diff --git "a/gradient-probe.json" "b/gradient-probe.json" new file mode 100644--- /dev/null +++ "b/gradient-probe.json" @@ -0,0 +1,8900 @@ +{ + "caveats": [ + "Only the pre-travel training-state probe was read; no final-test data or post-training teacher probe was used.", + "Quantities are conditional action-logit gradients, not measured parameter gradients, optimizer steps, or observed training batches.", + "The Monte Carlo estimator is unbiased here; high conditional variance does not prove it caused development regression.", + "Live-teacher drift remains untested. Different-step gradient norms on different prompts do not establish model collapse.", + "Exact full-logit reverse KL removes action-sampling variance, not prompt sampling, teacher errors, saturation, or shared-parameter interference." + ], + "conditional_recommendation": { + "anchor": "Existing pure-distillation, 50%-historical-replay branch; retain checkpoint, data, seed, lr, action mask, generation budget, and 64/192-step checks.", + "launched": false, + "one_change": "A separately registered development branch changes native distillation_mode from sampled_token to full_logits.", + "only_if": "All eight preregistered development candidates fail the promotion gate; not automatically checked or acted on here.", + "reason": "Compare an unbiased Monte Carlo gradient with its exact four-action expectation without changing the intended objective." + }, + "created_at": "2026-09-07T11:02:21.377151+00:00", + "input_filename": "prefix_teacher_probe.json", + "input_sha256": "fdd71492f24435542ab68e6ad520541e46d0a3dc0590eebdf00da086f3f4d65d", + "ledger_hash": "bd33e30f144155374d5af22484cd3d7cdaa575a32f037b51c104547361506ec0", + "method": "Exact enumeration of 4^4 iid action batches, using the saved conditional A-D distributions.", + "native_formula": "g(a)=log(p[a]/q[a])*(one_hot(a)-p); E[g]=p*(log(p/q)-KL(p||q)); q is stop-gradient.", + "native_source": "https://github.com/huggingface/trl/blob/4506f4af8ccc5b7f8b337f589a696a19913d8a4a/trl/experimental/sdpo/loss_utils.py", + "probe_selection_hash": "615639a9ca9409a5080730833ce147ce3c73051581e994923ed9338c1fda7119", + "script_sha256": "3ebe3034159a8c562564fc035a30586d63a9ea304d7b10644b611590c1827d4f", + "states": [ + { + "current_model_mistake": false, + "enumerated_mean_logit_gradient": [ + -0.29179943988251095, + 0.014352908206249168, + 0.1434601597355092, + 0.13398637194075322 + ], + "episode_id": "train-travel-2300105", + "exact_gradient_norm": 0.3519745507659723, + "exact_logit_gradient": [ + -0.29179943988251134, + 0.014352908206249152, + 0.14346015973550907, + 0.13398637194075314 + ], + "expected_cosine": 0.8743042827315961, + "feedback_advantage": 0.15829647175178885, + "id": "train-travel-2300105/frozen-warmstart-v2/seed-1707/step-3", + "kind": "repair", + "maximum_expectation_error": 3.885780586188048e-16, + "mc4_error_norm_quantiles": { + "0.1": 0.17509489192938096, + "0.25": 0.17509489192938096, + "0.5": 0.29846055392441406, + "0.75": 0.30557386603768055, + "0.9": 1.083005591054383 + }, + "mc4_gradient_norm_quantiles": { + "0.1": 0.05785139897382646, + "0.25": 0.05785139897382646, + "0.5": 0.3029804040318947, + "0.75": 0.5489855313084223, + "0.9": 1.3254905388424219 + }, + "mc4_gradient_rmse": 0.5114255564293229, + "noise_signal_ratio": 1.4530185643148088, + "optimal_choices": [ + "A" + ], + "p": [ + 0.7637291721343064, + 0.009613931393934743, + 0.19310097827988068, + 0.033555918191878156 + ], + "probability_all_four_same": 0.3416098517155346, + "probability_no_optimal_in_four": 0.0031163082690039835, + "probability_opposed_gradient": 0.0, + "q": [ + 0.9220256438860952, + 0.0017799282670200914, + 0.07568446990961133, + 0.0005099579372733557 + ], + "reverse_kl": 0.19371208214056254, + "student_entropy_nats": 0.6819790754458984, + "student_max_mass": 0.7637291721343064, + "student_optimal_mass": 0.7637291721343064, + "student_predicted_action_mass": 0.7637291721343064, + "teacher_entropy_nats": 0.2853426572454393, + "teacher_mass_on_student_prediction": 0.9220256438860952, + "teacher_max_mass": 0.9220256438860952, + "teacher_optimal_mass": 0.9220256438860952, + "variance_verification_error": 1.1102230246251565e-16 + }, + { + "current_model_mistake": false, + "enumerated_mean_logit_gradient": [ + -0.0001572255653929387, + 3.317613750809455e-05, + -3.739917368275727e-05, + 0.00016144860156760125 + ], + "episode_id": "train-travel-2300033", + "exact_gradient_norm": 0.00023083518720804316, + "exact_logit_gradient": [ + -0.0001572255653929386, + 3.317613750809445e-05, + -3.739917368275723e-05, + 0.00016144860156760136 + ], + "expected_cosine": -0.16154456051370122, + "feedback_advantage": -0.00040460666741481965, + "id": "train-travel-2300033/frozen-warmstart-v2/seed-1707/step-2", + "kind": "exercise", + "maximum_expectation_error": 1.0842021724855044e-19, + "mc4_error_norm_quantiles": { + "0.1": 0.00023088404967838686, + "0.25": 0.00023088404967838686, + "0.5": 0.00023088404967838686, + "0.75": 0.00023088404967838686, + "0.9": 0.00023088404967838686 + }, + "mc4_gradient_norm_quantiles": { + "0.1": 2.987314326338695e-07, + "0.25": 2.987314326338695e-07, + "0.5": 2.987314326338695e-07, + "0.75": 2.987314326338695e-07, + "0.9": 2.987314326338695e-07 + }, + "mc4_gradient_rmse": 0.013259854701317507, + "noise_signal_ratio": 57.44295253117929, + "optimal_choices": [ + "B" + ], + "p": [ + 0.00013976018715000123, + 0.9994175360722718, + 1.2212107232152854e-05, + 0.00043049163334600105 + ], + "probability_all_four_same": 0.997672179084163, + "probability_no_optimal_in_four": 1.1510021579285971e-13, + "probability_opposed_gradient": 0.9976721790841283, + "q": [ + 0.00043031738425974706, + 0.999012929404857, + 0.0002610006730919585, + 0.0002957525377913395 + ], + "reverse_kl": 0.00037172897171703154, + "student_entropy_nats": 0.0052974649130072835, + "student_max_mass": 0.9994175360722718, + "student_optimal_mass": 0.9994175360722718, + "student_predicted_action_mass": 0.9994175360722718, + "teacher_entropy_nats": 0.008878762682514894, + "teacher_mass_on_student_prediction": 0.999012929404857, + "teacher_max_mass": 0.999012929404857, + "teacher_optimal_mass": 0.999012929404857, + "variance_verification_error": 8.131516293641283e-20 + }, + { + "current_model_mistake": false, + "enumerated_mean_logit_gradient": [ + -0.00016174093856697653, + 0.001020687619279251, + -0.0013271147199357815, + 0.00046816803922350447 + ], + "episode_id": "train-travel-2300013", + "exact_gradient_norm": 0.00174596052079624, + "exact_logit_gradient": [ + -0.00016174093856697653, + 0.001020687619279252, + -0.00132711471993578, + 0.00046816803922350474 + ], + "expected_cosine": 0.9501465442487751, + "feedback_advantage": 0.000814515252221204, + "id": "train-travel-2300013/frozen-warmstart-v2/seed-1707/step-2", + "kind": "exercise", + "maximum_expectation_error": 1.5178830414797062e-18, + "mc4_error_norm_quantiles": { + "0.1": 0.0017439199514421866, + "0.25": 0.0017439199514421866, + "0.5": 0.0017439199514421866, + "0.75": 0.0017439199514421866, + "0.9": 0.0017439199514421866 + }, + "mc4_gradient_norm_quantiles": { + "0.1": 2.1409796039328926e-06, + "0.25": 2.1409796039328926e-06, + "0.5": 2.1409796039328926e-06, + "0.75": 2.1409796039328926e-06, + "0.9": 2.1409796039328926e-06 + }, + "mc4_gradient_rmse": 0.02548304874026197, + "noise_signal_ratio": 14.595432391930892, + "optimal_choices": [ + "C" + ], + "p": [ + 0.0004297848440399064, + 0.0011682762044508148, + 0.9977766057776156, + 0.0006253331738936379 + ], + "probability_all_four_same": 0.9911360400629231, + "probability_no_optimal_in_four": 2.4438012980552168e-11, + "probability_opposed_gradient": 0.0017088015632970074, + "q": [ + 0.0006258437035797225, + 0.00048740760927394605, + 0.9985911210298368, + 0.000295627657309495 + ], + "reverse_kl": 0.0005140747344204189, + "student_entropy_nats": 0.018054398851476947, + "student_max_mass": 0.9977766057776156, + "student_optimal_mass": 0.9977766057776156, + "student_predicted_action_mass": 0.9977766057776156, + "teacher_entropy_nats": 0.012143927385041194, + "teacher_mass_on_student_prediction": 0.9985911210298368, + "teacher_max_mass": 0.9985911210298368, + "teacher_optimal_mass": 0.9985911210298368, + "variance_verification_error": 2.168404344971009e-19 + }, + { + "current_model_mistake": false, + "enumerated_mean_logit_gradient": [ + -0.00014051510205564195, + -0.010868253307513327, + -8.286367246573695e-06, + 0.01101705477681555 + ], + "episode_id": "train-travel-2300011", + "exact_gradient_norm": 0.015476247577155664, + "exact_logit_gradient": [ + -0.0001405151020556424, + -0.010868253307513332, + -8.286367246573706e-06, + 0.011017054776815546 + ], + "expected_cosine": 0.9986795765487863, + "feedback_advantage": 0.004716670670575462, + "id": "train-travel-2300011/frozen-warmstart-v2/seed-1707/step-2", + "kind": "exercise", + "maximum_expectation_error": 5.204170427930421e-18, + "mc4_error_norm_quantiles": { + "0.1": 0.015436214136766446, + "0.25": 0.015436214136766446, + "0.5": 0.015436214136766446, + "0.75": 0.015436214136766446, + "0.9": 0.015436214136766446 + }, + "mc4_gradient_norm_quantiles": { + "0.1": 4.0053521929404146e-05, + "0.25": 4.0053521929404146e-05, + "0.5": 4.0053521929404146e-05, + "0.75": 4.0053521929404146e-05, + "0.9": 4.0053521929404146e-05 + }, + "mc4_gradient_rmse": 0.10116474066869896, + "noise_signal_ratio": 6.536774509734985, + "optimal_choices": [ + "B" + ], + "p": [ + 0.00013899527576388533, + 0.9939477047507816, + 3.0708055104854267e-06, + 0.005910229167944037 + ], + "probability_all_four_same": 0.976009716442743, + "probability_no_optimal_in_four": 1.3417772505118354e-09, + "probability_opposed_gradient": 0.0005581283830384063, + "q": [ + 0.00037962125357278843, + 0.9986643754213571, + 4.533929296974602e-05, + 0.0009106640321004089 + ], + "reverse_kl": 0.0062002644125231845, + "student_entropy_nats": 0.03763315294267296, + "student_max_mass": 0.9939477047507816, + "student_optimal_mass": 0.9939477047507816, + "student_predicted_action_mass": 0.9939477047507816, + "teacher_entropy_nats": 0.011154075842172926, + "teacher_mass_on_student_prediction": 0.9986643754213571, + "teacher_max_mass": 0.9986643754213571, + "teacher_optimal_mass": 0.9986643754213571, + "variance_verification_error": 5.204170427930421e-18 + }, + { + "current_model_mistake": false, + "enumerated_mean_logit_gradient": [ + 0.015993491868294574, + 0.03229769895673614, + -0.16646800701868497, + 0.11817681619365433 + ], + "episode_id": "train-travel-2300098", + "exact_gradient_norm": 0.2073072366958832, + "exact_logit_gradient": [ + 0.01599349186829457, + 0.032297698956736134, + -0.1664680070186851, + 0.11817681619365443 + ], + "expected_cosine": 0.9479346098659425, + "feedback_advantage": 0.08265589126296569, + "id": "train-travel-2300098/frozen-warmstart-v2/seed-1707/step-1", + "kind": "admin", + "maximum_expectation_error": 1.3877787807814457e-16, + "mc4_error_norm_quantiles": { + "0.1": 0.19647236353671704, + "0.25": 0.19647236353671704, + "0.5": 0.19647236353671704, + "0.75": 0.4312868846306149, + "0.9": 0.4312868846306149 + }, + "mc4_gradient_norm_quantiles": { + "0.1": 0.010870400029759315, + "0.25": 0.010870400029759315, + "0.5": 0.010870400029759315, + "0.75": 0.6277831065371587, + "0.9": 0.6277831065371587 + }, + "mc4_gradient_rmse": 0.35167299840212335, + "noise_signal_ratio": 1.6963855387162519, + "optimal_choices": [ + "C" + ], + "p": [ + 0.008812743296557296, + 0.02714516053909075, + 0.8989242373895548, + 0.06511785877479707 + ], + "probability_all_four_same": 0.6529872254957931, + "probability_no_optimal_in_four": 0.00010437298566973569, + "probability_opposed_gradient": 0.0, + "q": [ + 0.001302341298482554, + 0.007494456516878209, + 0.9815801286525205, + 0.009623073532118718 + ], + "reverse_kl": 0.097220910559366, + "student_entropy_nats": 0.41325789502446353, + "student_max_mass": 0.8989242373895548, + "student_optimal_mass": 0.8989242373895548, + "student_predicted_action_mass": 0.8989242373895548, + "teacher_entropy_nats": 0.10826183102829856, + "teacher_mass_on_student_prediction": 0.9815801286525205, + "teacher_max_mass": 0.9815801286525205, + "teacher_optimal_mass": 0.9815801286525205, + "variance_verification_error": 8.326672684688674e-17 + }, + { + "current_model_mistake": false, + "enumerated_mean_logit_gradient": [ + 0.012408120332915828, + -0.30387555944665884, + 0.24004201097169997, + 0.05142542814204264 + ], + "episode_id": "train-travel-2300067", + "exact_gradient_norm": 0.3908440338167467, + "exact_logit_gradient": [ + 0.012408120332915804, + -0.3038755594466588, + 0.2400420109717003, + 0.05142542814204266 + ], + "expected_cosine": 0.9540795496931246, + "feedback_advantage": 0.15620026335864545, + "id": "train-travel-2300067/frozen-warmstart-v2/seed-1707/step-0", + "kind": "focus", + "maximum_expectation_error": 3.3306690738754696e-16, + "mc4_error_norm_quantiles": { + "0.1": 0.16400698726015242, + "0.25": 0.16400698726015242, + "0.5": 0.3473598642610981, + "0.75": 0.3473598642610981, + "0.9": 0.7076470151768334 + }, + "mc4_gradient_norm_quantiles": { + "0.1": 0.04356671342739758, + "0.25": 0.04356671342739758, + "0.5": 0.5287917315525071, + "0.75": 0.5287917315525071, + "0.9": 1.0144462863380934 + }, + "mc4_gradient_rmse": 0.43738193170041256, + "noise_signal_ratio": 1.1190702527276797, + "optimal_choices": [ + "B" + ], + "p": [ + 0.0062013638913265366, + 0.8122184613195449, + 0.15993528730899123, + 0.021644887480137397 + ], + "probability_all_four_same": 0.43585706622556636, + "probability_no_optimal_in_four": 0.0012434020337377938, + "probability_opposed_gradient": 0.0, + "q": [ + 0.0006877461775305884, + 0.9684187246781903, + 0.029243711665146728, + 0.0016498174791322892 + ], + "reverse_kl": 0.19823511257763468, + "student_entropy_nats": 0.576575154416722, + "student_max_mass": 0.8122184613195449, + "student_optimal_mass": 0.8122184613195449, + "student_predicted_action_mass": 0.8122184613195449, + "teacher_entropy_nats": 0.14994745517046068, + "teacher_mass_on_student_prediction": 0.9684187246781903, + "teacher_max_mass": 0.9684187246781903, + "teacher_optimal_mass": 0.9684187246781903, + "variance_verification_error": 1.6653345369377348e-16 + }, + { + "current_model_mistake": true, + "enumerated_mean_logit_gradient": [ + -0.00031232628606707116, + -0.04819131868119233, + -4.848997463167357e-05, + 0.04855213494189114 + ], + "episode_id": "train-travel-2300049", + "exact_gradient_norm": 0.06840915803190727, + "exact_logit_gradient": [ + -0.0003123262860670711, + -0.04819131868119232, + -4.848997463167357e-05, + 0.048552134941891256 + ], + "expected_cosine": 0.9997384788792522, + "feedback_advantage": 0.897488391135733, + "id": "train-travel-2300049/frozen-warmstart-v2/seed-1707/step-1", + "kind": "admin", + "maximum_expectation_error": 1.1796119636642288e-16, + "mc4_error_norm_quantiles": { + "0.1": 0.046064185936844165, + "0.25": 0.046064185936844165, + "0.5": 0.046064185936844165, + "0.75": 0.046064185936844165, + "0.9": 0.046064185936844165 + }, + "mc4_gradient_norm_quantiles": { + "0.1": 0.022345391139556987, + "0.25": 0.022345391139556987, + "0.5": 0.022345391139556987, + "0.75": 0.022345391139556987, + "0.9": 0.022345391139556987 + }, + "mc4_gradient_rmse": 0.280260390232306, + "noise_signal_ratio": 4.096825604864008, + "optimal_choices": [ + "B" + ], + "p": [ + 7.434364868736843e-05, + 0.006692201354212919, + 2.2673530316689987e-05, + 0.993210781466783 + ], + "probability_all_four_same": 0.9731184391721788, + "probability_no_optimal_in_four": 0.9734987110867733, + "probability_opposed_gradient": 8.886245032262823e-05, + "q": [ + 0.0005000881283086045, + 0.9041805924899459, + 1.9390521220434366e-05, + 0.09529992886052506 + ], + "reverse_kl": 2.2950298254459622, + "student_entropy_nats": 0.0412219651593985, + "student_max_mass": 0.993210781466783, + "student_optimal_mass": 0.006692201354212919, + "student_predicted_action_mass": 0.993210781466783, + "teacher_entropy_nats": 0.3191101215673244, + "teacher_mass_on_student_prediction": 0.09529992886052506, + "teacher_max_mass": 0.9041805924899459, + "teacher_optimal_mass": 0.9041805924899459, + "variance_verification_error": 4.163336342344337e-17 + }, + { + "current_model_mistake": false, + "enumerated_mean_logit_gradient": [ + 0.02082392119556735, + -0.48298610173933004, + 0.2833020580344582, + 0.1788601225093049 + ], + "episode_id": "train-travel-2300068", + "exact_gradient_norm": 0.5881838230331716, + "exact_logit_gradient": [ + 0.020823921195567304, + -0.4829861017393306, + 0.28330205803445835, + 0.178860122509305 + ], + "expected_cosine": 0.8987975795681666, + "feedback_advantage": 0.19234428641104828, + "id": "train-travel-2300068/frozen-warmstart-v2/seed-1707/step-2", + "kind": "exercise", + "maximum_expectation_error": 5.551115123125783e-16, + "mc4_error_norm_quantiles": { + "0.1": 0.4951610298182248, + "0.25": 0.5343485187638508, + "0.5": 0.5343485187638508, + "0.75": 0.535951680087082, + "0.9": 0.9272258165631382 + }, + "mc4_gradient_norm_quantiles": { + "0.1": 0.05397330236860228, + "0.25": 0.05397330236860228, + "0.5": 0.8014416569656272, + "0.75": 0.942143976093293, + "0.9": 1.3888691235080994 + }, + "mc4_gradient_rmse": 0.6957378359855033, + "noise_signal_ratio": 1.1828578222326696, + "optimal_choices": [ + "B" + ], + "p": [ + 0.016452255677468938, + 0.7927141051257723, + 0.10728219025907647, + 0.08355144893768228 + ], + "probability_all_four_same": 0.3950623693564715, + "probability_no_optimal_in_four": 0.0018462010903970466, + "probability_opposed_gradient": 0.0, + "q": [ + 0.003135225157527096, + 0.9850583915368206, + 0.00516911204996443, + 0.006637271255687863 + ], + "reverse_kl": 0.39204329238886854, + "student_entropy_nats": 0.6986002783232113, + "student_max_mass": 0.7927141051257723, + "student_optimal_mass": 0.7927141051257723, + "student_predicted_action_mass": 0.7927141051257723, + "teacher_entropy_nats": 0.09340609775145609, + "teacher_mass_on_student_prediction": 0.9850583915368206, + "teacher_max_mass": 0.9850583915368206, + "teacher_optimal_mass": 0.9850583915368206, + "variance_verification_error": 2.220446049250313e-16 + }, + { + "current_model_mistake": false, + "enumerated_mean_logit_gradient": [ + -0.0003724094888398974, + 0.16112431822642348, + -2.814527459376944e-05, + -0.16072376346298992 + ], + "episode_id": "train-travel-2300042", + "exact_gradient_norm": 0.22758144376600817, + "exact_logit_gradient": [ + -0.00037240948883989736, + 0.16112431822642348, + -2.8145274593769447e-05, + -0.1607237634629898 + ], + "expected_cosine": 0.9989897089478503, + "feedback_advantage": 0.1158683586399426, + "id": "train-travel-2300042/frozen-warmstart-v2/seed-1707/step-2", + "kind": "exercise", + "maximum_expectation_error": 1.1102230246251565e-16, + "mc4_error_norm_quantiles": { + "0.1": 0.04537110594558701, + "0.25": 0.04537110594558701, + "0.5": 0.1888713554182189, + "0.75": 0.1888713554182189, + "0.9": 0.2796089495599117 + }, + "mc4_gradient_norm_quantiles": { + "0.1": 0.03871031299793804, + "0.25": 0.03871031299793804, + "0.5": 0.27294998745950944, + "0.75": 0.27294998745950944, + "0.9": 0.5071896903900507 + }, + "mc4_gradient_rmse": 0.1887007651842076, + "noise_signal_ratio": 0.8291570791607402, + "optimal_choices": [ + "D" + ], + "p": [ + 0.00023623860765030377, + 0.20176164187673776, + 1.9391643351532853e-05, + 0.7979827278722604 + ], + "probability_all_four_same": 0.4071413486104431, + "probability_no_optimal_in_four": 0.0016655359447370558, + "probability_opposed_gradient": 0.0005199563287349986, + "q": [ + 0.0010700096191843644, + 0.08500139263562859, + 7.751123298398139e-05, + 0.913851086512203 + ], + "reverse_kl": 0.06583190412553912, + "student_entropy_nats": 0.5052160469867346, + "student_max_mass": 0.7979827278722604, + "student_optimal_mass": 0.7979827278722604, + "student_predicted_action_mass": 0.7979827278722604, + "teacher_entropy_nats": 0.2999151854802204, + "teacher_mass_on_student_prediction": 0.913851086512203, + "teacher_max_mass": 0.913851086512203, + "teacher_optimal_mass": 0.913851086512203, + "variance_verification_error": 6.938893903907228e-18 + }, + { + "current_model_mistake": true, + "enumerated_mean_logit_gradient": [ + -0.02530322640971222, + -0.011449228697313546, + -0.747383722286977, + 0.7841361773940028 + ], + "episode_id": "train-travel-2300052", + "exact_gradient_norm": 1.0836158503557065, + "exact_logit_gradient": [ + -0.0253032264097122, + -0.011449228697313471, + -0.747383722286977, + 0.7841361773940028 + ], + "expected_cosine": 0.8551834755602493, + "feedback_advantage": 0.7977948455014461, + "id": "train-travel-2300052/frozen-warmstart-v2/seed-1707/step-2", + "kind": "exercise", + "maximum_expectation_error": 7.45931094670027e-17, + "mc4_error_norm_quantiles": { + "0.1": 0.4224167211374518, + "0.25": 0.5072306832099569, + "0.5": 0.508876903512235, + "0.75": 0.6197984512244105, + "0.9": 1.0559147952507288 + }, + "mc4_gradient_norm_quantiles": { + "0.1": 0.8000163285312458, + "0.25": 0.9980241347477529, + "0.5": 1.444530469711411, + "0.75": 1.495094430049431, + "0.9": 1.5865427001040446 + }, + "mc4_gradient_rmse": 0.6949973634353181, + "noise_signal_ratio": 0.641368768468253, + "optimal_choices": [ + "C" + ], + "p": [ + 0.029877883923900894, + 0.11816930616798783, + 0.1719353474610533, + 0.680017462447058 + ], + "probability_all_four_same": 0.21490541149471157, + "probability_no_optimal_in_four": 0.4701722418107565, + "probability_opposed_gradient": 0.011538383484934777, + "q": [ + 0.00508867728860605, + 0.009506901088639276, + 0.9697301929624994, + 0.015674228660255238 + ], + "reverse_kl": 2.6169887755259404, + "student_entropy_nats": 0.9222125778548519, + "student_max_mass": 0.680017462447058, + "student_optimal_mass": 0.1719353474610533, + "student_predicted_action_mass": 0.680017462447058, + "teacher_entropy_nats": 0.1660785632926465, + "teacher_mass_on_student_prediction": 0.015674228660255238, + "teacher_max_mass": 0.9697301929624994, + "teacher_optimal_mass": 0.9697301929624994, + "variance_verification_error": 5.551115123125783e-17 + }, + { + "current_model_mistake": false, + "enumerated_mean_logit_gradient": [ + -0.0007326827569054225, + 0.00043759176742709977, + 1.7883381453105306e-06, + 0.0002933026513330106 + ], + "episode_id": "train-travel-2300105", + "exact_gradient_norm": 0.0009024080122753778, + "exact_logit_gradient": [ + -0.0007326827569054217, + 0.00043759176742709993, + 1.7883381453105279e-06, + 0.0002933026513330113 + ], + "expected_cosine": 0.9715002113326107, + "feedback_advantage": 0.000542817785489591, + "id": "train-travel-2300105/frozen-warmstart-v2/seed-1707/step-0", + "kind": "focus", + "maximum_expectation_error": 7.589415207398531e-19, + "mc4_error_norm_quantiles": { + "0.1": 0.0009014207268570882, + "0.25": 0.0009014207268570882, + "0.5": 0.0009014207268570882, + "0.75": 0.0009014207268570882, + "0.9": 0.0009014207268570882 + }, + "mc4_gradient_norm_quantiles": { + "0.1": 1.0158211081278651e-06, + "0.25": 1.0158211081278651e-06, + "0.5": 1.0158211081278651e-06, + "0.75": 1.0158211081278651e-06, + "0.9": 1.0158211081278651e-06 + }, + "mc4_gradient_rmse": 0.015701602946467424, + "noise_signal_ratio": 17.39967147108612, + "optimal_choices": [ + "A" + ], + "p": [ + 0.9985651386223682, + 0.0011691994746718414, + 4.778251428891235e-06, + 0.0002608836515310601 + ], + "probability_all_four_same": 0.9942728956420966, + "probability_no_optimal_in_four": 4.2387693283634514e-12, + "probability_opposed_gradient": 0.0, + "q": [ + 0.9991079564078578, + 0.0008040151918209271, + 3.2858262909959606e-06, + 8.474257403034226e-05 + ], + "reverse_kl": 0.00019028548853822094, + "student_entropy_nats": 0.01153881206827373, + "student_max_mass": 0.9985651386223682, + "student_optimal_mass": 0.9985651386223682, + "student_predicted_action_mass": 0.9985651386223682, + "teacher_entropy_nats": 0.007456995090884982, + "teacher_mass_on_student_prediction": 0.9991079564078578, + "teacher_max_mass": 0.9991079564078578, + "teacher_optimal_mass": 0.9991079564078578, + "variance_verification_error": 2.168404344971009e-19 + }, + { + "current_model_mistake": false, + "enumerated_mean_logit_gradient": [ + -0.005012797759286699, + -0.0010807019141102117, + -0.0002555226541156751, + 0.006349022327512592 + ], + "episode_id": "train-travel-2300030", + "exact_gradient_norm": 0.008165257763518003, + "exact_logit_gradient": [ + -0.005012797759286703, + -0.001080701914110215, + -0.0002555226541156752, + 0.006349022327512593 + ], + "expected_cosine": 0.9734390433678075, + "feedback_advantage": 0.002913260982022048, + "id": "train-travel-2300030/frozen-warmstart-v2/seed-1707/step-1", + "kind": "admin", + "maximum_expectation_error": 4.336808689942018e-18, + "mc4_error_norm_quantiles": { + "0.1": 0.008055853622175887, + "0.25": 0.008055853622175887, + "0.5": 0.008055853622175887, + "0.75": 0.008055853622175887, + "0.9": 0.07708399079036952 + }, + "mc4_gradient_norm_quantiles": { + "0.1": 0.00011187075274321604, + "0.25": 0.00011187075274321604, + "0.5": 0.00011187075274321604, + "0.75": 0.00011187075274321604, + "0.9": 0.08510545574913767 + }, + "mc4_gradient_rmse": 0.03976057134254839, + "noise_signal_ratio": 4.869481465753206, + "optimal_choices": [ + "A" + ], + "p": [ + 0.9730323224369934, + 0.0007830312457861581, + 0.0002542130020440403, + 0.025930433315176315 + ], + "probability_all_four_same": 0.8964153549710596, + "probability_no_optimal_in_four": 5.289007559371888e-07, + "probability_opposed_gradient": 0.004134310399107986, + "q": [ + 0.9759455834190155, + 0.00310622087421601, + 0.0006930915504334978, + 0.020255104156335007 + ], + "reverse_kl": 0.0021621987249302458, + "student_entropy_nats": 0.12901215340836908, + "student_max_mass": 0.9730323224369934, + "student_optimal_mass": 0.9730323224369934, + "student_predicted_action_mass": 0.9730323224369934, + "teacher_entropy_nats": 0.1257226614134414, + "teacher_mass_on_student_prediction": 0.9759455834190155, + "teacher_max_mass": 0.9759455834190155, + "teacher_optimal_mass": 0.9759455834190155, + "variance_verification_error": 1.0842021724855044e-18 + }, + { + "current_model_mistake": true, + "enumerated_mean_logit_gradient": [ + 0.23441881911064882, + -0.04177046810475693, + -0.012958924674598308, + -0.17968942633129373 + ], + "episode_id": "train-travel-2300103", + "exact_gradient_norm": 0.29858529505471754, + "exact_logit_gradient": [ + 0.23441881911064857, + -0.04177046810475681, + -0.012958924674598341, + -0.17968942633129345 + ], + "expected_cosine": 0.9523035216967249, + "feedback_advantage": 0.3799800172462619, + "id": "train-travel-2300103/frozen-warmstart-v2/seed-1707/step-3", + "kind": "repair", + "maximum_expectation_error": 2.7755575615628914e-16, + "mc4_error_norm_quantiles": { + "0.1": 0.17812308691105305, + "0.25": 0.17886260720205063, + "0.5": 0.17886260720205063, + "0.75": 0.34709161554892126, + "0.9": 0.34709161554892126 + }, + "mc4_gradient_norm_quantiles": { + "0.1": 0.12374526404988935, + "0.25": 0.12374526404988935, + "0.5": 0.12374526404988935, + "0.75": 0.6362813792969093, + "0.9": 0.6362813792969093 + }, + "mc4_gradient_rmse": 0.3079891472109178, + "noise_signal_ratio": 1.0314946928464008, + "optimal_choices": [ + "D" + ], + "p": [ + 0.8665049148709926, + 0.04888486530628408, + 0.00401271417375135, + 0.08059750564897204 + ], + "probability_all_four_same": 0.5637947499562791, + "probability_no_optimal_in_four": 0.7145336903547949, + "probability_opposed_gradient": 0.0, + "q": [ + 0.4064582333690476, + 0.07063185502549844, + 0.06233238871022003, + 0.46057752289523396 + ], + "reverse_kl": 0.4864528601240988, + "student_entropy_nats": 0.4968189030298646, + "student_max_mass": 0.8665049148709926, + "student_optimal_mass": 0.08059750564897204, + "student_predicted_action_mass": 0.8665049148709926, + "teacher_entropy_nats": 1.083180877687151, + "teacher_mass_on_student_prediction": 0.4064582333690476, + "teacher_max_mass": 0.46057752289523396, + "teacher_optimal_mass": 0.46057752289523396, + "variance_verification_error": 9.71445146547012e-17 + }, + { + "current_model_mistake": false, + "enumerated_mean_logit_gradient": [ + -9.470913300632865e-05, + 4.513590183152324e-05, + 5.44495016565272e-05, + -4.876270481721823e-06 + ], + "episode_id": "train-travel-2300098", + "exact_gradient_norm": 0.000118302982859311, + "exact_logit_gradient": [ + -9.470913300632865e-05, + 4.5135901831523305e-05, + 5.4449501656527205e-05, + -4.876270481721831e-06 + ], + "expected_cosine": 0.30080905282894865, + "feedback_advantage": -0.0002435611877535937, + "id": "train-travel-2300098/frozen-warmstart-v2/seed-1707/step-0", + "kind": "focus", + "maximum_expectation_error": 6.776263578034403e-20, + "mc4_error_norm_quantiles": { + "0.1": 0.00011828977576059796, + "0.25": 0.00011828977576059796, + "0.5": 0.00011828977576059796, + "0.75": 0.00011828977576059796, + "0.9": 0.00011828977576059796 + }, + "mc4_gradient_norm_quantiles": { + "0.1": 4.393254663480912e-08, + "0.25": 4.393254663480912e-08, + "0.5": 4.393254663480912e-08, + "0.75": 4.393254663480912e-08, + "0.9": 4.393254663480912e-08 + }, + "mc4_gradient_rmse": 0.009562628254256763, + "noise_signal_ratio": 80.83167493442572, + "optimal_choices": [ + "B", + "C" + ], + "p": [ + 5.828415707731219e-05, + 0.9998198217369432, + 0.00010888915202276996, + 1.3004953956695169e-05 + ], + "probability_all_four_same": 0.9992794817096153, + "probability_no_optimal_in_four": 2.5828120064841875e-17, + "probability_opposed_gradient": 0.0, + "q": [ + 0.00029593200107304394, + 0.999619118355799, + 6.603134541340067e-05, + 1.891829771455445e-05 + ], + "reverse_kl": 0.00015561566512250187, + "student_entropy_nats": 0.00188838427096119, + "student_max_mass": 0.9998198217369432, + "student_optimal_mass": 0.999928710888966, + "student_predicted_action_mass": 0.9998198217369432, + "teacher_entropy_nats": 0.0036266898682119097, + "teacher_mass_on_student_prediction": 0.999619118355799, + "teacher_max_mass": 0.999619118355799, + "teacher_optimal_mass": 0.9996851497012124, + "variance_verification_error": 2.439454888092385e-19 + }, + { + "current_model_mistake": false, + "enumerated_mean_logit_gradient": [ + 0.013474619156485042, + -0.01740430700392289, + 0.0005916931632760389, + 0.0033379946841617953 + ], + "episode_id": "train-travel-2300003", + "exact_gradient_norm": 0.022270329432005992, + "exact_logit_gradient": [ + 0.013474619156485037, + -0.01740430700392287, + 0.0005916931632760401, + 0.0033379946841617953 + ], + "expected_cosine": 0.9881118471121266, + "feedback_advantage": 0.010361319681976067, + "id": "train-travel-2300003/frozen-warmstart-v2/seed-1707/step-0", + "kind": "focus", + "maximum_expectation_error": 2.0816681711721685e-17, + "mc4_error_norm_quantiles": { + "0.1": 0.02206519863312447, + "0.25": 0.02206519863312447, + "0.5": 0.02206519863312447, + "0.75": 0.02206519863312447, + "0.9": 0.02206519863312447 + }, + "mc4_gradient_norm_quantiles": { + "0.1": 0.00020613412762290285, + "0.25": 0.00020613412762290285, + "0.5": 0.00020613412762290285, + "0.75": 0.00020613412762290285, + "0.9": 0.00020613412762290285 + }, + "mc4_gradient_rmse": 0.10019239072796059, + "noise_signal_ratio": 4.498918214652373, + "optimal_choices": [ + "B" + ], + "p": [ + 0.010934370394379795, + 0.9842805704442221, + 0.00022693558639279288, + 0.004558123575005298 + ], + "probability_all_four_same": 0.9385894232418045, + "probability_no_optimal_in_four": 6.105864007950568e-08, + "probability_opposed_gradient": 0.0, + "q": [ + 0.0031657272167900847, + 0.9946418901261982, + 1.6612211600232937e-05, + 0.002175770445411524 + ], + "reverse_kl": 0.007210488097238576, + "student_entropy_nats": 0.09144945535824829, + "student_max_mass": 0.9842805704442221, + "student_optimal_mass": 0.9842805704442221, + "student_predicted_action_mass": 0.9842805704442221, + "teacher_entropy_nats": 0.03708477557484377, + "teacher_mass_on_student_prediction": 0.9946418901261982, + "teacher_max_mass": 0.9946418901261982, + "teacher_optimal_mass": 0.9946418901261982, + "variance_verification_error": 3.469446951953614e-18 + }, + { + "current_model_mistake": false, + "enumerated_mean_logit_gradient": [ + -0.0020493112558621824, + -0.0034870762542568794, + 0.0264349920811873, + -0.020898604571068246 + ], + "episode_id": "train-travel-2300047", + "exact_gradient_norm": 0.03393994485523976, + "exact_logit_gradient": [ + -0.0020493112558621807, + -0.0034870762542568777, + 0.026434992081187304, + -0.020898604571068246 + ], + "expected_cosine": 0.904454902500739, + "feedback_advantage": 0.004415656290430725, + "id": "train-travel-2300047/frozen-warmstart-v2/seed-1707/step-1", + "kind": "admin", + "maximum_expectation_error": 3.469446951953614e-18, + "mc4_error_norm_quantiles": { + "0.1": 0.03381329996647685, + "0.25": 0.03381329996647685, + "0.5": 0.03381329996647685, + "0.75": 0.03381329996647685, + "0.9": 0.03381329996647685 + }, + "mc4_gradient_norm_quantiles": { + "0.1": 0.00013629004707972953, + "0.25": 0.00013629004707972953, + "0.5": 0.00013629004707972953, + "0.75": 0.00013629004707972953, + "0.9": 0.00013629004707972953 + }, + "mc4_gradient_rmse": 0.1465996990238956, + "noise_signal_ratio": 4.319385303929363, + "optimal_choices": [ + "D" + ], + "p": [ + 0.0014675574366659467, + 0.004520395552278012, + 0.01787850919417323, + 0.9761335378168828 + ], + "probability_all_four_same": 0.9078978483034655, + "probability_no_optimal_in_four": 3.2445329318806837e-07, + "probability_opposed_gradient": 0.02248330763021639, + "q": [ + 0.005830558705192904, + 0.009612966695923018, + 0.004007280491570476, + 0.9805491941073136 + ], + "reverse_kl": 0.01689615713180947, + "student_entropy_nats": 0.12950616631084078, + "student_max_mass": 0.9761335378168828, + "student_optimal_mass": 0.9761335378168828, + "student_predicted_action_mass": 0.9761335378168828, + "teacher_entropy_nats": 0.11602408839959666, + "teacher_mass_on_student_prediction": 0.9805491941073136, + "teacher_max_mass": 0.9805491941073136, + "teacher_optimal_mass": 0.9805491941073136, + "variance_verification_error": 6.938893903907228e-18 + }, + { + "current_model_mistake": false, + "enumerated_mean_logit_gradient": [ + 0.0006018437680953504, + -0.00244254249139797, + 0.005013224764055906, + -0.003172526040753287 + ], + "episode_id": "train-travel-2300105", + "exact_gradient_norm": 0.00644403396621099, + "exact_logit_gradient": [ + 0.0006018437680953506, + -0.0024425424913979694, + 0.005013224764055906, + -0.0031725260407532876 + ], + "expected_cosine": 0.38496892935814414, + "feedback_advantage": 0.0006584259514138546, + "id": "train-travel-2300105/frozen-warmstart-v2/seed-1707/step-1", + "kind": "admin", + "maximum_expectation_error": 4.336808689942018e-19, + "mc4_error_norm_quantiles": { + "0.1": 0.006438941386930955, + "0.25": 0.006438941386930955, + "0.5": 0.006438941386930955, + "0.75": 0.006438941386930955, + "0.9": 0.006438941386930955 + }, + "mc4_gradient_norm_quantiles": { + "0.1": 1.294688532456474e-05, + "0.25": 1.294688532456474e-05, + "0.5": 1.294688532456474e-05, + "0.75": 1.294688532456474e-05, + "0.9": 1.294688532456474e-05 + }, + "mc4_gradient_rmse": 0.05860690688095274, + "noise_signal_ratio": 9.094754495127663, + "optimal_choices": [ + "D" + ], + "p": [ + 0.0024388305621309966, + 0.009645762386888456, + 0.004020951882083211, + 0.9838944551688973 + ], + "probability_all_four_same": 0.9371175179799742, + "probability_no_optimal_in_four": 6.728243248084133e-08, + "probability_opposed_gradient": 0.037292657437581145, + "q": [ + 0.001900634186565343, + 0.012393691839609752, + 0.0011527928535137674, + 0.9845528811203111 + ], + "reverse_kl": 0.0025554776617350405, + "student_entropy_nats": 0.09759651773281612, + "student_max_mass": 0.9838944551688973, + "student_optimal_mass": 0.9838944551688973, + "student_predicted_action_mass": 0.9838944551688973, + "teacher_entropy_nats": 0.08945038580748128, + "teacher_mass_on_student_prediction": 0.9845528811203111, + "teacher_max_mass": 0.9845528811203111, + "teacher_optimal_mass": 0.9845528811203111, + "variance_verification_error": 0.0 + }, + { + "current_model_mistake": false, + "enumerated_mean_logit_gradient": [ + -0.0010879534179755977, + 0.0044224712908988465, + 0.00034556907817624336, + -0.003680086951099491 + ], + "episode_id": "train-travel-2300097", + "exact_gradient_norm": 0.005865522390542348, + "exact_logit_gradient": [ + -0.0010879534179755977, + 0.004422471290898846, + 0.00034556907817624336, + -0.0036800869510994916 + ], + "expected_cosine": 0.9630670938821204, + "feedback_advantage": 0.0018811868302414103, + "id": "train-travel-2300097/frozen-warmstart-v2/seed-1707/step-2", + "kind": "exercise", + "maximum_expectation_error": 8.673617379884035e-19, + "mc4_error_norm_quantiles": { + "0.1": 0.005816401764904327, + "0.25": 0.005816401764904327, + "0.5": 0.005816401764904327, + "0.75": 0.005816401764904327, + "0.9": 0.005816401764904327 + }, + "mc4_gradient_norm_quantiles": { + "0.1": 5.077348823894307e-05, + "0.25": 5.077348823894307e-05, + "0.5": 5.077348823894307e-05, + "0.75": 5.077348823894307e-05, + "0.9": 5.077348823894307e-05 + }, + "mc4_gradient_rmse": 0.03662503983780746, + "noise_signal_ratio": 6.24412241556902, + "optimal_choices": [ + "D" + ], + "p": [ + 0.0007890850427561239, + 0.017959493353639274, + 0.000696365122827155, + 0.9805550564807775 + ], + "probability_all_four_same": 0.9244596989068754, + "probability_no_optimal_in_four": 1.4296401751977488e-07, + "probability_opposed_gradient": 0.003149354185604229, + "q": [ + 0.003126879277244545, + 0.014013700319546074, + 0.00042317709219048485, + 0.9824362433110189 + ], + "reverse_kl": 0.0018364112499849486, + "student_entropy_nats": 0.10214533388908396, + "student_max_mass": 0.9805550564807775, + "student_optimal_mass": 0.9805550564807775, + "student_predicted_action_mass": 0.9805550564807775, + "teacher_entropy_nats": 0.09853723429401573, + "teacher_mass_on_student_prediction": 0.9824362433110189, + "teacher_max_mass": 0.9824362433110189, + "teacher_optimal_mass": 0.9824362433110189, + "variance_verification_error": 2.168404344971009e-19 + }, + { + "current_model_mistake": false, + "enumerated_mean_logit_gradient": [ + -0.06769198622208332, + 0.005034553123846168, + 0.042138226133353565, + 0.020519206964883633 + ], + "episode_id": "train-travel-2300014", + "exact_gradient_norm": 0.08248769411240035, + "exact_logit_gradient": [ + -0.06769198622208339, + 0.005034553123846166, + 0.042138226133353565, + 0.02051920696488367 + ], + "expected_cosine": 0.977911094964513, + "feedback_advantage": 0.02235257188016193, + "id": "train-travel-2300014/frozen-warmstart-v2/seed-1707/step-0", + "kind": "focus", + "maximum_expectation_error": 6.938893903907228e-17, + "mc4_error_norm_quantiles": { + "0.1": 0.0818425611791268, + "0.25": 0.0818425611791268, + "0.5": 0.0818425611791268, + "0.75": 0.0818425611791268, + "0.9": 0.0818425611791268 + }, + "mc4_gradient_norm_quantiles": { + "0.1": 0.0006494898495900554, + "0.25": 0.0006494898495900554, + "0.5": 0.0006494898495900554, + "0.75": 0.0006494898495900554, + "0.9": 0.0006494898495900554 + }, + "mc4_gradient_rmse": 0.3141400759369444, + "noise_signal_ratio": 3.808326554854196, + "optimal_choices": [ + "A" + ], + "p": [ + 0.9757559760656355, + 0.003519125681925219, + 0.012282955109477018, + 0.008441943142962283 + ], + "probability_all_four_same": 0.9064939140114987, + "probability_no_optimal_in_four": 3.454767427861949e-07, + "probability_opposed_gradient": 0.0, + "q": [ + 0.9981085479457974, + 0.0008032109225813211, + 0.00037940997703734717, + 0.0007088311545839058 + ], + "reverse_kl": 0.04672438318884662, + "student_entropy_nats": 0.1381749887639865, + "student_max_mass": 0.9757559760656355, + "student_optimal_mass": 0.9757559760656355, + "student_predicted_action_mass": 0.9757559760656355, + "teacher_entropy_nats": 0.015743000303976247, + "teacher_mass_on_student_prediction": 0.9981085479457974, + "teacher_max_mass": 0.9981085479457974, + "teacher_optimal_mass": 0.9981085479457974, + "variance_verification_error": 5.134781488891349e-16 + }, + { + "current_model_mistake": false, + "enumerated_mean_logit_gradient": [ + -0.12493250086844179, + 0.11924559151212821, + 0.00045953575519281725, + 0.005227373601120859 + ], + "episode_id": "train-travel-2300005", + "exact_gradient_norm": 0.1727865083743419, + "exact_logit_gradient": [ + -0.12493250086844179, + 0.11924559151212809, + 0.00045953575519281693, + 0.0052273736011208645 + ], + "expected_cosine": 0.9922025808080299, + "feedback_advantage": 0.05092907871136121, + "id": "train-travel-2300005/frozen-warmstart-v2/seed-1707/step-3", + "kind": "repair", + "maximum_expectation_error": 1.249000902703301e-16, + "mc4_error_norm_quantiles": { + "0.1": 0.16866537453694858, + "0.25": 0.16866537453694858, + "0.5": 0.16866537453694858, + "0.75": 0.16866537453694858, + "0.9": 0.6061698150201553 + }, + "mc4_gradient_norm_quantiles": { + "0.1": 0.004122127060169079, + "0.25": 0.004122127060169079, + "0.5": 0.004122127060169079, + "0.75": 0.004122127060169079, + "0.9": 0.7787673935043461 + }, + "mc4_gradient_rmse": 0.3540270130964255, + "noise_signal_ratio": 2.048927410058117, + "optimal_choices": [ + "A" + ], + "p": [ + 0.942573072830452, + 0.05317633612043027, + 0.0012505875232469218, + 0.003000003525870861 + ], + "probability_all_four_same": 0.7893407330152087, + "probability_no_optimal_in_four": 1.0875827577359436e-05, + "probability_opposed_gradient": 0.0, + "q": [ + 0.9935021515418132, + 0.005213420745561644, + 0.0007995039913215533, + 0.0004849237213035647 + ], + "reverse_kl": 0.07992132088533857, + "student_entropy_nats": 0.23755895855536202, + "student_max_mass": 0.942573072830452, + "student_optimal_mass": 0.942573072830452, + "student_predicted_action_mass": 0.942573072830452, + "teacher_entropy_nats": 0.04328351961481103, + "teacher_mass_on_student_prediction": 0.9935021515418132, + "teacher_max_mass": 0.9935021515418132, + "teacher_optimal_mass": 0.9935021515418132, + "variance_verification_error": 2.7755575615628914e-17 + }, + { + "current_model_mistake": true, + "enumerated_mean_logit_gradient": [ + 0.3378857537489706, + -0.05633091718372716, + -0.00033255466642961536, + -0.28122228189881293 + ], + "episode_id": "train-travel-2300027", + "exact_gradient_norm": 0.4431997712615067, + "exact_logit_gradient": [ + 0.3378857537489705, + -0.05633091718372718, + -0.0003325546664296146, + -0.2812222818988137 + ], + "expected_cosine": 0.9700373527419379, + "feedback_advantage": 0.32136591051181357, + "id": "train-travel-2300027/frozen-warmstart-v2/seed-1707/step-3", + "kind": "repair", + "maximum_expectation_error": 7.771561172376096e-16, + "mc4_error_norm_quantiles": { + "0.1": 0.05679169935538693, + "0.25": 0.05679169935538693, + "0.5": 0.1683185108708732, + "0.75": 0.20711911592263857, + "0.9": 0.4051690749027038 + }, + "mc4_gradient_norm_quantiles": { + "0.1": 0.27566880294936724, + "0.25": 0.27566880294936724, + "0.5": 0.45350479686757295, + "0.75": 0.6325494191723509, + "0.9": 0.6325494191723509 + }, + "mc4_gradient_rmse": 0.21258563104466127, + "noise_signal_ratio": 0.47966096742235614, + "optimal_choices": [ + "D" + ], + "p": [ + 0.7539702372131349, + 0.029234596366958524, + 0.0007790756353545842, + 0.21601609078455192 + ], + "probability_all_four_same": 0.3253375741734931, + "probability_no_optimal_in_four": 0.3777709833186056, + "probability_opposed_gradient": 0.0, + "q": [ + 0.3259386387807604, + 0.1358714388197977, + 0.0008079211030763976, + 0.5373820012963655 + ], + "reverse_kl": 0.39050172354675633, + "student_entropy_nats": 0.6527910828186357, + "student_max_mass": 0.7539702372131349, + "student_optimal_mass": 0.21601609078455192, + "student_predicted_action_mass": 0.7539702372131349, + "teacher_entropy_nats": 0.9760901407137892, + "teacher_mass_on_student_prediction": 0.3259386387807604, + "teacher_max_mass": 0.5373820012963655, + "teacher_optimal_mass": 0.5373820012963655, + "variance_verification_error": 6.938893903907228e-17 + }, + { + "current_model_mistake": true, + "enumerated_mean_logit_gradient": [ + 0.2658931757315638, + -0.26525169297593215, + -7.418718255666529e-05, + -0.0005672955730736408 + ], + "episode_id": "train-travel-2300004", + "exact_gradient_norm": 0.3755768481352648, + "exact_logit_gradient": [ + 0.2658931757315632, + -0.2652516929759328, + -7.418718255666556e-05, + -0.0005672955730736402 + ], + "expected_cosine": 0.9980656138999212, + "feedback_advantage": 0.8990556412909598, + "id": "train-travel-2300004/frozen-warmstart-v2/seed-1707/step-0", + "kind": "focus", + "maximum_expectation_error": 6.661338147750939e-16, + "mc4_error_norm_quantiles": { + "0.1": 0.18144415045622936, + "0.25": 0.18144415045622936, + "0.5": 0.18144415045622936, + "0.75": 0.18144415045622936, + "0.9": 0.7781428186213195 + }, + "mc4_gradient_norm_quantiles": { + "0.1": 0.1941416888145443, + "0.25": 0.1941416888145443, + "0.5": 0.1941416888145443, + "0.75": 0.1941416888145443, + "0.9": 1.1537192424707883 + }, + "mc4_gradient_rmse": 0.4084492579571785, + "noise_signal_ratio": 1.0875251229811551, + "optimal_choices": [ + "B" + ], + "p": [ + 0.9521048658701523, + 0.047402507767502156, + 2.7908476022791746e-05, + 0.0004647178863226946 + ], + "probability_all_four_same": 0.8217539628999065, + "probability_no_optimal_in_four": 0.8234509510729375, + "probability_opposed_gradient": 0.0016056832131731568, + "q": [ + 0.05339551650366007, + 0.946458149058462, + 2.9532225739351397e-05, + 0.00011680221213855986 + ], + "reverse_kl": 2.601679608132438, + "student_entropy_nats": 0.19512239118994906, + "student_max_mass": 0.9521048658701523, + "student_optimal_mass": 0.047402507767502156, + "student_predicted_action_mass": 0.9521048658701523, + "teacher_entropy_nats": 0.20989825100506473, + "teacher_mass_on_student_prediction": 0.05339551650366007, + "teacher_max_mass": 0.946458149058462, + "teacher_optimal_mass": 0.946458149058462, + "variance_verification_error": 5.273559366969494e-16 + }, + { + "current_model_mistake": false, + "enumerated_mean_logit_gradient": [ + -6.707212763008498e-05, + 0.0010331514563725363, + 0.050173944181779256, + -0.05114002351052171 + ], + "episode_id": "train-travel-2300067", + "exact_gradient_norm": 0.07165053091230464, + "exact_logit_gradient": [ + -6.7072127630085e-05, + 0.0010331514563725367, + 0.050173944181779256, + -0.0511400235105217 + ], + "expected_cosine": 0.9946953242702508, + "feedback_advantage": 0.01423805877694384, + "id": "train-travel-2300067/frozen-warmstart-v2/seed-1707/step-2", + "kind": "exercise", + "maximum_expectation_error": 6.938893903907228e-18, + "mc4_error_norm_quantiles": { + "0.1": 0.07134958258241428, + "0.25": 0.07134958258241428, + "0.5": 0.07134958258241428, + "0.75": 0.07134958258241428, + "0.9": 0.07134958258241428 + }, + "mc4_gradient_norm_quantiles": { + "0.1": 0.00030168712058861654, + "0.25": 0.00030168712058861654, + "0.5": 0.00030168712058861654, + "0.75": 0.00030168712058861654, + "0.9": 0.00030168712058861654 + }, + "mc4_gradient_rmse": 0.2966737296175303, + "noise_signal_ratio": 4.1405656851397055, + "optimal_choices": [ + "D" + ], + "p": [ + 3.4807437204915996e-05, + 0.0014800498779424195, + 0.014042322304346004, + 0.9844428203805067 + ], + "probability_all_four_same": 0.9392084730545313, + "probability_no_optimal_in_four": 5.857650612076576e-08, + "probability_opposed_gradient": 0.00013343933218095826, + "q": [ + 0.00023025572350082998, + 0.000709237581614553, + 0.00037962753743410287, + 0.9986808791574505 + ], + "reverse_kl": 0.03758872004782009, + "student_entropy_nats": 0.08533643307609821, + "student_max_mass": 0.9844428203805067, + "student_optimal_mass": 0.9844428203805067, + "student_predicted_action_mass": 0.9844428203805067, + "teacher_entropy_nats": 0.011379922650208395, + "teacher_mass_on_student_prediction": 0.9986808791574505, + "teacher_max_mass": 0.9986808791574505, + "teacher_optimal_mass": 0.9986808791574505, + "variance_verification_error": 0.0 + }, + { + "current_model_mistake": true, + "enumerated_mean_logit_gradient": [ + 0.9978674167121667, + -0.025723751689821134, + -0.7295797351711724, + -0.2425639298511736 + ], + "episode_id": "train-travel-2300100", + "exact_gradient_norm": 1.259970215034505, + "exact_logit_gradient": [ + 0.997867416712167, + -0.02572375168982122, + -0.7295797351711723, + -0.24256392985117375 + ], + "expected_cosine": 0.9538503863769654, + "feedback_advantage": 0.667731609691164, + "id": "train-travel-2300100/frozen-warmstart-v2/seed-1707/step-3", + "kind": "repair", + "maximum_expectation_error": 3.3306690738754696e-16, + "mc4_error_norm_quantiles": { + "0.1": 0.15845058123945693, + "0.25": 0.18149916070740962, + "0.5": 0.4534921762433018, + "0.75": 0.5309044998445983, + "0.9": 0.7628203957061829 + }, + "mc4_gradient_norm_quantiles": { + "0.1": 0.7436178044605168, + "0.25": 1.070097306464292, + "0.5": 1.293616944624837, + "0.75": 1.6883186531317904, + "0.9": 1.8791644676183341 + }, + "mc4_gradient_rmse": 0.49837130381887623, + "noise_signal_ratio": 0.39554213097428503, + "optimal_choices": [ + "C" + ], + "p": [ + 0.5816723081205398, + 0.015501041824458904, + 0.2139853080083369, + 0.18884134204666436 + ], + "probability_all_four_same": 0.1178442312002484, + "probability_no_optimal_in_four": 0.38170043541530924, + "probability_opposed_gradient": 0.0017435476559064592, + "q": [ + 0.014251626269213056, + 0.011099178478391208, + 0.8817169176995009, + 0.09293227755289488 + ], + "reverse_kl": 1.993521483291967, + "student_entropy_nats": 1.0244711124698347, + "student_max_mass": 0.5816723081205398, + "student_optimal_mass": 0.2139853080083369, + "student_predicted_action_mass": 0.5816723081205398, + "teacher_entropy_nats": 0.44232871989232597, + "teacher_mass_on_student_prediction": 0.014251626269213056, + "teacher_max_mass": 0.8817169176995009, + "teacher_optimal_mass": 0.8817169176995009, + "variance_verification_error": 2.220446049250313e-16 + }, + { + "current_model_mistake": false, + "enumerated_mean_logit_gradient": [ + -3.3582964598896226e-05, + -0.03100505889890367, + 0.03148568944209106, + -0.0004470475785884617 + ], + "episode_id": "train-travel-2300109", + "exact_gradient_norm": 0.04419121288583654, + "exact_logit_gradient": [ + -3.358296459889629e-05, + -0.031005058898903714, + 0.03148568944209105, + -0.0004470475785884621 + ], + "expected_cosine": 0.9990476502896647, + "feedback_advantage": -0.0008899685486395414, + "id": "train-travel-2300109/frozen-warmstart-v2/seed-1707/step-1", + "kind": "admin", + "maximum_expectation_error": 4.5102810375396984e-17, + "mc4_error_norm_quantiles": { + "0.1": 0.04002569921978557, + "0.25": 0.04002569921978557, + "0.5": 0.04002569921978557, + "0.75": 0.04002569921978557, + "0.9": 0.04002569921978557 + }, + "mc4_gradient_norm_quantiles": { + "0.1": 0.004166928971764845, + "0.25": 0.004166928971764845, + "0.5": 0.004166928971764845, + "0.75": 0.004166928971764845, + "0.9": 0.004166928971764845 + }, + "mc4_gradient_rmse": 0.22809840189262814, + "noise_signal_ratio": 5.161623476637696, + "optimal_choices": [ + "B", + "C" + ], + "p": [ + 8.3316640061574e-06, + 0.007574655595042751, + 0.9920842055700893, + 0.0003328071708618576 + ], + "probability_all_four_same": 0.9687108042988676, + "probability_no_optimal_in_four": 1.3543304633965684e-14, + "probability_opposed_gradient": 0.0, + "q": [ + 0.0003310957872176103, + 0.3204263342682755, + 0.678342558348217, + 0.0009000115962898946 + ], + "reverse_kl": 0.3484186673671653, + "student_entropy_nats": 0.047633571160248724, + "student_max_mass": 0.9920842055700893, + "student_optimal_mass": 0.999658861165132, + "student_predicted_action_mass": 0.9920842055700893, + "teacher_entropy_nats": 0.6369098048604879, + "teacher_mass_on_student_prediction": 0.678342558348217, + "teacher_max_mass": 0.678342558348217, + "teacher_optimal_mass": 0.9987688926164925, + "variance_verification_error": 8.326672684688674e-17 + }, + { + "current_model_mistake": false, + "enumerated_mean_logit_gradient": [ + -0.14841011021967426, + -0.012701720529363632, + -0.0007730271705603843, + 0.16188485791959867 + ], + "episode_id": "train-travel-2300014", + "exact_gradient_norm": 0.21998681622878452, + "exact_logit_gradient": [ + -0.14841011021967448, + -0.012701720529363653, + -0.0007730271705603829, + 0.16188485791959853 + ], + "expected_cosine": 0.9511390641730724, + "feedback_advantage": 0.02514613525437226, + "id": "train-travel-2300014/frozen-warmstart-v2/seed-1707/step-3", + "kind": "repair", + "maximum_expectation_error": 2.220446049250313e-16, + "mc4_error_norm_quantiles": { + "0.1": 0.2176177333502271, + "0.25": 0.2176177333502271, + "0.5": 0.2176177333502271, + "0.75": 0.2176177333502271, + "0.9": 0.7254629160366342 + }, + "mc4_gradient_norm_quantiles": { + "0.1": 0.0024055206447534663, + "0.25": 0.0024055206447534663, + "0.5": 0.0024055206447534663, + "0.75": 0.0024055206447534663, + "0.9": 0.9448321642156433 + }, + "mc4_gradient_rmse": 0.4573936401938185, + "noise_signal_ratio": 2.0791865987011366, + "optimal_choices": [ + "A" + ], + "p": [ + 0.932422812628552, + 0.007119135433979432, + 0.0008502595554676532, + 0.05960779238200093 + ], + "probability_all_four_same": 0.7558903770271762, + "probability_no_optimal_in_four": 2.085453200004104e-05, + "probability_opposed_gradient": 0.02617507122671224, + "q": [ + 0.9575689478829242, + 0.03712897623330731, + 0.0018485429534995526, + 0.0034535329302688945 + ], + "reverse_kl": 0.13255475486783894, + "student_entropy_nats": 0.27454794108425273, + "student_max_mass": 0.932422812628552, + "student_optimal_mass": 0.932422812628552, + "student_predicted_action_mass": 0.932422812628552, + "teacher_entropy_nats": 0.19500624232250063, + "teacher_mass_on_student_prediction": 0.9575689478829242, + "teacher_max_mass": 0.9575689478829242, + "teacher_optimal_mass": 0.9575689478829242, + "variance_verification_error": 3.0531133177191805e-16 + }, + { + "current_model_mistake": false, + "enumerated_mean_logit_gradient": [ + -0.00046419249095728225, + 0.006242794361616413, + 0.008039116019668184, + -0.013817717890327309 + ], + "episode_id": "train-travel-2300092", + "exact_gradient_norm": 0.017168129489972046, + "exact_logit_gradient": [ + -0.0004641924909572824, + 0.006242794361616412, + 0.008039116019668183, + -0.013817717890327312 + ], + "expected_cosine": 0.9769440678627644, + "feedback_advantage": 0.006320707894505007, + "id": "train-travel-2300092/frozen-warmstart-v2/seed-1707/step-1", + "kind": "admin", + "maximum_expectation_error": 3.469446951953614e-18, + "mc4_error_norm_quantiles": { + "0.1": 0.017096789872557496, + "0.25": 0.017096789872557496, + "0.5": 0.017096789872557496, + "0.75": 0.017096789872557496, + "0.9": 0.017096789872557496 + }, + "mc4_gradient_norm_quantiles": { + "0.1": 7.230837374769348e-05, + "0.25": 7.230837374769348e-05, + "0.5": 7.230837374769348e-05, + "0.75": 7.230837374769348e-05, + "0.9": 7.230837374769348e-05 + }, + "mc4_gradient_rmse": 0.11050699164711454, + "noise_signal_ratio": 6.436751989298659, + "optimal_choices": [ + "D" + ], + "p": [ + 0.0009031846676340646, + 0.004586750771219427, + 0.004047792985059052, + 0.9904622715760875 + ], + "probability_all_four_same": 0.9623914343498661, + "probability_no_optimal_in_four": 8.275224962078443e-09, + "probability_opposed_gradient": 0.0035151539649399223, + "q": [ + 0.0014986026218353453, + 0.0011671127997515069, + 0.0005513051078207092, + 0.9967829794705925 + ], + "reverse_kl": 0.007589479023035825, + "student_entropy_nats": 0.06282244280689273, + "student_max_mass": 0.9904622715760875, + "student_optimal_mass": 0.9904622715760875, + "student_predicted_action_mass": 0.9904622715760875, + "teacher_entropy_nats": 0.02497592304579952, + "teacher_mass_on_student_prediction": 0.9967829794705925, + "teacher_max_mass": 0.9967829794705925, + "teacher_optimal_mass": 0.9967829794705925, + "variance_verification_error": 5.204170427930421e-18 + }, + { + "current_model_mistake": false, + "enumerated_mean_logit_gradient": [ + 0.25782981913261926, + -0.003731048411547306, + -0.03063151604357286, + -0.22346725467749934 + ], + "episode_id": "train-travel-2300089", + "exact_gradient_norm": 0.3425872736171536, + "exact_logit_gradient": [ + 0.2578298191326193, + -0.003731048411547303, + -0.03063151604357284, + -0.2234672546774992 + ], + "expected_cosine": 0.91513005609092, + "feedback_advantage": 0.11812208442139249, + "id": "train-travel-2300089/frozen-warmstart-v2/seed-1707/step-3", + "kind": "repair", + "maximum_expectation_error": 1.3877787807814457e-16, + "mc4_error_norm_quantiles": { + "0.1": 0.18726062936147714, + "0.25": 0.2003085636812756, + "0.5": 0.29658874507949046, + "0.75": 0.309128991690491, + "0.9": 0.6833249937034823 + }, + "mc4_gradient_norm_quantiles": { + "0.1": 0.038413282693475305, + "0.25": 0.038413282693475305, + "0.5": 0.05995627236420255, + "0.75": 0.5428305174044976, + "0.9": 1.0250224641005505 + }, + "mc4_gradient_rmse": 0.36235720036485053, + "noise_signal_ratio": 1.0577077091596523, + "optimal_choices": [ + "D" + ], + "p": [ + 0.15384604932310894, + 0.0028177885180914555, + 0.15384604932310894, + 0.6894901128356907 + ], + "probability_all_four_same": 0.2271223481847666, + "probability_no_optimal_in_four": 0.00929612026579849, + "probability_opposed_gradient": 0.007354531927773241, + "q": [ + 0.024387775488692393, + 0.008971762095930768, + 0.1590282651582936, + 0.8076121972570832 + ], + "reverse_kl": 0.16597545752315895, + "student_entropy_nats": 0.8488388862165889, + "student_max_mass": 0.6894901128356907, + "student_optimal_mass": 0.6894901128356907, + "student_predicted_action_mass": 0.6894901128356907, + "teacher_entropy_nats": 0.5978243683048636, + "teacher_mass_on_student_prediction": 0.8076121972570832, + "teacher_max_mass": 0.8076121972570832, + "teacher_optimal_mass": 0.8076121972570832, + "variance_verification_error": 2.7755575615628914e-17 + }, + { + "current_model_mistake": false, + "enumerated_mean_logit_gradient": [ + -0.007378057840646901, + 0.0019134380169807943, + 0.0003811977943420969, + 0.005083422029324013 + ], + "episode_id": "train-travel-2300034", + "exact_gradient_norm": 0.009169704129980779, + "exact_logit_gradient": [ + -0.0073780578406469005, + 0.0019134380169807939, + 0.0003811977943420969, + 0.005083422029324013 + ], + "expected_cosine": 0.6724677439567966, + "feedback_advantage": -0.04318017131666618, + "id": "train-travel-2300034/frozen-warmstart-v2/seed-1707/step-0", + "kind": "focus", + "maximum_expectation_error": 8.673617379884035e-19, + "mc4_error_norm_quantiles": { + "0.1": 0.008956543382270608, + "0.25": 0.008956543382270608, + "0.5": 0.008956543382270608, + "0.75": 0.008956543382270608, + "0.9": 0.008956543382270608 + }, + "mc4_gradient_norm_quantiles": { + "0.1": 0.00031605352513837127, + "0.25": 0.00031605352513837127, + "0.5": 0.00031605352513837127, + "0.75": 0.00031605352513837127, + "0.9": 0.00031605352513837127 + }, + "mc4_gradient_rmse": 0.10780765597342505, + "noise_signal_ratio": 11.756939421953959, + "optimal_choices": [ + "D" + ], + "p": [ + 0.002463552810921449, + 0.0021740777380017716, + 0.0014942203069396968, + 0.9938681491441371 + ], + "probability_all_four_same": 0.9756972734035159, + "probability_no_optimal_in_four": 1.4137295380387705e-09, + "probability_opposed_gradient": 0.014484870202386176, + "q": [ + 0.047331965889866996, + 0.0008669151952769148, + 0.001113141087385188, + 0.9506879778274709 + ], + "reverse_kl": 0.03930385630865504, + "student_entropy_nats": 0.043960702925218045, + "student_max_mass": 0.9938681491441371, + "student_optimal_mass": 0.9938681491441371, + "student_predicted_action_mass": 0.9938681491441371, + "teacher_entropy_nats": 0.20614737725103185, + "teacher_mass_on_student_prediction": 0.9506879778274709, + "teacher_max_mass": 0.9506879778274709, + "teacher_optimal_mass": 0.9506879778274709, + "variance_verification_error": 3.469446951953614e-18 + }, + { + "current_model_mistake": false, + "enumerated_mean_logit_gradient": [ + -0.8762209923754437, + 0.8803563576243907, + -0.020912214686751942, + 0.01677684943780459 + ], + "episode_id": "train-travel-2300022", + "exact_gradient_norm": 1.2423805082539152, + "exact_logit_gradient": [ + -0.8762209923754439, + 0.8803563576243914, + -0.020912214686751966, + 0.016776849437804595 + ], + "expected_cosine": 0.9856297130930259, + "feedback_advantage": 0.26619724282466095, + "id": "train-travel-2300022/frozen-warmstart-v2/seed-1707/step-3", + "kind": "repair", + "maximum_expectation_error": 6.661338147750939e-16, + "mc4_error_norm_quantiles": { + "0.1": 0.05557006866141852, + "0.25": 0.05557006866141852, + "0.5": 1.0246392844724383, + "0.75": 1.1175755959987208, + "0.9": 1.1448950604205017 + }, + "mc4_gradient_norm_quantiles": { + "0.1": 0.12550858013860328, + "0.25": 0.12550858013860328, + "0.5": 1.1952865030630944, + "0.75": 2.2659108667631234, + "0.9": 2.2659108667631234 + }, + "mc4_gradient_rmse": 0.9522731961212197, + "noise_signal_ratio": 0.7664907729915834, + "optimal_choices": [ + "A" + ], + "p": [ + 0.7068707699382506, + 0.2600432544745501, + 0.024187785671140934, + 0.00889818991605842 + ], + "probability_all_four_same": 0.2542395470314795, + "probability_no_optimal_in_four": 0.007383061892091442, + "probability_opposed_gradient": 0.0, + "q": [ + 0.9730680127629115, + 0.0035094311961328287, + 0.022884367360190373, + 0.0005381886807652323 + ], + "reverse_kl": 0.9199712280383934, + "student_entropy_nats": 0.7275139994986664, + "student_max_mass": 0.7068707699382506, + "student_optimal_mass": 0.7068707699382506, + "student_predicted_action_mass": 0.7068707699382506, + "teacher_entropy_nats": 0.13689464127936896, + "teacher_mass_on_student_prediction": 0.9730680127629115, + "teacher_max_mass": 0.9730680127629115, + "teacher_optimal_mass": 0.9730680127629115, + "variance_verification_error": 6.661338147750939e-16 + }, + { + "current_model_mistake": false, + "enumerated_mean_logit_gradient": [ + -0.010597872279289655, + 0.006680305289151855, + -5.8059229453833186e-05, + 0.003975626219591647 + ], + "episode_id": "train-travel-2300042", + "exact_gradient_norm": 0.013143452754753416, + "exact_logit_gradient": [ + -0.010597872279289643, + 0.006680305289151838, + -5.805922945383309e-05, + 0.003975626219591639 + ], + "expected_cosine": 0.9910193199875835, + "feedback_advantage": 0.004946933868665515, + "id": "train-travel-2300042/frozen-warmstart-v2/seed-1707/step-0", + "kind": "focus", + "maximum_expectation_error": 1.734723475976807e-17, + "mc4_error_norm_quantiles": { + "0.1": 0.013104790112144835, + "0.25": 0.013104790112144835, + "0.5": 0.013104790112144835, + "0.75": 0.013104790112144835, + "0.9": 0.013104790112144835 + }, + "mc4_gradient_norm_quantiles": { + "0.1": 3.8791584297097185e-05, + "0.25": 3.8791584297097185e-05, + "0.5": 3.8791584297097185e-05, + "0.75": 3.8791584297097185e-05, + "0.9": 3.8791584297097185e-05 + }, + "mc4_gradient_rmse": 0.09677700413329211, + "noise_signal_ratio": 7.363134021103555, + "optimal_choices": [ + "A" + ], + "p": [ + 0.9935261262887226, + 0.003583214605092831, + 0.0004279539748180446, + 0.002462705131366468 + ], + "probability_all_four_same": 0.9743548880509292, + "probability_no_optimal_in_four": 1.756535343417756e-09, + "probability_opposed_gradient": 0.0016798694242576157, + "q": [ + 0.9984730601573881, + 0.0005522398596608147, + 0.0004873499914755234, + 0.0004873499914755234 + ], + "reverse_kl": 0.005700115316064644, + "student_entropy_nats": 0.044743376368403365, + "student_max_mass": 0.9935261262887226, + "student_optimal_mass": 0.9935261262887226, + "student_predicted_action_mass": 0.9935261262887226, + "teacher_entropy_nats": 0.013101993024654722, + "teacher_mass_on_student_prediction": 0.9984730601573881, + "teacher_max_mass": 0.9984730601573881, + "teacher_optimal_mass": 0.9984730601573881, + "variance_verification_error": 1.9081958235744878e-17 + }, + { + "current_model_mistake": false, + "enumerated_mean_logit_gradient": [ + -0.00023126585615590868, + 0.0024405997994277734, + -0.006020880588161608, + 0.00381154664488974 + ], + "episode_id": "train-travel-2300105", + "exact_gradient_norm": 0.007535841171379739, + "exact_logit_gradient": [ + -0.00023126585615590857, + 0.002440599799427779, + -0.00602088058816161, + 0.0038115466448897396 + ], + "expected_cosine": 0.9912572808343121, + "feedback_advantage": 0.002057112729382138, + "id": "train-travel-2300105/frozen-warmstart-v2/seed-1707/step-2", + "kind": "exercise", + "maximum_expectation_error": 5.637851296924623e-18, + "mc4_error_norm_quantiles": { + "0.1": 0.007528525109108801, + "0.25": 0.007528525109108801, + "0.5": 0.007528525109108801, + "0.75": 0.007528525109108801, + "0.9": 0.007528525109108801 + }, + "mc4_gradient_norm_quantiles": { + "0.1": 7.360522435946972e-06, + "0.25": 7.360522435946972e-06, + "0.5": 7.360522435946972e-06, + "0.75": 7.360522435946972e-06, + "0.9": 7.360522435946972e-06 + }, + "mc4_gradient_rmse": 0.08521474269869302, + "noise_signal_ratio": 11.307927112679716, + "optimal_choices": [ + "C" + ], + "p": [ + 0.00022987771480888108, + 0.0010302404586715796, + 0.9970413023905915, + 0.0016985794359279827 + ], + "probability_all_four_same": 0.9882176293972418, + "probability_no_optimal_in_four": 7.663061716270434e-11, + "probability_opposed_gradient": 0.0009116885571949127, + "q": [ + 0.0006261615749632064, + 9.602499582722323e-05, + 0.9990984151199737, + 0.00017939830923595414 + ], + "reverse_kl": 0.003977655760073705, + "student_entropy_nats": 0.022799655951323265, + "student_max_mass": 0.9970413023905915, + "student_optimal_mass": 0.9970413023905915, + "student_predicted_action_mass": 0.9970413023905915, + "teacher_entropy_nats": 0.007955474883056677, + "teacher_mass_on_student_prediction": 0.9990984151199737, + "teacher_max_mass": 0.9990984151199737, + "teacher_optimal_mass": 0.9990984151199737, + "variance_verification_error": 1.734723475976807e-18 + }, + { + "current_model_mistake": false, + "enumerated_mean_logit_gradient": [ + 0.20369843728790235, + -0.001854121346311461, + 0.6040485896290494, + -0.8058929055706403 + ], + "episode_id": "train-travel-2300064", + "exact_gradient_norm": 1.0275381087830429, + "exact_logit_gradient": [ + 0.20369843728790252, + -0.0018541213463114609, + 0.6040485896290491, + -0.8058929055706402 + ], + "expected_cosine": 0.9084266281029794, + "feedback_advantage": 0.2595377430142579, + "id": "train-travel-2300064/frozen-warmstart-v2/seed-1707/step-1", + "kind": "admin", + "maximum_expectation_error": 3.3306690738754696e-16, + "mc4_error_norm_quantiles": { + "0.1": 0.37646695338388786, + "0.25": 0.37646695338388786, + "0.5": 0.9255847691483304, + "0.75": 0.9508158464106877, + "0.9": 1.3478445517454225 + }, + "mc4_gradient_norm_quantiles": { + "0.1": 0.10214813229127402, + "0.25": 0.10214813229127402, + "0.5": 1.188101232628722, + "0.75": 1.7369098459603591, + "0.9": 2.280936694339866 + }, + "mc4_gradient_rmse": 0.9921821205243262, + "noise_signal_ratio": 0.965591555236243, + "optimal_choices": [ + "D" + ], + "p": [ + 0.0769703816611041, + 0.008112618503227755, + 0.18464234024221, + 0.7302746595934582 + ], + "probability_all_four_same": 0.28560746217643823, + "probability_no_optimal_in_four": 0.005292818475050305, + "probability_opposed_gradient": 0.0, + "q": [ + 0.0024534998621739256, + 0.0045837413941498175, + 0.0031503561359601845, + 0.9898124026077161 + ], + "reverse_kl": 0.7994529554638221, + "student_entropy_nats": 0.7779079402743329, + "student_max_mass": 0.7302746595934582, + "student_optimal_mass": 0.7302746595934582, + "student_predicted_action_mass": 0.7302746595934582, + "teacher_entropy_nats": 0.06771300203350461, + "teacher_mass_on_student_prediction": 0.9898124026077161, + "teacher_max_mass": 0.9898124026077161, + "teacher_optimal_mass": 0.9898124026077161, + "variance_verification_error": 1.1102230246251565e-16 + }, + { + "current_model_mistake": true, + "enumerated_mean_logit_gradient": [ + -0.00024175978637244874, + 0.01883162011112871, + -0.01713872496420923, + -0.001451135360547036 + ], + "episode_id": "train-travel-2300005", + "exact_gradient_norm": 0.025505490605703153, + "exact_logit_gradient": [ + -0.00024175978637244863, + 0.01883162011112868, + -0.017138724964209238, + -0.0014511353605470355 + ], + "expected_cosine": 0.9940034839272041, + "feedback_advantage": 0.8442547458927114, + "id": "train-travel-2300005/frozen-warmstart-v2/seed-1707/step-1", + "kind": "admin", + "maximum_expectation_error": 3.122502256758253e-17, + "mc4_error_norm_quantiles": { + "0.1": 0.018837559857286315, + "0.25": 0.018837559857286315, + "0.5": 0.018837559857286315, + "0.75": 0.018837559857286315, + "0.9": 0.018837559857286315 + }, + "mc4_gradient_norm_quantiles": { + "0.1": 0.006714614136795115, + "0.25": 0.006714614136795115, + "0.5": 0.006714614136795115, + "0.75": 0.006714614136795115, + "0.9": 0.006714614136795115 + }, + "mc4_gradient_rmse": 0.19783263892525765, + "noise_signal_ratio": 7.75647259578772, + "optimal_choices": [ + "C" + ], + "p": [ + 3.5261909678299776e-05, + 0.9972963854291457, + 0.002181577014188315, + 0.000486775646987717 + ], + "probability_all_four_same": 0.9892293199345791, + "probability_no_optimal_in_four": 0.9913022061045822, + "probability_opposed_gradient": 0.0, + "q": [ + 0.005033094088474196, + 0.14708857712993326, + 0.8464363229068997, + 0.0014420058746927554 + ], + "reverse_kl": 1.8951303606487118, + "student_entropy_nats": 0.020142533976262124, + "student_max_mass": 0.9972963854291457, + "student_optimal_mass": 0.002181577014188315, + "student_predicted_action_mass": 0.9972963854291457, + "teacher_entropy_nats": 0.45911270949147354, + "teacher_mass_on_student_prediction": 0.14708857712993326, + "teacher_max_mass": 0.8464363229068997, + "teacher_optimal_mass": 0.8464363229068997, + "variance_verification_error": 6.938893903907228e-17 + }, + { + "current_model_mistake": true, + "enumerated_mean_logit_gradient": [ + 0.14775293609219553, + 6.90375792972434e-05, + 0.00139743079356909, + -0.14921940446506202 + ], + "episode_id": "train-travel-2300106", + "exact_gradient_norm": 0.20999837706940086, + "exact_logit_gradient": [ + 0.14775293609219545, + 6.903757929724365e-05, + 0.0013974307935690933, + -0.14921940446506177 + ], + "expected_cosine": 0.9865046723274769, + "feedback_advantage": 0.36978512802960584, + "id": "train-travel-2300106/frozen-warmstart-v2/seed-1707/step-0", + "kind": "focus", + "maximum_expectation_error": 2.498001805406602e-16, + "mc4_error_norm_quantiles": { + "0.1": 0.16086695974505943, + "0.25": 0.16086695974505943, + "0.5": 0.16086695974505943, + "0.75": 0.2576428727462663, + "0.9": 0.44264161141603725 + }, + "mc4_gradient_norm_quantiles": { + "0.1": 0.04918045342064891, + "0.25": 0.04918045342064891, + "0.5": 0.04918045342064891, + "0.75": 0.11670531454175051, + "0.9": 0.6526301085316878 + }, + "mc4_gradient_rmse": 0.3052025064176292, + "noise_signal_ratio": 1.453356500544597, + "optimal_choices": [ + "D" + ], + "p": [ + 0.9295424227274687, + 0.00203336589133058, + 0.0010883822110893122, + 0.0673358291701115 + ], + "probability_all_four_same": 0.7466014324629027, + "probability_no_optimal_in_four": 0.7566606915751338, + "probability_opposed_gradient": 0.01007981766675352, + "q": [ + 0.5612744259070631, + 0.0013912602273896235, + 0.00021335666582988718, + 0.43712095719971733 + ], + "reverse_kl": 0.34553015550726907, + "student_entropy_nats": 0.2696203281250995, + "student_max_mass": 0.9295424227274687, + "student_optimal_mass": 0.0673358291701115, + "student_predicted_action_mass": 0.9295424227274687, + "teacher_entropy_nats": 0.6968533099221963, + "teacher_mass_on_student_prediction": 0.5612744259070631, + "teacher_max_mass": 0.5612744259070631, + "teacher_optimal_mass": 0.43712095719971733, + "variance_verification_error": 1.3877787807814457e-16 + }, + { + "current_model_mistake": false, + "enumerated_mean_logit_gradient": [ + -0.007082111521726467, + 0.0038088151702956124, + -2.364487580533253e-05, + 0.0032969412272361863 + ], + "episode_id": "train-travel-2300002", + "exact_gradient_norm": 0.008691015886744556, + "exact_logit_gradient": [ + -0.007082111521726466, + 0.0038088151702956163, + -2.3644875805332464e-05, + 0.003296941227236182 + ], + "expected_cosine": 0.9953366965827455, + "feedback_advantage": 0.002707994252822088, + "id": "train-travel-2300002/frozen-warmstart-v2/seed-1707/step-0", + "kind": "focus", + "maximum_expectation_error": 4.336808689942018e-18, + "mc4_error_norm_quantiles": { + "0.1": 0.00868056327971664, + "0.25": 0.00868056327971664, + "0.5": 0.00868056327971664, + "0.75": 0.00868056327971664, + "0.9": 0.00868056327971664 + }, + "mc4_gradient_norm_quantiles": { + "0.1": 1.0473361786182919e-05, + "0.25": 1.0473361786182919e-05, + "0.5": 1.0473361786182919e-05, + "0.75": 1.0473361786182919e-05, + "0.9": 1.0473361786182919e-05 + }, + "mc4_gradient_rmse": 0.09140606930636627, + "noise_signal_ratio": 10.517305513821212, + "optimal_choices": [ + "A" + ], + "p": [ + 0.996800310291189, + 0.001698168937580159, + 0.00017898566270454098, + 0.0013225351085262669 + ], + "probability_all_four_same": 0.9872625383324704, + "probability_no_optimal_in_four": 1.0481693542837643e-10, + "probability_opposed_gradient": 0.0007092832438404427, + "q": [ + 0.9995083045440111, + 0.00017947190686978443, + 0.00020336832093491744, + 0.0001088552281842037 + ], + "reverse_kl": 0.004391841536732181, + "student_entropy_nats": 0.024336192919658772, + "student_max_mass": 0.996800310291189, + "student_optimal_mass": 0.996800310291189, + "student_predicted_action_mass": 0.996800310291189, + "teacher_entropy_nats": 0.004761696264664455, + "teacher_mass_on_student_prediction": 0.9995083045440111, + "teacher_max_mass": 0.9995083045440111, + "teacher_optimal_mass": 0.9995083045440111, + "variance_verification_error": 3.469446951953614e-18 + }, + { + "current_model_mistake": true, + "enumerated_mean_logit_gradient": [ + 0.3608407122605657, + -0.007715080604180267, + -0.4185907014259372, + 0.06546506976955163 + ], + "episode_id": "train-travel-2300035", + "exact_gradient_norm": 0.5565693063524356, + "exact_logit_gradient": [ + 0.3608407122605655, + -0.007715080604180244, + -0.4185907014259369, + 0.06546506976955166 + ], + "expected_cosine": 0.7694933720192478, + "feedback_advantage": 0.6495361229072096, + "id": "train-travel-2300035/frozen-warmstart-v2/seed-1707/step-3", + "kind": "repair", + "maximum_expectation_error": 2.7755575615628914e-16, + "mc4_error_norm_quantiles": { + "0.1": 0.1977219894778795, + "0.25": 0.23730707284301666, + "0.5": 0.43295309207180876, + "0.75": 0.5902606341285435, + "0.9": 0.7329292378277034 + }, + "mc4_gradient_norm_quantiles": { + "0.1": 0.38758429764100183, + "0.25": 0.6421996990953717, + "0.5": 0.69651019641845, + "0.75": 0.9033601598951685, + "0.9": 0.9808242483824586 + }, + "mc4_gradient_rmse": 0.5139771974218647, + "noise_signal_ratio": 0.9234738451358285, + "optimal_choices": [ + "C" + ], + "p": [ + 0.6204304625790887, + 0.1777563075226958, + 0.15686939481419257, + 0.044943835084022936 + ], + "probability_all_four_same": 0.14978217658724202, + "probability_no_optimal_in_four": 0.5053350442482901, + "probability_opposed_gradient": 0.029870049647962013, + "q": [ + 0.12366629153612706, + 0.06619379195620012, + 0.8064055177214022, + 0.0037343987862705685 + ], + "reverse_kl": 1.0312294420995018, + "student_entropy_nats": 1.0332101023154654, + "student_max_mass": 0.6204304625790887, + "student_optimal_mass": 0.15686939481419257, + "student_predicted_action_mass": 0.6204304625790887, + "teacher_entropy_nats": 0.6325997126593732, + "teacher_mass_on_student_prediction": 0.12366629153612706, + "teacher_max_mass": 0.8064055177214022, + "teacher_optimal_mass": 0.8064055177214022, + "variance_verification_error": 1.6653345369377348e-16 + }, + { + "current_model_mistake": false, + "enumerated_mean_logit_gradient": [ + -0.0014098701063391898, + -0.0018486006684186643, + 0.006646752687935227, + -0.003388281913177374 + ], + "episode_id": "train-travel-2300048", + "exact_gradient_norm": 0.00781439913018051, + "exact_logit_gradient": [ + -0.0014098701063391893, + -0.0018486006684186648, + 0.006646752687935228, + -0.003388281913177374 + ], + "expected_cosine": -0.6968715146506224, + "feedback_advantage": -0.0015291232768135377, + "id": "train-travel-2300048/frozen-warmstart-v2/seed-1707/step-2", + "kind": "exercise", + "maximum_expectation_error": 8.673617379884035e-19, + "mc4_error_norm_quantiles": { + "0.1": 0.007828798465395196, + "0.25": 0.007828798465395196, + "0.5": 0.007828798465395196, + "0.75": 0.007828798465395196, + "0.9": 0.007828798465395196 + }, + "mc4_gradient_norm_quantiles": { + "0.1": 1.9215804803199454e-05, + "0.25": 1.9215804803199454e-05, + "0.5": 1.9215804803199454e-05, + "0.75": 1.9215804803199454e-05, + "0.9": 1.9215804803199454e-05 + }, + "mc4_gradient_rmse": 0.07004631251429096, + "noise_signal_ratio": 8.963749016064515, + "optimal_choices": [ + "D" + ], + "p": [ + 0.001022813802645201, + 0.002453602756748243, + 0.006669583475726729, + 0.9898539999648798 + ], + "probability_all_four_same": 0.9600294826024025, + "probability_no_optimal_in_four": 1.0596914686536107e-08, + "probability_opposed_gradient": 0.9735873813968013, + "q": [ + 0.0040390579801369494, + 0.0051862527411043885, + 0.0024498125906924685, + 0.9883248766880662 + ], + "reverse_kl": 0.0049690030047890394, + "student_entropy_nats": 0.06529920063547548, + "student_max_mass": 0.9898539999648798, + "student_optimal_mass": 0.9898539999648798, + "student_predicted_action_mass": 0.9898539999648798, + "teacher_entropy_nats": 0.07588533403166493, + "teacher_mass_on_student_prediction": 0.9883248766880662, + "teacher_max_mass": 0.9883248766880662, + "teacher_optimal_mass": 0.9883248766880662, + "variance_verification_error": 8.673617379884035e-19 + }, + { + "current_model_mistake": true, + "enumerated_mean_logit_gradient": [ + 0.020959167671264548, + 0.00031579916777738684, + -3.0771965302355573e-05, + -0.02124419487373958 + ], + "episode_id": "train-travel-2300072", + "exact_gradient_norm": 0.02984465113439629, + "exact_logit_gradient": [ + 0.02095916767126455, + 0.00031579916777738765, + -3.077196530235554e-05, + -0.02124419487373958 + ], + "expected_cosine": 0.991337658991088, + "feedback_advantage": 0.08672328231940699, + "id": "train-travel-2300072/frozen-warmstart-v2/seed-1707/step-0", + "kind": "focus", + "maximum_expectation_error": 3.469446951953614e-18, + "mc4_error_norm_quantiles": { + "0.1": 0.028684363539193177, + "0.25": 0.028684363539193177, + "0.5": 0.028684363539193177, + "0.75": 0.028684363539193177, + "0.9": 0.028684363539193177 + }, + "mc4_gradient_norm_quantiles": { + "0.1": 0.0011652862899309065, + "0.25": 0.0011652862899309065, + "0.5": 0.0011652862899309065, + "0.75": 0.0011652862899309065, + "0.9": 0.0011652862899309065 + }, + "mc4_gradient_rmse": 0.15584063171048718, + "noise_signal_ratio": 5.221727371136168, + "optimal_choices": [ + "D" + ], + "p": [ + 0.9905816745050308, + 0.0007971537760653927, + 5.096034003460766e-05, + 0.008570211378869291 + ], + "probability_all_four_same": 0.9628555986097445, + "probability_no_optimal_in_four": 0.9661573331402065, + "probability_opposed_gradient": 0.003103110369135271, + "q": [ + 0.9041195554626359, + 0.0005000544173129482, + 8.689642177488053e-05, + 0.09529349369827628 + ], + "reverse_kl": 0.07017227253685511, + "student_entropy_nats": 0.05635441578536559, + "student_max_mass": 0.9905816745050308, + "student_optimal_mass": 0.008570211378869291, + "student_predicted_action_mass": 0.9905816745050308, + "teacher_entropy_nats": 0.319758242907484, + "teacher_mass_on_student_prediction": 0.9041195554626359, + "teacher_max_mass": 0.9041195554626359, + "teacher_optimal_mass": 0.09529349369827628, + "variance_verification_error": 4.85722573273506e-17 + }, + { + "current_model_mistake": false, + "enumerated_mean_logit_gradient": [ + -0.029352950363649117, + -0.00042237324012071636, + -9.595199915033082e-05, + 0.029871275602920144 + ], + "episode_id": "train-travel-2300024", + "exact_gradient_norm": 0.041881695370848813, + "exact_logit_gradient": [ + -0.029352950363649162, + -0.0004223732401207169, + -9.595199915033086e-05, + 0.029871275602920213 + ], + "expected_cosine": 0.9960628171764748, + "feedback_advantage": 0.014812380841931594, + "id": "train-travel-2300024/frozen-warmstart-v2/seed-1707/step-2", + "kind": "exercise", + "maximum_expectation_error": 6.938893903907228e-17, + "mc4_error_norm_quantiles": { + "0.1": 0.04144408979109826, + "0.25": 0.04144408979109826, + "0.5": 0.04144408979109826, + "0.75": 0.04144408979109826, + "0.9": 0.04144408979109826 + }, + "mc4_gradient_norm_quantiles": { + "0.1": 0.0004379089100389855, + "0.25": 0.0004379089100389855, + "0.5": 0.0004379089100389855, + "0.75": 0.0004379089100389855, + "0.9": 0.0004379089100389855 + }, + "mc4_gradient_rmse": 0.14592962749505695, + "noise_signal_ratio": 3.4843295192063617, + "optimal_choices": [ + "A" + ], + "p": [ + 0.9790877943060572, + 0.0005415181395046617, + 5.036903598664221e-05, + 0.020320318518451477 + ], + "probability_all_four_same": 0.9189388797221959, + "probability_no_optimal_in_four": 1.912490858877578e-07, + "probability_opposed_gradient": 0.002224117594745984, + "q": [ + 0.9939001751479888, + 0.0011637373772191003, + 0.0003334163614098545, + 0.004602671113382228 + ], + "reverse_kl": 0.014964437965055422, + "student_entropy_nats": 0.10443397786160737, + "student_max_mass": 0.9790877943060572, + "student_optimal_mass": 0.9790877943060572, + "student_predicted_action_mass": 0.9790877943060572, + "teacher_entropy_nats": 0.04138042012033426, + "teacher_mass_on_student_prediction": 0.9939001751479888, + "teacher_max_mass": 0.9939001751479888, + "teacher_optimal_mass": 0.9939001751479888, + "variance_verification_error": 3.469446951953614e-17 + }, + { + "current_model_mistake": false, + "enumerated_mean_logit_gradient": [ + -0.0019454184074194478, + -0.00019837907763855628, + 0.000227667555401616, + 0.0019161299296563922 + ], + "episode_id": "train-travel-2300050", + "exact_gradient_norm": 0.0027472519836153893, + "exact_logit_gradient": [ + -0.0019454184074194484, + -0.00019837907763855628, + 0.00022766755540161592, + 0.0019161299296563887 + ], + "expected_cosine": 0.9854701491546969, + "feedback_advantage": -0.0024270301264097682, + "id": "train-travel-2300050/frozen-warmstart-v2/seed-1707/step-1", + "kind": "admin", + "maximum_expectation_error": 3.469446951953614e-18, + "mc4_error_norm_quantiles": { + "0.1": 0.002727434905835206, + "0.25": 0.002727434905835206, + "0.5": 0.002727434905835206, + "0.75": 0.002727434905835206, + "0.9": 0.002727434905835206 + }, + "mc4_gradient_norm_quantiles": { + "0.1": 2.0054882382571803e-05, + "0.25": 2.0054882382571803e-05, + "0.5": 2.0054882382571803e-05, + "0.75": 2.0054882382571803e-05, + "0.9": 2.0054882382571803e-05 + }, + "mc4_gradient_rmse": 0.021896946451803595, + "noise_signal_ratio": 7.970490723965979, + "optimal_choices": [ + "D" + ], + "p": [ + 0.005214595837374895, + 0.0007996841658522994, + 0.0002596194166794704, + 0.9937261005800934 + ], + "probability_all_four_same": 0.975139587684885, + "probability_no_optimal_in_four": 1.5493523959469184e-09, + "probability_opposed_gradient": 0.001037963094069205, + "q": [ + 0.007568661418791174, + 0.0010243069615592996, + 0.0001079611659658851, + 0.9912990704536836 + ], + "reverse_kl": 0.0005171131799106122, + "student_entropy_nats": 0.04150990126922385, + "student_max_mass": 0.9937261005800934, + "student_optimal_mass": 0.9937261005800934, + "student_predicted_action_mass": 0.9937261005800934, + "teacher_entropy_nats": 0.053663484471515734, + "teacher_mass_on_student_prediction": 0.9912990704536836, + "teacher_max_mass": 0.9912990704536836, + "teacher_optimal_mass": 0.9912990704536836, + "variance_verification_error": 3.7947076036992655e-19 + }, + { + "current_model_mistake": false, + "enumerated_mean_logit_gradient": [ + 0.051169549223865934, + -0.029248738825824027, + 0.009372995211141143, + -0.03129380560918309 + ], + "episode_id": "train-travel-2300074", + "exact_gradient_norm": 0.06738669601196012, + "exact_logit_gradient": [ + 0.05116954922386597, + -0.02924873882582402, + 0.009372995211141141, + -0.03129380560918309 + ], + "expected_cosine": 0.432585736985595, + "feedback_advantage": 0.012649863747963508, + "id": "train-travel-2300074/frozen-warmstart-v2/seed-1707/step-3", + "kind": "repair", + "maximum_expectation_error": 3.469446951953614e-17, + "mc4_error_norm_quantiles": { + "0.1": 0.0615016649140957, + "0.25": 0.06536144986033489, + "0.5": 0.07162064368980403, + "0.75": 0.09538952449835111, + "0.9": 0.2798879037390059 + }, + "mc4_gradient_norm_quantiles": { + "0.1": 0.005175502217599003, + "0.25": 0.005175502217599003, + "0.5": 0.06471869166771999, + "0.75": 0.0752989516353021, + "0.9": 0.3389352911086298 + }, + "mc4_gradient_rmse": 0.1522759442940499, + "noise_signal_ratio": 2.259733052753072, + "optimal_choices": [ + "D" + ], + "p": [ + 0.04726231024957923, + 0.10005431510852038, + 0.11337640109434284, + 0.7393069735475576 + ], + "probability_all_four_same": 0.2990144508777924, + "probability_no_optimal_in_four": 0.004618677681973113, + "probability_opposed_gradient": 0.0, + "q": [ + 0.015606366446756566, + 0.13067050570258137, + 0.10176629055514091, + 0.7519568372955211 + ], + "reverse_kl": 0.025362868209147814, + "student_entropy_nats": 0.8447027915052953, + "student_max_mass": 0.7393069735475576, + "student_optimal_mass": 0.7393069735475576, + "student_predicted_action_mass": 0.7393069735475576, + "teacher_entropy_nats": 0.7777569900056194, + "teacher_mass_on_student_prediction": 0.7519568372955211, + "teacher_max_mass": 0.7519568372955211, + "teacher_optimal_mass": 0.7519568372955211, + "variance_verification_error": 3.122502256758253e-17 + }, + { + "current_model_mistake": false, + "enumerated_mean_logit_gradient": [ + -4.215915115427788e-06, + -0.00408012113278678, + 0.0029514525471887572, + 0.001132884500713453 + ], + "episode_id": "train-travel-2300110", + "exact_gradient_norm": 0.005161579764221154, + "exact_logit_gradient": [ + -4.2159151154277885e-06, + -0.004080121132786781, + 0.0029514525471887546, + 0.0011328845007134536 + ], + "expected_cosine": 0.8962033784664489, + "feedback_advantage": 0.0015207746697966096, + "id": "train-travel-2300110/frozen-warmstart-v2/seed-1707/step-0", + "kind": "focus", + "maximum_expectation_error": 2.6020852139652106e-18, + "mc4_error_norm_quantiles": { + "0.1": 0.005157053018147987, + "0.25": 0.005157053018147987, + "0.5": 0.005157053018147987, + "0.75": 0.005157053018147987, + "0.9": 0.005157053018147987 + }, + "mc4_gradient_norm_quantiles": { + "0.1": 5.0155533209747214e-06, + "0.25": 5.0155533209747214e-06, + "0.5": 5.0155533209747214e-06, + "0.75": 5.0155533209747214e-06, + "0.9": 5.0155533209747214e-06 + }, + "mc4_gradient_rmse": 0.07402011619458652, + "noise_signal_ratio": 14.34059330201122, + "optimal_choices": [ + "B" + ], + "p": [ + 0.001030353823811807, + 0.9971510780619579, + 0.0009092840571151346, + 0.0009092840571151346 + ], + "probability_all_four_same": 0.9886529179619944, + "probability_no_optimal_in_four": 6.587523811230645e-11, + "probability_opposed_gradient": 0.00409262867703174, + "q": [ + 0.0010319251722508418, + 0.9986718527317545, + 3.531054172758839e-05, + 0.0002609115542670526 + ], + "reverse_kl": 0.0025678204869940572, + "student_entropy_nats": 0.02266664726492304, + "student_max_mass": 0.9971510780619579, + "student_optimal_mass": 0.9971510780619579, + "student_predicted_action_mass": 0.9971510780619579, + "teacher_entropy_nats": 0.01093796907949321, + "teacher_mass_on_student_prediction": 0.9986718527317545, + "teacher_max_mass": 0.9986718527317545, + "teacher_optimal_mass": 0.9986718527317545, + "variance_verification_error": 1.1275702593849246e-17 + }, + { + "current_model_mistake": true, + "enumerated_mean_logit_gradient": [ + 0.13981884985675969, + -0.1366760257784998, + -4.3982799214865756e-05, + -0.003098841279044519 + ], + "episode_id": "train-travel-2300044", + "exact_gradient_norm": 0.19554859127498583, + "exact_logit_gradient": [ + 0.1398188498567591, + -0.13667602577849997, + -4.398279921486579e-05, + -0.0030988412790445338 + ], + "expected_cosine": 0.9921062584978286, + "feedback_advantage": 0.9315916123282266, + "id": "train-travel-2300044/frozen-warmstart-v2/seed-1707/step-2", + "kind": "exercise", + "maximum_expectation_error": 5.828670879282072e-16, + "mc4_error_norm_quantiles": { + "0.1": 0.10543532758438652, + "0.25": 0.10543532758438652, + "0.5": 0.10543532758438652, + "0.75": 0.10543532758438652, + "0.9": 0.10543532758438652 + }, + "mc4_gradient_norm_quantiles": { + "0.1": 0.09025469168198831, + "0.25": 0.09025469168198831, + "0.5": 0.09025469168198831, + "0.75": 0.09025469168198831, + "0.9": 0.09025469168198831 + }, + "mc4_gradient_rmse": 0.3700051362575641, + "noise_signal_ratio": 1.8921391038672974, + "optimal_choices": [ + "B" + ], + "p": [ + 0.9782165263684871, + 0.02030223602778511, + 1.0548527798231432e-05, + 0.0014706890759295266 + ], + "probability_all_four_same": 0.9156722721297913, + "probability_no_optimal_in_four": 0.9212308377416744, + "probability_opposed_gradient": 0.005519060389814809, + "q": [ + 0.047392003114413564, + 0.9518938483560118, + 3.813790735356167e-05, + 0.0006760106222211574 + ], + "reverse_kl": 2.884345124497957, + "student_entropy_nats": 0.11037552899846086, + "student_max_mass": 0.9782165263684871, + "student_optimal_mass": 0.02030223602778511, + "student_predicted_action_mass": 0.9782165263684871, + "teacher_entropy_nats": 0.19676498790698502, + "teacher_mass_on_student_prediction": 0.047392003114413564, + "teacher_max_mass": 0.9518938483560118, + "teacher_optimal_mass": 0.9518938483560118, + "variance_verification_error": 1.942890293094024e-16 + }, + { + "current_model_mistake": false, + "enumerated_mean_logit_gradient": [ + 0.31043611420495915, + -0.6440447123689974, + 0.0450576652868917, + 0.2885509328771471 + ], + "episode_id": "train-travel-2300028", + "exact_gradient_norm": 0.7723056432517614, + "exact_logit_gradient": [ + 0.31043611420495937, + -0.6440447123689979, + 0.04505766528689171, + 0.28855093287714684 + ], + "expected_cosine": 0.8332676239594775, + "feedback_advantage": 0.3527218747329721, + "id": "train-travel-2300028/frozen-warmstart-v2/seed-1707/step-3", + "kind": "repair", + "maximum_expectation_error": 4.440892098500626e-16, + "mc4_error_norm_quantiles": { + "0.1": 0.4212602574727744, + "0.25": 0.47779410836878333, + "0.5": 0.5757304701068372, + "0.75": 0.6965357884377604, + "0.9": 0.8137346186646748 + }, + "mc4_gradient_norm_quantiles": { + "0.1": 0.2305534570752647, + "0.25": 0.7147160720834825, + "0.5": 0.9605123966616761, + "0.75": 1.2026447502102329, + "0.9": 1.5440544025559917 + }, + "mc4_gradient_rmse": 0.6909541954322876, + "noise_signal_ratio": 0.8946641805219151, + "optimal_choices": [ + "B" + ], + "p": [ + 0.1889188438653946, + 0.581911027431488, + 0.11458506435155869, + 0.11458506435155869 + ], + "probability_all_four_same": 0.1162823846012764, + "probability_no_optimal_in_four": 0.0305544767911899, + "probability_opposed_gradient": 0.0, + "q": [ + 0.01939768726628952, + 0.9346329021644602, + 0.041064907075704066, + 0.004904503493546293 + ], + "reverse_kl": 0.6329389599937474, + "student_entropy_nats": 1.126372903990095, + "student_max_mass": 0.581911027431488, + "student_optimal_mass": 0.581911027431488, + "student_predicted_action_mass": 0.581911027431488, + "teacher_entropy_nats": 0.2968439572426678, + "teacher_mass_on_student_prediction": 0.9346329021644602, + "teacher_max_mass": 0.9346329021644602, + "teacher_optimal_mass": 0.9346329021644602, + "variance_verification_error": 6.661338147750939e-16 + }, + { + "current_model_mistake": true, + "enumerated_mean_logit_gradient": [ + -0.008676730915941884, + 0.08440407493376915, + 5.2898299634111204e-05, + -0.07578024231746149 + ], + "episode_id": "train-travel-2300032", + "exact_gradient_norm": 0.11376282982036047, + "exact_logit_gradient": [ + -0.008676730915941884, + 0.08440407493376936, + 5.289829963411141e-05, + -0.07578024231746146 + ], + "expected_cosine": 0.9819792685575933, + "feedback_advantage": 0.6558455845481231, + "id": "train-travel-2300032/frozen-warmstart-v2/seed-1707/step-0", + "kind": "focus", + "maximum_expectation_error": 2.0816681711721685e-16, + "mc4_error_norm_quantiles": { + "0.1": 0.0840522085809271, + "0.25": 0.0840522085809271, + "0.5": 0.0840522085809271, + "0.75": 0.0840522085809271, + "0.9": 0.0840522085809271 + }, + "mc4_gradient_norm_quantiles": { + "0.1": 0.030049004747121816, + "0.25": 0.030049004747121816, + "0.5": 0.030049004747121816, + "0.75": 0.030049004747121816, + "0.9": 0.030049004747121816 + }, + "mc4_gradient_rmse": 0.32582321242684803, + "noise_signal_ratio": 2.8640568535552946, + "optimal_choices": [ + "D" + ], + "p": [ + 0.004533800005929439, + 0.9790281025978073, + 0.0006135830378884609, + 0.015824514358374855 + ], + "probability_all_four_same": 0.9187146943676543, + "probability_no_optimal_in_four": 0.9381886460031981, + "probability_opposed_gradient": 0.002337314916586405, + "q": [ + 0.010856536535054445, + 0.31727452015436736, + 0.00019884440408021015, + 0.671670098906498 + ], + "reverse_kl": 1.040580847847811, + "student_entropy_nats": 0.1153654037839801, + "student_max_mass": 0.9790281025978073, + "student_optimal_mass": 0.015824514358374855, + "student_predicted_action_mass": 0.9790281025978073, + "teacher_entropy_nats": 0.6823426658620594, + "teacher_mass_on_student_prediction": 0.31727452015436736, + "teacher_max_mass": 0.671670098906498, + "teacher_optimal_mass": 0.671670098906498, + "variance_verification_error": 2.7755575615628914e-17 + }, + { + "current_model_mistake": false, + "enumerated_mean_logit_gradient": [ + 0.005419534573627144, + 0.047632733959448025, + -0.1801610280733966, + 0.12710875954032136 + ], + "episode_id": "train-travel-2300045", + "exact_gradient_norm": 0.22563882974299596, + "exact_logit_gradient": [ + 0.005419534573627141, + 0.047632733959448094, + -0.18016102807339657, + 0.12710875954032133 + ], + "expected_cosine": 0.9680385003361616, + "feedback_advantage": 0.0672901677736032, + "id": "train-travel-2300045/frozen-warmstart-v2/seed-1707/step-1", + "kind": "admin", + "maximum_expectation_error": 6.938893903907228e-17, + "mc4_error_norm_quantiles": { + "0.1": 0.21922881561982552, + "0.25": 0.21922881561982552, + "0.5": 0.21922881561982552, + "0.75": 0.451210937086542, + "0.9": 0.6383940967964189 + }, + "mc4_gradient_norm_quantiles": { + "0.1": 0.006413387786513846, + "0.25": 0.006413387786513846, + "0.5": 0.006413387786513846, + "0.75": 0.5320559772378849, + "0.9": 0.851447746257718 + }, + "mc4_gradient_rmse": 0.45622515126366936, + "noise_signal_ratio": 2.021926597400424, + "optimal_choices": [ + "C" + ], + "p": [ + 0.0037881916990436604, + 0.016977498786122855, + 0.9269399308040114, + 0.05229437871082216 + ], + "probability_all_four_same": 0.7382625067239711, + "probability_no_optimal_in_four": 2.8491828189016907e-05, + "probability_opposed_gradient": 0.0, + "q": [ + 0.0008000897896070585, + 0.0009066204682178887, + 0.9942300985776146, + 0.004063191164560443 + ], + "reverse_kl": 0.12428115714260658, + "student_entropy_nats": 0.31495789529329077, + "student_max_mass": 0.9269399308040114, + "student_optimal_mass": 0.9269399308040114, + "student_predicted_action_mass": 0.9269399308040114, + "teacher_entropy_nats": 0.04018114606982059, + "teacher_mass_on_student_prediction": 0.9942300985776146, + "teacher_max_mass": 0.9942300985776146, + "teacher_optimal_mass": 0.9942300985776146, + "variance_verification_error": 2.498001805406602e-16 + }, + { + "current_model_mistake": true, + "enumerated_mean_logit_gradient": [ + 0.700497003838679, + -0.050950085066363435, + -0.1131822916345267, + -0.5363646271377885 + ], + "episode_id": "train-travel-2300010", + "exact_gradient_norm": 0.8909484878148319, + "exact_logit_gradient": [ + 0.7004970038386782, + -0.05095008506636353, + -0.11318229163452678, + -0.5363646271377877 + ], + "expected_cosine": 0.9891436710880799, + "feedback_advantage": 0.7349780776709811, + "id": "train-travel-2300010/frozen-warmstart-v2/seed-1707/step-3", + "kind": "repair", + "maximum_expectation_error": 8.881784197001252e-16, + "mc4_error_norm_quantiles": { + "0.1": 0.1815048224055496, + "0.25": 0.1815048224055496, + "0.5": 0.20149932131944773, + "0.75": 0.2302127914555008, + "0.9": 0.3275143714263734 + }, + "mc4_gradient_norm_quantiles": { + "0.1": 0.716613339507799, + "0.25": 0.7168281496529835, + "0.5": 0.7168281496529835, + "0.75": 1.101689555902164, + "0.9": 1.101689555902164 + }, + "mc4_gradient_rmse": 0.2897003264839472, + "noise_signal_ratio": 0.32515945696756865, + "optimal_choices": [ + "D" + ], + "p": [ + 0.8191285179134189, + 0.019264055523634267, + 0.03598999624693764, + 0.12561743031600917 + ], + "probability_all_four_same": 0.45045360816389485, + "probability_no_optimal_in_four": 0.5845288707316668, + "probability_opposed_gradient": 0.0, + "q": [ + 0.033368909547031964, + 0.02598773259980377, + 0.08004784986617407, + 0.8605955079869902 + ], + "reverse_kl": 2.345442874769698, + "student_entropy_nats": 0.6197559114864926, + "student_max_mass": 0.8191285179134189, + "student_optimal_mass": 0.12561743031600917, + "student_predicted_action_mass": 0.8191285179134189, + "teacher_entropy_nats": 0.5396503429712639, + "teacher_mass_on_student_prediction": 0.033368909547031964, + "teacher_max_mass": 0.8605955079869902, + "teacher_optimal_mass": 0.8605955079869902, + "variance_verification_error": 1.1102230246251565e-16 + }, + { + "current_model_mistake": false, + "enumerated_mean_logit_gradient": [ + -1.0686076489892327, + 0.9559828082869198, + -0.045299820580092555, + 0.15792466128240706 + ], + "episode_id": "train-travel-2300027", + "exact_gradient_norm": 1.4431970446214863, + "exact_logit_gradient": [ + -1.0686076489892335, + 0.955982808286919, + -0.04529982058009252, + 0.15792466128240715 + ], + "expected_cosine": 0.9141354523811375, + "feedback_advantage": 0.4983248460311165, + "id": "train-travel-2300027/frozen-warmstart-v2/seed-1707/step-2", + "kind": "exercise", + "maximum_expectation_error": 8.881784197001252e-16, + "mc4_error_norm_quantiles": { + "0.1": 0.4088591331503421, + "0.25": 0.44757751679847524, + "0.5": 0.6255511987187788, + "0.75": 1.0041633093412803, + "0.9": 1.347100773764676 + }, + "mc4_gradient_norm_quantiles": { + "0.1": 0.9820917915595783, + "0.25": 1.092903306574678, + "0.5": 1.7083176490454546, + "0.75": 1.7882630912607753, + "0.9": 2.327470562232574 + }, + "mc4_gradient_rmse": 0.844175293939636, + "noise_signal_ratio": 0.5849341897461005, + "optimal_choices": [ + "A" + ], + "p": [ + 0.46527094342218256, + 0.41060016171038394, + 0.04327693646068776, + 0.08085195840674576 + ], + "probability_all_four_same": 0.07533198275765679, + "probability_no_optimal_in_four": 0.0817589179823143, + "probability_opposed_gradient": 0.0, + "q": [ + 0.963595789453299, + 0.008336737179732039, + 0.025678957987252442, + 0.00238851537971649 + ], + "reverse_kl": 1.5686907220383088, + "student_entropy_nats": 1.0607340561901464, + "student_max_mass": 0.46527094342218256, + "student_optimal_mass": 0.46527094342218256, + "student_predicted_action_mass": 0.46527094342218256, + "teacher_entropy_nats": 0.18410019410616654, + "teacher_mass_on_student_prediction": 0.963595789453299, + "teacher_max_mass": 0.963595789453299, + "teacher_optimal_mass": 0.963595789453299, + "variance_verification_error": 1.4432899320127035e-15 + }, + { + "current_model_mistake": true, + "enumerated_mean_logit_gradient": [ + 0.46403621399668976, + -0.09853201731927239, + -0.003108209191345184, + -0.3623959874860716 + ], + "episode_id": "train-travel-2300051", + "exact_gradient_norm": 0.5969746050275245, + "exact_logit_gradient": [ + 0.4640362139966893, + -0.09853201731927226, + -0.0031082091913451843, + -0.3623959874860719 + ], + "expected_cosine": 0.979150806528189, + "feedback_advantage": 0.48533119590495044, + "id": "train-travel-2300051/frozen-warmstart-v2/seed-1707/step-3", + "kind": "repair", + "maximum_expectation_error": 4.440892098500626e-16, + "mc4_error_norm_quantiles": { + "0.1": 0.05075921979375016, + "0.25": 0.12656949306505422, + "0.5": 0.14685960950364083, + "0.75": 0.21386239716374153, + "0.9": 0.3499800571012793 + }, + "mc4_gradient_norm_quantiles": { + "0.1": 0.4270404281392167, + "0.25": 0.4519068977770228, + "0.5": 0.5904074727892141, + "0.75": 0.7220872404874594, + "0.9": 0.9150159943166223 + }, + "mc4_gradient_rmse": 0.21998764508300117, + "noise_signal_ratio": 0.368504193026533, + "optimal_choices": [ + "D" + ], + "p": [ + 0.6586371426094243, + 0.14696181627991345, + 0.0056983282195416885, + 0.18870271289112053 + ], + "probability_all_four_same": 0.189919389037054, + "probability_no_optimal_in_four": 0.4332315684264591, + "probability_opposed_gradient": 7.666470095600251e-05, + "q": [ + 0.17042247183082856, + 0.15039730709332738, + 0.005146312279773139, + 0.674033908796071 + ], + "reverse_kl": 0.6473522796776567, + "student_entropy_nats": 0.9009706711719669, + "student_max_mass": 0.6586371426094243, + "student_optimal_mass": 0.18870271289112053, + "student_predicted_action_mass": 0.6586371426094243, + "teacher_entropy_nats": 0.8794899675904098, + "teacher_mass_on_student_prediction": 0.17042247183082856, + "teacher_max_mass": 0.674033908796071, + "teacher_optimal_mass": 0.674033908796071, + "variance_verification_error": 1.1102230246251565e-16 + }, + { + "current_model_mistake": false, + "enumerated_mean_logit_gradient": [ + 0.00010533713929276477, + 0.3353096412428237, + -0.3357135169803493, + 0.00029853859823286585 + ], + "episode_id": "train-travel-2300038", + "exact_gradient_norm": 0.4744852170667857, + "exact_logit_gradient": [ + 0.00010533713929276482, + 0.33530964124282425, + -0.3357135169803499, + 0.00029853859823286574 + ], + "expected_cosine": 0.9992379173410535, + "feedback_advantage": 0.15968340960752414, + "id": "train-travel-2300038/frozen-warmstart-v2/seed-1707/step-0", + "kind": "focus", + "maximum_expectation_error": 6.106226635438361e-16, + "mc4_error_norm_quantiles": { + "0.1": 0.15873324193483854, + "0.25": 0.15873324193483854, + "0.5": 0.4283862264126771, + "0.75": 0.4283862264126771, + "0.9": 0.745850616198395 + }, + "mc4_gradient_norm_quantiles": { + "0.1": 0.046099045460327795, + "0.25": 0.046099045460327795, + "0.5": 0.6332170128846428, + "0.75": 0.6332170128846428, + "0.9": 1.2203351446607107 + }, + "mc4_gradient_rmse": 0.4535961718429664, + "noise_signal_ratio": 0.9559753508171381, + "optimal_choices": [ + "C" + ], + "p": [ + 0.00031060204338678835, + 0.18231886753557908, + 0.8170964251081487, + 0.00027410531288540234 + ], + "probability_all_four_same": 0.44685684852709434, + "probability_no_optimal_in_four": 0.0011191512285012087, + "probability_opposed_gradient": 0.0, + "q": [ + 0.0001753907776464064, + 0.022971660740172612, + 0.9767798347156729, + 7.311376650813972e-05 + ], + "reverse_kl": 0.23235739916041798, + "student_entropy_nats": 0.480115291222093, + "student_max_mass": 0.8170964251081487, + "student_optimal_mass": 0.8170964251081487, + "student_predicted_action_mass": 0.8170964251081487, + "teacher_entropy_nats": 0.11184505381548003, + "teacher_mass_on_student_prediction": 0.9767798347156729, + "teacher_max_mass": 0.9767798347156729, + "teacher_optimal_mass": 0.9767798347156729, + "variance_verification_error": 3.0531133177191805e-16 + }, + { + "current_model_mistake": false, + "enumerated_mean_logit_gradient": [ + 0.00022451333523340223, + -0.002999340968475197, + 0.002399633832018009, + 0.00037519380122379197 + ], + "episode_id": "train-travel-2300091", + "exact_gradient_norm": 0.0038659365487697258, + "exact_logit_gradient": [ + 0.00022451333523340212, + -0.0029993409684752025, + 0.002399633832018008, + 0.0003751938012237924 + ], + "expected_cosine": 0.9737988324048764, + "feedback_advantage": 0.0004254325895903799, + "id": "train-travel-2300091/frozen-warmstart-v2/seed-1707/step-0", + "kind": "focus", + "maximum_expectation_error": 5.637851296924623e-18, + "mc4_error_norm_quantiles": { + "0.1": 0.0038592038828315282, + "0.25": 0.0038592038828315282, + "0.5": 0.0038592038828315282, + "0.75": 0.0038592038828315282, + "0.9": 0.0038592038828315282 + }, + "mc4_gradient_norm_quantiles": { + "0.1": 6.9014720586829e-06, + "0.25": 6.9014720586829e-06, + "0.5": 6.9014720586829e-06, + "0.75": 6.9014720586829e-06, + "0.9": 6.9014720586829e-06 + }, + "mc4_gradient_rmse": 0.041542548716104986, + "noise_signal_ratio": 10.745791657994298, + "optimal_choices": [ + "B", + "C" + ], + "p": [ + 0.0009089933777143577, + 0.9968322676359559, + 0.001924339015778721, + 0.00033439997055102267 + ], + "probability_all_four_same": 0.9873891506821041, + "probability_no_optimal_in_four": 2.390199541964891e-12, + "probability_opposed_gradient": 0.0, + "q": [ + 0.0007092012259865353, + 0.9986297127874771, + 0.0005523264538478326, + 0.00010875953268853427 + ], + "reverse_kl": 0.0012073389361175475, + "student_entropy_nats": 0.024238032455796716, + "student_max_mass": 0.9968322676359559, + "student_optimal_mass": 0.9987566066517346, + "student_predicted_action_mass": 0.9968322676359559, + "teacher_entropy_nats": 0.011647814995332833, + "teacher_mass_on_student_prediction": 0.9986297127874771, + "teacher_max_mass": 0.9986297127874771, + "teacher_optimal_mass": 0.999182039241325, + "variance_verification_error": 0.0 + }, + { + "current_model_mistake": true, + "enumerated_mean_logit_gradient": [ + 0.40000152350986157, + 0.015331286478717556, + 0.0015488179184504614, + -0.4168816279070315 + ], + "episode_id": "train-travel-2300038", + "exact_gradient_norm": 0.5779523835738225, + "exact_logit_gradient": [ + 0.4000015235098619, + 0.015331286478717463, + 0.0015488179184504616, + -0.4168816279070298 + ], + "expected_cosine": 0.7474851895614436, + "feedback_advantage": 0.7971835544659945, + "id": "train-travel-2300038/frozen-warmstart-v2/seed-1707/step-1", + "kind": "admin", + "maximum_expectation_error": 1.6653345369377348e-15, + "mc4_error_norm_quantiles": { + "0.1": 0.2534470129671211, + "0.25": 0.2534470129671211, + "0.5": 0.6058387705675374, + "0.75": 0.6280328748907031, + "0.9": 0.8995687758041782 + }, + "mc4_gradient_norm_quantiles": { + "0.1": 0.32526605067454817, + "0.25": 0.32526605067454817, + "0.5": 0.8093365209594187, + "0.75": 0.9783066787750341, + "0.9": 1.157461507375295 + }, + "mc4_gradient_rmse": 0.6454882599793311, + "noise_signal_ratio": 1.1168537033931658, + "optimal_choices": [ + "D" + ], + "p": [ + 0.6929802573511676, + 0.19854216847804107, + 0.0022056043523095395, + 0.1062719698184817 + ], + "probability_all_four_same": 0.23229422609740713, + "probability_no_optimal_in_four": 0.6380012301116318, + "probability_opposed_gradient": 0.14027784193617954, + "q": [ + 0.06544610318997536, + 0.030914547895359017, + 0.00018382463018928684, + 0.9034555242844763 + ], + "reverse_kl": 1.7825553384263513, + "student_entropy_nats": 0.8268736417592039, + "student_max_mass": 0.6929802573511676, + "student_optimal_mass": 0.1062719698184817, + "student_predicted_action_mass": 0.6929802573511676, + "teacher_entropy_nats": 0.3792235211619587, + "teacher_mass_on_student_prediction": 0.06544610318997536, + "teacher_max_mass": 0.9034555242844763, + "teacher_optimal_mass": 0.9034555242844763, + "variance_verification_error": 4.440892098500626e-16 + }, + { + "current_model_mistake": false, + "enumerated_mean_logit_gradient": [ + -0.00033243768450550375, + -1.568733478541877, + -8.627118802235748e-06, + 1.5690745433451836 + ], + "episode_id": "train-travel-2300060", + "exact_gradient_norm": 2.21876536836621, + "exact_logit_gradient": [ + -0.0003324376845055036, + -1.5687334785418776, + -8.627118802235755e-06, + 1.5690745433451856 + ], + "expected_cosine": 0.9999995195428689, + "feedback_advantage": 0.40623177842685, + "id": "train-travel-2300060/frozen-warmstart-v2/seed-1707/step-2", + "kind": "exercise", + "maximum_expectation_error": 1.9984014443252818e-15, + "mc4_error_norm_quantiles": { + "0.1": 0.4367693339436593, + "0.25": 0.4367693339436593, + "0.5": 0.7406172865870118, + "0.75": 1.6141559544073985, + "0.9": 1.9180039070613544 + }, + "mc4_gradient_norm_quantiles": { + "0.1": 0.30076150776280064, + "0.25": 1.4781480857975862, + "0.5": 2.655534701474559, + "0.75": 2.655534701474559, + "0.9": 3.8329213201052372 + }, + "mc4_gradient_rmse": 1.1569823100514236, + "noise_signal_ratio": 0.521453203906535, + "optimal_choices": [ + "B" + ], + "p": [ + 0.00015481470976223455, + 0.5925727996822291, + 3.4203025693628215e-06, + 0.40726896530543927 + ], + "probability_all_four_same": 0.15081326817974852, + "probability_no_optimal_in_four": 0.02755497863097846, + "probability_opposed_gradient": 6.269174247345769e-16, + "q": [ + 0.00015827189922174178, + 0.9988045781090791, + 5.0876420115380415e-06, + 0.00103206234968764 + ], + "reverse_kl": 2.125240759272016, + "student_entropy_nats": 0.6773258524276866, + "student_max_mass": 0.5925727996822291, + "student_optimal_mass": 0.5925727996822291, + "student_predicted_action_mass": 0.5925727996822291, + "teacher_entropy_nats": 0.009738450446801812, + "teacher_mass_on_student_prediction": 0.9988045781090791, + "teacher_max_mass": 0.9988045781090791, + "teacher_optimal_mass": 0.9988045781090791, + "variance_verification_error": 6.661338147750939e-16 + }, + { + "current_model_mistake": false, + "enumerated_mean_logit_gradient": [ + 0.02505366699039375, + -0.005420040851768691, + -0.565052910691787, + 0.545419284553162 + ], + "episode_id": "train-travel-2300056", + "exact_gradient_norm": 0.7857633555443042, + "exact_logit_gradient": [ + 0.025053666990393835, + -0.0054200408517686895, + -0.5650529106917873, + 0.5454192845531621 + ], + "expected_cosine": 0.9055851240851541, + "feedback_advantage": 0.46752207819647273, + "id": "train-travel-2300056/frozen-warmstart-v2/seed-1707/step-3", + "kind": "repair", + "maximum_expectation_error": 3.3306690738754696e-16, + "mc4_error_norm_quantiles": { + "0.1": 0.03532194571140697, + "0.25": 0.15775143953208784, + "0.5": 0.34502088334314274, + "0.75": 0.5168509803700074, + "0.9": 0.7042301999020986 + }, + "mc4_gradient_norm_quantiles": { + "0.1": 0.5741831231513559, + "0.25": 0.6318105973524771, + "0.5": 0.7723918209569453, + "0.75": 1.0821153948802649, + "0.9": 1.203928868685899 + }, + "mc4_gradient_rmse": 0.42572344822372366, + "noise_signal_ratio": 0.5417960066728001, + "optimal_choices": [ + "C" + ], + "p": [ + 0.2544052090782056, + 0.005280011359079725, + 0.37015738978135737, + 0.37015738978135737 + ], + "probability_all_four_same": 0.041735978231216185, + "probability_no_optimal_in_four": 0.15737224961049256, + "probability_opposed_gradient": 0.0, + "q": [ + 0.11336758288645755, + 0.007247347111394064, + 0.8376794679778301, + 0.04170560202431833 + ], + "reverse_kl": 0.7098134430200633, + "student_entropy_nats": 1.1116689842999914, + "student_max_mass": 0.37015738978135737, + "student_optimal_mass": 0.37015738978135737, + "student_predicted_action_mass": 0.37015738978135737, + "teacher_entropy_nats": 0.5633966271575032, + "teacher_mass_on_student_prediction": 0.8376794679778301, + "teacher_max_mass": 0.8376794679778301, + "teacher_optimal_mass": 0.8376794679778301, + "variance_verification_error": 5.551115123125783e-17 + }, + { + "current_model_mistake": false, + "enumerated_mean_logit_gradient": [ + -0.03829014569422298, + 0.02739567778624929, + 0.001487590569327077, + 0.0094068773386466 + ], + "episode_id": "train-travel-2300081", + "exact_gradient_norm": 0.04803499438554249, + "exact_logit_gradient": [ + -0.03829014569422289, + 0.027395677786249215, + 0.0014875905693270762, + 0.009406877338646591 + ], + "expected_cosine": 0.9909534295200938, + "feedback_advantage": 0.013541860890715962, + "id": "train-travel-2300081/frozen-warmstart-v2/seed-1707/step-0", + "kind": "focus", + "maximum_expectation_error": 9.020562075079397e-17, + "mc4_error_norm_quantiles": { + "0.1": 0.047788601790218946, + "0.25": 0.047788601790218946, + "0.5": 0.047788601790218946, + "0.75": 0.047788601790218946, + "0.9": 0.047788601790218946 + }, + "mc4_gradient_norm_quantiles": { + "0.1": 0.0002467555804595114, + "0.25": 0.0002467555804595114, + "0.5": 0.0002467555804595114, + "0.75": 0.0002467555804595114, + "0.9": 0.0002467555804595114 + }, + "mc4_gradient_rmse": 0.22390608651600888, + "noise_signal_ratio": 4.6613118077807005, + "optimal_choices": [ + "A" + ], + "p": [ + 0.9852957187565402, + 0.009659499333903055, + 0.0010181037386075298, + 0.004026678170949214 + ], + "probability_all_four_same": 0.9424675088699196, + "probability_no_optimal_in_four": 4.6749309742229706e-08, + "probability_opposed_gradient": 0.0, + "q": [ + 0.9988375796472562, + 0.0005524414203364405, + 0.00023029184213593524, + 0.0003796870902714833 + ], + "reverse_kl": 0.0252112129632341, + "student_entropy_nats": 0.08863483857518957, + "student_max_mass": 0.9852957187565402, + "student_optimal_mass": 0.9852957187565402, + "student_predicted_action_mass": 0.9852957187565402, + "teacher_entropy_nats": 0.010225137202818252, + "teacher_mass_on_student_prediction": 0.9988375796472562, + "teacher_max_mass": 0.9988375796472562, + "teacher_optimal_mass": 0.9988375796472562, + "variance_verification_error": 1.5959455978986625e-16 + }, + { + "current_model_mistake": true, + "enumerated_mean_logit_gradient": [ + 0.5030163766320608, + -0.7368949274386757, + -0.2200288688092202, + 0.45390741961583386 + ], + "episode_id": "train-travel-2300081", + "exact_gradient_norm": 1.0249313430355562, + "exact_logit_gradient": [ + 0.5030163766320614, + -0.7368949274386757, + -0.22002886880922015, + 0.4539074196158343 + ], + "expected_cosine": 0.7903900150108764, + "feedback_advantage": 0.57231821980171, + "id": "train-travel-2300081/frozen-warmstart-v2/seed-1707/step-3", + "kind": "repair", + "maximum_expectation_error": 6.661338147750939e-16, + "mc4_error_norm_quantiles": { + "0.1": 0.26129569322770674, + "0.25": 0.44900476470328626, + "0.5": 0.7729615702444427, + "0.75": 1.0093123577103884, + "0.9": 1.4663867877895356 + }, + "mc4_gradient_norm_quantiles": { + "0.1": 0.7672994112949866, + "0.25": 0.9789148295666507, + "0.5": 1.2386099473306067, + "0.75": 1.6511020107515255, + "0.9": 1.8500765711459184 + }, + "mc4_gradient_rmse": 0.9384926324830757, + "noise_signal_ratio": 0.9156639016458668, + "optimal_choices": [ + "B" + ], + "p": [ + 0.31483005069955955, + 0.27783654878274133, + 0.19095408062842129, + 0.21637931988927783 + ], + "probability_all_four_same": 0.019304849556107288, + "probability_no_optimal_in_four": 0.27198316283781837, + "probability_opposed_gradient": 0.0013295839814363272, + "q": [ + 0.013741471983846842, + 0.8501547685844514, + 0.13037546149013235, + 0.005728297941569398 + ], + "reverse_kl": 1.533875158268891, + "student_entropy_nats": 1.3670711533275415, + "student_max_mass": 0.31483005069955955, + "student_optimal_mass": 0.27783654878274133, + "student_predicted_action_mass": 0.31483005069955955, + "teacher_entropy_nats": 0.49211591202671257, + "teacher_mass_on_student_prediction": 0.013741471983846842, + "teacher_max_mass": 0.8501547685844514, + "teacher_optimal_mass": 0.8501547685844514, + "variance_verification_error": 3.3306690738754696e-16 + }, + { + "current_model_mistake": false, + "enumerated_mean_logit_gradient": [ + -0.3984055290973702, + 0.1906964621650697, + -1.045398242985208e-05, + 0.20771952091473017 + ], + "episode_id": "train-travel-2300077", + "exact_gradient_norm": 0.48809784446969207, + "exact_logit_gradient": [ + -0.39840552909736976, + 0.19069646216506939, + -1.0453982429852056e-05, + 0.20771952091473025 + ], + "expected_cosine": 0.8902851847283257, + "feedback_advantage": 0.18331094687457883, + "id": "train-travel-2300077/frozen-warmstart-v2/seed-1707/step-3", + "kind": "repair", + "maximum_expectation_error": 4.440892098500626e-16, + "mc4_error_norm_quantiles": { + "0.1": 0.2806267641890656, + "0.25": 0.2806267641890656, + "0.5": 0.43225207902759316, + "0.75": 0.5838686491983931, + "0.9": 0.8619561200911541 + }, + "mc4_gradient_norm_quantiles": { + "0.1": 0.058007152881202624, + "0.25": 0.058007152881202624, + "0.5": 0.4847659642028969, + "0.75": 0.9148076016823031, + "0.9": 1.31464382171635 + }, + "mc4_gradient_rmse": 0.6044601656692655, + "noise_signal_ratio": 1.2383995801620444, + "optimal_choices": [ + "A" + ], + "p": [ + 0.7810526309699114, + 0.15379837973577198, + 0.0010362853614592936, + 0.06411270393285734 + ], + "probability_all_four_same": 0.37272912559021804, + "probability_no_optimal_in_four": 0.002298047089051204, + "probability_opposed_gradient": 0.0, + "q": [ + 0.9643635778444902, + 0.032998707243412154, + 0.0007760552130394392, + 0.0018616596990581977 + ], + "reverse_kl": 0.29926209979397606, + "student_entropy_nats": 0.6641822620709803, + "student_max_mass": 0.7810526309699114, + "student_optimal_mass": 0.7810526309699114, + "student_predicted_action_mass": 0.7810526309699114, + "teacher_entropy_nats": 0.16482230320418492, + "teacher_mass_on_student_prediction": 0.9643635778444902, + "teacher_max_mass": 0.9643635778444902, + "teacher_optimal_mass": 0.9643635778444902, + "variance_verification_error": 6.661338147750939e-16 + }, + { + "current_model_mistake": false, + "enumerated_mean_logit_gradient": [ + -0.00019386649950737454, + -0.0002125657701145624, + 0.0056189695917799945, + -0.0052125373221580595 + ], + "episode_id": "train-travel-2300083", + "exact_gradient_norm": 0.007669819622031188, + "exact_logit_gradient": [ + -0.00019386649950737457, + -0.0002125657701145624, + 0.005618969591779998, + -0.00521253732215806 + ], + "expected_cosine": 0.9950073855655773, + "feedback_advantage": 0.0035047437301277684, + "id": "train-travel-2300083/frozen-warmstart-v2/seed-1707/step-2", + "kind": "exercise", + "maximum_expectation_error": 3.469446951953614e-18, + "mc4_error_norm_quantiles": { + "0.1": 0.007552497956018137, + "0.25": 0.007552497956018137, + "0.5": 0.007552497956018137, + "0.75": 0.007552497956018137, + "0.9": 0.007552497956018137 + }, + "mc4_gradient_norm_quantiles": { + "0.1": 0.00011760129272438289, + "0.25": 0.00011760129272438289, + "0.5": 0.00011760129272438289, + "0.75": 0.00011760129272438289, + "0.9": 0.00011760129272438289 + }, + "mc4_gradient_rmse": 0.031415963773930584, + "noise_signal_ratio": 4.096049884105454, + "optimal_choices": [ + "D" + ], + "p": [ + 6.450739312315921e-05, + 0.0004206410478587013, + 0.022966223669364745, + 0.9765486278896534 + ], + "probability_all_four_same": 0.909443303436771, + "probability_no_optimal_in_four": 3.024635403424057e-07, + "probability_opposed_gradient": 0.001826008184830036, + "q": [ + 0.0013003156139535915, + 0.0006960087966845013, + 0.0179503039695808, + 0.9800533716197811 + ], + "reverse_kl": 0.0017552302450786577, + "student_entropy_nats": 0.11373492106468482, + "student_max_mass": 0.9765486278896534, + "student_optimal_mass": 0.9765486278896534, + "student_predicted_action_mass": 0.9765486278896534, + "teacher_entropy_nats": 0.10561011826577972, + "teacher_mass_on_student_prediction": 0.9800533716197811, + "teacher_max_mass": 0.9800533716197811, + "teacher_optimal_mass": 0.9800533716197811, + "variance_verification_error": 0.0 + }, + { + "current_model_mistake": true, + "enumerated_mean_logit_gradient": [ + 0.3269675182573094, + 0.22400093118703532, + -0.6009498130694042, + 0.0499813636250591 + ], + "episode_id": "train-travel-2300039", + "exact_gradient_norm": 0.7216113841299718, + "exact_logit_gradient": [ + 0.32696751825730863, + 0.22400093118703562, + -0.6009498130694033, + 0.04998136362505912 + ], + "expected_cosine": 0.7296595664326799, + "feedback_advantage": 0.7306954498827651, + "id": "train-travel-2300039/frozen-warmstart-v2/seed-1707/step-3", + "kind": "repair", + "maximum_expectation_error": 8.881784197001252e-16, + "mc4_error_norm_quantiles": { + "0.1": 0.22874276858248022, + "0.25": 0.24981569608504983, + "0.5": 0.7481165906949885, + "0.75": 0.8912958645945398, + "0.9": 1.1222599416240198 + }, + "mc4_gradient_norm_quantiles": { + "0.1": 0.5871951205316214, + "0.25": 0.8387737181084953, + "0.5": 1.0798975764789622, + "0.75": 1.1314183591831208, + "0.9": 1.3495574247539817 + }, + "mc4_gradient_rmse": 0.805805662590793, + "noise_signal_ratio": 1.1166753744639604, + "optimal_choices": [ + "C" + ], + "p": [ + 0.5808061935722597, + 0.1885601722130173, + 0.1885601722130173, + 0.04207346200170568 + ], + "probability_all_four_same": 0.11632690606601859, + "probability_no_optimal_in_four": 0.43353611473261044, + "probability_opposed_gradient": 0.00023233527714717795, + "q": [ + 0.06659065675364269, + 0.011571721091189455, + 0.9192556220957825, + 0.0025820000593853227 + ], + "reverse_kl": 1.6028982488084647, + "student_entropy_nats": 1.0780413576172052, + "student_max_mass": 0.5808061935722597, + "student_optimal_mass": 0.1885601722130173, + "student_predicted_action_mass": 0.5808061935722597, + "teacher_entropy_nats": 0.32478704335659664, + "teacher_mass_on_student_prediction": 0.06659065675364269, + "teacher_max_mass": 0.9192556220957825, + "teacher_optimal_mass": 0.9192556220957825, + "variance_verification_error": 2.220446049250313e-16 + }, + { + "current_model_mistake": false, + "enumerated_mean_logit_gradient": [ + -2.8715894581312048e-05, + -0.00014788199172838321, + -4.052663155811548e-05, + 0.00021712451786781077 + ], + "episode_id": "train-travel-2300063", + "exact_gradient_norm": 0.0002673558493916584, + "exact_logit_gradient": [ + -2.871589458131207e-05, + -0.0001478819917283832, + -4.052663155811548e-05, + 0.00021712451786781072 + ], + "expected_cosine": 0.9993241968873857, + "feedback_advantage": -0.0014975724040160454, + "id": "train-travel-2300063/frozen-warmstart-v2/seed-1707/step-2", + "kind": "exercise", + "maximum_expectation_error": 5.421010862427522e-20, + "mc4_error_norm_quantiles": { + "0.1": 0.0002672197063511149, + "0.25": 0.0002672197063511149, + "0.5": 0.0002672197063511149, + "0.75": 0.0002672197063511149, + "0.9": 0.0002672197063511149 + }, + "mc4_gradient_norm_quantiles": { + "0.1": 1.3623031963191952e-07, + "0.25": 1.3623031963191952e-07, + "0.5": 1.3623031963191952e-07, + "0.75": 1.3623031963191952e-07, + "0.9": 1.3623031963191952e-07 + }, + "mc4_gradient_rmse": 0.018087986710731974, + "noise_signal_ratio": 67.65509994222825, + "optimal_choices": [ + "D" + ], + "p": [ + 6.961795324479202e-06, + 5.14410994815389e-05, + 1.4738120505016856e-05, + 0.9999268589846889 + ], + "probability_all_four_same": 0.9997074680348393, + "probability_no_optimal_in_four": 2.8618307045447777e-17, + "probability_opposed_gradient": 0.0, + "q": [ + 0.00043006599132489043, + 0.0009104497079421597, + 0.0002301977200600669, + 0.9984292865806729 + ], + "reverse_kl": 0.0012816641929145598, + "student_entropy_nats": 0.0008277574565303808, + "student_max_mass": 0.9999268589846889, + "student_optimal_mass": 0.9999268589846889, + "student_predicted_action_mass": 0.9999268589846889, + "teacher_entropy_nats": 0.013206013509351931, + "teacher_mass_on_student_prediction": 0.9984292865806729, + "teacher_max_mass": 0.9984292865806729, + "teacher_optimal_mass": 0.9984292865806729, + "variance_verification_error": 0.0 + }, + { + "current_model_mistake": false, + "enumerated_mean_logit_gradient": [ + -2.662079242099389e-05, + 0.001090144592037373, + -0.006150987828940298, + 0.005087464029323917 + ], + "episode_id": "train-travel-2300039", + "exact_gradient_norm": 0.008056430066696658, + "exact_logit_gradient": [ + -2.662079242099387e-05, + 0.0010901445920373728, + -0.0061509878289402985, + 0.00508746402932392 + ], + "expected_cosine": 0.9885358465493822, + "feedback_advantage": 0.002103304081625623, + "id": "train-travel-2300039/frozen-warmstart-v2/seed-1707/step-0", + "kind": "focus", + "maximum_expectation_error": 2.6020852139652106e-18, + "mc4_error_norm_quantiles": { + "0.1": 0.008049928358597207, + "0.25": 0.008049928358597207, + "0.5": 0.008049928358597207, + "0.75": 0.008049928358597207, + "0.9": 0.008049928358597207 + }, + "mc4_gradient_norm_quantiles": { + "0.1": 6.562912554916476e-06, + "0.25": 6.562912554916476e-06, + "0.5": 6.562912554916476e-06, + "0.75": 6.562912554916476e-06, + "0.9": 6.562912554916476e-06 + }, + "mc4_gradient_rmse": 0.09241910232296072, + "noise_signal_ratio": 11.471470807522929, + "optimal_choices": [ + "C" + ], + "p": [ + 0.00020295412201617725, + 0.0006251426559817868, + 0.9974725889940388, + 0.0016993142279632704 + ], + "probability_all_four_same": 0.9899286182853619, + "probability_no_optimal_in_four": 4.080407051513977e-11, + "probability_opposed_gradient": 0.0008059225782862981, + "q": [ + 0.0002304620820370265, + 0.00010886258462035443, + 0.9995758930756644, + 8.478225767823525e-05 + ], + "reverse_kl": 0.004060159876349124, + "student_entropy_nats": 0.019699275494913807, + "student_max_mass": 0.9974725889940388, + "student_optimal_mass": 0.9974725889940388, + "student_predicted_action_mass": 0.9974725889940388, + "teacher_entropy_nats": 0.004142521575924915, + "teacher_mass_on_student_prediction": 0.9995758930756644, + "teacher_max_mass": 0.9995758930756644, + "teacher_optimal_mass": 0.9995758930756644, + "variance_verification_error": 8.673617379884035e-18 + }, + { + "current_model_mistake": false, + "enumerated_mean_logit_gradient": [ + -2.609945394100268e-05, + -0.0006048395963651008, + -0.00014404503171963786, + 0.0007749840820257416 + ], + "episode_id": "train-travel-2300022", + "exact_gradient_norm": 0.0009939121779028114, + "exact_logit_gradient": [ + -2.6099453941002675e-05, + -0.0006048395963651009, + -0.00014404503171963788, + 0.0007749840820257417 + ], + "expected_cosine": 0.9943944486603719, + "feedback_advantage": -0.0036952708053559213, + "id": "train-travel-2300022/frozen-warmstart-v2/seed-1707/step-2", + "kind": "exercise", + "maximum_expectation_error": 1.0842021724855044e-19, + "mc4_error_norm_quantiles": { + "0.1": 0.0009923798728331136, + "0.25": 0.0009923798728331136, + "0.5": 0.0009923798728331136, + "0.75": 0.0009923798728331136, + "0.9": 0.0009923798728331136 + }, + "mc4_gradient_norm_quantiles": { + "0.1": 1.5407185933345832e-06, + "0.25": 1.5407185933345832e-06, + "0.5": 1.5407185933345832e-06, + "0.75": 1.5407185933345832e-06, + "0.9": 1.5407185933345832e-06 + }, + "mc4_gradient_rmse": 0.03088723326529668, + "noise_signal_ratio": 31.076420987687058, + "optimal_choices": [ + "D" + ], + "p": [ + 6.142168381280024e-06, + 0.00023048314511737715, + 9.607967591360751e-05, + 0.9996672950105877 + ], + "probability_all_four_same": 0.9986698440507139, + "probability_no_optimal_in_four": 1.2252853904147397e-14, + "probability_opposed_gradient": 0.0, + "q": [ + 0.00042900753580406563, + 0.003169960723160097, + 0.00042900753580406563, + 0.9959720242052318 + ], + "reverse_kl": 0.002928107582469479, + "student_entropy_nats": 0.003225499724270909, + "student_max_mass": 0.9996672950105877, + "student_optimal_mass": 0.9996672950105877, + "student_predicted_action_mass": 0.9996672950105877, + "teacher_entropy_nats": 0.02891300078033942, + "teacher_mass_on_student_prediction": 0.9959720242052318, + "teacher_max_mass": 0.9959720242052318, + "teacher_optimal_mass": 0.9959720242052318, + "variance_verification_error": 1.0842021724855044e-19 + }, + { + "current_model_mistake": true, + "enumerated_mean_logit_gradient": [ + 0.014700852908365758, + 0.010910129101099949, + 0.680636522824094, + -0.7062475048335601 + ], + "episode_id": "train-travel-2300031", + "exact_gradient_norm": 0.9810131295140447, + "exact_logit_gradient": [ + 0.01470085290836574, + 0.01091012910109996, + 0.6806365228240945, + -0.7062475048335602 + ], + "expected_cosine": 0.976006476181658, + "feedback_advantage": 0.45693457978879204, + "id": "train-travel-2300031/frozen-warmstart-v2/seed-1707/step-0", + "kind": "focus", + "maximum_expectation_error": 4.440892098500626e-16, + "mc4_error_norm_quantiles": { + "0.1": 0.0638738918312403, + "0.25": 0.0638738918312403, + "0.5": 0.23843478741946528, + "0.75": 0.31818148390987583, + "0.9": 0.5689131632176465 + }, + "mc4_gradient_norm_quantiles": { + "0.1": 0.7438445577638236, + "0.25": 0.7438445577638236, + "0.5": 1.01682924215185, + "0.75": 1.290158728883726, + "0.9": 1.290158728883726 + }, + "mc4_gradient_rmse": 0.33233583249983334, + "noise_signal_ratio": 0.33876797618851384, + "optimal_choices": [ + "D" + ], + "p": [ + 0.018704462657090554, + 0.01650663080434031, + 0.48239445326928454, + 0.48239445326928454 + ], + "probability_all_four_same": 0.10830288015167715, + "probability_no_optimal_in_four": 0.07177871621534286, + "probability_opposed_gradient": 0.0, + "q": [ + 0.0038388231863021394, + 0.0038388231863021394, + 0.05299332056931915, + 0.9393290330580766 + ], + "reverse_kl": 0.7976419795452597, + "student_entropy_nats": 0.8454925142348215, + "student_max_mass": 0.48239445326928454, + "student_optimal_mass": 0.48239445326928454, + "student_predicted_action_mass": 0.48239445326928454, + "teacher_entropy_nats": 0.2571723018502147, + "teacher_mass_on_student_prediction": 0.05299332056931915, + "teacher_max_mass": 0.9393290330580766, + "teacher_optimal_mass": 0.9393290330580766, + "variance_verification_error": 4.163336342344337e-17 + }, + { + "current_model_mistake": false, + "enumerated_mean_logit_gradient": [ + 0.02002079899672836, + 0.01122872975645332, + -0.0508811057981013, + 0.01963157704491964 + ], + "episode_id": "train-travel-2300103", + "exact_gradient_norm": 0.05917096001349866, + "exact_logit_gradient": [ + 0.02002079899672837, + 0.011228729756453323, + -0.05088110579810134, + 0.01963157704491965 + ], + "expected_cosine": 0.977447933873237, + "feedback_advantage": 0.0247094002681304, + "id": "train-travel-2300103/frozen-warmstart-v2/seed-1707/step-1", + "kind": "admin", + "maximum_expectation_error": 3.469446951953614e-17, + "mc4_error_norm_quantiles": { + "0.1": 0.05827863692815885, + "0.25": 0.05827863692815885, + "0.5": 0.05827863692815885, + "0.75": 0.05827863692815885, + "0.9": 0.46187640536795327 + }, + "mc4_gradient_norm_quantiles": { + "0.1": 0.0008942267533972165, + "0.25": 0.0008942267533972165, + "0.5": 0.0008942267533972165, + "0.75": 0.0008942267533972165, + "0.9": 0.510533875112966 + }, + "mc4_gradient_rmse": 0.2050333834122498, + "noise_signal_ratio": 3.4651015188105037, + "optimal_choices": [ + "C" + ], + "p": [ + 0.013831053003887622, + 0.005765640285410845, + 0.9696316714580113, + 0.010771635252690181 + ], + "probability_all_four_same": 0.8839489727954144, + "probability_no_optimal_in_four": 8.505180932356059e-07, + "probability_opposed_gradient": 0.0, + "q": [ + 0.0031647697939285787, + 0.0008001791297558835, + 0.9943410717261417, + 0.0016939793501738489 + ], + "reverse_kl": 0.027310674865470994, + "student_entropy_nats": 0.16764224009046583, + "student_max_mass": 0.9696316714580113, + "student_optimal_mass": 0.9696316714580113, + "student_predicted_action_mass": 0.9696316714580113, + "teacher_entropy_nats": 0.04037282125585986, + "teacher_mass_on_student_prediction": 0.9943410717261417, + "teacher_max_mass": 0.9943410717261417, + "teacher_optimal_mass": 0.9943410717261417, + "variance_verification_error": 3.469446951953614e-17 + }, + { + "current_model_mistake": false, + "enumerated_mean_logit_gradient": [ + -3.406312209206328e-06, + 0.02489266981511005, + -0.024948891238064166, + 5.962773516334278e-05 + ], + "episode_id": "train-travel-2300020", + "exact_gradient_norm": 0.03524337883351437, + "exact_logit_gradient": [ + -3.406312209206326e-06, + 0.024892669815110027, + -0.024948891238064166, + 5.96277351633428e-05 + ], + "expected_cosine": 0.9998332535468226, + "feedback_advantage": 0.016901645089948958, + "id": "train-travel-2300020/frozen-warmstart-v2/seed-1707/step-1", + "kind": "admin", + "maximum_expectation_error": 2.42861286636753e-17, + "mc4_error_norm_quantiles": { + "0.1": 0.03452677587163084, + "0.25": 0.03452677587163084, + "0.5": 0.03452677587163084, + "0.75": 0.03452677587163084, + "0.9": 0.2596487537357604 + }, + "mc4_gradient_norm_quantiles": { + "0.1": 0.0007166030303484287, + "0.25": 0.0007166030303484287, + "0.5": 0.0007166030303484287, + "0.75": 0.0007166030303484287, + "0.9": 0.29489204390546614 + }, + "mc4_gradient_rmse": 0.09936790867241382, + "noise_signal_ratio": 2.819477358905237, + "optimal_choices": [ + "C" + ], + "p": [ + 1.174305150263981e-06, + 0.029309784090482884, + 0.9706067164622976, + 8.232514206932153e-05 + ], + "probability_all_four_same": 0.8875105618910627, + "probability_no_optimal_in_four": 7.464357236135991e-07, + "probability_opposed_gradient": 4.685343249308331e-06, + "q": [ + 2.1177520572974143e-05, + 0.012430896113930202, + 0.9875083615522465, + 3.956481325024147e-05 + ], + "reverse_kl": 0.008440821522661594, + "student_entropy_nats": 0.13320596364422058, + "student_max_mass": 0.9706067164622976, + "student_optimal_mass": 0.9706067164622976, + "student_predicted_action_mass": 0.9706067164622976, + "teacher_entropy_nats": 0.06758373703606342, + "teacher_mass_on_student_prediction": 0.9875083615522465, + "teacher_max_mass": 0.9875083615522465, + "teacher_optimal_mass": 0.9875083615522465, + "variance_verification_error": 5.204170427930421e-18 + }, + { + "current_model_mistake": true, + "enumerated_mean_logit_gradient": [ + -0.14583836341636322, + -0.0014021100447375067, + -0.00015933997720080225, + 0.14739981343830164 + ], + "episode_id": "train-travel-2300050", + "exact_gradient_norm": 0.20735844460116704, + "exact_logit_gradient": [ + -0.14583836341636325, + -0.0014021100447375067, + -0.00015933997720080228, + 0.14739981343830164 + ], + "expected_cosine": 0.9994511711651409, + "feedback_advantage": 0.9141738026347064, + "id": "train-travel-2300050/frozen-warmstart-v2/seed-1707/step-2", + "kind": "exercise", + "maximum_expectation_error": 2.7755575615628914e-17, + "mc4_error_norm_quantiles": { + "0.1": 0.11604290687592642, + "0.25": 0.11604290687592642, + "0.5": 0.11604290687592642, + "0.75": 0.11604290687592642, + "0.9": 0.11604290687592642 + }, + "mc4_gradient_norm_quantiles": { + "0.1": 0.09131610962830106, + "0.25": 0.09131610962830106, + "0.5": 0.09131610962830106, + "0.75": 0.09131610962830106, + "0.9": 0.09131610962830106 + }, + "mc4_gradient_rmse": 0.37788846867803894, + "noise_signal_ratio": 1.8223924731151853, + "optimal_choices": [ + "A" + ], + "p": [ + 0.022969980633319925, + 0.00028914939489595637, + 3.2441660016679145e-05, + 0.9767084283117674 + ], + "probability_all_four_same": 0.9100387273434283, + "probability_no_optimal_in_four": 0.9112375982256933, + "probability_opposed_gradient": 0.0, + "q": [ + 0.9371437832680263, + 0.002632244137423116, + 0.00031437670881003275, + 0.05990959588574053 + ], + "reverse_kl": 2.6404366234106282, + "student_entropy_nats": 0.11238842796600466, + "student_max_mass": 0.9767084283117674, + "student_optimal_mass": 0.022969980633319925, + "student_predicted_action_mass": 0.9767084283117674, + "teacher_entropy_nats": 0.24764939622976076, + "teacher_mass_on_student_prediction": 0.05990959588574053, + "teacher_max_mass": 0.9371437832680263, + "teacher_optimal_mass": 0.9371437832680263, + "variance_verification_error": 2.7755575615628914e-17 + }, + { + "current_model_mistake": true, + "enumerated_mean_logit_gradient": [ + -0.04504936179502382, + -0.003697876378503185, + -0.25444814947233707, + 0.3031953876458647 + ], + "episode_id": "train-travel-2300108", + "exact_gradient_norm": 0.3983897879558068, + "exact_logit_gradient": [ + -0.04504936179502397, + -0.0036978763785031907, + -0.25444814947233746, + 0.30319538764586457 + ], + "expected_cosine": 0.5324512126919654, + "feedback_advantage": 0.6701165550363896, + "id": "train-travel-2300108/frozen-warmstart-v2/seed-1707/step-3", + "kind": "repair", + "maximum_expectation_error": 3.885780586188048e-16, + "mc4_error_norm_quantiles": { + "0.1": 0.18458102695336012, + "0.25": 0.38006479613177924, + "0.5": 0.6455592268956669, + "0.75": 0.7555805166777432, + "0.9": 0.8735415753124968 + }, + "mc4_gradient_norm_quantiles": { + "0.1": 0.2609708358336684, + "0.25": 0.38223167882048253, + "0.5": 0.5707689601939091, + "0.75": 0.897806312528604, + "0.9": 1.1660619327965283 + }, + "mc4_gradient_rmse": 0.686895251854707, + "noise_signal_ratio": 1.7241788635679185, + "optimal_choices": [ + "C" + ], + "p": [ + 0.6179018616676331, + 0.05072047561407089, + 0.07379785925149918, + 0.2575798034667969 + ], + "probability_all_four_same": 0.15021155976349657, + "probability_no_optimal_in_four": 0.7359073182630198, + "probability_opposed_gradient": 0.19985891505835052, + "q": [ + 0.21313504557110302, + 0.017495190464031838, + 0.7439144142878887, + 0.02545534967697636 + ], + "reverse_kl": 1.1373106484422033, + "student_entropy_nats": 0.9904296085898073, + "student_max_mass": 0.6179018616676331, + "student_optimal_mass": 0.07379785925149918, + "student_predicted_action_mass": 0.6179018616676331, + "teacher_entropy_nats": 0.713766865699619, + "teacher_mass_on_student_prediction": 0.21313504557110302, + "teacher_max_mass": 0.7439144142878887, + "teacher_optimal_mass": 0.7439144142878887, + "variance_verification_error": 6.106226635438361e-16 + }, + { + "current_model_mistake": false, + "enumerated_mean_logit_gradient": [ + -4.248026591474677e-05, + -1.2335663075396703, + 1.2329935772521194, + 0.0006152105534666494 + ], + "episode_id": "train-travel-2300059", + "exact_gradient_norm": 1.7441213767770418, + "exact_logit_gradient": [ + -4.2480265914746806e-05, + -1.233566307539673, + 1.2329935772521212, + 0.0006152105534666497 + ], + "expected_cosine": 0.9995330539863804, + "feedback_advantage": 0.37453072278937327, + "id": "train-travel-2300059/frozen-warmstart-v2/seed-1707/step-0", + "kind": "focus", + "maximum_expectation_error": 2.6645352591003757e-15, + "mc4_error_norm_quantiles": { + "0.1": 0.4848658221342704, + "0.25": 0.4848658221342704, + "0.5": 0.5038113975796964, + "0.75": 1.4735364260569022, + "0.9": 1.4924820854832206 + }, + "mc4_gradient_norm_quantiles": { + "0.1": 0.25163934470782284, + "0.25": 1.2403113279181672, + "0.5": 1.2403113279181672, + "0.75": 2.228983582895833, + "0.9": 3.2176558591272673 + }, + "mc4_gradient_rmse": 0.9591772191018071, + "noise_signal_ratio": 0.5499486629045672, + "optimal_choices": [ + "B" + ], + "p": [ + 1.1775014429140784e-05, + 0.6221769587648387, + 0.3773694120608382, + 0.00044185415989397045 + ], + "probability_all_four_same": 0.17012958227144184, + "probability_no_optimal_in_four": 0.020377634014439962, + "probability_opposed_gradient": 3.220382642694397e-10, + "q": [ + 9.579522192541592e-05, + 0.996707681554212, + 0.0031723023976098835, + 2.4220826252720602e-05 + ], + "reverse_kl": 1.5114285313408655, + "student_entropy_nats": 0.6665469182328836, + "student_max_mass": 0.6221769587648387, + "student_optimal_mass": 0.6221769587648387, + "student_predicted_action_mass": 0.6221769587648387, + "teacher_entropy_nats": 0.02268194058903448, + "teacher_mass_on_student_prediction": 0.996707681554212, + "teacher_max_mass": 0.996707681554212, + "teacher_optimal_mass": 0.996707681554212, + "variance_verification_error": 8.881784197001252e-16 + }, + { + "current_model_mistake": false, + "enumerated_mean_logit_gradient": [ + -0.31293726277877404, + 0.0011982829637023117, + 0.29252599280059444, + 0.019212987014476785 + ], + "episode_id": "train-travel-2300112", + "exact_gradient_norm": 0.4288027071410868, + "exact_logit_gradient": [ + -0.312937262778774, + 0.001198282963702313, + 0.2925259928005948, + 0.019212987014476827 + ], + "expected_cosine": 0.9751364246296171, + "feedback_advantage": 0.18465570675505882, + "id": "train-travel-2300112/frozen-warmstart-v2/seed-1707/step-2", + "kind": "exercise", + "maximum_expectation_error": 3.885780586188048e-16, + "mc4_error_norm_quantiles": { + "0.1": 0.05149509322291911, + "0.25": 0.05149509322291911, + "0.5": 0.3586933020373869, + "0.75": 0.4255677992341401, + "0.9": 0.44528902679839616 + }, + "mc4_gradient_norm_quantiles": { + "0.1": 0.0701160564016255, + "0.25": 0.0701160564016255, + "0.5": 0.4709042723860531, + "0.75": 0.4709042723860531, + "0.9": 0.8719088938639362 + }, + "mc4_gradient_rmse": 0.34769147062282973, + "noise_signal_ratio": 0.8108425269536149, + "optimal_choices": [ + "A" + ], + "p": [ + 0.7622589954966544, + 0.00352995977297949, + 0.21839086396148577, + 0.015820180768880247 + ], + "probability_all_four_same": 0.3398808942166405, + "probability_no_optimal_in_four": 0.0031945991621408192, + "probability_opposed_gradient": 0.0, + "q": [ + 0.9469147022517133, + 0.0020713675106462676, + 0.04714410601678711, + 0.003869824220853411 + ], + "reverse_kl": 0.19361665416616322, + "student_entropy_nats": 0.624734200510649, + "student_max_mass": 0.7622589954966544, + "student_optimal_mass": 0.7622589954966544, + "student_predicted_action_mass": 0.7622589954966544, + "teacher_entropy_nats": 0.22994973939233976, + "teacher_mass_on_student_prediction": 0.9469147022517133, + "teacher_max_mass": 0.9469147022517133, + "teacher_optimal_mass": 0.9469147022517133, + "variance_verification_error": 1.8041124150158794e-16 + }, + { + "current_model_mistake": true, + "enumerated_mean_logit_gradient": [ + 0.3547198817727627, + -0.3443089108825154, + -0.002303106385296777, + -0.008107864504948931 + ], + "episode_id": "train-travel-2300111", + "exact_gradient_norm": 0.49441466645300913, + "exact_logit_gradient": [ + 0.35471988177276254, + -0.34430891088251697, + -0.002303106385296768, + -0.008107864504948917 + ], + "expected_cosine": 0.9986667026879387, + "feedback_advantage": 0.8749602713266987, + "id": "train-travel-2300111/frozen-warmstart-v2/seed-1707/step-0", + "kind": "focus", + "maximum_expectation_error": 1.5543122344752192e-15, + "mc4_error_norm_quantiles": { + "0.1": 0.21396619383022378, + "0.25": 0.21396619383022378, + "0.5": 0.21396619383022378, + "0.75": 0.25611573358807704, + "0.9": 0.5845093511959601 + }, + "mc4_gradient_norm_quantiles": { + "0.1": 0.28058910250431124, + "0.25": 0.28058910250431124, + "0.5": 0.28058910250431124, + "0.75": 0.28058910250431124, + "0.9": 1.0787953339229093 + }, + "mc4_gradient_rmse": 0.4004032917144335, + "noise_signal_ratio": 0.8098531837394213, + "optimal_choices": [ + "B" + ], + "p": [ + 0.9286832400437213, + 0.06727358834332912, + 0.0010873762876861843, + 0.002955795325263472 + ], + "probability_all_four_same": 0.7438449011950024, + "probability_no_optimal_in_four": 0.7568626931230737, + "probability_opposed_gradient": 4.380241139150427e-08, + "q": [ + 0.05315719912775338, + 0.9422338596700278, + 0.0007582466926324588, + 0.003850694509586295 + ], + "reverse_kl": 2.4785541332490797, + "student_entropy_nats": 0.2749163492879753, + "student_max_mass": 0.9286832400437213, + "student_optimal_mass": 0.06727358834332912, + "student_predicted_action_mass": 0.9286832400437213, + "teacher_entropy_nats": 0.2389100491823568, + "teacher_mass_on_student_prediction": 0.05315719912775338, + "teacher_max_mass": 0.9422338596700278, + "teacher_optimal_mass": 0.9422338596700278, + "variance_verification_error": 5.551115123125783e-16 + }, + { + "current_model_mistake": false, + "enumerated_mean_logit_gradient": [ + -0.0005228428858265928, + -0.1004882882123177, + -3.165155705769543e-05, + 0.10104278265520182 + ], + "episode_id": "train-travel-2300082", + "exact_gradient_norm": 0.1425054888804375, + "exact_logit_gradient": [ + -0.0005228428858265925, + -0.10048828821231759, + -3.1651557057695455e-05, + 0.10104278265520188 + ], + "expected_cosine": 0.9954996068342523, + "feedback_advantage": 0.045103794916949824, + "id": "train-travel-2300082/frozen-warmstart-v2/seed-1707/step-2", + "kind": "exercise", + "maximum_expectation_error": 1.1102230246251565e-16, + "mc4_error_norm_quantiles": { + "0.1": 0.1389611460527513, + "0.25": 0.1389611460527513, + "0.5": 0.1389611460527513, + "0.75": 0.1389611460527513, + "0.9": 0.5136340684872037 + }, + "mc4_gradient_norm_quantiles": { + "0.1": 0.00354497218559655, + "0.25": 0.00354497218559655, + "0.5": 0.00354497218559655, + "0.75": 0.00354497218559655, + "0.9": 0.6561382049168507 + }, + "mc4_gradient_rmse": 0.2936076117144534, + "noise_signal_ratio": 2.060324932191145, + "optimal_choices": [ + "B" + ], + "p": [ + 0.0008624188487076085, + 0.9457571465889841, + 2.4464940169494172e-05, + 0.05335596962213872 + ], + "probability_all_four_same": 0.8000606786946833, + "probability_no_optimal_in_four": 8.657053650511058e-06, + "probability_opposed_gradient": 0.0030052212063977203, + "q": [ + 0.0014896991234305934, + 0.990860941505934, + 8.404307321260289e-05, + 0.007565316297422897 + ], + "reverse_kl": 0.05966329957738099, + "student_entropy_nats": 0.215463212035978, + "student_max_mass": 0.9457571465889841, + "student_optimal_mass": 0.9457571465889841, + "student_predicted_action_mass": 0.9457571465889841, + "teacher_entropy_nats": 0.056532941351068344, + "teacher_mass_on_student_prediction": 0.990860941505934, + "teacher_max_mass": 0.990860941505934, + "teacher_optimal_mass": 0.990860941505934, + "variance_verification_error": 4.163336342344337e-17 + }, + { + "current_model_mistake": false, + "enumerated_mean_logit_gradient": [ + -0.7185542121683568, + -0.004699732587939591, + -0.0004982191454460124, + 0.723752163901744 + ], + "episode_id": "train-travel-2300014", + "exact_gradient_norm": 1.0198821923565773, + "exact_logit_gradient": [ + -0.7185542121683571, + -0.0046997325879395944, + -0.0004982191454460118, + 0.7237521639017427 + ], + "expected_cosine": 0.9992951577089864, + "feedback_advantage": 0.2785588757568417, + "id": "train-travel-2300014/frozen-warmstart-v2/seed-1707/step-2", + "kind": "exercise", + "maximum_expectation_error": 1.3322676295501878e-15, + "mc4_error_norm_quantiles": { + "0.1": 0.12658945148537917, + "0.25": 0.12658945148537917, + "0.5": 0.626963431918347, + "0.75": 0.8801358717380313, + "0.9": 0.8801358717380313 + }, + "mc4_gradient_norm_quantiles": { + "0.1": 0.13979240718657776, + "0.25": 0.13979240718657776, + "0.5": 0.8932994067011727, + "0.75": 1.6468449417435977, + "0.9": 1.6468449417435977 + }, + "mc4_gradient_rmse": 0.6860871018746856, + "noise_signal_ratio": 0.6727121102971584, + "optimal_choices": [ + "A" + ], + "p": [ + 0.7013558429098083, + 0.006067916942928065, + 0.00020763269795201296, + 0.2923686074493116 + ], + "probability_all_four_same": 0.24927235771418596, + "probability_no_optimal_in_four": 0.007954558665566476, + "probability_opposed_gradient": 0.00029416287630392617, + "q": [ + 0.97991471866665, + 0.006602613213874736, + 0.0011473621635281531, + 0.012335305955947083 + ], + "reverse_kl": 0.6900714365653572, + "student_entropy_nats": 0.6410720573533747, + "student_max_mass": 0.7013558429098083, + "student_optimal_mass": 0.7013558429098083, + "student_predicted_action_mass": 0.7013558429098083, + "teacher_entropy_nats": 0.11501445724619513, + "teacher_mass_on_student_prediction": 0.97991471866665, + "teacher_max_mass": 0.97991471866665, + "teacher_optimal_mass": 0.97991471866665, + "variance_verification_error": 1.1102230246251565e-16 + }, + { + "current_model_mistake": false, + "enumerated_mean_logit_gradient": [ + 2.3811183075310462e-05, + -0.12643019411318854, + 1.76343389327095e-05, + 0.1263887485911806 + ], + "episode_id": "train-travel-2300084", + "exact_gradient_norm": 0.17876999365633356, + "exact_logit_gradient": [ + 2.3811183075310482e-05, + -0.1264301941131886, + 1.7634338932709536e-05, + 0.12638874859118057 + ], + "expected_cosine": 0.9995592299685225, + "feedback_advantage": 0.02582537683337016, + "id": "train-travel-2300084/frozen-warmstart-v2/seed-1707/step-1", + "kind": "admin", + "maximum_expectation_error": 5.551115123125783e-17, + "mc4_error_norm_quantiles": { + "0.1": 0.17780527484504507, + "0.25": 0.17780527484504507, + "0.5": 0.17780527484504507, + "0.75": 0.17780527484504507, + "0.9": 0.17780527484504507 + }, + "mc4_gradient_norm_quantiles": { + "0.1": 0.0009647404021785887, + "0.25": 0.0009647404021785887, + "0.5": 0.0009647404021785887, + "0.75": 0.0009647404021785887, + "0.9": 0.07846557626606893 + }, + "mc4_gradient_rmse": 0.5445335960877099, + "noise_signal_ratio": 3.0460010930834303, + "optimal_choices": [ + "B" + ], + "p": [ + 0.00019814290732924226, + 0.9738265927220864, + 2.366479542626597e-05, + 0.02595159957515809 + ], + "probability_all_four_same": 0.8993458571694772, + "probability_no_optimal_in_four": 4.6928973272611605e-07, + "probability_opposed_gradient": 0.0, + "q": [ + 0.00015840617361627897, + 0.9996519695554565, + 1.0126567165664017e-05, + 0.00017949770376147656 + ], + "reverse_kl": 0.10365430516723777, + "student_entropy_nats": 0.1225322291177002, + "student_max_mass": 0.9738265927220864, + "student_optimal_mass": 0.9738265927220864, + "student_predicted_action_mass": 0.9738265927220864, + "teacher_entropy_nats": 0.003398768269916631, + "teacher_mass_on_student_prediction": 0.9996519695554565, + "teacher_max_mass": 0.9996519695554565, + "teacher_optimal_mass": 0.9996519695554565, + "variance_verification_error": 2.220446049250313e-16 + }, + { + "current_model_mistake": false, + "enumerated_mean_logit_gradient": [ + -0.0004065540598109197, + 0.0018182803732465897, + -0.0019326724006706027, + 0.000520946087234935 + ], + "episode_id": "train-travel-2300008", + "exact_gradient_norm": 0.002734599998793565, + "exact_logit_gradient": [ + -0.00040655405981092043, + 0.0018182803732465897, + -0.0019326724006706044, + 0.000520946087234935 + ], + "expected_cosine": -0.9484452557914791, + "feedback_advantage": -2.4541199598826502e-05, + "id": "train-travel-2300008/frozen-warmstart-v2/seed-1707/step-0", + "kind": "focus", + "maximum_expectation_error": 1.734723475976807e-18, + "mc4_error_norm_quantiles": { + "0.1": 0.0027346558189445796, + "0.25": 0.0027346558189445796, + "0.5": 0.0027346558189445796, + "0.75": 0.0027346558189445796, + "0.9": 0.0027346558189445796 + }, + "mc4_gradient_norm_quantiles": { + "0.1": 5.8053940989052406e-08, + "0.25": 5.8053940989052406e-08, + "0.5": 5.8053940989052406e-08, + "0.75": 5.8053940989052406e-08, + "0.9": 5.8053940989052406e-08 + }, + "mc4_gradient_rmse": 0.0448230867433835, + "noise_signal_ratio": 16.39109440618675, + "optimal_choices": [ + "C" + ], + "p": [ + 0.00020308041824225012, + 0.0013242505814196313, + 0.9980932648306564, + 0.00037940416968166115 + ], + "probability_all_four_same": 0.9923948454441742, + "probability_no_optimal_in_four": 1.321787098203388e-11, + "probability_opposed_gradient": 0.9932027767036529, + "q": [ + 0.0015005356004335205, + 0.00033481473502353976, + 0.9980687236310576, + 9.592603348529963e-05 + ], + "reverse_kl": 0.0019609529198826776, + "student_entropy_nats": 0.015395706800202783, + "student_max_mass": 0.9980932648306564, + "student_optimal_mass": 0.9980932648306564, + "student_predicted_action_mass": 0.9980932648306564, + "teacher_entropy_nats": 0.015252458838692607, + "teacher_mass_on_student_prediction": 0.9980687236310576, + "teacher_max_mass": 0.9980687236310576, + "teacher_optimal_mass": 0.9980687236310576, + "variance_verification_error": 8.673617379884035e-19 + }, + { + "current_model_mistake": false, + "enumerated_mean_logit_gradient": [ + -6.072669315248995e-05, + -0.028628697151020944, + -1.055271793599227e-05, + 0.028699976562109396 + ], + "episode_id": "train-travel-2300078", + "exact_gradient_norm": 0.040537572131562345, + "exact_logit_gradient": [ + -6.072669315248995e-05, + -0.028628697151020923, + -1.0552717935992272e-05, + 0.028699976562109406 + ], + "expected_cosine": 0.9997094985690361, + "feedback_advantage": 0.008117156540829451, + "id": "train-travel-2300078/frozen-warmstart-v2/seed-1707/step-1", + "kind": "admin", + "maximum_expectation_error": 2.0816681711721685e-17, + "mc4_error_norm_quantiles": { + "0.1": 0.040438392510302436, + "0.25": 0.040438392510302436, + "0.5": 0.040438392510302436, + "0.75": 0.040438392510302436, + "0.9": 0.040438392510302436 + }, + "mc4_gradient_norm_quantiles": { + "0.1": 9.918156055838753e-05, + "0.25": 9.918156055838753e-05, + "0.5": 9.918156055838753e-05, + "0.75": 9.918156055838753e-05, + "0.9": 9.918156055838753e-05 + }, + "mc4_gradient_rmse": 0.21776722918374838, + "noise_signal_ratio": 5.371984993995137, + "optimal_choices": [ + "B" + ], + "p": [ + 3.9719781746747004e-05, + 0.9913762923388916, + 6.902263641830375e-06, + 0.008577085615719845 + ], + "probability_all_four_same": 0.9659488249780952, + "probability_no_optimal_in_four": 5.530649075796598e-09, + "probability_opposed_gradient": 0.00018171782784678512, + "q": [ + 0.00017946924518540225, + 0.999493448879721, + 3.118708167824952e-05, + 0.0002958947934153132 + ], + "reverse_kl": 0.020723302718079387, + "student_entropy_nats": 0.04988639074444729, + "student_max_mass": 0.9913762923388916, + "student_optimal_mass": 0.9913762923388916, + "student_predicted_action_mass": 0.9913762923388916, + "teacher_entropy_nats": 0.00478231284539364, + "teacher_mass_on_student_prediction": 0.999493448879721, + "teacher_max_mass": 0.999493448879721, + "teacher_optimal_mass": 0.999493448879721, + "variance_verification_error": 6.938893903907228e-18 + }, + { + "current_model_mistake": false, + "enumerated_mean_logit_gradient": [ + -4.789712539663171e-05, + 5.277023321414571e-05, + -4.876592864866178e-06, + 3.4850473522210706e-09 + ], + "episode_id": "train-travel-2300067", + "exact_gradient_norm": 7.143257873583472e-05, + "exact_logit_gradient": [ + -4.78971253966318e-05, + 5.277023321414581e-05, + -4.876592864866186e-06, + 3.4850473522210797e-09 + ], + "expected_cosine": 0.6875284044959842, + "feedback_advantage": -0.0003711005334182671, + "id": "train-travel-2300067/frozen-warmstart-v2/seed-1707/step-1", + "kind": "admin", + "maximum_expectation_error": 9.486769009248164e-20, + "mc4_error_norm_quantiles": { + "0.1": 7.140286834300142e-05, + "0.25": 7.140286834300142e-05, + "0.5": 7.140286834300142e-05, + "0.75": 7.140286834300142e-05, + "0.9": 7.140286834300142e-05 + }, + "mc4_gradient_norm_quantiles": { + "0.1": 4.320402141576934e-08, + "0.25": 4.320402141576934e-08, + "0.5": 4.320402141576934e-08, + "0.75": 4.320402141576934e-08, + "0.9": 4.320402141576934e-08 + }, + "mc4_gradient_rmse": 0.008872383045635588, + "noise_signal_ratio": 124.20639437428972, + "optimal_choices": [ + "B" + ], + "p": [ + 1.4737816291915159e-05, + 0.9999062057982658, + 1.300607740287897e-05, + 6.605030803936326e-05 + ], + "probability_all_four_same": 0.9996248759738765, + "probability_no_optimal_in_four": 7.739340711993441e-17, + "probability_opposed_gradient": 0.00026415306960066864, + "q": [ + 0.00037995223190918345, + 0.9995351052648476, + 1.8916708038220027e-05, + 6.602579520505415e-05 + ], + "reverse_kl": 0.00031842904830781647, + "student_entropy_nats": 0.0010398093975801037, + "student_max_mass": 0.9999062057982658, + "student_optimal_mass": 0.9999062057982658, + "student_predicted_action_mass": 0.9999062057982658, + "teacher_entropy_nats": 0.004298344146780182, + "teacher_mass_on_student_prediction": 0.9995351052648476, + "teacher_max_mass": 0.9995351052648476, + "teacher_optimal_mass": 0.9995351052648476, + "variance_verification_error": 5.421010862427522e-20 + }, + { + "current_model_mistake": true, + "enumerated_mean_logit_gradient": [ + -0.44033420776525506, + -0.00026152388095379434, + 0.0003880300109563215, + 0.4402077016352522 + ], + "episode_id": "train-travel-2300048", + "exact_gradient_norm": 0.622637337516211, + "exact_logit_gradient": [ + -0.440334207765255, + -0.0002615238809537943, + 0.0003880300109563214, + 0.4402077016352526 + ], + "expected_cosine": 0.99487441491828, + "feedback_advantage": 0.7036571542956169, + "id": "train-travel-2300048/frozen-warmstart-v2/seed-1707/step-1", + "kind": "admin", + "maximum_expectation_error": 3.885780586188048e-16, + "mc4_error_norm_quantiles": { + "0.1": 0.18068147874123203, + "0.25": 0.18068147874123203, + "0.5": 0.254426974979724, + "0.75": 0.254426974979724, + "0.9": 0.6157129703371086 + }, + "mc4_gradient_norm_quantiles": { + "0.1": 0.36828144356750625, + "0.25": 0.36828144356750625, + "0.5": 0.36828144356750625, + "0.75": 0.8033050006038859, + "0.9": 1.238349942557714 + }, + "mc4_gradient_rmse": 0.3129091796886849, + "noise_signal_ratio": 0.5025544740649898, + "optimal_choices": [ + "A" + ], + "p": [ + 0.1476659417449555, + 0.0011274434638945392, + 0.001447666261875436, + 0.8497589485292745 + ], + "probability_all_four_same": 0.5218898269799236, + "probability_no_optimal_in_four": 0.527763523700169, + "probability_opposed_gradient": 2.8789785204193496e-05, + "q": [ + 0.8513230960405724, + 0.0004155267806456877, + 0.00032361257368430257, + 0.14793776460509755 + ], + "reverse_kl": 1.2301227192403088, + "student_entropy_nats": 0.43791616481188433, + "student_max_mass": 0.8497589485292745, + "student_optimal_mass": 0.1476659417449555, + "student_predicted_action_mass": 0.8497589485292745, + "teacher_entropy_nats": 0.42557149182541676, + "teacher_mass_on_student_prediction": 0.14793776460509755, + "teacher_max_mass": 0.8513230960405724, + "teacher_optimal_mass": 0.8513230960405724, + "variance_verification_error": 2.7755575615628914e-17 + }, + { + "current_model_mistake": false, + "enumerated_mean_logit_gradient": [ + -0.00010812579068851573, + -0.001137968349015087, + -5.439657846090441e-06, + 0.0012515337975496919 + ], + "episode_id": "train-travel-2300043", + "exact_gradient_norm": 0.0016950013528781549, + "exact_logit_gradient": [ + -0.00010812579068851579, + -0.0011379683490150864, + -5.439657846090437e-06, + 0.0012515337975496925 + ], + "expected_cosine": 0.9855905656885309, + "feedback_advantage": 0.00046464238861665397, + "id": "train-travel-2300043/frozen-warmstart-v2/seed-1707/step-2", + "kind": "exercise", + "maximum_expectation_error": 6.505213034913027e-19, + "mc4_error_norm_quantiles": { + "0.1": 0.0016943761644455727, + "0.25": 0.0016943761644455727, + "0.5": 0.0016943761644455727, + "0.75": 0.0016943761644455727, + "0.9": 0.0016943761644455727 + }, + "mc4_gradient_norm_quantiles": { + "0.1": 6.339975482987568e-07, + "0.25": 6.339975482987568e-07, + "0.5": 6.339975482987568e-07, + "0.75": 6.339975482987568e-07, + "0.9": 6.339975482987568e-07 + }, + "mc4_gradient_rmse": 0.030397801517268088, + "noise_signal_ratio": 17.93379189087481, + "optimal_choices": [ + "B" + ], + "p": [ + 9.601459230953281e-05, + 0.9989901249431046, + 2.899389547155282e-06, + 0.0009109610750387552 + ], + "probability_all_four_same": 0.9959666147402557, + "probability_no_optimal_in_four": 1.0400891895171149e-12, + "probability_opposed_gradient": 0.00039451703769907123, + "q": [ + 0.00029588332705866925, + 0.9994547673317212, + 1.891518640778322e-05, + 0.00023043415481228057 + ], + "reverse_kl": 0.0006741147539774865, + "student_entropy_nats": 0.008312215047674041, + "student_max_mass": 0.9989901249431046, + "student_optimal_mass": 0.9989901249431046, + "student_predicted_action_mass": 0.9989901249431046, + "teacher_entropy_nats": 0.005085022087700366, + "teacher_mass_on_student_prediction": 0.9994547673317212, + "teacher_max_mass": 0.9994547673317212, + "teacher_optimal_mass": 0.9994547673317212, + "variance_verification_error": 1.0842021724855044e-19 + }, + { + "current_model_mistake": false, + "enumerated_mean_logit_gradient": [ + -0.0076886946658332594, + 0.00015492325070387664, + 0.006874903481567689, + 0.0006588679335616953 + ], + "episode_id": "train-travel-2300020", + "exact_gradient_norm": 0.010336267784493985, + "exact_logit_gradient": [ + -0.007688694665833237, + 0.00015492325070387664, + 0.006874903481567665, + 0.0006588679335616943 + ], + "expected_cosine": 0.9988568799355372, + "feedback_advantage": 0.003990976107786315, + "id": "train-travel-2300020/frozen-warmstart-v2/seed-1707/step-0", + "kind": "focus", + "maximum_expectation_error": 2.42861286636753e-17, + "mc4_error_norm_quantiles": { + "0.1": 0.010308513808481648, + "0.25": 0.010308513808481648, + "0.5": 0.010308513808481648, + "0.75": 0.010308513808481648, + "0.9": 0.010308513808481648 + }, + "mc4_gradient_norm_quantiles": { + "0.1": 2.7756787742047556e-05, + "0.25": 2.7756787742047556e-05, + "0.5": 2.7756787742047556e-05, + "0.75": 2.7756787742047556e-05, + "0.9": 2.7756787742047556e-05 + }, + "mc4_gradient_rmse": 0.07562362735119489, + "noise_signal_ratio": 7.316337862747919, + "optimal_choices": [ + "A" + ], + "p": [ + 0.9948361950629281, + 0.0001786329783619182, + 0.004607005901390053, + 0.0003781660573199747 + ], + "probability_all_four_same": 0.9795042199328126, + "probability_no_optimal_in_four": 7.110159015761402e-10, + "probability_opposed_gradient": 0.0, + "q": [ + 0.9988271711707144, + 7.476404513758454e-05, + 0.0010320857521651985, + 6.597903198286114e-05 + ], + "reverse_kl": 0.003724937345159065, + "student_entropy_nats": 0.03445860698650331, + "student_max_mass": 0.9948361950629281, + "student_optimal_mass": 0.9948361950629281, + "student_predicted_action_mass": 0.9948361950629281, + "teacher_entropy_nats": 0.009614413293678594, + "teacher_mass_on_student_prediction": 0.9988271711707144, + "teacher_max_mass": 0.9988271711707144, + "teacher_optimal_mass": 0.9988271711707144, + "variance_verification_error": 4.336808689942018e-18 + }, + { + "current_model_mistake": true, + "enumerated_mean_logit_gradient": [ + -0.2040903558370122, + -0.01284869278091376, + -0.012690930425903212, + 0.22962997904382917 + ], + "episode_id": "train-travel-2300044", + "exact_gradient_norm": 0.3077481912905436, + "exact_logit_gradient": [ + -0.2040903558370122, + -0.012848692780913763, + -0.012690930425903215, + 0.22962997904382892 + ], + "expected_cosine": 0.9923831423246474, + "feedback_advantage": 0.9392785316232912, + "id": "train-travel-2300044/frozen-warmstart-v2/seed-1707/step-1", + "kind": "admin", + "maximum_expectation_error": 2.498001805406602e-16, + "mc4_error_norm_quantiles": { + "0.1": 0.13397871814994575, + "0.25": 0.13397871814994575, + "0.5": 0.13397871814994575, + "0.75": 0.13397871814994575, + "0.9": 1.021606618396809 + }, + "mc4_gradient_norm_quantiles": { + "0.1": 0.17472527711776067, + "0.25": 0.17472527711776067, + "0.5": 0.17472527711776067, + "0.75": 0.17472527711776067, + "0.9": 1.3282917850086522 + }, + "mc4_gradient_rmse": 0.3894463454930071, + "noise_signal_ratio": 1.2654707859040921, + "optimal_choices": [ + "A" + ], + "p": [ + 0.029106838661514586, + 0.003939182249927146, + 0.003067838300877171, + 0.9638861407876811 + ], + "probability_all_four_same": 0.8631838468259141, + "probability_no_optimal_in_four": 0.8885579732637422, + "probability_opposed_gradient": 8.673641433710005e-05, + "q": [ + 0.9683853702848058, + 0.003082158463317984, + 0.005758229587238443, + 0.02277424166463781 + ], + "reverse_kl": 3.5071095201579405, + "student_entropy_nats": 0.1779616112888418, + "student_max_mass": 0.9638861407876811, + "student_optimal_mass": 0.029106838661514586, + "student_predicted_action_mass": 0.9638861407876811, + "teacher_entropy_nats": 0.16476190474349955, + "teacher_mass_on_student_prediction": 0.02277424166463781, + "teacher_max_mass": 0.9683853702848058, + "teacher_optimal_mass": 0.9683853702848058, + "variance_verification_error": 1.1102230246251565e-16 + }, + { + "current_model_mistake": false, + "enumerated_mean_logit_gradient": [ + 0.06172131542123895, + 0.034694836193407866, + -0.10102040968229586, + 0.0046042580676488596 + ], + "episode_id": "train-travel-2300090", + "exact_gradient_norm": 0.12344867273708486, + "exact_logit_gradient": [ + 0.061721315421239036, + 0.03469483619340792, + -0.1010204096822958, + 0.004604258067648859 + ], + "expected_cosine": 0.9727398978381718, + "feedback_advantage": 0.041759157354708254, + "id": "train-travel-2300090/frozen-warmstart-v2/seed-1707/step-0", + "kind": "focus", + "maximum_expectation_error": 8.326672684688674e-17, + "mc4_error_norm_quantiles": { + "0.1": 0.12092841553956618, + "0.25": 0.12092841553956618, + "0.5": 0.12092841553956618, + "0.75": 0.12092841553956618, + "0.9": 0.5533971965705881 + }, + "mc4_gradient_norm_quantiles": { + "0.1": 0.002532058622689476, + "0.25": 0.002532058622689476, + "0.5": 0.002532058622689476, + "0.75": 0.002532058622689476, + "0.9": 0.6660557289002529 + }, + "mc4_gradient_rmse": 0.32664792539811993, + "noise_signal_ratio": 2.6460221738778773, + "optimal_choices": [ + "C" + ], + "p": [ + 0.03258902322322157, + 0.011988831669042367, + 0.9523908936277701, + 0.0030312514799660417 + ], + "probability_all_four_same": 0.8227379745025586, + "probability_no_optimal_in_four": 5.137598000477353e-06, + "probability_opposed_gradient": 0.0, + "q": [ + 0.004603828342640711, + 0.0006230603374405071, + 0.9941500509824783, + 0.0006230603374405071 + ], + "reverse_kl": 0.06315772405240198, + "student_entropy_nats": 0.22864851524626278, + "student_max_mass": 0.9523908936277701, + "student_optimal_mass": 0.9523908936277701, + "student_predicted_action_mass": 0.9523908936277701, + "teacher_entropy_nats": 0.03980284415045393, + "teacher_mass_on_student_prediction": 0.9941500509824783, + "teacher_max_mass": 0.9941500509824783, + "teacher_optimal_mass": 0.9941500509824783, + "variance_verification_error": 5.551115123125783e-17 + }, + { + "current_model_mistake": false, + "enumerated_mean_logit_gradient": [ + -1.6493879476828643e-05, + -0.02759191569763784, + -1.5233920548802965e-05, + 0.027623643497663542 + ], + "episode_id": "train-travel-2300059", + "exact_gradient_norm": 0.039043309235667484, + "exact_logit_gradient": [ + -1.649387947682868e-05, + -0.02759191569763787, + -1.5233920548802955e-05, + 0.027623643497663504 + ], + "expected_cosine": 0.999948665013913, + "feedback_advantage": 0.010778118712939766, + "id": "train-travel-2300059/frozen-warmstart-v2/seed-1707/step-1", + "kind": "admin", + "maximum_expectation_error": 3.8163916471489756e-17, + "mc4_error_norm_quantiles": { + "0.1": 0.038852407251150974, + "0.25": 0.038852407251150974, + "0.5": 0.038852407251150974, + "0.75": 0.038852407251150974, + "0.9": 0.038852407251150974 + }, + "mc4_gradient_norm_quantiles": { + "0.1": 0.00019090214727299933, + "0.25": 0.00019090214727299933, + "0.5": 0.00019090214727299933, + "0.75": 0.00019090214727299933, + "0.9": 0.00019090214727299933 + }, + "mc4_gradient_rmse": 0.17340788245131394, + "noise_signal_ratio": 4.441423789275001, + "optimal_choices": [ + "B" + ], + "p": [ + 4.439280889414118e-06, + 0.9875598462589095, + 4.170318297371926e-06, + 0.012431544141903728 + ], + "probability_all_four_same": 0.9511602765950954, + "probability_no_optimal_in_four": 2.3949860624191467e-08, + "probability_opposed_gradient": 3.316950706590297e-05, + "q": [ + 0.00017926177293867826, + 0.9983379649718492, + 0.00015819796084519402, + 0.0013245752943668598 + ], + "reverse_kl": 0.017084724717979893, + "student_entropy_nats": 0.06701245186809049, + "student_max_mass": 0.9875598462589095, + "student_optimal_mass": 0.9875598462589095, + "student_predicted_action_mass": 0.9875598462589095, + "teacher_entropy_nats": 0.013369093989073075, + "teacher_mass_on_student_prediction": 0.9983379649718492, + "teacher_max_mass": 0.9983379649718492, + "teacher_optimal_mass": 0.9983379649718492, + "variance_verification_error": 4.5102810375396984e-17 + }, + { + "current_model_mistake": false, + "enumerated_mean_logit_gradient": [ + -0.002549779566855908, + 0.003810551920182794, + -0.003314806131917472, + 0.0020540337785905846 + ], + "episode_id": "train-travel-2300035", + "exact_gradient_norm": 0.006019026186332258, + "exact_logit_gradient": [ + -0.0025497795668559062, + 0.003810551920182793, + -0.0033148061319174713, + 0.002054033778590586 + ], + "expected_cosine": -0.5934261373160984, + "feedback_advantage": -0.004360806291487385, + "id": "train-travel-2300035/frozen-warmstart-v2/seed-1707/step-0", + "kind": "focus", + "maximum_expectation_error": 1.734723475976807e-18, + "mc4_error_norm_quantiles": { + "0.1": 0.006030012299498988, + "0.25": 0.006030012299498988, + "0.5": 0.006030012299498988, + "0.75": 0.006030012299498988, + "0.9": 0.006030012299498988 + }, + "mc4_gradient_norm_quantiles": { + "0.1": 1.7928939951426153e-05, + "0.25": 1.7928939951426153e-05, + "0.5": 1.7928939951426153e-05, + "0.75": 1.7928939951426153e-05, + "0.9": 1.7928939951426153e-05 + }, + "mc4_gradient_rmse": 0.09062161154494476, + "noise_signal_ratio": 15.055859326667221, + "optimal_choices": [ + "C", + "D" + ], + "p": [ + 0.0016960884020541236, + 0.0016960884020541236, + 0.9955790936788342, + 0.0010287295170574781 + ], + "probability_all_four_same": 0.9824332959751245, + "probability_no_optimal_in_four": 1.3240790692482487e-10, + "probability_opposed_gradient": 0.9891451952517438, + "q": [ + 0.007574839932981527, + 0.0001781431626142128, + 0.9921082788582242, + 0.00013873804618013948 + ], + "reverse_kl": 0.00682184380732853, + "student_entropy_nats": 0.03312834969062694, + "student_max_mass": 0.9955790936788342, + "student_optimal_mass": 0.9966078231958917, + "student_predicted_action_mass": 0.9955790936788342, + "teacher_entropy_nats": 0.04761815543620837, + "teacher_mass_on_student_prediction": 0.9921082788582242, + "teacher_max_mass": 0.9921082788582242, + "teacher_optimal_mass": 0.9922470169044043, + "variance_verification_error": 5.204170427930421e-18 + }, + { + "current_model_mistake": false, + "enumerated_mean_logit_gradient": [ + -0.0001959097946221977, + 0.00040315550311480767, + -0.00033053884969403276, + 0.00012329314120142248 + ], + "episode_id": "train-travel-2300109", + "exact_gradient_norm": 0.0005704140050393518, + "exact_logit_gradient": [ + -0.00019590979462219763, + 0.00040315550311480783, + -0.00033053884969403265, + 0.00012329314120142242 + ], + "expected_cosine": -0.6333908922901049, + "feedback_advantage": -2.3292143058073478e-05, + "id": "train-travel-2300109/frozen-warmstart-v2/seed-1707/step-0", + "kind": "focus", + "maximum_expectation_error": 1.6263032587282567e-19, + "mc4_error_norm_quantiles": { + "0.1": 0.000570424635999653, + "0.25": 0.000570424635999653, + "0.5": 0.000570424635999653, + "0.75": 0.000570424635999653, + "0.9": 0.000570424635999653 + }, + "mc4_gradient_norm_quantiles": { + "0.1": 1.671788026400076e-08, + "0.25": 1.671788026400076e-08, + "0.5": 1.671788026400076e-08, + "0.75": 1.671788026400076e-08, + "0.9": 1.671788026400076e-08 + }, + "mc4_gradient_rmse": 0.022078200468606264, + "noise_signal_ratio": 38.705572222202235, + "optimal_choices": [ + "C" + ], + "p": [ + 0.00026109792185281534, + 0.00023041810345613754, + 0.9993851500427376, + 0.00012333393195345171 + ], + "probability_all_four_same": 0.9975428674841682, + "probability_no_optimal_in_four": 1.4291459691662792e-13, + "probability_opposed_gradient": 0.998585742524318, + "q": [ + 0.0005527314163036175, + 4.003972387628832e-05, + 0.9993618578996796, + 4.53709601405328e-05 + ], + "reverse_kl": 0.00035404895116760684, + "student_entropy_nats": 0.005808853912992493, + "student_max_mass": 0.9993851500427376, + "student_optimal_mass": 0.9993851500427376, + "student_predicted_action_mass": 0.9993851500427376, + "teacher_entropy_nats": 0.005642943239186078, + "teacher_mass_on_student_prediction": 0.9993618578996796, + "teacher_max_mass": 0.9993618578996796, + "teacher_optimal_mass": 0.9993618578996796, + "variance_verification_error": 2.168404344971009e-19 + }, + { + "current_model_mistake": true, + "enumerated_mean_logit_gradient": [ + -0.21745319944427288, + -0.006331011349559136, + -0.023399717223609583, + 0.24718392801744174 + ], + "episode_id": "train-travel-2300061", + "exact_gradient_norm": 0.3301112186665818, + "exact_logit_gradient": [ + -0.217453199444273, + -0.006331011349559137, + -0.02339971722360958, + 0.24718392801744182 + ], + "expected_cosine": 0.9908989880509481, + "feedback_advantage": 0.8742359692131677, + "id": "train-travel-2300061/frozen-warmstart-v2/seed-1707/step-1", + "kind": "admin", + "maximum_expectation_error": 1.1102230246251565e-16, + "mc4_error_norm_quantiles": { + "0.1": 0.16125994855008227, + "0.25": 0.16125994855008227, + "0.5": 0.16125994855008227, + "0.75": 0.16125994855008227, + "0.9": 0.8836320234642078 + }, + "mc4_gradient_norm_quantiles": { + "0.1": 0.1690639044388701, + "0.25": 0.1690639044388701, + "0.5": 0.1690639044388701, + "0.75": 0.1690639044388701, + "0.9": 1.21196511763921 + }, + "mc4_gradient_rmse": 0.40438201303090915, + "noise_signal_ratio": 1.224987186634642, + "optimal_choices": [ + "A" + ], + "p": [ + 0.03706735670419262, + 0.0012683752901585627, + 0.005684463192777227, + 0.9559798048128716 + ], + "probability_all_four_same": 0.8352103232416813, + "probability_no_optimal_in_four": 0.8597726740754791, + "probability_opposed_gradient": 0.0, + "q": [ + 0.9113033259173603, + 0.012999043498141936, + 0.024285410775655283, + 0.05141221980884245 + ], + "reverse_kl": 2.6642948481945097, + "student_entropy_nats": 0.20302326372841561, + "student_max_mass": 0.9559798048128716, + "student_optimal_mass": 0.03706735670419262, + "student_predicted_action_mass": 0.9559798048128716, + "teacher_entropy_nats": 0.3839701550463422, + "teacher_mass_on_student_prediction": 0.05141221980884245, + "teacher_max_mass": 0.9113033259173603, + "teacher_optimal_mass": 0.9113033259173603, + "variance_verification_error": 5.551115123125783e-17 + }, + { + "current_model_mistake": false, + "enumerated_mean_logit_gradient": [ + -0.00046075780526193657, + 2.1256691350319787e-05, + -3.895502041279576e-06, + 0.0004433966159528963 + ], + "episode_id": "train-travel-2300095", + "exact_gradient_norm": 0.0006398166424934513, + "exact_logit_gradient": [ + -0.0004607578052619363, + 2.1256691350319784e-05, + -3.8955020412795695e-06, + 0.00044339661595289615 + ], + "expected_cosine": -0.1582396744028012, + "feedback_advantage": -0.0006635550464872964, + "id": "train-travel-2300095/frozen-warmstart-v2/seed-1707/step-0", + "kind": "focus", + "maximum_expectation_error": 2.710505431213761e-19, + "mc4_error_norm_quantiles": { + "0.1": 0.0006399596905755878, + "0.25": 0.0006399596905755878, + "0.5": 0.0006399596905755878, + "0.75": 0.0006399596905755878, + "0.9": 0.0006399596905755878 + }, + "mc4_gradient_norm_quantiles": { + "0.1": 8.849501351732758e-07, + "0.25": 8.849501351732758e-07, + "0.5": 8.849501351732758e-07, + "0.75": 8.849501351732758e-07, + "0.9": 8.849501351732758e-07 + }, + "mc4_gradient_rmse": 0.021325235271241156, + "noise_signal_ratio": 33.33022909209403, + "optimal_choices": [ + "B" + ], + "p": [ + 0.00033510179607039865, + 0.9989243184135644, + 3.1169322135560576e-05, + 0.0007094104682295522 + ], + "probability_all_four_same": 0.9957042112224695, + "probability_no_optimal_in_four": 1.338859293925901e-12, + "probability_opposed_gradient": 0.9957042112222037, + "q": [ + 0.0013244727891037567, + 0.9982607633670771, + 3.529600756213568e-05, + 0.00037946783625699803 + ], + "reverse_kl": 0.0006432107324445856, + "student_entropy_nats": 0.009223682505918603, + "student_max_mass": 0.9989243184135644, + "student_optimal_mass": 0.9989243184135644, + "student_predicted_action_mass": 0.9989243184135644, + "teacher_entropy_nats": 0.013865476419217246, + "teacher_mass_on_student_prediction": 0.9982607633670771, + "teacher_max_mass": 0.9982607633670771, + "teacher_optimal_mass": 0.9982607633670771, + "variance_verification_error": 3.7947076036992655e-19 + }, + { + "current_model_mistake": true, + "enumerated_mean_logit_gradient": [ + -0.9115051747990522, + 0.27340505959962996, + 0.005172280798860781, + 0.6329278344005609 + ], + "episode_id": "train-travel-2300087", + "exact_gradient_norm": 1.1428982484664907, + "exact_logit_gradient": [ + -0.9115051747990515, + 0.27340505959963, + 0.0051722807988607775, + 0.6329278344005607 + ], + "expected_cosine": 0.7987610017752023, + "feedback_advantage": 0.6211089991567667, + "id": "train-travel-2300087/frozen-warmstart-v2/seed-1707/step-3", + "kind": "repair", + "maximum_expectation_error": 7.771561172376096e-16, + "mc4_error_norm_quantiles": { + "0.1": 0.04724346058423541, + "0.25": 0.2862833874109072, + "0.5": 0.9595347220612652, + "0.75": 1.2909705144103865, + "0.9": 1.574023008150914 + }, + "mc4_gradient_norm_quantiles": { + "0.1": 1.1648724388835834, + "0.25": 1.1772809269006619, + "0.5": 1.2479406110460471, + "0.75": 1.627463011645794, + "0.9": 2.045660834826144 + }, + "mc4_gradient_rmse": 1.0206900154934309, + "noise_signal_ratio": 0.8930716420844678, + "optimal_choices": [ + "A" + ], + "p": [ + 0.333165408421924, + 0.4277928487501772, + 0.010060723908471778, + 0.22898101891942702 + ], + "probability_all_four_same": 0.048561437198557746, + "probability_no_optimal_in_four": 0.19772996152412062, + "probability_opposed_gradient": 0.0, + "q": [ + 0.9542744075786906, + 0.04192789123961212, + 0.0011173404654364514, + 0.0026803607162607704 + ], + "reverse_kl": 1.6835816181515761, + "student_entropy_nats": 1.1132483915767553, + "student_max_mass": 0.4277928487501772, + "student_optimal_mass": 0.333165408421924, + "student_predicted_action_mass": 0.4277928487501772, + "teacher_entropy_nats": 0.2011178371516103, + "teacher_mass_on_student_prediction": 0.04192789123961212, + "teacher_max_mass": 0.9542744075786906, + "teacher_optimal_mass": 0.9542744075786906, + "variance_verification_error": 1.7763568394002505e-15 + }, + { + "current_model_mistake": true, + "enumerated_mean_logit_gradient": [ + -0.00013029206016445028, + 0.07466805310703951, + -1.3215518163797437e-05, + -0.07452454552871128 + ], + "episode_id": "train-travel-2300056", + "exact_gradient_norm": 0.10549522828890179, + "exact_logit_gradient": [ + -0.00013029206016445025, + 0.07466805310703954, + -1.3215518163797434e-05, + -0.07452454552871131 + ], + "expected_cosine": 0.9999399139989253, + "feedback_advantage": 0.7401649974569259, + "id": "train-travel-2300056/frozen-warmstart-v2/seed-1707/step-1", + "kind": "admin", + "maximum_expectation_error": 2.7755575615628914e-17, + "mc4_error_norm_quantiles": { + "0.1": 0.0777625498299529, + "0.25": 0.0777625498299529, + "0.5": 0.0777625498299529, + "0.75": 0.0777625498299529, + "0.9": 0.0777625498299529 + }, + "mc4_gradient_norm_quantiles": { + "0.1": 0.027732679325521644, + "0.25": 0.027732679325521644, + "0.5": 0.027732679325521644, + "0.75": 0.027732679325521644, + "0.9": 0.027732679325521644 + }, + "mc4_gradient_rmse": 0.32551778118851216, + "noise_signal_ratio": 3.0856161597856553, + "optimal_choices": [ + "D" + ], + "p": [ + 2.7148339725912717e-05, + 0.9859055017596253, + 4.163332394314917e-06, + 0.01406318656825449 + ], + "probability_all_four_same": 0.9448028151369326, + "probability_no_optimal_in_four": 0.9449228068533838, + "probability_opposed_gradient": 0.0, + "q": [ + 0.0008831104023762828, + 0.24486203794872544, + 2.6667623717960876e-05, + 0.7542281840251803 + ], + "reverse_kl": 1.3171300614058727, + "student_entropy_nats": 0.07429989081152509, + "student_max_mass": 0.9859055017596253, + "student_optimal_mass": 0.01406318656825449, + "student_predicted_action_mass": 0.9859055017596253, + "teacher_entropy_nats": 0.5637644594300629, + "teacher_mass_on_student_prediction": 0.24486203794872544, + "teacher_max_mass": 0.7542281840251803, + "teacher_optimal_mass": 0.7542281840251803, + "variance_verification_error": 1.249000902703301e-16 + }, + { + "current_model_mistake": true, + "enumerated_mean_logit_gradient": [ + -0.0029273164983299946, + -1.0046889296638577, + 1.0073616723213052, + 0.0002545738408828871 + ], + "episode_id": "train-travel-2300024", + "exact_gradient_norm": 1.4227389142920086, + "exact_logit_gradient": [ + -0.0029273164983299933, + -1.004688929663857, + 1.0073616723213041, + 0.00025457384088288716 + ], + "expected_cosine": 0.9992200000552189, + "feedback_advantage": 0.7149499331102416, + "id": "train-travel-2300024/frozen-warmstart-v2/seed-1707/step-0", + "kind": "focus", + "maximum_expectation_error": 1.1102230246251565e-15, + "mc4_error_norm_quantiles": { + "0.1": 0.006921276085145846, + "0.25": 0.006921276085145846, + "0.5": 0.022871692921727457, + "0.75": 0.03585846435468713, + "0.9": 0.03585846435468713 + }, + "mc4_gradient_norm_quantiles": { + "0.1": 1.3998882487561344, + "0.25": 1.3998882487561344, + "0.5": 1.4290000955939526, + "0.75": 1.4581181615655239, + "0.9": 1.4581181615655239 + }, + "mc4_gradient_rmse": 0.06678320383782932, + "noise_signal_ratio": 0.04693988697923705, + "optimal_choices": [ + "B" + ], + "p": [ + 0.0018081055511778854, + 0.26834665726961215, + 0.7294417943376518, + 0.0004034428415582176 + ], + "probability_all_four_same": 0.28830022962151497, + "probability_no_optimal_in_four": 0.28656387714792864, + "probability_opposed_gradient": 1.960221961052502e-08, + "q": [ + 0.0007912912442166093, + 0.9832965903798537, + 0.015893508990560995, + 1.8609385368639253e-05 + ], + "reverse_kl": 2.4453652882846066, + "student_entropy_nats": 0.5976967968211073, + "student_max_mass": 0.7294417943376518, + "student_optimal_mass": 0.26834665726961215, + "student_predicted_action_mass": 0.7294417943376518, + "teacher_entropy_nats": 0.08824553660486163, + "teacher_mass_on_student_prediction": 0.015893508990560995, + "teacher_max_mass": 0.9832965903798537, + "teacher_optimal_mass": 0.9832965903798537, + "variance_verification_error": 1.3010426069826053e-17 + }, + { + "current_model_mistake": false, + "enumerated_mean_logit_gradient": [ + -0.00013643325884501217, + 0.00823677007109376, + -0.015142287399805016, + 0.007041950587556252 + ], + "episode_id": "train-travel-2300034", + "exact_gradient_norm": 0.018620980935870775, + "exact_logit_gradient": [ + -0.0001364332588450122, + 0.008236770071093778, + -0.015142287399805016, + 0.007041950587556249 + ], + "expected_cosine": 0.9823782139197327, + "feedback_advantage": 0.007290978959020866, + "id": "train-travel-2300034/frozen-warmstart-v2/seed-1707/step-1", + "kind": "admin", + "maximum_expectation_error": 1.734723475976807e-17, + "mc4_error_norm_quantiles": { + "0.1": 0.018531350986190475, + "0.25": 0.018531350986190475, + "0.5": 0.018531350986190475, + "0.75": 0.018531350986190475, + "0.9": 0.018531350986190475 + }, + "mc4_gradient_norm_quantiles": { + "0.1": 9.07700231170888e-05, + "0.25": 9.07700231170888e-05, + "0.5": 9.07700231170888e-05, + "0.75": 9.07700231170888e-05, + "0.9": 9.07700231170888e-05 + }, + "mc4_gradient_rmse": 0.11351102422932814, + "noise_signal_ratio": 6.095867055567662, + "optimal_choices": [ + "C" + ], + "p": [ + 0.00010782730887425303, + 0.0066710386756537975, + 0.9900699583458206, + 0.003151175669651377 + ], + "probability_all_four_same": 0.9608675629130768, + "probability_no_optimal_in_four": 9.72308944723869e-09, + "probability_opposed_gradient": 0.00041865695415969293, + "q": [ + 0.000379125782093296, + 0.0019253595807784925, + 0.9973609373048414, + 0.0003345773322867307 + ], + "reverse_kl": 0.007957036841705815, + "student_entropy_nats": 0.06243805122901782, + "student_max_mass": 0.9900699583458206, + "student_optimal_mass": 0.9900699583458206, + "student_predicted_action_mass": 0.9900699583458206, + "teacher_entropy_nats": 0.02033828267841367, + "teacher_mass_on_student_prediction": 0.9973609373048414, + "teacher_max_mass": 0.9973609373048414, + "teacher_optimal_mass": 0.9973609373048414, + "variance_verification_error": 5.204170427930421e-18 + }, + { + "current_model_mistake": false, + "enumerated_mean_logit_gradient": [ + -6.883604383183073e-05, + 0.0008433705919435552, + -0.0006904436207139522, + -8.409092739777185e-05 + ], + "episode_id": "train-travel-2300007", + "exact_gradient_norm": 0.001095352013619754, + "exact_logit_gradient": [ + -6.883604383183076e-05, + 0.0008433705919435549, + -0.0006904436207139527, + -8.40909273977717e-05 + ], + "expected_cosine": 0.9985734223736494, + "feedback_advantage": -0.002539917932157021, + "id": "train-travel-2300007/frozen-warmstart-v2/seed-1707/step-1", + "kind": "admin", + "maximum_expectation_error": 4.336808689942018e-19, + "mc4_error_norm_quantiles": { + "0.1": 0.0010931858433079348, + "0.25": 0.0010931858433079348, + "0.5": 0.0010931858433079348, + "0.75": 0.0010931858433079348, + "0.9": 0.0010931858433079348 + }, + "mc4_gradient_norm_quantiles": { + "0.1": 2.168909882102839e-06, + "0.25": 2.168909882102839e-06, + "0.5": 2.168909882102839e-06, + "0.75": 2.168909882102839e-06, + "0.9": 2.168909882102839e-06 + }, + "mc4_gradient_rmse": 0.024861038884486163, + "noise_signal_ratio": 22.696848661764136, + "optimal_choices": [ + "B" + ], + "p": [ + 1.6690940181623504e-05, + 0.999355777405804, + 0.0005527280398056947, + 7.48036142087045e-05 + ], + "probability_all_four_same": 0.9974255986905185, + "probability_no_optimal_in_four": 1.7224388374193723e-13, + "probability_opposed_gradient": 0.0, + "q": [ + 0.0010300074928578336, + 0.9968158594736469, + 0.0019243073168641873, + 0.00022982571663101532 + ], + "reverse_kl": 0.0017008762333174901, + "student_entropy_nats": 0.005684125186589149, + "student_max_mass": 0.999355777405804, + "student_optimal_mass": 0.999355777405804, + "student_predicted_action_mass": 0.999355777405804, + "teacher_entropy_nats": 0.024222233288175958, + "teacher_mass_on_student_prediction": 0.9968158594736469, + "teacher_max_mass": 0.9968158594736469, + "teacher_optimal_mass": 0.9968158594736469, + "variance_verification_error": 1.0842021724855044e-18 + }, + { + "current_model_mistake": false, + "enumerated_mean_logit_gradient": [ + 0.6806349874375518, + 0.27905318567644655, + -0.8701601855791945, + -0.0895279875348041 + ], + "episode_id": "train-travel-2300053", + "exact_gradient_norm": 1.1429473634772267, + "exact_logit_gradient": [ + 0.680634987437553, + 0.27905318567644694, + -0.8701601855791956, + -0.08952798753480398 + ], + "expected_cosine": 0.8009990799801388, + "feedback_advantage": 0.5815802590217309, + "id": "train-travel-2300053/frozen-warmstart-v2/seed-1707/step-1", + "kind": "admin", + "maximum_expectation_error": 1.2212453270876722e-15, + "mc4_error_norm_quantiles": { + "0.1": 0.1713892623107099, + "0.25": 0.5398826440360164, + "0.5": 0.807118429374573, + "0.75": 1.140137892723027, + "0.9": 1.316159078569187 + }, + "mc4_gradient_norm_quantiles": { + "0.1": 0.8993787866251717, + "0.25": 1.0541382749352242, + "0.5": 1.2785029810691826, + "0.75": 1.8257645290740632, + "0.9": 1.9951677787076558 + }, + "mc4_gradient_rmse": 0.9731829810313191, + "noise_signal_ratio": 0.8514678909364402, + "optimal_choices": [ + "C" + ], + "p": [ + 0.26503734990843086, + 0.23389465409831242, + 0.34031470595852414, + 0.16075329003473252 + ], + "probability_all_four_same": 0.022007850403136513, + "probability_no_optimal_in_four": 0.18938571195124448, + "probability_opposed_gradient": 0.000667789337506519, + "q": [ + 0.0042692208543348474, + 0.01490104380557022, + 0.921894964980255, + 0.058934770359839954 + ], + "reverse_kl": 1.5603674552634061, + "student_entropy_nats": 1.3524192827810813, + "student_max_mass": 0.34031470595852414, + "student_optimal_mass": 0.34031470595852414, + "student_predicted_action_mass": 0.34031470595852414, + "teacher_entropy_nats": 0.3278084723299808, + "teacher_mass_on_student_prediction": 0.921894964980255, + "teacher_max_mass": 0.921894964980255, + "teacher_optimal_mass": 0.921894964980255, + "variance_verification_error": 1.1102230246251565e-16 + }, + { + "current_model_mistake": false, + "enumerated_mean_logit_gradient": [ + -8.596590617609031e-05, + -0.0007528697316410847, + -0.00011498068743906665, + 0.0009538163252562419 + ], + "episode_id": "train-travel-2300112", + "exact_gradient_norm": 0.0012235967925159315, + "exact_logit_gradient": [ + -8.596590617609028e-05, + -0.0007528697316410845, + -0.00011498068743906664, + 0.0009538163252562414 + ], + "expected_cosine": 0.9767278326297244, + "feedback_advantage": -0.0027851359889374816, + "id": "train-travel-2300112/frozen-warmstart-v2/seed-1707/step-0", + "kind": "focus", + "maximum_expectation_error": 4.336808689942018e-19, + "mc4_error_norm_quantiles": { + "0.1": 0.001221294289803069, + "0.25": 0.001221294289803069, + "0.5": 0.001221294289803069, + "0.75": 0.001221294289803069, + "0.9": 0.001221294289803069 + }, + "mc4_gradient_norm_quantiles": { + "0.1": 2.3566087693781574e-06, + "0.25": 2.3566087693781574e-06, + "0.5": 2.3566087693781574e-06, + "0.75": 2.3566087693781574e-06, + "0.9": 2.3566087693781574e-06 + }, + "mc4_gradient_rmse": 0.028607533448510445, + "noise_signal_ratio": 23.379869597147515, + "optimal_choices": [ + "D" + ], + "p": [ + 2.75174956903655e-05, + 0.0004304460382672396, + 0.00023040114616690028, + 0.9993116353198755 + ], + "probability_all_four_same": 0.9972493830506457, + "probability_no_optimal_in_four": 2.2452996807170252e-13, + "probability_opposed_gradient": 0.0, + "q": [ + 0.0006245497247072033, + 0.0024701423417332613, + 0.0003788086026215151, + 0.9965264993309381 + ], + "reverse_kl": 0.0018364722156335362, + "student_entropy_nats": 0.006243101817394599, + "student_max_mass": 0.9993116353198755, + "student_optimal_mass": 0.9993116353198755, + "student_predicted_action_mass": 0.9993116353198755, + "teacher_entropy_nats": 0.02588957319704802, + "teacher_mass_on_student_prediction": 0.9965264993309381, + "teacher_max_mass": 0.9965264993309381, + "teacher_optimal_mass": 0.9965264993309381, + "variance_verification_error": 1.0842021724855044e-19 + }, + { + "current_model_mistake": true, + "enumerated_mean_logit_gradient": [ + 0.2915103454943772, + -0.4244186426834809, + -0.044733394627627564, + 0.17764169181673117 + ], + "episode_id": "train-travel-2300006", + "exact_gradient_norm": 0.5465044492535815, + "exact_logit_gradient": [ + 0.29151034549437693, + -0.4244186426834805, + -0.04473339462762757, + 0.17764169181673117 + ], + "expected_cosine": 0.8446599043655213, + "feedback_advantage": 0.34920825878864575, + "id": "train-travel-2300006/frozen-warmstart-v2/seed-1707/step-3", + "kind": "repair", + "maximum_expectation_error": 3.885780586188048e-16, + "mc4_error_norm_quantiles": { + "0.1": 0.11002833851598678, + "0.25": 0.11002833851598678, + "0.5": 0.3226932680803967, + "0.75": 0.399769006889043, + "0.9": 0.6033130291399741 + }, + "mc4_gradient_norm_quantiles": { + "0.1": 0.46227682000343145, + "0.25": 0.5465084576293947, + "0.5": 0.6825516666787217, + "0.75": 0.7364642405149499, + "0.9": 0.7960319389652373 + }, + "mc4_gradient_rmse": 0.3949921863463617, + "noise_signal_ratio": 0.7227611538860186, + "optimal_choices": [ + "B", + "C", + "D" + ], + "p": [ + 0.5103194031997125, + 0.2731543137535661, + 0.028790251800581316, + 0.18773603124614008 + ], + "probability_all_four_same": 0.07463167150841804, + "probability_no_optimal_in_four": 0.06782164589178531, + "probability_opposed_gradient": 0.0, + "q": [ + 0.1611111444110668, + 0.722050062693096, + 0.07610351507250017, + 0.040735277823337045 + ], + "reverse_kl": 0.5817111840679332, + "student_entropy_nats": 1.113947883248574, + "student_max_mass": 0.5103194031997125, + "student_optimal_mass": 0.48968059680028747, + "student_predicted_action_mass": 0.5103194031997125, + "teacher_entropy_nats": 0.8556743554401491, + "teacher_mass_on_student_prediction": 0.1611111444110668, + "teacher_max_mass": 0.722050062693096, + "teacher_optimal_mass": 0.8388888555889332, + "variance_verification_error": 2.7755575615628914e-16 + }, + { + "current_model_mistake": false, + "enumerated_mean_logit_gradient": [ + -0.001114698848221802, + 0.005639785482411302, + -0.003735447276340494, + -0.000789639357848995 + ], + "episode_id": "train-travel-2300046", + "exact_gradient_norm": 0.006901219492207733, + "exact_logit_gradient": [ + -0.0011146988482218043, + 0.005639785482411291, + -0.0037354472763404937, + -0.0007896393578489941 + ], + "expected_cosine": 0.9561783468527061, + "feedback_advantage": -0.019234405787590614, + "id": "train-travel-2300046/frozen-warmstart-v2/seed-1707/step-1", + "kind": "admin", + "maximum_expectation_error": 1.1275702593849246e-17, + "mc4_error_norm_quantiles": { + "0.1": 0.006816516982537701, + "0.25": 0.006816516982537701, + "0.5": 0.006816516982537701, + "0.75": 0.006816516982537701, + "0.9": 0.006816516982537701 + }, + "mc4_gradient_norm_quantiles": { + "0.1": 8.83861385929355e-05, + "0.25": 8.83861385929355e-05, + "0.5": 8.83861385929355e-05, + "0.75": 8.83861385929355e-05, + "0.9": 8.83861385929355e-05 + }, + "mc4_gradient_rmse": 0.0727919557775121, + "noise_signal_ratio": 10.547694629869772, + "optimal_choices": [ + "B" + ], + "p": [ + 0.0014975705192846, + 0.996096536517329, + 0.0014975705192846, + 0.0009083224441018614 + ], + "probability_all_four_same": 0.9844773305664701, + "probability_no_optimal_in_four": 2.321669966912193e-10, + "probability_opposed_gradient": 0.0, + "q": [ + 0.0031091379875147008, + 0.9768621307297384, + 0.017891854014449093, + 0.0021368772682978264 + ], + "reverse_kl": 0.013836763480656485, + "student_entropy_nats": 0.029737775870523214, + "student_max_mass": 0.996096536517329, + "student_optimal_mass": 0.996096536517329, + "student_predicted_action_mass": 0.996096536517329, + "teacher_entropy_nats": 0.12594308459004566, + "teacher_mass_on_student_prediction": 0.9768621307297384, + "teacher_max_mass": 0.9768621307297384, + "teacher_optimal_mass": 0.9768621307297384, + "variance_verification_error": 3.469446951953614e-18 + } + ], + "status": "completed", + "summary": { + "all": { + "count": 96, + "metrics": { + "exact_gradient_norm": { + "count": 96, + "maximum": 2.21876536836621, + "mean": 0.33082494821223496, + "minimum": 7.143257873583472e-05, + "quantiles": { + "0.1": 0.001459299072697043, + "0.25": 0.008138050839312667, + "0.5": 0.10962902905463112, + "0.75": 0.4896770499655213, + "0.9": 1.0262347259092994 + } + }, + "expected_cosine": { + "count": 96, + "maximum": 0.9999995195428689, + "mean": 0.8348410296868836, + "minimum": -0.9484452557914791, + "quantiles": { + "0.1": 0.6024594783243811, + "0.25": 0.8981490292927372, + "0.5": 0.9771960008680007, + "0.75": 0.9949076575801044, + "0.9": 0.999309677298186 + } + }, + "feedback_advantage": { + "count": 96, + "maximum": 0.9392785316232912, + "mean": 0.24884076504679128, + "minimum": -0.04318017131666618, + "quantiles": { + "0.1": -0.0007767617975634189, + "0.25": 0.002091756243564752, + "0.5": 0.048016436814155516, + "0.75": 0.4719743576235922, + "0.9": 0.7686742759614602 + } + }, + "mc4_gradient_rmse": { + "count": 96, + "maximum": 1.1569823100514236, + "mean": 0.3120661285223722, + "minimum": 0.008872383045635588, + "quantiles": { + "0.1": 0.027045291094386206, + "0.25": 0.08281696386181849, + "0.5": 0.2541793960624671, + "0.75": 0.42863806909289587, + "0.9": 0.6953675997104107 + } + }, + "noise_signal_ratio": { + "count": 96, + "maximum": 124.20639437428972, + "mean": 8.74506546631662, + "minimum": 0.04693988697923705, + "quantiles": { + "0.1": 0.5674414263253338, + "0.25": 0.9631875041314668, + "0.5": 2.841767106230266, + "0.75": 7.809977127832285, + "0.9": 17.666731680980465 + } + }, + "probability_all_four_same": { + "count": 96, + "maximum": 0.9997074680348393, + "mean": 0.6853491713021231, + "minimum": 0.019304849556107288, + "quantiles": { + "0.1": 0.1338132038937452, + "0.25": 0.33624506420585365, + "0.5": 0.8978806060702684, + "0.75": 0.9752790091145427, + "0.9": 0.9949885534322831 + } + }, + "probability_no_optimal_in_four": { + "count": 96, + "maximum": 0.9913022061045822, + "mean": 0.1873585057142297, + "minimum": 2.5828120064841875e-17, + "quantiles": { + "0.1": 1.189474241721508e-12, + "0.25": 1.5154466814698814e-09, + "0.5": 2.4673180094528973e-05, + "0.75": 0.2756283414153459, + "0.9": 0.7901568220980056 + } + }, + "probability_opposed_gradient": { + "count": 96, + "maximum": 0.998585742524318, + "mean": 0.06766391024951433, + "minimum": 0.0, + "quantiles": { + "0.1": 0.0, + "0.25": 0.0, + "0.5": 5.491710401095274e-05, + "0.75": 0.0022524169252060894, + "0.9": 0.028022560437337127 + } + }, + "reverse_kl": { + "count": 96, + "maximum": 3.5071095201579405, + "mean": 0.6218334224574268, + "minimum": 0.00015561566512250187, + "quantiles": { + "0.1": 0.0012445015645160535, + "0.25": 0.004824712637774824, + "0.5": 0.0885711157223523, + "0.75": 1.0335672935365792, + "0.9": 2.210135292358989 + } + }, + "student_entropy_nats": { + "count": 96, + "maximum": 1.3670711533275415, + "mean": 0.3515898874130362, + "minimum": 0.0008277574565303808, + "quantiles": { + "0.1": 0.010381247287096166, + "0.25": 0.04143791724176751, + "0.5": 0.13135606497753066, + "0.75": 0.6556388776317219, + "0.9": 1.007450360529821 + } + }, + "student_max_mass": { + "count": 96, + "maximum": 0.9999268589846889, + "mean": 0.8653542619621928, + "minimum": 0.31483005069955955, + "quantiles": { + "0.1": 0.5872419135568585, + "0.25": 0.7601868059257746, + "0.5": 0.9734294575795399, + "0.75": 0.9937616127211043, + "0.9": 0.9987447285179663 + } + }, + "student_optimal_mass": { + "count": 96, + "maximum": 0.999928710888966, + "mean": 0.6784077964095541, + "minimum": 0.002181577014188315, + "quantiles": { + "0.1": 0.05733804805541564, + "0.25": 0.27546407590445904, + "0.5": 0.9296813717162817, + "0.75": 0.9937616127211043, + "0.9": 0.9989572216783345 + } + }, + "student_predicted_action_mass": { + "count": 96, + "maximum": 0.9999268589846889, + "mean": 0.8653542619621928, + "minimum": 0.31483005069955955, + "quantiles": { + "0.1": 0.5872419135568585, + "0.25": 0.7601868059257746, + "0.5": 0.9734294575795399, + "0.75": 0.9937616127211043, + "0.9": 0.9987447285179663 + } + }, + "teacher_entropy_nats": { + "count": 96, + "maximum": 1.083180877687151, + "mean": 0.20908336337336503, + "minimum": 0.003398768269916631, + "quantiles": { + "0.1": 0.008417118782785785, + "0.25": 0.015620364937655336, + "0.5": 0.10693597464703913, + "0.75": 0.3047139195019964, + "0.9": 0.6152120404821184 + } + }, + "teacher_mass_on_student_prediction": { + "count": 96, + "maximum": 0.9996519695554565, + "mean": 0.735213684703471, + "minimum": 0.013741471983846842, + "quantiles": { + "0.1": 0.05307525984853627, + "0.25": 0.32377260912416217, + "0.5": 0.9803012828635473, + "0.75": 0.9980786797097425, + "0.9": 0.9990556722624153 + } + }, + "teacher_max_mass": { + "count": 96, + "maximum": 0.9996519695554565, + "mean": 0.9324043667424093, + "minimum": 0.46057752289523396, + "quantiles": { + "0.1": 0.7803168508732913, + "0.25": 0.9212351292591369, + "0.5": 0.981064661379917, + "0.75": 0.9980786797097425, + "0.9": 0.9990556722624153 + } + }, + "teacher_optimal_mass": { + "count": 96, + "maximum": 0.9996851497012124, + "mean": 0.9272485614563453, + "minimum": 0.09529349369827628, + "quantiles": { + "0.1": 0.8070088574892427, + "0.25": 0.9219929741596351, + "0.5": 0.9820081859817698, + "0.75": 0.9981466018011174, + "0.9": 0.9991031857639157 + } + } + }, + "noise_signal_above_one": 71, + "student_max_above_99_percent": 32, + "teacher_optimal_above_99_percent": 41 + }, + "by_kind": { + "admin": { + "count": 26, + "metrics": { + "exact_gradient_norm": { + "count": 26, + "maximum": 1.1429473634772267, + "mean": 0.19743848371185474, + "minimum": 7.143257873583472e-05, + "quantiles": { + "0.1": 0.00459564297491319, + "0.25": 0.017531342351446728, + "0.5": 0.042364392508699444, + "0.75": 0.22105593148121777, + "0.9": 0.6002948605450167 + } + }, + "expected_cosine": { + "count": 26, + "maximum": 0.999948665013913, + "mean": 0.9296232361958978, + "minimum": 0.38496892935814414, + "quantiles": { + "0.1": 0.7742421347707912, + "0.25": 0.9499955441126333, + "0.5": 0.9839241815372148, + "0.75": 0.9989290933106609, + "0.9": 0.9997858662130374 + } + }, + "feedback_advantage": { + "count": 26, + "maximum": 0.9392785316232912, + "mean": 0.2642229119134767, + "minimum": -0.019234405787590614, + "quantiles": { + "0.1": -0.0016584993375246548, + "0.25": 0.003288859809124217, + "0.5": 0.02080552267903968, + "0.75": 0.6731379304771454, + "0.9": 0.8592453575529395 + } + }, + "mc4_gradient_rmse": { + "count": 26, + "maximum": 0.9921821205243262, + "mean": 0.284412145336261, + "minimum": 0.008872383045635588, + "quantiles": { + "0.1": 0.03231080511351728, + "0.25": 0.102152679416089, + "0.5": 0.2114003062979991, + "0.75": 0.38000300872028614, + "0.9": 0.5950109280335205 + } + }, + "noise_signal_ratio": { + "count": 26, + "maximum": 124.20639437428972, + "mean": 9.427978193192184, + "minimum": 0.5025544740649898, + "quantiles": { + "0.1": 1.0412226293147044, + "0.25": 1.7777708033872948, + "0.5": 4.208105454396685, + "0.75": 6.35153075586591, + "0.9": 9.821224562498717 + } + }, + "probability_all_four_same": { + "count": 26, + "maximum": 0.9996248759738765, + "mean": 0.81679139428266, + "minimum": 0.022007850403136513, + "quantiles": { + "0.1": 0.40374864457818094, + "0.25": 0.8422037041377395, + "0.5": 0.9225076831417198, + "0.75": 0.9680203094686745, + "0.9": 0.9868533252505246 + } + }, + "probability_no_optimal_in_four": { + "count": 26, + "maximum": 0.9913022061045822, + "mean": 0.23148590602217214, + "minimum": 7.739340711993441e-17, + "quantiles": { + "0.1": 1.1616962028748061e-10, + "0.25": 8.637191083368504e-09, + "0.5": 6.37668239775394e-07, + "0.75": 0.4431690707629379, + "0.9": 0.916740390058563 + } + }, + "probability_opposed_gradient": { + "count": 26, + "maximum": 0.14027784193617954, + "mean": 0.008096761351976423, + "minimum": 0.0, + "quantiles": { + "0.1": 0.0, + "0.25": 0.0, + "0.5": 3.097964613504823e-05, + "0.75": 0.0006055062416698124, + "0.9": 0.01330880901466219 + } + }, + "reverse_kl": { + "count": 26, + "maximum": 3.5071095201579405, + "mean": 0.6866100453704946, + "minimum": 0.00031842904830781647, + "quantiles": { + "0.1": 0.0019315374791238678, + "0.25": 0.00807798301194476, + "0.5": 0.0622657927124185, + "0.75": 1.2953782258644817, + "0.9": 2.095080093047337 + } + }, + "student_entropy_nats": { + "count": 26, + "maximum": 1.3524192827810813, + "mean": 0.22258622595175867, + "minimum": 0.0010398093975801037, + "quantiles": { + "0.1": 0.024940154923392667, + "0.25": 0.048196776056298365, + "0.5": 0.11006437342525816, + "0.75": 0.19675785061852216, + "0.9": 0.6079120525431085 + } + }, + "student_max_mass": { + "count": 26, + "maximum": 0.9999062057982658, + "mean": 0.9278166863207137, + "minimum": 0.34031470595852414, + "quantiles": { + "0.1": 0.7900168040613663, + "0.25": 0.957956388806574, + "0.5": 0.98001399649289, + "0.75": 0.9919072272622899, + "0.9": 0.9966964609732374 + } + }, + "student_optimal_mass": { + "count": 26, + "maximum": 0.9999062057982658, + "mean": 0.6936476827562635, + "minimum": 0.002181577014188315, + "quantiles": { + "0.1": 0.02158501261488454, + "0.25": 0.19582813279834765, + "0.5": 0.9718195194496455, + "0.75": 0.9903641932685208, + "0.9": 0.9977261569615665 + } + }, + "student_predicted_action_mass": { + "count": 26, + "maximum": 0.9999062057982658, + "mean": 0.9278166863207137, + "minimum": 0.34031470595852414, + "quantiles": { + "0.1": 0.7900168040613663, + "0.25": 0.957956388806574, + "0.5": 0.98001399649289, + "0.75": 0.9919072272622899, + "0.9": 0.9966964609732374 + } + }, + "teacher_entropy_nats": { + "count": 26, + "maximum": 0.6369098048604879, + "mean": 0.17655899387791257, + "minimum": 0.003398768269916631, + "quantiles": { + "0.1": 0.009075703417233356, + "0.25": 0.028777228801804788, + "0.5": 0.09885610841788992, + "0.75": 0.3256338846393167, + "0.9": 0.44234210065844515 + } + }, + "teacher_mass_on_student_prediction": { + "count": 26, + "maximum": 0.9996519695554565, + "mean": 0.739219907092441, + "minimum": 0.02277424166463781, + "quantiles": { + "0.1": 0.08037301602525021, + "0.25": 0.35323216804859836, + "0.5": 0.9830665048864158, + "0.75": 0.9961725025344798, + "0.9": 0.9989157069257851 + } + }, + "teacher_max_mass": { + "count": 26, + "maximum": 0.9996519695554565, + "mean": 0.9455465048901911, + "minimum": 0.678342558348217, + "quantiles": { + "0.1": 0.8488797094737361, + "0.25": 0.913951235683084, + "0.5": 0.9830665048864158, + "0.75": 0.9961725025344798, + "0.9": 0.9989157069257851 + } + }, + "teacher_optimal_mass": { + "count": 26, + "maximum": 0.9996519695554565, + "mean": 0.9578705946697402, + "minimum": 0.7542281840251803, + "quantiles": { + "0.1": 0.8773893101625243, + "0.25": 0.9335175663063927, + "0.5": 0.9860306213362788, + "0.75": 0.9968076394728833, + "0.9": 0.9991311707481068 + } + } + }, + "noise_signal_above_one": 23, + "student_max_above_99_percent": 10, + "teacher_optimal_above_99_percent": 11 + }, + "exercise": { + "count": 22, + "metrics": { + "exact_gradient_norm": { + "count": 22, + "maximum": 2.21876536836621, + "mean": 0.35083036707542986, + "minimum": 0.00023083518720804316, + "quantiles": { + "0.1": 0.0010640210954003459, + "0.25": 0.006283102085751696, + "0.5": 0.05676611314157673, + "0.75": 0.3784973912973172, + "0.9": 1.0772424845557937 + } + }, + "expected_cosine": { + "count": 22, + "maximum": 0.9999995195428689, + "mean": 0.8472001597015618, + "minimum": -0.6968715146506224, + "quantiles": { + "0.1": 0.8595448859610411, + "0.25": 0.9533766816571114, + "0.5": 0.9932503535791002, + "0.75": 0.9980253867057084, + "0.9": 0.9993212929695457 + } + }, + "feedback_advantage": { + "count": 22, + "maximum": 0.9315916123282266, + "mean": 0.20000048388493757, + "minimum": -0.0036952708053559213, + "quantiles": { + "0.1": -0.0013882758303559227, + "0.25": 0.0010811831467262556, + "0.5": 0.014525219809437717, + "0.75": 0.2570052284203933, + "0.9": 0.7678478455544138 + } + }, + "mc4_gradient_rmse": { + "count": 22, + "maximum": 1.1569823100514236, + "mean": 0.29732088360384296, + "minimum": 0.013259854701317507, + "quantiles": { + "0.1": 0.02597452401796258, + "0.25": 0.032718232789899805, + "0.5": 0.16731519633963227, + "0.75": 0.37591763557292024, + "0.9": 0.6956637887304848 + } + }, + "noise_signal_ratio": { + "count": 22, + "maximum": 67.65509994222825, + "mean": 11.113427186197216, + "minimum": 0.521453203906535, + "quantiles": { + "0.1": 0.6445031026511435, + "0.25": 0.9175822649287225, + "0.5": 3.7901897016559074, + "0.75": 10.721882588525915, + "0.9": 29.76215807800586 + } + }, + "probability_all_four_same": { + "count": 22, + "maximum": 0.9997074680348393, + "mean": 0.7298926654560643, + "minimum": 0.07533198275765679, + "quantiles": { + "0.1": 0.21834210611665902, + "0.25": 0.3980821141699644, + "0.5": 0.9173055759259936, + "0.75": 0.9851656511586171, + "0.9": 0.9975016226497723 + } + }, + "probability_no_optimal_in_four": { + "count": 22, + "maximum": 0.9212308377416744, + "mean": 0.11030112880009031, + "minimum": 2.8618307045447777e-17, + "quantiles": { + "0.1": 2.0759911316528531e-13, + "0.25": 3.9291727549998713e-10, + "0.5": 2.4685631311508176e-07, + "0.75": 0.0067645687897100615, + "0.9": 0.4313309094279131 + } + }, + "probability_opposed_gradient": { + "count": 22, + "maximum": 0.9976721790841283, + "mean": 0.09104738180025941, + "minimum": 0.0, + "quantiles": { + "0.1": 0.0, + "0.25": 1.5672935618364423e-16, + "0.5": 0.0005390423558867025, + "0.75": 0.0028099453034847862, + "0.9": 0.010936451175422794 + } + }, + "reverse_kl": { + "count": 22, + "maximum": 2.884345124497957, + "mean": 0.6051813638404083, + "minimum": 0.00037172897171703154, + "quantiles": { + "0.1": 0.000734869697871194, + "0.25": 0.0021093353331060813, + "0.5": 0.026276579006437754, + "0.75": 0.6155644005212351, + "0.9": 2.567813973900549 + } + }, + "student_entropy_nats": { + "count": 22, + "maximum": 1.0607340561901464, + "mean": 0.2788737386391365, + "minimum": 0.0008277574565303808, + "quantiles": { + "0.1": 0.00559893992647396, + "0.25": 0.02650803019916069, + "0.5": 0.10740475343003411, + "0.75": 0.5948546621296704, + "0.9": 0.6964728357336588 + } + }, + "student_max_mass": { + "count": 22, + "maximum": 0.9999268589846889, + "mean": 0.8904595775080327, + "minimum": 0.46527094342218256, + "quantiles": { + "0.1": 0.6821513004933331, + "0.25": 0.7940312608123944, + "0.5": 0.9786521603372722, + "0.75": 0.996267902980639, + "0.9": 0.9993747949593551 + } + }, + "student_optimal_mass": { + "count": 22, + "maximum": 0.9999268589846889, + "mean": 0.7804716296441621, + "minimum": 0.02030223602778511, + "quantiles": { + "0.1": 0.20126890705716624, + "0.25": 0.7165816310565198, + "0.5": 0.9778182110978553, + "0.75": 0.996267902980639, + "0.9": 0.9993747949593551 + } + }, + "student_predicted_action_mass": { + "count": 22, + "maximum": 0.9999268589846889, + "mean": 0.8904595775080327, + "minimum": 0.46527094342218256, + "quantiles": { + "0.1": 0.6821513004933331, + "0.25": 0.7940312608123944, + "0.5": 0.9786521603372722, + "0.75": 0.996267902980639, + "0.9": 0.9993747949593551 + } + }, + "teacher_entropy_nats": { + "count": 22, + "maximum": 0.2999151854802204, + "mean": 0.09178542362435549, + "minimum": 0.005085022087700366, + "quantiles": { + "0.1": 0.008964731458943586, + "0.25": 0.011570923833916594, + "0.5": 0.06620913769136663, + "0.75": 0.15331253678103365, + "0.9": 0.22663126424380436 + } + }, + "teacher_mass_on_student_prediction": { + "count": 22, + "maximum": 0.9994547673317212, + "mean": 0.8561179318506393, + "minimum": 0.015674228660255238, + "quantiles": { + "0.1": 0.14530374494838685, + "0.25": 0.9676755217566368, + "0.5": 0.9895929090970002, + "0.75": 0.998646061823477, + "0.9": 0.9989920942752792 + } + }, + "teacher_max_mass": { + "count": 22, + "maximum": 0.9994547673317212, + "mean": 0.9804721135290997, + "minimum": 0.913851086512203, + "quantiles": { + "0.1": 0.9474126168621431, + "0.25": 0.9722763243885371, + "0.5": 0.9895929090970002, + "0.75": 0.998646061823477, + "0.9": 0.9989920942752792 + } + }, + "teacher_optimal_mass": { + "count": 22, + "maximum": 0.9994547673317212, + "mean": 0.9804721135290997, + "minimum": 0.913851086512203, + "quantiles": { + "0.1": 0.9474126168621431, + "0.25": 0.9722763243885371, + "0.5": 0.9895929090970002, + "0.75": 0.998646061823477, + "0.9": 0.9989920942752792 + } + } + }, + "noise_signal_above_one": 16, + "student_max_above_99_percent": 7, + "teacher_optimal_above_99_percent": 11 + }, + "focus": { + "count": 28, + "metrics": { + "exact_gradient_norm": { + "count": 28, + "maximum": 1.7441213767770418, + "mean": 0.23513122451078794, + "minimum": 0.000118302982859311, + "quantiles": { + "0.1": 0.0008236306013408, + "0.25": 0.004837668960358297, + "0.5": 0.017706891093379702, + "0.75": 0.25139299483586686, + "0.9": 0.6403942053713202 + } + }, + "expected_cosine": { + "count": 28, + "maximum": 0.9995330539863804, + "mean": 0.7230750363582332, + "minimum": -0.9484452557914791, + "quantiles": { + "0.1": -0.28879561327679026, + "0.25": 0.9396105068864556, + "0.5": 0.9799451817610532, + "0.75": 0.9923374183890024, + "0.9": 0.9989658159714417 + } + }, + "feedback_advantage": { + "count": 28, + "maximum": 0.8990556412909598, + "mean": 0.17148717486281145, + "minimum": -0.04318017131666618, + "quantiles": { + "0.1": -0.0013000293292223515, + "0.25": 0.00031325140642826654, + "0.5": 0.007654126775320791, + "0.75": 0.21220883921304456, + "0.9": 0.6735768891167587 + } + }, + "mc4_gradient_rmse": { + "count": 28, + "maximum": 0.9591772191018071, + "mean": 0.20079270573084998, + "minimum": 0.009562628254256763, + "quantiles": { + "0.1": 0.02185231090939673, + "0.25": 0.061293174564217866, + "0.5": 0.09848469743062635, + "0.75": 0.326029390669666, + "0.9": 0.4171290600801488 + } + }, + "noise_signal_ratio": { + "count": 28, + "maximum": 80.83167493442572, + "mean": 11.738119412489414, + "minimum": 0.04693988697923705, + "quantiles": { + "0.1": 0.7318818274889651, + "0.25": 1.3697849385903678, + "0.5": 6.269032616942043, + "0.75": 14.51940980817522, + "0.9": 26.364977445631474 + } + }, + "probability_all_four_same": { + "count": 28, + "maximum": 0.9992794817096153, + "mean": 0.8248275930048489, + "minimum": 0.10830288015167715, + "quantiles": { + "0.1": 0.391590015244351, + "0.25": 0.8029658302906556, + "0.5": 0.9686052433303368, + "0.75": 0.9889718430428363, + "0.9": 0.9961677627709224 + } + }, + "probability_no_optimal_in_four": { + "count": 28, + "maximum": 0.9661573331402065, + "mean": 0.16508602466728325, + "minimum": 2.5828120064841875e-17, + "quantiles": { + "0.1": 1.0045604961696418e-12, + "0.25": 3.3907520631863294e-11, + "0.5": 2.425292254282373e-08, + "0.75": 0.033227904564665685, + "0.9": 0.776839170508033 + } + }, + "probability_opposed_gradient": { + "count": 28, + "maximum": 0.998585742524318, + "mean": 0.14341201749000138, + "minimum": 0.0, + "quantiles": { + "0.1": 0.0, + "0.25": 0.0, + "0.5": 3.170231550101465e-08, + "0.75": 0.003350489946109388, + "0.9": 0.9903624696873166 + } + }, + "reverse_kl": { + "count": 28, + "maximum": 2.601679608132438, + "mean": 0.42631312985964814, + "minimum": 0.00015561566512250187, + "quantiles": { + "0.1": 0.0005564621980614921, + "0.25": 0.0024161035952162124, + "0.5": 0.01621085053023634, + "0.75": 0.2606505882471307, + "0.9": 1.7916095584239886 + } + }, + "student_entropy_nats": { + "count": 28, + "maximum": 0.8454925142348215, + "mean": 0.1757872530901954, + "minimum": 0.00188838427096119, + "quantiles": { + "0.1": 0.008329508299361403, + "0.25": 0.02192480432242073, + "0.5": 0.05054889607688448, + "0.75": 0.23889146846597198, + "0.9": 0.5829116471380377 + } + }, + "student_max_mass": { + "count": 28, + "maximum": 0.9998198217369432, + "mean": 0.9321841676986331, + "minimum": 0.48239445326928454, + "quantiles": { + "0.1": 0.787385461224977, + "0.25": 0.9464642550844814, + "0.5": 0.9920539003968767, + "0.75": 0.9972314557949782, + "0.9": 0.9990405134854577 + } + }, + "student_optimal_mass": { + "count": 28, + "maximum": 0.999928710888966, + "mean": 0.752485352194734, + "minimum": 0.008570211378869291, + "quantiles": { + "0.1": 0.06131226417058104, + "0.25": 0.5872313323909502, + "0.5": 0.9894109225226314, + "0.75": 0.9976277579531931, + "0.9": 0.9990405134854577 + } + }, + "student_predicted_action_mass": { + "count": 28, + "maximum": 0.9998198217369432, + "mean": 0.9321841676986331, + "minimum": 0.48239445326928454, + "quantiles": { + "0.1": 0.787385461224977, + "0.25": 0.9464642550844814, + "0.5": 0.9920539003968767, + "0.75": 0.9972314557949782, + "0.9": 0.9990405134854577 + } + }, + "teacher_entropy_nats": { + "count": 28, + "maximum": 0.6968533099221963, + "mean": 0.11643630863269767, + "minimum": 0.0036266898682119097, + "quantiles": { + "0.1": 0.005378569146829592, + "0.25": 0.01075976111032447, + "0.5": 0.02428575689304125, + "0.75": 0.1639974356906035, + "0.9": 0.2759480841673956 + } + }, + "teacher_mass_on_student_prediction": { + "count": 28, + "maximum": 0.999619118355799, + "mean": 0.8147564116610814, + "minimum": 0.015893508990560995, + "quantiles": { + "0.1": 0.05332402129088806, + "0.25": 0.9390458722362622, + "0.5": 0.9966170904425751, + "0.75": 0.9987106823414945, + "0.9": 0.9994057918929791 + } + }, + "teacher_max_mass": { + "count": 28, + "maximum": 0.999619118355799, + "mean": 0.9572661854370549, + "minimum": 0.5612744259070631, + "quantiles": { + "0.1": 0.9287661897794444, + "0.25": 0.9639860379655105, + "0.5": 0.9966170904425751, + "0.75": 0.9987106823414945, + "0.9": 0.9994057918929791 + } + }, + "teacher_optimal_mass": { + "count": 28, + "maximum": 0.9996851497012124, + "mean": 0.9239725270575454, + "minimum": 0.09529349369827628, + "quantiles": { + "0.1": 0.859031352812603, + "0.25": 0.9639860379655105, + "0.5": 0.9966170904425751, + "0.75": 0.9988297732898499, + "0.9": 0.9994057918929791 + } + } + }, + "noise_signal_above_one": 23, + "student_max_above_99_percent": 15, + "teacher_optimal_above_99_percent": 18 + }, + "repair": { + "count": 20, + "metrics": { + "exact_gradient_norm": { + "count": 20, + "maximum": 1.259970215034505, + "mean": 0.6161926044952407, + "minimum": 0.06738669601196012, + "quantiles": { + "0.1": 0.21526678544334027, + "0.25": 0.3496277314787676, + "0.5": 0.5515368778030085, + "0.75": 0.8120596386119361, + "0.9": 1.1528464744452334 + } + }, + "expected_cosine": { + "count": 20, + "maximum": 0.9922025808080299, + "mean": 0.8545015088691293, + "minimum": 0.432585736985595, + "quantiles": { + "0.1": 0.7099387310586085, + "0.25": 0.7966682550841208, + "0.5": 0.89793515440674, + "0.75": 0.9578971279682085, + "0.9": 0.9859811088925313 + } + }, + "feedback_advantage": { + "count": 20, + "maximum": 0.7349780776709811, + "mean": 0.39086330965571103, + "minimum": 0.012649863747963508, + "quantiles": { + "0.1": 0.048350784365662314, + "0.25": 0.17705732809388133, + "0.5": 0.36635094598961704, + "0.75": 0.6282157800943774, + "0.9": 0.6761744445210272 + } + }, + "mc4_gradient_rmse": { + "count": 20, + "maximum": 1.0206900154934309, + "mean": 0.52001886798283, + "minimum": 0.1522759442940499, + "quantiles": { + "0.1": 0.21924744367916718, + "0.25": 0.3425175466250486, + "0.5": 0.47788247200634737, + "0.75": 0.6879099877491022, + "0.9": 0.9398706888468901 + } + }, + "noise_signal_ratio": { + "count": 20, + "maximum": 2.259733052753072, + "mean": 1.0618055048678168, + "minimum": 0.32515945696756865, + "quantiles": { + "0.1": 0.39283833717950983, + "0.25": 0.677519867082714, + "0.5": 0.9195688733908476, + "0.75": 1.2920543262002355, + "0.9": 2.051953328922419 + } + }, + "probability_all_four_same": { + "count": 20, + "maximum": 0.7893407330152087, + "mean": 0.2702066474742739, + "minimum": 0.019304849556107288, + "quantiles": { + "0.1": 0.04787889130182359, + "0.25": 0.11631577569983303, + "0.5": 0.2085208686109103, + "0.75": 0.34938967018420547, + "0.9": 0.583004312663369 + } + }, + "probability_no_optimal_in_four": { + "count": 20, + "maximum": 0.7359073182630198, + "mean": 0.24593747338518293, + "minimum": 1.0875827577359436e-05, + "quantiles": { + "0.1": 0.002070327833346088, + "0.25": 0.00669196583956186, + "0.5": 0.17755110556730658, + "0.75": 0.4333077050029969, + "0.9": 0.5975293526939798 + } + }, + "probability_opposed_gradient": { + "count": 20, + "maximum": 0.19985891505835052, + "mean": 0.013332034973812199, + "minimum": 0.0, + "quantiles": { + "0.1": 0.0, + "0.25": 0.0, + "0.5": 0.0, + "0.75": 0.0014330749000538603, + "0.9": 0.026544569068837225 + } + }, + "reverse_kl": { + "count": 20, + "maximum": 2.345442874769698, + "mean": 0.8296694867860487, + "minimum": 0.025362868209147814, + "quantiles": { + "0.1": 0.12729141146958892, + "0.25": 0.27287459538062264, + "0.5": 0.6401456198357021, + "0.75": 1.2364517758988751, + "0.9": 1.7145756046656158 + } + }, + "student_entropy_nats": { + "count": 20, + "maximum": 1.3670711533275415, + "mean": 0.8454060990159636, + "minimum": 0.23755895855536202, + "quantiles": { + "0.1": 0.47459180683530344, + "0.25": 0.6613344672578941, + "0.5": 0.8749047786942779, + "0.75": 1.0864482642879016, + "0.9": 1.115190385322726 + } + }, + "student_max_mass": { + "count": 20, + "maximum": 0.942573072830452, + "mean": 0.6629753951646754, + "minimum": 0.31483005069955955, + "quantiles": { + "0.1": 0.4220293028532952, + "0.25": 0.5814557794834698, + "0.5": 0.6740636277225576, + "0.75": 0.7680600368432076, + "0.9": 0.8730967046467486 + } + }, + "student_optimal_mass": { + "count": 20, + "maximum": 0.942573072830452, + "mean": 0.44261714950151093, + "minimum": 0.07379785925149918, + "quantiles": { + "0.1": 0.12111543784930545, + "0.25": 0.18866707772159472, + "0.5": 0.3516613991016407, + "0.75": 0.7149798208405773, + "0.9": 0.7961896491357757 + } + }, + "student_predicted_action_mass": { + "count": 20, + "maximum": 0.942573072830452, + "mean": 0.6629753951646754, + "minimum": 0.31483005069955955, + "quantiles": { + "0.1": 0.4220293028532952, + "0.25": 0.5814557794834698, + "0.5": 0.6740636277225576, + "0.75": 0.7680600368432076, + "0.9": 0.8730967046467486 + } + }, + "teacher_entropy_nats": { + "count": 20, + "maximum": 1.083180877687151, + "mean": 0.510098654078298, + "minimum": 0.04328351961481103, + "quantiles": { + "0.1": 0.16202953701170333, + "0.25": 0.26428645222198205, + "0.5": 0.5158831274989882, + "0.75": 0.7297643967761192, + "0.9": 0.8891499849027479 + } + }, + "teacher_mass_on_student_prediction": { + "count": 20, + "maximum": 0.9935021515418132, + "mean": 0.48565110599527045, + "minimum": 0.013741471983846842, + "quantiles": { + "0.1": 0.03145718121925008, + "0.25": 0.10939738284050596, + "0.5": 0.366198436074904, + "0.75": 0.9251774584556864, + "0.9": 0.9652340213363324 + } + }, + "teacher_max_mass": { + "count": 20, + "maximum": 0.9935021515418132, + "mean": 0.8276385195124301, + "minimum": 0.46057752289523396, + "quantiles": { + "0.1": 0.6603687180461004, + "0.25": 0.749946231543613, + "0.5": 0.8553751382857209, + "0.75": 0.9395432785180178, + "0.9": 0.9652340213363324 + } + }, + "teacher_optimal_mass": { + "count": 20, + "maximum": 0.9935021515418132, + "mean": 0.8334804591572219, + "minimum": 0.46057752289523396, + "quantiles": { + "0.1": 0.6603687180461004, + "0.25": 0.7927933476149319, + "0.5": 0.8553751382857209, + "0.75": 0.9395432785180178, + "0.9": 0.9652340213363324 + } + } + }, + "noise_signal_above_one": 9, + "student_max_above_99_percent": 0, + "teacher_optimal_above_99_percent": 1 + } + }, + "current_correct": { + "count": 68, + "metrics": { + "exact_gradient_norm": { + "count": 68, + "maximum": 2.21876536836621, + "mean": 0.24598354153475901, + "minimum": 7.143257873583472e-05, + "quantiles": { + "0.1": 0.0009664609282145813, + "0.25": 0.005980650237384781, + "0.5": 0.034591661844377064, + "0.75": 0.226124483248749, + "0.9": 0.8559990065879871 + } + }, + "expected_cosine": { + "count": 68, + "maximum": 0.9999995195428689, + "mean": 0.7990734283768914, + "minimum": -0.9484452557914791, + "quantiles": { + "0.1": 0.3597209663993855, + "0.25": 0.9030405717675959, + "0.5": 0.9759321286296707, + "0.75": 0.9950897133198694, + "0.9": 0.9993038694625062 + } + }, + "feedback_advantage": { + "count": 68, + "maximum": 0.5815802590217309, + "mean": 0.07659213239605377, + "minimum": -0.04318017131666618, + "quantiles": { + "0.1": -0.0017984953316924068, + "0.25": 0.00045483993886008545, + "0.5": 0.007704067749925159, + "0.75": 0.09095900810720992, + "0.9": 0.2699057327043152 + } + }, + "mc4_gradient_rmse": { + "count": 68, + "maximum": 1.1569823100514236, + "mean": 0.2658131738947489, + "minimum": 0.008872383045635588, + "quantiles": { + "0.1": 0.022023824263565462, + "0.25": 0.04400295223656387, + "0.5": 0.12972032586219254, + "0.75": 0.37819876232956884, + "0.9": 0.6923892875982524 + } + }, + "noise_signal_ratio": { + "count": 68, + "maximum": 124.20639437428972, + "mean": 11.697105746961249, + "minimum": 0.521453203906535, + "quantiles": { + "0.1": 0.8236627134986026, + "0.25": 1.635543795115891, + "0.5": 4.580115011216536, + "0.75": 11.348813036390519, + "0.9": 25.68883501430941 + } + }, + "probability_all_four_same": { + "count": 68, + "maximum": 0.9997074680348393, + "mean": 0.7587740203548244, + "minimum": 0.022007850403136513, + "quantiles": { + "0.1": 0.24262735485536016, + "0.25": 0.44410690295171235, + "0.5": 0.9378534706108894, + "0.75": 0.9872941914198787, + "0.9": 0.9973022477426076 + } + }, + "probability_no_optimal_in_four": { + "count": 68, + "maximum": 0.18938571195124448, + "mean": 0.00817963296426272, + "minimum": 2.5828120064841875e-17, + "quantiles": { + "0.1": 1.6344509769434443e-13, + "0.25": 7.394177240010488e-11, + "0.5": 6.417053628017351e-08, + "0.75": 0.001150213929810355, + "0.9": 0.008357027145636086 + } + }, + "probability_opposed_gradient": { + "count": 68, + "maximum": 0.998585742524318, + "mean": 0.08952880328824331, + "minimum": 0.0, + "quantiles": { + "0.1": 0.0, + "0.25": 0.0, + "0.5": 1.892742515760565e-05, + "0.75": 0.0019255355373090231, + "0.9": 0.02951034708997296 + } + }, + "reverse_kl": { + "count": 68, + "maximum": 2.125240759272016, + "mean": 0.20072740518539292, + "minimum": 0.00015561566512250187, + "quantiles": { + "0.1": 0.0006053814666843936, + "0.25": 0.002457157927533842, + "0.5": 0.01593029754843245, + "0.75": 0.14090993053166895, + "0.9": 0.6959940385017691 + } + }, + "student_entropy_nats": { + "count": 68, + "maximum": 1.3524192827810813, + "mean": 0.26589542953553214, + "minimum": 0.0008277574565303808, + "quantiles": { + "0.1": 0.006112827446073967, + "0.25": 0.024311652803693257, + "0.5": 0.0945229865455322, + "0.75": 0.4863904801632534, + "0.9": 0.7426321817313665 + } + }, + "student_max_mass": { + "count": 68, + "maximum": 0.9999268589846889, + "mean": 0.9008276237729351, + "minimum": 0.34031470595852414, + "quantiles": { + "0.1": 0.6977961238875731, + "0.25": 0.8158769341609977, + "0.5": 0.9840875128065597, + "0.75": 0.9968082996273807, + "0.9": 0.999324877945654 + } + }, + "student_optimal_mass": { + "count": 68, + "maximum": 0.999928710888966, + "mean": 0.9009840445564631, + "minimum": 0.34031470595852414, + "quantiles": { + "0.1": 0.6977961238875731, + "0.25": 0.8158769341609977, + "0.5": 0.9840875128065597, + "0.75": 0.9970687463084331, + "0.9": 0.9993645891968841 + } + }, + "student_predicted_action_mass": { + "count": 68, + "maximum": 0.9999268589846889, + "mean": 0.9008276237729351, + "minimum": 0.34031470595852414, + "quantiles": { + "0.1": 0.6977961238875731, + "0.25": 0.8158769341609977, + "0.5": 0.9840875128065597, + "0.75": 0.9968082996273807, + "0.9": 0.999324877945654 + } + }, + "teacher_entropy_nats": { + "count": 68, + "maximum": 0.7777569900056194, + "mean": 0.10488255059741668, + "minimum": 0.003398768269916631, + "quantiles": { + "0.1": 0.005475566893740364, + "0.25": 0.011580841909051724, + "0.5": 0.04027698366284022, + "0.75": 0.11844873165305785, + "0.9": 0.2887930472446079 + } + }, + "teacher_mass_on_student_prediction": { + "count": 68, + "maximum": 0.9996519695554565, + "mean": 0.9728528912155505, + "minimum": 0.678342558348217, + "quantiles": { + "0.1": 0.9308507246809506, + "0.25": 0.976841556726222, + "0.5": 0.9941900747800465, + "0.75": 0.9986383784459472, + "0.9": 0.999389730729292 + } + }, + "teacher_max_mass": { + "count": 68, + "maximum": 0.9996519695554565, + "mean": 0.9728528912155505, + "minimum": 0.678342558348217, + "quantiles": { + "0.1": 0.9308507246809506, + "0.25": 0.976841556726222, + "0.5": 0.9941900747800465, + "0.75": 0.9986383784459472, + "0.9": 0.999389730729292 + } + }, + "teacher_optimal_mass": { + "count": 68, + "maximum": 0.9996851497012124, + "mean": 0.9775761769525169, + "minimum": 0.7519568372955211, + "quantiles": { + "0.1": 0.9432301622255373, + "0.25": 0.9791515716824222, + "0.5": 0.9942855851518781, + "0.75": 0.9986741093381786, + "0.9": 0.999389730729292 + } + } + }, + "noise_signal_above_one": 56, + "student_max_above_99_percent": 29, + "teacher_optimal_above_99_percent": 41 + }, + "current_mistakes": { + "count": 28, + "metrics": { + "exact_gradient_norm": { + "count": 28, + "maximum": 1.4227389142920086, + "mean": 0.5368683644289622, + "minimum": 0.025505490605703153, + "quantiles": { + "0.1": 0.09436940721180344, + "0.25": 0.2093383939523424, + "0.5": 0.46880721885725796, + "0.75": 0.7639456600511869, + "0.9": 1.1014005697889417 + } + }, + "expected_cosine": { + "count": 28, + "maximum": 0.9999399139989253, + "mean": 0.9217052042968644, + "minimum": 0.5324512126919654, + "quantiles": { + "0.1": 0.7628909172819066, + "0.25": 0.8525525827615673, + "0.5": 0.9842419704425351, + "0.75": 0.9942212166749731, + "0.9": 0.9992893513881955 + } + }, + "feedback_advantage": { + "count": 28, + "maximum": 0.9392785316232912, + "mean": 0.667158872912868, + "minimum": 0.08672328231940699, + "quantiles": { + "0.1": 0.3636120672573178, + "0.25": 0.5505714638275201, + "0.5": 0.7093035437029293, + "0.75": 0.8517500517228255, + "0.9": 0.9035910896940837 + } + }, + "mc4_gradient_rmse": { + "count": 28, + "maximum": 1.0206900154934309, + "mean": 0.42439473261802874, + "minimum": 0.06678320383782932, + "quantiles": { + "0.1": 0.2081597334088402, + "0.25": 0.3013269614342087, + "0.5": 0.37394680246780154, + "0.75": 0.5022727772196234, + "0.9": 0.7282398531819607 + } + }, + "noise_signal_ratio": { + "count": 28, + "maximum": 7.75647259578772, + "mean": 1.5758247847510929, + "minimum": 0.04693988697923705, + "quantiles": { + "0.1": 0.35958332797512726, + "0.25": 0.6066651948674372, + "0.5": 1.059509907913778, + "0.75": 1.748732265954735, + "0.9": 3.388978993309162 + } + }, + "probability_all_four_same": { + "count": 28, + "maximum": 0.9892293199345791, + "mean": 0.5070316807455628, + "minimum": 0.019304849556107288, + "quantiles": { + "0.1": 0.09820151755869942, + "0.25": 0.15010421396943294, + "0.5": 0.48617171757190925, + "0.75": 0.8748975669552926, + "0.9": 0.9502186501787762 + } + }, + "probability_no_optimal_in_four": { + "count": 28, + "maximum": 0.9913022061045822, + "mean": 0.6225071966784352, + "minimum": 0.06782164589178531, + "quantiles": { + "0.1": 0.24970720244370906, + "0.25": 0.42034878517367164, + "0.5": 0.6762674602332133, + "0.75": 0.89422787950423, + "0.9": 0.9512931647394306 + } + }, + "probability_opposed_gradient": { + "count": 28, + "maximum": 0.19985891505835052, + "mean": 0.014563455726886818, + "minimum": 0.0, + "quantiles": { + "0.1": 0.0, + "0.25": 0.0, + "0.5": 8.170055764655128e-05, + "0.75": 0.0025287637797236215, + "0.9": 0.01703788333384296 + } + }, + "reverse_kl": { + "count": 28, + "maximum": 3.5071095201579405, + "mean": 1.6445194644037946, + "minimum": 0.07017227253685511, + "quantiles": { + "0.1": 0.45766751915089604, + "0.25": 0.9728325764609413, + "0.5": 1.6432399334800203, + "0.75": 2.453662499525725, + "0.9": 2.647594090845793 + } + }, + "student_entropy_nats": { + "count": 28, + "maximum": 1.3670711533275415, + "mean": 0.5597049994012604, + "minimum": 0.020142533976262124, + "quantiles": { + "0.1": 0.06891624830367725, + "0.25": 0.16231255941262637, + "0.5": 0.547257849925486, + "0.75": 0.9392668355385907, + "0.9": 1.0886034678050702 + } + }, + "student_max_mass": { + "count": 28, + "maximum": 0.9972963854291457, + "mean": 0.7792046689932471, + "minimum": 0.31483005069955955, + "quantiles": { + "0.1": 0.5019419182205841, + "0.25": 0.6197983123512248, + "0.5": 0.8344437332213467, + "0.75": 0.9670917126687026, + "0.9": 0.987308353583247 + } + }, + "student_optimal_mass": { + "count": 28, + "maximum": 0.48968059680028747, + "mean": 0.13786547948134623, + "minimum": 0.002181577014188315, + "quantiles": { + "0.1": 0.012415294011438932, + "0.25": 0.02757262415446592, + "0.5": 0.09343473773372687, + "0.75": 0.1950233616704246, + "0.9": 0.2944352066744962 + } + }, + "student_predicted_action_mass": { + "count": 28, + "maximum": 0.9972963854291457, + "mean": 0.7792046689932471, + "minimum": 0.31483005069955955, + "quantiles": { + "0.1": 0.5019419182205841, + "0.25": 0.6197983123512248, + "0.5": 0.8344437332213467, + "0.75": 0.9670917126687026, + "0.9": 0.987308353583247 + } + }, + "teacher_entropy_nats": { + "count": 28, + "maximum": 1.083180877687151, + "mean": 0.4621424801149539, + "minimum": 0.08824553660486163, + "quantiles": { + "0.1": 0.18755906052268348, + "0.25": 0.24546455946790977, + "0.5": 0.40477082343587945, + "0.75": 0.6450354509600448, + "0.9": 0.8628190390852274 + } + }, + "teacher_mass_on_student_prediction": { + "count": 28, + "maximum": 0.9041195554626359, + "mean": 0.1580898974598495, + "minimum": 0.013741471983846842, + "quantiles": { + "0.1": 0.015827724891469268, + "0.25": 0.0460259751457132, + "0.5": 0.06601837997180902, + "0.75": 0.18110061526589719, + "0.9": 0.3500945171572466 + } + }, + "teacher_max_mass": { + "count": 28, + "maximum": 0.9832965903798537, + "mean": 0.8341722358790667, + "minimum": 0.46057752289523396, + "quantiles": { + "0.1": 0.6385513970066675, + "0.25": 0.7516497415908574, + "0.5": 0.8925862209919886, + "0.75": 0.9400552397110644, + "0.9": 0.9585076963905251 + } + }, + "teacher_optimal_mass": { + "count": 28, + "maximum": 0.9832965903798537, + "mean": 0.8050243523942143, + "minimum": 0.09529349369827628, + "quantiles": { + "0.1": 0.514340657776026, + "0.25": 0.7516497415908574, + "0.5": 0.8711562128432455, + "0.75": 0.9400552397110644, + "0.9": 0.9585076963905251 + } + } + }, + "noise_signal_above_one": 15, + "student_max_above_99_percent": 3, + "teacher_optimal_above_99_percent": 0 + }, + "mistakes_by_kind": { + "admin": { + "count": 7, + "metrics": { + "exact_gradient_norm": { + "count": 7, + "maximum": 0.622637337516211, + "mean": 0.29112271542481016, + "minimum": 0.025505490605703153, + "quantiles": { + "0.1": 0.051247691061425626, + "0.25": 0.08695219316040453, + "0.5": 0.3077481912905436, + "0.75": 0.45403180112020214, + "0.9": 0.5958263651507779 + } + }, + "expected_cosine": { + "count": 7, + "maximum": 0.9999399139989253, + "mean": 0.9599033730943858, + "minimum": 0.7474851895614436, + "quantiles": { + "0.1": 0.8935334686551464, + "0.25": 0.9916410651877978, + "0.5": 0.9940034839272041, + "0.75": 0.997306446898766, + "0.9": 0.9998190529271215 + } + }, + "feedback_advantage": { + "count": 7, + "maximum": 0.9392785316232912, + "mean": 0.8280376205833486, + "minimum": 0.7036571542956169, + "quantiles": { + "0.1": 0.7255618601924023, + "0.25": 0.7686742759614602, + "0.5": 0.8442547458927114, + "0.75": 0.8858621801744504, + "0.9": 0.9142044473307563 + } + }, + "mc4_gradient_rmse": { + "count": 7, + "maximum": 0.6454882599793311, + "mean": 0.3651195155054297, + "minimum": 0.19783263892525765, + "quantiles": { + "0.1": 0.24728928970948666, + "0.25": 0.29658478496049545, + "0.5": 0.32551778118851216, + "0.75": 0.39691417926195816, + "0.9": 0.500824511810278 + } + }, + "noise_signal_ratio": { + "count": 7, + "maximum": 7.75647259578772, + "mean": 2.7212543586334674, + "minimum": 0.5025544740649898, + "quantiles": { + "0.1": 0.8711340116618955, + "0.25": 1.170920445013904, + "0.5": 1.2654707859040921, + "0.75": 3.5912208823248317, + "0.9": 5.560684401233494 + } + }, + "probability_all_four_same": { + "count": 7, + "maximum": 0.9892293199345791, + "mean": 0.7656755424840881, + "minimum": 0.23229422609740713, + "quantiles": { + "0.1": 0.40605158662691704, + "0.25": 0.6785500751108025, + "0.5": 0.8631838468259141, + "0.75": 0.9589606271545557, + "0.9": 0.9795627914771389 + } + }, + "probability_no_optimal_in_four": { + "count": 7, + "maximum": 0.9913022061045822, + "mean": 0.8319741607422516, + "minimum": 0.527763523700169, + "quantiles": { + "0.1": 0.5939061475470466, + "0.25": 0.7488869520935555, + "0.5": 0.8885579732637422, + "0.75": 0.9592107589700786, + "0.9": 0.9806201090938969 + } + }, + "probability_opposed_gradient": { + "count": 7, + "maximum": 0.14027784193617954, + "mean": 0.020068890083720495, + "minimum": 0.0, + "quantiles": { + "0.1": 0.0, + "0.25": 0.0, + "0.5": 2.8789785204193496e-05, + "0.75": 8.779943232986413e-05, + "0.9": 0.056164454244665436 + } + }, + "reverse_kl": { + "count": 7, + "maximum": 3.5071095201579405, + "mean": 2.0987675247885225, + "minimum": 1.2301227192403088, + "quantiles": { + "0.1": 1.2823271245396473, + "0.25": 1.5498426999161121, + "0.5": 1.8951303606487118, + "0.75": 2.479662336820236, + "0.9": 3.001420716979882 + } + }, + "student_entropy_nats": { + "count": 7, + "maximum": 0.8268736417592039, + "mean": 0.25449129593364733, + "minimum": 0.020142533976262124, + "quantiles": { + "0.1": 0.03279019268614395, + "0.25": 0.0577609279854618, + "0.5": 0.1779616112888418, + "0.75": 0.32046971427015, + "0.9": 0.5934991555908122 + } + }, + "student_max_mass": { + "count": 7, + "maximum": 0.9972963854291457, + "mean": 0.9198596885909356, + "minimum": 0.6929802573511676, + "quantiles": { + "0.1": 0.7870474720580317, + "0.25": 0.902869376671073, + "0.5": 0.9638861407876811, + "0.75": 0.9895581416132042, + "0.9": 0.9948450230517281 + } + }, + "student_optimal_mass": { + "count": 7, + "maximum": 0.1476659417449555, + "mean": 0.049007010266542875, + "minimum": 0.002181577014188315, + "quantiles": { + "0.1": 0.004887951618203077, + "0.25": 0.010377693961233704, + "0.5": 0.029106838661514586, + "0.75": 0.07166966326133717, + "0.9": 0.12282955858907124 + } + }, + "student_predicted_action_mass": { + "count": 7, + "maximum": 0.9972963854291457, + "mean": 0.9198596885909356, + "minimum": 0.6929802573511676, + "quantiles": { + "0.1": 0.7870474720580317, + "0.25": 0.902869376671073, + "0.5": 0.9638861407876811, + "0.75": 0.9895581416132042, + "0.9": 0.9948450230517281 + } + }, + "teacher_entropy_nats": { + "count": 7, + "maximum": 0.5637644594300629, + "mean": 0.3850734804665826, + "minimum": 0.16476190474349955, + "quantiles": { + "0.1": 0.25737083483779444, + "0.25": 0.3491668213646415, + "0.5": 0.3839701550463422, + "0.75": 0.44234210065844515, + "0.9": 0.5009734094669094 + } + }, + "teacher_mass_on_student_prediction": { + "count": 7, + "maximum": 0.24486203794872544, + "mean": 0.11068869617253385, + "minimum": 0.02277424166463781, + "quantiles": { + "0.1": 0.039957028551160595, + "0.25": 0.05842916149940891, + "0.5": 0.09529992886052506, + "0.75": 0.14751317086751542, + "0.9": 0.18670747394254875 + } + }, + "teacher_max_mass": { + "count": 7, + "maximum": 0.9683853702848058, + "mean": 0.8770446308498915, + "minimum": 0.7542281840251803, + "quantiles": { + "0.1": 0.809553067354212, + "0.25": 0.8488797094737361, + "0.5": 0.9034555242844763, + "0.75": 0.9077419592036531, + "0.9": 0.9341361436643385 + } + }, + "teacher_optimal_mass": { + "count": 7, + "maximum": 0.9683853702848058, + "mean": 0.8770446308498915, + "minimum": 0.7542281840251803, + "quantiles": { + "0.1": 0.809553067354212, + "0.25": 0.8488797094737361, + "0.5": 0.9034555242844763, + "0.75": 0.9077419592036531, + "0.9": 0.9341361436643385 + } + } + }, + "noise_signal_above_one": 6, + "student_max_above_99_percent": 2, + "teacher_optimal_above_99_percent": 0 + }, + "exercise": { + "count": 3, + "metrics": { + "exact_gradient_norm": { + "count": 3, + "maximum": 1.0836158503557065, + "mean": 0.49550762874395315, + "minimum": 0.19554859127498583, + "quantiles": { + "0.1": 0.19791056194022208, + "0.25": 0.20145351793807642, + "0.5": 0.20735844460116704, + "0.75": 0.6454871474784367, + "0.9": 0.9083643692047987 + } + }, + "expected_cosine": { + "count": 3, + "maximum": 0.9994511711651409, + "mean": 0.9489136350744063, + "minimum": 0.8551834755602493, + "quantiles": { + "0.1": 0.8825680321477651, + "0.25": 0.9236448670290389, + "0.5": 0.9921062584978286, + "0.75": 0.9957787148314847, + "0.9": 0.9979821886316784 + } + }, + "feedback_advantage": { + "count": 3, + "maximum": 0.9315916123282266, + "mean": 0.8811867534881264, + "minimum": 0.7977948455014461, + "quantiles": { + "0.1": 0.8210706369280981, + "0.25": 0.8559843240680762, + "0.5": 0.9141738026347064, + "0.75": 0.9228827074814665, + "0.9": 0.9281080503895226 + } + }, + "mc4_gradient_rmse": { + "count": 3, + "maximum": 0.6949973634353181, + "mean": 0.4809636561236404, + "minimum": 0.3700051362575641, + "quantiles": { + "0.1": 0.3715818027416591, + "0.25": 0.37394680246780154, + "0.5": 0.37788846867803894, + "0.75": 0.5364429160566786, + "0.9": 0.6315755844838623 + } + }, + "noise_signal_ratio": { + "count": 3, + "maximum": 1.8921391038672974, + "mean": 1.451966781816912, + "minimum": 0.641368768468253, + "quantiles": { + "0.1": 0.8775735093976395, + "0.25": 1.2318806207917192, + "0.5": 1.8223924731151853, + "0.75": 1.8572657884912414, + "0.9": 1.878189777716875 + } + }, + "probability_all_four_same": { + "count": 3, + "maximum": 0.9156722721297913, + "mean": 0.6802054703226437, + "minimum": 0.21490541149471157, + "quantiles": { + "0.1": 0.3539320746644549, + "0.25": 0.5624720694190699, + "0.5": 0.9100387273434283, + "0.75": 0.9128554997366098, + "0.9": 0.9145455631725187 + } + }, + "probability_no_optimal_in_four": { + "count": 3, + "maximum": 0.9212308377416744, + "mean": 0.767546892592708, + "minimum": 0.4701722418107565, + "quantiles": { + "0.1": 0.5583853130937438, + "0.25": 0.6907049200182249, + "0.5": 0.9112375982256933, + "0.75": 0.9162342179836838, + "0.9": 0.9192321898384781 + } + }, + "probability_opposed_gradient": { + "count": 3, + "maximum": 0.011538383484934777, + "mean": 0.005685814624916529, + "minimum": 0.0, + "quantiles": { + "0.1": 0.0011038120779629618, + "0.25": 0.0027595301949074045, + "0.5": 0.005519060389814809, + "0.75": 0.008528721937374794, + "0.9": 0.010334518865910785 + } + }, + "reverse_kl": { + "count": 3, + "maximum": 2.884345124497957, + "mean": 2.7139235078115087, + "minimum": 2.6169887755259404, + "quantiles": { + "0.1": 2.621678345102878, + "0.25": 2.6287126994682843, + "0.5": 2.6404366234106282, + "0.75": 2.762390873954293, + "0.9": 2.8355634242804912 + } + }, + "student_entropy_nats": { + "count": 3, + "maximum": 0.9222125778548519, + "mean": 0.3816588449397725, + "minimum": 0.11037552899846086, + "quantiles": { + "0.1": 0.11077810879196962, + "0.25": 0.11138197848223276, + "0.5": 0.11238842796600466, + "0.75": 0.5173005029104283, + "0.9": 0.7602477478770825 + } + }, + "student_max_mass": { + "count": 3, + "maximum": 0.9782165263684871, + "mean": 0.8783141390424375, + "minimum": 0.680017462447058, + "quantiles": { + "0.1": 0.7393556556199998, + "0.25": 0.8283629453794127, + "0.5": 0.9767084283117674, + "0.75": 0.9774624773401273, + "0.9": 0.9779149067571432 + } + }, + "student_optimal_mass": { + "count": 3, + "maximum": 0.1719353474610533, + "mean": 0.07173585470738611, + "minimum": 0.02030223602778511, + "quantiles": { + "0.1": 0.020835784948892072, + "0.25": 0.021636108330552517, + "0.5": 0.022969980633319925, + "0.75": 0.09745266404718661, + "0.9": 0.14214227409550662 + } + }, + "student_predicted_action_mass": { + "count": 3, + "maximum": 0.9782165263684871, + "mean": 0.8783141390424375, + "minimum": 0.680017462447058, + "quantiles": { + "0.1": 0.7393556556199998, + "0.25": 0.8283629453794127, + "0.5": 0.9767084283117674, + "0.75": 0.9774624773401273, + "0.9": 0.9779149067571432 + } + }, + "teacher_entropy_nats": { + "count": 3, + "maximum": 0.24764939622976076, + "mean": 0.20349764914313076, + "minimum": 0.1660785632926465, + "quantiles": { + "0.1": 0.1722158482155142, + "0.25": 0.18142177559981576, + "0.5": 0.19676498790698502, + "0.75": 0.2222071920683729, + "0.9": 0.23747251456520563 + } + }, + "teacher_mass_on_student_prediction": { + "count": 3, + "maximum": 0.05990959588574053, + "mean": 0.04099194255346978, + "minimum": 0.015674228660255238, + "quantiles": { + "0.1": 0.022017783551086904, + "0.25": 0.0315331158873344, + "0.5": 0.047392003114413564, + "0.75": 0.05365079950007705, + "0.9": 0.05740607733147514 + } + }, + "teacher_max_mass": { + "count": 3, + "maximum": 0.9697301929624994, + "mean": 0.9529226081955124, + "minimum": 0.9371437832680263, + "quantiles": { + "0.1": 0.9400937962856234, + "0.25": 0.944518815812019, + "0.5": 0.9518938483560118, + "0.75": 0.9608120206592556, + "0.9": 0.9661629240412019 + } + }, + "teacher_optimal_mass": { + "count": 3, + "maximum": 0.9697301929624994, + "mean": 0.9529226081955124, + "minimum": 0.9371437832680263, + "quantiles": { + "0.1": 0.9400937962856234, + "0.25": 0.944518815812019, + "0.5": 0.9518938483560118, + "0.75": 0.9608120206592556, + "0.9": 0.9661629240412019 + } + } + }, + "noise_signal_above_one": 2, + "student_max_above_99_percent": 0, + "teacher_optimal_above_99_percent": 0 + }, + "focus": { + "count": 7, + "metrics": { + "exact_gradient_norm": { + "count": 7, + "maximum": 1.4227389142920086, + "mean": 0.5181927737740692, + "minimum": 0.02984465113439629, + "quantiles": { + "0.1": 0.0801955583459748, + "0.25": 0.16188060344488067, + "0.5": 0.3755768481352648, + "0.75": 0.7377138979835269, + "0.9": 1.1577034434252305 + } + }, + "expected_cosine": { + "count": 7, + "maximum": 0.9992200000552189, + "mean": 0.9902543418144136, + "minimum": 0.976006476181658, + "quantiles": { + "0.1": 0.9795901516072192, + "0.25": 0.9842419704425351, + "0.5": 0.991337658991088, + "0.75": 0.9983661582939299, + "0.9": 0.9988880216348508 + } + }, + "feedback_advantage": { + "count": 7, + "maximum": 0.8990556412909598, + "mean": 0.5797506314876897, + "minimum": 0.08672328231940699, + "quantiles": { + "0.1": 0.2565603897455263, + "0.25": 0.41335985390919894, + "0.5": 0.6558455845481231, + "0.75": 0.7949551022184702, + "0.9": 0.8845984193124031 + } + }, + "mc4_gradient_rmse": { + "count": 7, + "maximum": 0.4084492579571785, + "mean": 0.2849768480806056, + "minimum": 0.06678320383782932, + "quantiles": { + "0.1": 0.12021766056142405, + "0.25": 0.23052156906405818, + "0.5": 0.32582321242684803, + "0.75": 0.36636956210713345, + "0.9": 0.4036216782115315 + } + }, + "noise_signal_ratio": { + "count": 7, + "maximum": 5.221727371136168, + "mean": 1.6888895564463409, + "minimum": 0.04693988697923705, + "quantiles": { + "0.1": 0.22203674050480315, + "0.25": 0.5743105799639676, + "0.5": 1.0875251229811551, + "0.75": 2.158706677049946, + "0.9": 3.8071250605876448 + } + }, + "probability_all_four_same": { + "count": 7, + "maximum": 0.9628555986097445, + "mean": 0.6557676713297718, + "minimum": 0.10830288015167715, + "quantiles": { + "0.1": 0.21630128983357985, + "0.25": 0.5160725654082586, + "0.5": 0.7466014324629027, + "0.75": 0.8702343286337804, + "0.9": 0.9363710560644903 + } + }, + "probability_no_optimal_in_four": { + "count": 7, + "maximum": 0.9661573331402065, + "mean": 0.6570947011825459, + "minimum": 0.07177871621534286, + "quantiles": { + "0.1": 0.20064981277489435, + "0.25": 0.5216122843615312, + "0.5": 0.7568626931230737, + "0.75": 0.8808197985380678, + "0.9": 0.9493761208580014 + } + }, + "probability_opposed_gradient": { + "count": 7, + "maximum": 0.01007981766675352, + "mean": 0.0024465699386113362, + "minimum": 0.0, + "quantiles": { + "0.1": 1.1761331766315014e-08, + "0.25": 3.170231550101465e-08, + "0.5": 0.0016056832131731568, + "0.75": 0.002720212642860838, + "0.9": 0.005893793288182573 + } + }, + "reverse_kl": { + "count": 7, + "maximum": 2.601679608132438, + "mean": 1.3970748978719028, + "minimum": 0.07017227253685511, + "quantiles": { + "0.1": 0.23538700231910353, + "0.25": 0.5715860675262644, + "0.5": 1.040580847847811, + "0.75": 2.461959710766843, + "0.9": 2.527804323202423 + } + }, + "student_entropy_nats": { + "count": 7, + "maximum": 0.8454925142348215, + "mean": 0.33636688560404265, + "minimum": 0.05635441578536559, + "quantiles": { + "0.1": 0.0917610085845343, + "0.25": 0.15524389748696457, + "0.5": 0.2696203281250995, + "0.75": 0.43630657305454135, + "0.9": 0.6968150837865931 + } + }, + "student_max_mass": { + "count": 7, + "maximum": 0.9905816745050308, + "mean": 0.8559680790501595, + "minimum": 0.48239445326928454, + "quantiles": { + "0.1": 0.6306228579103049, + "0.25": 0.8290625171906865, + "0.5": 0.9295424227274687, + "0.75": 0.9655664842339797, + "0.9": 0.9836495313606967 + } + }, + "student_optimal_mass": { + "count": 7, + "maximum": 0.48239445326928454, + "mean": 0.1367353945081548, + "minimum": 0.008570211378869291, + "quantiles": { + "0.1": 0.01292279316657263, + "0.25": 0.031613511062938505, + "0.5": 0.06727358834332912, + "0.75": 0.16784124321986182, + "0.9": 0.3539657756694812 + } + }, + "student_predicted_action_mass": { + "count": 7, + "maximum": 0.9905816745050308, + "mean": 0.8559680790501595, + "minimum": 0.48239445326928454, + "quantiles": { + "0.1": 0.6306228579103049, + "0.25": 0.8290625171906865, + "0.5": 0.9295424227274687, + "0.75": 0.9655664842339797, + "0.9": 0.9836495313606967 + } + }, + "teacher_entropy_nats": { + "count": 7, + "maximum": 0.6968533099221963, + "mean": 0.35616862247631964, + "minimum": 0.08824553660486163, + "quantiles": { + "0.1": 0.1612371652449835, + "0.25": 0.22440415009371079, + "0.5": 0.2571723018502147, + "0.75": 0.5010504543847717, + "0.9": 0.6881469234861142 + } + }, + "teacher_mass_on_student_prediction": { + "count": 7, + "maximum": 0.9041195554626359, + "mean": 0.2797297209593371, + "minimum": 0.015893508990560995, + "quantiles": { + "0.1": 0.03815339593781589, + "0.25": 0.05307525984853627, + "0.5": 0.05339551650366007, + "0.75": 0.4392744730307152, + "0.9": 0.6984124777292924 + } + }, + "teacher_max_mass": { + "count": 7, + "maximum": 0.9832965903798537, + "mean": 0.8497688160632311, + "minimum": 0.5612744259070631, + "quantiles": { + "0.1": 0.627511829706724, + "0.25": 0.7878948271845669, + "0.5": 0.9393290330580766, + "0.75": 0.9443460043642449, + "0.9": 0.9611935255870188 + } + }, + "teacher_optimal_mass": { + "count": 7, + "maximum": 0.9832965903798537, + "mean": 0.7164860259958445, + "minimum": 0.09529349369827628, + "quantiles": { + "0.1": 0.30038997179914095, + "0.25": 0.5543955280531077, + "0.5": 0.9393290330580766, + "0.75": 0.9443460043642449, + "0.9": 0.9611935255870188 + } + } + }, + "noise_signal_above_one": 4, + "student_max_above_99_percent": 1, + "teacher_optimal_above_99_percent": 0 + }, + "repair": { + "count": 11, + "metrics": { + "exact_gradient_norm": { + "count": 11, + "maximum": 1.259970215034505, + "mean": 0.7164166266715389, + "minimum": 0.29858529505471754, + "quantiles": { + "0.1": 0.3983897879558068, + "0.25": 0.49485211025754416, + "0.5": 0.5969746050275245, + "0.75": 0.9579399154251941, + "0.9": 1.1428982484664907 + } + }, + "expected_cosine": { + "count": 11, + "maximum": 0.9891436710880799, + "mean": 0.8463546191570355, + "minimum": 0.5324512126919654, + "quantiles": { + "0.1": 0.7296595664326799, + "0.25": 0.779941693515062, + "0.5": 0.8446599043655213, + "0.75": 0.9619438695594517, + "0.9": 0.979150806528189 + } + }, + "feedback_advantage": { + "count": 11, + "maximum": 0.7349780776709811, + "mean": 0.5620336742362416, + "minimum": 0.32136591051181357, + "quantiles": { + "0.1": 0.34920825878864575, + "0.25": 0.4326556065756062, + "0.5": 0.6211089991567667, + "0.75": 0.6689240823637768, + "0.9": 0.7306954498827651 + } + }, + "mc4_gradient_rmse": { + "count": 11, + "maximum": 1.0206900154934309, + "mean": 0.5354079090756033, + "minimum": 0.21258563104466127, + "quantiles": { + "0.1": 0.21998764508300117, + "0.25": 0.2988447368474325, + "0.5": 0.49837130381887623, + "0.75": 0.74635045722275, + "0.9": 0.9384926324830757 + } + }, + "noise_signal_ratio": { + "count": 11, + "maximum": 1.7241788635679185, + "mean": 0.8087442020019276, + "minimum": 0.32515945696756865, + "quantiles": { + "0.1": 0.368504193026533, + "0.25": 0.4376015491983206, + "0.5": 0.8930716420844678, + "0.75": 0.9774842689911147, + "0.9": 1.1166753744639604 + } + }, + "probability_all_four_same": { + "count": 11, + "maximum": 0.5637947499562791, + "mean": 0.2005607412009827, + "minimum": 0.019304849556107288, + "quantiles": { + "0.1": 0.048561437198557746, + "0.25": 0.09547928878721831, + "0.5": 0.14978217658724202, + "0.75": 0.2576284816052736, + "0.9": 0.45045360816389485 + } + }, + "probability_no_optimal_in_four": { + "count": 11, + "maximum": 0.7359073182630198, + "mean": 0.4276435268858618, + "minimum": 0.06782164589178531, + "quantiles": { + "0.1": 0.19772996152412062, + "0.25": 0.324877073078212, + "0.5": 0.4332315684264591, + "0.75": 0.5449319574899785, + "0.9": 0.7145336903547949 + } + }, + "probability_opposed_gradient": { + "count": 11, + "maximum": 0.19985891505835052, + "mean": 0.02119191784743259, + "minimum": 0.0, + "quantiles": { + "0.1": 0.0, + "0.25": 0.0, + "0.5": 7.666470095600251e-05, + "0.75": 0.0015365658186713932, + "0.9": 0.029870049647962013 + } + }, + "reverse_kl": { + "count": 11, + "maximum": 2.345442874769698, + "mean": 1.2212615928407953, + "minimum": 0.39050172354675633, + "quantiles": { + "0.1": 0.4864528601240988, + "0.25": 0.614531731872795, + "0.5": 1.1373106484422033, + "0.75": 1.6432399334800203, + "0.9": 1.993521483291967 + } + }, + "student_entropy_nats": { + "count": 11, + "maximum": 1.3670711533275415, + "mean": 0.9446141979683766, + "minimum": 0.4968189030298646, + "quantiles": { + "0.1": 0.6197559114864926, + "0.25": 0.7768808769953013, + "0.5": 1.0244711124698347, + "0.75": 1.0956448745969802, + "0.9": 1.113947883248574 + } + }, + "student_max_mass": { + "count": 11, + "maximum": 0.8665049148709926, + "mean": 0.6138176310178128, + "minimum": 0.31483005069955955, + "quantiles": { + "0.1": 0.4277928487501772, + "0.25": 0.5455627983859861, + "0.5": 0.6179018616676331, + "0.75": 0.7063036899112796, + "0.9": 0.8191285179134189 + } + }, + "student_optimal_mass": { + "count": 11, + "maximum": 0.48968059680028747, + "mean": 0.21316627526660475, + "minimum": 0.07379785925149918, + "quantiles": { + "0.1": 0.08059750564897204, + "0.25": 0.14124341256510087, + "0.5": 0.18870271289112053, + "0.75": 0.24692631978364663, + "0.9": 0.333165408421924 + } + }, + "student_predicted_action_mass": { + "count": 11, + "maximum": 0.8665049148709926, + "mean": 0.6138176310178128, + "minimum": 0.31483005069955955, + "quantiles": { + "0.1": 0.4277928487501772, + "0.25": 0.5455627983859861, + "0.5": 0.6179018616676331, + "0.75": 0.7063036899112796, + "0.9": 0.8191285179134189 + } + }, + "teacher_entropy_nats": { + "count": 11, + "maximum": 1.083180877687151, + "mean": 0.6491637977444547, + "minimum": 0.2011178371516103, + "quantiles": { + "0.1": 0.32478704335659664, + "0.25": 0.46722231595951924, + "0.5": 0.6325997126593732, + "0.75": 0.8675821615152794, + "0.9": 0.9760901407137892 + } + }, + "teacher_mass_on_student_prediction": { + "count": 11, + "maximum": 0.4064582333690476, + "mean": 0.14278294375384365, + "minimum": 0.013741471983846842, + "quantiles": { + "0.1": 0.014251626269213056, + "0.25": 0.03764840039332204, + "0.5": 0.12366629153612706, + "0.75": 0.19177875870096578, + "0.9": 0.3259386387807604 + } + }, + "teacher_max_mass": { + "count": 11, + "maximum": 0.9542744075786906, + "mean": 0.7645782410577703, + "minimum": 0.46057752289523396, + "quantiles": { + "0.1": 0.5373820012963655, + "0.25": 0.6980419857445834, + "0.5": 0.8064055177214022, + "0.75": 0.8711562128432455, + "0.9": 0.9192556220957825 + } + }, + "teacher_optimal_mass": { + "count": 11, + "maximum": 0.9542744075786906, + "mean": 0.7751999495028464, + "minimum": 0.46057752289523396, + "quantiles": { + "0.1": 0.5373820012963655, + "0.25": 0.7089741615419798, + "0.5": 0.8388888555889332, + "0.75": 0.8711562128432455, + "0.9": 0.9192556220957825 + } + } + }, + "noise_signal_above_one": 3, + "student_max_above_99_percent": 0, + "teacher_optimal_above_99_percent": 0 + } + } + }, + "verification": { + "maximum_gradient_expectation_error": 2.6645352591003757e-15, + "maximum_variance_error": 1.7763568394002505e-15, + "passed": true + } +}