thegovind's picture
Publish the weight-space programming labs
bb102c0 verified
Raw History Blame Contribute Delete
41.1 kB
{ "problems": 256, "samples": 512, "acc": 0.4296875, "acc_se": 0.02603094141885692, "cut_share": 0.525390625, "mean_tokens": 22698.728515625, "closed_share": 0.53515625, "p_finish": 0.474609375, "p_correct_given_finished": 0.8930041152263375, "p_correct_given_cut": 0.011152416356877323, "no_answer_finished_share": 0.0078125, "loop_end_share_of_cut": 0.8141263940520446, "tail_dup8_cut_mean": 0.8259728624535316, "median_tokens_correct": 10278.5, "by_source": { "gsm8k": { "problems": 128, "samples": 256, "acc": 0.5546875, "acc_se": 0.03646638807723636, "cut_share": 0.375, "mean_tokens": 18912.36328125 }, "math": { "problems": 128, "samples": 256, "acc": 0.3046875, "acc_se": 0.03384154569911248, "cut_share": 0.67578125, "mean_tokens": 26485.09375 } }, "run": { "name": "dev/capA-a6", "model": "qwen35-capA-a6", "data": "dev256", "n": 2, "shards": 1, "est": 8 }, "sampling": { "temperature": 0.6, "top_p": 0.95, "top_k": 20, "presence_penalty": 1.5, "repetition_penalty": 1.0, "max_tokens": 32768, "thinking": true }, "edit_note": { "name": "capA-a6", "spec": { "about": "Lab 8 recipe A: u and 64 neurons from the base model's own loops (windows just before the loop), alpha 6", "layers": { "15": { "alpha": 6, "dir": [ -0.19642147, 0.03620983, 0.15367946, 0.06497408, 0.03057081, -0.15543503, 0.01351306, -0.05578131, -0.01658088, 0.01544302, 0.05921284, 0.06464835, 0.02034978, 0.16278124, 0.02162938, 0.1445464, 0.07718109, 0.13444427, -0.10969324, -0.06296548, 0.07529427, 0.09108829, 0.0843537, 0.08002103, -0.02452753, 0.02076042, 0.00602258, -0.00699884, 0.0446047, 0.07109267, 0.00730817, -0.14850335, 0.02196117, -0.06272466, -0.08063199, -0.01963586, -0.06232904, 0.18691534, -0.02989487, -0.02307681, -0.09384636, 0.01869328, -0.08615778, -0.01516586, 0.02899578, 0.00498247, 0.0662872, -0.04447446, 0.03304262, 0.06277169, 0.00584157, 0.01387291, -0.04707102, 0.01839207, 0.03542451, 0.08492478, 0.03609524, 0.07776414, -0.05058173, -0.02504171, -0.02387556, -0.06714687, -0.0498366, 0.02289074, 0.04711203, 0.03032538, 0.06156486, -0.00659584, -0.03459039, -0.03311691, 0.02699054, 0.02207826, -0.04704108, -0.02012696, -0.10484155, 0.01460513, -0.05287892, -0.00257121, 0.03665803, -8.442e-05, 0.0344651, 0.09257579, 0.02871555, -0.05414261, 0.01085017, 0.12585324, 0.05451579, -0.0301715, 0.00942063, 0.04333999, 0.05119767, -0.05102875, -0.04207597, 0.02894551, 0.05760689, -0.01165443, -0.0400944, 0.01880827, -0.00536408, 0.00873167, -0.06115466, -0.00820906, -0.01046184, 0.00953318, -0.04116342, -0.00371408, 0.05087092, 0.02443181, -0.02275142, 0.0121256, 0.03931541, 0.01642587, 0.00594028, -0.01517512, -0.00460996, 0.05321185, -0.00303453, -0.04462391, -0.02955657, 0.06756701, 0.01269387, 0.00439294, -0.01688814, 0.00599055, -0.05860509, -0.00640408, -0.02288292, -0.01169652, 0.01191018, 0.06990831, -0.03254844, 0.01306989, 0.0324331, 0.01259914, 0.012772, 0.01562133, 0.02279549, -0.06174612, 0.10596453, 0.01665649, 0.05464435, 0.03043335, 0.00261869, 0.04120243, -0.01908323, -0.01475778, 0.01024658, 0.00602525, 0.034647, -0.01601795, -0.0098665, 0.00715095, 0.0386551, 0.00813517, -0.02485966, -0.00370103, -0.04929582, 0.00873361, 0.00677907, -0.01257964, 0.06989538, -0.03499186, 0.01892586, 0.00830449, -0.02865393, 0.0222334, 0.02955934, 0.01670051, -0.03454896, 0.00511459, -0.06118869, -0.04395781, -0.00728062, 0.03459383, -0.01677352, -0.05499182, 0.02996176, -0.00793099, 0.01103673, -0.02698723, -0.03706948, 0.0058741, 0.01856517, -0.0184078, -0.02116514, -0.0109629, -0.06174179, -0.01890302, 0.03024366, 0.00644585, -0.04642926, 0.01424284, 0.04885947, 0.01391168, 0.00777366, 0.02742206, 0.00707238, -0.01929868, 0.03457972, -0.04974987, 0.00105937, -0.01246471, -0.02007423, 0.000219, 0.02734674, 0.05963851, 0.03654126, -0.0129875, -0.04307868, -0.00316628, 0.01014835, -0.00251832, -0.03931477, -0.00972052, -0.00568248, 0.00316628, 0.00346336, 0.03684881, -0.00336378, -0.05937261, -0.00259194, 0.00196418, 0.02412233, 0.00198911, 0.00796683, -0.03575343, 0.0077577, -0.02936143, -0.04127926, -0.03727643, -0.04707571, 0.04822274, 0.0111492, 0.0357839, -0.01515097, -0.0181218, 0.02394596, 0.02694717, -0.035783, 0.0281493, -0.01310454, 0.02221958, 0.02990598, -0.01907246, 0.04340885, 0.0115768, 0.00898189, -0.06997719, 0.01230609, -0.00925507, 0.05233327, -0.02267844, -0.05345995, 0.00955621, 0.01006574, 0.10575184, -0.0681143, -0.01524912, -0.01217928, 0.00278406, -0.00136194, 0.02119698, -0.00788542, 0.00100464, 0.02298328, 0.00925969, 0.02953216, -0.00704557, 0.00990218, -0.00553312, -0.01701421, -0.00671327, -0.00800176, -0.03354774, -0.00255209, 0.01456584, -0.02189159, -0.0115431, -0.01603708, -0.03584647, 0.04013485, -0.01710701, -0.05626024, 0.02047668, 0.01898202, -0.00604367, 0.06082913, 0.00710128, -0.02479753, 0.01761165, 0.01341764, -0.01323848, -0.02196406, -0.02244893, -0.02151935, -0.01825762, 0.00698251, -0.02361639, 0.00114175, -0.02762575, 0.16642642, -0.00773265, 0.00217698, 0.04885407, -0.01389956, 0.01700542, 0.01211646, 0.00832499, -0.00395274, 0.01165037, -0.00379319, 0.0132295, -0.00373486, 0.02491317, -0.02714441, 0.03632208, -0.08712547, -0.06559645, -0.02572042, 0.04448447, -0.00740151, 0.02038549, 0.03381699, -0.00414533, 0.02067941, -0.04021452, 0.04214275, -0.00379636, 0.01809693, -0.01443073, 0.02897453, -0.01338227, 0.00038025, 0.02115615, 0.01903441, -0.01938041, 0.02999728, 0.01710915, -0.00018278, 0.03359207, -0.15671353, 0.06294922, -0.02928597, -0.00917316, -0.02511222, 0.04767934, 0.01702067, 0.01511116, -0.01941296, 0.04285015, -0.00288132, -0.01123376, -0.01124302, -0.00706538, -0.02094487, -0.06044375, -0.01406492, -0.02563945, 0.02763499, -0.01873456, 0.00962685, 0.00600576, 0.02742081, -0.03916538, 0.00491569, 0.01069831, 0.01297124, 0.0156616, 0.01917705, 0.02480869, -0.0106574, -0.00759733, 0.00509676, 0.01549563, -0.02802384, -0.01822686, 0.00713354, -0.00076786, 0.01179229, -0.01646931, -0.02218645, -0.01974743, -0.02323301, -0.0161015, 0.00569559, -0.0318373, 0.00214495, 0.02582066, 0.0120695, -0.00323667, -0.0181305, 0.00456708, 0.00088165, -0.00181784, -0.0323725, 0.03889608, -0.00593832, -0.00065643, -0.00818366, 0.00936741, -0.01945619, 0.02748532, 0.01352952, 0.0104318, -0.04723194, -0.04889061, -0.03551376, 0.01957176, 0.03616486, -0.00134334, 0.02413643, -0.03108473, 0.00653157, -0.00807491, 0.01170466, 0.01715695, -0.02850865, -0.02597558, -0.01232146, -0.00075774, 0.00142594, 0.01942839, -0.01098431, -0.00176267, -0.08830237, 0.00896412, 0.0134145, 0.00978285, 0.0046121, -0.00490226, -0.01901206, -0.01399694, 0.01516195, -0.00334541, -0.01679288, 0.03267581, -0.01082043, 0.01129618, 0.03566024, 0.02445141, 0.01040449, 0.05409379, 0.01997331, -0.00243477, 0.01394332, 0.02577617, 0.03131029, 0.00099272, 0.0257875, 0.00535434, 0.01975382, 0.01685537, -0.00746932, 0.0180197, -0.00473603, 0.00482269, -0.02610375, -0.00263356, 0.04015455, 0.01175972, 0.00810278, 0.03288561, -0.00391045, -0.02447399, 0.0166478, 0.01949768, 0.02602378, -0.00563907, -0.00467757, 0.00259495, -0.0125137, 0.05420196, -0.00224606, -0.00816949, 0.01439246, -0.0126056, 0.00068195, 0.01309173, -0.00121671, -0.03312248, -0.01834775, 0.0016364, -0.02588954, 0.00480775, -0.01843944, -0.0041554, 0.01282341, 0.03219848, -0.01286189, 0.00484094, -0.01602968, -0.01150733, -0.03129844, 0.0092206, 0.03618083, 0.02785339, 0.02549848, 0.00144244, 0.03333158, 0.01289997, 0.00317834, 0.00372877, 1.953e-05, 0.00166176, 0.00610443, 0.01573688, 0.00230932, -0.00716901, -0.01175373, 0.00542746, -0.01469357, -0.0064027, -0.00792847, 0.0059745, -0.02925546, 0.01021165, -0.06110986, 0.01822174, 0.00379858, 0.02198827, 0.04593732, 0.01543745, -0.00720788, 0.00476402, -0.0038778, 0.01266388, -0.00248264, 0.05027047, -0.01237561, -0.00957299, -0.01446781, -0.00401733, 0.01843788, -0.02245559, 0.00411258, -0.02207132, 0.01917735, 0.01551384, 0.06460287, 0.02049114, 0.00455079, -0.00079158, -0.00720685, -0.02944683, -0.00974075, -0.00361087, -0.02188736, 0.00510789, 0.00083505, 0.01718823, -0.01092549, -0.06818206, 0.02002659, -0.00443351, 0.02195536, -0.03922576, 0.02893249, -0.00792199, -0.01345805, 0.03962484, 0.01178997, 0.03471842, 0.02131379, 0.0221365, -0.01394667, -0.0011956, -0.00585808, -0.00686869, -0.00681893, 0.00314807, 0.02003211, -0.01824619, -0.01195085, -0.00977994, -0.02926364, 0.00122858, 0.0081286, -0.0084117, 0.03174116, 0.01527151, 0.01818123, -0.01315631, 0.00535931, -0.01808719, -0.01626276, 0.01888465, -0.00875709, 0.0206139, 0.01838784, -0.03064582, 0.01166973, 0.02911546, 0.02594003, -0.00576419, -0.00914466, -0.0237985, 0.01038998, -0.01717332, 0.02492017, -0.00753594, 0.01322463, 0.01343332, 0.00121454, 0.01580115, 0.00307425, 0.0345596, -0.01798728, 0.0323272, -0.0068936, -0.00869091, 0.0045136, -0.00837262, -0.00855466, 0.0034192, 0.02522525, -0.00705909, -0.019891, 0.00191272, -0.01211489, 0.0285136, -0.00566154, 0.04103695, -0.0092919, 0.00241761, 0.01635534, -0.00316568, -0.01597946, -0.03456803, -0.03958578, 0.01394486, -0.01306217, -0.00343395, 0.01033765, 0.01521762, 0.00855863, -0.00389284, 0.02527569, -0.01169722, 0.00200877, 0.04353902, 0.02844795, 0.02484205, 0.01893727, -0.03412919, 0.01272242, 0.00202661, 0.023077, 0.02425507, -0.0236186, -0.04173858, 0.00582972, 0.00735155, -0.00016013, 0.01041111, 0.0020433, 0.01619346, 0.01290293, 0.0304944, -0.02799492, 0.00890391, 0.01989347, 0.01319829, 0.0316516, 0.01294222, 0.01560815, 6.969e-05, 0.00689163, 0.0201618, 0.01609466, -0.00964703, 0.00868139, -0.03052312, -0.00467848, -0.01110903, -0.00782298, -0.01644403, -0.02660462, -0.00318245, -0.03656901, -0.01252536, 2.685e-05, 0.00575581, 0.00867378, -0.03588792, -0.00962772, 0.00612709, 0.02778116, 0.00617909, -0.02462822, 0.01033108, 0.02212686, 0.00757206, 0.0081198, 0.02414416, -0.00085268, -0.01033239, -0.01519087, 0.01407947, 0.01629492, -0.00385944, 0.00988953, 0.01378492, 0.04593561, 0.00968383, 0.03686356, -0.01484743, 0.01556203, -0.01520757, 0.02010669, -0.0176068, 0.02796162, -0.00331875, -0.01562665, -0.02325287, -0.03160228, 0.01461223, -0.00678877, -0.0085217, 0.00715341, -0.003018, 0.01534117, 0.00859851, 0.00273759, -0.00587398, 0.0363301, 0.04355061, -0.01432952, 0.02605049, 0.02288688, 0.03925271, 0.01841203, -0.03189914, 0.01263224, 0.01408128, 0.0248368, 0.01680999, -0.01338161, 0.00324507, -0.02825224, -0.01734889, 0.00191852, -0.01146315, -0.0038348, -0.00840155, -0.0083625, -0.0499628, -0.02321059, -0.01606922, 0.01277077, -0.00190077, 0.01025087, -0.03114786, 0.00791134, 0.01744954, 0.02002841, -0.0255163, -0.02600448, -0.01578891, -0.02690942, -0.02880245, -0.00431575, 0.00600727, -0.03090141, 0.01105587, -0.0015371, 0.00296751, 0.01277654, -0.00477245, -0.02422091, 0.02559535, -0.00133811, -0.00772111, -0.04095183, 0.03351398, 0.01896126, 0.00341619, 0.01234442, -0.03707758, 0.00587088, -0.00318547, 0.0377168, -0.01426577, -0.02044055, -0.0175332, 0.01608709, 0.03591258, 0.00580495, 0.01636545, -0.01722871, 0.00606186, -0.00400982, 0.0109228, 0.02489203, 0.01245983, -0.00455665, -0.01384578, -0.00659455, 0.00777224, -0.00334753, -0.0177411, -0.00531191, 0.00629907, 0.02441626, -0.01576012, 0.03351801, 0.02288313, -0.05304268, -0.0071294, 0.03232586, 0.00891935, -0.01977362, -0.0027413, -0.01511885, 0.02056648, 0.00657744, 0.01876156, 0.01019189, -0.01308637, 0.02896509, 0.00256687, -0.00765172, 0.03474916, 0.00256524, 0.00674338, 0.04313926, -0.00284791, 0.0150191, 0.01572495, 0.02367985, -0.01331006, 0.01604667, -0.02399078, 0.03464028, -0.00468134, 0.00330221, 0.00213209, 0.00835461, -0.02771784, 0.00901233, 0.0105306, -0.00924321, 0.02637529, -0.02676393, 0.00356408, -0.04889319, -0.02941449, 0.00112647, -0.02983064, 0.03998261, -0.02925883, -0.00396614, 0.01319757, 0.01322922, -0.00243386, -0.00769942, 0.01422793, -0.02085671, -0.01465598, -0.00793447, -0.0158751, -0.04526321, 0.03444614, 0.01128489, -0.02307381, 0.02501738, 0.00990441, 0.02356402, -0.00317668, -0.0126682, 0.02966011, 0.00138062, -0.02201164, -0.02099483, -0.00252512, 0.0012899, 0.01235682, -0.01162883, -0.02899506, -0.05046245, 0.03096318, -0.0059149, -0.02507442, -0.00721694, -0.03142115, -0.01031928, 0.00791632, 0.01309902, 0.02127181, -0.01971884, -0.02214627, -0.02675827, 0.00386679, 0.00320048, -0.00816859, -0.01619648, -0.01310949, 0.00787304, 0.00961308, -0.0018458, 0.01286562, 0.03139392, -0.00560179, -0.00183139, -0.00443225, -0.00790803, -0.01636733, -0.00759358, 0.00203761, 0.01884115, -0.01272084, 0.00250993, 0.03664702, -0.04066893, 0.00552276, 0.0137794, -0.02775645, -0.00558242, -0.01088545, -0.01883953, -0.00373733, 0.00910788, 0.01156108, -0.01827489, -0.02444516, -0.00397615, -0.02902949, -0.01167935, -0.00136049, 0.00129787, 0.00911742, -0.00392505, -0.00567445, 0.00135177, 0.01242116, 0.00707127, 0.01174442, 0.00069741, -0.02878679, -0.02556429, 0.00227394, 0.00793164, -0.00369597, 0.03255857, 0.0202454, -0.01635662, -0.02483233, 0.00268604, 0.01370827, -0.00288547, -0.0091429, -0.04589674, -0.00381021, -0.02784827, -0.02464681, 0.03012316, -0.02526064, -0.0019681, 0.00598796, -0.01061581, -0.01781175, -0.01317692, -0.01122522, -0.00949888, -0.04640295, 0.00405447, 0.00692723, -0.01286737, 0.0194275, 0.03252542, -0.03120079, -0.00596153, -0.02418105, 0.01823442, -0.00854398, 0.03457795, -0.00829153, 0.01389347, 0.01776138, -0.0100889, -0.00340556, -0.00454865, -0.00269702, -0.01599278, 0.03753144, 0.02651554, -0.01864239, -0.00012952, 0.03227106, 0.00020344, 0.03920548, 0.00676807, -0.01476055, -0.02480705, -0.0294454, -0.02185388, 0.01376991, 0.0019891, -0.00082901, -0.00209755, -0.00537243, 0.00898147, -0.03283366, -0.02692142, -0.03934251, -0.03625376, -0.00753794, 0.00549959, 0.00160269, -0.01382472, -0.03401972, -0.01955991, -0.01109684, 0.02179758, -0.03753874, -0.06081887, -0.00784457, -0.01305669, -0.04447476, -0.00143666, -0.01756143, -0.02385547, 0.01329127, -0.01187246, -0.01682425 ], "cols": [ 0, 11, 25, 35, 37, 38, 39, 42, 46, 56, 61, 63, 65, 68, 81, 105, 110, 129, 135, 169, 215, 259, 264, 266, 294, 371, 414, 426, 452, 496, 504, 505, 509, 516, 580, 582, 601, 638, 682, 733, 738, 750, 755, 882, 891, 907, 926, 956, 1047, 1060, 1087, 1130, 1155, 1175, 1184, 1702, 1727, 1757, 1787, 2248, 2288, 2431, 2636, 3003 ] }, "19": { "alpha": 6, "dir": [ -0.41399729, 0.04963015, 0.12075837, 0.02570651, 0.07051032, -0.14113681, 0.03799108, 0.01167555, -0.0064611, -0.01368775, 0.06116361, 0.03473959, 0.05443978, 0.1034804, 0.03853016, 0.10406735, 0.07458493, 0.19364277, -0.10859208, -0.07088345, 0.09971021, 0.04388057, 0.06559257, 0.07909922, -0.0305314, 0.00847475, 0.0029426, 0.01185544, 0.01822992, 0.04678907, 0.00322332, -0.12947491, 0.05207255, -0.04346941, -0.08363152, 0.01037417, -0.024745, 0.13930042, -0.03795761, -0.02326387, -0.08854304, 0.00513338, -0.04700642, -0.0410671, -0.00597825, -0.01444194, 0.03166021, -0.06466202, 0.03836314, 0.10043173, 0.00429663, 0.01563108, -0.04114208, 0.01129955, 0.03724522, 0.07670557, 0.04311912, 0.07472885, -0.02922567, -0.03833525, 0.00456427, -0.07277104, -0.04912667, 0.03210241, 0.06677274, 0.04299912, 0.01544243, -0.00771099, -0.04890209, -0.05095252, 0.03187749, 0.01289586, -0.04323121, 0.01584703, -0.09161146, 0.01245053, -0.06768124, 0.00860595, -0.00343637, 0.01398703, 0.03606768, 0.0810465, 0.02315859, -0.04040698, 0.03198423, 0.11273056, 0.07610082, -0.0318647, 0.01234657, 0.02210993, 0.04747062, -0.06642445, -0.06831807, -0.00169164, 0.04582261, -0.01536702, -0.02826523, -0.01215465, -0.011371, 0.01469242, -0.05166398, -0.02717075, -0.00141099, 0.02742627, -0.02786403, -0.00629218, 0.03265667, 0.01514308, -0.0149828, -0.00829845, 0.04983302, 0.04255297, 0.00411385, -0.03872541, 0.0036479, 0.04089405, -0.02173937, -0.07732391, -0.02259742, 0.05235458, 0.0173907, 0.02537671, -0.02115237, 0.01312889, -0.03039251, 0.00446135, -0.03203883, 0.02627006, -0.00464935, 0.06962904, -0.03380452, -0.01486555, 0.03601762, 0.04170777, 0.02079187, 0.01055915, 0.013812, -0.07132905, 0.05949075, 0.00309522, 0.0485229, 0.02636794, 0.02857509, 0.04520793, -0.00615302, -0.00280603, 0.00502628, -0.01417321, 0.02896457, -0.01341004, -0.01564776, 0.01597291, 0.02518996, 0.02494553, -0.00542398, -0.00570583, -0.01915297, 0.00438684, 0.00933441, -0.01003382, 0.07714284, -0.00258696, 0.01232391, 0.01856954, -0.02422703, 0.00045015, -0.00747694, -0.04692789, -0.03999501, -0.00415803, -0.02718486, -0.02438036, -0.00504409, 0.01552685, -0.01566504, -0.03189168, 0.01812186, -0.02202662, 0.02614452, -0.00671965, -0.03856579, 0.02344356, -0.00223916, -0.00331282, -0.02047138, -0.01889345, -0.04949783, -0.02935014, 0.0107663, 0.00219746, -0.06389891, 0.01158193, 0.04443077, 0.01606618, -0.01341348, 0.02656645, 0.01186088, -0.0103251, 0.00045351, -0.05710696, 0.01312309, -0.01828778, -0.02093595, 0.01126861, 0.00185225, 0.06209056, 0.01881377, -0.0147169, -0.04910628, -0.00019105, -0.00031153, 0.01395955, -0.04751529, -0.02592737, -0.02079906, -0.01138995, 0.02150493, 0.0168967, 0.01505595, -0.07899936, -0.01489948, -0.01328334, 0.03457796, -0.0035607, 0.02112566, -0.05852021, -0.00032909, -0.01642394, -0.02042259, -0.03190672, -0.07005456, 0.05231471, 0.01389211, 0.03463357, 0.00445699, -0.01345005, 0.01217057, 0.02445503, -0.04094464, 0.03871498, -0.00529151, 0.04551385, 0.02378383, -0.0063007, 0.0368812, 0.00015905, 0.02571216, -0.04776682, -0.00851591, 0.00099652, 0.04026485, -0.01877851, -0.0219236, -0.00297592, 0.01029671, 0.10225255, -0.07007081, -0.02340342, 0.00304178, 0.01930523, -0.02637254, 0.00015534, 0.01005215, 0.01588354, 0.01971863, 0.00551752, 0.01135873, 0.00747602, 0.01103365, 0.01211043, 0.00037698, -0.02620154, 0.00256552, -0.02569012, -0.02783686, 0.00628542, -0.00104915, -0.0084195, -0.03021691, -0.06425054, 0.02496984, -0.0123, -0.03871801, 0.00344185, 0.0094332, -0.01717056, 0.04348284, -0.01649027, -0.01774609, -0.00925723, 0.01012276, -0.02162685, -0.00469051, -0.00543928, 0.00891075, -0.02664486, 0.00453793, -0.00719609, -0.0229699, -0.02674694, 0.08312266, -0.02648537, -0.01864358, 0.05058388, -0.02575401, 0.02509584, -0.00238742, -0.01080631, -0.02850006, 0.03767575, -0.01170178, -0.04227291, -0.00819114, 0.0155615, -0.01753663, 0.01993192, -0.08148444, -0.0680353, 0.00388821, 0.03550627, 0.00862947, 0.00714255, 0.010537, 0.00096149, -0.00442716, -0.01539658, 0.03665021, 0.00852256, 0.01478875, -0.03822764, 0.04834064, -0.00460326, 0.01477617, 0.04410122, 0.01627444, -0.00093999, -0.00024491, -0.00553846, -0.00497731, 0.04350387, -0.14360385, 0.02938055, -0.0352683, -0.00509822, -0.02180302, 0.00219526, -0.00115064, -0.00144354, -0.02192711, 0.04223738, -0.02610195, -0.01616591, -0.01219281, -0.0051492, -0.02957796, -0.02786715, -0.02184732, -0.01340924, 0.01838329, -0.02598088, 0.02937649, 0.01490723, 0.01670144, -0.03311621, 0.02081646, 0.01394797, -0.00322997, 0.00261211, 0.03853096, 0.02314228, 0.00417184, -0.02041819, 0.00740087, 0.00630271, -0.02640118, -0.01714695, 0.0081078, 0.02775934, -0.00849896, -0.02215351, -0.0097992, 0.0137838, -0.00961078, 0.00229612, -0.00337052, -0.01665382, 0.00472534, 0.01867853, 0.02347246, 0.00167132, -0.00106219, -0.00794656, -0.00257331, -0.00023658, 0.00359527, 0.04330531, -0.01585915, 0.00389967, -0.00603891, 0.01601479, -0.00080251, -0.00465458, 0.00355762, -0.00012541, -0.01363729, -0.03129135, -0.0451249, 0.03790808, 0.01233148, -0.00681186, 0.02245213, -0.02457226, 0.0053979, -0.00176179, -0.01983261, -0.00289942, -0.054424, -0.01660743, -0.01866144, -0.0056431, -0.01305195, -0.01278731, -0.01718922, -0.01980148, -0.12208816, 0.01950914, 0.00548164, -0.00599155, 0.01228508, -0.01484066, -0.02760676, 0.02717005, -0.00303716, -0.00324011, 0.00126969, 0.02935875, -0.01822443, 0.01723574, 0.01369331, 0.01482911, -0.00349897, -0.0075394, -0.00909447, -0.01029597, 0.02345074, 0.03705572, 0.00569269, 0.02199166, 0.00893563, -0.00376267, 0.00459211, 0.03037536, -0.01397452, -0.00245161, -0.00043594, -0.00770201, -0.02184014, 0.00735113, 0.03283777, -0.0031054, 0.00327932, 0.02279597, 0.00313427, -0.02228454, 0.0249339, 0.00893045, 0.0193599, -0.00423225, 0.01214966, 0.00211539, -0.01705077, 0.05321393, 0.00832467, 0.00691531, 0.00440044, -0.0036028, 0.00648574, -0.00184864, 0.01262479, -0.04579851, -0.01826578, -0.00640297, -0.03256388, 0.00699382, -0.01327709, -0.01331763, -0.00193405, -0.00183068, -0.00744029, 0.00887535, -0.03525895, -0.00613091, 0.00389063, 0.0061255, -0.00145237, 0.02164902, 0.00551578, -0.02099774, 0.02638274, -0.01244471, 0.01595411, 0.00367786, 0.00150957, 0.01250048, 0.01080034, -0.00269924, -0.01266212, -0.02325669, -0.00212248, 0.05340743, -0.00488497, -0.00560996, -0.01019677, -0.00038237, -0.01917257, 0.00416132, -0.06233919, 0.02279506, 0.01383749, 0.0223188, 0.03706111, 0.00759379, -0.01933291, -0.00911448, -0.00832836, 0.00218727, 0.00528866, 0.0012676, -0.00689148, -0.01066677, -0.03458137, -0.00148494, 0.00298672, -0.02046814, 0.01484077, -0.00187794, 0.00895328, 0.01091576, 0.02540395, 0.01473134, -0.01313322, 0.00882845, 0.00047043, -0.03233552, -0.00381481, -0.00340058, 0.00045749, 0.01598782, 0.01730791, 0.03945119, -0.00992342, -0.04769488, 0.01061766, -0.00912597, -0.00186258, -0.02074894, -0.00693543, 0.00157427, -0.02019794, 0.03320531, 0.00160058, 0.03022134, 0.01365437, 0.01773841, -0.00126859, -0.00960154, 0.00160947, 0.02864504, 0.00388886, 0.01610713, 0.02257901, -0.01016495, 0.0029654, -0.00020181, 0.00606368, 0.01182888, 0.01258298, 0.06290828, 0.02865976, 0.02483223, 0.00040623, 0.00811181, -0.01250675, -0.01186497, -0.01285955, 0.01597996, -0.04972612, 0.00089825, 0.00958841, -0.02032982, -0.00527198, 0.02549261, 0.04237567, 0.0055739, -0.01983182, -0.02598889, 0.01521344, 0.01940417, 0.00654279, -0.01277678, 0.01001596, 0.00197745, 0.00052592, 0.01791021, 0.00403065, 0.01413665, -0.00654267, 0.01739347, -0.00468635, 0.01130318, 0.01190673, 0.00176921, 0.00215732, -0.01801398, 0.00423987, -0.02487228, -0.03574106, -0.03761583, -0.01580498, 0.0112633, -0.00802143, 0.04035542, -0.02123511, 0.01391431, 0.02576919, -0.01363103, -0.01700656, -0.03270056, -0.01692577, 0.0040207, -0.00770615, -0.02023523, -0.01287303, -0.00399295, -0.0097395, -0.00157304, 0.01470541, -0.02900858, -0.01364948, 0.01622227, 0.04555156, 0.02300258, 0.00070742, -0.03894363, 0.01395334, -0.011304, 0.01820423, 0.02882553, -0.0265112, -0.02328012, -0.0057346, -0.0111037, -0.00239438, 0.00067644, -0.00284005, 0.02123305, 0.019265, 0.01552788, -0.01986999, 0.03097746, 0.03139879, 0.01663943, 0.02280164, 0.01902663, 0.00619938, -0.01090529, -0.02094108, 0.01410988, 0.0052609, -0.01526535, 0.03699815, -0.03132536, 0.01252833, -0.02781519, -0.01810875, -0.009691, -0.02077244, 0.00926613, -0.04685893, -0.00726184, -0.00891186, -0.00769646, 0.01439862, -0.02565044, 0.00493228, 0.00974353, -0.00995644, 0.02049081, -0.04404585, 0.00751193, 0.00638864, 0.00514695, 0.05242157, 0.00289454, -0.00064507, 0.03833183, -0.04490974, -0.00162655, -0.01159293, 0.01254596, -0.01223043, 0.0235262, 0.02629913, 0.01253196, -0.02080517, -0.00689689, -0.0068256, 0.01760788, 0.01303217, -0.00408248, 0.01806383, 0.00275279, -0.02245613, -0.03216419, -0.02313394, 0.01341674, -0.02875509, 0.00821746, -0.00258162, -0.0166717, -0.00257669, -0.01655585, 0.01896282, -0.01112964, 0.00746154, 0.01104538, -0.02793135, -0.00074981, 0.00759404, 0.04883965, 0.02882431, -0.01466329, -0.0079204, -0.01327058, 0.01041078, 0.01971299, -0.00478428, -0.02979348, -0.0033707, -0.01569781, 0.00752487, 0.00428126, 0.02388609, 0.0027208, -0.00341252, -0.04374065, -0.01116345, -0.00532541, 0.02159592, 0.00406952, 0.00415196, -0.0033014, 0.00208155, -0.02914683, 0.02400334, -0.00277937, -0.01673185, -0.01697779, -0.02679257, -0.0103722, -0.03774866, 0.02358133, -0.02143352, -0.00031711, -0.00352962, -0.00598108, 0.01242257, 0.0164373, 0.00470733, 0.02708453, -0.02024415, 0.01169108, -0.03683765, 0.02083484, 0.00937352, -0.02739683, -0.0067206, -0.01986139, 0.00547282, -0.01628444, 0.00153831, -0.00249261, -0.02345225, -0.01357254, 0.01328131, 0.00521324, 0.00352823, -0.01862443, -0.00389278, 0.00673196, 0.00429158, 0.00606037, 0.03048845, -0.01037415, -0.01091966, 0.0005297, -0.0454992, -0.03626728, 0.00287546, 0.00962111, -0.01698189, -0.00070894, 0.01815648, -0.01495158, 0.01427327, 0.02984908, -0.01813151, -0.00496254, 0.0179225, -0.00915224, -0.03176316, 0.00395478, 0.00483918, 0.00103693, -0.03188303, 0.00852776, -0.00672803, -0.01914714, 0.00857762, -0.0096268, 0.00650117, -0.01087384, -0.011639, 0.01615791, 0.02781934, -0.00755544, -0.00014848, -0.00793268, 0.00750446, -0.0103245, 0.0168096, -0.02983714, 0.03803321, -0.01768607, 0.00270445, 0.01078355, 0.01566177, -0.0133848, 0.01774806, -0.00812243, 0.02237075, 0.01115108, 0.00381729, 0.01070505, -0.03039744, -0.04433396, 0.02080437, -0.01398751, 0.02386202, 9.26e-05, -0.01348085, 0.00854043, 0.03121264, -0.00824868, -0.0059332, 0.01327891, -0.01346949, -0.00999715, -0.00842401, -0.02804655, -0.03686436, 0.02232404, -0.0279701, -0.02540337, -0.00054304, -0.00756712, 0.02793047, -0.01626339, -0.01397534, 0.02358809, 0.04023468, -0.03251312, -0.01672773, 0.00377267, 0.01258042, -0.00385397, -0.02168425, -0.01490837, -0.01752106, -0.0036807, -0.01502097, -0.03276443, -0.01766985, -0.02547744, -0.00919792, -0.00224599, 0.0251513, -0.01591504, 0.00427116, -0.01306467, 0.01153746, -0.02659429, -0.00869732, 0.00801325, -0.00534268, 0.00309692, 0.0168271, 0.00267658, -0.00784185, -0.01638011, -0.00787242, -0.01431224, -0.00839908, 0.00343782, 0.0076508, 0.00512584, -0.00750912, 0.01765133, 0.00432924, -0.00314923, -0.02161724, 0.02385022, -0.02513087, -0.01094419, 0.01210399, -0.00285633, 0.00383794, -0.03123457, -0.01198237, -0.01510271, -0.01148189, 0.00541213, -0.01798316, 0.00496864, -0.0015806, -0.01432069, -0.02940245, -0.01776116, -0.00566956, 0.0015968, -0.01688569, -0.00069971, -0.00806095, 0.01892859, 0.03505353, 0.00642723, -0.00281842, -0.03205081, -0.01303378, -0.00905993, 0.02895496, 0.00354829, 0.01735854, 0.00953999, -0.01426204, 0.01120116, -0.00881752, 0.02241647, -0.01657692, -0.00571749, -0.04898428, 0.00551105, -0.02055241, -0.00990897, 0.00936325, -0.0096167, 0.01104994, 0.01584025, 0.01587972, 0.00194767, -0.00631707, -0.04304481, 0.0018921, 0.01500908, -0.01644222, 0.00668909, -0.03100232, -0.00048992, 0.0197435, -0.01865795, -0.00662279, -0.00430736, 0.01438951, -0.02331211, 0.00492494, -0.01072574, -0.00245177, -0.00950786, -0.00477532, -0.0100874, -0.01600529, 0.00570134, -0.01740529, 0.03467067, 0.01419344, 0.00161084, -0.00347943, 0.02319815, -0.00585438, 0.00335634, 0.03062036, -0.01164215, 0.00494848, 0.01392082, -0.04225956, -0.00249652, -0.00907702, 0.03732198, -0.00339302, -0.01306331, 0.02235401, -0.01099497, -0.01200077, -0.05103194, -0.00398947, -0.00825615, 0.02447357, -0.00067779, -0.00834806, -0.0175427, -0.01224584, 0.00016019, 0.01329831, -0.0194385, -0.01699858, 0.00808758, -0.0110953, -0.00470004, 0.0187731, 0.00288963, -0.04354201, -0.0111457, 0.0091344, 0.00545016 ], "cols": [ 1, 11, 13, 15, 17, 24, 34, 36, 41, 53, 58, 61, 63, 64, 66, 69, 76, 79, 92, 95, 97, 98, 99, 114, 119, 123, 135, 145, 146, 148, 152, 161, 171, 173, 177, 192, 246, 253, 259, 265, 268, 286, 287, 302, 316, 324, 343, 351, 355, 450, 500, 579, 602, 614, 795, 933, 1003, 1110, 2055, 2072, 2175, 2595, 3242, 3343 ] } } }, "built": { "15": { "alpha": 6.0, "n_cols": 64, "dir": "custom", "relative_change": 0.10540313273668289 }, "19": { "alpha": 6.0, "n_cols": 64, "dir": "custom", "relative_change": 0.0964626893401146 } }, "edited_sha256": { "model.language_model.layers.15.mlp.down_proj.weight": "468df0e1776c53c712a8d1e769f839a76ae044bf27a2fba47e00b8c0da7eeb2c", "model.language_model.layers.19.mlp.down_proj.weight": "f27f357d1fd4685e8ba7fb390ee17a1eb593bfa3d700795299cddf923e36cc6a" } }, "tokens_total": 11621749, "gpu_seconds": 1911, "tok_per_s_per_gpu": 6277 }
{
"problems": 256,
"samples": 512,
"acc": 0.4296875,
"acc_se": 0.02603094141885692,
"cut_share": 0.525390625,
"mean_tokens": 22698.728515625,
"closed_share": 0.53515625,
"p_finish": 0.474609375,
"p_correct_given_finished": 0.8930041152263375,
"p_correct_given_cut": 0.011152416356877323,
"no_answer_finished_share": 0.0078125,
"loop_end_share_of_cut": 0.8141263940520446,
"tail_dup8_cut_mean": 0.8259728624535316,
"median_tokens_correct": 10278.5,
"by_source": {
"gsm8k": {
"problems": 128,
"samples": 256,
"acc": 0.5546875,
"acc_se": 0.03646638807723636,
"cut_share": 0.375,
"mean_tokens": 18912.36328125
},
"math": {
"problems": 128,
"samples": 256,
"acc": 0.3046875,
"acc_se": 0.03384154569911248,
"cut_share": 0.67578125,
"mean_tokens": 26485.09375
}
},
"run": {
"name": "dev/capA-a6",
"model": "qwen35-capA-a6",
"data": "dev256",
"n": 2,
"shards": 1,
"est": 8
},
"sampling": {
"temperature": 0.6,
"top_p": 0.95,
"top_k": 20,
"presence_penalty": 1.5,
"repetition_penalty": 1.0,
"max_tokens": 32768,
"thinking": true
},
"edit_note": {
"name": "capA-a6",
"spec": {
"about": "Lab 8 recipe A: u and 64 neurons from the base model's own loops (windows just before the loop), alpha 6",
"layers": {
"15": {
"alpha": 6,
"dir": [
-0.19642147,
0.03620983,
0.15367946,
0.06497408,
0.03057081,
-0.15543503,
0.01351306,
-0.05578131,
-0.01658088,
0.01544302,
0.05921284,
0.06464835,
0.02034978,
0.16278124,
0.02162938,
0.1445464,
0.07718109,
0.13444427,
-0.10969324,
-0.06296548,
0.07529427,
0.09108829,
0.0843537,
0.08002103,
-0.02452753,
0.02076042,
0.00602258,
-0.00699884,
0.0446047,
0.07109267,
0.00730817,
-0.14850335,
0.02196117,
-0.06272466,
-0.08063199,
-0.01963586,
-0.06232904,
0.18691534,
-0.02989487,
-0.02307681,
-0.09384636,
0.01869328,