026-04-12T22:03:24.880179+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": "sft",
-  "wandb_run_id": "7dqmkyo7",
-  "wandb_url": "https://wandb.ai/nlp_and_interpretability/sorl-arithmetic/runs/7dqmkyo7",
-  "final_accuracy": 0.6270833333333333,
-  "sft_accuracy": 0.6270833333333333,
+  "wandb_run_id": "j0499k85",
+  "wandb_url": "https://wandb.ai/nlp_and_interpretability/sorl-arithmetic/runs/j0499k85",
+  "final_accuracy": 0.7241666666666666,
+  "sft_accuracy": 0.7241666666666666,
   "eval_method": "ArithmeticEvaluator"
 }