diff options
| author | YurenHao0426 <blackhao0426@gmail.com> | 2026-06-13 12:35:36 -0500 |
|---|---|---|
| committer | YurenHao0426 <blackhao0426@gmail.com> | 2026-06-13 12:35:36 -0500 |
| commit | 66e0d8b9fd4d0f7a2231d689c055e26fdf1cf04a (patch) | |
| tree | c29cba61124018755a19b02c9d33e3ad5f2e05cc /research/flossing/spectrum_microscope/checkpoint_summary.json | |
Curated export for clone-and-run Maze training (2x A6000) + diagnostics.
trm/hrm pretrain.py carry trajectory-augmentation code (backward-compatible).
Heavy artifacts (checkpoints/wandb/npz) gitignored; see PROVENANCE.md.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Diffstat (limited to 'research/flossing/spectrum_microscope/checkpoint_summary.json')
| -rw-r--r-- | research/flossing/spectrum_microscope/checkpoint_summary.json | 982 |
1 files changed, 982 insertions, 0 deletions
diff --git a/research/flossing/spectrum_microscope/checkpoint_summary.json b/research/flossing/spectrum_microscope/checkpoint_summary.json new file mode 100644 index 0000000..69a643d --- /dev/null +++ b/research/flossing/spectrum_microscope/checkpoint_summary.json @@ -0,0 +1,982 @@ +[ + { + "kind": "HRM", + "file": "/home/yurenh2/rrm/research/flossing/diag_hrm_step_2604_512.npz", + "step": 2604, + "n": 512, + "k": 8, + "acc": 0.015625, + "n_success": 8, + "n_failure": 504, + "raw_monotone_adjacent_fraction": 0.6551339285714286, + "raw_col0_is_sample_max_fraction": 0.62890625, + "raw_col0_success_mean": -0.12265677284449339, + "raw_col0_failure_mean": -0.0977591768882814, + "raw_col0_delta_failure_minus_success": 0.024897595956211993, + "raw_col0_auc_failure": 0.6889880952380952, + "lambda_max_success_mean": -0.1213220115751028, + "lambda_max_failure_mean": -0.0949796658704087, + "lambda_max_delta_failure_minus_success": 0.026342345704694112, + "lambda_max_auc_failure": 0.6932043650793651, + "mean8_success_mean": -0.19951309508178383, + "mean8_failure_mean": -0.11536999984130648, + "mean8_delta_failure_minus_success": 0.08414309524047735, + "mean8_auc_failure": 0.9992559523809523, + "tail_mean_5_8_success_mean": -0.23467964679002762, + "tail_mean_5_8_failure_mean": -0.12418151276171326, + "tail_mean_5_8_delta_failure_minus_success": 0.11049813402831436, + "tail_mean_5_8_auc_failure": 1.0, + "positive_sum_success_mean": 0.0, + "positive_sum_failure_mean": 7.118396313181e-05, + "positive_sum_delta_failure_minus_success": 7.118396313181e-05, + "positive_sum_auc_failure": 0.5009920634920635, + "positive_count_success_mean": 0.0, + "positive_count_failure_mean": 0.001984126984126984, + "positive_count_delta_failure_minus_success": 0.001984126984126984, + "positive_count_auc_failure": 0.5009920634920635, + "spread_success_mean": 0.13017353788018227, + "spread_failure_mean": 0.034761710632001126, + "spread_delta_failure_minus_success": -0.09541182724818115, + "spread_auc_failure": 0.019345238095238096, + "gap12_success_mean": 0.03395136073231697, + "gap12_failure_mean": 0.01004286463712416, + "gap12_delta_failure_minus_success": -0.023908496095192813, + "gap12_auc_failure": 0.12177579365079365, + "lambda_max_corr_token_acc": 0.11864046173536506, + "mean8_corr_token_acc": -0.24317640561416, + "tail_mean_5_8_corr_token_acc": -0.364745516214904, + "positive_sum_corr_token_acc": 0.01654115981917942, + "positive_count_corr_token_acc": 0.01654115981917943 + }, + { + "kind": "HRM", + "file": "/home/yurenh2/rrm/research/flossing/diag_hrm_step_5208_512.npz", + "step": 5208, + "n": 512, + "k": 8, + "acc": 0.048828125, + "n_success": 25, + "n_failure": 487, + "raw_monotone_adjacent_fraction": 0.7368861607142857, + "raw_col0_is_sample_max_fraction": 0.798828125, + "raw_col0_success_mean": -0.18248083293437958, + "raw_col0_failure_mean": -0.1252742203170514, + "raw_col0_delta_failure_minus_success": 0.05720661261732818, + "raw_col0_auc_failure": 0.8788501026694046, + "lambda_max_success_mean": -0.18110741019248963, + "lambda_max_failure_mean": -0.12377287614975867, + "lambda_max_delta_failure_minus_success": 0.057334534042730964, + "lambda_max_auc_failure": 0.8758110882956879, + "mean8_success_mean": -0.22203324407339095, + "mean8_failure_mean": -0.16840960289919615, + "mean8_delta_failure_minus_success": 0.0536236411741948, + "mean8_auc_failure": 0.9536755646817249, + "tail_mean_5_8_success_mean": -0.24040259554982185, + "tail_mean_5_8_failure_mean": -0.1877109366994986, + "tail_mean_5_8_delta_failure_minus_success": 0.05269165885032326, + "tail_mean_5_8_auc_failure": 0.9394661190965092, + "positive_sum_success_mean": 0.0, + "positive_sum_failure_mean": 0.0, + "positive_sum_delta_failure_minus_success": 0.0, + "positive_sum_auc_failure": 0.5, + "positive_count_success_mean": 0.0, + "positive_count_failure_mean": 0.0, + "positive_count_delta_failure_minus_success": 0.0, + "positive_count_auc_failure": 0.5, + "spread_success_mean": 0.06972571730613708, + "spread_failure_mean": 0.07497160217019078, + "spread_delta_failure_minus_success": 0.005245884864053696, + "spread_auc_failure": 0.5605749486652978, + "gap12_success_mean": 0.01896007537841797, + "gap12_failure_mean": 0.0215270076581952, + "gap12_delta_failure_minus_success": 0.00256693227977723, + "gap12_auc_failure": 0.5172073921971253, + "lambda_max_corr_token_acc": -0.19170281345239037, + "mean8_corr_token_acc": -0.3935742986571875, + "tail_mean_5_8_corr_token_acc": -0.44039154437692224, + "positive_sum_corr_token_acc": NaN, + "positive_count_corr_token_acc": NaN + }, + { + "kind": "HRM", + "file": "/home/yurenh2/rrm/research/flossing/diag_hrm_step_7812_512.npz", + "step": 7812, + "n": 512, + "k": 8, + "acc": 0.15234375, + "n_success": 78, + "n_failure": 434, + "raw_monotone_adjacent_fraction": 0.7536272321428571, + "raw_col0_is_sample_max_fraction": 0.826171875, + "raw_col0_success_mean": -0.19744304562799442, + "raw_col0_failure_mean": -0.09758298332957993, + "raw_col0_delta_failure_minus_success": 0.09986006229841449, + "raw_col0_auc_failure": 0.9227519792035921, + "lambda_max_success_mean": -0.19638217866229704, + "lambda_max_failure_mean": -0.09583942211473957, + "lambda_max_delta_failure_minus_success": 0.10054275654755747, + "lambda_max_auc_failure": 0.9273307337823467, + "mean8_success_mean": -0.2626351442558166, + "mean8_failure_mean": -0.15164301766584318, + "mean8_delta_failure_minus_success": 0.1109921265899734, + "mean8_auc_failure": 0.9941214699279215, + "tail_mean_5_8_success_mean": -0.2882901179866913, + "tail_mean_5_8_failure_mean": -0.1752684197264127, + "tail_mean_5_8_delta_failure_minus_success": 0.11302169826027861, + "tail_mean_5_8_auc_failure": 0.9917582417582418, + "positive_sum_success_mean": 0.0, + "positive_sum_failure_mean": 0.0003272232148648253, + "positive_sum_delta_failure_minus_success": 0.0003272232148648253, + "positive_sum_auc_failure": 0.5138248847926268, + "positive_count_success_mean": 0.0, + "positive_count_failure_mean": 0.027649769585253458, + "positive_count_delta_failure_minus_success": 0.027649769585253458, + "positive_count_auc_failure": 0.5138248847926268, + "spread_success_mean": 0.10413952138370429, + "spread_failure_mean": 0.09246484074762525, + "spread_delta_failure_minus_success": -0.011674680636079043, + "spread_auc_failure": 0.4651719248493442, + "gap12_success_mean": 0.03829955409925718, + "gap12_failure_mean": 0.028275593728860324, + "gap12_delta_failure_minus_success": -0.010023960370396854, + "gap12_auc_failure": 0.4361928394186459, + "lambda_max_corr_token_acc": -0.538963951651198, + "mean8_corr_token_acc": -0.6951799773061379, + "tail_mean_5_8_corr_token_acc": -0.6975383575513661, + "positive_sum_corr_token_acc": -0.06688266082458418, + "positive_count_corr_token_acc": -0.07401140024719328 + }, + { + "kind": "HRM", + "file": "/home/yurenh2/rrm/research/flossing/diag_hrm_step_10416_512.npz", + "step": 10416, + "n": 512, + "k": 8, + "acc": 0.1796875, + "n_success": 92, + "n_failure": 420, + "raw_monotone_adjacent_fraction": 0.7572544642857143, + "raw_col0_is_sample_max_fraction": 0.841796875, + "raw_col0_success_mean": -0.1734262867020848, + "raw_col0_failure_mean": -0.0592695716998562, + "raw_col0_delta_failure_minus_success": 0.1141567150022286, + "raw_col0_auc_failure": 0.9278985507246377, + "lambda_max_success_mean": -0.1725775988408081, + "lambda_max_failure_mean": -0.05791120557031328, + "lambda_max_delta_failure_minus_success": 0.1146663932704948, + "lambda_max_auc_failure": 0.9305900621118013, + "mean8_success_mean": -0.22779642522547636, + "mean8_failure_mean": -0.11871804687665038, + "mean8_delta_failure_minus_success": 0.10907837834882597, + "mean8_auc_failure": 0.9611024844720497, + "tail_mean_5_8_success_mean": -0.25038242412974004, + "tail_mean_5_8_failure_mean": -0.1428806266010118, + "tail_mean_5_8_delta_failure_minus_success": 0.10750179752872824, + "tail_mean_5_8_auc_failure": 0.9642857142857143, + "positive_sum_success_mean": 0.0, + "positive_sum_failure_mean": 0.005729037300833235, + "positive_sum_delta_failure_minus_success": 0.005729037300833235, + "positive_sum_auc_failure": 0.580952380952381, + "positive_count_success_mean": 0.0, + "positive_count_failure_mean": 0.20476190476190476, + "positive_count_delta_failure_minus_success": 0.20476190476190476, + "positive_count_auc_failure": 0.580952380952381, + "spread_success_mean": 0.09030715536083216, + "spread_failure_mean": 0.09668695457983717, + "spread_delta_failure_minus_success": 0.006379799219005014, + "spread_auc_failure": 0.555667701863354, + "gap12_success_mean": 0.028749073491148327, + "gap12_failure_mean": 0.03182062698594693, + "gap12_delta_failure_minus_success": 0.003071553494798606, + "gap12_auc_failure": 0.5320393374741201, + "lambda_max_corr_token_acc": -0.4191290312385547, + "mean8_corr_token_acc": -0.47253422425178293, + "tail_mean_5_8_corr_token_acc": -0.4888621923983516, + "positive_sum_corr_token_acc": -0.0452883997405569, + "positive_count_corr_token_acc": -0.059274922302819555 + }, + { + "kind": "HRM", + "file": "/home/yurenh2/rrm/research/flossing/diag_hrm_step_13020_512.npz", + "step": 13020, + "n": 512, + "k": 8, + "acc": 0.30078125, + "n_success": 154, + "n_failure": 358, + "raw_monotone_adjacent_fraction": 0.7329799107142857, + "raw_col0_is_sample_max_fraction": 0.779296875, + "raw_col0_success_mean": -0.16317386656331334, + "raw_col0_failure_mean": -0.012345420176165918, + "raw_col0_delta_failure_minus_success": 0.15082844638714743, + "raw_col0_auc_failure": 0.9784879924544729, + "lambda_max_success_mean": -0.16130138939328179, + "lambda_max_failure_mean": -0.010260718098612039, + "lambda_max_delta_failure_minus_success": 0.15104067129466975, + "lambda_max_auc_failure": 0.9784698541681782, + "mean8_success_mean": -0.20199005053763794, + "mean8_failure_mean": -0.06562561695864545, + "mean8_delta_failure_minus_success": 0.13636443357899247, + "mean8_auc_failure": 0.9897337299571937, + "tail_mean_5_8_success_mean": -0.21928635291610057, + "tail_mean_5_8_failure_mean": -0.08840406447549293, + "tail_mean_5_8_delta_failure_minus_success": 0.13088228844060765, + "tail_mean_5_8_auc_failure": 0.9910578248567076, + "positive_sum_success_mean": 0.0, + "positive_sum_failure_mean": 0.03084314924092586, + "positive_sum_delta_failure_minus_success": 0.03084314924092586, + "positive_sum_auc_failure": 0.7067039106145251, + "positive_count_success_mean": 0.0, + "positive_count_failure_mean": 0.9134078212290503, + "positive_count_delta_failure_minus_success": 0.9134078212290503, + "positive_count_auc_failure": 0.7067039106145251, + "spread_success_mean": 0.06808332314294485, + "spread_failure_mean": 0.08949387786655578, + "spread_delta_failure_minus_success": 0.021410554723610933, + "spread_auc_failure": 0.6710440397591235, + "gap12_success_mean": 0.020334010348842756, + "gap12_failure_mean": 0.02894391759471739, + "gap12_delta_failure_minus_success": 0.008609907245874633, + "gap12_auc_failure": 0.5919248349415948, + "lambda_max_corr_token_acc": -0.6439558752057271, + "mean8_corr_token_acc": -0.6897349633120721, + "tail_mean_5_8_corr_token_acc": -0.6989600499840098, + "positive_sum_corr_token_acc": -0.14942778693878858, + "positive_count_corr_token_acc": -0.20197511884575872 + }, + { + "kind": "HRM", + "file": "/home/yurenh2/rrm/research/flossing/diag_hrm_step_15624_512.npz", + "step": 15624, + "n": 512, + "k": 8, + "acc": 0.333984375, + "n_success": 171, + "n_failure": 341, + "raw_monotone_adjacent_fraction": 0.7368861607142857, + "raw_col0_is_sample_max_fraction": 0.78515625, + "raw_col0_success_mean": -0.17926316590140962, + "raw_col0_failure_mean": 0.013238977896144526, + "raw_col0_delta_failure_minus_success": 0.19250214379755415, + "raw_col0_auc_failure": 0.9800209222959647, + "lambda_max_success_mean": -0.1768394878504482, + "lambda_max_failure_mean": 0.015099927341163328, + "lambda_max_delta_failure_minus_success": 0.19193941519161153, + "lambda_max_auc_failure": 0.9809298417108264, + "mean8_success_mean": -0.2279172917539989, + "mean8_failure_mean": -0.04059797345319265, + "mean8_delta_failure_minus_success": 0.18731931830080625, + "mean8_auc_failure": 0.9898475416302241, + "tail_mean_5_8_success_mean": -0.24842964750796295, + "tail_mean_5_8_failure_mean": -0.06421948799691395, + "tail_mean_5_8_delta_failure_minus_success": 0.18421015951104902, + "tail_mean_5_8_auc_failure": 0.9895731508634734, + "positive_sum_success_mean": 0.0007992622970837598, + "positive_sum_failure_mean": 0.054435018036423026, + "positive_sum_delta_failure_minus_success": 0.053635755739339264, + "positive_sum_auc_failure": 0.7755826516437722, + "positive_count_success_mean": 0.023391812865497075, + "positive_count_failure_mean": 1.5073313782991202, + "positive_count_delta_failure_minus_success": 1.4839395654336232, + "positive_count_auc_failure": 0.7767059388451578, + "spread_success_mean": 0.08114021638005275, + "spread_failure_mean": 0.0924473905291844, + "spread_delta_failure_minus_success": 0.01130717414913164, + "spread_auc_failure": 0.5924782631064465, + "gap12_success_mean": 0.026074302491693818, + "gap12_failure_mean": 0.027355855661720895, + "gap12_delta_failure_minus_success": 0.001281553170027077, + "gap12_auc_failure": 0.5126305499819931, + "lambda_max_corr_token_acc": -0.7050637096452648, + "mean8_corr_token_acc": -0.7549696612898014, + "tail_mean_5_8_corr_token_acc": -0.758163657007401, + "positive_sum_corr_token_acc": -0.24676472967716626, + "positive_count_corr_token_acc": -0.2994831724010142 + }, + { + "kind": "HRM", + "file": "/home/yurenh2/rrm/research/flossing/diag_hrm_step_18228_512.npz", + "step": 18228, + "n": 512, + "k": 8, + "acc": 0.474609375, + "n_success": 243, + "n_failure": 269, + "raw_monotone_adjacent_fraction": 0.7474888392857143, + "raw_col0_is_sample_max_fraction": 0.818359375, + "raw_col0_success_mean": -0.07771555354618448, + "raw_col0_failure_mean": 0.016638744400673956, + "raw_col0_delta_failure_minus_success": 0.09435429794685843, + "raw_col0_auc_failure": 0.8841617329845335, + "lambda_max_success_mean": -0.07614383449056474, + "lambda_max_failure_mean": 0.018023035920062363, + "lambda_max_delta_failure_minus_success": 0.0941668704106271, + "lambda_max_auc_failure": 0.8879250998210106, + "mean8_success_mean": -0.14575748546156192, + "mean8_failure_mean": -0.03353973620840138, + "mean8_delta_failure_minus_success": 0.11221774925316054, + "mean8_auc_failure": 0.9716523628130402, + "tail_mean_5_8_success_mean": -0.1721063231313486, + "tail_mean_5_8_failure_mean": -0.05434106856267585, + "tail_mean_5_8_delta_failure_minus_success": 0.11776525456867276, + "tail_mean_5_8_auc_failure": 0.9799899031621461, + "positive_sum_success_mean": 0.00525456639471437, + "positive_sum_failure_mean": 0.050987723279165634, + "positive_sum_delta_failure_minus_success": 0.045733156884451266, + "positive_sum_auc_failure": 0.749567824743372, + "positive_count_success_mean": 0.13580246913580246, + "positive_count_failure_mean": 1.4981412639405205, + "positive_count_delta_failure_minus_success": 1.362338794804718, + "positive_count_auc_failure": 0.7569645233833586, + "spread_success_mean": 0.10785400053824455, + "spread_failure_mean": 0.0830327885870487, + "spread_delta_failure_minus_success": -0.024821211951195854, + "spread_auc_failure": 0.35938623464439245, + "gap12_success_mean": 0.03951045055418571, + "gap12_failure_mean": 0.027345439433070017, + "gap12_delta_failure_minus_success": -0.01216501112111569, + "gap12_auc_failure": 0.4258111891321309, + "lambda_max_corr_token_acc": -0.6086543875688578, + "mean8_corr_token_acc": -0.7665403099919734, + "tail_mean_5_8_corr_token_acc": -0.786002777751629, + "positive_sum_corr_token_acc": -0.3211841461864595, + "positive_count_corr_token_acc": -0.4146342203458471 + }, + { + "kind": "HRM", + "file": "/home/yurenh2/rrm/research/flossing/diag_hrm_step_20832_512.npz", + "step": 20832, + "n": 512, + "k": 8, + "acc": 0.45703125, + "n_success": 234, + "n_failure": 278, + "raw_monotone_adjacent_fraction": 0.7497209821428571, + "raw_col0_is_sample_max_fraction": 0.80078125, + "raw_col0_success_mean": -0.0847591630423951, + "raw_col0_failure_mean": 0.011107698112080065, + "raw_col0_delta_failure_minus_success": 0.09586686115447517, + "raw_col0_auc_failure": 0.9046762589928058, + "lambda_max_success_mean": -0.08308592252134799, + "lambda_max_failure_mean": 0.012636878445990327, + "lambda_max_delta_failure_minus_success": 0.09572280096733832, + "lambda_max_auc_failure": 0.9064287031912931, + "mean8_success_mean": -0.14093293187680603, + "mean8_failure_mean": -0.044163071447564366, + "mean8_delta_failure_minus_success": 0.09676986042924166, + "mean8_auc_failure": 0.9825831642378405, + "tail_mean_5_8_success_mean": -0.16335714651812983, + "tail_mean_5_8_failure_mean": -0.06769531056555371, + "tail_mean_5_8_delta_failure_minus_success": 0.09566183595257612, + "tail_mean_5_8_auc_failure": 0.9836438541474513, + "positive_sum_success_mean": 0.002644224575935648, + "positive_sum_failure_mean": 0.04743396344555085, + "positive_sum_delta_failure_minus_success": 0.0447897388696152, + "positive_sum_auc_failure": 0.7440432269568961, + "positive_count_success_mean": 0.0811965811965812, + "positive_count_failure_mean": 1.2589928057553956, + "positive_count_delta_failure_minus_success": 1.1777962245588145, + "positive_count_auc_failure": 0.747501998401279, + "spread_success_mean": 0.0904661258792059, + "spread_failure_mean": 0.09196829576401849, + "spread_delta_failure_minus_success": 0.0015021698848125958, + "spread_auc_failure": 0.5366168603578676, + "gap12_success_mean": 0.03158770251569426, + "gap12_failure_mean": 0.029151334776909496, + "gap12_delta_failure_minus_success": -0.0024363677387847643, + "gap12_auc_failure": 0.49451208264157903, + "lambda_max_corr_token_acc": -0.5961668167415688, + "mean8_corr_token_acc": -0.7490505096842796, + "tail_mean_5_8_corr_token_acc": -0.7670744302809093, + "positive_sum_corr_token_acc": -0.25827536960316083, + "positive_count_corr_token_acc": -0.3352751212871484 + }, + { + "kind": "HRM", + "file": "/home/yurenh2/rrm/research/flossing/diag_hrm_step_23436_512.npz", + "step": 23436, + "n": 512, + "k": 8, + "acc": 0.505859375, + "n_success": 259, + "n_failure": 253, + "raw_monotone_adjacent_fraction": 0.7444196428571429, + "raw_col0_is_sample_max_fraction": 0.8125, + "raw_col0_success_mean": -0.1396155291305258, + "raw_col0_failure_mean": -0.002783365376173506, + "raw_col0_delta_failure_minus_success": 0.13683216375435228, + "raw_col0_auc_failure": 0.9710501014848841, + "lambda_max_success_mean": -0.1384992553999928, + "lambda_max_failure_mean": -0.0009893162803058476, + "lambda_max_delta_failure_minus_success": 0.13750993911968695, + "lambda_max_auc_failure": 0.9724541028888855, + "mean8_success_mean": -0.2008695753030298, + "mean8_failure_mean": -0.05231515268090541, + "mean8_delta_failure_minus_success": 0.14855442262212437, + "mean8_auc_failure": 0.9922932531628184, + "tail_mean_5_8_success_mean": -0.22361706499433173, + "tail_mean_5_8_failure_mean": -0.07373835104911058, + "tail_mean_5_8_delta_failure_minus_success": 0.14987871394522115, + "tail_mean_5_8_auc_failure": 0.9920032963511224, + "positive_sum_success_mean": 0.0005239612893938558, + "positive_sum_failure_mean": 0.02692777939243448, + "positive_sum_delta_failure_minus_success": 0.026403818103040624, + "positive_sum_auc_failure": 0.7254032688815297, + "positive_count_success_mean": 0.019305019305019305, + "positive_count_failure_mean": 0.8932806324110671, + "positive_count_delta_failure_minus_success": 0.8739756131060479, + "positive_count_auc_failure": 0.7251133120698338, + "spread_success_mean": 0.09569484527981534, + "spread_failure_mean": 0.08370462378894082, + "spread_delta_failure_minus_success": -0.011990221490874517, + "spread_auc_failure": 0.4474338822164909, + "gap12_success_mean": 0.0377779275387468, + "gap12_failure_mean": 0.025563243349588635, + "gap12_delta_failure_minus_success": -0.012214684189158165, + "gap12_auc_failure": 0.42294016207059687, + "lambda_max_corr_token_acc": -0.735475615818423, + "mean8_corr_token_acc": -0.8331365987869843, + "tail_mean_5_8_corr_token_acc": -0.8388368318549599, + "positive_sum_corr_token_acc": -0.2844985120608491, + "positive_count_corr_token_acc": -0.35566859763323194 + }, + { + "kind": "HRM", + "file": "/home/yurenh2/rrm/research/flossing/diag_hrm_step_26040_512.npz", + "step": 26040, + "n": 512, + "k": 8, + "acc": 0.5, + "n_success": 256, + "n_failure": 256, + "raw_monotone_adjacent_fraction": 0.748046875, + "raw_col0_is_sample_max_fraction": 0.787109375, + "raw_col0_success_mean": -0.14642834789538028, + "raw_col0_failure_mean": 0.03254939207545249, + "raw_col0_delta_failure_minus_success": 0.17897773997083277, + "raw_col0_auc_failure": 0.9890289306640625, + "lambda_max_success_mean": -0.14456235250236205, + "lambda_max_failure_mean": 0.034372387226426326, + "lambda_max_delta_failure_minus_success": 0.17893473972878837, + "lambda_max_auc_failure": 0.9896240234375, + "mean8_success_mean": -0.18957092847483636, + "mean8_failure_mean": -0.025070655147768406, + "mean8_delta_failure_minus_success": 0.16450027332706796, + "mean8_auc_failure": 0.997589111328125, + "tail_mean_5_8_success_mean": -0.21117772247089306, + "tail_mean_5_8_failure_mean": -0.04990981457450516, + "tail_mean_5_8_delta_failure_minus_success": 0.1612679078963879, + "tail_mean_5_8_auc_failure": 0.9984893798828125, + "positive_sum_success_mean": 0.00045363006938714534, + "positive_sum_failure_mean": 0.074183922140719, + "positive_sum_delta_failure_minus_success": 0.07373029207133186, + "positive_sum_auc_failure": 0.865997314453125, + "positive_count_success_mean": 0.01953125, + "positive_count_failure_mean": 1.94140625, + "positive_count_delta_failure_minus_success": 1.921875, + "positive_count_auc_failure": 0.8662796020507812, + "spread_success_mean": 0.07883575186588132, + "spread_failure_mean": 0.09741729416225553, + "spread_delta_failure_minus_success": 0.01858154229637421, + "spread_auc_failure": 0.62774658203125, + "gap12_success_mean": 0.01608559737815085, + "gap12_failure_mean": 0.02983991140987996, + "gap12_delta_failure_minus_success": 0.013754314031729109, + "gap12_auc_failure": 0.631378173828125, + "lambda_max_corr_token_acc": -0.8022434699246608, + "mean8_corr_token_acc": -0.8493972056528871, + "tail_mean_5_8_corr_token_acc": -0.8567110926525265, + "positive_sum_corr_token_acc": -0.37718525164073874, + "positive_count_corr_token_acc": -0.4879767533856408 + }, + { + "kind": "TRM", + "file": "/home/yurenh2/rrm/research/flossing/diag_trm_singleGPU_step26041_512.npz", + "step": 26041, + "n": 512, + "k": 8, + "acc": 0.576171875, + "n_success": 295, + "n_failure": 217, + "raw_monotone_adjacent_fraction": 0.7438616071428571, + "raw_col0_is_sample_max_fraction": 0.791015625, + "raw_col0_success_mean": -0.03554810737259686, + "raw_col0_failure_mean": 0.034024751214321254, + "raw_col0_delta_failure_minus_success": 0.06957285858691811, + "raw_col0_auc_failure": 0.9836132156525814, + "lambda_max_success_mean": -0.03484145596473462, + "lambda_max_failure_mean": 0.034498983389517714, + "lambda_max_delta_failure_minus_success": 0.06934043935425233, + "lambda_max_auc_failure": 0.9838475357338123, + "mean8_success_mean": -0.044594623737682894, + "mean8_failure_mean": 0.011831644308724274, + "mean8_delta_failure_minus_success": 0.05642626804640717, + "mean8_auc_failure": 0.9900960712333047, + "tail_mean_5_8_success_mean": -0.04878217898325509, + "tail_mean_5_8_failure_mean": 0.002079399892583064, + "tail_mean_5_8_delta_failure_minus_success": 0.050861578875838157, + "tail_mean_5_8_auc_failure": 0.9916582051081778, + "positive_sum_success_mean": 0.002200093488031367, + "positive_sum_failure_mean": 0.11402314361281402, + "positive_sum_delta_failure_minus_success": 0.11182305012478265, + "positive_sum_auc_failure": 0.9693353120362415, + "positive_count_success_mean": 0.1694915254237288, + "positive_count_failure_mean": 5.539170506912442, + "positive_count_delta_failure_minus_success": 5.369678981488713, + "positive_count_auc_failure": 0.9706006404748887, + "spread_success_mean": 0.01622369096148759, + "spread_failure_mean": 0.03719716329264318, + "spread_delta_failure_minus_success": 0.02097347233115559, + "spread_auc_failure": 0.8946653128173084, + "gap12_success_mean": 0.004580460641320037, + "gap12_failure_mean": 0.010477446547843793, + "gap12_delta_failure_minus_success": 0.005896985906523756, + "gap12_auc_failure": 0.7136608607357651, + "lambda_max_corr_token_acc": -0.8221535026414836, + "mean8_corr_token_acc": -0.8532299277645251, + "tail_mean_5_8_corr_token_acc": -0.8618572814599218, + "positive_sum_corr_token_acc": -0.6425677667973041, + "positive_count_corr_token_acc": -0.7985750358664757 + }, + { + "kind": "TRM", + "file": "/home/yurenh2/rrm/research/flossing/diag_trm_singleGPU_step52082_512.npz", + "step": 52082, + "n": 512, + "k": 8, + "acc": 0.6484375, + "n_success": 332, + "n_failure": 180, + "raw_monotone_adjacent_fraction": 0.7435825892857143, + "raw_col0_is_sample_max_fraction": 0.78125, + "raw_col0_success_mean": -0.0007075830156292839, + "raw_col0_failure_mean": 0.05957649097674423, + "raw_col0_delta_failure_minus_success": 0.06028407399237352, + "raw_col0_auc_failure": 0.981425702811245, + "lambda_max_success_mean": -0.00011202936298378704, + "lambda_max_failure_mean": 0.059987526981987886, + "lambda_max_delta_failure_minus_success": 0.060099556344971675, + "lambda_max_auc_failure": 0.9824129852744311, + "mean8_success_mean": -0.00932737177132882, + "mean8_failure_mean": 0.03324007580221304, + "mean8_delta_failure_minus_success": 0.04256744757354186, + "mean8_auc_failure": 0.993591030789826, + "tail_mean_5_8_success_mean": -0.01301439359083114, + "tail_mean_5_8_failure_mean": 0.02160164021022663, + "tail_mean_5_8_delta_failure_minus_success": 0.034616033801057766, + "tail_mean_5_8_auc_failure": 0.9948627844712182, + "positive_sum_success_mean": 0.011575991871503936, + "positive_sum_failure_mean": 0.26611879217404444, + "positive_sum_delta_failure_minus_success": 0.2545428003025405, + "positive_sum_auc_failure": 0.9932730923694779, + "positive_count_success_mean": 0.9879518072289156, + "positive_count_failure_mean": 7.916666666666667, + "positive_count_delta_failure_minus_success": 6.9287148594377514, + "positive_count_auc_failure": 0.9850485274431058, + "spread_success_mean": 0.014868822472832896, + "spread_failure_mean": 0.04419851013889355, + "spread_delta_failure_minus_success": 0.029329687666060658, + "spread_auc_failure": 0.9461680053547523, + "gap12_success_mean": 0.0048919110970939116, + "gap12_failure_mean": 0.01239234626862324, + "gap12_delta_failure_minus_success": 0.007500435171529329, + "gap12_auc_failure": 0.7928547523427042, + "lambda_max_corr_token_acc": -0.822862756235417, + "mean8_corr_token_acc": -0.8543835399263785, + "tail_mean_5_8_corr_token_acc": -0.8502975145711263, + "positive_sum_corr_token_acc": -0.83615142839377, + "positive_count_corr_token_acc": -0.8756439805499547 + }, + { + "kind": "TRM", + "file": "/home/yurenh2/rrm/research/flossing/diag_trm_singleGPU_step78123_512.npz", + "step": 78123, + "n": 512, + "k": 8, + "acc": 0.6875, + "n_success": 352, + "n_failure": 160, + "raw_monotone_adjacent_fraction": 0.7522321428571429, + "raw_col0_is_sample_max_fraction": 0.94921875, + "raw_col0_success_mean": -0.0022565727597132286, + "raw_col0_failure_mean": 0.08176288979593664, + "raw_col0_delta_failure_minus_success": 0.08401946255564988, + "raw_col0_auc_failure": 0.9937144886363637, + "lambda_max_success_mean": -0.002104297534184628, + "lambda_max_failure_mean": 0.08205510303378105, + "lambda_max_delta_failure_minus_success": 0.08415940056796568, + "lambda_max_auc_failure": 0.99375, + "mean8_success_mean": -0.03085258081190225, + "mean8_failure_mean": 0.04898583621179568, + "mean8_delta_failure_minus_success": 0.07983841702369793, + "mean8_auc_failure": 0.9991299715909091, + "tail_mean_5_8_success_mean": -0.03740866408387236, + "tail_mean_5_8_failure_mean": 0.03495748526038369, + "tail_mean_5_8_delta_failure_minus_success": 0.07236614934425606, + "tail_mean_5_8_auc_failure": 0.9993607954545455, + "positive_sum_success_mean": 0.011536654059641065, + "positive_sum_failure_mean": 0.39188668969436546, + "positive_sum_delta_failure_minus_success": 0.3803500356347244, + "positive_sum_auc_failure": 0.9991299715909091, + "positive_count_success_mean": 0.6619318181818182, + "positive_count_failure_mean": 8.0, + "positive_count_delta_failure_minus_success": 7.338068181818182, + "positive_count_auc_failure": 0.9900568181818182, + "spread_success_mean": 0.037154979946411586, + "spread_failure_mean": 0.05358391968184151, + "spread_delta_failure_minus_success": 0.016428939735429922, + "spread_auc_failure": 0.8060191761363636, + "gap12_success_mean": 0.026502143868418152, + "gap12_failure_mean": 0.015829400252550842, + "gap12_delta_failure_minus_success": -0.01067274361586731, + "gap12_auc_failure": 0.2559303977272727, + "lambda_max_corr_token_acc": -0.8695160653352085, + "mean8_corr_token_acc": -0.9167732333138554, + "tail_mean_5_8_corr_token_acc": -0.9212713093750954, + "positive_sum_corr_token_acc": -0.9333667142066921, + "positive_count_corr_token_acc": -0.916788261375625 + }, + { + "kind": "TRM", + "file": "/home/yurenh2/rrm/research/flossing/diag_trm_singleGPU_step104164_512.npz", + "step": 104164, + "n": 512, + "k": 8, + "acc": 0.7109375, + "n_success": 364, + "n_failure": 148, + "raw_monotone_adjacent_fraction": 0.7659040178571429, + "raw_col0_is_sample_max_fraction": 0.873046875, + "raw_col0_success_mean": 0.0007209903823832564, + "raw_col0_failure_mean": 0.09478743761979244, + "raw_col0_delta_failure_minus_success": 0.09406644723740919, + "raw_col0_auc_failure": 0.9850942975942976, + "lambda_max_success_mean": 0.0011589818968986748, + "lambda_max_failure_mean": 0.0948834198753576, + "lambda_max_delta_failure_minus_success": 0.09372443797845893, + "lambda_max_auc_failure": 0.9852242352242352, + "mean8_success_mean": -0.013257209616115657, + "mean8_failure_mean": 0.05609436892523632, + "mean8_delta_failure_minus_success": 0.06935157854135197, + "mean8_auc_failure": 0.9912013662013662, + "tail_mean_5_8_success_mean": -0.018433471481510245, + "tail_mean_5_8_failure_mean": 0.04002377992288235, + "tail_mean_5_8_delta_failure_minus_success": 0.05845725140439259, + "tail_mean_5_8_auc_failure": 0.9914241164241164, + "positive_sum_success_mean": 0.024264690904521637, + "positive_sum_failure_mean": 0.44875495140189053, + "positive_sum_delta_failure_minus_success": 0.4244902604973689, + "positive_sum_auc_failure": 0.9912013662013662, + "positive_count_success_mean": 1.0631868131868132, + "positive_count_failure_mean": 8.0, + "positive_count_delta_failure_minus_success": 6.936813186813187, + "positive_count_auc_failure": 0.9766483516483516, + "spread_success_mean": 0.02201517710918679, + "spread_failure_mean": 0.06216927877048383, + "spread_delta_failure_minus_success": 0.04015410166129704, + "spread_auc_failure": 0.9606660231660231, + "gap12_success_mean": 0.008753806905887968, + "gap12_failure_mean": 0.020252479617861478, + "gap12_delta_failure_minus_success": 0.01149867271197351, + "gap12_auc_failure": 0.8213171963171964, + "lambda_max_corr_token_acc": -0.8714207860201784, + "mean8_corr_token_acc": -0.8978503025935065, + "tail_mean_5_8_corr_token_acc": -0.9018662195261188, + "positive_sum_corr_token_acc": -0.9164290683745636, + "positive_count_corr_token_acc": -0.8552081309221397 + }, + { + "kind": "TRM", + "file": "/home/yurenh2/rrm/research/flossing/diag_trm_singleGPU_step130205_512.npz", + "step": 130205, + "n": 512, + "k": 8, + "acc": 0.755859375, + "n_success": 387, + "n_failure": 125, + "raw_monotone_adjacent_fraction": 0.7661830357142857, + "raw_col0_is_sample_max_fraction": 0.83203125, + "raw_col0_success_mean": 0.013444696512909148, + "raw_col0_failure_mean": 0.10679112273454666, + "raw_col0_delta_failure_minus_success": 0.09334642622163751, + "raw_col0_auc_failure": 0.988857881136951, + "lambda_max_success_mean": 0.014031398803655426, + "lambda_max_failure_mean": 0.10683381146192551, + "lambda_max_delta_failure_minus_success": 0.09280241265827009, + "lambda_max_auc_failure": 0.9889198966408269, + "mean8_success_mean": 0.0006696076976612102, + "mean8_failure_mean": 0.05997022696118802, + "mean8_delta_failure_minus_success": 0.05930061926352681, + "mean8_auc_failure": 0.9935503875968992, + "tail_mean_5_8_success_mean": -0.004693869635777962, + "tail_mean_5_8_failure_mean": 0.04096000522933901, + "tail_mean_5_8_delta_failure_minus_success": 0.04565387486511697, + "tail_mean_5_8_auc_failure": 0.9943152454780362, + "positive_sum_success_mean": 0.04017261892318711, + "positive_sum_failure_mean": 0.47976181568950416, + "positive_sum_delta_failure_minus_success": 0.439589196766317, + "positive_sum_auc_failure": 0.9935503875968992, + "positive_count_success_mean": 2.4108527131782944, + "positive_count_failure_mean": 8.0, + "positive_count_delta_failure_minus_success": 5.589147286821706, + "positive_count_auc_failure": 0.9521963824289406, + "spread_success_mean": 0.021535266655820777, + "spread_failure_mean": 0.07427249775826932, + "spread_delta_failure_minus_success": 0.05273723110244854, + "spread_auc_failure": 0.9726511627906976, + "gap12_success_mean": 0.007057571094765433, + "gap12_failure_mean": 0.02309795269370079, + "gap12_delta_failure_minus_success": 0.016040381598935356, + "gap12_auc_failure": 0.8932093023255814, + "lambda_max_corr_token_acc": -0.868494410453853, + "mean8_corr_token_acc": -0.8995673015890842, + "tail_mean_5_8_corr_token_acc": -0.9101679267135595, + "positive_sum_corr_token_acc": -0.9091943845695494, + "positive_count_corr_token_acc": -0.7232326485538275 + }, + { + "kind": "TRM", + "file": "/home/yurenh2/rrm/research/flossing/diag_trm_singleGPU_step156246_512.npz", + "step": 156246, + "n": 512, + "k": 8, + "acc": 0.7578125, + "n_success": 388, + "n_failure": 124, + "raw_monotone_adjacent_fraction": 0.8465401785714286, + "raw_col0_is_sample_max_fraction": 0.984375, + "raw_col0_success_mean": 0.033355438992845106, + "raw_col0_failure_mean": 0.1080104663006721, + "raw_col0_delta_failure_minus_success": 0.074655027307827, + "raw_col0_auc_failure": 0.9934527768540073, + "lambda_max_success_mean": 0.03340256920781724, + "lambda_max_failure_mean": 0.10808430672172577, + "lambda_max_delta_failure_minus_success": 0.07468173751390852, + "lambda_max_auc_failure": 0.9935359161955437, + "mean8_success_mean": -0.01719525050584332, + "mean8_failure_mean": 0.059510113856841566, + "mean8_delta_failure_minus_success": 0.07670536436268488, + "mean8_auc_failure": 0.9990646824077153, + "tail_mean_5_8_success_mean": -0.03258687625807819, + "tail_mean_5_8_failure_mean": 0.03992759636434306, + "tail_mean_5_8_delta_failure_minus_success": 0.07251447262242125, + "tail_mean_5_8_auc_failure": 0.9987944795477219, + "positive_sum_success_mean": 0.0501136147264783, + "positive_sum_failure_mean": 0.47608091085473253, + "positive_sum_delta_failure_minus_success": 0.42596729612825424, + "positive_sum_auc_failure": 0.9990646824077153, + "positive_count_success_mean": 1.7268041237113403, + "positive_count_failure_mean": 8.0, + "positive_count_delta_failure_minus_success": 6.27319587628866, + "positive_count_auc_failure": 0.9806701030927835, + "spread_success_mean": 0.07044461364565689, + "spread_failure_mean": 0.07696695457167563, + "spread_delta_failure_minus_success": 0.006522340926018735, + "spread_auc_failure": 0.5740563684735617, + "gap12_success_mean": 0.03767048126359648, + "gap12_failure_mean": 0.024864713691415324, + "gap12_delta_failure_minus_success": -0.012805767572181152, + "gap12_auc_failure": 0.30377036913867644, + "lambda_max_corr_token_acc": -0.8015386582153512, + "mean8_corr_token_acc": -0.8887892447157182, + "tail_mean_5_8_corr_token_acc": -0.8908349686685592, + "positive_sum_corr_token_acc": -0.9311284896491179, + "positive_count_corr_token_acc": -0.8411922875634292 + }, + { + "kind": "TRM", + "file": "/home/yurenh2/rrm/research/flossing/diag_trm_singleGPU_step182287_512.npz", + "step": 182287, + "n": 512, + "k": 8, + "acc": 0.7421875, + "n_success": 380, + "n_failure": 132, + "raw_monotone_adjacent_fraction": 0.7837611607142857, + "raw_col0_is_sample_max_fraction": 0.9296875, + "raw_col0_success_mean": 0.00821161610291902, + "raw_col0_failure_mean": 0.10131858266664273, + "raw_col0_delta_failure_minus_success": 0.0931069665637237, + "raw_col0_auc_failure": 0.9476076555023923, + "lambda_max_success_mean": 0.008431412303494461, + "lambda_max_failure_mean": 0.10132696061874881, + "lambda_max_delta_failure_minus_success": 0.09289554831525434, + "lambda_max_auc_failure": 0.9476275917065391, + "mean8_success_mean": -0.018691345271909503, + "mean8_failure_mean": 0.049928791549256966, + "mean8_delta_failure_minus_success": 0.06862013682116647, + "mean8_auc_failure": 0.993122009569378, + "tail_mean_5_8_success_mean": -0.026891200006043378, + "tail_mean_5_8_failure_mean": 0.029759585490286223, + "tail_mean_5_8_delta_failure_minus_success": 0.0566507854963296, + "tail_mean_5_8_auc_failure": 0.993421052631579, + "positive_sum_success_mean": 0.027419095844602212, + "positive_sum_failure_mean": 0.3994350857048465, + "positive_sum_delta_failure_minus_success": 0.37201598986024426, + "positive_sum_auc_failure": 0.993122009569378, + "positive_count_success_mean": 0.8973684210526316, + "positive_count_failure_mean": 7.992424242424242, + "positive_count_delta_failure_minus_success": 7.095055821371611, + "positive_count_auc_failure": 0.9867125199362041, + "spread_success_mean": 0.03768573965873513, + "spread_failure_mean": 0.07997999362255954, + "spread_delta_failure_minus_success": 0.04229425396382441, + "spread_auc_failure": 0.8919258373205742, + "gap12_success_mean": 0.019669575117656057, + "gap12_failure_mean": 0.027337144778081864, + "gap12_delta_failure_minus_success": 0.007667569660425807, + "gap12_auc_failure": 0.7587121212121212, + "lambda_max_corr_token_acc": -0.7443356021278564, + "mean8_corr_token_acc": -0.8907098916862968, + "tail_mean_5_8_corr_token_acc": -0.9075392398447194, + "positive_sum_corr_token_acc": -0.9063416990478439, + "positive_count_corr_token_acc": -0.8813795117907826 + }, + { + "kind": "TRM", + "file": "/home/yurenh2/rrm/research/flossing/diag_trm_singleGPU_step208328_512.npz", + "step": 208328, + "n": 512, + "k": 8, + "acc": 0.75, + "n_success": 384, + "n_failure": 128, + "raw_monotone_adjacent_fraction": 0.8138950892857143, + "raw_col0_is_sample_max_fraction": 0.970703125, + "raw_col0_success_mean": 0.03753090985757505, + "raw_col0_failure_mean": 0.10676543856970966, + "raw_col0_delta_failure_minus_success": 0.06923452871213462, + "raw_col0_auc_failure": 0.9824422200520834, + "lambda_max_success_mean": 0.037651085844269495, + "lambda_max_failure_mean": 0.10677664534887299, + "lambda_max_delta_failure_minus_success": 0.0691255595046035, + "lambda_max_auc_failure": 0.9825439453125, + "mean8_success_mean": -0.003334924222756793, + "mean8_failure_mean": 0.0556086836418217, + "mean8_delta_failure_minus_success": 0.058943607864578494, + "mean8_auc_failure": 0.997314453125, + "tail_mean_5_8_success_mean": -0.015420041531266785, + "tail_mean_5_8_failure_mean": 0.03501016539212287, + "tail_mean_5_8_delta_failure_minus_success": 0.050430206923389655, + "tail_mean_5_8_auc_failure": 0.9977620442708334, + "positive_sum_success_mean": 0.0587247395355727, + "positive_sum_failure_mean": 0.4448694691345736, + "positive_sum_delta_failure_minus_success": 0.38614472959900087, + "positive_sum_auc_failure": 0.997314453125, + "positive_count_success_mean": 2.3385416666666665, + "positive_count_failure_mean": 8.0, + "positive_count_delta_failure_minus_success": 5.661458333333334, + "positive_count_auc_failure": 0.97265625, + "spread_success_mean": 0.05609253943835787, + "spread_failure_mean": 0.08039409609409631, + "spread_delta_failure_minus_success": 0.024301556655738445, + "spread_auc_failure": 0.8315836588541666, + "gap12_success_mean": 0.029487703329541166, + "gap12_failure_mean": 0.026957579655572772, + "gap12_delta_failure_minus_success": -0.0025301236739683937, + "gap12_auc_failure": 0.4844767252604167, + "lambda_max_corr_token_acc": -0.7847679828439686, + "mean8_corr_token_acc": -0.9025430240366568, + "tail_mean_5_8_corr_token_acc": -0.9116696180830379, + "positive_sum_corr_token_acc": -0.918178990916473, + "positive_count_corr_token_acc": -0.8241555473543875 + }, + { + "kind": "TRM", + "file": "/home/yurenh2/rrm/research/flossing/diag_trm_singleGPU_step234369_512.npz", + "step": 234369, + "n": 512, + "k": 8, + "acc": 0.7734375, + "n_success": 396, + "n_failure": 116, + "raw_monotone_adjacent_fraction": 0.7837611607142857, + "raw_col0_is_sample_max_fraction": 0.943359375, + "raw_col0_success_mean": 0.017852102687761037, + "raw_col0_failure_mean": 0.10551282866247769, + "raw_col0_delta_failure_minus_success": 0.08766072597471665, + "raw_col0_auc_failure": 0.9871125043538836, + "lambda_max_success_mean": 0.018100257435352516, + "lambda_max_failure_mean": 0.10551282866247769, + "lambda_max_delta_failure_minus_success": 0.08741257122712517, + "lambda_max_auc_failure": 0.9871125043538836, + "mean8_success_mean": -0.011827684605397524, + "mean8_failure_mean": 0.05244761205414824, + "mean8_delta_failure_minus_success": 0.06427529665954576, + "mean8_auc_failure": 0.9948624172762104, + "tail_mean_5_8_success_mean": -0.020207434739923615, + "tail_mean_5_8_failure_mean": 0.03131334408957126, + "tail_mean_5_8_delta_failure_minus_success": 0.05152077882949488, + "tail_mean_5_8_auc_failure": 0.995276036224312, + "positive_sum_success_mean": 0.03539385316928405, + "positive_sum_failure_mean": 0.4195808964331859, + "positive_sum_delta_failure_minus_success": 0.38418704326390185, + "positive_sum_auc_failure": 0.9948624172762104, + "positive_count_success_mean": 1.4595959595959596, + "positive_count_failure_mean": 8.0, + "positive_count_delta_failure_minus_success": 6.540404040404041, + "positive_count_auc_failure": 0.976010101010101, + "spread_success_mean": 0.04073868896878162, + "spread_failure_mean": 0.08256442746115399, + "spread_delta_failure_minus_success": 0.041825738492372366, + "spread_auc_failure": 0.9344740508533612, + "gap12_success_mean": 0.022704324246853612, + "gap12_failure_mean": 0.026842519994182832, + "gap12_delta_failure_minus_success": 0.00413819574732922, + "gap12_auc_failure": 0.6134839777081157, + "lambda_max_corr_token_acc": -0.8095679410757006, + "mean8_corr_token_acc": -0.8841140003061905, + "tail_mean_5_8_corr_token_acc": -0.8897475061573502, + "positive_sum_corr_token_acc": -0.9111627943782784, + "positive_count_corr_token_acc": -0.8264057387186906 + }, + { + "kind": "TRM", + "file": "/home/yurenh2/rrm/research/flossing/diag_trm_singleGPU_step260410_512.npz", + "step": 260410, + "n": 512, + "k": 8, + "acc": 0.76953125, + "n_success": 394, + "n_failure": 118, + "raw_monotone_adjacent_fraction": 0.7299107142857143, + "raw_col0_is_sample_max_fraction": 0.767578125, + "raw_col0_success_mean": 0.012635910749513134, + "raw_col0_failure_mean": 0.10288177159125522, + "raw_col0_delta_failure_minus_success": 0.09024586084174209, + "raw_col0_auc_failure": 0.9893530069689409, + "lambda_max_success_mean": 0.013151870022697326, + "lambda_max_failure_mean": 0.10296485964524543, + "lambda_max_delta_failure_minus_success": 0.0898129896225481, + "lambda_max_auc_failure": 0.9894175341994321, + "mean8_success_mean": 0.0016685908707722002, + "mean8_failure_mean": 0.05065993657996641, + "mean8_delta_failure_minus_success": 0.04899134570919421, + "mean8_auc_failure": 0.9931170954142648, + "tail_mean_5_8_success_mean": -0.002553942017630222, + "tail_mean_5_8_failure_mean": 0.030341437701096412, + "tail_mean_5_8_delta_failure_minus_success": 0.032895379718726636, + "tail_mean_5_8_auc_failure": 0.9938484040264992, + "positive_sum_success_mean": 0.030020373667639433, + "positive_sum_failure_mean": 0.40529236062114143, + "positive_sum_delta_failure_minus_success": 0.375271986953502, + "positive_sum_auc_failure": 0.9931170954142648, + "positive_count_success_mean": 3.182741116751269, + "positive_count_failure_mean": 7.991525423728813, + "positive_count_delta_failure_minus_success": 4.808784306977544, + "positive_count_auc_failure": 0.9728770541168373, + "spread_success_mean": 0.01786690075157378, + "spread_failure_mean": 0.08103766449001003, + "spread_delta_failure_minus_success": 0.06317076373843625, + "spread_auc_failure": 0.9816742665404801, + "gap12_success_mean": 0.006782306742966769, + "gap12_failure_mean": 0.028344342895483567, + "gap12_delta_failure_minus_success": 0.021562036152516798, + "gap12_auc_failure": 0.9238363589434742, + "lambda_max_corr_token_acc": -0.8797240829367476, + "mean8_corr_token_acc": -0.9146835924338959, + "tail_mean_5_8_corr_token_acc": -0.9119410149316228, + "positive_sum_corr_token_acc": -0.9168566073790785, + "positive_count_corr_token_acc": -0.6899483781164985 + } +]
\ No newline at end of file |