Upload add_sub_baseline_10K

add_sub_baseline_10K/metrics.json

@@ -65,132 +65,132 @@
       3100
     ],
     "loss": [
-      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
+      7.406970977783203,
+      4.287904262542725,
+      2.028409719467163,
+      1.9122778177261353,
+      1.7641465663909912,
+      1.6944198608398438,
+      1.5549181699752808,
+      1.4273613691329956,
+      1.1411476135253906,
+      0.7707713842391968,
+      0.652624249458313,
+      0.5090309977531433,
+      0.4863796830177307,
+      0.4217025935649872,
+      0.3141794204711914,
+      0.26072242856025696,
+      0.20925267040729523,
+      0.1722976416349411,
+      0.16701845824718475,
+      0.14609448611736298,
+      0.15161116421222687,
+      0.11793585866689682,
+      0.12784767150878906,
+      0.17903172969818115,
+      0.07992793619632721,
+      0.09357694536447525,
+      0.07849029451608658,
+      0.08744285255670547,
+      0.047625500708818436,
+      0.06857192516326904,
+      0.058801162987947464,
+      0.08054264634847641,
+      0.0829831138253212,
+      0.08300244808197021,
+      0.04127458110451698,
+      0.045694876462221146,
+      0.0545954629778862,
+      0.04024777188897133,
+      0.07766951620578766,
+      0.024591432884335518,
+      0.06888258457183838,
+      0.03272942453622818,
+      0.05081992596387863,
+      0.07547096163034439,
+      0.04898810759186745,
+      0.04332876205444336,
+      0.018458085134625435,
+      0.03549834340810776,
+      0.020787600427865982,
+      0.01855045184493065,
+      0.018782520666718483,
+      0.013612723909318447,
+      0.0217971820384264,
+      0.01920527033507824,
+      0.01275918073952198,
+      0.023383263498544693,
+      0.017817409709095955,
+      0.01498924009501934,
+      0.021258754655718803,
+      0.022163305431604385,
+      0.012838104739785194,
+      0.015554169192910194
     ],
     "base_loss": [
-      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
+      7.406970977783203,
+      4.287904262542725,
+      2.028409719467163,
+      1.9122778177261353,
+      1.7641465663909912,
+      1.6944198608398438,
+      1.5549181699752808,
+      1.4273613691329956,
+      1.1411476135253906,
+      0.7707713842391968,
+      0.652624249458313,
+      0.5090309977531433,
+      0.4863796830177307,
+      0.4217025935649872,
+      0.3141794204711914,
+      0.26072242856025696,
+      0.20925267040729523,
+      0.1722976416349411,
+      0.16701845824718475,
+      0.14609448611736298,
+      0.15161116421222687,
+      0.11793585866689682,
+      0.12784767150878906,
+      0.17903172969818115,
+      0.07992793619632721,
+      0.09357694536447525,
+      0.07849029451608658,
+      0.08744285255670547,
+      0.047625500708818436,
+      0.06857192516326904,
+      0.058801162987947464,
+      0.08054264634847641,
+      0.0829831138253212,
+      0.08300244808197021,
+      0.04127458110451698,
+      0.045694876462221146,
+      0.0545954629778862,
+      0.04024777188897133,
+      0.07766951620578766,
+      0.024591432884335518,
+      0.06888258457183838,
+      0.03272942453622818,
+      0.05081992596387863,
+      0.07547096163034439,
+      0.04898810759186745,
+      0.04332876205444336,
+      0.018458085134625435,
+      0.03549834340810776,
+      0.020787600427865982,
+      0.01855045184493065,
+      0.018782520666718483,
+      0.013612723909318447,
+      0.0217971820384264,
+      0.01920527033507824,
+      0.01275918073952198,
+      0.023383263498544693,
+      0.017817409709095955,
+      0.01498924009501934,
+      0.021258754655718803,
+      0.022163305431604385,
+      0.012838104739785194,
+      0.015554169192910194
     ],
     "lr": [
       3.9200000000000004e-05,
@@ -301,595 +301,595 @@
       20
     ],
     "eval_accuracy": [
-      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
+      0.005263157894736842,
+      0.005263157894736842,
+      0.023157894736842106,
+      0.10421052631578948,
+      0.4073684210526316,
+      0.6189473684210526,
+      0.6663157894736842,
+      0.7210526315789474,
+      0.7315789473684211,
+      0.76,
+      0.7884210526315789,
+      0.783157894736842,
+      0.7631578947368421,
+      0.8126315789473684,
+      0.8147368421052632,
+      0.8126315789473684,
+      0.8105263157894737,
+      0.8189473684210526,
+      0.82,
+      0.8094736842105263
     ]
   },
-  "final_accuracy": 0.7241666666666666,
+  "final_accuracy": 0.76,
   "sft_eval": {
     "config": {
       "ops": "add_sub",
       "K": null,
       "mode": "sft",
       "n_digits": 6,
-      "n_per_split": 50
+      "n_per_split": 100
     },
     "splits": {
       "add_S0": {
-        "full_accuracy": 0.94,
-        "digit_accuracy": 0.9914285714285714,
-        "n_examples": 50,
+        "full_accuracy": 0.97,
+        "digit_accuracy": 0.9957142857142857,
+        "n_examples": 100,
         "per_subtask": {
           "SA": {
-            "accuracy": 0.9898305084745763,
-            "count": 295
+            "accuracy": 0.9950413223140496,
+            "count": 605
           },
           "SS": {
             "accuracy": 1.0,
-            "count": 55
+            "count": 95
           }
         }
       },
       "add_S1": {
-        "full_accuracy": 0.98,
-        "digit_accuracy": 0.9971428571428571,
-        "n_examples": 50,
+        "full_accuracy": 1.0,
+        "digit_accuracy": 1.0,
+        "n_examples": 100,
         "per_subtask": {
           "SA": {
             "accuracy": 1.0,
-            "count": 126
+            "count": 204
           },
           "SC": {
             "accuracy": 1.0,
-            "count": 79
+            "count": 169
           },
           "SS": {
             "accuracy": 1.0,
-            "count": 21
+            "count": 31
           },
           "UC": {
-            "accuracy": 0.9919354838709677,
-            "count": 124
+            "accuracy": 1.0,
+            "count": 296
           }
         }
       },
       "add_S2": {
-        "full_accuracy": 0.9,
-        "digit_accuracy": 0.9828571428571429,
-        "n_examples": 50,
+        "full_accuracy": 0.97,
+        "digit_accuracy": 0.9957142857142857,
+        "n_examples": 100,
         "per_subtask": {
           "SA": {
-            "accuracy": 0.9866666666666667,
-            "count": 75
+            "accuracy": 1.0,
+            "count": 163
           },
           "SC": {
-            "accuracy": 1.0,
-            "count": 62
+            "accuracy": 0.9846153846153847,
+            "count": 130
           },
           "SS": {
-            "accuracy": 1.0,
-            "count": 39
+            "accuracy": 0.9885057471264368,
+            "count": 87
           },
           "UC": {
-            "accuracy": 0.954954954954955,
-            "count": 111
+            "accuracy": 1.0,
+            "count": 203
           },
           "US": {
             "accuracy": 1.0,
-            "count": 63
+            "count": 117
           }
         }
       },
       "add_S3": {
-        "full_accuracy": 0.7,
-        "digit_accuracy": 0.9571428571428572,
-        "n_examples": 50,
+        "full_accuracy": 0.67,
+        "digit_accuracy": 0.9528571428571428,
+        "n_examples": 100,
         "per_subtask": {
           "SA": {
             "accuracy": 1.0,
-            "count": 60
+            "count": 121
           },
           "SC": {
-            "accuracy": 1.0,
-            "count": 57
+            "accuracy": 0.9834710743801653,
+            "count": 121
           },
           "SS": {
             "accuracy": 1.0,
-            "count": 19
+            "count": 49
           },
           "UC": {
-            "accuracy": 0.8557692307692307,
-            "count": 104
+            "accuracy": 0.8333333333333334,
+            "count": 186
           },
           "US": {
             "accuracy": 1.0,
-            "count": 110
+            "count": 223
           }
         }
       },
       "add_S4": {
-        "full_accuracy": 0.64,
-        "digit_accuracy": 0.9342857142857143,
-        "n_examples": 50,
+        "full_accuracy": 0.6,
+        "digit_accuracy": 0.9171428571428571,
+        "n_examples": 100,
         "per_subtask": {
           "SA": {
             "accuracy": 1.0,
-            "count": 48
+            "count": 104
           },
           "SC": {
             "accuracy": 1.0,
-            "count": 52
+            "count": 106
           },
           "SS": {
             "accuracy": 1.0,
-            "count": 7
+            "count": 23
           },
           "UC": {
-            "accuracy": 0.8202247191011236,
-            "count": 89
+            "accuracy": 0.8,
+            "count": 160
           },
           "US": {
-            "accuracy": 0.9545454545454546,
-            "count": 154
+            "accuracy": 0.9153094462540716,
+            "count": 307
           }
         }
       },
       "add_S5": {
-        "full_accuracy": 0.3,
-        "digit_accuracy": 0.7342857142857143,
-        "n_examples": 50,
+        "full_accuracy": 0.5,
+        "digit_accuracy": 0.8371428571428572,
+        "n_examples": 100,
         "per_subtask": {
           "SA": {
             "accuracy": 1.0,
-            "count": 50
+            "count": 100
           },
           "SC": {
             "accuracy": 1.0,
-            "count": 50
+            "count": 100
           },
           "UC": {
-            "accuracy": 0.38,
-            "count": 50
+            "accuracy": 0.65,
+            "count": 100
           },
           "US": {
-            "accuracy": 0.69,
-            "count": 200
+            "accuracy": 0.8025,
+            "count": 400
           }
         }
       },
       "add_S6": {
-        "full_accuracy": 0.36,
-        "digit_accuracy": 0.6685714285714286,
-        "n_examples": 50,
+        "full_accuracy": 0.5,
+        "digit_accuracy": 0.8142857142857143,
+        "n_examples": 100,
         "per_subtask": {
           "SC": {
             "accuracy": 1.0,
-            "count": 50
+            "count": 100
           },
           "UC": {
-            "accuracy": 0.46,
-            "count": 50
+            "accuracy": 0.76,
+            "count": 100
           },
           "US": {
-            "accuracy": 0.644,
-            "count": 250
+            "accuracy": 0.788,
+            "count": 500
           }
         }
       },
       "add_random": {
-        "full_accuracy": 0.96,
-        "digit_accuracy": 0.9935714285714285,
+        "full_accuracy": 0.97,
+        "digit_accuracy": 0.9957142857142857,
         "n_examples": 200,
         "per_subtask": {
           "SA": {
-            "accuracy": 0.9953596287703016,
-            "count": 431
+            "accuracy": 0.9955257270693513,
+            "count": 447
           },
           "SC": {
-            "accuracy": 1.0,
-            "count": 316
+            "accuracy": 0.99375,
+            "count": 320
           },
           "SS": {
             "accuracy": 1.0,
-            "count": 39
+            "count": 56
           },
           "UC": {
-            "accuracy": 0.9875,
-            "count": 560
+            "accuracy": 0.996219281663516,
+            "count": 529
           },
           "US": {
             "accuracy": 1.0,
-            "count": 54
+            "count": 48
           }
         }
       },
       "add_C1": {
-        "full_accuracy": 1.0,
-        "digit_accuracy": 1.0,
-        "n_examples": 50,
+        "full_accuracy": 0.99,
+        "digit_accuracy": 0.9985714285714286,
+        "n_examples": 100,
         "per_subtask": {
           "SA": {
-            "accuracy": 1.0,
-            "count": 250
+            "accuracy": 0.998,
+            "count": 500
           },
           "SC": {
             "accuracy": 1.0,
-            "count": 50
+            "count": 100
           },
           "UC": {
             "accuracy": 1.0,
-            "count": 50
+            "count": 100
           }
         }
       },
       "add_C2": {
-        "full_accuracy": 0.98,
-        "digit_accuracy": 0.9971428571428571,
-        "n_examples": 50,
+        "full_accuracy": 0.93,
+        "digit_accuracy": 0.99,
+        "n_examples": 100,
         "per_subtask": {
           "SA": {
-            "accuracy": 1.0,
-            "count": 200
+            "accuracy": 0.995,
+            "count": 400
           },
           "SC": {
             "accuracy": 1.0,
-            "count": 50
+            "count": 100
           },
           "UC": {
-            "accuracy": 0.9879518072289156,
-            "count": 83
+            "accuracy": 0.9743589743589743,
+            "count": 156
           },
           "US": {
-            "accuracy": 1.0,
-            "count": 17
+            "accuracy": 0.9772727272727273,
+            "count": 44
           }
         }
       },
       "add_C3": {
-        "full_accuracy": 0.78,
-        "digit_accuracy": 0.9685714285714285,
-        "n_examples": 50,
+        "full_accuracy": 0.85,
+        "digit_accuracy": 0.9785714285714285,
+        "n_examples": 100,
         "per_subtask": {
           "SA": {
-            "accuracy": 0.9933333333333333,
-            "count": 150
+            "accuracy": 1.0,
+            "count": 300
           },
           "SC": {
             "accuracy": 1.0,
-            "count": 50
+            "count": 100
           },
           "UC": {
-            "accuracy": 0.9,
-            "count": 100
+            "accuracy": 0.9246231155778895,
+            "count": 199
           },
           "US": {
             "accuracy": 1.0,
-            "count": 50
+            "count": 101
           }
         }
       },
       "add_C4": {
-        "full_accuracy": 0.88,
-        "digit_accuracy": 0.9828571428571429,
-        "n_examples": 50,
+        "full_accuracy": 0.78,
+        "digit_accuracy": 0.9657142857142857,
+        "n_examples": 100,
         "per_subtask": {
           "SA": {
             "accuracy": 1.0,
-            "count": 100
+            "count": 200
           },
           "SC": {
             "accuracy": 1.0,
-            "count": 50
+            "count": 100
           },
           "UC": {
-            "accuracy": 0.9545454545454546,
-            "count": 132
+            "accuracy": 0.9242424242424242,
+            "count": 264
           },
           "US": {
-            "accuracy": 1.0,
-            "count": 68
+            "accuracy": 0.9705882352941176,
+            "count": 136
           }
         }
       },
       "add_C5": {
-        "full_accuracy": 0.76,
-        "digit_accuracy": 0.9571428571428572,
-        "n_examples": 50,
+        "full_accuracy": 0.79,
+        "digit_accuracy": 0.9614285714285714,
+        "n_examples": 100,
         "per_subtask": {
           "SA": {
             "accuracy": 1.0,
-            "count": 50
+            "count": 100
           },
           "SC": {
             "accuracy": 1.0,
-            "count": 50
+            "count": 100
           },
           "UC": {
-            "accuracy": 0.9178082191780822,
-            "count": 146
+            "accuracy": 0.9354838709677419,
+            "count": 310
           },
           "US": {
-            "accuracy": 0.9711538461538461,
-            "count": 104
+            "accuracy": 0.9631578947368421,
+            "count": 190
           }
         }
       },
       "add_C6": {
-        "full_accuracy": 0.92,
-        "digit_accuracy": 0.9885714285714285,
-        "n_examples": 50,
+        "full_accuracy": 0.74,
+        "digit_accuracy": 0.9542857142857143,
+        "n_examples": 100,
         "per_subtask": {
           "SC": {
             "accuracy": 1.0,
-            "count": 50
+            "count": 100
           },
           "UC": {
-            "accuracy": 0.9788359788359788,
-            "count": 189
+            "accuracy": 0.9459459459459459,
+            "count": 370
           },
           "US": {
-            "accuracy": 1.0,
-            "count": 111
+            "accuracy": 0.9478260869565217,
+            "count": 230
           }
         }
       },
       "sub_M0": {
-        "full_accuracy": 1.0,
-        "digit_accuracy": 1.0,
-        "n_examples": 50,
+        "full_accuracy": 0.97,
+        "digit_accuracy": 0.9957142857142857,
+        "n_examples": 100,
         "per_subtask": {
           "MD": {
-            "accuracy": 1.0,
-            "count": 303
+            "accuracy": 0.9951219512195122,
+            "count": 615
           },
           "ME": {
             "accuracy": 1.0,
-            "count": 47
+            "count": 85
           }
         }
       },
       "sub_M1": {
-        "full_accuracy": 0.92,
-        "digit_accuracy": 0.9885714285714285,
-        "n_examples": 50,
+        "full_accuracy": 0.97,
+        "digit_accuracy": 0.9957142857142857,
+        "n_examples": 100,
         "per_subtask": {
           "MD": {
-            "accuracy": 0.9858156028368794,
-            "count": 141
+            "accuracy": 0.9931506849315068,
+            "count": 292
           },
           "MB": {
-            "accuracy": 0.9722222222222222,
-            "count": 72
+            "accuracy": 0.9930555555555556,
+            "count": 144
           },
           "ME": {
             "accuracy": 1.0,
-            "count": 18
+            "count": 25
           },
           "UB": {
             "accuracy": 1.0,
-            "count": 119
+            "count": 239
           }
         }
       },
       "sub_M2": {
-        "full_accuracy": 0.76,
-        "digit_accuracy": 0.9657142857142857,
-        "n_examples": 50,
+        "full_accuracy": 0.94,
+        "digit_accuracy": 0.9914285714285714,
+        "n_examples": 100,
         "per_subtask": {
           "MD": {
             "accuracy": 1.0,
-            "count": 112
+            "count": 211
           },
           "MB": {
-            "accuracy": 0.9056603773584906,
-            "count": 53
+            "accuracy": 1.0,
+            "count": 115
           },
           "ME": {
-            "accuracy": 1.0,
-            "count": 47
+            "accuracy": 0.9882352941176471,
+            "count": 85
           },
           "UB": {
-            "accuracy": 0.9176470588235294,
-            "count": 85
+            "accuracy": 0.9723756906077348,
+            "count": 181
           },
           "UD": {
             "accuracy": 1.0,
-            "count": 53
+            "count": 108
           }
         }
       },
       "sub_M3": {
-        "full_accuracy": 0.42,
-        "digit_accuracy": 0.9171428571428571,
-        "n_examples": 50,
+        "full_accuracy": 0.38,
+        "digit_accuracy": 0.9028571428571428,
+        "n_examples": 100,
         "per_subtask": {
           "MD": {
             "accuracy": 1.0,
-            "count": 97
+            "count": 179
           },
           "MB": {
             "accuracy": 1.0,
-            "count": 51
+            "count": 103
           },
           "ME": {
             "accuracy": 1.0,
-            "count": 27
+            "count": 56
           },
           "UB": {
-            "accuracy": 0.6081081081081081,
-            "count": 74
+            "accuracy": 0.610738255033557,
+            "count": 149
           },
           "UD": {
-            "accuracy": 1.0,
-            "count": 101
+            "accuracy": 0.9530516431924883,
+            "count": 213
           }
         }
       },
       "sub_M4": {
-        "full_accuracy": 0.1,
-        "digit_accuracy": 0.8228571428571428,
-        "n_examples": 50,
+        "full_accuracy": 0.18,
+        "digit_accuracy": 0.7942857142857143,
+        "n_examples": 100,
         "per_subtask": {
           "MD": {
             "accuracy": 1.0,
-            "count": 100
+            "count": 200
           },
           "MB": {
             "accuracy": 1.0,
-            "count": 50
+            "count": 100
           },
           "UB": {
-            "accuracy": 0.44,
-            "count": 50
+            "accuracy": 0.47,
+            "count": 100
           },
           "UD": {
-            "accuracy": 0.7733333333333333,
-            "count": 150
+            "accuracy": 0.6966666666666667,
+            "count": 300
           }
         }
       },
       "sub_M5": {
-        "full_accuracy": 0.06,
-        "digit_accuracy": 0.6885714285714286,
-        "n_examples": 50,
+        "full_accuracy": 0.13,
+        "digit_accuracy": 0.6685714285714286,
+        "n_examples": 100,
         "per_subtask": {
           "MD": {
             "accuracy": 1.0,
-            "count": 50
+            "count": 100
           },
           "MB": {
             "accuracy": 1.0,
-            "count": 50
+            "count": 100
           },
           "UB": {
-            "accuracy": 0.38,
-            "count": 50
+            "accuracy": 0.46,
+            "count": 100
           },
           "UD": {
-            "accuracy": 0.61,
-            "count": 200
+            "accuracy": 0.555,
+            "count": 400
           }
         }
       },
       "sub_random": {
-        "full_accuracy": 0.96,
-        "digit_accuracy": 0.9942857142857143,
+        "full_accuracy": 0.95,
+        "digit_accuracy": 0.9928571428571429,
         "n_examples": 200,
         "per_subtask": {
           "MD": {
-            "accuracy": 0.9964912280701754,
-            "count": 570
+            "accuracy": 0.9883333333333333,
+            "count": 600
           },
           "MB": {
-            "accuracy": 0.9819494584837545,
-            "count": 277
+            "accuracy": 0.9962546816479401,
+            "count": 267
           },
           "ME": {
             "accuracy": 1.0,
             "count": 53
           },
           "UB": {
-            "accuracy": 0.9978768577494692,
-            "count": 471
+            "accuracy": 0.9954441913439636,
+            "count": 439
           },
           "UD": {
             "accuracy": 1.0,
-            "count": 29
+            "count": 41
           }
         }
       },
       "sub_B3": {
-        "full_accuracy": 0.82,
-        "digit_accuracy": 0.9742857142857143,
-        "n_examples": 50,
+        "full_accuracy": 0.79,
+        "digit_accuracy": 0.9685714285714285,
+        "n_examples": 100,
         "per_subtask": {
           "MD": {
-            "accuracy": 1.0,
-            "count": 150
+            "accuracy": 0.9966666666666667,
+            "count": 300
           },
           "MB": {
             "accuracy": 1.0,
-            "count": 50
+            "count": 100
           },
           "UB": {
-            "accuracy": 0.9108910891089109,
-            "count": 101
+            "accuracy": 0.8984771573604061,
+            "count": 197
           },
           "UD": {
-            "accuracy": 1.0,
-            "count": 49
+            "accuracy": 0.9902912621359223,
+            "count": 103
           }
         }
       },
       "sub_B4": {
-        "full_accuracy": 0.62,
-        "digit_accuracy": 0.9371428571428572,
-        "n_examples": 50,
+        "full_accuracy": 0.61,
+        "digit_accuracy": 0.9342857142857143,
+        "n_examples": 100,
         "per_subtask": {
           "MD": {
-            "accuracy": 1.0,
-            "count": 100
+            "accuracy": 0.98,
+            "count": 200
           },
           "MB": {
             "accuracy": 1.0,
-            "count": 50
+            "count": 100
           },
           "UB": {
-            "accuracy": 0.859504132231405,
-            "count": 121
+            "accuracy": 0.8744939271255061,
+            "count": 247
           },
           "UD": {
-            "accuracy": 0.9367088607594937,
-            "count": 79
+            "accuracy": 0.9281045751633987,
+            "count": 153
           }
         }
       },
       "sub_B5": {
-        "full_accuracy": 0.46,
-        "digit_accuracy": 0.9142857142857143,
-        "n_examples": 50,
+        "full_accuracy": 0.66,
+        "digit_accuracy": 0.9342857142857143,
+        "n_examples": 100,
         "per_subtask": {
           "MD": {
             "accuracy": 1.0,
-            "count": 50
+            "count": 100
           },
           "MB": {
             "accuracy": 1.0,
-            "count": 50
+            "count": 100
           },
           "UB": {
-            "accuracy": 0.8289473684210527,
-            "count": 152
+            "accuracy": 0.8993288590604027,
+            "count": 298
           },
           "UD": {
-            "accuracy": 0.9591836734693877,
-            "count": 98
+            "accuracy": 0.9207920792079208,
+            "count": 202
           }
         }
       }
     },
     "summary": {
-      "overall_accuracy": 0.766,
-      "digit_accuracy": 0.9439047619047619,
-      "total_examples": 1500,
+      "overall_accuracy": 0.76,
+      "digit_accuracy": 0.9431318681318681,
+      "total_examples": 2600,
       "n_splits": 24
     }
   }

add_sub_baseline_10K/model.safetensors

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:99090d5772338ac21e3d890eee50dfa2204eeafe4c53123e495571b22267bec9
+oid sha256:9024dda0f3d40530abfd72b50df46a8e6a8a4dde5fed1a429b758171421411ab
 size 650266922

add_sub_baseline_10K/train_config.json

@@ -36,7 +36,7 @@
   "eval_every": 156,
   "save_every": 999999,
   "eval_samples": 100,
-  "output_dir": "ckpt/sweep/add_sub_baseline_10K",
+  "output_dir": "ckpt/sweep/as_baseline_10K_2L3H510d",
   "eval_K": 4,
   "alpha_traj": 0.0,
   "corrupt_method": "shuffle",
@@ -69,16 +69,20 @@
   "no_wandb": false,
   "n_params": 162490082,
   "run_name": "add_sub_baseline_10K",
-  "git_commit": "8d5ee5420119746ef4e2c87570eb250c9718f643",
-  "timestamp": "2026-04-12T22:03:24.880179+00:00",
+  "git_commit": "f835493c19eb98267697007042c9d440cad2afbb",
+  "timestamp": "2026-04-16T01:10:30.945547+00:00",
   "tokenizer": "Qwen/Qwen3-0.6B",
   "dataset_repo": "thoughtworks/arithmetic-sorl-data",
   "dataset_config": "add_sub_6digit",
+  "train_dataset": "fixed_train/train_10K_seed42.pt",
   "model_repo": "thoughtworks/arithmetic-sorl",
   "trainer_version": "sft",
-  "wandb_run_id": "j0499k85",
-  "wandb_url": "https://wandb.ai/nlp_and_interpretability/sorl-arithmetic/runs/j0499k85",
-  "final_accuracy": 0.7241666666666666,
-  "sft_accuracy": 0.7241666666666666,
+  "wandb_run_id": "dvsplayi",
+  "wandb_url": "https://wandb.ai/nlp_and_interpretability/sorl-arithmetic/runs/dvsplayi",
+  "eval_final_dataset": "eval_sets/eval_add_sub_6d_N100_seed42.json",
+  "eval_epoch_dataset": "eval_sets/eval_add_sub_6d_N25_seed42.json",
+  "eval_hf_repo": "thoughtworks/arithmetic-sorl-data",
+  "final_accuracy": 0.76,
+  "sft_accuracy": 0.76,
   "eval_method": "ArithmeticEvaluator"
 }