Upload add_sub_sorl_v6_abs30_K1_25K

add_sub_sorl_v6_abs30_K1_25K/metrics.json

@@ -1717,6 +1717,7 @@
     "splits": {
       "add_S0": {
         "full_accuracy": 0.0,
+        "digit_accuracy": 0.1442857142857143,
         "n_examples": 100,
         "per_subtask": {
           "SA": {
@@ -1731,6 +1732,7 @@
       },
       "add_S1": {
         "full_accuracy": 0.0,
+        "digit_accuracy": 0.11571428571428571,
         "n_examples": 100,
         "per_subtask": {
           "SA": {
@@ -1753,6 +1755,7 @@
       },
       "add_S2": {
         "full_accuracy": 0.0,
+        "digit_accuracy": 0.12142857142857143,
         "n_examples": 100,
         "per_subtask": {
           "SA": {
@@ -1779,6 +1782,7 @@
       },
       "add_S3": {
         "full_accuracy": 0.0,
+        "digit_accuracy": 0.10428571428571429,
         "n_examples": 100,
         "per_subtask": {
           "SA": {
@@ -1805,6 +1809,7 @@
       },
       "add_S4": {
         "full_accuracy": 0.0,
+        "digit_accuracy": 0.13142857142857142,
         "n_examples": 100,
         "per_subtask": {
           "SA": {
@@ -1831,6 +1836,7 @@
       },
       "add_S5": {
         "full_accuracy": 0.0,
+        "digit_accuracy": 0.09857142857142857,
         "n_examples": 100,
         "per_subtask": {
           "SA": {
@@ -1853,6 +1859,7 @@
       },
       "add_S6": {
         "full_accuracy": 0.0,
+        "digit_accuracy": 0.11571428571428571,
         "n_examples": 100,
         "per_subtask": {
           "SC": {
@@ -1871,6 +1878,7 @@
       },
       "add_random": {
         "full_accuracy": 0.0,
+        "digit_accuracy": 0.13214285714285715,
         "n_examples": 200,
         "per_subtask": {
           "SA": {
@@ -1897,6 +1905,7 @@
       },
       "add_C3": {
         "full_accuracy": 0.0,
+        "digit_accuracy": 0.15142857142857144,
         "n_examples": 100,
         "per_subtask": {
           "SA": {
@@ -1919,6 +1928,7 @@
       },
       "add_C4": {
         "full_accuracy": 0.0,
+        "digit_accuracy": 0.15285714285714286,
         "n_examples": 100,
         "per_subtask": {
           "SA": {
@@ -1941,6 +1951,7 @@
       },
       "add_C5": {
         "full_accuracy": 0.0,
+        "digit_accuracy": 0.11571428571428571,
         "n_examples": 100,
         "per_subtask": {
           "SA": {
@@ -1963,6 +1974,7 @@
       },
       "add_C6": {
         "full_accuracy": 0.0,
+        "digit_accuracy": 0.11285714285714285,
         "n_examples": 100,
         "per_subtask": {
           "SC": {
@@ -1981,6 +1993,7 @@
       },
       "sub_M0": {
         "full_accuracy": 0.0,
+        "digit_accuracy": 0.12857142857142856,
         "n_examples": 100,
         "per_subtask": {
           "MD": {
@@ -1995,6 +2008,7 @@
       },
       "sub_M1": {
         "full_accuracy": 0.0,
+        "digit_accuracy": 0.13142857142857142,
         "n_examples": 100,
         "per_subtask": {
           "MD": {
@@ -2017,6 +2031,7 @@
       },
       "sub_M2": {
         "full_accuracy": 0.0,
+        "digit_accuracy": 0.12,
         "n_examples": 100,
         "per_subtask": {
           "MD": {
@@ -2043,6 +2058,7 @@
       },
       "sub_M3": {
         "full_accuracy": 0.0,
+        "digit_accuracy": 0.13714285714285715,
         "n_examples": 100,
         "per_subtask": {
           "MD": {
@@ -2069,6 +2085,7 @@
       },
       "sub_M4": {
         "full_accuracy": 0.0,
+        "digit_accuracy": 0.12571428571428572,
         "n_examples": 100,
         "per_subtask": {
           "MD": {
@@ -2091,6 +2108,7 @@
       },
       "sub_M5": {
         "full_accuracy": 0.0,
+        "digit_accuracy": 0.14142857142857143,
         "n_examples": 100,
         "per_subtask": {
           "MD": {
@@ -2113,6 +2131,7 @@
       },
       "sub_random": {
         "full_accuracy": 0.0,
+        "digit_accuracy": 0.13714285714285715,
         "n_examples": 200,
         "per_subtask": {
           "MD": {
@@ -2139,6 +2158,7 @@
       },
       "sub_B3": {
         "full_accuracy": 0.0,
+        "digit_accuracy": 0.14,
         "n_examples": 100,
         "per_subtask": {
           "MD": {
@@ -2161,6 +2181,7 @@
       },
       "sub_B4": {
         "full_accuracy": 0.0,
+        "digit_accuracy": 0.13714285714285715,
         "n_examples": 100,
         "per_subtask": {
           "MD": {
@@ -2183,6 +2204,7 @@
       },
       "sub_B5": {
         "full_accuracy": 0.0,
+        "digit_accuracy": 0.16428571428571428,
         "n_examples": 100,
         "per_subtask": {
           "MD": {
@@ -2206,6 +2228,7 @@
     },
     "summary": {
       "overall_accuracy": 0.0,
+      "digit_accuracy": 0.13023809523809524,
       "total_examples": 2400,
       "n_splits": 22
     }
@@ -2221,6 +2244,7 @@
     "splits": {
       "add_S0": {
         "full_accuracy": 0.99,
+        "digit_accuracy": 0.9985714285714286,
         "n_examples": 100,
         "per_subtask": {
           "SA": {
@@ -2235,6 +2259,7 @@
       },
       "add_S1": {
         "full_accuracy": 0.98,
+        "digit_accuracy": 0.9971428571428571,
         "n_examples": 100,
         "per_subtask": {
           "SA": {
@@ -2257,6 +2282,7 @@
       },
       "add_S2": {
         "full_accuracy": 0.96,
+        "digit_accuracy": 0.9942857142857143,
         "n_examples": 100,
         "per_subtask": {
           "SA": {
@@ -2283,6 +2309,7 @@
       },
       "add_S3": {
         "full_accuracy": 0.86,
+        "digit_accuracy": 0.9757142857142858,
         "n_examples": 100,
         "per_subtask": {
           "SA": {
@@ -2309,6 +2336,7 @@
       },
       "add_S4": {
         "full_accuracy": 0.68,
+        "digit_accuracy": 0.9328571428571428,
         "n_examples": 100,
         "per_subtask": {
           "SA": {
@@ -2335,6 +2363,7 @@
       },
       "add_S5": {
         "full_accuracy": 0.4,
+        "digit_accuracy": 0.8,
         "n_examples": 100,
         "per_subtask": {
           "SA": {
@@ -2357,6 +2386,7 @@
       },
       "add_S6": {
         "full_accuracy": 0.61,
+        "digit_accuracy": 0.8271428571428572,
         "n_examples": 100,
         "per_subtask": {
           "SC": {
@@ -2375,6 +2405,7 @@
       },
       "add_random": {
         "full_accuracy": 0.97,
+        "digit_accuracy": 0.9957142857142857,
         "n_examples": 200,
         "per_subtask": {
           "SA": {
@@ -2401,6 +2432,7 @@
       },
       "add_C3": {
         "full_accuracy": 0.86,
+        "digit_accuracy": 0.9742857142857143,
         "n_examples": 100,
         "per_subtask": {
           "SA": {
@@ -2423,6 +2455,7 @@
       },
       "add_C4": {
         "full_accuracy": 0.85,
+        "digit_accuracy": 0.9714285714285714,
         "n_examples": 100,
         "per_subtask": {
           "SA": {
@@ -2445,6 +2478,7 @@
       },
       "add_C5": {
         "full_accuracy": 0.85,
+        "digit_accuracy": 0.9685714285714285,
         "n_examples": 100,
         "per_subtask": {
           "SA": {
@@ -2467,6 +2501,7 @@
       },
       "add_C6": {
         "full_accuracy": 0.9,
+        "digit_accuracy": 0.9757142857142858,
         "n_examples": 100,
         "per_subtask": {
           "SC": {
@@ -2485,6 +2520,7 @@
       },
       "sub_M0": {
         "full_accuracy": 1.0,
+        "digit_accuracy": 1.0,
         "n_examples": 100,
         "per_subtask": {
           "MD": {
@@ -2499,6 +2535,7 @@
       },
       "sub_M1": {
         "full_accuracy": 1.0,
+        "digit_accuracy": 1.0,
         "n_examples": 100,
         "per_subtask": {
           "MD": {
@@ -2521,6 +2558,7 @@
       },
       "sub_M2": {
         "full_accuracy": 1.0,
+        "digit_accuracy": 1.0,
         "n_examples": 100,
         "per_subtask": {
           "MD": {
@@ -2547,6 +2585,7 @@
       },
       "sub_M3": {
         "full_accuracy": 0.76,
+        "digit_accuracy": 0.9614285714285714,
         "n_examples": 100,
         "per_subtask": {
           "MD": {
@@ -2573,6 +2612,7 @@
       },
       "sub_M4": {
         "full_accuracy": 0.43,
+        "digit_accuracy": 0.87,
         "n_examples": 100,
         "per_subtask": {
           "MD": {
@@ -2595,6 +2635,7 @@
       },
       "sub_M5": {
         "full_accuracy": 0.07,
+        "digit_accuracy": 0.7042857142857143,
         "n_examples": 100,
         "per_subtask": {
           "MD": {
@@ -2617,6 +2658,7 @@
       },
       "sub_random": {
         "full_accuracy": 0.985,
+        "digit_accuracy": 0.9971428571428571,
         "n_examples": 200,
         "per_subtask": {
           "MD": {
@@ -2643,6 +2685,7 @@
       },
       "sub_B3": {
         "full_accuracy": 0.85,
+        "digit_accuracy": 0.9785714285714285,
         "n_examples": 100,
         "per_subtask": {
           "MD": {
@@ -2665,6 +2708,7 @@
       },
       "sub_B4": {
         "full_accuracy": 0.8,
+        "digit_accuracy": 0.9657142857142857,
         "n_examples": 100,
         "per_subtask": {
           "MD": {
@@ -2687,6 +2731,7 @@
       },
       "sub_B5": {
         "full_accuracy": 0.8,
+        "digit_accuracy": 0.9614285714285714,
         "n_examples": 100,
         "per_subtask": {
           "MD": {
@@ -2710,6 +2755,7 @@
     },
     "summary": {
       "overall_accuracy": 0.815,
+      "digit_accuracy": 0.9516666666666667,
       "total_examples": 2400,
       "n_splits": 22
     }

add_sub_sorl_v6_abs30_K1_25K/train_config.json

@@ -69,15 +69,15 @@
   "no_wandb": false,
   "n_params": 162519662,
   "run_name": "add_sub_sorl_v6_abs30_K1_25K",
-  "git_commit": "17e935f460a7f9595b705c1d614101a6b0e520f7",
-  "timestamp": "2026-04-14T10:04:30.519578+00:00",
+  "git_commit": "ec82dee57b1073e52f06ffb0efedc7bce16fff21",
+  "timestamp": "2026-04-14T10:38:56.263883+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": "v6",
-  "wandb_run_id": "0l1r06x6",
-  "wandb_url": "https://wandb.ai/nlp_and_interpretability/sorl-arithmetic/runs/0l1r06x6",
+  "wandb_run_id": "55qc1hsd",
+  "wandb_url": "https://wandb.ai/nlp_and_interpretability/sorl-arithmetic/runs/55qc1hsd",
   "final_accuracy": 0.815,
   "sft_accuracy": 0.0,
   "eval_method": "ArithmeticEvaluator"