File size: 3,570 Bytes
a48e00f | 1 2 | {"gold_dir": "results_decrypted/composition/op-difficulty-10B/graphs", "generated_path": "results/Qwen3-100M/composition", "results_dir": "results/Qwen3-100M/composition/graphs", "per_file": [{"file": "results/Qwen3-100M/composition/checkpoint-21414_id_generations.jsonl", "total": 1152000, "missing_solution": 0, "missing_gold": 0, "parse_fail": 0, "perfect": 1000024, "value_mismatch": 26024, "dependency_mismatch": 149100, "answer_mismatch": 23802, "extra_nodes": 11190, "missing_nodes": 10769, "perfect_with_extra": 313, "unused_gold_entries": 10000, "template_stats": {"crazy_zootopia": {"total": 379008, "perfect": 329006, "dependency_mismatch": 49118, "value_mismatch": 9389, "answer_mismatch": 8732, "missing_nodes": 3471, "extra_nodes": 3651, "perfect_with_extra": 103}, "teachers_in_school": {"total": 390656, "perfect": 339613, "dependency_mismatch": 50020, "value_mismatch": 8367, "answer_mismatch": 7574, "extra_nodes": 3807, "missing_nodes": 3843, "perfect_with_extra": 47}, "movie_festival_awards": {"total": 382336, "perfect": 331405, "value_mismatch": 8268, "dependency_mismatch": 49962, "answer_mismatch": 7496, "extra_nodes": 3732, "missing_nodes": 3455, "perfect_with_extra": 163}}, "results_path": "results/Qwen3-100M/composition/graphs/checkpoint-21414_id_generations.jsonl.graph_results.jsonl"}, {"file": "results/Qwen3-100M/composition/checkpoint-21414_ood_generations.jsonl", "total": 1280000, "missing_solution": 0, "missing_gold": 0, "parse_fail": 0, "perfect": 177121, "value_mismatch": 938826, "dependency_mismatch": 1078759, "answer_mismatch": 803644, "extra_nodes": 52171, "missing_nodes": 681350, "perfect_with_extra": 60, "unused_gold_entries": 9000, "template_stats": {"crazy_zootopia": {"total": 430720, "value_mismatch": 316566, "dependency_mismatch": 365270, "answer_mismatch": 271691, "extra_nodes": 18289, "perfect": 58140, "missing_nodes": 241666, "perfect_with_extra": 16}, "movie_festival_awards": {"total": 418816, "dependency_mismatch": 352010, "value_mismatch": 305912, "answer_mismatch": 262289, "perfect": 58778, "missing_nodes": 214731, "extra_nodes": 18046, "perfect_with_extra": 2}, "teachers_in_school": {"total": 430464, "perfect": 60203, "dependency_mismatch": 361479, "value_mismatch": 316348, "answer_mismatch": 269664, "extra_nodes": 15836, "missing_nodes": 224953, "perfect_with_extra": 42}}, "results_path": "results/Qwen3-100M/composition/graphs/checkpoint-21414_ood_generations.jsonl.graph_results.jsonl"}], "aggregate": {"files": 2, "total": 2432000, "missing_solution": 0, "missing_gold": 0, "parse_fail": 0, "perfect": 1177145, "value_mismatch": 964850, "dependency_mismatch": 1227859, "answer_mismatch": 827446, "extra_nodes": 63361, "missing_nodes": 692119, "perfect_with_extra": 373, "skipped": 0, "unused_gold_entries": 0, "aggregated": true, "report_path": "results/Qwen3-100M/composition/graphs/composition_score_report.jsonl", "template_stats": {"crazy_zootopia": {"total": 809728, "perfect": 387146, "dependency_mismatch": 414388, "value_mismatch": 325955, "answer_mismatch": 280423, "missing_nodes": 245137, "extra_nodes": 21940, "perfect_with_extra": 119}, "teachers_in_school": {"total": 821120, "perfect": 399816, "dependency_mismatch": 411499, "value_mismatch": 324715, "answer_mismatch": 277238, "extra_nodes": 19643, "missing_nodes": 228796, "perfect_with_extra": 89}, "movie_festival_awards": {"total": 801152, "perfect": 390183, "value_mismatch": 314180, "dependency_mismatch": 401972, "answer_mismatch": 269785, "extra_nodes": 21778, "missing_nodes": 218186, "perfect_with_extra": 165}}}}
|