thegovind's picture
Publish the weight-space programming labs
bb102c0 verified
Raw History Blame Contribute Delete
41.1 kB
{ "problems": 256, "samples": 512, "acc": 0.537109375, "acc_se": 0.027922737013093644, "cut_share": 0.396484375, "mean_tokens": 20725.197265625, "closed_share": 0.70703125, "p_finish": 0.603515625, "p_correct_given_finished": 0.8867313915857605, "p_correct_given_cut": 0.0049261083743842365, "no_answer_finished_share": 0.00390625, "loop_end_share_of_cut": 0.7684729064039408, "tail_dup8_cut_mean": 0.7750477832512315, "median_tokens_correct": 11535, "by_source": { "gsm8k": { "problems": 128, "samples": 256, "acc": 0.69140625, "acc_se": 0.03641858274265708, "cut_share": 0.203125, "mean_tokens": 16059.3515625 }, "math": { "problems": 128, "samples": 256, "acc": 0.3828125, "acc_se": 0.037812174626048516, "cut_share": 0.58984375, "mean_tokens": 25391.04296875 } }, "run": { "name": "dev/capB-max-samples-16", "model": "qwen35-capB-max-samples-16", "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": "capB-max-samples-16", "spec": { "about": "Lab 8 recipe B: u and 64 neurons from the base model's own loops (windows well before the loop), alpha 6", "layers": { "15": { "alpha": 6, "dir": [ -0.21637826, 0.0208582, 0.14215979, 0.03148801, 0.05498521, -0.20134567, 0.02097962, -0.04125047, -0.00052325, 0.00733103, 0.07352018, 0.09851355, 0.05241353, 0.15527157, 0.02442253, 0.18563084, 0.05889465, 0.20897448, -0.13037334, -0.07115818, 0.07148564, 0.1075747, 0.09280334, 0.09322366, -0.03304736, 0.0135476, -0.0039322, -0.02402865, 0.02404705, 0.08518188, 0.03829363, -0.15843356, 0.02383504, -0.04695624, -0.09074076, -0.01772078, -0.07573749, 0.16334917, -0.03774936, -0.00817425, -0.09472003, 0.00438293, -0.02289957, -0.01206419, 0.02433615, -0.00402353, 0.11656889, -0.04101976, 0.03317771, 0.04635753, 0.0136615, -0.01092584, -0.05370047, 0.02088208, 0.02476078, 0.08169514, 0.04900517, 0.08017961, -0.06309437, -0.03409009, -0.01955112, -0.07093485, -0.07129309, 0.01417559, 0.03233548, 0.01469945, 0.03210051, -0.01336099, -0.0527927, -0.06246308, 0.0312902, 0.00981067, -0.05429796, -0.03030176, -0.08878677, 0.027743, -0.04120322, 0.01580702, 0.0270111, 0.00517031, 0.05249692, 0.11200938, 0.02801427, -0.04316639, 0.02306494, 0.10611688, 0.06860207, -0.03525721, 0.01403375, 0.05267483, 0.03644583, -0.04387341, -0.03257689, 0.02230091, 0.05894295, -0.02715016, -0.01069902, 0.02121824, 0.01928968, 0.01017583, -0.06540359, -0.0304288, -0.02855779, 0.00934084, -0.03657155, 0.01117012, 0.05167731, 0.01095467, -0.03311194, 0.03558973, 0.03556742, 0.01650866, 0.00263433, -0.0188665, -0.01134456, 0.05503199, -0.00063654, -0.02584132, -0.01402893, 0.06095159, -0.00308138, 0.00242266, -0.00146307, 0.00524068, -0.06716381, -0.00451794, -0.02463672, -0.01660852, 0.00293268, 0.04098685, -0.04678087, 0.01451141, 0.01131659, 0.00731429, 0.03095794, -0.00482354, 0.00319243, -0.06602572, 0.09552348, 0.00376164, 0.00205593, 0.0248206, 0.0606396, 0.02391795, -0.03181076, -0.01020631, 0.01061545, 0.01111651, 0.01812949, -0.01816203, -0.0087539, -0.0095157, 0.02544892, 0.01982855, -0.02565813, -0.00071948, -0.04701146, 0.00710719, 0.00102724, -0.00315379, 0.06567564, -0.01805431, 0.00854577, -0.00391886, -0.03749699, 0.01716518, 0.03549364, 0.02678264, -0.0193614, 0.00598351, -0.05132948, -0.0409944, -0.01740579, 0.05001541, -0.00146954, -0.05465335, 0.08600657, 0.01201068, 0.01541469, -0.02613408, -0.02651178, 0.00280248, 0.01720738, -0.01958507, -0.01702372, -0.00038944, -0.06105636, 0.00287163, 0.00761413, 0.02329392, -0.04148633, 0.01394837, 0.02558159, 0.01322039, 0.00660018, 0.04995323, -0.00177992, -0.00260812, 0.01295732, -0.03431334, 0.00483544, -0.0189087, -0.01810008, -0.00301407, 0.01489605, 0.04309268, 0.00916058, -0.00763202, -0.03752465, -0.00257878, -0.00185424, -0.00939396, -0.02871411, -0.01694912, 0.00617886, 0.01543686, -0.00318275, -0.00119812, -0.00914472, -0.06660454, -0.02198981, 0.01329394, 0.02651368, -0.00987455, 0.00744685, -0.06965589, 0.01270202, -0.01686738, -0.02549471, -0.00066508, -0.02412686, 0.05300555, 0.0009579, 0.0321163, -0.00716791, -0.01013476, 0.01229769, 0.02446852, -0.01693287, 0.02369243, -0.01564124, 0.00635829, 0.02192136, -0.00652901, 0.05100792, -0.00297848, 0.00627843, -0.04704197, 0.00929152, -0.01633348, 0.0380201, -0.01274131, -0.04084484, 0.00495194, 0.01178072, 0.08586691, -0.03260215, -0.02283864, 0.02895299, -0.00908572, -0.01062437, 0.02229035, -0.01363111, -0.00026542, 0.01407236, 0.00810807, 0.03858791, -0.00178614, 0.00746921, -0.011037, -0.00596729, -0.00478139, 0.00604533, -0.03742398, 0.00761321, 0.02869093, -0.02771991, -0.01507655, -0.01780738, -0.04108199, 0.03030084, -0.01594888, -0.06110862, 0.02679177, 0.01771371, 0.01325282, 0.04883391, 0.00990529, -0.01405532, 0.02551616, 0.01493111, 0.00115052, -0.01985505, -0.00212461, -0.00808253, -0.0195281, 0.01100378, -0.01349443, 0.00853589, -0.02471553, 0.1474885, -0.0028093, 0.00431209, 0.04638571, -0.00894456, 0.01005976, 0.00514061, 0.00342722, -0.01534806, 0.02596342, 0.00351884, 0.02283818, 0.00081903, 0.02703408, -0.02058055, 0.00455945, -0.05770486, -0.04616447, -0.03020212, 0.04096419, -0.00021511, 0.01669699, 0.01905221, 0.00096826, 0.00021619, -0.01689276, 0.02666213, 0.0030318, 0.01700302, -0.0252526, 0.01408874, 0.0034899, 0.00164022, 0.0194878, 0.02702537, -0.02587424, 0.02682163, 0.01517034, 0.00895178, 0.03934432, -0.14160426, 0.05098933, -0.02668699, -0.02009902, -0.0110586, 0.04004399, 0.02103234, -0.00054014, -0.00304157, 0.04513934, 0.00115051, -0.00763194, -0.01521879, -0.00248921, -0.01612355, -0.0726915, -0.00779943, -0.0390397, 0.02930136, -0.01531973, 0.01127179, 0.00116985, 0.04106858, -0.0327399, 0.00517233, 0.02065066, 0.0139229, 0.0153218, 0.0300049, 0.01230022, -0.01084541, 0.00537872, -0.00136067, 0.02197347, -0.04333798, -0.02579584, -0.00508774, 0.00833374, 2.418e-05, -0.01854045, -0.00749867, -0.00814498, -0.02718223, -0.01018141, 0.01809509, -0.02845303, -0.00640503, 0.028403, 9.423e-05, 0.01023048, -0.0148559, 0.01151288, -0.00960036, 0.00259169, -0.02387808, 0.02720485, 0.00708053, -0.00603507, -0.00515519, 0.00921025, -0.01155539, 0.03224582, 0.01222016, -0.0018756, -0.03686034, -0.03205812, -0.02737863, 0.00654368, 0.02653143, -0.01576704, 0.01697567, -0.0326158, 0.00554613, -0.00331781, -0.01196258, 0.01424106, -0.03876435, -0.0360326, -0.0211947, 0.00411794, 0.00194809, 0.01709978, 0.00240725, -0.01357874, -0.08201197, 0.00028075, 0.00433373, 0.00884444, 0.00636668, 0.00866142, -0.01676651, -0.02009776, 0.01145247, -0.00339291, -0.02142267, 0.02839607, -0.00955048, 0.01357461, 0.036533, 0.02180281, 0.02530635, 0.05473747, 0.01639242, -0.0064534, 0.01218683, 0.01897812, 0.02747294, 0.00920427, 0.02827672, 0.01947163, 0.03674255, 0.01123746, 0.00064828, 0.02112044, 0.01015427, 0.00795799, -0.01490599, 0.01035441, 0.03797704, 0.00894573, 0.00049762, 0.00639881, 0.00384943, -0.01778555, 0.01663044, 0.00974921, 0.02123337, -0.01016194, 0.00549926, -0.00854986, -0.00565578, 0.04361366, -0.00171666, 0.00527589, 0.00582117, -0.00980804, 0.00212335, -0.00456517, -0.00473958, -0.04356948, -0.01715631, 0.01194917, -0.0202417, 0.0005836, -0.01327072, -0.00214214, 0.00217193, 0.04005405, -0.00075428, 0.000556, -0.01876217, -0.00930269, -0.04083677, 0.00474346, 0.0197421, 0.03920778, 0.02430846, -0.00335513, 0.02624585, 0.0277168, 0.00965792, -0.00575595, 0.00957528, 0.01172352, 0.0187465, 0.01345966, 0.00756121, 0.00613374, -0.00859714, 0.00250639, -0.01373302, -0.0221034, 0.00576238, -0.00549897, -0.03135081, 0.00580172, -0.05249637, 0.01283482, 0.00825837, 0.0311162, 0.04761201, 0.01806747, -0.01647371, 0.00901918, 0.00402425, 0.01729331, 0.00285937, 0.03775802, -0.00719376, -0.00443339, -0.00292074, -0.014518, 0.02157991, -0.00644157, 0.0058136, -0.0140032, 0.01613103, 0.00490574, 0.05473264, 0.01168561, 0.00156543, -0.01618162, -0.00696151, -0.0413295, -0.00997392, -0.01876063, -0.02228089, 0.01021953, -0.00523615, 0.01207606, 0.00043311, -0.06499377, 0.0299222, -0.00841037, 0.03009915, -0.03534083, 0.0241453, -0.01399117, 0.00172493, 0.01968024, 0.0231289, 0.02813661, 0.02161097, 0.00998347, -0.023444, 0.00622823, -0.01743298, -0.01636894, 0.0057391, 0.00622567, 0.01908882, -0.01835142, -0.00238498, 0.00416884, -0.01642993, 0.00663776, 0.0078424, 0.02521407, 0.02109439, 0.01216206, 0.02298219, -0.00553557, 0.00653391, -0.00815363, -0.01787233, 0.00021755, 0.02368685, 0.00455294, 0.01300119, -0.01815774, 0.00836542, 0.02003333, 0.01744072, -0.00205533, 0.00294184, 0.00422834, -0.00681667, 7.038e-05, 0.01449687, -0.00348955, 0.02712471, 0.02293792, 0.0117723, 0.02824472, 0.0050289, 0.02710769, -0.02985661, 0.02778633, 0.00556758, -0.01496237, 0.00734616, 0.00228278, 0.01276059, -0.00595078, 0.02245223, -0.00324678, -0.00257048, 0.00411377, -0.01252966, 0.03350377, 0.01627953, 0.03409994, -0.01535645, -0.00115265, 0.0304749, -0.00707587, -0.02790663, -0.0440765, -0.02894666, 0.02789892, -0.01641196, -0.00374821, 0.00833597, 0.00653626, 0.00927995, -0.00510176, 0.01337331, -0.01163289, -0.00044067, 0.04714484, 0.02263353, 0.0198896, 0.02757403, -0.03915639, 0.01243817, -0.00015133, 0.01012254, 0.02740928, -0.00713481, -0.01387607, 0.0058999, 0.01401392, 0.0016258, 0.01238173, 0.00079326, 0.03022914, 0.01588522, 0.02195257, -0.01408552, -0.00434917, 0.00528793, -0.00420233, 0.0131251, 0.00447973, 0.03269675, 0.00391985, -0.00536354, 0.02376903, 0.01611643, 0.0004442, 0.0073623, -0.02064884, -0.01050849, -0.01081939, -0.00053128, -0.01569698, -0.01748251, -0.01115517, -0.0368048, -0.00608349, -0.00691206, 0.00119363, 0.01784694, -0.01563957, -0.00105553, -0.00412624, 0.02558698, 0.00535417, -0.01815499, 0.02173047, 0.02038895, 0.01184773, 0.01170407, 0.03102636, -0.01295243, -0.0268425, -0.01411292, 0.01309746, 0.01036491, -0.01859779, 0.01766764, -0.00614697, 0.03774634, 0.01979585, 0.04686963, -0.01473713, 0.02444057, -0.00396742, 0.01549287, -0.00996807, 0.01067487, 0.00674213, -0.01101863, -0.01666162, -0.03999902, 0.0134171, -0.00202965, -0.01242103, 0.01976216, 0.00524476, 0.00758222, 0.01076278, -0.00505364, -0.00425184, 0.03837146, 0.03884887, -0.0162502, 0.02975437, 0.00444965, 0.03610103, 0.02642073, -0.03548933, 0.01547195, 0.00644196, 0.02706225, 0.01800843, -0.00781093, -0.00486106, -0.0268721, -0.01905215, -0.00731676, -0.01124513, 0.01282586, 0.01430456, 0.00293278, -0.03970284, -0.02299645, -0.02275939, 0.00757372, -0.00179313, 0.00823983, -0.04368165, -0.00490686, 0.01017395, 0.02006527, -0.02266996, -0.01984322, -0.00431914, -0.03618793, -0.02906649, 0.00019875, 0.01462264, -0.02227929, 0.01260245, 0.00904468, 0.00200294, 0.02408413, -0.00533964, -0.0058466, 0.03226778, 0.00264232, -0.00519984, -0.03621732, 0.03478916, 0.01705477, 0.00978686, 0.01098218, -0.02563523, 0.00848929, -0.0126275, 0.02042658, -0.0066765, -0.00933558, -0.01677838, 0.02286891, 0.03010353, 0.01382014, 0.01426606, -0.02179327, 0.01253089, -0.00388169, 0.01392385, 0.0108657, 0.00157205, 0.01041973, -0.00554368, -0.00028152, 0.00545052, -0.00374843, -0.02227512, -0.00266097, 0.00254435, 0.00698552, -0.02063029, 0.02130756, 0.0179806, -0.03979089, -0.01482454, 0.01742368, 0.01346612, -0.02217621, -0.00303823, -0.02330962, 0.01857341, 0.01308789, 0.02105988, -0.0015126, -0.01697375, 0.02212925, 0.00411078, -0.01711467, 0.04253754, 0.00250948, 0.02619134, 0.03438265, -0.0087473, 0.01052142, 0.01052836, 0.0064118, -0.01756983, 0.01307795, -0.01589733, 0.02563347, -0.0133308, -0.00259087, -0.01713379, -0.01071607, -0.03534602, 0.00652011, 0.0097325, 0.00910617, 0.01646658, -0.02396664, 0.00715146, -0.04202237, -0.02859035, 0.00242289, -0.02271108, 0.04036607, -0.02535962, -0.00285476, 0.00719117, 0.02030975, -0.00159972, -0.01081652, 0.01695991, -0.01595064, -0.01562011, 0.00363326, -0.01818914, -0.0451778, 0.01674915, 0.01408524, -0.00497298, 0.02133562, 0.01122899, 0.02453887, -0.0120907, -0.02230058, 0.02245758, -0.00365998, -0.01320692, -0.02177522, -0.00991325, 0.0139793, 0.00438002, -0.01485207, -0.01792047, -0.03305126, 0.02019338, -0.00272138, -0.02303404, -0.00690035, -0.01574001, 0.00067245, -0.00380218, 0.01910563, 0.02367547, -0.01223903, -0.02684288, -0.03173235, -0.00261731, 0.00736793, -0.01114917, -0.01880739, -0.0106492, 0.01610846, 0.0127876, -0.01013898, 0.02557062, 0.03280301, -0.01123896, -0.00435923, -0.01225559, 0.00107175, -0.01544293, -0.00554472, 0.00800751, 0.00411687, 0.00238706, 0.02265125, 0.01943326, -0.04440862, 0.01470459, 0.01201839, -0.01385308, 0.00590084, -0.00615393, -0.0043049, -0.00260483, 0.01390211, -0.00013944, -0.01767793, -0.01445556, 0.00995654, -0.02865898, -0.008709, 0.00910513, 0.00847967, 0.01707621, -0.00887391, -0.01253253, 0.00312656, 0.01035588, 0.00641921, -0.01029581, 0.01132845, -0.02236251, -0.02060476, 0.01289726, 0.00812421, -0.00296098, 0.02756017, 0.01764038, -0.01906771, -0.03088378, 0.00600311, 0.00992661, 0.0098976, -0.00691109, -0.03716392, -0.01981835, -0.01740552, -0.03400804, 0.02310871, -0.03407235, -0.00059299, 0.0083054, -0.00597733, -0.02159307, -0.00415138, 0.00054489, -0.0221254, -0.03017242, 0.02797676, 0.01107227, -0.00976945, 0.01308582, 0.04042145, -0.02927984, -0.00921222, -0.02420728, 0.00908838, -0.00090499, 0.02408376, 0.00399322, 0.01176275, 0.02528179, -0.01991752, -0.00282033, -0.00487063, -0.01095138, -0.00946819, 0.01733616, 0.0195776, -0.0128392, -0.00805821, 0.02574589, 0.00500291, 0.03921548, 0.00336303, -0.01169733, -0.00991731, -0.02538016, -0.0163031, 0.01803913, -0.00319162, -0.00572383, -0.01124951, -0.00694349, -0.00085283, -0.02370997, -0.01627219, -0.03391844, -0.03988779, -0.00684243, -0.00188227, 0.00997175, -0.00956665, -0.02754596, -0.00666886, -0.0042144, 0.01658166, -0.03475415, -0.04894758, -0.01744534, -0.02482471, -0.03251565, -0.01131866, -0.00662228, -0.02328805, 0.00759126, -0.01911985, -0.0215375 ], "cols": [ 0, 11, 25, 35, 37, 38, 39, 46, 53, 61, 63, 65, 67, 68, 79, 84, 105, 129, 135, 169, 215, 225, 259, 264, 266, 294, 304, 322, 338, 371, 406, 409, 414, 426, 452, 496, 504, 505, 509, 582, 592, 601, 682, 733, 738, 881, 882, 892, 926, 956, 1060, 1087, 1130, 1184, 1341, 1702, 1787, 2066, 2159, 2252, 2288, 2431, 2612, 3535 ] }, "19": { "alpha": 6, "dir": [ -0.40303805, 0.0329975, 0.10403553, -0.0209099, 0.09365643, -0.192155, 0.02698611, 0.02667853, -0.00437309, -0.0299947, 0.07137594, 0.05299167, 0.0937297, 0.08620736, 0.042626, 0.1429681, 0.04721639, 0.25699237, -0.12434925, -0.07475132, 0.08786446, 0.06968278, 0.07975617, 0.08826555, -0.0397184, -0.00341501, -0.01464794, 0.00140585, 0.00736103, 0.06114507, 0.02299195, -0.13341559, 0.05640099, -0.03847188, -0.08698785, 0.00476994, -0.03285822, 0.1037427, -0.04276746, -0.01313698, -0.10109258, -0.00824235, 0.00531195, -0.02284941, -0.01404601, -0.00765076, 0.05599665, -0.06144537, 0.03866914, 0.07693519, 0.01647117, -0.01187289, -0.04359513, -0.00133371, 0.02434327, 0.06511034, 0.05916858, 0.07504672, -0.05050368, -0.05953808, 0.01576725, -0.07817139, -0.05695913, 0.02084892, 0.05844348, 0.02503907, -0.00118718, -0.00996915, -0.05700012, -0.06182868, 0.050662, 0.00675236, -0.05990036, 0.00560643, -0.07600475, 0.02052813, -0.04989911, 0.03428475, -0.01068215, 0.0253167, 0.054955, 0.1112799, 0.02270513, -0.02924658, 0.03459894, 0.09027974, 0.09684429, -0.03744385, 0.01228896, 0.02916334, 0.04298693, -0.03321001, -0.05384923, -0.00912412, 0.03467786, -0.01946466, -0.01537559, -0.00564796, 0.00591098, 0.01540753, -0.05764904, -0.04066702, -0.0189424, 0.02351499, -0.03018603, 0.00088564, 0.03036759, 0.00726154, -0.03324075, 0.01276938, 0.03604248, 0.04482879, 0.006666, -0.03563202, -0.00748893, 0.04663388, -0.01966599, -0.04702299, -0.00993462, 0.04537556, -0.00204271, 0.03050143, -0.00201608, 0.03898253, -0.047811, 0.00189243, -0.03170019, 0.01476422, -0.00809932, 0.04813007, -0.04051394, 0.00068748, 0.023313, 0.0306307, 0.02844512, -0.00848589, -0.01803911, -0.06040058, 0.05202504, -0.00238838, 0.01033624, 0.02365543, 0.08331686, 0.04461651, -0.01148987, 0.00163412, -0.00153335, -0.00516465, 0.01470824, -0.00489359, -0.01607419, 0.00020207, 0.02367819, 0.02733325, 0.00063969, -0.00831591, -0.0209804, 0.00126793, -0.00726657, 0.00066406, 0.06348248, 0.00201157, 0.00396431, -0.00344394, -0.04016386, -0.00017567, -0.00910502, -0.03662831, -0.02763264, -0.00310353, -0.03082627, -0.02265084, -0.00451601, 0.02336263, 0.00673821, -0.03130874, 0.05797688, -0.00218501, 0.02092489, -0.00539545, -0.03473163, 0.01805203, -0.00012734, -0.00400688, -0.02478947, -0.01183948, -0.04739719, -0.00670978, 0.00084123, 0.01346655, -0.04946645, 0.01839454, 0.03365371, 0.00902308, -0.01371482, 0.03675059, 0.0009716, -0.00247339, -0.01003343, -0.04753535, 0.01870654, -0.02161889, -0.02098623, 0.0019909, -0.00476361, 0.04148345, -0.00323742, -0.014203, -0.04067614, -0.00110696, -0.01084116, 0.00099532, -0.04544856, -0.02752233, -0.01037184, 0.00340519, 0.01071695, -3.053e-05, 0.00802704, -0.08500948, -0.02814225, -0.00555921, 0.02564316, -0.02478302, 0.02557177, -0.07124162, 0.01407428, 0.0010604, -0.01927022, 0.0063734, -0.03984567, 0.06196889, -0.00235534, 0.02643645, -0.00265604, -0.01699017, 0.00643226, 0.01197971, -0.02531807, 0.03990724, 0.00612159, 0.02886316, 0.02454276, -0.00286661, 0.04416312, -0.01598255, 0.01194422, -0.03382059, -0.01258989, 0.00065155, 0.03416109, -0.00918627, -0.01756595, -0.00544739, 0.00968638, 0.08575365, -0.04310172, -0.02123114, 0.03103097, 0.01038746, -0.03642382, 0.00219577, 0.01220632, 0.01007034, 0.01126756, 0.00783913, 0.01929826, 0.0019949, 0.00045626, 0.00400553, 0.00912561, -0.01650603, -0.00431264, -0.0388132, -0.02495558, 0.01603597, 0.00626655, -0.03401203, -0.03112821, -0.06809703, 0.02446333, -0.01712414, -0.04576444, 0.01292116, 0.01258606, -0.01130493, 0.03311132, -0.0188331, -0.00239761, 0.00029516, 0.01813175, -0.00711609, 0.00427899, -0.00198381, 0.01185463, -0.02622689, 0.01642408, -0.00308381, -0.01048345, -0.02013568, 0.07661538, -0.01446139, -0.00600802, 0.05260718, -0.02086512, 0.01684219, -0.01044171, -0.0230399, -0.04302485, 0.04375228, -0.00500612, -0.03674692, -0.0032724, 0.0205769, -0.01249505, 0.00432687, -0.06169629, -0.05315479, 0.01014823, 0.04117028, 0.01495056, -0.001776, 0.00032922, -0.0059673, -0.02366105, -0.00312147, 0.02290104, 0.00848917, 0.00966943, -0.03708085, 0.03881599, 0.00190149, 0.00513811, 0.04708545, 0.01801702, -0.01748442, 0.00842807, 0.01615084, 0.0050975, 0.03528532, -0.14008042, 0.02364798, -0.04108557, -0.01698508, -0.01404503, 0.00869193, -0.00010824, -0.01547313, -0.01626009, 0.04450187, -0.01098249, -0.01527852, -0.00475868, -0.0030468, -0.02795796, -0.04963853, -0.01585494, -0.02109603, 0.01524505, -0.02236662, 0.01400957, 0.0064209, 0.03387895, -0.02655072, 0.02563305, 0.0312883, -0.00533046, 0.00821784, 0.04589652, 0.00789291, -0.00199905, -0.00011294, -0.00419002, 0.00910988, -0.03101155, -0.02826713, -0.00271063, 0.02451408, -0.01221345, -0.02056752, 0.00471987, 0.02008954, -0.00956004, 0.00519779, 0.00242066, -0.00584248, -0.0048598, 0.0114145, 0.00469483, 0.00931662, -0.00462288, 0.00622943, -0.00053509, -0.00149505, 0.00647583, 0.03319455, -0.01123795, -0.00840665, -0.00263404, 0.01159186, 0.01139623, 0.00565191, 0.00258509, -0.00489478, -0.00853801, -0.0091461, -0.05054704, 0.02477333, 0.0021998, -0.01492881, 0.02183418, -0.02420258, 0.00471695, -0.00113598, -0.03845035, -0.00117403, -0.06199903, -0.01917586, -0.02454664, 0.00087716, -0.01349168, -0.01287688, -0.01245939, -0.02629074, -0.09974691, 0.01091907, -0.00391323, -0.01096682, 0.01214593, -0.00946139, -0.02915651, 0.00937881, 0.00562696, -0.00936563, 0.00256885, 0.0268971, -0.0116727, 0.01549586, 0.01316128, 0.02284234, 0.00909617, -0.00809629, -0.01953212, -0.0129853, 0.01894005, 0.0325468, 0.01911852, 0.02675558, 0.01520315, 0.00495247, 0.01502663, 0.01590631, -0.00530423, 0.0019599, 0.0162559, -0.0012955, -0.01707534, 0.01930814, 0.03617071, -0.0019372, -0.00707838, 0.00701538, 0.00695747, -0.00277397, 0.02455544, 0.01111011, 0.02052771, 0.00082167, 0.02250673, -0.00798365, -0.00801462, 0.0508317, -9.375e-05, 0.01904196, -0.00919574, 0.01167577, 0.00546223, -0.01508579, -0.0006264, -0.04727374, -0.03377324, 0.0027254, -0.03818843, 0.00818787, -0.00689099, -0.01197335, -0.00516053, 0.00646393, 0.00846315, 0.01195741, -0.0365892, -0.00745085, -0.00428318, 0.007973, -0.01314605, 0.02908652, 0.01298066, -0.0173705, 0.02419879, 0.00218602, 0.01928374, 0.00508567, -0.00211529, 0.01487939, 0.00837229, -0.00011635, -0.01281008, -0.01408992, -0.00351064, 0.04980747, -0.00465233, -0.01279383, -0.00057706, -0.01141001, -0.02609113, 0.00264699, -0.0565442, 0.01816659, 0.00806187, 0.02342308, 0.02896993, 0.01568547, -0.02374927, -0.00248609, -0.00241215, 0.01282237, 0.00571339, -0.00210891, 0.0047637, 0.00098846, -0.02166379, -0.0131998, 0.00795409, -0.01523779, 0.01452046, -0.00845477, 0.02079317, 0.00692318, 0.02653175, 0.0042664, -0.01262448, -0.00510448, 4.957e-05, -0.03905357, -0.00251511, -0.01699689, -0.00689903, 0.02518062, 0.01212352, 0.03058587, -0.00656732, -0.05452856, 0.01735215, 0.00558136, 0.00143205, -0.02730398, -0.00217664, 0.00567189, -0.00826461, 0.02103444, 0.00645421, 0.02649493, 0.005601, 0.00488648, -0.01254205, -0.00431054, -0.01591754, 0.01170847, 0.00488227, 0.01092794, 0.01418996, -0.00660805, 0.00773353, 0.01112791, 0.00906591, 0.02489558, 0.01874612, 0.07885447, 0.02199026, 0.02615559, 0.01782136, 0.00516706, -0.01399359, -0.00202788, -0.02244174, 0.00386115, -0.0341826, -0.00116838, 0.01111056, -0.01181473, 0.00046735, 0.01640572, 0.03063136, 0.00704706, -0.01418224, -0.00518598, 0.00525866, 0.02534688, 0.00566562, -0.01091902, 0.01498393, -0.00909562, 0.00326286, 0.02358589, 0.00790946, 0.02306728, -0.0262363, 0.00915681, -0.00484231, 0.00258546, 0.0079408, 0.00690506, 0.00095181, -0.01958587, 0.00112982, -0.02001894, -0.02493037, -0.0326503, -0.01394145, 0.0077762, 0.01214608, 0.02613667, -0.01132605, 0.00931333, 0.0387103, -0.01500592, -0.01783975, -0.04186295, -0.02266708, 0.02314912, -0.01698104, -0.01665973, -0.005301, 0.00042177, 0.00232282, -0.00772069, -0.0053295, -0.03100577, -0.01421619, 0.02114789, 0.03915536, 0.02150632, 0.00759725, -0.0334801, 0.00841852, -0.01916282, 0.01262707, 0.03581546, -0.01396338, -0.0083547, 0.00014665, -0.00206262, 0.00047464, 0.00686827, 0.01449366, 0.0249184, 0.01602543, 0.02268857, -0.0116243, 0.0245544, 0.01862229, 0.01024355, 0.00836354, 0.01124888, 0.01107886, -0.01351656, -0.01871679, 0.0171045, 0.01118574, -0.01734518, 0.02602814, -0.02101769, 0.0154846, -0.02667392, -0.01784633, -0.01317136, -0.01633694, 0.008699, -0.0527638, -0.00203045, -0.0157266, -0.02169789, 0.02152181, -0.01646204, 0.00960654, 0.00509686, -0.00608442, 0.02444478, -0.03974782, 0.01612161, 0.00677621, 0.01147696, 0.04470504, 0.00256618, -0.00952843, 0.02406729, -0.04679559, -0.00774345, -0.01386489, 0.01363496, -0.00101011, 0.01204833, 0.00770687, 0.01788205, -0.01635344, -0.00641703, 0.00463655, 0.01557082, 0.01394424, 0.00123676, 0.01102833, 0.00125298, -0.01286629, -0.02460117, -0.01925873, 0.01671567, -0.01477258, -0.00107901, 0.00754077, -0.01119798, -0.00935859, -0.00847786, 0.00715216, -0.00157592, 0.00621346, 0.01044471, -0.03625395, -0.00521711, -0.00452813, 0.03532019, 0.03946002, -0.00864522, -0.00942822, -0.01687333, 0.02638818, 0.02313282, -0.00988636, -0.02514545, -0.00451056, -0.01856106, -0.00306914, -0.00260156, 0.02972909, 0.01877441, 0.00071083, -0.03811838, -0.00868523, -0.01375061, 0.02080228, 0.00818848, -0.00023008, -0.00979582, -0.00510938, -0.03489744, 0.01566195, 0.00826539, -0.01311246, -0.01605817, -0.03151313, -0.01793739, -0.02454815, 0.04060522, -0.02754511, 0.00577689, 0.00037588, 0.00284941, 0.01398198, 0.01095516, 0.00507469, 0.01926889, -0.02177124, 0.00352636, -0.02421063, 0.01897777, 0.00794042, -0.01903835, -0.01019955, -0.01215394, 0.0181435, -0.01626652, -0.01552044, -0.00056849, -0.00855363, -0.01600707, 0.01713308, 0.00498583, 0.01981701, -0.01577345, -0.01453055, -0.00358987, 0.01257273, 0.01170709, 0.01619244, -0.02788386, 0.00302789, -0.00795681, -0.03881671, -0.0254103, -0.00649084, 0.00401415, -0.00885045, 0.00897933, 0.00846715, -0.01944916, 0.01098515, 0.02958035, 0.002796, -0.00970202, 0.00264912, -0.00209761, -0.0297788, 0.00745154, 0.00078796, 0.0077956, -0.02616478, 0.01565293, -0.01397188, -0.01799043, 0.01114042, -0.01150207, -0.00730181, -0.01888783, -0.00477981, 0.01119853, 0.02606427, -0.00990112, 0.00452353, 0.00097743, 0.00404155, -0.02092475, 0.0198313, -0.02845686, 0.02988892, -0.01967429, -0.00234447, -0.00169516, -0.00404113, -0.02696731, 0.01209176, 0.00248254, 0.03847836, 0.00579242, -0.00249322, 0.01219832, -0.03304773, -0.02388715, 0.02146677, -0.01525156, 0.01962912, -0.00011136, -0.00730731, 0.00743123, 0.02491707, -0.02086556, 0.0011956, 0.01434948, -0.00067252, -0.01052126, -0.00089142, -0.02716547, -0.04152113, 0.01699185, -0.02839208, -0.00994837, 0.0032557, -0.00084267, 0.03056953, -0.02806186, -0.00636801, 0.0102405, 0.0392765, -0.02917243, -0.02395096, 0.0030348, 0.01891886, 0.004632, -0.0252702, -0.01215409, 0.0046654, -0.0112583, -0.01865133, -0.03766386, -0.01420537, -0.00806902, -0.00121897, -0.00733954, 0.03316591, -0.01875471, 0.00280505, -0.0154234, 0.00519485, -0.03003224, -0.00622006, -0.00091026, -0.01525642, -0.00659069, 0.0192284, 0.0172484, -0.00925321, -0.00728299, -0.00394578, -0.0163736, 0.00108747, -0.00069319, 0.00864509, 0.0077902, -0.01124784, 0.02520256, -0.00707292, 0.0137952, -0.00608825, 0.01473028, -0.03668444, 0.0025801, 0.01427094, 0.00367407, 0.02259825, -0.02711322, -0.00902057, -0.01900061, -0.00815246, -0.00609889, -0.01662255, -0.00451035, 0.00151863, -0.01451367, -0.02145989, -0.01448709, -0.00681982, -0.00208977, -0.00806661, 0.00479471, -0.00269823, 0.00959441, 0.02306974, -0.00716286, 0.00366802, -0.02434608, -0.01507821, -0.00585343, 0.01796961, 0.01575135, 0.00706694, -0.00562256, -0.00824497, 0.00831951, -0.0044277, 0.01774421, -0.00207182, -0.00403484, -0.04845684, -0.00237786, -0.01226677, -0.00758332, 0.00172534, -0.01722374, 0.01717514, 0.02007657, 0.02328251, -0.00864593, 0.00052561, -0.03738995, -0.00440814, 0.02000018, -0.0119994, 0.01025173, -0.03096017, -0.00707121, 0.02796363, -0.01488446, -0.00993766, -0.01532069, 0.00665802, -0.01918009, -0.00327633, -0.00208455, -0.00894725, 0.00117816, -0.00573215, -0.01161137, -0.01661021, 0.00335749, -0.01907716, 0.01830916, 0.00507185, 0.00775102, -0.0079882, 0.01377143, -0.00037081, 0.0005024, 0.01686083, 0.00347498, 0.00391057, 0.00955494, -0.03267493, 0.00041566, -0.0044906, 0.02599197, -0.00566409, -0.01121515, 0.01122707, -0.01300336, -0.00945701, -0.05133087, -0.00125735, -0.00112834, 0.01224202, -0.0007026, 0.00103709, -0.00757639, 0.00300311, 0.00511677, 0.01669963, -0.01687949, -0.01683812, 0.00910018, -0.02204714, 0.00895432, 0.00274409, 0.00975537, -0.04031643, -0.01076295, 0.00777576, 0.00186366 ], "cols": [ 1, 3, 11, 13, 14, 15, 20, 24, 26, 34, 36, 41, 53, 57, 61, 63, 64, 66, 69, 76, 79, 95, 98, 99, 101, 119, 123, 135, 145, 146, 152, 161, 171, 180, 182, 225, 226, 246, 253, 265, 268, 280, 324, 343, 351, 355, 380, 450, 489, 585, 602, 614, 705, 795, 836, 998, 1003, 1088, 1246, 1973, 2055, 2107, 2661, 2899 ] } } }, "built": { "15": { "alpha": 6.0, "n_cols": 64, "dir": "custom", "relative_change": 0.09710153937339783 }, "19": { "alpha": 6.0, "n_cols": 64, "dir": "custom", "relative_change": 0.09152013063430786 } }, "edited_sha256": { "model.language_model.layers.15.mlp.down_proj.weight": "7740fda5dd25780bd915fee4c7fb04c15a1e9cd40968b75310595cd42107874d", "model.language_model.layers.19.mlp.down_proj.weight": "c71e2cfeff579a6fd22723e1076bb2601809e053205b0d633d55f21e78d41b4c" } }, "tokens_total": 10611301, "gpu_seconds": 1684, "tok_per_s_per_gpu": 6551 }
{
"problems": 256,
"samples": 512,
"acc": 0.537109375,
"acc_se": 0.027922737013093644,
"cut_share": 0.396484375,
"mean_tokens": 20725.197265625,
"closed_share": 0.70703125,
"p_finish": 0.603515625,
"p_correct_given_finished": 0.8867313915857605,
"p_correct_given_cut": 0.0049261083743842365,
"no_answer_finished_share": 0.00390625,
"loop_end_share_of_cut": 0.7684729064039408,
"tail_dup8_cut_mean": 0.7750477832512315,
"median_tokens_correct": 11535,
"by_source": {
"gsm8k": {
"problems": 128,
"samples": 256,
"acc": 0.69140625,
"acc_se": 0.03641858274265708,
"cut_share": 0.203125,
"mean_tokens": 16059.3515625
},
"math": {
"problems": 128,
"samples": 256,
"acc": 0.3828125,
"acc_se": 0.037812174626048516,
"cut_share": 0.58984375,
"mean_tokens": 25391.04296875
}
},
"run": {
"name": "dev/capB-max-samples-16",
"model": "qwen35-capB-max-samples-16",
"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": "capB-max-samples-16",
"spec": {
"about": "Lab 8 recipe B: u and 64 neurons from the base model's own loops (windows well before the loop), alpha 6",
"layers": {
"15": {
"alpha": 6,
"dir": [
-0.21637826,
0.0208582,
0.14215979,
0.03148801,
0.05498521,
-0.20134567,
0.02097962,
-0.04125047,
-0.00052325,
0.00733103,
0.07352018,
0.09851355,
0.05241353,
0.15527157,
0.02442253,
0.18563084,
0.05889465,
0.20897448,
-0.13037334,
-0.07115818,
0.07148564,
0.1075747,
0.09280334,
0.09322366,
-0.03304736,
0.0135476,
-0.0039322,
-0.02402865,
0.02404705,
0.08518188,
0.03829363,
-0.15843356,
0.02383504,
-0.04695624,
-0.09074076,
-0.01772078,
-0.07573749,
0.16334917,
-0.03774936,
-0.00817425,
-0.09472003,
0.00438293,