{
  "args": {
    "runs": "main_teacher_softpos_low_s900",
    "eval_size": 256,
    "eval_lengths": "24",
    "eval_seed": 92001,
    "eval_batch_size": 8,
    "output_csv": "experiments/qwen_teacher_distilled_slot_compiler/analysis/selected_retest_metrics.csv",
    "output_json": "experiments/qwen_teacher_distilled_slot_compiler/analysis/selected_retest_results.json",
    "append_existing": true
  },
  "rows": [
    {
      "run": "main_control_light_state_s900",
      "variant": "copy_trace_state_scheduled",
      "selected_step": "800",
      "selected_tag": "eval_trace_state_scheduled_long",
      "state_loss_schedule": "constant",
      "teacher_position_loss_weight": "0.0",
      "teacher_rep_loss_weight": "0.0",
      "selection_metric": "paired_len24_executor_accuracy",
      "selection_value": "0.3046875",
      "split": "fresh_standard_len24",
      "n": "256.0",
      "direct_accuracy": "nan",
      "direct_target_mass": "nan",
      "executor_accuracy": "0.3046875",
      "executor_target_mass": "0.010309270379366353",
      "init_accuracy": "1.0",
      "init_pos_accuracy": "1.0",
      "op_accuracy": "1.0",
      "arg_accuracy": "0.951171875",
      "op_pos_accuracy": "1.0",
      "arg_pos_accuracy": "0.9510091145833334",
      "program_exact": "0.30078125",
      "state_accuracy": "0.57177734375",
      "state_all_exact": "0.30078125",
      "state_prefix_fraction": "0.5711263020833334",
      "state_mean_correct_prefix": "13.70703125",
      "pair_true_answer_consistency": "",
      "pair_true_length_consistency": "",
      "executor_pair_answer_consistency": "",
      "executor_pair_both_correct": "",
      "compiler_pair_program_consistency": "",
      "compiler_pair_state_consistency": ""
    },
    {
      "run": "main_control_light_state_s900",
      "variant": "copy_trace_state_scheduled",
      "selected_step": "800",
      "selected_tag": "eval_trace_state_scheduled_long",
      "state_loss_schedule": "constant",
      "teacher_position_loss_weight": "0.0",
      "teacher_rep_loss_weight": "0.0",
      "selection_metric": "paired_len24_executor_accuracy",
      "selection_value": "0.3046875",
      "split": "fresh_paraphrase_len24",
      "n": "256.0",
      "direct_accuracy": "nan",
      "direct_target_mass": "nan",
      "executor_accuracy": "0.27734375",
      "executor_target_mass": "0.010309266479453072",
      "init_accuracy": "1.0",
      "init_pos_accuracy": "1.0",
      "op_accuracy": "1.0",
      "arg_accuracy": "0.9495442708333334",
      "op_pos_accuracy": "1.0",
      "arg_pos_accuracy": "0.9431966145833334",
      "program_exact": "0.265625",
      "state_accuracy": "0.5520833333333334",
      "state_all_exact": "0.265625",
      "state_prefix_fraction": "0.5480143229166666",
      "state_mean_correct_prefix": "13.15234375",
      "pair_true_answer_consistency": "",
      "pair_true_length_consistency": "",
      "executor_pair_answer_consistency": "",
      "executor_pair_both_correct": "",
      "compiler_pair_program_consistency": "",
      "compiler_pair_state_consistency": ""
    },
    {
      "run": "main_control_light_state_s900",
      "variant": "copy_trace_state_scheduled",
      "selected_step": "800",
      "selected_tag": "eval_trace_state_scheduled_long",
      "state_loss_schedule": "constant",
      "teacher_position_loss_weight": "0.0",
      "teacher_rep_loss_weight": "0.0",
      "selection_metric": "paired_len24_executor_accuracy",
      "selection_value": "0.3046875",
      "split": "fresh_paired_len24",
      "n": "512.0",
      "direct_accuracy": "nan",
      "direct_target_mass": "nan",
      "executor_accuracy": "0.271484375",
      "executor_target_mass": "0.010309285964467563",
      "init_accuracy": "1.0",
      "init_pos_accuracy": "1.0",
      "op_accuracy": "0.9999186197916666",
      "arg_accuracy": "0.95068359375",
      "op_pos_accuracy": "1.0",
      "arg_pos_accuracy": "0.948486328125",
      "program_exact": "0.267578125",
      "state_accuracy": "0.548583984375",
      "state_all_exact": "0.267578125",
      "state_prefix_fraction": "0.5479329427083334",
      "state_mean_correct_prefix": "13.150390625",
      "pair_true_answer_consistency": "1.0",
      "pair_true_length_consistency": "1.0",
      "executor_pair_answer_consistency": "0.7265625",
      "executor_pair_both_correct": "0.2578125",
      "compiler_pair_program_consistency": "0.7265625",
      "compiler_pair_state_consistency": "0.7265625"
    },
    {
      "run": "main_teacher_slot_distill_s900",
      "variant": "copy_trace_state_teacher",
      "selected_step": "800",
      "selected_tag": "eval_trace_state_teacher_long",
      "state_loss_schedule": "constant",
      "teacher_position_loss_weight": "0.1",
      "teacher_rep_loss_weight": "0.05",
      "selection_metric": "paired_len24_executor_accuracy",
      "selection_value": "0.28125",
      "split": "fresh_standard_len24",
      "n": "256.0",
      "direct_accuracy": "nan",
      "direct_target_mass": "nan",
      "executor_accuracy": "0.3046875",
      "executor_target_mass": "0.010309271223377436",
      "init_accuracy": "1.0",
      "init_pos_accuracy": "1.0",
      "op_accuracy": "1.0",
      "arg_accuracy": "0.95068359375",
      "op_pos_accuracy": "1.0",
      "arg_pos_accuracy": "0.9510091145833334",
      "program_exact": "0.296875",
      "state_accuracy": "0.56982421875",
      "state_all_exact": "0.296875",
      "state_prefix_fraction": "0.5685221354166666",
      "state_mean_correct_prefix": "13.64453125",
      "pair_true_answer_consistency": "",
      "pair_true_length_consistency": "",
      "executor_pair_answer_consistency": "",
      "executor_pair_both_correct": "",
      "compiler_pair_program_consistency": "",
      "compiler_pair_state_consistency": ""
    },
    {
      "run": "main_teacher_slot_distill_s900",
      "variant": "copy_trace_state_teacher",
      "selected_step": "800",
      "selected_tag": "eval_trace_state_teacher_long",
      "state_loss_schedule": "constant",
      "teacher_position_loss_weight": "0.1",
      "teacher_rep_loss_weight": "0.05",
      "selection_metric": "paired_len24_executor_accuracy",
      "selection_value": "0.28125",
      "split": "fresh_paraphrase_len24",
      "n": "256.0",
      "direct_accuracy": "nan",
      "direct_target_mass": "nan",
      "executor_accuracy": "0.28125",
      "executor_target_mass": "0.010309272824088112",
      "init_accuracy": "1.0",
      "init_pos_accuracy": "1.0",
      "op_accuracy": "0.9993489583333334",
      "arg_accuracy": "0.94775390625",
      "op_pos_accuracy": "1.0",
      "arg_pos_accuracy": "0.9475911458333334",
      "program_exact": "0.26953125",
      "state_accuracy": "0.5608723958333334",
      "state_all_exact": "0.26953125",
      "state_prefix_fraction": "0.55859375",
      "state_mean_correct_prefix": "13.40625",
      "pair_true_answer_consistency": "",
      "pair_true_length_consistency": "",
      "executor_pair_answer_consistency": "",
      "executor_pair_both_correct": "",
      "compiler_pair_program_consistency": "",
      "compiler_pair_state_consistency": ""
    },
    {
      "run": "main_teacher_slot_distill_s900",
      "variant": "copy_trace_state_teacher",
      "selected_step": "800",
      "selected_tag": "eval_trace_state_teacher_long",
      "state_loss_schedule": "constant",
      "teacher_position_loss_weight": "0.1",
      "teacher_rep_loss_weight": "0.05",
      "selection_metric": "paired_len24_executor_accuracy",
      "selection_value": "0.28125",
      "split": "fresh_paired_len24",
      "n": "512.0",
      "direct_accuracy": "nan",
      "direct_target_mass": "nan",
      "executor_accuracy": "0.279296875",
      "executor_target_mass": "0.010309286823030561",
      "init_accuracy": "1.0",
      "init_pos_accuracy": "1.0",
      "op_accuracy": "0.9995930989583334",
      "arg_accuracy": "0.949462890625",
      "op_pos_accuracy": "1.0",
      "arg_pos_accuracy": "0.9495442708333334",
      "program_exact": "0.2734375",
      "state_accuracy": "0.5567220052083334",
      "state_all_exact": "0.2734375",
      "state_prefix_fraction": "0.5548502604166666",
      "state_mean_correct_prefix": "13.31640625",
      "pair_true_answer_consistency": "1.0",
      "pair_true_length_consistency": "1.0",
      "executor_pair_answer_consistency": "0.5546875",
      "executor_pair_both_correct": "0.26171875",
      "compiler_pair_program_consistency": "0.5546875",
      "compiler_pair_state_consistency": "0.5546875"
    },
    {
      "run": "main_teacher_softpos_low_s900",
      "variant": "copy_trace_state_teacher",
      "selected_step": 800,
      "selected_tag": "eval_trace_state_teacher_long",
      "state_loss_schedule": "constant",
      "teacher_position_loss_weight": 0.03,
      "teacher_rep_loss_weight": 0.0,
      "selection_metric": "paired_len24_executor_accuracy",
      "selection_value": 0.234375,
      "split": "fresh_standard_len24",
      "n": 256.0,
      "direct_accuracy": NaN,
      "direct_target_mass": NaN,
      "executor_accuracy": 0.26953125,
      "executor_target_mass": 0.010309266042895615,
      "init_accuracy": 1.0,
      "init_pos_accuracy": 1.0,
      "op_accuracy": 1.0,
      "arg_accuracy": 0.94775390625,
      "op_pos_accuracy": 1.0,
      "arg_pos_accuracy": 0.9503580729166666,
      "program_exact": 0.26171875,
      "state_accuracy": 0.5628255208333334,
      "state_all_exact": 0.26171875,
      "state_prefix_fraction": 0.5598958333333334,
      "state_mean_correct_prefix": 13.4375
    },
    {
      "run": "main_teacher_softpos_low_s900",
      "variant": "copy_trace_state_teacher",
      "selected_step": 800,
      "selected_tag": "eval_trace_state_teacher_long",
      "state_loss_schedule": "constant",
      "teacher_position_loss_weight": 0.03,
      "teacher_rep_loss_weight": 0.0,
      "selection_metric": "paired_len24_executor_accuracy",
      "selection_value": 0.234375,
      "split": "fresh_paraphrase_len24",
      "n": 256.0,
      "direct_accuracy": NaN,
      "direct_target_mass": NaN,
      "executor_accuracy": 0.1171875,
      "executor_target_mass": 0.010309278150089085,
      "init_accuracy": 1.0,
      "init_pos_accuracy": 1.0,
      "op_accuracy": 0.99462890625,
      "arg_accuracy": 0.92041015625,
      "op_pos_accuracy": 0.9972330729166666,
      "arg_pos_accuracy": 0.9200846354166666,
      "program_exact": 0.10546875,
      "state_accuracy": 0.5260416666666666,
      "state_all_exact": 0.10546875,
      "state_prefix_fraction": 0.52197265625,
      "state_mean_correct_prefix": 12.52734375
    },
    {
      "run": "main_teacher_softpos_low_s900",
      "variant": "copy_trace_state_teacher",
      "selected_step": 800,
      "selected_tag": "eval_trace_state_teacher_long",
      "state_loss_schedule": "constant",
      "teacher_position_loss_weight": 0.03,
      "teacher_rep_loss_weight": 0.0,
      "selection_metric": "paired_len24_executor_accuracy",
      "selection_value": 0.234375,
      "split": "fresh_paired_len24",
      "n": 512.0,
      "direct_accuracy": NaN,
      "direct_target_mass": NaN,
      "executor_accuracy": 0.18359375,
      "executor_target_mass": 0.010309282064554282,
      "init_accuracy": 1.0,
      "init_pos_accuracy": 1.0,
      "op_accuracy": 0.9965006510416666,
      "arg_accuracy": 0.935546875,
      "op_pos_accuracy": 0.9986979166666666,
      "arg_pos_accuracy": 0.9345703125,
      "program_exact": 0.169921875,
      "state_accuracy": 0.5284016927083334,
      "state_all_exact": 0.169921875,
      "state_prefix_fraction": 0.527099609375,
      "state_mean_correct_prefix": 12.650390625,
      "pair_true_answer_consistency": 1.0,
      "pair_true_length_consistency": 1.0,
      "executor_pair_answer_consistency": 0.42578125,
      "executor_pair_both_correct": 0.11328125,
      "compiler_pair_program_consistency": 0.41015625,
      "compiler_pair_state_consistency": 0.41015625
    }
  ]
}