| {"step": 845000, "seed_holdout": {"answer_byte_loss": 0.005443127962936972, "teacher_forced_byte_accuracy": 1.0, "generation_exact_match": 1.0, "generation_truncated": 0, "episodes": 12}, "composition_ood": {"answer_byte_loss": 1.6167928483751086, "teacher_forced_byte_accuracy": 0.4666666666666667, "generation_exact_match": 0.0, "generation_truncated": 1, "episodes": 3}, "families": [{"group": "seed_holdout", "family": "gen_three_rule_comp", "surface": "symbolic", "exact": true, "generated_bytes": 11, "target_bytes": 11}, {"group": "seed_holdout", "family": "gen_sort_rows", "surface": "symbolic", "exact": true, "generated_bytes": 54, "target_bytes": 54}, {"group": "seed_holdout", "family": "gen_negative_space", "surface": "symbolic", "exact": true, "generated_bytes": 41, "target_bytes": 41}, {"group": "seed_holdout", "family": "gen_occlusion_complete", "surface": "symbolic", "exact": true, "generated_bytes": 40, "target_bytes": 40}, {"group": "seed_holdout", "family": "gen_single_fact", "surface": "symbolic", "exact": true, "generated_bytes": 2, "target_bytes": 2}, {"group": "seed_holdout", "family": "gen_two_hop", "surface": "symbolic", "exact": true, "generated_bytes": 2, "target_bytes": 2}, {"group": "seed_holdout", "family": "gen_multi_hop", "surface": "symbolic", "exact": true, "generated_bytes": 2, "target_bytes": 2}, {"group": "seed_holdout", "family": "gen_inverse", "surface": "instruction_json", "exact": true, "generated_bytes": 6, "target_bytes": 6}, {"group": "seed_holdout", "family": "gen_transitivity", "surface": "symbolic", "exact": true, "generated_bytes": 2, "target_bytes": 2}, {"group": "seed_holdout", "family": "gen_symmetric", "surface": "instruction_json", "exact": true, "generated_bytes": 5, "target_bytes": 5}, {"group": "seed_holdout", "family": "gen_temporal_order", "surface": "symbolic", "exact": true, "generated_bytes": 2, "target_bytes": 2}, {"group": "seed_holdout", "family": "gen_relative_rank", "surface": "symbolic", "exact": true, "generated_bytes": 2, "target_bytes": 2}, {"group": "composition_ood", "family": "compositional_execution", "surface": "symbolic", "exact": false, "generated_bytes": 24, "target_bytes": 18}, {"group": "composition_ood", "family": "compositional_execution", "surface": "symbolic", "exact": false, "generated_bytes": 28, "target_bytes": 56}, {"group": "composition_ood", "family": "compositional_execution", "surface": "symbolic", "exact": false, "generated_bytes": 192, "target_bytes": 16}]} |
| {"step": 850000, "seed_holdout": {"answer_byte_loss": 0.08846963951273702, "teacher_forced_byte_accuracy": 0.972972972972973, "generation_exact_match": 0.9166666666666666, "generation_truncated": 0, "episodes": 12}, "composition_ood": {"answer_byte_loss": 1.7895467758178711, "teacher_forced_byte_accuracy": 0.45555555555555555, "generation_exact_match": 0.0, "generation_truncated": 0, "episodes": 3}, "families": [{"group": "seed_holdout", "family": "gen_relation_comp", "surface": "symbolic", "exact": true, "generated_bytes": 2, "target_bytes": 2}, {"group": "seed_holdout", "family": "gen_exactly_one", "surface": "symbolic", "exact": true, "generated_bytes": 2, "target_bytes": 2}, {"group": "seed_holdout", "family": "gen_at_least", "surface": "symbolic", "exact": true, "generated_bytes": 2, "target_bytes": 2}, {"group": "seed_holdout", "family": "gen_at_most", "surface": "symbolic", "exact": true, "generated_bytes": 2, "target_bytes": 2}, {"group": "seed_holdout", "family": "gen_quantifier", "surface": "instruction_json", "exact": true, "generated_bytes": 6, "target_bytes": 6}, {"group": "seed_holdout", "family": "gen_ordering_csp", "surface": "symbolic", "exact": true, "generated_bytes": 2, "target_bytes": 2}, {"group": "seed_holdout", "family": "gen_assignment_matching", "surface": "symbolic", "exact": true, "generated_bytes": 2, "target_bytes": 2}, {"group": "seed_holdout", "family": "gen_small_sat", "surface": "instruction_json", "exact": false, "generated_bytes": 10, "target_bytes": 8}, {"group": "seed_holdout", "family": "gen_contradiction", "surface": "instruction_json", "exact": true, "generated_bytes": 5, "target_bytes": 5}, {"group": "seed_holdout", "family": "gen_kv_lookup", "surface": "instruction_json", "exact": true, "generated_bytes": 2, "target_bytes": 2}, {"group": "seed_holdout", "family": "gen_role_filler", "surface": "instruction_json", "exact": true, "generated_bytes": 2, "target_bytes": 2}, {"group": "seed_holdout", "family": "gen_assoc_distractors", "surface": "instruction_json", "exact": true, "generated_bytes": 2, "target_bytes": 2}, {"group": "composition_ood", "family": "compositional_execution", "surface": "symbolic", "exact": false, "generated_bytes": 6, "target_bytes": 18}, {"group": "composition_ood", "family": "compositional_execution", "surface": "symbolic", "exact": false, "generated_bytes": 21, "target_bytes": 56}, {"group": "composition_ood", "family": "compositional_execution", "surface": "symbolic", "exact": false, "generated_bytes": 15, "target_bytes": 16}]} |
| {"step": 855000, "seed_holdout": {"answer_byte_loss": 0.10454198174835953, "teacher_forced_byte_accuracy": 0.9699248120300752, "generation_exact_match": 0.75, "generation_truncated": 0, "episodes": 12}, "composition_ood": {"answer_byte_loss": 1.9086882485283745, "teacher_forced_byte_accuracy": 0.37777777777777777, "generation_exact_match": 0.0, "generation_truncated": 0, "episodes": 3}, "families": [{"group": "seed_holdout", "family": "gen_nested_lookup", "surface": "instruction_json", "exact": true, "generated_bytes": 2, "target_bytes": 2}, {"group": "seed_holdout", "family": "gen_mutable_dictionary", "surface": "instruction_json", "exact": true, "generated_bytes": 2, "target_bytes": 2}, {"group": "seed_holdout", "family": "gen_queue", "surface": "instruction_json", "exact": true, "generated_bytes": 21, "target_bytes": 21}, {"group": "seed_holdout", "family": "gen_deque", "surface": "instruction_json", "exact": true, "generated_bytes": 18, "target_bytes": 18}, {"group": "seed_holdout", "family": "gen_delayed_recall", "surface": "instruction_json", "exact": true, "generated_bytes": 2, "target_bytes": 2}, {"group": "seed_holdout", "family": "gen_register_swap", "surface": "instruction_json", "exact": false, "generated_bytes": 8, "target_bytes": 8}, {"group": "seed_holdout", "family": "gen_register_update", "surface": "instruction_json", "exact": true, "generated_bytes": 8, "target_bytes": 8}, {"group": "seed_holdout", "family": "gen_conditional_branch", "surface": "instruction_json", "exact": true, "generated_bytes": 8, "target_bytes": 8}, {"group": "seed_holdout", "family": "gen_loop_counter", "surface": "instruction_json", "exact": false, "generated_bytes": 20, "target_bytes": 12}, {"group": "seed_holdout", "family": "gen_multi_register", "surface": "instruction_json", "exact": false, "generated_bytes": 8, "target_bytes": 8}, {"group": "seed_holdout", "family": "gen_program_trace", "surface": "instruction_json", "exact": true, "generated_bytes": 42, "target_bytes": 42}, {"group": "seed_holdout", "family": "gen_stack_expr", "surface": "instruction_json", "exact": true, "generated_bytes": 2, "target_bytes": 2}, {"group": "composition_ood", "family": "compositional_execution", "surface": "symbolic", "exact": false, "generated_bytes": 37, "target_bytes": 18}, {"group": "composition_ood", "family": "compositional_execution", "surface": "symbolic", "exact": false, "generated_bytes": 12, "target_bytes": 56}, {"group": "composition_ood", "family": "compositional_execution", "surface": "symbolic", "exact": false, "generated_bytes": 15, "target_bytes": 16}]} |
|
|
