Upload add_sub_sorl_v1_abs10_K1_10K

add_sub_sorl_v1_abs10_K1_10K/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_K1_10K/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_K1_10K/metrics.json

@@ -0,0 +1,1297 @@
+{
+  "history": {
+    "step": [
+      50,
+      100,
+      150,
+      207,
+      257,
+      307,
+      364,
+      414,
+      464,
+      521,
+      571,
+      621,
+      678,
+      728,
+      778,
+      835,
+      885,
+      935,
+      992,
+      1042,
+      1092,
+      1149,
+      1199,
+      1249,
+      1306,
+      1356,
+      1406,
+      1463,
+      1513,
+      1563
+    ],
+    "loss": [
+      8.773706436157227,
+      4.004265308380127,
+      3.2570762634277344,
+      3.041517734527588,
+      2.8982486724853516,
+      2.4261059761047363,
+      -1.0320148468017578,
+      -6.537303924560547,
+      -7.572015285491943,
+      -8.820948600769043,
+      -9.133442878723145,
+      -9.698243141174316,
+      -10.157323837280273,
+      -9.581876754760742,
+      -10.299582481384277,
+      -8.835819244384766,
+      -7.875391006469727,
+      -6.175548553466797,
+      -4.922896385192871,
+      -5.274993896484375,
+      -3.509742021560669,
+      -2.9279913902282715,
+      -2.248248815536499,
+      -1.4432270526885986,
+      -0.9849211573600769,
+      -0.7613394260406494,
+      -0.9093202948570251,
+      -0.20168478786945343,
+      -0.5009633302688599,
+      -0.4359404742717743
+    ],
+    "base_loss": [
+      6.400631904602051,
+      2.4112229347229004,
+      1.8464170694351196,
+      1.8753893375396729,
+      1.8903639316558838,
+      1.8746846914291382,
+      1.8820133209228516,
+      1.855467438697815,
+      1.7718533277511597,
+      1.7219583988189697,
+      1.7517011165618896,
+      1.6724598407745361,
+      1.647750973701477,
+      1.4754587411880493,
+      1.4356294870376587,
+      1.2360666990280151,
+      1.1430914402008057,
+      0.8601657748222351,
+      0.6953770518302917,
+      0.6893697381019592,
+      0.4810207784175873,
+      0.39060816168785095,
+      0.3150084316730499,
+      0.2136605978012085,
+      0.163777157664299,
+      0.12215958535671234,
+      0.13819025456905365,
+      0.08133143931627274,
+      0.08307908475399017,
+      0.07490883022546768
+    ],
+    "info_loss": [
+      -0.21113967895507812,
+      -0.04423642158508301,
+      -0.04978370666503906,
+      -0.0719677209854126,
+      -0.087005615234375,
+      -0.13271820545196533,
+      -0.4793226718902588,
+      -1.025976300239563,
+      -1.1140811443328857,
+      -1.2247107028961182,
+      -1.256365180015564,
+      -1.3006340265274048,
+      -1.3388965129852295,
+      -1.2538009881973267,
+      -1.3090555667877197,
+      -1.1411150693893433,
+      -1.0133163928985596,
+      -0.8001939654350281,
+      -0.6398270130157471,
+      -0.6569589376449585,
+      -0.44588613510131836,
+      -0.368633508682251,
+      -0.2934108078479767,
+      -0.1994001418352127,
+      -0.14748595654964447,
+      -0.11586211621761322,
+      -0.12713156640529633,
+      -0.04881247505545616,
+      -0.07891618460416794,
+      -0.06897100806236267
+    ],
+    "abs_loss": [
+      2.0780529975891113,
+      1.8440923690795898,
+      1.8333244323730469,
+      1.8426448106765747,
+      1.8403841257095337,
+      1.8378067016601562,
+      1.8460584878921509,
+      1.7702795267105103,
+      1.5657355785369873,
+      1.3797943592071533,
+      1.3185033798217773,
+      1.2104538679122925,
+      1.125502586364746,
+      0.9860450029373169,
+      0.8728498220443726,
+      0.7434267997741699,
+      0.7268884778022766,
+      0.5966061949729919,
+      0.5761818885803223,
+      0.4872754216194153,
+      0.42560893297195435,
+      0.3502724766731262,
+      0.29791387915611267,
+      0.267427533864975,
+      0.24475598335266113,
+      0.22809018194675446,
+      0.13244910538196564,
+      0.15626277029514313,
+      0.18973781168460846,
+      0.15587982535362244
+    ],
+    "zipf_loss": [
+      4.276666164398193,
+      1.8509973287582397,
+      1.7251636981964111,
+      1.7015411853790283,
+      1.6939024925231934,
+      1.6948225498199463,
+      1.6945927143096924,
+      1.6899633407592773,
+      1.6403698921203613,
+      1.5662205219268799,
+      1.546657681465149,
+      1.514591932296753,
+      1.4713406562805176,
+      1.3820693492889404,
+      1.2680578231811523,
+      1.26492178440094,
+      1.0419921875,
+      0.9065651893615723,
+      0.7223787903785706,
+      0.5564987659454346,
+      0.42553776502609253,
+      0.33270835876464844,
+      0.3410593867301941,
+      0.3103709816932678,
+      0.30168575048446655,
+      0.25231313705444336,
+      0.21056024730205536,
+      0.18948225677013397,
+      0.18614569306373596,
+      0.16327276825904846
+    ],
+    "denoise_loss": [],
+    "ortho_loss": [
+      0.3211989104747772,
+      0.19110506772994995,
+      0.11888417601585388,
+      0.0979497954249382,
+      0.09260708838701248,
+      0.10610710084438324,
+      0.14702288806438446,
+      0.21001708507537842,
+      0.2301187366247177,
+      0.24748258292675018,
+      0.2629227936267853,
+      0.26881012320518494,
+      0.269039511680603,
+      0.26941126585006714,
+      0.26823461055755615,
+      0.26427075266838074,
+      0.2743089199066162,
+      0.2734338045120239,
+      0.27655836939811707,
+      0.2692953944206238,
+      0.27160370349884033,
+      0.27964547276496887,
+      0.2874664068222046,
+      0.2921176552772522,
+      0.2956086993217468,
+      0.29919639229774475,
+      0.3037196099758148,
+      0.3071216940879822,
+      0.311845988035202,
+      0.3125845789909363
+    ],
+    "lr": [
+      7.840000000000001e-05,
+      8e-05,
+      8e-05,
+      8e-05,
+      8e-05,
+      8e-05,
+      8e-05,
+      8e-05,
+      8e-05,
+      8e-05,
+      8e-05,
+      8e-05,
+      8e-05,
+      8e-05,
+      8e-05,
+      8e-05,
+      8e-05,
+      8e-05,
+      7.656578947368422e-05,
+      7.064473684210527e-05,
+      6.472368421052631e-05,
+      5.7973684210526325e-05,
+      5.2052631578947374e-05,
+      4.613157894736842e-05,
+      3.938157894736842e-05,
+      3.346052631578948e-05,
+      2.7539473684210523e-05,
+      2.078947368421053e-05,
+      1.4868421052631572e-05,
+      8.947368421052635e-06
+    ],
+    "emb_lr": [],
+    "eval_step": [
+      150,
+      307,
+      464,
+      621,
+      778,
+      935,
+      1092,
+      1199,
+      1356,
+      1513
+    ],
+    "eval_accuracy": [
+      0.0,
+      0.0,
+      0.0,
+      0.0,
+      0.0,
+      0.0,
+      0.0,
+      0.0,
+      0.0,
+      0.0
+    ]
+  },
+  "final_accuracy": 0.0,
+  "sft_eval": {
+    "config": {
+      "ops": "add_sub",
+      "K": null,
+      "mode": "sft",
+      "n_digits": 6,
+      "n_per_split": 250
+    },
+    "splits": {
+      "add_S0": {
+        "full_accuracy": 0.944,
+        "n_examples": 250,
+        "per_subtask": {
+          "SA": {
+            "accuracy": 0.9908076165462902,
+            "count": 1523
+          },
+          "SS": {
+            "accuracy": 1.0,
+            "count": 227
+          }
+        }
+      },
+      "add_S1": {
+        "full_accuracy": 0.952,
+        "n_examples": 250,
+        "per_subtask": {
+          "SA": {
+            "accuracy": 0.9870848708487084,
+            "count": 542
+          },
+          "SC": {
+            "accuracy": 0.9952038369304557,
+            "count": 417
+          },
+          "SS": {
+            "accuracy": 1.0,
+            "count": 70
+          },
+          "UC": {
+            "accuracy": 0.9958391123439667,
+            "count": 721
+          }
+        }
+      },
+      "add_S2": {
+        "full_accuracy": 0.912,
+        "n_examples": 250,
+        "per_subtask": {
+          "SA": {
+            "accuracy": 0.970108695652174,
+            "count": 368
+          },
+          "SC": {
+            "accuracy": 0.9781931464174455,
+            "count": 321
+          },
+          "SS": {
+            "accuracy": 0.9736842105263158,
+            "count": 228
+          },
+          "UC": {
+            "accuracy": 0.9962335216572504,
+            "count": 531
+          },
+          "US": {
+            "accuracy": 1.0,
+            "count": 302
+          }
+        }
+      },
+      "add_S3": {
+        "full_accuracy": 0.756,
+        "n_examples": 250,
+        "per_subtask": {
+          "SA": {
+            "accuracy": 0.990228013029316,
+            "count": 307
+          },
+          "SC": {
+            "accuracy": 0.993127147766323,
+            "count": 291
+          },
+          "SS": {
+            "accuracy": 0.9823008849557522,
+            "count": 113
+          },
+          "UC": {
+            "accuracy": 0.8799171842650103,
+            "count": 483
+          },
+          "US": {
+            "accuracy": 1.0,
+            "count": 556
+          }
+        }
+      },
+      "add_S4": {
+        "full_accuracy": 0.544,
+        "n_examples": 250,
+        "per_subtask": {
+          "SA": {
+            "accuracy": 1.0,
+            "count": 238
+          },
+          "SC": {
+            "accuracy": 0.988929889298893,
+            "count": 271
+          },
+          "SS": {
+            "accuracy": 1.0,
+            "count": 59
+          },
+          "UC": {
+            "accuracy": 0.782716049382716,
+            "count": 405
+          },
+          "US": {
+            "accuracy": 0.9047619047619048,
+            "count": 777
+          }
+        }
+      },
+      "add_S5": {
+        "full_accuracy": 0.26,
+        "n_examples": 250,
+        "per_subtask": {
+          "SA": {
+            "accuracy": 0.988,
+            "count": 250
+          },
+          "SC": {
+            "accuracy": 1.0,
+            "count": 250
+          },
+          "UC": {
+            "accuracy": 0.452,
+            "count": 250
+          },
+          "US": {
+            "accuracy": 0.736,
+            "count": 1000
+          }
+        }
+      },
+      "add_S6": {
+        "full_accuracy": 0.656,
+        "n_examples": 250,
+        "per_subtask": {
+          "SC": {
+            "accuracy": 1.0,
+            "count": 250
+          },
+          "UC": {
+            "accuracy": 0.812,
+            "count": 250
+          },
+          "US": {
+            "accuracy": 0.8448,
+            "count": 1250
+          }
+        }
+      },
+      "add_random": {
+        "full_accuracy": 0.91,
+        "n_examples": 200,
+        "per_subtask": {
+          "SA": {
+            "accuracy": 0.9886621315192744,
+            "count": 441
+          },
+          "SC": {
+            "accuracy": 0.9936908517350158,
+            "count": 317
+          },
+          "SS": {
+            "accuracy": 0.9814814814814815,
+            "count": 54
+          },
+          "UC": {
+            "accuracy": 0.981203007518797,
+            "count": 532
+          },
+          "US": {
+            "accuracy": 0.9642857142857143,
+            "count": 56
+          }
+        }
+      },
+      "add_C3": {
+        "full_accuracy": 0.772,
+        "n_examples": 250,
+        "per_subtask": {
+          "SA": {
+            "accuracy": 0.9786666666666667,
+            "count": 750
+          },
+          "SC": {
+            "accuracy": 0.996,
+            "count": 250
+          },
+          "UC": {
+            "accuracy": 0.9096638655462185,
+            "count": 476
+          },
+          "US": {
+            "accuracy": 1.0,
+            "count": 274
+          }
+        }
+      },
+      "add_C4": {
+        "full_accuracy": 0.836,
+        "n_examples": 250,
+        "per_subtask": {
+          "SA": {
+            "accuracy": 0.986,
+            "count": 500
+          },
+          "SC": {
+            "accuracy": 0.996,
+            "count": 250
+          },
+          "UC": {
+            "accuracy": 0.9515885022692889,
+            "count": 661
+          },
+          "US": {
+            "accuracy": 0.9616519174041298,
+            "count": 339
+          }
+        }
+      },
+      "add_C5": {
+        "full_accuracy": 0.78,
+        "n_examples": 250,
+        "per_subtask": {
+          "SA": {
+            "accuracy": 0.996,
+            "count": 250
+          },
+          "SC": {
+            "accuracy": 1.0,
+            "count": 250
+          },
+          "UC": {
+            "accuracy": 0.9344262295081968,
+            "count": 732
+          },
+          "US": {
+            "accuracy": 0.9305019305019305,
+            "count": 518
+          }
+        }
+      },
+      "add_C6": {
+        "full_accuracy": 0.86,
+        "n_examples": 250,
+        "per_subtask": {
+          "SC": {
+            "accuracy": 1.0,
+            "count": 250
+          },
+          "UC": {
+            "accuracy": 0.9673423423423423,
+            "count": 888
+          },
+          "US": {
+            "accuracy": 0.9673202614379085,
+            "count": 612
+          }
+        }
+      },
+      "sub_M0": {
+        "full_accuracy": 0.988,
+        "n_examples": 250,
+        "per_subtask": {
+          "MD": {
+            "accuracy": 0.998671096345515,
+            "count": 1505
+          },
+          "ME": {
+            "accuracy": 0.9959183673469387,
+            "count": 245
+          }
+        }
+      },
+      "sub_M1": {
+        "full_accuracy": 0.932,
+        "n_examples": 250,
+        "per_subtask": {
+          "MD": {
+            "accuracy": 0.9929971988795518,
+            "count": 714
+          },
+          "MB": {
+            "accuracy": 0.9786096256684492,
+            "count": 374
+          },
+          "ME": {
+            "accuracy": 0.9866666666666667,
+            "count": 75
+          },
+          "UB": {
+            "accuracy": 0.9948892674616695,
+            "count": 587
+          }
+        }
+      },
+      "sub_M2": {
+        "full_accuracy": 0.896,
+        "n_examples": 250,
+        "per_subtask": {
+          "MD": {
+            "accuracy": 0.9909747292418772,
+            "count": 554
+          },
+          "MB": {
+            "accuracy": 0.9743589743589743,
+            "count": 273
+          },
+          "ME": {
+            "accuracy": 0.9908675799086758,
+            "count": 219
+          },
+          "UB": {
+            "accuracy": 0.9674418604651163,
+            "count": 430
+          },
+          "UD": {
+            "accuracy": 0.9963503649635036,
+            "count": 274
+          }
+        }
+      },
+      "sub_M3": {
+        "full_accuracy": 0.272,
+        "n_examples": 250,
+        "per_subtask": {
+          "MD": {
+            "accuracy": 0.9978165938864629,
+            "count": 458
+          },
+          "MB": {
+            "accuracy": 0.9923371647509579,
+            "count": 261
+          },
+          "ME": {
+            "accuracy": 1.0,
+            "count": 124
+          },
+          "UB": {
+            "accuracy": 0.5438144329896907,
+            "count": 388
+          },
+          "UD": {
+            "accuracy": 0.9730250481695568,
+            "count": 519
+          }
+        }
+      },
+      "sub_M4": {
+        "full_accuracy": 0.06,
+        "n_examples": 250,
+        "per_subtask": {
+          "MD": {
+            "accuracy": 1.0,
+            "count": 500
+          },
+          "MB": {
+            "accuracy": 0.98,
+            "count": 250
+          },
+          "UB": {
+            "accuracy": 0.192,
+            "count": 250
+          },
+          "UD": {
+            "accuracy": 0.6426666666666667,
+            "count": 750
+          }
+        }
+      },
+      "sub_M5": {
+        "full_accuracy": 0.016,
+        "n_examples": 250,
+        "per_subtask": {
+          "MD": {
+            "accuracy": 1.0,
+            "count": 250
+          },
+          "MB": {
+            "accuracy": 0.984,
+            "count": 250
+          },
+          "UB": {
+            "accuracy": 0.096,
+            "count": 250
+          },
+          "UD": {
+            "accuracy": 0.476,
+            "count": 1000
+          }
+        }
+      },
+      "sub_random": {
+        "full_accuracy": 0.935,
+        "n_examples": 200,
+        "per_subtask": {
+          "MD": {
+            "accuracy": 0.993103448275862,
+            "count": 580
+          },
+          "MB": {
+            "accuracy": 0.9887640449438202,
+            "count": 267
+          },
+          "ME": {
+            "accuracy": 1.0,
+            "count": 63
+          },
+          "UB": {
+            "accuracy": 0.984375,
+            "count": 448
+          },
+          "UD": {
+            "accuracy": 1.0,
+            "count": 42
+          }
+        }
+      },
+      "sub_B3": {
+        "full_accuracy": 0.724,
+        "n_examples": 250,
+        "per_subtask": {
+          "MD": {
+            "accuracy": 0.9986666666666667,
+            "count": 750
+          },
+          "MB": {
+            "accuracy": 0.996,
+            "count": 250
+          },
+          "UB": {
+            "accuracy": 0.8700787401574803,
+            "count": 508
+          },
+          "UD": {
+            "accuracy": 0.9628099173553719,
+            "count": 242
+          }
+        }
+      },
+      "sub_B4": {
+        "full_accuracy": 0.56,
+        "n_examples": 250,
+        "per_subtask": {
+          "MD": {
+            "accuracy": 0.988,
+            "count": 500
+          },
+          "MB": {
+            "accuracy": 0.976,
+            "count": 250
+          },
+          "UB": {
+            "accuracy": 0.8257328990228013,
+            "count": 614
+          },
+          "UD": {
+            "accuracy": 0.8860103626943006,
+            "count": 386
+          }
+        }
+      },
+      "sub_B5": {
+        "full_accuracy": 0.52,
+        "n_examples": 250,
+        "per_subtask": {
+          "MD": {
+            "accuracy": 1.0,
+            "count": 250
+          },
+          "MB": {
+            "accuracy": 0.992,
+            "count": 250
+          },
+          "UB": {
+            "accuracy": 0.8597168597168597,
+            "count": 777
+          },
+          "UD": {
+            "accuracy": 0.8752642706131079,
+            "count": 473
+          }
+        }
+      }
+    },
+    "summary": {
+      "overall_accuracy": 0.6812962962962963,
+      "total_examples": 5400,
+      "n_splits": 22
+    }
+  },
+  "sorl_eval": {
+    "config": {
+      "ops": "add_sub",
+      "K": 1,
+      "mode": "sorl",
+      "n_digits": 6,
+      "n_per_split": 250
+    },
+    "splits": {
+      "add_S0": {
+        "full_accuracy": 0.0,
+        "n_examples": 250,
+        "per_subtask": {
+          "SA": {
+            "accuracy": 0.2593565331582403,
+            "count": 1523
+          },
+          "SS": {
+            "accuracy": 0.23348017621145375,
+            "count": 227
+          }
+        }
+      },
+      "add_S1": {
+        "full_accuracy": 0.0,
+        "n_examples": 250,
+        "per_subtask": {
+          "SA": {
+            "accuracy": 0.31365313653136534,
+            "count": 542
+          },
+          "SC": {
+            "accuracy": 0.09112709832134293,
+            "count": 417
+          },
+          "SS": {
+            "accuracy": 0.12857142857142856,
+            "count": 70
+          },
+          "UC": {
+            "accuracy": 0.2565880721220527,
+            "count": 721
+          }
+        }
+      },
+      "add_S2": {
+        "full_accuracy": 0.0,
+        "n_examples": 250,
+        "per_subtask": {
+          "SA": {
+            "accuracy": 0.29891304347826086,
+            "count": 368
+          },
+          "SC": {
+            "accuracy": 0.13707165109034267,
+            "count": 321
+          },
+          "SS": {
+            "accuracy": 0.20614035087719298,
+            "count": 228
+          },
+          "UC": {
+            "accuracy": 0.3465160075329567,
+            "count": 531
+          },
+          "US": {
+            "accuracy": 0.0695364238410596,
+            "count": 302
+          }
+        }
+      },
+      "add_S3": {
+        "full_accuracy": 0.0,
+        "n_examples": 250,
+        "per_subtask": {
+          "SA": {
+            "accuracy": 0.2931596091205212,
+            "count": 307
+          },
+          "SC": {
+            "accuracy": 0.12027491408934708,
+            "count": 291
+          },
+          "SS": {
+            "accuracy": 0.3185840707964602,
+            "count": 113
+          },
+          "UC": {
+            "accuracy": 0.40993788819875776,
+            "count": 483
+          },
+          "US": {
+            "accuracy": 0.04316546762589928,
+            "count": 556
+          }
+        }
+      },
+      "add_S4": {
+        "full_accuracy": 0.0,
+        "n_examples": 250,
+        "per_subtask": {
+          "SA": {
+            "accuracy": 0.2857142857142857,
+            "count": 238
+          },
+          "SC": {
+            "accuracy": 0.08487084870848709,
+            "count": 271
+          },
+          "SS": {
+            "accuracy": 0.4576271186440678,
+            "count": 59
+          },
+          "UC": {
+            "accuracy": 0.5308641975308642,
+            "count": 405
+          },
+          "US": {
+            "accuracy": 0.04504504504504504,
+            "count": 777
+          }
+        }
+      },
+      "add_S5": {
+        "full_accuracy": 0.0,
+        "n_examples": 250,
+        "per_subtask": {
+          "SA": {
+            "accuracy": 0.588,
+            "count": 250
+          },
+          "SC": {
+            "accuracy": 0.084,
+            "count": 250
+          },
+          "UC": {
+            "accuracy": 0.488,
+            "count": 250
+          },
+          "US": {
+            "accuracy": 0.059,
+            "count": 1000
+          }
+        }
+      },
+      "add_S6": {
+        "full_accuracy": 0.0,
+        "n_examples": 250,
+        "per_subtask": {
+          "SC": {
+            "accuracy": 0.08,
+            "count": 250
+          },
+          "UC": {
+            "accuracy": 0.992,
+            "count": 250
+          },
+          "US": {
+            "accuracy": 0.1616,
+            "count": 1250
+          }
+        }
+      },
+      "add_random": {
+        "full_accuracy": 0.0,
+        "n_examples": 200,
+        "per_subtask": {
+          "SA": {
+            "accuracy": 0.2925170068027211,
+            "count": 441
+          },
+          "SC": {
+            "accuracy": 0.11356466876971609,
+            "count": 317
+          },
+          "SS": {
+            "accuracy": 0.2962962962962963,
+            "count": 54
+          },
+          "UC": {
+            "accuracy": 0.2725563909774436,
+            "count": 532
+          },
+          "US": {
+            "accuracy": 0.07142857142857142,
+            "count": 56
+          }
+        }
+      },
+      "add_C3": {
+        "full_accuracy": 0.0,
+        "n_examples": 250,
+        "per_subtask": {
+          "SA": {
+            "accuracy": 0.33866666666666667,
+            "count": 750
+          },
+          "SC": {
+            "accuracy": 0.044,
+            "count": 250
+          },
+          "UC": {
+            "accuracy": 0.17016806722689076,
+            "count": 476
+          },
+          "US": {
+            "accuracy": 0.0036496350364963502,
+            "count": 274
+          }
+        }
+      },
+      "add_C4": {
+        "full_accuracy": 0.0,
+        "n_examples": 250,
+        "per_subtask": {
+          "SA": {
+            "accuracy": 0.412,
+            "count": 500
+          },
+          "SC": {
+            "accuracy": 0.056,
+            "count": 250
+          },
+          "UC": {
+            "accuracy": 0.21331316187594554,
+            "count": 661
+          },
+          "US": {
+            "accuracy": 0.0,
+            "count": 339
+          }
+        }
+      },
+      "add_C5": {
+        "full_accuracy": 0.0,
+        "n_examples": 250,
+        "per_subtask": {
+          "SA": {
+            "accuracy": 0.62,
+            "count": 250
+          },
+          "SC": {
+            "accuracy": 0.092,
+            "count": 250
+          },
+          "UC": {
+            "accuracy": 0.22814207650273224,
+            "count": 732
+          },
+          "US": {
+            "accuracy": 0.0694980694980695,
+            "count": 518
+          }
+        }
+      },
+      "add_C6": {
+        "full_accuracy": 0.0,
+        "n_examples": 250,
+        "per_subtask": {
+          "SC": {
+            "accuracy": 0.1,
+            "count": 250
+          },
+          "UC": {
+            "accuracy": 0.3536036036036036,
+            "count": 888
+          },
+          "US": {
+            "accuracy": 0.1323529411764706,
+            "count": 612
+          }
+        }
+      },
+      "sub_M0": {
+        "full_accuracy": 0.0,
+        "n_examples": 250,
+        "per_subtask": {
+          "MD": {
+            "accuracy": 0.3282392026578073,
+            "count": 1505
+          },
+          "ME": {
+            "accuracy": 0.5918367346938775,
+            "count": 245
+          }
+        }
+      },
+      "sub_M1": {
+        "full_accuracy": 0.0,
+        "n_examples": 250,
+        "per_subtask": {
+          "MD": {
+            "accuracy": 0.4887955182072829,
+            "count": 714
+          },
+          "MB": {
+            "accuracy": 0.26737967914438504,
+            "count": 374
+          },
+          "ME": {
+            "accuracy": 0.52,
+            "count": 75
+          },
+          "UB": {
+            "accuracy": 0.1839863713798978,
+            "count": 587
+          }
+        }
+      },
+      "sub_M2": {
+        "full_accuracy": 0.0,
+        "n_examples": 250,
+        "per_subtask": {
+          "MD": {
+            "accuracy": 0.5379061371841155,
+            "count": 554
+          },
+          "MB": {
+            "accuracy": 0.18681318681318682,
+            "count": 273
+          },
+          "ME": {
+            "accuracy": 0.410958904109589,
+            "count": 219
+          },
+          "UB": {
+            "accuracy": 0.19534883720930232,
+            "count": 430
+          },
+          "UD": {
+            "accuracy": 0.4124087591240876,
+            "count": 274
+          }
+        }
+      },
+      "sub_M3": {
+        "full_accuracy": 0.0,
+        "n_examples": 250,
+        "per_subtask": {
+          "MD": {
+            "accuracy": 0.611353711790393,
+            "count": 458
+          },
+          "MB": {
+            "accuracy": 0.19157088122605365,
+            "count": 261
+          },
+          "ME": {
+            "accuracy": 0.41935483870967744,
+            "count": 124
+          },
+          "UB": {
+            "accuracy": 0.18814432989690721,
+            "count": 388
+          },
+          "UD": {
+            "accuracy": 0.23121387283236994,
+            "count": 519
+          }
+        }
+      },
+      "sub_M4": {
+        "full_accuracy": 0.0,
+        "n_examples": 250,
+        "per_subtask": {
+          "MD": {
+            "accuracy": 0.602,
+            "count": 500
+          },
+          "MB": {
+            "accuracy": 0.16,
+            "count": 250
+          },
+          "UB": {
+            "accuracy": 0.06,
+            "count": 250
+          },
+          "UD": {
+            "accuracy": 0.13466666666666666,
+            "count": 750
+          }
+        }
+      },
+      "sub_M5": {
+        "full_accuracy": 0.0,
+        "n_examples": 250,
+        "per_subtask": {
+          "MD": {
+            "accuracy": 1.0,
+            "count": 250
+          },
+          "MB": {
+            "accuracy": 0.072,
+            "count": 250
+          },
+          "UB": {
+            "accuracy": 0.084,
+            "count": 250
+          },
+          "UD": {
+            "accuracy": 0.413,
+            "count": 1000
+          }
+        }
+      },
+      "sub_random": {
+        "full_accuracy": 0.0,
+        "n_examples": 200,
+        "per_subtask": {
+          "MD": {
+            "accuracy": 0.47586206896551725,
+            "count": 580
+          },
+          "MB": {
+            "accuracy": 0.20599250936329588,
+            "count": 267
+          },
+          "ME": {
+            "accuracy": 0.5714285714285714,
+            "count": 63
+          },
+          "UB": {
+            "accuracy": 0.18080357142857142,
+            "count": 448
+          },
+          "UD": {
+            "accuracy": 0.30952380952380953,
+            "count": 42
+          }
+        }
+      },
+      "sub_B3": {
+        "full_accuracy": 0.0,
+        "n_examples": 250,
+        "per_subtask": {
+          "MD": {
+            "accuracy": 0.44,
+            "count": 750
+          },
+          "MB": {
+            "accuracy": 0.212,
+            "count": 250
+          },
+          "UB": {
+            "accuracy": 0.20078740157480315,
+            "count": 508
+          },
+          "UD": {
+            "accuracy": 0.16942148760330578,
+            "count": 242
+          }
+        }
+      },
+      "sub_B4": {
+        "full_accuracy": 0.0,
+        "n_examples": 250,
+        "per_subtask": {
+          "MD": {
+            "accuracy": 0.622,
+            "count": 500
+          },
+          "MB": {
+            "accuracy": 0.184,
+            "count": 250
+          },
+          "UB": {
+            "accuracy": 0.14006514657980457,
+            "count": 614
+          },
+          "UD": {
+            "accuracy": 0.19948186528497408,
+            "count": 386
+          }
+        }
+      },
+      "sub_B5": {
+        "full_accuracy": 0.0,
+        "n_examples": 250,
+        "per_subtask": {
+          "MD": {
+            "accuracy": 1.0,
+            "count": 250
+          },
+          "MB": {
+            "accuracy": 0.068,
+            "count": 250
+          },
+          "UB": {
+            "accuracy": 0.17631917631917632,
+            "count": 777
+          },
+          "UD": {
+            "accuracy": 0.3678646934460888,
+            "count": 473
+          }
+        }
+      }
+    },
+    "summary": {
+      "overall_accuracy": 0.0,
+      "total_examples": 5400,
+      "n_splits": 22
+    }
+  },
+  "sorl_overall_accuracy": 0.0,
+  "sft_overall_accuracy": 0.6812962962962963
+}

add_sub_sorl_v1_abs10_K1_10K/model.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:fdc469755c92ef1653fb3f6ae13eac568959c2ae6791d4bfece6a727903c7bc8
+size 650303660

add_sub_sorl_v1_abs10_K1_10K/train_config.json

@@ -0,0 +1,35 @@
+{
+  "mode": "sorl",
+  "ops": "add_sub",
+  "n_digits": 6,
+  "n_layer": 2,
+  "n_head": 3,
+  "n_embd": 510,
+  "abs_vocab": 10,
+  "K": 1,
+  "alpha_info_gain": 10.0,
+  "alpha_abs": 0.1,
+  "alpha_soft_zipf": 1.0,
+  "batch_size": 64,
+  "num_epochs": 10,
+  "dataset_size": 10000,
+  "lr": 8e-05,
+  "output_dir": "ckpt/sweep/as_sorl_abs10_K1_10K",
+  "device": "cuda",
+  "push_to_hub": true,
+  "no_wandb": false,
+  "n_params": 162499262,
+  "run_name": "add_sub_sorl_v1_abs10_K1_10K",
+  "git_commit": "800625019270114adcda289bbd550c4f1109a514",
+  "timestamp": "2026-04-11T21:33:55.579231+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": "ulvxvuq3",
+  "wandb_url": "https://wandb.ai/nlp_and_interpretability/sorl-arithmetic/runs/ulvxvuq3",
+  "final_accuracy": 0.0,
+  "sft_accuracy": 0.6812962962962963,
+  "eval_method": "ArithmeticEvaluator"
+}