Upload extrapolation artifact: 050302 OPD Ref-KL OP13-20 step200 score report

050302/evals/opd050302_refkl_s200_merged_hf/graph_results/opd050302_refkl_s200_mergedhf_paper_eval_r113_score_report.jsonl

@@ -0,0 +1 @@
+{"gold_dir": "/mnt/hdfs/output/work/opd050302_refkl_s200_mergedhf_paper_eval_r113/gold_graphs", "generated_path": "/mnt/hdfs/output/work/opd050302_refkl_s200_mergedhf_paper_eval_r113", "results_dir": "/mnt/hdfs/output/work/opd050302_refkl_s200_mergedhf_paper_eval_r113/graph_results", "per_file": [{"file": "/mnt/hdfs/output/work/opd050302_refkl_s200_mergedhf_paper_eval_r113/opd050302_refkl_s200_merged_hf_ood_generations.jsonl", "total": 1280000, "missing_solution": 0, "missing_gold": 0, "parse_fail": 0, "perfect": 202356, "value_mismatch": 1043156, "dependency_mismatch": 1000555, "answer_mismatch": 840967, "extra_nodes": 107655, "missing_nodes": 581408, "perfect_with_extra": 495, "unused_gold_entries": 9000, "template_stats": {"crazy_zootopia": {"total": 430720, "value_mismatch": 350208, "dependency_mismatch": 336133, "answer_mismatch": 281767, "extra_nodes": 37939, "perfect": 68910, "missing_nodes": 194363, "perfect_with_extra": 145}, "teachers_in_school": {"total": 430464, "perfect": 70442, "dependency_mismatch": 331990, "value_mismatch": 348710, "answer_mismatch": 280800, "missing_nodes": 194129, "extra_nodes": 34529, "perfect_with_extra": 111}, "movie_festival_awards": {"total": 418816, "perfect": 63004, "value_mismatch": 344238, "dependency_mismatch": 332432, "answer_mismatch": 278400, "extra_nodes": 35187, "missing_nodes": 192916, "perfect_with_extra": 239}}, "results_path": "/mnt/hdfs/output/work/opd050302_refkl_s200_mergedhf_paper_eval_r113/graph_results/opd050302_refkl_s200_merged_hf_ood_generations.jsonl.graph_results.jsonl"}, {"file": "/mnt/hdfs/output/work/opd050302_refkl_s200_mergedhf_paper_eval_r113/opd050302_refkl_s200_merged_hf_id_generations.jsonl", "total": 1152000, "missing_solution": 0, "missing_gold": 0, "parse_fail": 0, "perfect": 672559, "value_mismatch": 454953, "dependency_mismatch": 231178, "answer_mismatch": 334831, "extra_nodes": 96884, "missing_nodes": 50221, "perfect_with_extra": 13296, "unused_gold_entries": 10000, "template_stats": {"crazy_zootopia": {"total": 379008, "perfect": 223136, "value_mismatch": 147432, "dependency_mismatch": 75445, "answer_mismatch": 108475, "extra_nodes": 33813, "missing_nodes": 16180, "perfect_with_extra": 5017}, "teachers_in_school": {"total": 390656, "perfect": 227331, "dependency_mismatch": 75809, "extra_nodes": 27794, "value_mismatch": 155883, "answer_mismatch": 114027, "missing_nodes": 15076, "perfect_with_extra": 2207}, "movie_festival_awards": {"total": 382336, "perfect": 222092, "dependency_mismatch": 79924, "value_mismatch": 151638, "answer_mismatch": 112329, "extra_nodes": 35277, "missing_nodes": 18965, "perfect_with_extra": 6072}}, "results_path": "/mnt/hdfs/output/work/opd050302_refkl_s200_mergedhf_paper_eval_r113/graph_results/opd050302_refkl_s200_merged_hf_id_generations.jsonl.graph_results.jsonl"}], "aggregate": {"files": 2, "total": 2432000, "missing_solution": 0, "missing_gold": 0, "parse_fail": 0, "perfect": 874915, "value_mismatch": 1498109, "dependency_mismatch": 1231733, "answer_mismatch": 1175798, "extra_nodes": 204539, "missing_nodes": 631629, "perfect_with_extra": 13791, "skipped": 0, "unused_gold_entries": 0, "aggregated": true, "report_path": "/mnt/hdfs/output/work/opd050302_refkl_s200_mergedhf_paper_eval_r113/graph_results/opd050302_refkl_s200_mergedhf_paper_eval_r113_score_report.jsonl", "template_stats": {"crazy_zootopia": {"total": 809728, "value_mismatch": 497640, "dependency_mismatch": 411578, "answer_mismatch": 390242, "extra_nodes": 71752, "perfect": 292046, "missing_nodes": 210543, "perfect_with_extra": 5162}, "teachers_in_school": {"total": 821120, "perfect": 297773, "dependency_mismatch": 407799, "value_mismatch": 504593, "answer_mismatch": 394827, "missing_nodes": 209205, "extra_nodes": 62323, "perfect_with_extra": 2318}, "movie_festival_awards": {"total": 801152, "perfect": 285096, "value_mismatch": 495876, "dependency_mismatch": 412356, "answer_mismatch": 390729, "extra_nodes": 70464, "missing_nodes": 211881, "perfect_with_extra": 6311}}}}