Upload add_sub_sorl_v1_abs10_25K

add_sub_sorl_v1_abs10_25K/config.json

@@ -0,0 +1,37 @@
+{
+  "architectures": [
+    "SorlModelWrapper"
+  ],
+  "attention_bias": false,
+  "attention_dropout": 0.0,
+  "bos_token_id": null,
+  "dtype": "float32",
+  "eos_token_id": null,
+  "head_dim": 128,
+  "hidden_act": "silu",
+  "hidden_size": 510,
+  "initializer_range": 0.02,
+  "intermediate_size": 2040,
+  "layer_types": [
+    "full_attention",
+    "full_attention"
+  ],
+  "max_position_embeddings": 128,
+  "max_window_layers": 28,
+  "model_type": "qwen3",
+  "num_attention_heads": 3,
+  "num_hidden_layers": 2,
+  "num_key_value_heads": 3,
+  "pad_token_id": null,
+  "rms_norm_eps": 1e-06,
+  "rope_parameters": {
+    "rope_theta": 10000.0,
+    "rope_type": "default"
+  },
+  "sliding_window": null,
+  "tie_word_embeddings": false,
+  "transformers_version": "5.5.0",
+  "use_cache": true,
+  "use_sliding_window": false,
+  "vocab_size": 151654
+}

add_sub_sorl_v1_abs10_25K/generation_config.json

@@ -0,0 +1,7 @@
+{
+  "_from_model_config": true,
+  "output_attentions": false,
+  "output_hidden_states": false,
+  "transformers_version": "5.5.0",
+  "use_cache": true
+}

add_sub_sorl_v1_abs10_25K/metrics.json

@@ -0,0 +1,1617 @@
+{
+  "history": {
+    "step": [
+      50,
+      100,
+      150,
+      200,
+      250,
+      300,
+      350,
+      441,
+      491,
+      541,
+      591,
+      641,
+      691,
+      741,
+      832,
+      882,
+      932,
+      982,
+      1032,
+      1082,
+      1132,
+      1223,
+      1273,
+      1323,
+      1373,
+      1423,
+      1473,
+      1523,
+      1614,
+      1664,
+      1714,
+      1764,
+      1814,
+      1864,
+      1914,
+      2005,
+      2055,
+      2105,
+      2155,
+      2205,
+      2255,
+      2305,
+      2396,
+      2446,
+      2496,
+      2546,
+      2596,
+      2646,
+      2696,
+      2787,
+      2837,
+      2887,
+      2937,
+      2987,
+      3037,
+      3087,
+      3178,
+      3228,
+      3278,
+      3328,
+      3378,
+      3428,
+      3478,
+      3569,
+      3619,
+      3669,
+      3719,
+      3769,
+      3819,
+      3869
+    ],
+    "loss": [
+      14.256978034973145,
+      8.812192916870117,
+      5.189567565917969,
+      3.4612081050872803,
+      2.9352636337280273,
+      3.010927438735962,
+      2.87626314163208,
+      2.229984998703003,
+      1.9719955921173096,
+      1.5822030305862427,
+      1.6735219955444336,
+      1.3240095376968384,
+      1.6034573316574097,
+      1.584771752357483,
+      1.3375678062438965,
+      1.5081782341003418,
+      1.2838478088378906,
+      1.442265510559082,
+      1.8373520374298096,
+      1.5873284339904785,
+      1.0017588138580322,
+      0.5844299793243408,
+      0.4599916934967041,
+      0.4710693061351776,
+      -0.016015514731407166,
+      -0.03885940462350845,
+      -1.8101730346679688,
+      -6.263668537139893,
+      -8.383745193481445,
+      -8.508172035217285,
+      -9.05787181854248,
+      -9.528923988342285,
+      -10.230069160461426,
+      -10.578496932983398,
+      -10.426430702209473,
+      -10.69153118133545,
+      -11.126657485961914,
+      -10.420424461364746,
+      -10.6273775100708,
+      -8.421341896057129,
+      -7.00389289855957,
+      -6.054770469665527,
+      -3.3519654273986816,
+      -2.9612250328063965,
+      -2.2057387828826904,
+      -1.278799057006836,
+      -1.6747382879257202,
+      -0.8737431168556213,
+      -0.8648754358291626,
+      -0.9642059803009033,
+      -1.3165982961654663,
+      -0.8842989802360535,
+      -0.5905858278274536,
+      -0.7463504076004028,
+      -1.1404083967208862,
+      -0.4939427971839905,
+      -0.732154905796051,
+      -0.5326835513114929,
+      -0.6656410694122314,
+      -0.5514172315597534,
+      -0.4456074833869934,
+      -0.8364714980125427,
+      -0.6598624587059021,
+      -0.48375943303108215,
+      -0.8276984691619873,
+      -0.6905249953269958,
+      -0.7523627281188965,
+      -0.6108655333518982,
+      -0.634128749370575,
+      -0.6320534944534302
+    ],
+    "base_loss": [
+      8.179495811462402,
+      6.115806579589844,
+      4.016663551330566,
+      2.338233232498169,
+      2.049792528152466,
+      1.9500830173492432,
+      1.8704925775527954,
+      1.915109634399414,
+      1.9027211666107178,
+      1.822627067565918,
+      1.7563773393630981,
+      1.8559844493865967,
+      1.810850739479065,
+      1.8188568353652954,
+      1.8670803308486938,
+      1.7376196384429932,
+      1.7410252094268799,
+      1.6789716482162476,
+      1.7815697193145752,
+      1.7297338247299194,
+      1.76444411277771,
+      1.7341679334640503,
+      1.745019555091858,
+      1.7294776439666748,
+      1.7048413753509521,
+      1.7179557085037231,
+      1.7174227237701416,
+      1.8109327554702759,
+      1.6843860149383545,
+      1.691207766532898,
+      1.6398429870605469,
+      1.6830883026123047,
+      1.7041343450546265,
+      1.7096806764602661,
+      1.6330136060714722,
+      1.572530746459961,
+      1.6290007829666138,
+      1.4542722702026367,
+      1.4407005310058594,
+      1.222497820854187,
+      1.0295950174331665,
+      0.9284564256668091,
+      0.6118221282958984,
+      0.5249842405319214,
+      0.4513547718524933,
+      0.31663909554481506,
+      0.3410135805606842,
+      0.3127397894859314,
+      0.23937572538852692,
+      0.23965051770210266,
+      0.2704034447669983,
+      0.22560283541679382,
+      0.20714624226093292,
+      0.18898002803325653,
+      0.21477903425693512,
+      0.18460628390312195,
+      0.20580320060253143,
+      0.16052566468715668,
+      0.17262044548988342,
+      0.14774015545845032,
+      0.14181342720985413,
+      0.2022370994091034,
+      0.15932251513004303,
+      0.11304645985364914,
+      0.16339722275733948,
+      0.16573865711688995,
+      0.17047688364982605,
+      0.13443663716316223,
+      0.14277073740959167,
+      0.12313588708639145
+    ],
+    "info_loss": [
+      -0.23002290725708008,
+      -0.16527986526489258,
+      -0.11838364601135254,
+      -0.08787274360656738,
+      -0.10383403301239014,
+      -0.0843881368637085,
+      -0.08847904205322266,
+      -0.1563432216644287,
+      -0.1807953119277954,
+      -0.21196186542510986,
+      -0.19581389427185059,
+      -0.24078822135925293,
+      -0.2084348201751709,
+      -0.21077227592468262,
+      -0.2398141622543335,
+      -0.2098095417022705,
+      -0.231908917427063,
+      -0.20753967761993408,
+      -0.17290937900543213,
+      -0.17037701606750488,
+      -0.12913298606872559,
+      -0.12954926490783691,
+      -0.1401742696762085,
+      -0.13750088214874268,
+      -0.18243885040283203,
+      -0.18511807918548584,
+      -0.36848652362823486,
+      -0.8207617402076721,
+      -1.0179661512374878,
+      -1.0324950218200684,
+      -1.079706072807312,
+      -1.1315070390701294,
+      -1.2022056579589844,
+      -1.2376500368118286,
+      -1.2149099111557007,
+      -1.234117865562439,
+      -1.2836066484451294,
+      -1.1954413652420044,
+      -1.2143906354904175,
+      -0.9723633527755737,
+      -0.8120318651199341,
+      -0.7064701318740845,
+      -0.40395891666412354,
+      -0.3564797043800354,
+      -0.2730256915092468,
+      -0.1676408350467682,
+      -0.2091994285583496,
+      -0.12594006955623627,
+      -0.1181057021021843,
+      -0.12815025448799133,
+      -0.16598695516586304,
+      -0.11813508719205856,
+      -0.08626975864171982,
+      -0.09994277358055115,
+      -0.14211693406105042,
+      -0.07376568019390106,
+      -0.10002028197050095,
+      -0.07552778720855713,
+      -0.08972552418708801,
+      -0.0756625384092331,
+      -0.06394617259502411,
+      -0.10925082862377167,
+      -0.08743764460086823,
+      -0.06474234163761139,
+      -0.10444813966751099,
+      -0.09072288870811462,
+      -0.09688038378953934,
+      -0.07976917922496796,
+      -0.08310963958501816,
+      -0.08014579117298126
+    ],
+    "abs_loss": [
+      2.2766036987304688,
+      2.091233730316162,
+      1.9168267250061035,
+      1.879494547843933,
+      1.8019380569458008,
+      1.82309889793396,
+      1.8274445533752441,
+      1.8011195659637451,
+      1.8016701936721802,
+      1.8573944568634033,
+      1.8407838344573975,
+      1.8531955480575562,
+      1.851827621459961,
+      1.843540906906128,
+      1.8486919403076172,
+      1.8462592363357544,
+      1.8527169227600098,
+      1.8005094528198242,
+      1.8157070875167847,
+      1.8012362718582153,
+      1.0213181972503662,
+      0.5417872667312622,
+      0.5030105113983154,
+      0.5118759274482727,
+      0.4955863952636719,
+      0.44640424847602844,
+      0.6110158562660217,
+      0.558056116104126,
+      0.486600399017334,
+      0.562961757183075,
+      0.47528234124183655,
+      0.49910861253738403,
+      0.46877163648605347,
+      0.44574135541915894,
+      0.5675186514854431,
+      0.4594317078590393,
+      0.49148818850517273,
+      0.5518755912780762,
+      0.5050919055938721,
+      0.4664192795753479,
+      0.5244419574737549,
+      0.47135239839553833,
+      0.5251164436340332,
+      0.5231102705001831,
+      0.5319302678108215,
+      0.5147315263748169,
+      0.4716704487800598,
+      0.5043060779571533,
+      0.5314348936080933,
+      0.6154720783233643,
+      0.6017305850982666,
+      0.5053713321685791,
+      0.5168009996414185,
+      0.5287708044052124,
+      0.5177637338638306,
+      0.46657508611679077,
+      0.4951359033584595,
+      0.4861401319503784,
+      0.40949398279190063,
+      0.38447850942611694,
+      0.40168479084968567,
+      0.3637966215610504,
+      0.41874241828918457,
+      0.4040564298629761,
+      0.3729878067970276,
+      0.36216187477111816,
+      0.3261665105819702,
+      0.3961992561817169,
+      0.40525567531585693,
+      0.347781777381897
+    ],
+    "zipf_loss": [
+      8.15005111694336,
+      4.140061378479004,
+      2.16505765914917,
+      1.8137528896331787,
+      1.7436177730560303,
+      1.7224159240722656,
+      1.707816481590271,
+      1.6981955766677856,
+      1.697060465812683,
+      1.6934552192687988,
+      1.6912052631378174,
+      1.6905877590179443,
+      1.6917719841003418,
+      1.6892836093902588,
+      1.6837599277496338,
+      1.6840280294418335,
+      1.6766400337219238,
+      1.658639669418335,
+      1.6033053398132324,
+      1.3812410831451416,
+      0.426512748003006,
+      0.0915759801864624,
+      0.06641378998756409,
+      0.06541288644075394,
+      0.05397297441959381,
+      0.049725256860256195,
+      0.09616798162460327,
+      0.07721105217933655,
+      0.06286955624818802,
+      0.06927378475666046,
+      0.05181824788451195,
+      0.05314686521887779,
+      0.04097625985741615,
+      0.04374873265624046,
+      0.032902587205171585,
+      0.03117380663752556,
+      0.03125934302806854,
+      0.02452976070344448,
+      0.025319188833236694,
+      0.03315161168575287,
+      0.03438675403594971,
+      0.03433896601200104,
+      0.023289574310183525,
+      0.02627655677497387,
+      0.019970480352640152,
+      0.029496990144252777,
+      0.029075412079691887,
+      0.022487249225378036,
+      0.02366235852241516,
+      0.016098806634545326,
+      0.01269473321735859,
+      0.02091185189783573,
+      0.01328546553850174,
+      0.011220241896808147,
+      0.014205502346158028,
+      0.01245021354407072,
+      0.012731088325381279,
+      0.013454608619213104,
+      0.01804427243769169,
+      0.019020183011889458,
+      0.011872355826199055,
+      0.017420047894120216,
+      0.013317206874489784,
+      0.010211849585175514,
+      0.016086921095848083,
+      0.014749030582606792,
+      0.013347554951906204,
+      0.012769741006195545,
+      0.013671365566551685,
+      0.011490320786833763
+    ],
+    "denoise_loss": [],
+    "ortho_loss": [
+      0.5640959739685059,
+      0.2631226181983948,
+      0.14999699592590332,
+      0.12239736318588257,
+      0.09714409708976746,
+      0.07303234189748764,
+      0.06550216674804688,
+      0.06051327660679817,
+      0.07507379353046417,
+      0.084007628262043,
+      0.08215909451246262,
+      0.08330931514501572,
+      0.08743671327829361,
+      0.09553566575050354,
+      0.08920977264642715,
+      0.09277433902025223,
+      0.09217482805252075,
+      0.09592001140117645,
+      0.10014844685792923,
+      0.10119500756263733,
+      0.09707954525947571,
+      0.097630575299263,
+      0.09751289337873459,
+      0.09296217560768127,
+      0.08584143221378326,
+      0.08582761883735657,
+      0.08925554901361465,
+      0.09694031625986099,
+      0.10842876881361008,
+      0.11049525439739227,
+      0.11342362314462662,
+      0.11486317962408066,
+      0.11290319263935089,
+      0.1154816746711731,
+      0.11996712535619736,
+      0.13185496628284454,
+      0.12799306213855743,
+      0.13368353247642517,
+      0.13123612105846405,
+      0.13329772651195526,
+      0.1350865513086319,
+      0.13771839439868927,
+      0.1451946496963501,
+      0.15061640739440918,
+      0.1515263319015503,
+      0.1512250006198883,
+      0.14959852397441864,
+      0.15471874177455902,
+      0.1546550989151001,
+      0.15259236097335815,
+      0.1592143177986145,
+      0.16225320100784302,
+      0.16515889763832092,
+      0.16539309918880463,
+      0.16331565380096436,
+      0.16385121643543243,
+      0.17084619402885437,
+      0.17373836040496826,
+      0.17300517857074738,
+      0.17302757501602173,
+      0.17285387217998505,
+      0.17389743030071259,
+      0.17504020035266876,
+      0.17735148966312408,
+      0.18059377372264862,
+      0.18277068436145782,
+      0.1846488118171692,
+      0.18622788786888123,
+      0.18851174414157867,
+      0.19029369950294495
+    ],
+    "lr": [
+      1.6752136752136756e-05,
+      3.384615384615385e-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,
+      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,
+      4e-05,
+      4e-05,
+      4e-05,
+      4e-05,
+      4e-05,
+      4e-05,
+      4e-05,
+      4e-05,
+      3.9947798576324814e-05,
+      3.8761402583706826e-05,
+      3.757500659108885e-05,
+      3.6388610598470864e-05,
+      3.5202214605852884e-05,
+      3.401581861323491e-05,
+      3.282942262061693e-05,
+      3.0670181914052204e-05,
+      2.948378592143422e-05,
+      2.8297389928816243e-05,
+      2.711099393619826e-05,
+      2.5924597943580284e-05,
+      2.4738201950962303e-05,
+      2.3551805958344316e-05,
+      2.1392565251779595e-05,
+      2.020616925916161e-05,
+      1.901977326654364e-05,
+      1.783337727392566e-05,
+      1.6646981281307675e-05,
+      1.546058528868969e-05,
+      1.427418929607171e-05,
+      1.2114948589506984e-05,
+      1.0928552596889013e-05,
+      9.742156604271029e-06,
+      8.555760611653046e-06,
+      7.369364619035064e-06,
+      6.182968626417082e-06,
+      4.996572633799099e-06
+    ],
+    "emb_lr": [],
+    "eval_step": [
+      350,
+      741,
+      1132,
+      1523,
+      1914,
+      2305,
+      2696,
+      3087,
+      3478,
+      3869
+    ],
+    "eval_accuracy": [
+      0.01,
+      0.0,
+      0.01,
+      0.12,
+      0.26,
+      0.6,
+      0.59,
+      0.76,
+      0.78,
+      0.83
+    ]
+  },
+  "final_accuracy": 0.6191666666666666,
+  "sft_eval": {
+    "config": {
+      "ops": "add_sub",
+      "K": null,
+      "mode": "sft",
+      "n_digits": 6,
+      "n_per_split": 100
+    },
+    "splits": {
+      "add_S0": {
+        "full_accuracy": 0.82,
+        "n_examples": 100,
+        "per_subtask": {
+          "SA": {
+            "accuracy": 0.9735537190082645,
+            "count": 605
+          },
+          "SS": {
+            "accuracy": 0.9473684210526315,
+            "count": 95
+          }
+        }
+      },
+      "add_S1": {
+        "full_accuracy": 0.83,
+        "n_examples": 100,
+        "per_subtask": {
+          "SA": {
+            "accuracy": 0.9754901960784313,
+            "count": 204
+          },
+          "SC": {
+            "accuracy": 0.9881656804733728,
+            "count": 169
+          },
+          "SS": {
+            "accuracy": 1.0,
+            "count": 31
+          },
+          "UC": {
+            "accuracy": 0.9628378378378378,
+            "count": 296
+          }
+        }
+      },
+      "add_S2": {
+        "full_accuracy": 0.59,
+        "n_examples": 100,
+        "per_subtask": {
+          "SA": {
+            "accuracy": 0.9815950920245399,
+            "count": 163
+          },
+          "SC": {
+            "accuracy": 0.9384615384615385,
+            "count": 130
+          },
+          "SS": {
+            "accuracy": 0.9425287356321839,
+            "count": 87
+          },
+          "UC": {
+            "accuracy": 0.8669950738916257,
+            "count": 203
+          },
+          "US": {
+            "accuracy": 0.9145299145299145,
+            "count": 117
+          }
+        }
+      },
+      "add_S3": {
+        "full_accuracy": 0.38,
+        "n_examples": 100,
+        "per_subtask": {
+          "SA": {
+            "accuracy": 0.9586776859504132,
+            "count": 121
+          },
+          "SC": {
+            "accuracy": 0.9752066115702479,
+            "count": 121
+          },
+          "SS": {
+            "accuracy": 0.9387755102040817,
+            "count": 49
+          },
+          "UC": {
+            "accuracy": 0.7204301075268817,
+            "count": 186
+          },
+          "US": {
+            "accuracy": 0.7982062780269058,
+            "count": 223
+          }
+        }
+      },
+      "add_S4": {
+        "full_accuracy": 0.31,
+        "n_examples": 100,
+        "per_subtask": {
+          "SA": {
+            "accuracy": 0.9807692307692307,
+            "count": 104
+          },
+          "SC": {
+            "accuracy": 0.9716981132075472,
+            "count": 106
+          },
+          "SS": {
+            "accuracy": 1.0,
+            "count": 23
+          },
+          "UC": {
+            "accuracy": 0.63125,
+            "count": 160
+          },
+          "US": {
+            "accuracy": 0.6742671009771987,
+            "count": 307
+          }
+        }
+      },
+      "add_S5": {
+        "full_accuracy": 0.28,
+        "n_examples": 100,
+        "per_subtask": {
+          "SA": {
+            "accuracy": 1.0,
+            "count": 100
+          },
+          "SC": {
+            "accuracy": 1.0,
+            "count": 100
+          },
+          "UC": {
+            "accuracy": 0.46,
+            "count": 100
+          },
+          "US": {
+            "accuracy": 0.51,
+            "count": 400
+          }
+        }
+      },
+      "add_S6": {
+        "full_accuracy": 0.41,
+        "n_examples": 100,
+        "per_subtask": {
+          "SC": {
+            "accuracy": 1.0,
+            "count": 100
+          },
+          "UC": {
+            "accuracy": 0.44,
+            "count": 100
+          },
+          "US": {
+            "accuracy": 0.582,
+            "count": 500
+          }
+        }
+      },
+      "add_random": {
+        "full_accuracy": 0.785,
+        "n_examples": 200,
+        "per_subtask": {
+          "SA": {
+            "accuracy": 0.9753914988814317,
+            "count": 447
+          },
+          "SC": {
+            "accuracy": 0.98125,
+            "count": 320
+          },
+          "SS": {
+            "accuracy": 0.9821428571428571,
+            "count": 56
+          },
+          "UC": {
+            "accuracy": 0.9489603024574669,
+            "count": 529
+          },
+          "US": {
+            "accuracy": 0.7708333333333334,
+            "count": 48
+          }
+        }
+      },
+      "add_C3": {
+        "full_accuracy": 0.56,
+        "n_examples": 100,
+        "per_subtask": {
+          "SA": {
+            "accuracy": 0.9966666666666667,
+            "count": 300
+          },
+          "SC": {
+            "accuracy": 1.0,
+            "count": 100
+          },
+          "UC": {
+            "accuracy": 0.772020725388601,
+            "count": 193
+          },
+          "US": {
+            "accuracy": 0.8037383177570093,
+            "count": 107
+          }
+        }
+      },
+      "add_C4": {
+        "full_accuracy": 0.54,
+        "n_examples": 100,
+        "per_subtask": {
+          "SA": {
+            "accuracy": 1.0,
+            "count": 200
+          },
+          "SC": {
+            "accuracy": 1.0,
+            "count": 100
+          },
+          "UC": {
+            "accuracy": 0.8359375,
+            "count": 256
+          },
+          "US": {
+            "accuracy": 0.8611111111111112,
+            "count": 144
+          }
+        }
+      },
+      "add_C5": {
+        "full_accuracy": 0.4,
+        "n_examples": 100,
+        "per_subtask": {
+          "SA": {
+            "accuracy": 1.0,
+            "count": 100
+          },
+          "SC": {
+            "accuracy": 1.0,
+            "count": 100
+          },
+          "UC": {
+            "accuracy": 0.803921568627451,
+            "count": 306
+          },
+          "US": {
+            "accuracy": 0.7731958762886598,
+            "count": 194
+          }
+        }
+      },
+      "add_C6": {
+        "full_accuracy": 0.52,
+        "n_examples": 100,
+        "per_subtask": {
+          "SC": {
+            "accuracy": 1.0,
+            "count": 100
+          },
+          "UC": {
+            "accuracy": 0.8688524590163934,
+            "count": 366
+          },
+          "US": {
+            "accuracy": 0.8675213675213675,
+            "count": 234
+          }
+        }
+      },
+      "sub_M0": {
+        "full_accuracy": 0.9,
+        "n_examples": 100,
+        "per_subtask": {
+          "MD": {
+            "accuracy": 0.9850249584026622,
+            "count": 601
+          },
+          "ME": {
+            "accuracy": 0.9797979797979798,
+            "count": 99
+          }
+        }
+      },
+      "sub_M1": {
+        "full_accuracy": 0.9,
+        "n_examples": 100,
+        "per_subtask": {
+          "MD": {
+            "accuracy": 0.989247311827957,
+            "count": 279
+          },
+          "MB": {
+            "accuracy": 0.9793103448275862,
+            "count": 145
+          },
+          "ME": {
+            "accuracy": 0.9583333333333334,
+            "count": 24
+          },
+          "UB": {
+            "accuracy": 0.9880952380952381,
+            "count": 252
+          }
+        }
+      },
+      "sub_M2": {
+        "full_accuracy": 0.65,
+        "n_examples": 100,
+        "per_subtask": {
+          "MD": {
+            "accuracy": 0.9953051643192489,
+            "count": 213
+          },
+          "MB": {
+            "accuracy": 1.0,
+            "count": 113
+          },
+          "ME": {
+            "accuracy": 0.9764705882352941,
+            "count": 85
+          },
+          "UB": {
+            "accuracy": 0.8176795580110497,
+            "count": 181
+          },
+          "UD": {
+            "accuracy": 0.9537037037037037,
+            "count": 108
+          }
+        }
+      },
+      "sub_M3": {
+        "full_accuracy": 0.24,
+        "n_examples": 100,
+        "per_subtask": {
+          "MD": {
+            "accuracy": 1.0,
+            "count": 179
+          },
+          "MB": {
+            "accuracy": 0.9902912621359223,
+            "count": 103
+          },
+          "ME": {
+            "accuracy": 1.0,
+            "count": 56
+          },
+          "UB": {
+            "accuracy": 0.5234899328859061,
+            "count": 149
+          },
+          "UD": {
+            "accuracy": 0.7934272300469484,
+            "count": 213
+          }
+        }
+      },
+      "sub_M4": {
+        "full_accuracy": 0.01,
+        "n_examples": 100,
+        "per_subtask": {
+          "MD": {
+            "accuracy": 1.0,
+            "count": 200
+          },
+          "MB": {
+            "accuracy": 0.99,
+            "count": 100
+          },
+          "UB": {
+            "accuracy": 0.39,
+            "count": 100
+          },
+          "UD": {
+            "accuracy": 0.43666666666666665,
+            "count": 300
+          }
+        }
+      },
+      "sub_M5": {
+        "full_accuracy": 0.0,
+        "n_examples": 100,
+        "per_subtask": {
+          "MD": {
+            "accuracy": 1.0,
+            "count": 100
+          },
+          "MB": {
+            "accuracy": 1.0,
+            "count": 100
+          },
+          "UB": {
+            "accuracy": 0.39,
+            "count": 100
+          },
+          "UD": {
+            "accuracy": 0.275,
+            "count": 400
+          }
+        }
+      },
+      "sub_random": {
+        "full_accuracy": 0.81,
+        "n_examples": 200,
+        "per_subtask": {
+          "MD": {
+            "accuracy": 0.9933333333333333,
+            "count": 600
+          },
+          "MB": {
+            "accuracy": 0.9625468164794008,
+            "count": 267
+          },
+          "ME": {
+            "accuracy": 0.9622641509433962,
+            "count": 53
+          },
+          "UB": {
+            "accuracy": 0.9476082004555809,
+            "count": 439
+          },
+          "UD": {
+            "accuracy": 0.975609756097561,
+            "count": 41
+          }
+        }
+      },
+      "sub_B3": {
+        "full_accuracy": 0.62,
+        "n_examples": 100,
+        "per_subtask": {
+          "MD": {
+            "accuracy": 0.9966666666666667,
+            "count": 300
+          },
+          "MB": {
+            "accuracy": 1.0,
+            "count": 100
+          },
+          "UB": {
+            "accuracy": 0.8274111675126904,
+            "count": 197
+          },
+          "UD": {
+            "accuracy": 0.7669902912621359,
+            "count": 103
+          }
+        }
+      },
+      "sub_B4": {
+        "full_accuracy": 0.38,
+        "n_examples": 100,
+        "per_subtask": {
+          "MD": {
+            "accuracy": 1.0,
+            "count": 200
+          },
+          "MB": {
+            "accuracy": 1.0,
+            "count": 100
+          },
+          "UB": {
+            "accuracy": 0.7732793522267206,
+            "count": 247
+          },
+          "UD": {
+            "accuracy": 0.7450980392156863,
+            "count": 153
+          }
+        }
+      },
+      "sub_B5": {
+        "full_accuracy": 0.37,
+        "n_examples": 100,
+        "per_subtask": {
+          "MD": {
+            "accuracy": 1.0,
+            "count": 100
+          },
+          "MB": {
+            "accuracy": 1.0,
+            "count": 100
+          },
+          "UB": {
+            "accuracy": 0.785234899328859,
+            "count": 298
+          },
+          "UD": {
+            "accuracy": 0.7524752475247525,
+            "count": 202
+          }
+        }
+      }
+    },
+    "summary": {
+      "overall_accuracy": 0.5375,
+      "total_examples": 2400,
+      "n_splits": 22
+    }
+  },
+  "sorl_eval": {
+    "config": {
+      "ops": "add_sub",
+      "K": 4,
+      "mode": "sorl",
+      "n_digits": 6,
+      "n_per_split": 100
+    },
+    "splits": {
+      "add_S0": {
+        "full_accuracy": 0.97,
+        "n_examples": 100,
+        "per_subtask": {
+          "SA": {
+            "accuracy": 0.9950413223140496,
+            "count": 605
+          },
+          "SS": {
+            "accuracy": 1.0,
+            "count": 95
+          }
+        }
+      },
+      "add_S1": {
+        "full_accuracy": 0.9,
+        "n_examples": 100,
+        "per_subtask": {
+          "SA": {
+            "accuracy": 0.9852941176470589,
+            "count": 204
+          },
+          "SC": {
+            "accuracy": 0.9822485207100592,
+            "count": 169
+          },
+          "SS": {
+            "accuracy": 1.0,
+            "count": 31
+          },
+          "UC": {
+            "accuracy": 0.9864864864864865,
+            "count": 296
+          }
+        }
+      },
+      "add_S2": {
+        "full_accuracy": 0.76,
+        "n_examples": 100,
+        "per_subtask": {
+          "SA": {
+            "accuracy": 0.9815950920245399,
+            "count": 163
+          },
+          "SC": {
+            "accuracy": 0.9615384615384616,
+            "count": 130
+          },
+          "SS": {
+            "accuracy": 0.9425287356321839,
+            "count": 87
+          },
+          "UC": {
+            "accuracy": 0.9261083743842364,
+            "count": 203
+          },
+          "US": {
+            "accuracy": 1.0,
+            "count": 117
+          }
+        }
+      },
+      "add_S3": {
+        "full_accuracy": 0.57,
+        "n_examples": 100,
+        "per_subtask": {
+          "SA": {
+            "accuracy": 0.9917355371900827,
+            "count": 121
+          },
+          "SC": {
+            "accuracy": 0.9834710743801653,
+            "count": 121
+          },
+          "SS": {
+            "accuracy": 0.9591836734693877,
+            "count": 49
+          },
+          "UC": {
+            "accuracy": 0.7956989247311828,
+            "count": 186
+          },
+          "US": {
+            "accuracy": 0.9192825112107623,
+            "count": 223
+          }
+        }
+      },
+      "add_S4": {
+        "full_accuracy": 0.38,
+        "n_examples": 100,
+        "per_subtask": {
+          "SA": {
+            "accuracy": 1.0,
+            "count": 104
+          },
+          "SC": {
+            "accuracy": 1.0,
+            "count": 106
+          },
+          "SS": {
+            "accuracy": 1.0,
+            "count": 23
+          },
+          "UC": {
+            "accuracy": 0.68125,
+            "count": 160
+          },
+          "US": {
+            "accuracy": 0.7654723127035831,
+            "count": 307
+          }
+        }
+      },
+      "add_S5": {
+        "full_accuracy": 0.19,
+        "n_examples": 100,
+        "per_subtask": {
+          "SA": {
+            "accuracy": 1.0,
+            "count": 100
+          },
+          "SC": {
+            "accuracy": 1.0,
+            "count": 100
+          },
+          "UC": {
+            "accuracy": 0.32,
+            "count": 100
+          },
+          "US": {
+            "accuracy": 0.505,
+            "count": 400
+          }
+        }
+      },
+      "add_S6": {
+        "full_accuracy": 0.38,
+        "n_examples": 100,
+        "per_subtask": {
+          "SC": {
+            "accuracy": 1.0,
+            "count": 100
+          },
+          "UC": {
+            "accuracy": 0.59,
+            "count": 100
+          },
+          "US": {
+            "accuracy": 0.634,
+            "count": 500
+          }
+        }
+      },
+      "add_random": {
+        "full_accuracy": 0.885,
+        "n_examples": 200,
+        "per_subtask": {
+          "SA": {
+            "accuracy": 0.9865771812080537,
+            "count": 447
+          },
+          "SC": {
+            "accuracy": 0.98125,
+            "count": 320
+          },
+          "SS": {
+            "accuracy": 0.9642857142857143,
+            "count": 56
+          },
+          "UC": {
+            "accuracy": 0.9792060491493384,
+            "count": 529
+          },
+          "US": {
+            "accuracy": 0.9375,
+            "count": 48
+          }
+        }
+      },
+      "add_C3": {
+        "full_accuracy": 0.71,
+        "n_examples": 100,
+        "per_subtask": {
+          "SA": {
+            "accuracy": 1.0,
+            "count": 300
+          },
+          "SC": {
+            "accuracy": 1.0,
+            "count": 100
+          },
+          "UC": {
+            "accuracy": 0.8497409326424871,
+            "count": 193
+          },
+          "US": {
+            "accuracy": 0.9345794392523364,
+            "count": 107
+          }
+        }
+      },
+      "add_C4": {
+        "full_accuracy": 0.76,
+        "n_examples": 100,
+        "per_subtask": {
+          "SA": {
+            "accuracy": 1.0,
+            "count": 200
+          },
+          "SC": {
+            "accuracy": 1.0,
+            "count": 100
+          },
+          "UC": {
+            "accuracy": 0.91015625,
+            "count": 256
+          },
+          "US": {
+            "accuracy": 0.9236111111111112,
+            "count": 144
+          }
+        }
+      },
+      "add_C5": {
+        "full_accuracy": 0.54,
+        "n_examples": 100,
+        "per_subtask": {
+          "SA": {
+            "accuracy": 1.0,
+            "count": 100
+          },
+          "SC": {
+            "accuracy": 1.0,
+            "count": 100
+          },
+          "UC": {
+            "accuracy": 0.8594771241830066,
+            "count": 306
+          },
+          "US": {
+            "accuracy": 0.8402061855670103,
+            "count": 194
+          }
+        }
+      },
+      "add_C6": {
+        "full_accuracy": 0.68,
+        "n_examples": 100,
+        "per_subtask": {
+          "SC": {
+            "accuracy": 1.0,
+            "count": 100
+          },
+          "UC": {
+            "accuracy": 0.9098360655737705,
+            "count": 366
+          },
+          "US": {
+            "accuracy": 0.9487179487179487,
+            "count": 234
+          }
+        }
+      },
+      "sub_M0": {
+        "full_accuracy": 0.98,
+        "n_examples": 100,
+        "per_subtask": {
+          "MD": {
+            "accuracy": 0.9966722129783694,
+            "count": 601
+          },
+          "ME": {
+            "accuracy": 1.0,
+            "count": 99
+          }
+        }
+      },
+      "sub_M1": {
+        "full_accuracy": 0.97,
+        "n_examples": 100,
+        "per_subtask": {
+          "MD": {
+            "accuracy": 1.0,
+            "count": 279
+          },
+          "MB": {
+            "accuracy": 0.9862068965517241,
+            "count": 145
+          },
+          "ME": {
+            "accuracy": 0.9583333333333334,
+            "count": 24
+          },
+          "UB": {
+            "accuracy": 1.0,
+            "count": 252
+          }
+        }
+      },
+      "sub_M2": {
+        "full_accuracy": 0.63,
+        "n_examples": 100,
+        "per_subtask": {
+          "MD": {
+            "accuracy": 0.9906103286384976,
+            "count": 213
+          },
+          "MB": {
+            "accuracy": 0.9911504424778761,
+            "count": 113
+          },
+          "ME": {
+            "accuracy": 1.0,
+            "count": 85
+          },
+          "UB": {
+            "accuracy": 0.8121546961325967,
+            "count": 181
+          },
+          "UD": {
+            "accuracy": 1.0,
+            "count": 108
+          }
+        }
+      },
+      "sub_M3": {
+        "full_accuracy": 0.2,
+        "n_examples": 100,
+        "per_subtask": {
+          "MD": {
+            "accuracy": 0.994413407821229,
+            "count": 179
+          },
+          "MB": {
+            "accuracy": 0.9902912621359223,
+            "count": 103
+          },
+          "ME": {
+            "accuracy": 1.0,
+            "count": 56
+          },
+          "UB": {
+            "accuracy": 0.4966442953020134,
+            "count": 149
+          },
+          "UD": {
+            "accuracy": 0.8075117370892019,
+            "count": 213
+          }
+        }
+      },
+      "sub_M4": {
+        "full_accuracy": 0.06,
+        "n_examples": 100,
+        "per_subtask": {
+          "MD": {
+            "accuracy": 1.0,
+            "count": 200
+          },
+          "MB": {
+            "accuracy": 1.0,
+            "count": 100
+          },
+          "UB": {
+            "accuracy": 0.25,
+            "count": 100
+          },
+          "UD": {
+            "accuracy": 0.5,
+            "count": 300
+          }
+        }
+      },
+      "sub_M5": {
+        "full_accuracy": 0.08,
+        "n_examples": 100,
+        "per_subtask": {
+          "MD": {
+            "accuracy": 1.0,
+            "count": 100
+          },
+          "MB": {
+            "accuracy": 1.0,
+            "count": 100
+          },
+          "UB": {
+            "accuracy": 0.22,
+            "count": 100
+          },
+          "UD": {
+            "accuracy": 0.425,
+            "count": 400
+          }
+        }
+      },
+      "sub_random": {
+        "full_accuracy": 0.87,
+        "n_examples": 200,
+        "per_subtask": {
+          "MD": {
+            "accuracy": 0.9933333333333333,
+            "count": 600
+          },
+          "MB": {
+            "accuracy": 0.9887640449438202,
+            "count": 267
+          },
+          "ME": {
+            "accuracy": 1.0,
+            "count": 53
+          },
+          "UB": {
+            "accuracy": 0.9567198177676538,
+            "count": 439
+          },
+          "UD": {
+            "accuracy": 0.975609756097561,
+            "count": 41
+          }
+        }
+      },
+      "sub_B3": {
+        "full_accuracy": 0.67,
+        "n_examples": 100,
+        "per_subtask": {
+          "MD": {
+            "accuracy": 0.9933333333333333,
+            "count": 300
+          },
+          "MB": {
+            "accuracy": 1.0,
+            "count": 100
+          },
+          "UB": {
+            "accuracy": 0.8629441624365483,
+            "count": 197
+          },
+          "UD": {
+            "accuracy": 0.8349514563106796,
+            "count": 103
+          }
+        }
+      },
+      "sub_B4": {
+        "full_accuracy": 0.55,
+        "n_examples": 100,
+        "per_subtask": {
+          "MD": {
+            "accuracy": 1.0,
+            "count": 200
+          },
+          "MB": {
+            "accuracy": 1.0,
+            "count": 100
+          },
+          "UB": {
+            "accuracy": 0.8259109311740891,
+            "count": 247
+          },
+          "UD": {
+            "accuracy": 0.8169934640522876,
+            "count": 153
+          }
+        }
+      },
+      "sub_B5": {
+        "full_accuracy": 0.38,
+        "n_examples": 100,
+        "per_subtask": {
+          "MD": {
+            "accuracy": 1.0,
+            "count": 100
+          },
+          "MB": {
+            "accuracy": 1.0,
+            "count": 100
+          },
+          "UB": {
+            "accuracy": 0.7986577181208053,
+            "count": 298
+          },
+          "UD": {
+            "accuracy": 0.7574257425742574,
+            "count": 202
+          }
+        }
+      }
+    },
+    "summary": {
+      "overall_accuracy": 0.6191666666666666,
+      "total_examples": 2400,
+      "n_splits": 22
+    }
+  },
+  "sorl_overall_accuracy": 0.6191666666666666,
+  "sft_overall_accuracy": 0.5375
+}

add_sub_sorl_v1_abs10_25K/model.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:c44e7d1362819ab8f5d7105db96ac5f8cd27c038610420dcca40684cc135fc24
+size 650303660

add_sub_sorl_v1_abs10_25K/train_config.json

@@ -0,0 +1,84 @@
+{
+  "num_rollouts": 4,
+  "K": 4,
+  "max_iterations": 2,
+  "memory_span_abs": 1792,
+  "memory_span_traj": 1792,
+  "temperature": 1.0,
+  "ar_search": false,
+  "response_only_abs": false,
+  "alpha_info_gain": 10.0,
+  "alpha_abs": 0.1,
+  "alpha_soft_zipf": 1.0,
+  "alpha_ortho": 0.0,
+  "alpha_anchor": 0.0,
+  "alpha_jacobi": 0.0,
+  "decay": 0.8,
+  "target_vocab_util": 0.8,
+  "min_abs_ppl": 0.0,
+  "zipf_alpha": 1.0,
+  "lr": 4e-05,
+  "emb_lr_mult": 1.0,
+  "weight_decay": 0.01,
+  "warmup_steps": 117,
+  "cooldown_frac": 0.4,
+  "max_grad_norm": 1.0,
+  "vq_abs_pretrain_steps": 0,
+  "vq_abs_pretrain_lr": 0.001,
+  "vq_abs_pretrain_layer": -1,
+  "vq_abs_pretrain_batch_size": 256,
+  "vq_abs_pretrain_target_vectors": 20000,
+  "batch_size": 64,
+  "gradient_accumulation_steps": 1,
+  "num_epochs": 10,
+  "emb_warmup_steps": 0,
+  "log_every": 50,
+  "eval_every": 390,
+  "save_every": 999999,
+  "eval_samples": 100,
+  "output_dir": "ckpt/sweep/as_sorl_abs10_K4_25K",
+  "eval_K": 4,
+  "alpha_traj": 0.0,
+  "corrupt_method": "shuffle",
+  "corrupt_ratio": 0.3,
+  "alpha_contrastive": 1.0,
+  "gamma_contrastive": 0.5,
+  "alpha_masked_traj": 0.0,
+  "mask_nl_ratio": 0.3,
+  "mask_nl_mode": "fixed",
+  "mask_nl_fixed_id": 0,
+  "use_ste": true,
+  "n_inner": 1,
+  "random_K": null,
+  "strip_suffix": null,
+  "compress_prefix": null,
+  "random_mem_span": null,
+  "warmup_ratio": 0.03,
+  "beta2": 0.999,
+  "seed": 42,
+  "n_digits": 6,
+  "n_layer": 2,
+  "n_head": 3,
+  "n_embd": 510,
+  "ops": "add_sub",
+  "abs_vocab": 10,
+  "dataset_size": 25000,
+  "mode": "sorl",
+  "device": "cuda",
+  "push_to_hub": true,
+  "no_wandb": false,
+  "n_params": 162499262,
+  "run_name": "add_sub_sorl_v1_abs10_25K",
+  "git_commit": "78d46f8665a87f4b44bd5894bd34f393f2dea51f",
+  "timestamp": "2026-04-12T08:59:10.961931+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": "pmvjbi05",
+  "wandb_url": "https://wandb.ai/nlp_and_interpretability/sorl-arithmetic/runs/pmvjbi05",
+  "final_accuracy": 0.6191666666666666,
+  "sft_accuracy": 0.5375,
+  "eval_method": "ArithmeticEvaluator"
+}