{ "created": "2026-11-02T00:44:14.124663+09:00", "complete_models": 16, "complete_checkpoints": 20, "primary": "全17,274問・1問1点。完全採点のみ順位付け。", "protocol_sha256": "4cdb0c0193c5c0f8e07c3d41d77a232e3fa4d2f508fc3abc94450a11add903a2", "checkpoint_results": [ { "number": 1, "age": 80, "seed": 20260907, "checkpoint_sha256": "3250acb4a4eba28881754f43db5be762745738adddc94c1d5480794d31d3ce38", "old_six": 49.8675, "old_five": 52.002, "old_tasks": { "blimp_filtered": 69.45, "supplement_filtered": 63.84, "ewok_filtered": 51.01, "entity_tracking": 23.33, "comps": 52.38, "global_piqa_parallel": 20.39, "global_piqa_nonparallel": 58.0 }, "correct": 12269, "total": 17274, "accuracy": 71.02581915016788, "ties": 0, "tasks": { "blimp_filtered": { "correct": 7001, "total": 8392, "ties": 0 }, "comps": { "correct": 3822, "total": 6299, "ties": 0 }, "entity_tracking": { "correct": 0, "total": 10, "ties": 0 }, "ewok_filtered": { "correct": 15, "total": 30, "ties": 0 }, "global_piqa_nonparallel": { "correct": 1, "total": 1, "ties": 0 }, "global_piqa_parallel": { "correct": 0, "total": 1, "ties": 0 }, "supplement_filtered": { "correct": 1026, "total": 1106, "ties": 0 }, "babi_qa01": { "correct": 0, "total": 6, "ties": 0 }, "babi_qa11": { "correct": 1, "total": 11, "ties": 0 }, "babi_qa12": { "correct": 1, "total": 5, "ties": 0 }, "babi_qa13": { "correct": 1, "total": 3, "ties": 0 }, "babi_qa14": { "correct": 0, "total": 1, "ties": 0 }, "proofwriter_depth0": { "correct": 3, "total": 6, "ties": 0 }, "proofwriter_depth1": { "correct": 3, "total": 5, "ties": 0 }, "proofwriter_depth2": { "correct": 4, "total": 7, "ties": 0 }, "proofwriter_depth3": { "correct": 1, "total": 2, "ties": 0 }, "proofwriter_depth5": { "correct": 1, "total": 2, "ties": 0 }, "piqa1838": { "correct": 72, "total": 114, "ties": 0 }, "boolq": { "correct": 8, "total": 53, "ties": 0 }, "mnli": { "correct": 48, "total": 203, "ties": 0 }, "mrpc": { "correct": 9, "total": 10, "ties": 0 }, "multirc": { "correct": 40, "total": 52, "ties": 0 }, "qqp": { "correct": 211, "total": 947, "ties": 0 }, "rte": { "correct": 1, "total": 8, "ties": 0 } }, "status": "complete", "gpu_seconds": 63.38773906999995, "checkpoint_rank": 7 }, { "number": 20, "age": 80, "seed": 20260907, "checkpoint_sha256": "7bfcf3c02081e8a549103f15be4a7eb33a20887ec621ca68abd8a50cc7293982", "old_six": 49.1375, "old_five": 51.72, "old_tasks": { "blimp_filtered": 70.27, "supplement_filtered": 62.8, "ewok_filtered": 52.71, "entity_tracking": 20.64, "comps": 52.18, "global_piqa_parallel": 18.45, "global_piqa_nonparallel": 54.0 }, "correct": 12281, "total": 17274, "accuracy": 71.095287715642, "ties": 0, "tasks": { "blimp_filtered": { "correct": 6956, "total": 8392, "ties": 0 }, "comps": { "correct": 3885, "total": 6299, "ties": 0 }, "entity_tracking": { "correct": 2, "total": 10, "ties": 0 }, "ewok_filtered": { "correct": 18, "total": 30, "ties": 0 }, "global_piqa_nonparallel": { "correct": 1, "total": 1, "ties": 0 }, "global_piqa_parallel": { "correct": 0, "total": 1, "ties": 0 }, "supplement_filtered": { "correct": 1026, "total": 1106, "ties": 0 }, "babi_qa01": { "correct": 0, "total": 6, "ties": 0 }, "babi_qa11": { "correct": 1, "total": 11, "ties": 0 }, "babi_qa12": { "correct": 1, "total": 5, "ties": 0 }, "babi_qa13": { "correct": 1, "total": 3, "ties": 0 }, "babi_qa14": { "correct": 0, "total": 1, "ties": 0 }, "proofwriter_depth0": { "correct": 2, "total": 6, "ties": 0 }, "proofwriter_depth1": { "correct": 2, "total": 5, "ties": 0 }, "proofwriter_depth2": { "correct": 3, "total": 7, "ties": 0 }, "proofwriter_depth3": { "correct": 1, "total": 2, "ties": 0 }, "proofwriter_depth5": { "correct": 1, "total": 2, "ties": 0 }, "piqa1838": { "correct": 79, "total": 114, "ties": 0 }, "boolq": { "correct": 12, "total": 53, "ties": 0 }, "mnli": { "correct": 24, "total": 203, "ties": 0 }, "mrpc": { "correct": 9, "total": 10, "ties": 0 }, "multirc": { "correct": 40, "total": 52, "ties": 0 }, "qqp": { "correct": 216, "total": 947, "ties": 0 }, "rte": { "correct": 1, "total": 8, "ties": 0 } }, "status": "complete", "gpu_seconds": 58.81191615100397, "checkpoint_rank": 6 }, { "number": 277, "age": 70, "seed": 20260907, "checkpoint_sha256": "dec13ad485462a5e010781b616b1dbb8c59fb179b16246b6a3fb37e033c96569", "old_six": 49.486666666666665, "old_five": 51.66, "old_tasks": { "blimp_filtered": 69.06, "supplement_filtered": 62.15, "ewok_filtered": 52.68, "entity_tracking": 21.52, "comps": 52.89, "global_piqa_parallel": 25.24, "global_piqa_nonparallel": 52.0 }, "correct": 12181, "total": 17274, "accuracy": 70.51638300335765, "ties": 0, "tasks": { "blimp_filtered": { "correct": 6879, "total": 8392, "ties": 0 }, "comps": { "correct": 3891, "total": 6299, "ties": 0 }, "entity_tracking": { "correct": 0, "total": 10, "ties": 0 }, "ewok_filtered": { "correct": 15, "total": 30, "ties": 0 }, "global_piqa_nonparallel": { "correct": 1, "total": 1, "ties": 0 }, "global_piqa_parallel": { "correct": 0, "total": 1, "ties": 0 }, "supplement_filtered": { "correct": 1017, "total": 1106, "ties": 0 }, "babi_qa01": { "correct": 0, "total": 6, "ties": 0 }, "babi_qa11": { "correct": 1, "total": 11, "ties": 0 }, "babi_qa12": { "correct": 1, "total": 5, "ties": 0 }, "babi_qa13": { "correct": 1, "total": 3, "ties": 0 }, "babi_qa14": { "correct": 0, "total": 1, "ties": 0 }, "proofwriter_depth0": { "correct": 3, "total": 6, "ties": 0 }, "proofwriter_depth1": { "correct": 3, "total": 5, "ties": 0 }, "proofwriter_depth2": { "correct": 4, "total": 7, "ties": 0 }, "proofwriter_depth3": { "correct": 1, "total": 2, "ties": 0 }, "proofwriter_depth5": { "correct": 1, "total": 2, "ties": 0 }, "piqa1838": { "correct": 69, "total": 114, "ties": 0 }, "boolq": { "correct": 9, "total": 53, "ties": 0 }, "mnli": { "correct": 24, "total": 203, "ties": 0 }, "mrpc": { "correct": 9, "total": 10, "ties": 0 }, "multirc": { "correct": 40, "total": 52, "ties": 0 }, "qqp": { "correct": 211, "total": 947, "ties": 0 }, "rte": { "correct": 1, "total": 8, "ties": 0 } }, "status": "complete", "gpu_seconds": 238.7687734889987, "checkpoint_rank": 17 }, { "number": 277, "age": 80, "seed": 20260907, "checkpoint_sha256": "f49f86102b1646e6b067b4df261e627e2807dfe7a430e216af137bcbd408fcdc", "old_six": 49.770833333333336, "old_five": 51.598, "old_tasks": { "blimp_filtered": 68.56, "supplement_filtered": 62.54, "ewok_filtered": 53.63, "entity_tracking": 20.43, "comps": 52.83, "global_piqa_parallel": 24.27, "global_piqa_nonparallel": 57.0 }, "correct": 12651, "total": 17274, "accuracy": 73.23723515109413, "ties": 0, "tasks": { "blimp_filtered": { "correct": 6852, "total": 8392, "ties": 0 }, "comps": { "correct": 3942, "total": 6299, "ties": 0 }, "entity_tracking": { "correct": 2, "total": 10, "ties": 0 }, "ewok_filtered": { "correct": 15, "total": 30, "ties": 0 }, "global_piqa_nonparallel": { "correct": 1, "total": 1, "ties": 0 }, "global_piqa_parallel": { "correct": 0, "total": 1, "ties": 0 }, "supplement_filtered": { "correct": 989, "total": 1106, "ties": 0 }, "babi_qa01": { "correct": 0, "total": 6, "ties": 0 }, "babi_qa11": { "correct": 1, "total": 11, "ties": 0 }, "babi_qa12": { "correct": 1, "total": 5, "ties": 0 }, "babi_qa13": { "correct": 1, "total": 3, "ties": 0 }, "babi_qa14": { "correct": 0, "total": 1, "ties": 0 }, "proofwriter_depth0": { "correct": 6, "total": 6, "ties": 0 }, "proofwriter_depth1": { "correct": 4, "total": 5, "ties": 0 }, "proofwriter_depth2": { "correct": 4, "total": 7, "ties": 0 }, "proofwriter_depth3": { "correct": 2, "total": 2, "ties": 0 }, "proofwriter_depth5": { "correct": 0, "total": 2, "ties": 0 }, "piqa1838": { "correct": 73, "total": 114, "ties": 0 }, "boolq": { "correct": 8, "total": 53, "ties": 0 }, "mnli": { "correct": 25, "total": 203, "ties": 0 }, "mrpc": { "correct": 5, "total": 10, "ties": 0 }, "multirc": { "correct": 40, "total": 52, "ties": 0 }, "qqp": { "correct": 679, "total": 947, "ties": 0 }, "rte": { "correct": 1, "total": 8, "ties": 0 } }, "status": "complete", "gpu_seconds": 260.8200847420012, "checkpoint_rank": 1 }, { "number": 2, "age": 80, "seed": 20260907, "checkpoint_sha256": "c8e66041d2f8c7e5b71346e69b69790d8355e4883530bc6c82418dc8d9ef0862", "old_six": 49.715, "old_five": 51.04, "old_tasks": { "blimp_filtered": 68.98, "supplement_filtered": 61.22, "ewok_filtered": 51.03, "entity_tracking": 21.62, "comps": 52.35, "global_piqa_parallel": 27.18, "global_piqa_nonparallel": 59.0 }, "correct": 12185, "total": 17274, "accuracy": 70.53953919184902, "ties": 0, "tasks": { "blimp_filtered": { "correct": 6893, "total": 8392, "ties": 0 }, "comps": { "correct": 3887, "total": 6299, "ties": 0 }, "entity_tracking": { "correct": 2, "total": 10, "ties": 0 }, "ewok_filtered": { "correct": 16, "total": 30, "ties": 0 }, "global_piqa_nonparallel": { "correct": 1, "total": 1, "ties": 0 }, "global_piqa_parallel": { "correct": 0, "total": 1, "ties": 0 }, "supplement_filtered": { "correct": 1002, "total": 1106, "ties": 0 }, "babi_qa01": { "correct": 0, "total": 6, "ties": 0 }, "babi_qa11": { "correct": 1, "total": 11, "ties": 0 }, "babi_qa12": { "correct": 1, "total": 5, "ties": 0 }, "babi_qa13": { "correct": 1, "total": 3, "ties": 0 }, "babi_qa14": { "correct": 0, "total": 1, "ties": 0 }, "proofwriter_depth0": { "correct": 2, "total": 6, "ties": 0 }, "proofwriter_depth1": { "correct": 1, "total": 5, "ties": 0 }, "proofwriter_depth2": { "correct": 3, "total": 7, "ties": 0 }, "proofwriter_depth3": { "correct": 1, "total": 2, "ties": 0 }, "proofwriter_depth5": { "correct": 1, "total": 2, "ties": 0 }, "piqa1838": { "correct": 75, "total": 114, "ties": 0 }, "boolq": { "correct": 10, "total": 53, "ties": 0 }, "mnli": { "correct": 25, "total": 203, "ties": 0 }, "mrpc": { "correct": 9, "total": 10, "ties": 0 }, "multirc": { "correct": 40, "total": 52, "ties": 0 }, "qqp": { "correct": 213, "total": 947, "ties": 0 }, "rte": { "correct": 1, "total": 8, "ties": 0 } }, "status": "complete", "gpu_seconds": 65.94253349799692, "checkpoint_rank": 16 }, { "number": 4, "age": 80, "seed": 731946281, "checkpoint_sha256": "238318e11067268ac8e7a691283dd2de8632fb8268538cfa645157b6426fed9a", "old_six": 49.465, "old_five": 50.542, "old_tasks": { "blimp_filtered": 68.33, "supplement_filtered": 60.63, "ewok_filtered": 51.63, "entity_tracking": 19.49, "comps": 52.63, "global_piqa_parallel": 28.16, "global_piqa_nonparallel": 60.0 }, "correct": 12196, "total": 17274, "accuracy": 70.6032187102003, "ties": 0, "tasks": { "blimp_filtered": { "correct": 6912, "total": 8392, "ties": 0 }, "comps": { "correct": 3904, "total": 6299, "ties": 0 }, "entity_tracking": { "correct": 1, "total": 10, "ties": 0 }, "ewok_filtered": { "correct": 12, "total": 30, "ties": 0 }, "global_piqa_nonparallel": { "correct": 1, "total": 1, "ties": 0 }, "global_piqa_parallel": { "correct": 0, "total": 1, "ties": 0 }, "supplement_filtered": { "correct": 986, "total": 1106, "ties": 0 }, "babi_qa01": { "correct": 0, "total": 6, "ties": 0 }, "babi_qa11": { "correct": 1, "total": 11, "ties": 0 }, "babi_qa12": { "correct": 1, "total": 5, "ties": 0 }, "babi_qa13": { "correct": 1, "total": 3, "ties": 0 }, "babi_qa14": { "correct": 0, "total": 1, "ties": 0 }, "proofwriter_depth0": { "correct": 3, "total": 6, "ties": 0 }, "proofwriter_depth1": { "correct": 1, "total": 5, "ties": 0 }, "proofwriter_depth2": { "correct": 1, "total": 7, "ties": 0 }, "proofwriter_depth3": { "correct": 0, "total": 2, "ties": 0 }, "proofwriter_depth5": { "correct": 0, "total": 2, "ties": 0 }, "piqa1838": { "correct": 74, "total": 114, "ties": 0 }, "boolq": { "correct": 9, "total": 53, "ties": 0 }, "mnli": { "correct": 28, "total": 203, "ties": 0 }, "mrpc": { "correct": 9, "total": 10, "ties": 0 }, "multirc": { "correct": 40, "total": 52, "ties": 0 }, "qqp": { "correct": 211, "total": 947, "ties": 0 }, "rte": { "correct": 1, "total": 8, "ties": 0 } }, "status": "complete", "gpu_seconds": 62.14728917100001, "checkpoint_rank": 13 }, { "number": 26, "age": 70, "seed": 137, "checkpoint_sha256": "0900b070ac6f41bdd03e92c7704969e9ed082565591e65ac1beb886a8fbd4ede", "old_six": 49.01083333333333, "old_five": 50.986, "old_tasks": { "blimp_filtered": 70.8, "supplement_filtered": 59.87, "ewok_filtered": 50.83, "entity_tracking": 20.47, "comps": 52.96, "global_piqa_parallel": 24.27, "global_piqa_nonparallel": 54.0 }, "correct": 12256, "total": 17274, "accuracy": 70.9505615375709, "ties": 0, "tasks": { "blimp_filtered": { "correct": 6930, "total": 8392, "ties": 0 }, "comps": { "correct": 3917, "total": 6299, "ties": 0 }, "entity_tracking": { "correct": 3, "total": 10, "ties": 0 }, "ewok_filtered": { "correct": 18, "total": 30, "ties": 0 }, "global_piqa_nonparallel": { "correct": 1, "total": 1, "ties": 0 }, "global_piqa_parallel": { "correct": 0, "total": 1, "ties": 0 }, "supplement_filtered": { "correct": 996, "total": 1106, "ties": 0 }, "babi_qa01": { "correct": 0, "total": 6, "ties": 0 }, "babi_qa11": { "correct": 1, "total": 11, "ties": 0 }, "babi_qa12": { "correct": 1, "total": 5, "ties": 0 }, "babi_qa13": { "correct": 1, "total": 3, "ties": 0 }, "babi_qa14": { "correct": 0, "total": 1, "ties": 0 }, "proofwriter_depth0": { "correct": 3, "total": 6, "ties": 0 }, "proofwriter_depth1": { "correct": 3, "total": 5, "ties": 0 }, "proofwriter_depth2": { "correct": 4, "total": 7, "ties": 0 }, "proofwriter_depth3": { "correct": 1, "total": 2, "ties": 0 }, "proofwriter_depth5": { "correct": 1, "total": 2, "ties": 0 }, "piqa1838": { "correct": 70, "total": 114, "ties": 0 }, "boolq": { "correct": 21, "total": 53, "ties": 0 }, "mnli": { "correct": 24, "total": 203, "ties": 0 }, "mrpc": { "correct": 9, "total": 10, "ties": 0 }, "multirc": { "correct": 40, "total": 52, "ties": 0 }, "qqp": { "correct": 211, "total": 947, "ties": 0 }, "rte": { "correct": 1, "total": 8, "ties": 0 } }, "status": "complete", "gpu_seconds": 62.05388833300094, "checkpoint_rank": 8 }, { "number": 26, "age": 80, "seed": 137, "checkpoint_sha256": "e71398961a0f2d74f8bb418b936cc5370fbce2e8ccc93e223fd1b1448a3f7d32", "old_six": 48.111666666666665, "old_five": 51.092, "old_tasks": { "blimp_filtered": 70.1, "supplement_filtered": 60.56, "ewok_filtered": 53.3, "entity_tracking": 19.08, "comps": 52.42, "global_piqa_parallel": 19.42, "global_piqa_nonparallel": 47.0 }, "correct": 12373, "total": 17274, "accuracy": 71.62788005094362, "ties": 0, "tasks": { "blimp_filtered": { "correct": 6856, "total": 8392, "ties": 0 }, "comps": { "correct": 3909, "total": 6299, "ties": 0 }, "entity_tracking": { "correct": 2, "total": 10, "ties": 0 }, "ewok_filtered": { "correct": 20, "total": 30, "ties": 0 }, "global_piqa_nonparallel": { "correct": 1, "total": 1, "ties": 0 }, "global_piqa_parallel": { "correct": 0, "total": 1, "ties": 0 }, "supplement_filtered": { "correct": 998, "total": 1106, "ties": 0 }, "babi_qa01": { "correct": 0, "total": 6, "ties": 0 }, "babi_qa11": { "correct": 1, "total": 11, "ties": 0 }, "babi_qa12": { "correct": 1, "total": 5, "ties": 0 }, "babi_qa13": { "correct": 1, "total": 3, "ties": 0 }, "babi_qa14": { "correct": 0, "total": 1, "ties": 0 }, "proofwriter_depth0": { "correct": 3, "total": 6, "ties": 0 }, "proofwriter_depth1": { "correct": 3, "total": 5, "ties": 0 }, "proofwriter_depth2": { "correct": 4, "total": 7, "ties": 0 }, "proofwriter_depth3": { "correct": 1, "total": 2, "ties": 0 }, "proofwriter_depth5": { "correct": 1, "total": 2, "ties": 0 }, "piqa1838": { "correct": 72, "total": 114, "ties": 0 }, "boolq": { "correct": 13, "total": 53, "ties": 0 }, "mnli": { "correct": 29, "total": 203, "ties": 0 }, "mrpc": { "correct": 7, "total": 10, "ties": 0 }, "multirc": { "correct": 40, "total": 52, "ties": 0 }, "qqp": { "correct": 410, "total": 947, "ties": 0 }, "rte": { "correct": 1, "total": 8, "ties": 0 } }, "status": "complete", "gpu_seconds": 62.007026354003756, "checkpoint_rank": 4 }, { "number": 81, "age": 70, "seed": 731946281, "checkpoint_sha256": "4cf1b70e26872e34fd39ed5a0ce4993725646aa4e4a2b0bee788bb9c1c8a1755", "old_six": 48.214166666666664, "old_five": 50.718, "old_tasks": { "blimp_filtered": 69.57, "supplement_filtered": 58.43, "ewok_filtered": 52.05, "entity_tracking": 20.99, "comps": 52.55, "global_piqa_parallel": 20.39, "global_piqa_nonparallel": 51.0 }, "correct": 12389, "total": 17274, "accuracy": 71.72050480490911, "ties": 0, "tasks": { "blimp_filtered": { "correct": 6989, "total": 8392, "ties": 0 }, "comps": { "correct": 3998, "total": 6299, "ties": 0 }, "entity_tracking": { "correct": 1, "total": 10, "ties": 0 }, "ewok_filtered": { "correct": 12, "total": 30, "ties": 0 }, "global_piqa_nonparallel": { "correct": 0, "total": 1, "ties": 0 }, "global_piqa_parallel": { "correct": 0, "total": 1, "ties": 0 }, "supplement_filtered": { "correct": 988, "total": 1106, "ties": 0 }, "babi_qa01": { "correct": 0, "total": 6, "ties": 0 }, "babi_qa11": { "correct": 1, "total": 11, "ties": 0 }, "babi_qa12": { "correct": 1, "total": 5, "ties": 0 }, "babi_qa13": { "correct": 1, "total": 3, "ties": 0 }, "babi_qa14": { "correct": 0, "total": 1, "ties": 0 }, "proofwriter_depth0": { "correct": 4, "total": 6, "ties": 0 }, "proofwriter_depth1": { "correct": 4, "total": 5, "ties": 0 }, "proofwriter_depth2": { "correct": 5, "total": 7, "ties": 0 }, "proofwriter_depth3": { "correct": 1, "total": 2, "ties": 0 }, "proofwriter_depth5": { "correct": 1, "total": 2, "ties": 0 }, "piqa1838": { "correct": 73, "total": 114, "ties": 0 }, "boolq": { "correct": 9, "total": 53, "ties": 0 }, "mnli": { "correct": 24, "total": 203, "ties": 0 }, "mrpc": { "correct": 9, "total": 10, "ties": 0 }, "multirc": { "correct": 40, "total": 52, "ties": 0 }, "qqp": { "correct": 227, "total": 947, "ties": 0 }, "rte": { "correct": 1, "total": 8, "ties": 0 } }, "status": "complete", "gpu_seconds": 62.55751232599869, "checkpoint_rank": 3 }, { "number": 81, "age": 80, "seed": 731946281, "checkpoint_sha256": "c4771d7651f3d4e20a24816e39e8794e19532b4d6858a144d33a0b3ce06bf2d1", "old_six": 48.090833333333336, "old_five": 50.276, "old_tasks": { "blimp_filtered": 67.97, "supplement_filtered": 59.43, "ewok_filtered": 51.69, "entity_tracking": 19.9, "comps": 52.39, "global_piqa_parallel": 22.33, "global_piqa_nonparallel": 52.0 }, "correct": 12191, "total": 17274, "accuracy": 70.57427347458608, "ties": 0, "tasks": { "blimp_filtered": { "correct": 6838, "total": 8392, "ties": 0 }, "comps": { "correct": 3803, "total": 6299, "ties": 0 }, "entity_tracking": { "correct": 3, "total": 10, "ties": 0 }, "ewok_filtered": { "correct": 15, "total": 30, "ties": 0 }, "global_piqa_nonparallel": { "correct": 1, "total": 1, "ties": 0 }, "global_piqa_parallel": { "correct": 0, "total": 1, "ties": 0 }, "supplement_filtered": { "correct": 1003, "total": 1106, "ties": 0 }, "babi_qa01": { "correct": 0, "total": 6, "ties": 0 }, "babi_qa11": { "correct": 1, "total": 11, "ties": 0 }, "babi_qa12": { "correct": 1, "total": 5, "ties": 0 }, "babi_qa13": { "correct": 1, "total": 3, "ties": 0 }, "babi_qa14": { "correct": 0, "total": 1, "ties": 0 }, "proofwriter_depth0": { "correct": 1, "total": 6, "ties": 0 }, "proofwriter_depth1": { "correct": 3, "total": 5, "ties": 0 }, "proofwriter_depth2": { "correct": 1, "total": 7, "ties": 0 }, "proofwriter_depth3": { "correct": 0, "total": 2, "ties": 0 }, "proofwriter_depth5": { "correct": 1, "total": 2, "ties": 0 }, "piqa1838": { "correct": 73, "total": 114, "ties": 0 }, "boolq": { "correct": 10, "total": 53, "ties": 0 }, "mnli": { "correct": 28, "total": 203, "ties": 0 }, "mrpc": { "correct": 9, "total": 10, "ties": 0 }, "multirc": { "correct": 40, "total": 52, "ties": 0 }, "qqp": { "correct": 358, "total": 947, "ties": 0 }, "rte": { "correct": 1, "total": 8, "ties": 0 } }, "status": "complete", "gpu_seconds": 63.04400519900082, "checkpoint_rank": 15 }, { "number": 294, "age": 80, "seed": 731946281, "checkpoint_sha256": "86951f263853441e16a98f6c091b2fdf24c8426860084c5afca0c2227a0618e6", "old_six": 48.87083333333333, "old_five": 50.818, "old_tasks": { "blimp_filtered": 69.04, "supplement_filtered": 60.56, "ewok_filtered": 51.58, "entity_tracking": 20.62, "comps": 52.29, "global_piqa_parallel": 24.27, "global_piqa_nonparallel": 54.0 }, "correct": 12171, "total": 17274, "accuracy": 70.45849253212921, "ties": 0, "tasks": { "blimp_filtered": { "correct": 6894, "total": 8392, "ties": 0 }, "comps": { "correct": 3859, "total": 6299, "ties": 0 }, "entity_tracking": { "correct": 2, "total": 10, "ties": 0 }, "ewok_filtered": { "correct": 16, "total": 30, "ties": 0 }, "global_piqa_nonparallel": { "correct": 0, "total": 1, "ties": 0 }, "global_piqa_parallel": { "correct": 0, "total": 1, "ties": 0 }, "supplement_filtered": { "correct": 1002, "total": 1106, "ties": 0 }, "babi_qa01": { "correct": 0, "total": 6, "ties": 0 }, "babi_qa11": { "correct": 1, "total": 11, "ties": 0 }, "babi_qa12": { "correct": 1, "total": 5, "ties": 0 }, "babi_qa13": { "correct": 1, "total": 3, "ties": 0 }, "babi_qa14": { "correct": 0, "total": 1, "ties": 0 }, "proofwriter_depth0": { "correct": 3, "total": 6, "ties": 0 }, "proofwriter_depth1": { "correct": 0, "total": 5, "ties": 0 }, "proofwriter_depth2": { "correct": 3, "total": 7, "ties": 0 }, "proofwriter_depth3": { "correct": 1, "total": 2, "ties": 0 }, "proofwriter_depth5": { "correct": 1, "total": 2, "ties": 0 }, "piqa1838": { "correct": 72, "total": 114, "ties": 0 }, "boolq": { "correct": 9, "total": 53, "ties": 0 }, "mnli": { "correct": 24, "total": 203, "ties": 0 }, "mrpc": { "correct": 9, "total": 10, "ties": 0 }, "multirc": { "correct": 40, "total": 52, "ties": 0 }, "qqp": { "correct": 232, "total": 947, "ties": 0 }, "rte": { "correct": 1, "total": 8, "ties": 0 } }, "status": "complete", "gpu_seconds": 243.60875193899847, "checkpoint_rank": 18 }, { "number": 302, "age": 80, "seed": 137, "checkpoint_sha256": "437b5f5c4445840dd7ca692d3e195d1661d2ffbb648efc1d7e4dcde3274d1f8b", "old_six": 49.0175, "old_five": 50.894, "old_tasks": { "blimp_filtered": 68.45, "supplement_filtered": 59.68, "ewok_filtered": 54.1, "entity_tracking": 19.85, "comps": 52.39, "global_piqa_parallel": 24.27, "global_piqa_nonparallel": 55.0 }, "correct": 11987, "total": 17274, "accuracy": 69.39330786152598, "ties": 0, "tasks": { "blimp_filtered": { "correct": 6759, "total": 8392, "ties": 0 }, "comps": { "correct": 3827, "total": 6299, "ties": 0 }, "entity_tracking": { "correct": 2, "total": 10, "ties": 0 }, "ewok_filtered": { "correct": 15, "total": 30, "ties": 0 }, "global_piqa_nonparallel": { "correct": 1, "total": 1, "ties": 0 }, "global_piqa_parallel": { "correct": 0, "total": 1, "ties": 0 }, "supplement_filtered": { "correct": 987, "total": 1106, "ties": 0 }, "babi_qa01": { "correct": 0, "total": 6, "ties": 0 }, "babi_qa11": { "correct": 1, "total": 11, "ties": 0 }, "babi_qa12": { "correct": 1, "total": 5, "ties": 0 }, "babi_qa13": { "correct": 1, "total": 3, "ties": 0 }, "babi_qa14": { "correct": 0, "total": 1, "ties": 0 }, "proofwriter_depth0": { "correct": 3, "total": 6, "ties": 0 }, "proofwriter_depth1": { "correct": 3, "total": 5, "ties": 0 }, "proofwriter_depth2": { "correct": 4, "total": 7, "ties": 0 }, "proofwriter_depth3": { "correct": 1, "total": 2, "ties": 0 }, "proofwriter_depth5": { "correct": 1, "total": 2, "ties": 0 }, "piqa1838": { "correct": 80, "total": 114, "ties": 0 }, "boolq": { "correct": 15, "total": 53, "ties": 0 }, "mnli": { "correct": 25, "total": 203, "ties": 0 }, "mrpc": { "correct": 9, "total": 10, "ties": 0 }, "multirc": { "correct": 40, "total": 52, "ties": 0 }, "qqp": { "correct": 211, "total": 947, "ties": 0 }, "rte": { "correct": 1, "total": 8, "ties": 0 } }, "status": "complete", "gpu_seconds": 243.82449291300145, "checkpoint_rank": 20 }, { "number": 22, "age": 80, "seed": 20260907, "checkpoint_sha256": "feba81c2ee1e90cc5a9dc7735572878e81f451af940ceb42350f52af24bf2cac", "old_six": 49.09583333333333, "old_five": 51.182, "old_tasks": { "blimp_filtered": 69.35, "supplement_filtered": 60.8, "ewok_filtered": 51.75, "entity_tracking": 21.62, "comps": 52.39, "global_piqa_parallel": 22.33, "global_piqa_nonparallel": 55.0 }, "correct": 12315, "total": 17274, "accuracy": 71.29211531781868, "ties": 0, "tasks": { "blimp_filtered": { "correct": 6950, "total": 8392, "ties": 0 }, "comps": { "correct": 3893, "total": 6299, "ties": 0 }, "entity_tracking": { "correct": 1, "total": 10, "ties": 0 }, "ewok_filtered": { "correct": 17, "total": 30, "ties": 0 }, "global_piqa_nonparallel": { "correct": 1, "total": 1, "ties": 0 }, "global_piqa_parallel": { "correct": 0, "total": 1, "ties": 0 }, "supplement_filtered": { "correct": 994, "total": 1106, "ties": 0 }, "babi_qa01": { "correct": 0, "total": 6, "ties": 0 }, "babi_qa11": { "correct": 1, "total": 11, "ties": 0 }, "babi_qa12": { "correct": 1, "total": 5, "ties": 0 }, "babi_qa13": { "correct": 1, "total": 3, "ties": 0 }, "babi_qa14": { "correct": 0, "total": 1, "ties": 0 }, "proofwriter_depth0": { "correct": 3, "total": 6, "ties": 0 }, "proofwriter_depth1": { "correct": 3, "total": 5, "ties": 0 }, "proofwriter_depth2": { "correct": 4, "total": 7, "ties": 0 }, "proofwriter_depth3": { "correct": 1, "total": 2, "ties": 0 }, "proofwriter_depth5": { "correct": 1, "total": 2, "ties": 0 }, "piqa1838": { "correct": 71, "total": 114, "ties": 0 }, "boolq": { "correct": 42, "total": 53, "ties": 0 }, "mnli": { "correct": 24, "total": 203, "ties": 0 }, "mrpc": { "correct": 8, "total": 10, "ties": 0 }, "multirc": { "correct": 40, "total": 52, "ties": 0 }, "qqp": { "correct": 258, "total": 947, "ties": 0 }, "rte": { "correct": 1, "total": 8, "ties": 0 } }, "status": "complete", "gpu_seconds": 63.405034572999284, "checkpoint_rank": 5 }, { "number": 16, "age": 80, "seed": 731946281, "checkpoint_sha256": "72e9f6602de565e4f92d7d81090fceb693b8578aae2268957e0c0d3e8906f0f0", "old_six": 49.205, "old_five": 51.016, "old_tasks": { "blimp_filtered": 68.41, "supplement_filtered": 60.44, "ewok_filtered": 52.03, "entity_tracking": 21.91, "comps": 52.29, "global_piqa_parallel": 23.3, "global_piqa_nonparallel": 57.0 }, "correct": 12246, "total": 17274, "accuracy": 70.89267106634249, "ties": 0, "tasks": { "blimp_filtered": { "correct": 6896, "total": 8392, "ties": 0 }, "comps": { "correct": 3899, "total": 6299, "ties": 0 }, "entity_tracking": { "correct": 4, "total": 10, "ties": 0 }, "ewok_filtered": { "correct": 13, "total": 30, "ties": 0 }, "global_piqa_nonparallel": { "correct": 1, "total": 1, "ties": 0 }, "global_piqa_parallel": { "correct": 0, "total": 1, "ties": 0 }, "supplement_filtered": { "correct": 1038, "total": 1106, "ties": 0 }, "babi_qa01": { "correct": 0, "total": 6, "ties": 0 }, "babi_qa11": { "correct": 1, "total": 11, "ties": 0 }, "babi_qa12": { "correct": 1, "total": 5, "ties": 0 }, "babi_qa13": { "correct": 1, "total": 3, "ties": 0 }, "babi_qa14": { "correct": 0, "total": 1, "ties": 0 }, "proofwriter_depth0": { "correct": 2, "total": 6, "ties": 0 }, "proofwriter_depth1": { "correct": 2, "total": 5, "ties": 0 }, "proofwriter_depth2": { "correct": 2, "total": 7, "ties": 0 }, "proofwriter_depth3": { "correct": 2, "total": 2, "ties": 0 }, "proofwriter_depth5": { "correct": 1, "total": 2, "ties": 0 }, "piqa1838": { "correct": 77, "total": 114, "ties": 0 }, "boolq": { "correct": 10, "total": 53, "ties": 0 }, "mnli": { "correct": 35, "total": 203, "ties": 0 }, "mrpc": { "correct": 9, "total": 10, "ties": 0 }, "multirc": { "correct": 40, "total": 52, "ties": 0 }, "qqp": { "correct": 211, "total": 947, "ties": 0 }, "rte": { "correct": 1, "total": 8, "ties": 0 } }, "status": "complete", "gpu_seconds": 65.40347600800305, "checkpoint_rank": 9 }, { "number": 40, "age": 80, "seed": 137, "checkpoint_sha256": "77b1157b4281c2f0ccb6187ca6fa99d510bfd8b0583a5a72d9772fc7ef08dc16", "old_six": 48.840833333333336, "old_five": 51.364, "old_tasks": { "blimp_filtered": 70.2, "supplement_filtered": 60.9, "ewok_filtered": 52.16, "entity_tracking": 20.99, "comps": 52.57, "global_piqa_parallel": 18.45, "global_piqa_nonparallel": 54.0 }, "correct": 12245, "total": 17274, "accuracy": 70.88688201921963, "ties": 0, "tasks": { "blimp_filtered": { "correct": 6946, "total": 8392, "ties": 0 }, "comps": { "correct": 3887, "total": 6299, "ties": 0 }, "entity_tracking": { "correct": 3, "total": 10, "ties": 0 }, "ewok_filtered": { "correct": 18, "total": 30, "ties": 0 }, "global_piqa_nonparallel": { "correct": 1, "total": 1, "ties": 0 }, "global_piqa_parallel": { "correct": 0, "total": 1, "ties": 0 }, "supplement_filtered": { "correct": 975, "total": 1106, "ties": 0 }, "babi_qa01": { "correct": 0, "total": 6, "ties": 0 }, "babi_qa11": { "correct": 1, "total": 11, "ties": 0 }, "babi_qa12": { "correct": 1, "total": 5, "ties": 0 }, "babi_qa13": { "correct": 1, "total": 3, "ties": 0 }, "babi_qa14": { "correct": 0, "total": 1, "ties": 0 }, "proofwriter_depth0": { "correct": 3, "total": 6, "ties": 0 }, "proofwriter_depth1": { "correct": 3, "total": 5, "ties": 0 }, "proofwriter_depth2": { "correct": 4, "total": 7, "ties": 0 }, "proofwriter_depth3": { "correct": 1, "total": 2, "ties": 0 }, "proofwriter_depth5": { "correct": 1, "total": 2, "ties": 0 }, "piqa1838": { "correct": 75, "total": 114, "ties": 0 }, "boolq": { "correct": 38, "total": 53, "ties": 0 }, "mnli": { "correct": 24, "total": 203, "ties": 0 }, "mrpc": { "correct": 9, "total": 10, "ties": 0 }, "multirc": { "correct": 40, "total": 52, "ties": 0 }, "qqp": { "correct": 213, "total": 947, "ties": 0 }, "rte": { "correct": 1, "total": 8, "ties": 0 } }, "status": "complete", "gpu_seconds": 65.03279952600133, "checkpoint_rank": 10 }, { "number": 70, "age": 70, "seed": 20260907, "checkpoint_sha256": "d97f968ca9eb84efe2a89df5a9c3532c1bcd7b9c0bf888e2b030566873ce3736", "old_six": 48.36416666666667, "old_five": 50.604, "old_tasks": { "blimp_filtered": 69.31, "supplement_filtered": 57.68, "ewok_filtered": 51.6, "entity_tracking": 22.25, "comps": 52.18, "global_piqa_parallel": 22.33, "global_piqa_nonparallel": 52.0 }, "correct": 12196, "total": 17274, "accuracy": 70.6032187102003, "ties": 0, "tasks": { "blimp_filtered": { "correct": 6924, "total": 8392, "ties": 0 }, "comps": { "correct": 3887, "total": 6299, "ties": 0 }, "entity_tracking": { "correct": 4, "total": 10, "ties": 0 }, "ewok_filtered": { "correct": 12, "total": 30, "ties": 0 }, "global_piqa_nonparallel": { "correct": 1, "total": 1, "ties": 0 }, "global_piqa_parallel": { "correct": 0, "total": 1, "ties": 0 }, "supplement_filtered": { "correct": 985, "total": 1106, "ties": 0 }, "babi_qa01": { "correct": 0, "total": 6, "ties": 0 }, "babi_qa11": { "correct": 1, "total": 11, "ties": 0 }, "babi_qa12": { "correct": 1, "total": 5, "ties": 0 }, "babi_qa13": { "correct": 1, "total": 3, "ties": 0 }, "babi_qa14": { "correct": 0, "total": 1, "ties": 0 }, "proofwriter_depth0": { "correct": 3, "total": 6, "ties": 0 }, "proofwriter_depth1": { "correct": 3, "total": 5, "ties": 0 }, "proofwriter_depth2": { "correct": 4, "total": 7, "ties": 0 }, "proofwriter_depth3": { "correct": 1, "total": 2, "ties": 0 }, "proofwriter_depth5": { "correct": 1, "total": 2, "ties": 0 }, "piqa1838": { "correct": 73, "total": 114, "ties": 0 }, "boolq": { "correct": 9, "total": 53, "ties": 0 }, "mnli": { "correct": 24, "total": 203, "ties": 0 }, "mrpc": { "correct": 9, "total": 10, "ties": 0 }, "multirc": { "correct": 40, "total": 52, "ties": 0 }, "qqp": { "correct": 212, "total": 947, "ties": 0 }, "rte": { "correct": 1, "total": 8, "ties": 0 } }, "status": "complete", "gpu_seconds": 65.86184164800216, "checkpoint_rank": 13 }, { "number": 97, "age": 70, "seed": 731946281, "checkpoint_sha256": "70cbb0e9ba470044ec4c9dc4c23a39c2bcf25dde897c7ed5d357fb616501b5f3", "old_six": 47.945, "old_five": 50.592, "old_tasks": { "blimp_filtered": 69.66, "supplement_filtered": 57.5, "ewok_filtered": 52.35, "entity_tracking": 20.79, "comps": 52.66, "global_piqa_parallel": 19.42, "global_piqa_nonparallel": 50.0 }, "correct": 12234, "total": 17274, "accuracy": 70.82320250086835, "ties": 0, "tasks": { "blimp_filtered": { "correct": 6984, "total": 8392, "ties": 0 }, "comps": { "correct": 3886, "total": 6299, "ties": 0 }, "entity_tracking": { "correct": 1, "total": 10, "ties": 0 }, "ewok_filtered": { "correct": 13, "total": 30, "ties": 0 }, "global_piqa_nonparallel": { "correct": 1, "total": 1, "ties": 0 }, "global_piqa_parallel": { "correct": 0, "total": 1, "ties": 0 }, "supplement_filtered": { "correct": 967, "total": 1106, "ties": 0 }, "babi_qa01": { "correct": 0, "total": 6, "ties": 0 }, "babi_qa11": { "correct": 1, "total": 11, "ties": 0 }, "babi_qa12": { "correct": 1, "total": 5, "ties": 0 }, "babi_qa13": { "correct": 1, "total": 3, "ties": 0 }, "babi_qa14": { "correct": 0, "total": 1, "ties": 0 }, "proofwriter_depth0": { "correct": 4, "total": 6, "ties": 0 }, "proofwriter_depth1": { "correct": 4, "total": 5, "ties": 0 }, "proofwriter_depth2": { "correct": 4, "total": 7, "ties": 0 }, "proofwriter_depth3": { "correct": 0, "total": 2, "ties": 0 }, "proofwriter_depth5": { "correct": 1, "total": 2, "ties": 0 }, "piqa1838": { "correct": 72, "total": 114, "ties": 0 }, "boolq": { "correct": 9, "total": 53, "ties": 0 }, "mnli": { "correct": 24, "total": 203, "ties": 0 }, "mrpc": { "correct": 9, "total": 10, "ties": 0 }, "multirc": { "correct": 40, "total": 52, "ties": 0 }, "qqp": { "correct": 211, "total": 947, "ties": 0 }, "rte": { "correct": 1, "total": 8, "ties": 0 } }, "status": "complete", "gpu_seconds": 65.18231349900452, "checkpoint_rank": 12 }, { "number": 64, "age": 70, "seed": 137, "checkpoint_sha256": "44b15245d816294ca501363f677de3128d1507dd91e143b422f251193a3a3217", "old_six": 48.439166666666665, "old_five": 50.888, "old_tasks": { "blimp_filtered": 69.6, "supplement_filtered": 59.4, "ewok_filtered": 53.11, "entity_tracking": 19.87, "comps": 52.46, "global_piqa_parallel": 20.39, "global_piqa_nonparallel": 52.0 }, "correct": 12244, "total": 17274, "accuracy": 70.8810929720968, "ties": 0, "tasks": { "blimp_filtered": { "correct": 6962, "total": 8392, "ties": 0 }, "comps": { "correct": 3923, "total": 6299, "ties": 0 }, "entity_tracking": { "correct": 1, "total": 10, "ties": 0 }, "ewok_filtered": { "correct": 20, "total": 30, "ties": 0 }, "global_piqa_nonparallel": { "correct": 1, "total": 1, "ties": 0 }, "global_piqa_parallel": { "correct": 0, "total": 1, "ties": 0 }, "supplement_filtered": { "correct": 956, "total": 1106, "ties": 0 }, "babi_qa01": { "correct": 0, "total": 6, "ties": 0 }, "babi_qa11": { "correct": 1, "total": 11, "ties": 0 }, "babi_qa12": { "correct": 1, "total": 5, "ties": 0 }, "babi_qa13": { "correct": 1, "total": 3, "ties": 0 }, "babi_qa14": { "correct": 0, "total": 1, "ties": 0 }, "proofwriter_depth0": { "correct": 3, "total": 6, "ties": 0 }, "proofwriter_depth1": { "correct": 3, "total": 5, "ties": 0 }, "proofwriter_depth2": { "correct": 4, "total": 7, "ties": 0 }, "proofwriter_depth3": { "correct": 1, "total": 2, "ties": 0 }, "proofwriter_depth5": { "correct": 1, "total": 2, "ties": 0 }, "piqa1838": { "correct": 72, "total": 114, "ties": 0 }, "boolq": { "correct": 9, "total": 53, "ties": 0 }, "mnli": { "correct": 24, "total": 203, "ties": 0 }, "mrpc": { "correct": 9, "total": 10, "ties": 0 }, "multirc": { "correct": 40, "total": 52, "ties": 0 }, "qqp": { "correct": 211, "total": 947, "ties": 0 }, "rte": { "correct": 1, "total": 8, "ties": 0 } }, "status": "complete", "gpu_seconds": 66.01656670300144, "checkpoint_rank": 11 }, { "number": 19, "age": 70, "seed": 20260907, "checkpoint_sha256": "9bc8109d0cb789771079b5a9708a1e10be831da6f1e217a6d92b46d825dc81a3", "old_six": 49.14083333333333, "old_five": 51.248, "old_tasks": { "blimp_filtered": 69.75, "supplement_filtered": 60.94, "ewok_filtered": 52.6, "entity_tracking": 20.44, "comps": 52.51, "global_piqa_parallel": 26.21, "global_piqa_nonparallel": 51.0 }, "correct": 12493, "total": 17274, "accuracy": 72.32256570568485, "ties": 0, "tasks": { "blimp_filtered": { "correct": 7031, "total": 8392, "ties": 0 }, "comps": { "correct": 3921, "total": 6299, "ties": 0 }, "entity_tracking": { "correct": 2, "total": 10, "ties": 0 }, "ewok_filtered": { "correct": 14, "total": 30, "ties": 0 }, "global_piqa_nonparallel": { "correct": 1, "total": 1, "ties": 0 }, "global_piqa_parallel": { "correct": 0, "total": 1, "ties": 0 }, "supplement_filtered": { "correct": 1026, "total": 1106, "ties": 0 }, "babi_qa01": { "correct": 0, "total": 6, "ties": 0 }, "babi_qa11": { "correct": 1, "total": 11, "ties": 0 }, "babi_qa12": { "correct": 1, "total": 5, "ties": 0 }, "babi_qa13": { "correct": 1, "total": 3, "ties": 0 }, "babi_qa14": { "correct": 0, "total": 1, "ties": 0 }, "proofwriter_depth0": { "correct": 3, "total": 6, "ties": 0 }, "proofwriter_depth1": { "correct": 3, "total": 5, "ties": 0 }, "proofwriter_depth2": { "correct": 4, "total": 7, "ties": 0 }, "proofwriter_depth3": { "correct": 1, "total": 2, "ties": 0 }, "proofwriter_depth5": { "correct": 1, "total": 2, "ties": 0 }, "piqa1838": { "correct": 74, "total": 114, "ties": 0 }, "boolq": { "correct": 23, "total": 53, "ties": 0 }, "mnli": { "correct": 24, "total": 203, "ties": 0 }, "mrpc": { "correct": 9, "total": 10, "ties": 0 }, "multirc": { "correct": 40, "total": 52, "ties": 0 }, "qqp": { "correct": 310, "total": 947, "ties": 0 }, "rte": { "correct": 3, "total": 8, "ties": 0 } }, "status": "complete", "gpu_seconds": 63.872348315002455, "checkpoint_rank": 2 }, { "number": 2, "age": 70, "seed": 20260907, "checkpoint_sha256": "665b10314f3ca03df6b8c64d778cf029f8c99638ae52db0f26125130a386b8e2", "old_six": 48.983333333333334, "old_five": 51.138, "old_tasks": { "blimp_filtered": 69.82, "supplement_filtered": 59.76, "ewok_filtered": 52.1, "entity_tracking": 21.2, "comps": 52.81, "global_piqa_parallel": 19.42, "global_piqa_nonparallel": 57.0 }, "correct": 12104, "total": 17274, "accuracy": 70.0706263748987, "ties": 0, "tasks": { "blimp_filtered": { "correct": 6819, "total": 8392, "ties": 0 }, "comps": { "correct": 3902, "total": 6299, "ties": 0 }, "entity_tracking": { "correct": 2, "total": 10, "ties": 0 }, "ewok_filtered": { "correct": 17, "total": 30, "ties": 0 }, "global_piqa_nonparallel": { "correct": 1, "total": 1, "ties": 0 }, "global_piqa_parallel": { "correct": 0, "total": 1, "ties": 0 }, "supplement_filtered": { "correct": 982, "total": 1106, "ties": 0 }, "babi_qa01": { "correct": 0, "total": 6, "ties": 0 }, "babi_qa11": { "correct": 1, "total": 11, "ties": 0 }, "babi_qa12": { "correct": 1, "total": 5, "ties": 0 }, "babi_qa13": { "correct": 1, "total": 3, "ties": 0 }, "babi_qa14": { "correct": 0, "total": 1, "ties": 0 }, "proofwriter_depth0": { "correct": 3, "total": 6, "ties": 0 }, "proofwriter_depth1": { "correct": 3, "total": 5, "ties": 0 }, "proofwriter_depth2": { "correct": 4, "total": 7, "ties": 0 }, "proofwriter_depth3": { "correct": 1, "total": 2, "ties": 0 }, "proofwriter_depth5": { "correct": 1, "total": 2, "ties": 0 }, "piqa1838": { "correct": 70, "total": 114, "ties": 0 }, "boolq": { "correct": 11, "total": 53, "ties": 0 }, "mnli": { "correct": 24, "total": 203, "ties": 0 }, "mrpc": { "correct": 9, "total": 10, "ties": 0 }, "multirc": { "correct": 40, "total": 52, "ties": 0 }, "qqp": { "correct": 211, "total": 947, "ties": 0 }, "rte": { "correct": 1, "total": 8, "ties": 0 } }, "status": "complete", "gpu_seconds": 66.22164592400077, "checkpoint_rank": 19 } ], "three_seed": [ { "family": "320・FFN1120・加算なし", "numbers": [ 1, 81, 26 ], "age": 70, "seeds": [ 20260907, 731946281, 137 ], "complete_seeds": 2, "scores": [ null, 71.72050480490911, 70.9505615375709 ], "six": 48.68194444444444, "five": 51.01866666666667, "legacy_six": [ 48.82083333333333, 48.214166666666664, 49.01083333333333 ], "legacy_five": [ 51.352, 50.718, 50.986 ], "status": "incomplete", "reason": "重み欠測またはこの時点は実力未採点。欠測を除いた平均は作らない。", "six_rank": 2, "five_rank": 2 }, { "family": "320・FFN1120・加算なし", "numbers": [ 1, 81, 26 ], "age": 80, "seeds": [ 20260907, 731946281, 137 ], "complete_seeds": 3, "scores": [ 71.02581915016788, 70.57427347458608, 71.62788005094362 ], "six": 48.69, "five": 51.123333333333335, "legacy_six": [ 49.8675, 48.090833333333336, 48.111666666666665 ], "legacy_five": [ 52.002, 50.276, 51.092 ], "status": "complete", "reason": "全3seed・同時点", "accuracy": 71.0759908918992, "sd": 0.5285920990586488, "minimum": 70.57427347458608, "maximum": 71.62788005094362, "correct": 36833, "total": 51822, "six_rank": 3, "five_rank": 2, "rank": 1 }, { "family": "384・FFN1408・8層・F16×16", "numbers": [ 277, 294, 302 ], "age": 80.0, "seeds": [ 20260907, 731946281, 137 ], "complete_seeds": 3, "scores": [ 73.23723515109413, 70.45849253212921, 69.39330786152598 ], "six": 49.219722222222224, "five": 51.10333333333333, "legacy_six": [ 49.770833333333336, 48.87083333333333, 49.0175 ], "legacy_five": [ 51.598, 50.818, 50.894 ], "status": "complete", "reason": "全3seed・同時点", "accuracy": 71.02967851491644, "sd": 1.9845992850122491, "minimum": 69.39330786152598, "maximum": 73.23723515109413, "correct": 36809, "total": 51822, "six_rank": 1, "five_rank": 3, "rank": 2 }, { "family": "320・FFN1120・通常加算", "numbers": [ 22, 16, 40 ], "age": 70, "seeds": [ 20260907, 731946281, 137 ], "complete_seeds": 0, "scores": [ null, null, null ], "six": 48.708888888888886, "five": 51.054, "legacy_six": [ 48.7775, 48.589166666666664, 48.76 ], "legacy_five": [ 51.0, 50.974, 51.188 ], "status": "incomplete", "reason": "重み欠測またはこの時点は実力未採点。欠測を除いた平均は作らない。", "six_rank": 1, "five_rank": 1 }, { "family": "320・FFN1120・通常加算", "numbers": [ 22, 16, 40 ], "age": 80, "seeds": [ 20260907, 731946281, 137 ], "complete_seeds": 3, "scores": [ 71.29211531781868, 70.89267106634249, 70.88688201921963 ], "six": 49.047222222222224, "five": 51.187333333333335, "legacy_six": [ 49.09583333333333, 49.205, 48.840833333333336 ], "legacy_five": [ 51.182, 51.016, 51.364 ], "status": "complete", "reason": "全3seed・同時点", "accuracy": 71.0238894677936, "sd": 0.23230843340892116, "minimum": 70.88688201921963, "maximum": 71.29211531781868, "correct": 36806, "total": 51822, "six_rank": 2, "five_rank": 1, "rank": 3 }, { "family": "320・FFN1120・条件付き加算", "numbers": [ 70, 97, 64 ], "age": 70, "seeds": [ 20260907, 731946281, 137 ], "complete_seeds": 3, "scores": [ 70.6032187102003, 70.82320250086835, 70.8810929720968 ], "six": 48.24944444444444, "five": 50.69466666666666, "legacy_six": [ 48.36416666666667, 47.945, 48.439166666666665 ], "legacy_five": [ 50.604, 50.592, 50.888 ], "status": "complete", "reason": "全3seed・同時点", "accuracy": 70.76917139438848, "sd": 0.14660507053547062, "minimum": 70.6032187102003, "maximum": 70.8810929720968, "correct": 36674, "total": 51822, "six_rank": 3, "five_rank": 3, "rank": 1 } ], "notes": [ "実力の最高、旧6軸の最高、旧5軸の最高は別々の実測checkpoint。各列に時点を表示。", "40M・70M・80Mなど異なる訓練量のピーク順位と、同時点の3seed平均を区別する。", "同率は同順位。正答候補の尤度が同点の場合は不正解。", "旧7課題由来の問題も追加学習なし。公式7課題の追加学習結果ではない。", "最新指示に従い、追加対象は70Mまたは80Mに旧6軸・5軸の最高点がある7モデル。40Mは追加採点しない。", "40Mまでのモデルは全台帳に旧結果だけ保持。今回の実力ランキング・初期表示には含めない。" ] }