diff --git "a/checkpoints/codi-single-1.5b/checkpoint-2500/trainer_state.json" "b/checkpoints/codi-single-1.5b/checkpoint-2500/trainer_state.json" new file mode 100644--- /dev/null +++ "b/checkpoints/codi-single-1.5b/checkpoint-2500/trainer_state.json" @@ -0,0 +1,5046 @@ +{ + "best_global_step": null, + "best_metric": null, + "best_model_checkpoint": null, + "epoch": 0.720876585928489, + "eval_steps": 500, + "global_step": 2500, + "is_hyper_param_search": false, + "is_local_process_zero": true, + "is_world_process_zero": true, + "log_history": [ + { + "epoch": 0.0014417531718569781, + "grad_norm": 91136.0, + "kd_loss": 0.53515625, + "learning_rate": 1.3333333333333334e-06, + "loss": 1.4874, + "step": 5, + "student_loss": 0.6224480271339417, + "teacher_loss": 0.00038480176590383053 + }, + { + "epoch": 0.0028835063437139563, + "grad_norm": 29312.0, + "kd_loss": 0.54296875, + "learning_rate": 3e-06, + "loss": 1.3555, + "step": 10, + "student_loss": 0.7942929863929749, + "teacher_loss": 0.0007905907114036381 + }, + { + "epoch": 0.004325259515570935, + "grad_norm": 25088.0, + "kd_loss": 0.515625, + "learning_rate": 4.666666666666667e-06, + "loss": 1.3876, + "step": 15, + "student_loss": 0.2712247967720032, + "teacher_loss": 0.0006367590394802392 + }, + { + "epoch": 0.0057670126874279125, + "grad_norm": 11328.0, + "kd_loss": 0.5234375, + "learning_rate": 6.333333333333333e-06, + "loss": 1.3195, + "step": 20, + "student_loss": 0.991253137588501, + "teacher_loss": 0.00029781984630972147 + }, + { + "epoch": 0.00720876585928489, + "grad_norm": 4896.0, + "kd_loss": 0.46875, + "learning_rate": 8.000000000000001e-06, + "loss": 1.2722, + "step": 25, + "student_loss": 0.24368858337402344, + "teacher_loss": 0.0009928465588018298 + }, + { + "epoch": 0.00865051903114187, + "grad_norm": 35584.0, + "kd_loss": 0.48828125, + "learning_rate": 9.666666666666667e-06, + "loss": 1.2823, + "step": 30, + "student_loss": 0.7547083497047424, + "teacher_loss": 0.0003548521490301937 + }, + { + "epoch": 0.010092272202998846, + "grad_norm": 11584.0, + "kd_loss": 0.458984375, + "learning_rate": 9.99958042442916e-06, + "loss": 1.1068, + "step": 35, + "student_loss": 0.48039907217025757, + "teacher_loss": 0.00026277394499629736 + }, + { + "epoch": 0.011534025374855825, + "grad_norm": 33536.0, + "kd_loss": 0.5, + "learning_rate": 9.997876019358083e-06, + "loss": 1.174, + "step": 40, + "student_loss": 0.7645292282104492, + "teacher_loss": 0.07346436381340027 + }, + { + "epoch": 0.012975778546712802, + "grad_norm": 3664.0, + "kd_loss": 0.478515625, + "learning_rate": 9.99486100792044e-06, + "loss": 0.9907, + "step": 45, + "student_loss": 1.340754508972168, + "teacher_loss": 0.0009389458573423326 + }, + { + "epoch": 0.01441753171856978, + "grad_norm": 1584.0, + "kd_loss": 0.486328125, + "learning_rate": 9.990536180750724e-06, + "loss": 0.9709, + "step": 50, + "student_loss": 0.31134045124053955, + "teacher_loss": 0.00372113729827106 + }, + { + "epoch": 0.015859284890426758, + "grad_norm": 672.0, + "kd_loss": 0.44921875, + "learning_rate": 9.984902671959911e-06, + "loss": 0.7719, + "step": 55, + "student_loss": 0.11045902222394943, + "teacher_loss": 0.0004030822601635009 + }, + { + "epoch": 0.01730103806228374, + "grad_norm": 648.0, + "kd_loss": 0.447265625, + "learning_rate": 9.97796195883804e-06, + "loss": 0.725, + "step": 60, + "student_loss": 0.15876038372516632, + "teacher_loss": 0.0018200079211965203 + }, + { + "epoch": 0.018742791234140715, + "grad_norm": 490.0, + "kd_loss": 0.447265625, + "learning_rate": 9.969715861466839e-06, + "loss": 0.6844, + "step": 65, + "student_loss": 0.2451532483100891, + "teacher_loss": 0.056328896433115005 + }, + { + "epoch": 0.020184544405997693, + "grad_norm": 210.0, + "kd_loss": 0.4453125, + "learning_rate": 9.96016654224243e-06, + "loss": 0.6719, + "step": 70, + "student_loss": 0.036203403025865555, + "teacher_loss": 0.0021477588452398777 + }, + { + "epoch": 0.02162629757785467, + "grad_norm": 366.0, + "kd_loss": 0.44921875, + "learning_rate": 9.94931650530827e-06, + "loss": 0.6026, + "step": 75, + "student_loss": 0.009898507036268711, + "teacher_loss": 0.0007659209077246487 + }, + { + "epoch": 0.02306805074971165, + "grad_norm": 1816.0, + "kd_loss": 0.462890625, + "learning_rate": 9.93716859589851e-06, + "loss": 0.6433, + "step": 80, + "student_loss": 0.030777160078287125, + "teacher_loss": 0.0008218760485760868 + }, + { + "epoch": 0.024509803921568627, + "grad_norm": 360.0, + "kd_loss": 0.470703125, + "learning_rate": 9.923725999591846e-06, + "loss": 0.5588, + "step": 85, + "student_loss": 0.07332021743059158, + "teacher_loss": 0.00044353402336128056 + }, + { + "epoch": 0.025951557093425604, + "grad_norm": 732.0, + "kd_loss": 0.39453125, + "learning_rate": 9.908992241476189e-06, + "loss": 0.5723, + "step": 90, + "student_loss": 0.07177069783210754, + "teacher_loss": 0.00044491401058621705 + }, + { + "epoch": 0.027393310265282585, + "grad_norm": 250.0, + "kd_loss": 0.4296875, + "learning_rate": 9.892971185224244e-06, + "loss": 0.5615, + "step": 95, + "student_loss": 0.25651854276657104, + "teacher_loss": 0.0204803254455328 + }, + { + "epoch": 0.02883506343713956, + "grad_norm": 195.0, + "kd_loss": 0.42578125, + "learning_rate": 9.875667032080354e-06, + "loss": 0.5376, + "step": 100, + "student_loss": 0.0018209030386060476, + "teacher_loss": 0.0005951338680461049 + }, + { + "epoch": 0.03027681660899654, + "grad_norm": 74.5, + "kd_loss": 0.416015625, + "learning_rate": 9.857084319758772e-06, + "loss": 0.5554, + "step": 105, + "student_loss": 0.009238015860319138, + "teacher_loss": 0.000378329394152388 + }, + { + "epoch": 0.031718569780853516, + "grad_norm": 134.0, + "kd_loss": 0.3984375, + "learning_rate": 9.837227921253747e-06, + "loss": 0.5376, + "step": 110, + "student_loss": 0.008933513425290585, + "teacher_loss": 0.054748453199863434 + }, + { + "epoch": 0.03316032295271049, + "grad_norm": 53.5, + "kd_loss": 0.380859375, + "learning_rate": 9.816103043561648e-06, + "loss": 0.5101, + "step": 115, + "student_loss": 0.0015179987531155348, + "teacher_loss": 0.0003994991711806506 + }, + { + "epoch": 0.03460207612456748, + "grad_norm": 90.5, + "kd_loss": 0.34765625, + "learning_rate": 9.79371522631553e-06, + "loss": 0.4575, + "step": 120, + "student_loss": 0.005917699541896582, + "teacher_loss": 0.0011113588698208332 + }, + { + "epoch": 0.036043829296424454, + "grad_norm": 105.0, + "kd_loss": 0.333984375, + "learning_rate": 9.770070340332457e-06, + "loss": 0.4262, + "step": 125, + "student_loss": 0.15691499412059784, + "teacher_loss": 0.0019031435949727893 + }, + { + "epoch": 0.03748558246828143, + "grad_norm": 16.375, + "kd_loss": 0.3125, + "learning_rate": 9.745174586073982e-06, + "loss": 0.4434, + "step": 130, + "student_loss": 0.0024075880646705627, + "teacher_loss": 0.007021783851087093 + }, + { + "epoch": 0.03892733564013841, + "grad_norm": 21.625, + "kd_loss": 0.294921875, + "learning_rate": 9.719034492020183e-06, + "loss": 0.3819, + "step": 135, + "student_loss": 0.002887926297262311, + "teacher_loss": 0.00043111268314532936 + }, + { + "epoch": 0.040369088811995385, + "grad_norm": 11.875, + "kd_loss": 0.26953125, + "learning_rate": 9.691656912957686e-06, + "loss": 0.3717, + "step": 140, + "student_loss": 0.0013271740172058344, + "teacher_loss": 0.0008426732383668423 + }, + { + "epoch": 0.04181084198385236, + "grad_norm": 11.6875, + "kd_loss": 0.259765625, + "learning_rate": 9.663049028182112e-06, + "loss": 0.3423, + "step": 145, + "student_loss": 0.11146840453147888, + "teacher_loss": 0.0059180548414587975 + }, + { + "epoch": 0.04325259515570934, + "grad_norm": 7.09375, + "kd_loss": 0.2333984375, + "learning_rate": 9.633218339615433e-06, + "loss": 0.3012, + "step": 150, + "student_loss": 0.001806147862225771, + "teacher_loss": 0.001238134689629078 + }, + { + "epoch": 0.04469434832756632, + "grad_norm": 6.84375, + "kd_loss": 0.201171875, + "learning_rate": 9.602172669838721e-06, + "loss": 0.3157, + "step": 155, + "student_loss": 0.0033397674560546875, + "teacher_loss": 0.0009877807460725307 + }, + { + "epoch": 0.0461361014994233, + "grad_norm": 7.0625, + "kd_loss": 0.22265625, + "learning_rate": 9.569920160040815e-06, + "loss": 0.2821, + "step": 160, + "student_loss": 0.22769513726234436, + "teacher_loss": 0.041321102529764175 + }, + { + "epoch": 0.04757785467128028, + "grad_norm": 5.96875, + "kd_loss": 0.1943359375, + "learning_rate": 9.536469267883432e-06, + "loss": 0.2727, + "step": 165, + "student_loss": 0.00636801915243268, + "teacher_loss": 0.00474645895883441 + }, + { + "epoch": 0.049019607843137254, + "grad_norm": 6.375, + "kd_loss": 0.166015625, + "learning_rate": 9.501828765283295e-06, + "loss": 0.2549, + "step": 170, + "student_loss": 0.00595078757032752, + "teacher_loss": 0.002222016453742981 + }, + { + "epoch": 0.05046136101499423, + "grad_norm": 5.875, + "kd_loss": 0.154296875, + "learning_rate": 9.466007736111846e-06, + "loss": 0.249, + "step": 175, + "student_loss": 0.0170090701431036, + "teacher_loss": 0.001197761739604175 + }, + { + "epoch": 0.05190311418685121, + "grad_norm": 6.5, + "kd_loss": 0.15234375, + "learning_rate": 9.429015573813163e-06, + "loss": 0.2458, + "step": 180, + "student_loss": 0.0050661382265388966, + "teacher_loss": 0.002651061164215207 + }, + { + "epoch": 0.05334486735870819, + "grad_norm": 4.25, + "kd_loss": 0.169921875, + "learning_rate": 9.390861978940687e-06, + "loss": 0.2313, + "step": 185, + "student_loss": 0.3254599869251251, + "teacher_loss": 0.0007031414425000548 + }, + { + "epoch": 0.05478662053056517, + "grad_norm": 7.71875, + "kd_loss": 0.177734375, + "learning_rate": 9.351556956613423e-06, + "loss": 0.2414, + "step": 190, + "student_loss": 0.003914886154234409, + "teacher_loss": 0.001010580570437014 + }, + { + "epoch": 0.056228373702422146, + "grad_norm": 6.84375, + "kd_loss": 0.1435546875, + "learning_rate": 9.31111081389227e-06, + "loss": 0.2156, + "step": 195, + "student_loss": 0.06280706077814102, + "teacher_loss": 0.0005038110539317131 + }, + { + "epoch": 0.05767012687427912, + "grad_norm": 3.5625, + "kd_loss": 0.140625, + "learning_rate": 9.269534157077177e-06, + "loss": 0.1981, + "step": 200, + "student_loss": 0.002929725218564272, + "teacher_loss": 0.0005093662184663117 + }, + { + "epoch": 0.0591118800461361, + "grad_norm": 4.96875, + "kd_loss": 0.150390625, + "learning_rate": 9.226837888925813e-06, + "loss": 0.2157, + "step": 205, + "student_loss": 0.00641452893614769, + "teacher_loss": 0.03549756482243538 + }, + { + "epoch": 0.06055363321799308, + "grad_norm": 6.375, + "kd_loss": 0.1318359375, + "learning_rate": 9.183033205794525e-06, + "loss": 0.2094, + "step": 210, + "student_loss": 0.13332298398017883, + "teacher_loss": 0.0007853205897845328 + }, + { + "epoch": 0.061995386389850055, + "grad_norm": 6.46875, + "kd_loss": 0.1259765625, + "learning_rate": 9.13813159470227e-06, + "loss": 0.2075, + "step": 215, + "student_loss": 0.020972760394215584, + "teacher_loss": 0.0003814305819105357 + }, + { + "epoch": 0.06343713956170703, + "grad_norm": 7.8125, + "kd_loss": 0.1357421875, + "learning_rate": 9.092144830318357e-06, + "loss": 0.2394, + "step": 220, + "student_loss": 0.3116561472415924, + "teacher_loss": 0.008577825501561165 + }, + { + "epoch": 0.06487889273356401, + "grad_norm": 6.125, + "kd_loss": 0.11572265625, + "learning_rate": 9.045084971874738e-06, + "loss": 0.2193, + "step": 225, + "student_loss": 0.17015616595745087, + "teacher_loss": 0.0012981987092643976 + }, + { + "epoch": 0.06632064590542099, + "grad_norm": 5.0625, + "kd_loss": 0.11962890625, + "learning_rate": 8.99696436000368e-06, + "loss": 0.1794, + "step": 230, + "student_loss": 0.00222015380859375, + "teacher_loss": 0.003056387882679701 + }, + { + "epoch": 0.06776239907727798, + "grad_norm": 4.9375, + "kd_loss": 0.15234375, + "learning_rate": 8.947795613501658e-06, + "loss": 0.2096, + "step": 235, + "student_loss": 0.005798778962343931, + "teacher_loss": 0.00040503445779904723 + }, + { + "epoch": 0.06920415224913495, + "grad_norm": 4.34375, + "kd_loss": 0.1103515625, + "learning_rate": 8.897591626020284e-06, + "loss": 0.2009, + "step": 240, + "student_loss": 0.0019766136538237333, + "teacher_loss": 0.0007733534439466894 + }, + { + "epoch": 0.07064590542099193, + "grad_norm": 4.875, + "kd_loss": 0.1376953125, + "learning_rate": 8.846365562685178e-06, + "loss": 0.1982, + "step": 245, + "student_loss": 0.002786125522106886, + "teacher_loss": 0.0018817130476236343 + }, + { + "epoch": 0.07208765859284891, + "grad_norm": 5.0625, + "kd_loss": 0.1162109375, + "learning_rate": 8.794130856643635e-06, + "loss": 0.1736, + "step": 250, + "student_loss": 0.003304118989035487, + "teacher_loss": 0.0007230375776998699 + }, + { + "epoch": 0.07352941176470588, + "grad_norm": 6.875, + "kd_loss": 0.12890625, + "learning_rate": 8.74090120554202e-06, + "loss": 0.2065, + "step": 255, + "student_loss": 0.0011023435508832335, + "teacher_loss": 0.0011509901378303766 + }, + { + "epoch": 0.07497116493656286, + "grad_norm": 4.9375, + "kd_loss": 0.1259765625, + "learning_rate": 8.686690567933803e-06, + "loss": 0.2005, + "step": 260, + "student_loss": 0.001012590597383678, + "teacher_loss": 0.035696372389793396 + }, + { + "epoch": 0.07641291810841984, + "grad_norm": 4.8125, + "kd_loss": 0.11669921875, + "learning_rate": 8.63151315961915e-06, + "loss": 0.1962, + "step": 265, + "student_loss": 0.001487129949964583, + "teacher_loss": 0.0009277480421587825 + }, + { + "epoch": 0.07785467128027682, + "grad_norm": 4.84375, + "kd_loss": 0.1298828125, + "learning_rate": 8.575383449917103e-06, + "loss": 0.196, + "step": 270, + "student_loss": 0.016224455088377, + "teacher_loss": 0.0017208521021530032 + }, + { + "epoch": 0.07929642445213379, + "grad_norm": 6.5625, + "kd_loss": 0.12451171875, + "learning_rate": 8.518316157871232e-06, + "loss": 0.2031, + "step": 275, + "student_loss": 0.0010892748832702637, + "teacher_loss": 0.035929761826992035 + }, + { + "epoch": 0.08073817762399077, + "grad_norm": 4.90625, + "kd_loss": 0.1279296875, + "learning_rate": 8.460326248389825e-06, + "loss": 0.217, + "step": 280, + "student_loss": 0.0011528143659234047, + "teacher_loss": 0.0003817000542767346 + }, + { + "epoch": 0.08217993079584775, + "grad_norm": 6.625, + "kd_loss": 0.11572265625, + "learning_rate": 8.401428928321607e-06, + "loss": 0.206, + "step": 285, + "student_loss": 0.0025788608472794294, + "teacher_loss": 0.001419232808984816 + }, + { + "epoch": 0.08362168396770472, + "grad_norm": 4.59375, + "kd_loss": 0.13671875, + "learning_rate": 8.341639642468002e-06, + "loss": 0.2497, + "step": 290, + "student_loss": 0.012355645187199116, + "teacher_loss": 0.004034335725009441 + }, + { + "epoch": 0.0850634371395617, + "grad_norm": 6.875, + "kd_loss": 0.1259765625, + "learning_rate": 8.280974069532999e-06, + "loss": 0.1999, + "step": 295, + "student_loss": 0.0029606728348881006, + "teacher_loss": 0.0018796900985762477 + }, + { + "epoch": 0.08650519031141868, + "grad_norm": 7.40625, + "kd_loss": 0.12451171875, + "learning_rate": 8.219448118011687e-06, + "loss": 0.1898, + "step": 300, + "student_loss": 0.05260760709643364, + "teacher_loss": 0.0028262475971132517 + }, + { + "epoch": 0.08794694348327567, + "grad_norm": 5.75, + "kd_loss": 0.11474609375, + "learning_rate": 8.157077922018537e-06, + "loss": 0.1993, + "step": 305, + "student_loss": 0.07766762375831604, + "teacher_loss": 0.0020029095467180014 + }, + { + "epoch": 0.08938869665513265, + "grad_norm": 3.796875, + "kd_loss": 0.10595703125, + "learning_rate": 8.093879837056486e-06, + "loss": 0.1971, + "step": 310, + "student_loss": 0.0014958757674321532, + "teacher_loss": 0.0014671633252874017 + }, + { + "epoch": 0.09083044982698962, + "grad_norm": 7.15625, + "kd_loss": 0.1103515625, + "learning_rate": 8.029870435728018e-06, + "loss": 0.204, + "step": 315, + "student_loss": 0.2129756659269333, + "teacher_loss": 0.0017302327323704958 + }, + { + "epoch": 0.0922722029988466, + "grad_norm": 5.0, + "kd_loss": 0.10400390625, + "learning_rate": 7.965066503389264e-06, + "loss": 0.2075, + "step": 320, + "student_loss": 0.0016070405254140496, + "teacher_loss": 0.0009597347816452384 + }, + { + "epoch": 0.09371395617070358, + "grad_norm": 6.59375, + "kd_loss": 0.171875, + "learning_rate": 7.89948503374835e-06, + "loss": 0.1917, + "step": 325, + "student_loss": 0.003808736801147461, + "teacher_loss": 0.01614256761968136 + }, + { + "epoch": 0.09515570934256055, + "grad_norm": 6.03125, + "kd_loss": 0.115234375, + "learning_rate": 7.833143224409076e-06, + "loss": 0.2017, + "step": 330, + "student_loss": 0.021043118089437485, + "teacher_loss": 0.0004570994933601469 + }, + { + "epoch": 0.09659746251441753, + "grad_norm": 4.21875, + "kd_loss": 0.12109375, + "learning_rate": 7.766058472361154e-06, + "loss": 0.176, + "step": 335, + "student_loss": 0.0018800155958160758, + "teacher_loss": 0.0022506280802190304 + }, + { + "epoch": 0.09803921568627451, + "grad_norm": 5.03125, + "kd_loss": 0.1083984375, + "learning_rate": 7.698248369418146e-06, + "loss": 0.1834, + "step": 340, + "student_loss": 0.0805514007806778, + "teacher_loss": 0.006284959614276886 + }, + { + "epoch": 0.09948096885813149, + "grad_norm": 5.6875, + "kd_loss": 0.10888671875, + "learning_rate": 7.629730697604314e-06, + "loss": 0.2074, + "step": 345, + "student_loss": 0.13202711939811707, + "teacher_loss": 0.0005228903028182685 + }, + { + "epoch": 0.10092272202998846, + "grad_norm": 5.25, + "kd_loss": 0.12890625, + "learning_rate": 7.560523424491595e-06, + "loss": 0.1831, + "step": 350, + "student_loss": 0.081477090716362, + "teacher_loss": 0.0003300213429611176 + }, + { + "epoch": 0.10236447520184544, + "grad_norm": 3.796875, + "kd_loss": 0.10693359375, + "learning_rate": 7.490644698487909e-06, + "loss": 0.1853, + "step": 355, + "student_loss": 0.002148553030565381, + "teacher_loss": 0.0014131986536085606 + }, + { + "epoch": 0.10380622837370242, + "grad_norm": 5.65625, + "kd_loss": 0.111328125, + "learning_rate": 7.420112844078066e-06, + "loss": 0.1865, + "step": 360, + "student_loss": 0.18958300352096558, + "teacher_loss": 0.001118413987569511 + }, + { + "epoch": 0.1052479815455594, + "grad_norm": 7.625, + "kd_loss": 0.1123046875, + "learning_rate": 7.348946357018479e-06, + "loss": 0.1824, + "step": 365, + "student_loss": 0.0013260500272735953, + "teacher_loss": 0.029322339221835136 + }, + { + "epoch": 0.10668973471741638, + "grad_norm": 4.875, + "kd_loss": 0.1318359375, + "learning_rate": 7.277163899486975e-06, + "loss": 0.189, + "step": 370, + "student_loss": 0.16300825774669647, + "teacher_loss": 0.000458209979115054 + }, + { + "epoch": 0.10813148788927336, + "grad_norm": 5.875, + "kd_loss": 0.15625, + "learning_rate": 7.204784295188959e-06, + "loss": 0.1865, + "step": 375, + "student_loss": 0.3203120529651642, + "teacher_loss": 0.02350226417183876 + }, + { + "epoch": 0.10957324106113034, + "grad_norm": 4.34375, + "kd_loss": 0.10546875, + "learning_rate": 7.1318265244212305e-06, + "loss": 0.1864, + "step": 380, + "student_loss": 0.0027754041366279125, + "teacher_loss": 0.001105214236304164 + }, + { + "epoch": 0.11101499423298732, + "grad_norm": 4.8125, + "kd_loss": 0.1171875, + "learning_rate": 7.05830971909472e-06, + "loss": 0.1872, + "step": 385, + "student_loss": 0.001283544348552823, + "teacher_loss": 0.0009954526321962476 + }, + { + "epoch": 0.11245674740484429, + "grad_norm": 5.25, + "kd_loss": 0.125, + "learning_rate": 6.9842531577174865e-06, + "loss": 0.1764, + "step": 390, + "student_loss": 0.0008525378652848303, + "teacher_loss": 0.001041764859110117 + }, + { + "epoch": 0.11389850057670127, + "grad_norm": 6.59375, + "kd_loss": 0.1171875, + "learning_rate": 6.9096762603392595e-06, + "loss": 0.195, + "step": 395, + "student_loss": 0.0015218615299090743, + "teacher_loss": 0.0010145456762984395 + }, + { + "epoch": 0.11534025374855825, + "grad_norm": 4.6875, + "kd_loss": 0.12451171875, + "learning_rate": 6.834598583458862e-06, + "loss": 0.1822, + "step": 400, + "student_loss": 0.0013333633542060852, + "teacher_loss": 0.00030270780553109944 + }, + { + "epoch": 0.11678200692041522, + "grad_norm": 2.703125, + "kd_loss": 0.103515625, + "learning_rate": 6.7590398148958625e-06, + "loss": 0.196, + "step": 405, + "student_loss": 0.026790756732225418, + "teacher_loss": 0.0005501789273694158 + }, + { + "epoch": 0.1182237600922722, + "grad_norm": 8.125, + "kd_loss": 0.1787109375, + "learning_rate": 6.6830197686277945e-06, + "loss": 0.2152, + "step": 410, + "student_loss": 0.5714533925056458, + "teacher_loss": 0.008862318471074104 + }, + { + "epoch": 0.11966551326412918, + "grad_norm": 4.1875, + "kd_loss": 0.1279296875, + "learning_rate": 6.6065583795942625e-06, + "loss": 0.2006, + "step": 415, + "student_loss": 0.019465278834104538, + "teacher_loss": 0.0042519038543105125 + }, + { + "epoch": 0.12110726643598616, + "grad_norm": 5.21875, + "kd_loss": 0.1181640625, + "learning_rate": 6.52967569846937e-06, + "loss": 0.1764, + "step": 420, + "student_loss": 0.008550797589123249, + "teacher_loss": 0.023795029148459435 + }, + { + "epoch": 0.12254901960784313, + "grad_norm": 8.5625, + "kd_loss": 0.126953125, + "learning_rate": 6.452391886403767e-06, + "loss": 0.1854, + "step": 425, + "student_loss": 0.04949700087308884, + "teacher_loss": 0.004153509624302387 + }, + { + "epoch": 0.12399077277970011, + "grad_norm": 7.25, + "kd_loss": 0.18359375, + "learning_rate": 6.374727209737743e-06, + "loss": 0.2107, + "step": 430, + "student_loss": 0.002016805112361908, + "teacher_loss": 0.06358348578214645 + }, + { + "epoch": 0.1254325259515571, + "grad_norm": 6.21875, + "kd_loss": 0.11083984375, + "learning_rate": 6.296702034686726e-06, + "loss": 0.1934, + "step": 435, + "student_loss": 0.0017790297279134393, + "teacher_loss": 0.0017155191162601113 + }, + { + "epoch": 0.12687427912341406, + "grad_norm": 4.125, + "kd_loss": 0.18359375, + "learning_rate": 6.218336822000598e-06, + "loss": 0.2068, + "step": 440, + "student_loss": 0.565355658531189, + "teacher_loss": 0.008367877453565598 + }, + { + "epoch": 0.12831603229527105, + "grad_norm": 5.09375, + "kd_loss": 0.1357421875, + "learning_rate": 6.139652121598219e-06, + "loss": 0.2072, + "step": 445, + "student_loss": 0.00040582873043604195, + "teacher_loss": 0.017554111778736115 + }, + { + "epoch": 0.12975778546712802, + "grad_norm": 4.03125, + "kd_loss": 0.1396484375, + "learning_rate": 6.060668567178561e-06, + "loss": 0.194, + "step": 450, + "student_loss": 0.0020873546600341797, + "teacher_loss": 0.0007538718055002391 + }, + { + "epoch": 0.131199538638985, + "grad_norm": 4.09375, + "kd_loss": 0.10595703125, + "learning_rate": 5.981406870809889e-06, + "loss": 0.1896, + "step": 455, + "student_loss": 0.010908172465860844, + "teacher_loss": 0.0012792785419151187 + }, + { + "epoch": 0.13264129181084197, + "grad_norm": 6.6875, + "kd_loss": 0.12255859375, + "learning_rate": 5.9018878174983674e-06, + "loss": 0.1893, + "step": 460, + "student_loss": 0.01578596420586109, + "teacher_loss": 0.0009114979766309261 + }, + { + "epoch": 0.13408304498269896, + "grad_norm": 4.4375, + "kd_loss": 0.14453125, + "learning_rate": 5.822132259737565e-06, + "loss": 0.2189, + "step": 465, + "student_loss": 0.0021727595012634993, + "teacher_loss": 0.0004909814451821148 + }, + { + "epoch": 0.13552479815455595, + "grad_norm": 5.09375, + "kd_loss": 0.1005859375, + "learning_rate": 5.742161112040237e-06, + "loss": 0.2169, + "step": 470, + "student_loss": 0.0009243786334991455, + "teacher_loss": 0.000744891760405153 + }, + { + "epoch": 0.13696655132641292, + "grad_norm": 4.9375, + "kd_loss": 0.1396484375, + "learning_rate": 5.661995345453867e-06, + "loss": 0.1752, + "step": 475, + "student_loss": 0.003345559583976865, + "teacher_loss": 0.0005117281689308584 + }, + { + "epoch": 0.1384083044982699, + "grad_norm": 4.46875, + "kd_loss": 0.1376953125, + "learning_rate": 5.581655982061367e-06, + "loss": 0.211, + "step": 480, + "student_loss": 0.0018078088760375977, + "teacher_loss": 0.029633358120918274 + }, + { + "epoch": 0.13985005767012687, + "grad_norm": 6.59375, + "kd_loss": 0.11767578125, + "learning_rate": 5.501164089468406e-06, + "loss": 0.1795, + "step": 485, + "student_loss": 0.3106631636619568, + "teacher_loss": 0.0037072307895869017 + }, + { + "epoch": 0.14129181084198386, + "grad_norm": 6.0, + "kd_loss": 0.1328125, + "learning_rate": 5.4205407752787884e-06, + "loss": 0.1896, + "step": 490, + "student_loss": 0.0010533903259783983, + "teacher_loss": 0.0012765543069690466 + }, + { + "epoch": 0.14273356401384082, + "grad_norm": 5.25, + "kd_loss": 0.1376953125, + "learning_rate": 5.339807181559359e-06, + "loss": 0.194, + "step": 495, + "student_loss": 0.08354002982378006, + "teacher_loss": 0.0006759578245691955 + }, + { + "epoch": 0.14417531718569782, + "grad_norm": 6.125, + "kd_loss": 0.12060546875, + "learning_rate": 5.258984479295853e-06, + "loss": 0.1865, + "step": 500, + "student_loss": 0.003352736122906208, + "teacher_loss": 0.0017620434518903494 + }, + { + "epoch": 0.14561707035755478, + "grad_norm": 5.71875, + "kd_loss": 0.11572265625, + "learning_rate": 5.1780938628411795e-06, + "loss": 0.2201, + "step": 505, + "student_loss": 0.002489902079105377, + "teacher_loss": 0.0007855825824663043 + }, + { + "epoch": 0.14705882352941177, + "grad_norm": 5.59375, + "kd_loss": 0.10498046875, + "learning_rate": 5.097156544357567e-06, + "loss": 0.2023, + "step": 510, + "student_loss": 0.0014551215572282672, + "teacher_loss": 0.0008335533202625811 + }, + { + "epoch": 0.14850057670126873, + "grad_norm": 4.75, + "kd_loss": 0.12353515625, + "learning_rate": 5.016193748254045e-06, + "loss": 0.1779, + "step": 515, + "student_loss": 0.01823529414832592, + "teacher_loss": 0.0005908762104809284 + }, + { + "epoch": 0.14994232987312572, + "grad_norm": 3.875, + "kd_loss": 0.1181640625, + "learning_rate": 4.935226705620699e-06, + "loss": 0.1875, + "step": 520, + "student_loss": 0.48260822892189026, + "teacher_loss": 0.011817601509392262 + }, + { + "epoch": 0.1513840830449827, + "grad_norm": 3.828125, + "kd_loss": 0.1142578125, + "learning_rate": 4.8542766486612035e-06, + "loss": 0.179, + "step": 525, + "student_loss": 0.0011587169719859958, + "teacher_loss": 0.0004872040299233049 + }, + { + "epoch": 0.15282583621683968, + "grad_norm": 5.78125, + "kd_loss": 0.11767578125, + "learning_rate": 4.773364805125025e-06, + "loss": 0.1752, + "step": 530, + "student_loss": 0.0030523419845849276, + "teacher_loss": 0.0013172540348023176 + }, + { + "epoch": 0.15426758938869667, + "grad_norm": 3.1875, + "kd_loss": 0.10302734375, + "learning_rate": 4.6925123927408265e-06, + "loss": 0.1654, + "step": 535, + "student_loss": 0.0017982972785830498, + "teacher_loss": 0.00047424182412214577 + }, + { + "epoch": 0.15570934256055363, + "grad_norm": 5.59375, + "kd_loss": 0.11962890625, + "learning_rate": 4.611740613652485e-06, + "loss": 0.1655, + "step": 540, + "student_loss": 0.013529052957892418, + "teacher_loss": 0.0009085286292247474 + }, + { + "epoch": 0.15715109573241062, + "grad_norm": 9.0, + "kd_loss": 0.115234375, + "learning_rate": 4.531070648859186e-06, + "loss": 0.1973, + "step": 545, + "student_loss": 0.004623454995453358, + "teacher_loss": 0.007325619924813509 + }, + { + "epoch": 0.15859284890426759, + "grad_norm": 5.8125, + "kd_loss": 0.126953125, + "learning_rate": 4.450523652661086e-06, + "loss": 0.1622, + "step": 550, + "student_loss": 0.0009506479254923761, + "teacher_loss": 0.008248833939433098 + }, + { + "epoch": 0.16003460207612458, + "grad_norm": 4.34375, + "kd_loss": 0.11376953125, + "learning_rate": 4.370120747111956e-06, + "loss": 0.1848, + "step": 555, + "student_loss": 0.005332108587026596, + "teacher_loss": 0.0016086830291897058 + }, + { + "epoch": 0.16147635524798154, + "grad_norm": 10.375, + "kd_loss": 0.1201171875, + "learning_rate": 4.289883016480291e-06, + "loss": 0.2032, + "step": 560, + "student_loss": 0.12518270313739777, + "teacher_loss": 0.0005838426877744496 + }, + { + "epoch": 0.16291810841983853, + "grad_norm": 6.0625, + "kd_loss": 0.12353515625, + "learning_rate": 4.209831501720328e-06, + "loss": 0.1825, + "step": 565, + "student_loss": 0.029691526666283607, + "teacher_loss": 0.021172240376472473 + }, + { + "epoch": 0.1643598615916955, + "grad_norm": 8.9375, + "kd_loss": 0.1708984375, + "learning_rate": 4.129987194954421e-06, + "loss": 0.189, + "step": 570, + "student_loss": 0.13193272054195404, + "teacher_loss": 0.00722926901653409 + }, + { + "epoch": 0.16580161476355249, + "grad_norm": 7.84375, + "kd_loss": 0.10302734375, + "learning_rate": 4.050371033968216e-06, + "loss": 0.1851, + "step": 575, + "student_loss": 0.0010820090537890792, + "teacher_loss": 0.0006335995858535171 + }, + { + "epoch": 0.16724336793540945, + "grad_norm": 4.96875, + "kd_loss": 0.330078125, + "learning_rate": 3.9710038967200825e-06, + "loss": 0.1666, + "step": 580, + "student_loss": 0.0031647607684135437, + "teacher_loss": 0.0028759294655174017 + }, + { + "epoch": 0.16868512110726644, + "grad_norm": 3.1875, + "kd_loss": 0.1142578125, + "learning_rate": 3.89190659586623e-06, + "loss": 0.1868, + "step": 585, + "student_loss": 0.040488943457603455, + "teacher_loss": 0.0005194384139031172 + }, + { + "epoch": 0.1701268742791234, + "grad_norm": 7.5625, + "kd_loss": 0.11181640625, + "learning_rate": 3.8130998733029517e-06, + "loss": 0.1949, + "step": 590, + "student_loss": 0.00226792530156672, + "teacher_loss": 0.0028023580089211464 + }, + { + "epoch": 0.1715686274509804, + "grad_norm": 4.0625, + "kd_loss": 0.11181640625, + "learning_rate": 3.734604394727419e-06, + "loss": 0.2049, + "step": 595, + "student_loss": 0.0009982635965570807, + "teacher_loss": 0.001097838394343853 + }, + { + "epoch": 0.17301038062283736, + "grad_norm": 6.5625, + "kd_loss": 0.12353515625, + "learning_rate": 3.656440744218464e-06, + "loss": 0.1982, + "step": 600, + "student_loss": 0.34091848134994507, + "teacher_loss": 0.009622580371797085 + }, + { + "epoch": 0.17445213379469435, + "grad_norm": 6.625, + "kd_loss": 0.1181640625, + "learning_rate": 3.578629418838757e-06, + "loss": 0.1972, + "step": 605, + "student_loss": 0.2827480435371399, + "teacher_loss": 0.039488162845373154 + }, + { + "epoch": 0.17589388696655134, + "grad_norm": 4.96875, + "kd_loss": 0.1201171875, + "learning_rate": 3.5011908232598124e-06, + "loss": 0.1603, + "step": 610, + "student_loss": 0.106364406645298, + "teacher_loss": 0.0008834014879539609 + }, + { + "epoch": 0.1773356401384083, + "grad_norm": 4.75, + "kd_loss": 0.1240234375, + "learning_rate": 3.4241452644112085e-06, + "loss": 0.1596, + "step": 615, + "student_loss": 0.0008959124679677188, + "teacher_loss": 0.0007150223245844245 + }, + { + "epoch": 0.1787773933102653, + "grad_norm": 3.34375, + "kd_loss": 0.1533203125, + "learning_rate": 3.3475129461554567e-06, + "loss": 0.1941, + "step": 620, + "student_loss": 0.00717555359005928, + "teacher_loss": 0.008744844235479832 + }, + { + "epoch": 0.18021914648212226, + "grad_norm": 4.46875, + "kd_loss": 0.109375, + "learning_rate": 3.271313963989886e-06, + "loss": 0.1711, + "step": 625, + "student_loss": 0.007324306294322014, + "teacher_loss": 0.005477549973875284 + }, + { + "epoch": 0.18166089965397925, + "grad_norm": 4.875, + "kd_loss": 0.11962890625, + "learning_rate": 3.195568299776945e-06, + "loss": 0.1813, + "step": 630, + "student_loss": 0.1220104992389679, + "teacher_loss": 0.005055113695561886 + }, + { + "epoch": 0.1831026528258362, + "grad_norm": 4.1875, + "kd_loss": 0.1005859375, + "learning_rate": 3.1202958165043053e-06, + "loss": 0.2012, + "step": 635, + "student_loss": 0.0011498430976644158, + "teacher_loss": 0.0006288467557169497 + }, + { + "epoch": 0.1845444059976932, + "grad_norm": 4.84375, + "kd_loss": 0.11767578125, + "learning_rate": 3.045516253076137e-06, + "loss": 0.1779, + "step": 640, + "student_loss": 0.0011653146939352155, + "teacher_loss": 0.0009915747214108706 + }, + { + "epoch": 0.18598615916955016, + "grad_norm": 9.125, + "kd_loss": 0.119140625, + "learning_rate": 2.9712492191369245e-06, + "loss": 0.1795, + "step": 645, + "student_loss": 0.004314988851547241, + "teacher_loss": 0.0008632438839413226 + }, + { + "epoch": 0.18742791234140715, + "grad_norm": 5.71875, + "kd_loss": 0.138671875, + "learning_rate": 2.8975141899291777e-06, + "loss": 0.1767, + "step": 650, + "student_loss": 0.004328088369220495, + "teacher_loss": 0.0019480936462059617 + }, + { + "epoch": 0.18886966551326412, + "grad_norm": 5.3125, + "kd_loss": 0.1064453125, + "learning_rate": 2.8243305011863843e-06, + "loss": 0.1858, + "step": 655, + "student_loss": 0.07007281482219696, + "teacher_loss": 0.002063432242721319 + }, + { + "epoch": 0.1903114186851211, + "grad_norm": 5.34375, + "kd_loss": 0.12060546875, + "learning_rate": 2.751717344062552e-06, + "loss": 0.1979, + "step": 660, + "student_loss": 0.0020055994391441345, + "teacher_loss": 0.0012256632326170802 + }, + { + "epoch": 0.19175317185697807, + "grad_norm": 4.96875, + "kd_loss": 0.1259765625, + "learning_rate": 2.6796937600996587e-06, + "loss": 0.1824, + "step": 665, + "student_loss": 0.0013414303539320827, + "teacher_loss": 0.0005829873844049871 + }, + { + "epoch": 0.19319492502883506, + "grad_norm": 6.21875, + "kd_loss": 0.11181640625, + "learning_rate": 2.6082786362343377e-06, + "loss": 0.2091, + "step": 670, + "student_loss": 0.01750928722321987, + "teacher_loss": 0.01849350705742836 + }, + { + "epoch": 0.19463667820069205, + "grad_norm": 5.9375, + "kd_loss": 0.1083984375, + "learning_rate": 2.5374906998451094e-06, + "loss": 0.1855, + "step": 675, + "student_loss": 0.0015071257948875427, + "teacher_loss": 0.0012101252796128392 + }, + { + "epoch": 0.19607843137254902, + "grad_norm": 3.984375, + "kd_loss": 0.14453125, + "learning_rate": 2.467348513841447e-06, + "loss": 0.1808, + "step": 680, + "student_loss": 0.09338736534118652, + "teacher_loss": 0.038049884140491486 + }, + { + "epoch": 0.197520184544406, + "grad_norm": 3.75, + "kd_loss": 0.177734375, + "learning_rate": 2.3978704717959777e-06, + "loss": 0.1863, + "step": 685, + "student_loss": 0.0010167881846427917, + "teacher_loss": 0.03185700252652168 + }, + { + "epoch": 0.19896193771626297, + "grad_norm": 3.984375, + "kd_loss": 0.11572265625, + "learning_rate": 2.329074793121085e-06, + "loss": 0.1721, + "step": 690, + "student_loss": 0.017529672011733055, + "teacher_loss": 0.008856060914695263 + }, + { + "epoch": 0.20040369088811996, + "grad_norm": 6.65625, + "kd_loss": 0.1171875, + "learning_rate": 2.260979518291186e-06, + "loss": 0.2122, + "step": 695, + "student_loss": 0.028023820370435715, + "teacher_loss": 0.005092882085591555 + }, + { + "epoch": 0.20184544405997693, + "grad_norm": 4.84375, + "kd_loss": 0.1015625, + "learning_rate": 2.1936025041119268e-06, + "loss": 0.1889, + "step": 700, + "student_loss": 0.004128373693674803, + "teacher_loss": 0.0009403788135387003 + }, + { + "epoch": 0.20328719723183392, + "grad_norm": 5.03125, + "kd_loss": 0.12451171875, + "learning_rate": 2.1269614190375477e-06, + "loss": 0.1781, + "step": 705, + "student_loss": 0.0009666193509474397, + "teacher_loss": 0.0012199536431580782 + }, + { + "epoch": 0.20472895040369088, + "grad_norm": 5.46875, + "kd_loss": 0.12890625, + "learning_rate": 2.061073738537635e-06, + "loss": 0.214, + "step": 710, + "student_loss": 0.20427227020263672, + "teacher_loss": 0.003926330246031284 + }, + { + "epoch": 0.20617070357554787, + "grad_norm": 4.875, + "kd_loss": 0.10693359375, + "learning_rate": 1.9959567405144825e-06, + "loss": 0.2167, + "step": 715, + "student_loss": 0.034982144832611084, + "teacher_loss": 0.012018893845379353 + }, + { + "epoch": 0.20761245674740483, + "grad_norm": 4.90625, + "kd_loss": 0.140625, + "learning_rate": 1.931627500772263e-06, + "loss": 0.1911, + "step": 720, + "student_loss": 0.0010867691598832607, + "teacher_loss": 0.005945001263171434 + }, + { + "epoch": 0.20905420991926182, + "grad_norm": 3.375, + "kd_loss": 0.1171875, + "learning_rate": 1.8681028885391905e-06, + "loss": 0.1776, + "step": 725, + "student_loss": 0.0011550028575584292, + "teacher_loss": 0.0008522791904397309 + }, + { + "epoch": 0.2104959630911188, + "grad_norm": 4.625, + "kd_loss": 0.111328125, + "learning_rate": 1.8053995620438625e-06, + "loss": 0.1726, + "step": 730, + "student_loss": 0.03152952715754509, + "teacher_loss": 0.00232282024808228 + }, + { + "epoch": 0.21193771626297578, + "grad_norm": 4.8125, + "kd_loss": 0.11328125, + "learning_rate": 1.743533964146924e-06, + "loss": 0.1841, + "step": 735, + "student_loss": 0.0014305273070931435, + "teacher_loss": 0.0003280507226008922 + }, + { + "epoch": 0.21337946943483277, + "grad_norm": 6.9375, + "kd_loss": 0.138671875, + "learning_rate": 1.6825223180292138e-06, + "loss": 0.1674, + "step": 740, + "student_loss": 0.00881188828498125, + "teacher_loss": 0.00042216398287564516 + }, + { + "epoch": 0.21482122260668973, + "grad_norm": 6.71875, + "kd_loss": 0.12158203125, + "learning_rate": 1.6223806229375182e-06, + "loss": 0.1744, + "step": 745, + "student_loss": 0.16076479852199554, + "teacher_loss": 0.021096019074320793 + }, + { + "epoch": 0.21626297577854672, + "grad_norm": 6.28125, + "kd_loss": 0.12890625, + "learning_rate": 1.563124649989043e-06, + "loss": 0.1968, + "step": 750, + "student_loss": 0.11668777465820312, + "teacher_loss": 0.006358537822961807 + }, + { + "epoch": 0.2177047289504037, + "grad_norm": 5.4375, + "kd_loss": 0.0966796875, + "learning_rate": 1.5047699380357134e-06, + "loss": 0.1956, + "step": 755, + "student_loss": 0.11215528845787048, + "teacher_loss": 0.0056000882759690285 + }, + { + "epoch": 0.21914648212226068, + "grad_norm": 5.125, + "kd_loss": 0.10107421875, + "learning_rate": 1.4473317895893773e-06, + "loss": 0.1792, + "step": 760, + "student_loss": 0.26354143023490906, + "teacher_loss": 0.0006397212855517864 + }, + { + "epoch": 0.22058823529411764, + "grad_norm": 4.53125, + "kd_loss": 0.12060546875, + "learning_rate": 1.39082526680899e-06, + "loss": 0.1971, + "step": 765, + "student_loss": 0.08930032700300217, + "teacher_loss": 0.0006192077999003232 + }, + { + "epoch": 0.22202998846597463, + "grad_norm": 3.6875, + "kd_loss": 0.1142578125, + "learning_rate": 1.3352651875508204e-06, + "loss": 0.1708, + "step": 770, + "student_loss": 0.004422426223754883, + "teacher_loss": 0.0008745273225940764 + }, + { + "epoch": 0.2234717416378316, + "grad_norm": 5.4375, + "kd_loss": 0.11328125, + "learning_rate": 1.2806661214827286e-06, + "loss": 0.1885, + "step": 775, + "student_loss": 0.0018669217824935913, + "teacher_loss": 0.0006652303854934871 + }, + { + "epoch": 0.22491349480968859, + "grad_norm": 2.9375, + "kd_loss": 0.1083984375, + "learning_rate": 1.2270423862635188e-06, + "loss": 0.1836, + "step": 780, + "student_loss": 0.009803751483559608, + "teacher_loss": 0.0004910013522021472 + }, + { + "epoch": 0.22635524798154555, + "grad_norm": 5.6875, + "kd_loss": 0.09912109375, + "learning_rate": 1.1744080437883859e-06, + "loss": 0.1669, + "step": 785, + "student_loss": 0.0010796686401590705, + "teacher_loss": 0.001094150822609663 + }, + { + "epoch": 0.22779700115340254, + "grad_norm": 6.0625, + "kd_loss": 0.1279296875, + "learning_rate": 1.1227768965014246e-06, + "loss": 0.2026, + "step": 790, + "student_loss": 0.07495569437742233, + "teacher_loss": 0.0015408035833388567 + }, + { + "epoch": 0.2292387543252595, + "grad_norm": 3.390625, + "kd_loss": 0.1083984375, + "learning_rate": 1.0721624837761768e-06, + "loss": 0.1999, + "step": 795, + "student_loss": 0.0036292201839387417, + "teacher_loss": 0.0005642004543915391 + }, + { + "epoch": 0.2306805074971165, + "grad_norm": 6.5625, + "kd_loss": 0.18359375, + "learning_rate": 1.0225780783651689e-06, + "loss": 0.2151, + "step": 800, + "student_loss": 0.062444765120744705, + "teacher_loss": 0.04929126426577568 + }, + { + "epoch": 0.23212226066897348, + "grad_norm": 4.875, + "kd_loss": 0.10546875, + "learning_rate": 9.740366829193587e-07, + "loss": 0.2096, + "step": 805, + "student_loss": 0.0012999593745917082, + "teacher_loss": 0.001006675767712295 + }, + { + "epoch": 0.23356401384083045, + "grad_norm": 6.09375, + "kd_loss": 0.109375, + "learning_rate": 9.265510265784189e-07, + "loss": 0.2063, + "step": 810, + "student_loss": 0.0013730658683925867, + "teacher_loss": 0.00053932867012918 + }, + { + "epoch": 0.23500576701268744, + "grad_norm": 10.0, + "kd_loss": 0.103515625, + "learning_rate": 8.801335616327378e-07, + "loss": 0.1942, + "step": 815, + "student_loss": 0.05159832164645195, + "teacher_loss": 0.010285490192472935 + }, + { + "epoch": 0.2364475201845444, + "grad_norm": 4.9375, + "kd_loss": 0.125, + "learning_rate": 8.347964602580245e-07, + "loss": 0.1808, + "step": 820, + "student_loss": 0.037393856793642044, + "teacher_loss": 0.0004633679345715791 + }, + { + "epoch": 0.2378892733564014, + "grad_norm": 6.5625, + "kd_loss": 0.142578125, + "learning_rate": 7.905516113233652e-07, + "loss": 0.1747, + "step": 825, + "student_loss": 0.0011921566911041737, + "teacher_loss": 0.021351948380470276 + }, + { + "epoch": 0.23933102652825836, + "grad_norm": 3.484375, + "kd_loss": 0.111328125, + "learning_rate": 7.474106172735746e-07, + "loss": 0.1797, + "step": 830, + "student_loss": 0.03779162839055061, + "teacher_loss": 0.003403074573725462 + }, + { + "epoch": 0.24077277970011535, + "grad_norm": 6.03125, + "kd_loss": 0.10009765625, + "learning_rate": 7.053847910866513e-07, + "loss": 0.1667, + "step": 835, + "student_loss": 0.11626744270324707, + "teacher_loss": 0.00203131721355021 + }, + { + "epoch": 0.2422145328719723, + "grad_norm": 4.28125, + "kd_loss": 0.1376953125, + "learning_rate": 6.644851533071556e-07, + "loss": 0.1761, + "step": 840, + "student_loss": 0.0023884603288024664, + "teacher_loss": 0.0004405094077810645 + }, + { + "epoch": 0.2436562860438293, + "grad_norm": 8.6875, + "kd_loss": 0.140625, + "learning_rate": 6.24722429156251e-07, + "loss": 0.2435, + "step": 845, + "student_loss": 0.14598870277404785, + "teacher_loss": 0.0010793671244755387 + }, + { + "epoch": 0.24509803921568626, + "grad_norm": 4.6875, + "kd_loss": 0.103515625, + "learning_rate": 5.861070457192081e-07, + "loss": 0.186, + "step": 850, + "student_loss": 0.06827586144208908, + "teacher_loss": 0.00046239409130066633 + }, + { + "epoch": 0.24653979238754326, + "grad_norm": 3.203125, + "kd_loss": 0.125, + "learning_rate": 5.486491292110796e-07, + "loss": 0.1726, + "step": 855, + "student_loss": 0.0007081849034875631, + "teacher_loss": 0.0005193519755266607 + }, + { + "epoch": 0.24798154555940022, + "grad_norm": 4.40625, + "kd_loss": 0.11474609375, + "learning_rate": 5.123585023212785e-07, + "loss": 0.2129, + "step": 860, + "student_loss": 0.002149001695215702, + "teacher_loss": 0.0017558797262609005 + }, + { + "epoch": 0.2494232987312572, + "grad_norm": 3.921875, + "kd_loss": 0.111328125, + "learning_rate": 4.772446816377408e-07, + "loss": 0.1792, + "step": 865, + "student_loss": 0.0010927807306870818, + "teacher_loss": 0.0006015551625750959 + }, + { + "epoch": 0.2508650519031142, + "grad_norm": 4.0, + "kd_loss": 0.12890625, + "learning_rate": 4.4331687515137614e-07, + "loss": 0.1958, + "step": 870, + "student_loss": 0.04162781313061714, + "teacher_loss": 0.0015390625922009349 + }, + { + "epoch": 0.25230680507497116, + "grad_norm": 7.9375, + "kd_loss": 0.10888671875, + "learning_rate": 4.1058397984142405e-07, + "loss": 0.1771, + "step": 875, + "student_loss": 0.0009604351944290102, + "teacher_loss": 0.0005376276094466448 + }, + { + "epoch": 0.2537485582468281, + "grad_norm": 3.953125, + "kd_loss": 0.1162109375, + "learning_rate": 3.790545793423761e-07, + "loss": 0.1917, + "step": 880, + "student_loss": 0.0019418075680732727, + "teacher_loss": 0.0007360613089986145 + }, + { + "epoch": 0.25519031141868515, + "grad_norm": 3.25, + "kd_loss": 0.1435546875, + "learning_rate": 3.4873694169306915e-07, + "loss": 0.1832, + "step": 885, + "student_loss": 0.017352323979139328, + "teacher_loss": 0.05972852557897568 + }, + { + "epoch": 0.2566320645905421, + "grad_norm": 4.75, + "kd_loss": 0.111328125, + "learning_rate": 3.196390171685343e-07, + "loss": 0.1981, + "step": 890, + "student_loss": 0.0014260741882026196, + "teacher_loss": 0.0012508125510066748 + }, + { + "epoch": 0.25807381776239907, + "grad_norm": 4.25, + "kd_loss": 0.1435546875, + "learning_rate": 2.917684361951728e-07, + "loss": 0.1799, + "step": 895, + "student_loss": 0.014447882771492004, + "teacher_loss": 0.0008786572143435478 + }, + { + "epoch": 0.25951557093425603, + "grad_norm": 4.3125, + "kd_loss": 0.1240234375, + "learning_rate": 2.65132507349814e-07, + "loss": 0.2243, + "step": 900, + "student_loss": 0.001752070034854114, + "teacher_loss": 0.028893902897834778 + }, + { + "epoch": 0.26095732410611305, + "grad_norm": 3.890625, + "kd_loss": 0.10888671875, + "learning_rate": 2.397382154431621e-07, + "loss": 0.1707, + "step": 905, + "student_loss": 0.07611552625894547, + "teacher_loss": 0.0018923009047284722 + }, + { + "epoch": 0.26239907727797, + "grad_norm": 3.84375, + "kd_loss": 0.130859375, + "learning_rate": 2.1559221968815547e-07, + "loss": 0.1867, + "step": 910, + "student_loss": 0.0012913525570183992, + "teacher_loss": 0.0014879581285640597 + }, + { + "epoch": 0.263840830449827, + "grad_norm": 5.1875, + "kd_loss": 0.11669921875, + "learning_rate": 1.9270085195370048e-07, + "loss": 0.1647, + "step": 915, + "student_loss": 0.03545321896672249, + "teacher_loss": 0.0013070907443761826 + }, + { + "epoch": 0.26528258362168394, + "grad_norm": 3.390625, + "kd_loss": 0.1630859375, + "learning_rate": 1.7107011510424766e-07, + "loss": 0.1914, + "step": 920, + "student_loss": 0.013124704360961914, + "teacher_loss": 0.015305249951779842 + }, + { + "epoch": 0.26672433679354096, + "grad_norm": 3.84375, + "kd_loss": 0.1064453125, + "learning_rate": 1.5070568142564912e-07, + "loss": 0.1662, + "step": 925, + "student_loss": 0.0009416788816452026, + "teacher_loss": 0.0009266760898754001 + }, + { + "epoch": 0.2681660899653979, + "grad_norm": 5.90625, + "kd_loss": 0.11376953125, + "learning_rate": 1.3161289113769405e-07, + "loss": 0.1771, + "step": 930, + "student_loss": 0.04463067650794983, + "teacher_loss": 0.0009097782894968987 + }, + { + "epoch": 0.2696078431372549, + "grad_norm": 5.59375, + "kd_loss": 0.126953125, + "learning_rate": 1.1379675099373489e-07, + "loss": 0.1749, + "step": 935, + "student_loss": 0.02378019131720066, + "teacher_loss": 0.0022270630579441786 + }, + { + "epoch": 0.2710495963091119, + "grad_norm": 5.78125, + "kd_loss": 0.2138671875, + "learning_rate": 9.726193296774767e-08, + "loss": 0.1876, + "step": 940, + "student_loss": 0.002607885980978608, + "teacher_loss": 0.0031158654019236565 + }, + { + "epoch": 0.27249134948096887, + "grad_norm": 7.5625, + "kd_loss": 0.10595703125, + "learning_rate": 8.201277302919086e-08, + "loss": 0.1904, + "step": 945, + "student_loss": 0.07421658933162689, + "teacher_loss": 0.002074115676805377 + }, + { + "epoch": 0.27393310265282583, + "grad_norm": 3.203125, + "kd_loss": 0.13671875, + "learning_rate": 6.805327000596995e-08, + "loss": 0.17, + "step": 950, + "student_loss": 0.005115017760545015, + "teacher_loss": 0.0004907959373667836 + }, + { + "epoch": 0.2753748558246828, + "grad_norm": 4.5625, + "kd_loss": 0.095703125, + "learning_rate": 5.538708453581787e-08, + "loss": 0.1903, + "step": 955, + "student_loss": 0.011558901518583298, + "teacher_loss": 0.0004379775491543114 + }, + { + "epoch": 0.2768166089965398, + "grad_norm": 7.9375, + "kd_loss": 0.12451171875, + "learning_rate": 4.40175381063529e-08, + "loss": 0.1861, + "step": 960, + "student_loss": 0.0017829686403274536, + "teacher_loss": 0.0036916364915668964 + }, + { + "epoch": 0.2782583621683968, + "grad_norm": 4.96875, + "kd_loss": 0.1728515625, + "learning_rate": 3.394761218407705e-08, + "loss": 0.2026, + "step": 965, + "student_loss": 0.2359560877084732, + "teacher_loss": 0.008938993327319622 + }, + { + "epoch": 0.27970011534025374, + "grad_norm": 4.375, + "kd_loss": 0.1123046875, + "learning_rate": 2.5179947432540376e-08, + "loss": 0.1889, + "step": 970, + "student_loss": 0.0006114219431765378, + "teacher_loss": 0.00041617779061198235 + }, + { + "epoch": 0.2811418685121107, + "grad_norm": 4.5, + "kd_loss": 0.10888671875, + "learning_rate": 1.7716843019867646e-08, + "loss": 0.1982, + "step": 975, + "student_loss": 0.07514014840126038, + "teacher_loss": 0.001045848592184484 + }, + { + "epoch": 0.2825836216839677, + "grad_norm": 5.5, + "kd_loss": 0.107421875, + "learning_rate": 1.156025601584676e-08, + "loss": 0.1779, + "step": 980, + "student_loss": 0.0017143742879852653, + "teacher_loss": 0.0004985960549674928 + }, + { + "epoch": 0.2840253748558247, + "grad_norm": 6.0625, + "kd_loss": 0.107421875, + "learning_rate": 6.711800878718144e-09, + "loss": 0.1914, + "step": 985, + "student_loss": 0.0008509993785992265, + "teacher_loss": 0.0007709055789746344 + }, + { + "epoch": 0.28546712802768165, + "grad_norm": 11.8125, + "kd_loss": 0.0986328125, + "learning_rate": 3.1727490318111953e-09, + "loss": 0.1871, + "step": 990, + "student_loss": 0.007576147560030222, + "teacher_loss": 0.015172326937317848 + }, + { + "epoch": 0.2869088811995386, + "grad_norm": 4.625, + "kd_loss": 0.1318359375, + "learning_rate": 9.440285301370865e-10, + "loss": 0.2025, + "step": 995, + "student_loss": 0.0030154017731547356, + "teacher_loss": 0.0012625143863260746 + }, + { + "epoch": 0.28835063437139563, + "grad_norm": 5.40625, + "kd_loss": 0.10546875, + "learning_rate": 2.622381702066523e-11, + "loss": 0.1678, + "step": 1000, + "student_loss": 0.08579359203577042, + "teacher_loss": 0.0004960879450663924 + }, + { + "epoch": 0.28835063437139563, + "kd_loss": 0.10546875, + "step": 1000, + "student_loss": 0.08579359203577042, + "teacher_loss": 0.0004960879450663924, + "total_flos": 0.0, + "train_loss": 0.2803848307132721, + "train_runtime": 4898.5131, + "train_samples_per_second": 3.266, + "train_steps_per_second": 0.204 + }, + { + "epoch": 0.2897923875432526, + "grad_norm": 4.21875, + "kd_loss": 0.107421875, + "learning_rate": 7.75705864825114e-06, + "loss": 0.1907, + "step": 1005, + "student_loss": 0.04834694042801857, + "teacher_loss": 0.07516621053218842 + }, + { + "epoch": 0.29123414071510956, + "grad_norm": 10.375, + "kd_loss": 0.11767578125, + "learning_rate": 7.734502946076656e-06, + "loss": 0.1897, + "step": 1010, + "student_loss": 0.001301786513067782, + "teacher_loss": 0.0013126098783686757 + }, + { + "epoch": 0.2926758938869666, + "grad_norm": 4.3125, + "kd_loss": 0.1748046875, + "learning_rate": 7.711867567242769e-06, + "loss": 0.1779, + "step": 1015, + "student_loss": 0.026407891884446144, + "teacher_loss": 0.008747180923819542 + }, + { + "epoch": 0.29411764705882354, + "grad_norm": 5.5, + "kd_loss": 0.2119140625, + "learning_rate": 7.689153171288487e-06, + "loss": 0.1807, + "step": 1020, + "student_loss": 0.029385080561041832, + "teacher_loss": 0.03383686766028404 + }, + { + "epoch": 0.2955594002306805, + "grad_norm": 9.375, + "kd_loss": 0.11474609375, + "learning_rate": 7.666360420055188e-06, + "loss": 0.1903, + "step": 1025, + "student_loss": 0.0012202368816360831, + "teacher_loss": 0.0004073530144523829 + }, + { + "epoch": 0.29700115340253747, + "grad_norm": 5.28125, + "kd_loss": 0.10400390625, + "learning_rate": 7.643489977667327e-06, + "loss": 0.1909, + "step": 1030, + "student_loss": 0.03574004024267197, + "teacher_loss": 0.0005011596367694438 + }, + { + "epoch": 0.2984429065743945, + "grad_norm": 5.71875, + "kd_loss": 0.11376953125, + "learning_rate": 7.6205425105130855e-06, + "loss": 0.1876, + "step": 1035, + "student_loss": 0.009626907296478748, + "teacher_loss": 0.0027181527111679316 + }, + { + "epoch": 0.29988465974625145, + "grad_norm": 5.5, + "kd_loss": 0.1044921875, + "learning_rate": 7.597518687224959e-06, + "loss": 0.2002, + "step": 1040, + "student_loss": 0.04977378249168396, + "teacher_loss": 0.0008017396903596818 + }, + { + "epoch": 0.3013264129181084, + "grad_norm": 5.84375, + "kd_loss": 0.0927734375, + "learning_rate": 7.574419178660269e-06, + "loss": 0.1864, + "step": 1045, + "student_loss": 0.00125790829770267, + "teacher_loss": 0.0010273143416270614 + }, + { + "epoch": 0.3027681660899654, + "grad_norm": 7.5, + "kd_loss": 0.10302734375, + "learning_rate": 7.551244657881618e-06, + "loss": 0.2226, + "step": 1050, + "student_loss": 0.396968275308609, + "teacher_loss": 0.0008409225265495479 + }, + { + "epoch": 0.3042099192618224, + "grad_norm": 5.46875, + "kd_loss": 0.11279296875, + "learning_rate": 7.527995800137287e-06, + "loss": 0.1736, + "step": 1055, + "student_loss": 0.001924316049553454, + "teacher_loss": 0.0013026405358687043 + }, + { + "epoch": 0.30565167243367936, + "grad_norm": 7.28125, + "kd_loss": 0.1083984375, + "learning_rate": 7.504673282841544e-06, + "loss": 0.179, + "step": 1060, + "student_loss": 0.07162957638502121, + "teacher_loss": 0.0005437894142232835 + }, + { + "epoch": 0.3070934256055363, + "grad_norm": 5.5, + "kd_loss": 0.1376953125, + "learning_rate": 7.481277785554918e-06, + "loss": 0.1812, + "step": 1065, + "student_loss": 0.051106277853250504, + "teacher_loss": 0.0007818956510163844 + }, + { + "epoch": 0.30853517877739334, + "grad_norm": 7.65625, + "kd_loss": 0.11865234375, + "learning_rate": 7.457809989964393e-06, + "loss": 0.2085, + "step": 1070, + "student_loss": 0.07073640078306198, + "teacher_loss": 0.0005237783188931644 + }, + { + "epoch": 0.3099769319492503, + "grad_norm": 4.5625, + "kd_loss": 0.10546875, + "learning_rate": 7.434270579863549e-06, + "loss": 0.1846, + "step": 1075, + "student_loss": 0.007236347068101168, + "teacher_loss": 0.0034689689055085182 + }, + { + "epoch": 0.31141868512110726, + "grad_norm": 7.4375, + "kd_loss": 0.11376953125, + "learning_rate": 7.4106602411326345e-06, + "loss": 0.1724, + "step": 1080, + "student_loss": 0.002105340361595154, + "teacher_loss": 0.01056791003793478 + }, + { + "epoch": 0.3128604382929642, + "grad_norm": 4.46875, + "kd_loss": 0.10302734375, + "learning_rate": 7.386979661718585e-06, + "loss": 0.1935, + "step": 1085, + "student_loss": 0.00163954496383667, + "teacher_loss": 0.0008159097633324564 + }, + { + "epoch": 0.31430219146482125, + "grad_norm": 5.59375, + "kd_loss": 0.1044921875, + "learning_rate": 7.363229531614973e-06, + "loss": 0.175, + "step": 1090, + "student_loss": 0.035642288625240326, + "teacher_loss": 0.0027311204466968775 + }, + { + "epoch": 0.3157439446366782, + "grad_norm": 3.6875, + "kd_loss": 0.12353515625, + "learning_rate": 7.339410542841906e-06, + "loss": 0.1985, + "step": 1095, + "student_loss": 0.14903123676776886, + "teacher_loss": 0.007329413667321205 + }, + { + "epoch": 0.31718569780853517, + "grad_norm": 7.9375, + "kd_loss": 0.11279296875, + "learning_rate": 7.315523389425867e-06, + "loss": 0.1747, + "step": 1100, + "student_loss": 0.0011135012609884143, + "teacher_loss": 0.0004325899062678218 + }, + { + "epoch": 0.31862745098039214, + "grad_norm": 3.328125, + "kd_loss": 0.099609375, + "learning_rate": 7.291568767379484e-06, + "loss": 0.1604, + "step": 1105, + "student_loss": 0.012246989645063877, + "teacher_loss": 0.0004818146117031574 + }, + { + "epoch": 0.32006920415224915, + "grad_norm": 4.15625, + "kd_loss": 0.107421875, + "learning_rate": 7.267547374681259e-06, + "loss": 0.1857, + "step": 1110, + "student_loss": 0.0022737043909728527, + "teacher_loss": 0.0010464838705956936 + }, + { + "epoch": 0.3215109573241061, + "grad_norm": 3.625, + "kd_loss": 0.1083984375, + "learning_rate": 7.24345991125522e-06, + "loss": 0.1756, + "step": 1115, + "student_loss": 0.013854903168976307, + "teacher_loss": 0.0004721402656286955 + }, + { + "epoch": 0.3229527104959631, + "grad_norm": 4.78125, + "kd_loss": 0.11474609375, + "learning_rate": 7.219307078950536e-06, + "loss": 0.175, + "step": 1120, + "student_loss": 0.08585968613624573, + "teacher_loss": 0.0014684133930131793 + }, + { + "epoch": 0.32439446366782004, + "grad_norm": 4.4375, + "kd_loss": 0.11572265625, + "learning_rate": 7.195089581521064e-06, + "loss": 0.1717, + "step": 1125, + "student_loss": 0.04334001988172531, + "teacher_loss": 0.021173296496272087 + }, + { + "epoch": 0.32583621683967706, + "grad_norm": 6.65625, + "kd_loss": 0.1220703125, + "learning_rate": 7.170808124604842e-06, + "loss": 0.1814, + "step": 1130, + "student_loss": 0.0865454375743866, + "teacher_loss": 0.000596641271840781 + }, + { + "epoch": 0.327277970011534, + "grad_norm": 5.90625, + "kd_loss": 0.099609375, + "learning_rate": 7.14646341570353e-06, + "loss": 0.1876, + "step": 1135, + "student_loss": 0.06569283455610275, + "teacher_loss": 0.005785806570202112 + }, + { + "epoch": 0.328719723183391, + "grad_norm": 4.03125, + "kd_loss": 0.1279296875, + "learning_rate": 7.122056164161795e-06, + "loss": 0.1885, + "step": 1140, + "student_loss": 0.03503354638814926, + "teacher_loss": 0.003907355945557356 + }, + { + "epoch": 0.330161476355248, + "grad_norm": 9.0625, + "kd_loss": 0.0908203125, + "learning_rate": 7.097587081146636e-06, + "loss": 0.1845, + "step": 1145, + "student_loss": 0.0037424068432301283, + "teacher_loss": 0.0011356781469658017 + }, + { + "epoch": 0.33160322952710497, + "grad_norm": 5.1875, + "kd_loss": 0.08984375, + "learning_rate": 7.073056879626681e-06, + "loss": 0.2248, + "step": 1150, + "student_loss": 0.002525654621422291, + "teacher_loss": 0.0010445830412209034 + }, + { + "epoch": 0.33304498269896193, + "grad_norm": 5.1875, + "kd_loss": 0.1044921875, + "learning_rate": 7.048466274351389e-06, + "loss": 0.1782, + "step": 1155, + "student_loss": 0.05463318154215813, + "teacher_loss": 0.00036173113039694726 + }, + { + "epoch": 0.3344867358708189, + "grad_norm": 5.84375, + "kd_loss": 0.11083984375, + "learning_rate": 7.023815981830236e-06, + "loss": 0.2102, + "step": 1160, + "student_loss": 0.002071135677397251, + "teacher_loss": 0.030623162165284157 + }, + { + "epoch": 0.3359284890426759, + "grad_norm": 4.5, + "kd_loss": 0.119140625, + "learning_rate": 6.999106720311846e-06, + "loss": 0.1978, + "step": 1165, + "student_loss": 0.06917373836040497, + "teacher_loss": 0.0006929833325557411 + }, + { + "epoch": 0.3373702422145329, + "grad_norm": 4.9375, + "kd_loss": 0.126953125, + "learning_rate": 6.974339209763043e-06, + "loss": 0.1661, + "step": 1170, + "student_loss": 0.047193318605422974, + "teacher_loss": 0.01937233842909336 + }, + { + "epoch": 0.33881199538638984, + "grad_norm": 5.0625, + "kd_loss": 0.10302734375, + "learning_rate": 6.949514171847891e-06, + "loss": 0.1856, + "step": 1175, + "student_loss": 0.0045834011398255825, + "teacher_loss": 0.0021127781365066767 + }, + { + "epoch": 0.3402537485582468, + "grad_norm": 4.4375, + "kd_loss": 0.10888671875, + "learning_rate": 6.924632329906657e-06, + "loss": 0.1824, + "step": 1180, + "student_loss": 0.03225143998861313, + "teacher_loss": 0.0007702379371039569 + }, + { + "epoch": 0.3416955017301038, + "grad_norm": 4.53125, + "kd_loss": 0.109375, + "learning_rate": 6.899694408934734e-06, + "loss": 0.1567, + "step": 1185, + "student_loss": 0.011755186133086681, + "teacher_loss": 0.0007617790251970291 + }, + { + "epoch": 0.3431372549019608, + "grad_norm": 4.46875, + "kd_loss": 0.1123046875, + "learning_rate": 6.874701135561524e-06, + "loss": 0.1666, + "step": 1190, + "student_loss": 0.0012172460556030273, + "teacher_loss": 0.00039286207174882293 + }, + { + "epoch": 0.34457900807381775, + "grad_norm": 5.34375, + "kd_loss": 0.09814453125, + "learning_rate": 6.849653238029261e-06, + "loss": 0.1611, + "step": 1195, + "student_loss": 0.005217494908720255, + "teacher_loss": 0.0005248200614005327 + }, + { + "epoch": 0.3460207612456747, + "grad_norm": 4.9375, + "kd_loss": 0.10546875, + "learning_rate": 6.824551446171788e-06, + "loss": 0.2491, + "step": 1200, + "student_loss": 0.00212606368586421, + "teacher_loss": 0.0009664412937127054 + }, + { + "epoch": 0.34746251441753173, + "grad_norm": 5.25, + "kd_loss": 0.10302734375, + "learning_rate": 6.7993964913932975e-06, + "loss": 0.206, + "step": 1205, + "student_loss": 0.0657915249466896, + "teacher_loss": 0.0004779589653480798 + }, + { + "epoch": 0.3489042675893887, + "grad_norm": 6.71875, + "kd_loss": 0.1025390625, + "learning_rate": 6.774189106647021e-06, + "loss": 0.1767, + "step": 1210, + "student_loss": 0.054130807518959045, + "teacher_loss": 0.0005891909822821617 + }, + { + "epoch": 0.35034602076124566, + "grad_norm": 4.8125, + "kd_loss": 0.10400390625, + "learning_rate": 6.748930026413865e-06, + "loss": 0.1911, + "step": 1215, + "student_loss": 0.10985054075717926, + "teacher_loss": 0.0005408765282481909 + }, + { + "epoch": 0.3517877739331027, + "grad_norm": 4.9375, + "kd_loss": 0.115234375, + "learning_rate": 6.7236199866810185e-06, + "loss": 0.1839, + "step": 1220, + "student_loss": 0.04293162375688553, + "teacher_loss": 0.010158946737647057 + }, + { + "epoch": 0.35322952710495964, + "grad_norm": 5.46875, + "kd_loss": 0.11328125, + "learning_rate": 6.698259724920503e-06, + "loss": 0.1899, + "step": 1225, + "student_loss": 0.025801874697208405, + "teacher_loss": 0.0039003598503768444 + }, + { + "epoch": 0.3546712802768166, + "grad_norm": 7.03125, + "kd_loss": 0.103515625, + "learning_rate": 6.672849980067685e-06, + "loss": 0.1782, + "step": 1230, + "student_loss": 0.0011705292854458094, + "teacher_loss": 0.0013004811480641365 + }, + { + "epoch": 0.35611303344867357, + "grad_norm": 4.96875, + "kd_loss": 0.134765625, + "learning_rate": 6.647391492499746e-06, + "loss": 0.1692, + "step": 1235, + "student_loss": 0.0030034382361918688, + "teacher_loss": 0.0015800351975485682 + }, + { + "epoch": 0.3575547866205306, + "grad_norm": 5.6875, + "kd_loss": 0.1015625, + "learning_rate": 6.621885004014113e-06, + "loss": 0.1994, + "step": 1240, + "student_loss": 0.0015772647457197309, + "teacher_loss": 0.0011423062533140182 + }, + { + "epoch": 0.35899653979238755, + "grad_norm": 4.25, + "kd_loss": 0.11669921875, + "learning_rate": 6.596331257806837e-06, + "loss": 0.1877, + "step": 1245, + "student_loss": 0.0013259351253509521, + "teacher_loss": 0.0011870627058669925 + }, + { + "epoch": 0.3604382929642445, + "grad_norm": 3.96875, + "kd_loss": 0.1162109375, + "learning_rate": 6.570730998450945e-06, + "loss": 0.1859, + "step": 1250, + "student_loss": 0.26618441939353943, + "teacher_loss": 0.002830672077834606 + }, + { + "epoch": 0.3618800461361015, + "grad_norm": 4.75, + "kd_loss": 0.11962890625, + "learning_rate": 6.545084971874738e-06, + "loss": 0.1923, + "step": 1255, + "student_loss": 0.04056844487786293, + "teacher_loss": 0.006723723839968443 + }, + { + "epoch": 0.3633217993079585, + "grad_norm": 5.0, + "kd_loss": 0.1142578125, + "learning_rate": 6.519393925340067e-06, + "loss": 0.1923, + "step": 1260, + "student_loss": 0.0006198959308676422, + "teacher_loss": 0.0006204553064890206 + }, + { + "epoch": 0.36476355247981546, + "grad_norm": 6.21875, + "kd_loss": 0.10888671875, + "learning_rate": 6.49365860742055e-06, + "loss": 0.1947, + "step": 1265, + "student_loss": 0.41350483894348145, + "teacher_loss": 0.020098216831684113 + }, + { + "epoch": 0.3662053056516724, + "grad_norm": 4.65625, + "kd_loss": 0.10107421875, + "learning_rate": 6.467879767979764e-06, + "loss": 0.1638, + "step": 1270, + "student_loss": 0.0016804474871605635, + "teacher_loss": 0.0008275846485048532 + }, + { + "epoch": 0.36764705882352944, + "grad_norm": 4.8125, + "kd_loss": 0.107421875, + "learning_rate": 6.442058158149396e-06, + "loss": 0.1518, + "step": 1275, + "student_loss": 0.07305289059877396, + "teacher_loss": 0.0009938895236700773 + }, + { + "epoch": 0.3690888119953864, + "grad_norm": 4.53125, + "kd_loss": 0.10693359375, + "learning_rate": 6.4161945303073535e-06, + "loss": 0.1549, + "step": 1280, + "student_loss": 0.003565654158592224, + "teacher_loss": 0.0004424946673680097 + }, + { + "epoch": 0.37053056516724336, + "grad_norm": 4.15625, + "kd_loss": 0.11865234375, + "learning_rate": 6.390289638055851e-06, + "loss": 0.1723, + "step": 1285, + "student_loss": 0.0022811174858361483, + "teacher_loss": 0.000994974747300148 + }, + { + "epoch": 0.3719723183391003, + "grad_norm": 6.40625, + "kd_loss": 0.08837890625, + "learning_rate": 6.364344236199441e-06, + "loss": 0.1828, + "step": 1290, + "student_loss": 0.4067663848400116, + "teacher_loss": 0.015121642500162125 + }, + { + "epoch": 0.37341407151095735, + "grad_norm": 6.5, + "kd_loss": 0.091796875, + "learning_rate": 6.3383590807230264e-06, + "loss": 0.2023, + "step": 1295, + "student_loss": 0.0015801489353179932, + "teacher_loss": 0.007307669147849083 + }, + { + "epoch": 0.3748558246828143, + "grad_norm": 4.75, + "kd_loss": 0.1259765625, + "learning_rate": 6.3123349287698345e-06, + "loss": 0.1514, + "step": 1300, + "student_loss": 0.002292349934577942, + "teacher_loss": 0.0012089353986084461 + }, + { + "epoch": 0.3762975778546713, + "grad_norm": 4.875, + "kd_loss": 0.12451171875, + "learning_rate": 6.286272538619351e-06, + "loss": 0.1734, + "step": 1305, + "student_loss": 0.004027123097330332, + "teacher_loss": 0.30039486289024353 + }, + { + "epoch": 0.37773933102652824, + "grad_norm": 5.6875, + "kd_loss": 0.111328125, + "learning_rate": 6.260172669665233e-06, + "loss": 0.1584, + "step": 1310, + "student_loss": 0.00176389014814049, + "teacher_loss": 0.001472116680815816 + }, + { + "epoch": 0.37918108419838525, + "grad_norm": 6.40625, + "kd_loss": 0.11669921875, + "learning_rate": 6.234036082393171e-06, + "loss": 0.1926, + "step": 1315, + "student_loss": 0.11009548604488373, + "teacher_loss": 0.0014415290206670761 + }, + { + "epoch": 0.3806228373702422, + "grad_norm": 4.78125, + "kd_loss": 0.11865234375, + "learning_rate": 6.207863538358741e-06, + "loss": 0.1882, + "step": 1320, + "student_loss": 0.012124598026275635, + "teacher_loss": 0.0018767216242849827 + }, + { + "epoch": 0.3820645905420992, + "grad_norm": 5.1875, + "kd_loss": 0.11328125, + "learning_rate": 6.181655800165207e-06, + "loss": 0.1981, + "step": 1325, + "student_loss": 0.009575133211910725, + "teacher_loss": 0.0010605778079479933 + }, + { + "epoch": 0.38350634371395614, + "grad_norm": 6.09375, + "kd_loss": 0.111328125, + "learning_rate": 6.155413631441307e-06, + "loss": 0.1758, + "step": 1330, + "student_loss": 0.06053456291556358, + "teacher_loss": 0.0006676834309473634 + }, + { + "epoch": 0.38494809688581316, + "grad_norm": 5.53125, + "kd_loss": 0.099609375, + "learning_rate": 6.129137796818997e-06, + "loss": 0.1761, + "step": 1335, + "student_loss": 0.0009542008046992123, + "teacher_loss": 0.0006002294248901308 + }, + { + "epoch": 0.3863898500576701, + "grad_norm": 5.75, + "kd_loss": 0.1650390625, + "learning_rate": 6.102829061911176e-06, + "loss": 0.1853, + "step": 1340, + "student_loss": 0.005108754616230726, + "teacher_loss": 0.005764763802289963 + }, + { + "epoch": 0.3878316032295271, + "grad_norm": 6.65625, + "kd_loss": 0.12353515625, + "learning_rate": 6.076488193289375e-06, + "loss": 0.1821, + "step": 1345, + "student_loss": 0.0015393177745863795, + "teacher_loss": 0.0016605017008259892 + }, + { + "epoch": 0.3892733564013841, + "grad_norm": 8.4375, + "kd_loss": 0.11865234375, + "learning_rate": 6.050115958461423e-06, + "loss": 0.1673, + "step": 1350, + "student_loss": 0.048770517110824585, + "teacher_loss": 0.0007486655958928168 + }, + { + "epoch": 0.39071510957324107, + "grad_norm": 6.1875, + "kd_loss": 0.103515625, + "learning_rate": 6.02371312584908e-06, + "loss": 0.1894, + "step": 1355, + "student_loss": 0.09447399526834488, + "teacher_loss": 0.0003722972178366035 + }, + { + "epoch": 0.39215686274509803, + "grad_norm": 8.5, + "kd_loss": 0.11328125, + "learning_rate": 5.997280464765655e-06, + "loss": 0.1885, + "step": 1360, + "student_loss": 0.14768695831298828, + "teacher_loss": 0.0006404736777767539 + }, + { + "epoch": 0.393598615916955, + "grad_norm": 7.8125, + "kd_loss": 0.109375, + "learning_rate": 5.970818745393579e-06, + "loss": 0.1734, + "step": 1365, + "student_loss": 0.01714850589632988, + "teacher_loss": 0.029827572405338287 + }, + { + "epoch": 0.395040369088812, + "grad_norm": 7.6875, + "kd_loss": 0.09619140625, + "learning_rate": 5.9443287387619754e-06, + "loss": 0.1706, + "step": 1370, + "student_loss": 0.32004305720329285, + "teacher_loss": 0.018626194447278976 + }, + { + "epoch": 0.396482122260669, + "grad_norm": 4.5, + "kd_loss": 0.10791015625, + "learning_rate": 5.9178112167241805e-06, + "loss": 0.1781, + "step": 1375, + "student_loss": 0.09515227377414703, + "teacher_loss": 0.00036773309693671763 + }, + { + "epoch": 0.39792387543252594, + "grad_norm": 4.6875, + "kd_loss": 0.10546875, + "learning_rate": 5.8912669519352725e-06, + "loss": 0.1764, + "step": 1380, + "student_loss": 0.011620831675827503, + "teacher_loss": 0.0029276730492711067 + }, + { + "epoch": 0.3993656286043829, + "grad_norm": 4.46875, + "kd_loss": 0.10791015625, + "learning_rate": 5.864696717829539e-06, + "loss": 0.1843, + "step": 1385, + "student_loss": 0.10878758132457733, + "teacher_loss": 0.013216445222496986 + }, + { + "epoch": 0.4008073817762399, + "grad_norm": 6.53125, + "kd_loss": 0.126953125, + "learning_rate": 5.838101288597951e-06, + "loss": 0.1739, + "step": 1390, + "student_loss": 0.0017183037707582116, + "teacher_loss": 0.0011319200275465846 + }, + { + "epoch": 0.4022491349480969, + "grad_norm": 4.5, + "kd_loss": 0.1171875, + "learning_rate": 5.8114814391656046e-06, + "loss": 0.1812, + "step": 1395, + "student_loss": 0.0009771361947059631, + "teacher_loss": 0.0006259285728447139 + }, + { + "epoch": 0.40369088811995385, + "grad_norm": 4.59375, + "kd_loss": 0.10888671875, + "learning_rate": 5.78483794516914e-06, + "loss": 0.1759, + "step": 1400, + "student_loss": 0.0028622837271541357, + "teacher_loss": 0.00048269989201799035 + }, + { + "epoch": 0.40513264129181087, + "grad_norm": 5.84375, + "kd_loss": 0.1064453125, + "learning_rate": 5.75817158293414e-06, + "loss": 0.1843, + "step": 1405, + "student_loss": 0.0025167667772620916, + "teacher_loss": 0.0007314765243791044 + }, + { + "epoch": 0.40657439446366783, + "grad_norm": 4.625, + "kd_loss": 0.10693359375, + "learning_rate": 5.731483129452514e-06, + "loss": 0.1836, + "step": 1410, + "student_loss": 0.0036077857948839664, + "teacher_loss": 0.0006345498841255903 + }, + { + "epoch": 0.4080161476355248, + "grad_norm": 9.625, + "kd_loss": 0.1083984375, + "learning_rate": 5.704773362359854e-06, + "loss": 0.1652, + "step": 1415, + "student_loss": 0.011436261236667633, + "teacher_loss": 0.01508229598402977 + }, + { + "epoch": 0.40945790080738176, + "grad_norm": 5.90625, + "kd_loss": 0.1201171875, + "learning_rate": 5.678043059912776e-06, + "loss": 0.202, + "step": 1420, + "student_loss": 0.001543233753181994, + "teacher_loss": 0.0010519566712900996 + }, + { + "epoch": 0.4108996539792388, + "grad_norm": 4.1875, + "kd_loss": 0.09765625, + "learning_rate": 5.6512930009662524e-06, + "loss": 0.1847, + "step": 1425, + "student_loss": 0.1836177557706833, + "teacher_loss": 0.000503146555274725 + }, + { + "epoch": 0.41234140715109574, + "grad_norm": 5.09375, + "kd_loss": 0.11279296875, + "learning_rate": 5.624523964950903e-06, + "loss": 0.1717, + "step": 1430, + "student_loss": 0.005288061685860157, + "teacher_loss": 0.0005534209776669741 + }, + { + "epoch": 0.4137831603229527, + "grad_norm": 5.125, + "kd_loss": 0.1044921875, + "learning_rate": 5.597736731850295e-06, + "loss": 0.1881, + "step": 1435, + "student_loss": 0.019315902143716812, + "teacher_loss": 0.00202935584820807 + }, + { + "epoch": 0.41522491349480967, + "grad_norm": 8.5, + "kd_loss": 0.10107421875, + "learning_rate": 5.570932082178219e-06, + "loss": 0.1903, + "step": 1440, + "student_loss": 0.08893950283527374, + "teacher_loss": 0.000398140778997913 + }, + { + "epoch": 0.4166666666666667, + "grad_norm": 9.5625, + "kd_loss": 0.12890625, + "learning_rate": 5.5441107969559315e-06, + "loss": 0.1919, + "step": 1445, + "student_loss": 0.2631194293498993, + "teacher_loss": 0.0044821687042713165 + }, + { + "epoch": 0.41810841983852365, + "grad_norm": 5.1875, + "kd_loss": 0.1826171875, + "learning_rate": 5.517273657689419e-06, + "loss": 0.1631, + "step": 1450, + "student_loss": 0.006265243049710989, + "teacher_loss": 0.0023199233692139387 + }, + { + "epoch": 0.4195501730103806, + "grad_norm": 5.84375, + "kd_loss": 0.12890625, + "learning_rate": 5.490421446346608e-06, + "loss": 0.1744, + "step": 1455, + "student_loss": 0.0026093143969774246, + "teacher_loss": 0.0007649324252270162 + }, + { + "epoch": 0.4209919261822376, + "grad_norm": 5.8125, + "kd_loss": 0.134765625, + "learning_rate": 5.463554945334589e-06, + "loss": 0.1769, + "step": 1460, + "student_loss": 0.0007114948239177465, + "teacher_loss": 0.0003705586714204401 + }, + { + "epoch": 0.4224336793540946, + "grad_norm": 13.4375, + "kd_loss": 0.1435546875, + "learning_rate": 5.43667493747682e-06, + "loss": 0.2009, + "step": 1465, + "student_loss": 0.0005334956222213805, + "teacher_loss": 0.013963425531983376 + }, + { + "epoch": 0.42387543252595156, + "grad_norm": 8.375, + "kd_loss": 0.1015625, + "learning_rate": 5.409782205990317e-06, + "loss": 0.1964, + "step": 1470, + "student_loss": 0.0011849907459691167, + "teacher_loss": 0.004421910271048546 + }, + { + "epoch": 0.4253171856978085, + "grad_norm": 4.5625, + "kd_loss": 0.11376953125, + "learning_rate": 5.3828775344628245e-06, + "loss": 0.1738, + "step": 1475, + "student_loss": 0.0008997453842312098, + "teacher_loss": 0.0010891201673075557 + }, + { + "epoch": 0.42675893886966554, + "grad_norm": 4.78125, + "kd_loss": 0.1123046875, + "learning_rate": 5.355961706829997e-06, + "loss": 0.1813, + "step": 1480, + "student_loss": 0.000982648809440434, + "teacher_loss": 0.0016894500004127622 + }, + { + "epoch": 0.4282006920415225, + "grad_norm": 13.5625, + "kd_loss": 0.11279296875, + "learning_rate": 5.329035507352548e-06, + "loss": 0.185, + "step": 1485, + "student_loss": 0.0005971972714178264, + "teacher_loss": 0.0009129931568168104 + }, + { + "epoch": 0.42964244521337946, + "grad_norm": 7.09375, + "kd_loss": 0.119140625, + "learning_rate": 5.3020997205933985e-06, + "loss": 0.1721, + "step": 1490, + "student_loss": 0.004597559571266174, + "teacher_loss": 0.001885790377855301 + }, + { + "epoch": 0.43108419838523643, + "grad_norm": 4.6875, + "kd_loss": 0.1015625, + "learning_rate": 5.275155131394825e-06, + "loss": 0.18, + "step": 1495, + "student_loss": 0.0022636561188846827, + "teacher_loss": 0.0010887808166444302 + }, + { + "epoch": 0.43252595155709345, + "grad_norm": 6.21875, + "kd_loss": 0.1103515625, + "learning_rate": 5.248202524855578e-06, + "loss": 0.1757, + "step": 1500, + "student_loss": 0.04894772171974182, + "teacher_loss": 0.004637535195797682 + }, + { + "epoch": 0.4339677047289504, + "grad_norm": 5.0, + "kd_loss": 0.10302734375, + "learning_rate": 5.221242686308019e-06, + "loss": 0.172, + "step": 1505, + "student_loss": 0.13838815689086914, + "teacher_loss": 0.0005776140023954213 + }, + { + "epoch": 0.4354094579008074, + "grad_norm": 5.46875, + "kd_loss": 0.1435546875, + "learning_rate": 5.194276401295231e-06, + "loss": 0.1758, + "step": 1510, + "student_loss": 0.1576482057571411, + "teacher_loss": 0.013320242054760456 + }, + { + "epoch": 0.43685121107266434, + "grad_norm": 5.375, + "kd_loss": 0.10791015625, + "learning_rate": 5.167304455548128e-06, + "loss": 0.1679, + "step": 1515, + "student_loss": 0.0007608433370478451, + "teacher_loss": 0.0003724750131368637 + }, + { + "epoch": 0.43829296424452135, + "grad_norm": 6.25, + "kd_loss": 0.1181640625, + "learning_rate": 5.14032763496257e-06, + "loss": 0.1759, + "step": 1520, + "student_loss": 0.005186790134757757, + "teacher_loss": 0.0005813217139802873 + }, + { + "epoch": 0.4397347174163783, + "grad_norm": 7.0, + "kd_loss": 0.146484375, + "learning_rate": 5.11334672557645e-06, + "loss": 0.1712, + "step": 1525, + "student_loss": 0.012267577461898327, + "teacher_loss": 0.0004309048526920378 + }, + { + "epoch": 0.4411764705882353, + "grad_norm": 9.125, + "kd_loss": 0.1630859375, + "learning_rate": 5.086362513546807e-06, + "loss": 0.1671, + "step": 1530, + "student_loss": 0.30388641357421875, + "teacher_loss": 0.007112511433660984 + }, + { + "epoch": 0.4426182237600923, + "grad_norm": 6.78125, + "kd_loss": 0.1123046875, + "learning_rate": 5.059375785126907e-06, + "loss": 0.1674, + "step": 1535, + "student_loss": 0.20476478338241577, + "teacher_loss": 0.012248540297150612 + }, + { + "epoch": 0.44405997693194926, + "grad_norm": 4.875, + "kd_loss": 0.11669921875, + "learning_rate": 5.032387326643331e-06, + "loss": 0.1778, + "step": 1540, + "student_loss": 0.10948977619409561, + "teacher_loss": 0.0005565967294387519 + }, + { + "epoch": 0.4455017301038062, + "grad_norm": 6.40625, + "kd_loss": 0.115234375, + "learning_rate": 5.005397924473082e-06, + "loss": 0.1764, + "step": 1545, + "student_loss": 0.0020793951116502285, + "teacher_loss": 0.00420406274497509 + }, + { + "epoch": 0.4469434832756632, + "grad_norm": 5.8125, + "kd_loss": 0.1005859375, + "learning_rate": 4.978408365020651e-06, + "loss": 0.2026, + "step": 1550, + "student_loss": 0.060113225132226944, + "teacher_loss": 0.0008351522265002131 + }, + { + "epoch": 0.4483852364475202, + "grad_norm": 8.9375, + "kd_loss": 0.12255859375, + "learning_rate": 4.951419434695115e-06, + "loss": 0.1805, + "step": 1555, + "student_loss": 0.008758403360843658, + "teacher_loss": 0.00422658259049058 + }, + { + "epoch": 0.44982698961937717, + "grad_norm": 5.3125, + "kd_loss": 0.10595703125, + "learning_rate": 4.924431919887216e-06, + "loss": 0.1829, + "step": 1560, + "student_loss": 0.003116800682619214, + "teacher_loss": 0.0007468878175131977 + }, + { + "epoch": 0.45126874279123413, + "grad_norm": 4.75, + "kd_loss": 0.10400390625, + "learning_rate": 4.897446606946459e-06, + "loss": 0.1619, + "step": 1565, + "student_loss": 0.01210244745016098, + "teacher_loss": 0.0016926496755331755 + }, + { + "epoch": 0.4527104959630911, + "grad_norm": 5.71875, + "kd_loss": 0.13671875, + "learning_rate": 4.870464282158184e-06, + "loss": 0.1666, + "step": 1570, + "student_loss": 0.0017574841622263193, + "teacher_loss": 0.02678789757192135 + }, + { + "epoch": 0.4541522491349481, + "grad_norm": 5.71875, + "kd_loss": 0.1513671875, + "learning_rate": 4.84348573172067e-06, + "loss": 0.1654, + "step": 1575, + "student_loss": 0.04304865747690201, + "teacher_loss": 0.023975025862455368 + }, + { + "epoch": 0.4555940023068051, + "grad_norm": 4.65625, + "kd_loss": 0.09326171875, + "learning_rate": 4.816511741722215e-06, + "loss": 0.1877, + "step": 1580, + "student_loss": 0.15135310590267181, + "teacher_loss": 0.0006109736277721822 + }, + { + "epoch": 0.45703575547866204, + "grad_norm": 5.625, + "kd_loss": 0.10400390625, + "learning_rate": 4.7895430981182415e-06, + "loss": 0.194, + "step": 1585, + "student_loss": 0.0017708265222609043, + "teacher_loss": 0.0013553814496845007 + }, + { + "epoch": 0.458477508650519, + "grad_norm": 5.625, + "kd_loss": 0.1396484375, + "learning_rate": 4.762580586708389e-06, + "loss": 0.1849, + "step": 1590, + "student_loss": 0.002241171430796385, + "teacher_loss": 0.0015448500635102391 + }, + { + "epoch": 0.459919261822376, + "grad_norm": 6.375, + "kd_loss": 0.125, + "learning_rate": 4.73562499311362e-06, + "loss": 0.1746, + "step": 1595, + "student_loss": 0.001443210057914257, + "teacher_loss": 0.001196635770611465 + }, + { + "epoch": 0.461361014994233, + "grad_norm": 7.71875, + "kd_loss": 0.09765625, + "learning_rate": 4.708677102753331e-06, + "loss": 0.1837, + "step": 1600, + "student_loss": 0.06453961879014969, + "teacher_loss": 0.0021809807512909174 + }, + { + "epoch": 0.46280276816608995, + "grad_norm": 7.28125, + "kd_loss": 0.123046875, + "learning_rate": 4.681737700822464e-06, + "loss": 0.1922, + "step": 1605, + "student_loss": 0.0005613799439743161, + "teacher_loss": 0.025183849036693573 + }, + { + "epoch": 0.46424452133794697, + "grad_norm": 6.1875, + "kd_loss": 0.1513671875, + "learning_rate": 4.654807572268628e-06, + "loss": 0.1916, + "step": 1610, + "student_loss": 0.005002737045288086, + "teacher_loss": 0.0012489790096879005 + }, + { + "epoch": 0.46568627450980393, + "grad_norm": 4.09375, + "kd_loss": 0.11083984375, + "learning_rate": 4.627887501769231e-06, + "loss": 0.1804, + "step": 1615, + "student_loss": 0.007485619746148586, + "teacher_loss": 0.0014420546358451247 + }, + { + "epoch": 0.4671280276816609, + "grad_norm": 6.59375, + "kd_loss": 0.1435546875, + "learning_rate": 4.600978273708612e-06, + "loss": 0.1793, + "step": 1620, + "student_loss": 0.02008935436606407, + "teacher_loss": 0.0020190745126456022 + }, + { + "epoch": 0.46856978085351786, + "grad_norm": 9.1875, + "kd_loss": 0.1298828125, + "learning_rate": 4.574080672155189e-06, + "loss": 0.1876, + "step": 1625, + "student_loss": 0.004716619849205017, + "teacher_loss": 0.0006995577132329345 + }, + { + "epoch": 0.4700115340253749, + "grad_norm": 7.0625, + "kd_loss": 0.0966796875, + "learning_rate": 4.547195480838612e-06, + "loss": 0.184, + "step": 1630, + "student_loss": 0.06263118237257004, + "teacher_loss": 0.0006376359378919005 + }, + { + "epoch": 0.47145328719723184, + "grad_norm": 6.125, + "kd_loss": 0.119140625, + "learning_rate": 4.520323483126928e-06, + "loss": 0.2015, + "step": 1635, + "student_loss": 0.15935464203357697, + "teacher_loss": 0.0003686147683765739 + }, + { + "epoch": 0.4728950403690888, + "grad_norm": 6.5, + "kd_loss": 0.115234375, + "learning_rate": 4.493465462003756e-06, + "loss": 0.1754, + "step": 1640, + "student_loss": 0.0011656845454126596, + "teacher_loss": 0.0015453645028173923 + }, + { + "epoch": 0.47433679354094577, + "grad_norm": 6.625, + "kd_loss": 0.1328125, + "learning_rate": 4.4666222000454685e-06, + "loss": 0.1864, + "step": 1645, + "student_loss": 0.0029342269990593195, + "teacher_loss": 0.0005665006465278566 + }, + { + "epoch": 0.4757785467128028, + "grad_norm": 9.4375, + "kd_loss": 0.10107421875, + "learning_rate": 4.4397944793983946e-06, + "loss": 0.1819, + "step": 1650, + "student_loss": 0.0005228065419942141, + "teacher_loss": 0.0007448173710145056 + }, + { + "epoch": 0.47722029988465975, + "grad_norm": 5.3125, + "kd_loss": 0.10498046875, + "learning_rate": 4.4129830817560284e-06, + "loss": 0.1901, + "step": 1655, + "student_loss": 0.001164126442745328, + "teacher_loss": 0.0010993466712534428 + }, + { + "epoch": 0.4786620530565167, + "grad_norm": 5.59375, + "kd_loss": 0.1201171875, + "learning_rate": 4.386188788336251e-06, + "loss": 0.1714, + "step": 1660, + "student_loss": 0.07953157275915146, + "teacher_loss": 0.0009381614509038627 + }, + { + "epoch": 0.4801038062283737, + "grad_norm": 5.59375, + "kd_loss": 0.119140625, + "learning_rate": 4.359412379858569e-06, + "loss": 0.1704, + "step": 1665, + "student_loss": 0.0008934318902902305, + "teacher_loss": 0.0003539333411026746 + }, + { + "epoch": 0.4815455594002307, + "grad_norm": 8.9375, + "kd_loss": 0.17578125, + "learning_rate": 4.332654636521365e-06, + "loss": 0.187, + "step": 1670, + "student_loss": 0.12339435517787933, + "teacher_loss": 0.004934420809149742 + }, + { + "epoch": 0.48298731257208766, + "grad_norm": 8.4375, + "kd_loss": 0.1181640625, + "learning_rate": 4.3059163379791676e-06, + "loss": 0.1854, + "step": 1675, + "student_loss": 0.0012413784861564636, + "teacher_loss": 0.0013670932967215776 + }, + { + "epoch": 0.4844290657439446, + "grad_norm": 8.3125, + "kd_loss": 0.1376953125, + "learning_rate": 4.279198263319932e-06, + "loss": 0.2255, + "step": 1680, + "student_loss": 0.05413307249546051, + "teacher_loss": 0.0009388276375830173 + }, + { + "epoch": 0.48587081891580164, + "grad_norm": 8.25, + "kd_loss": 0.1015625, + "learning_rate": 4.252501191042334e-06, + "loss": 0.1783, + "step": 1685, + "student_loss": 0.002770025050267577, + "teacher_loss": 0.0007223464781418443 + }, + { + "epoch": 0.4873125720876586, + "grad_norm": 6.34375, + "kd_loss": 0.10205078125, + "learning_rate": 4.2258258990331015e-06, + "loss": 0.1702, + "step": 1690, + "student_loss": 0.001255531096830964, + "teacher_loss": 0.001056861481629312 + }, + { + "epoch": 0.48875432525951557, + "grad_norm": 3.625, + "kd_loss": 0.12353515625, + "learning_rate": 4.199173164544331e-06, + "loss": 0.1609, + "step": 1695, + "student_loss": 0.00047228377661667764, + "teacher_loss": 0.0005455633508972824 + }, + { + "epoch": 0.49019607843137253, + "grad_norm": 4.1875, + "kd_loss": 0.1259765625, + "learning_rate": 4.1725437641708535e-06, + "loss": 0.1627, + "step": 1700, + "student_loss": 0.002585329581052065, + "teacher_loss": 0.0003256455820519477 + }, + { + "epoch": 0.49163783160322955, + "grad_norm": 6.1875, + "kd_loss": 0.10498046875, + "learning_rate": 4.145938473827598e-06, + "loss": 0.1941, + "step": 1705, + "student_loss": 0.04190181940793991, + "teacher_loss": 0.0010891788406297565 + }, + { + "epoch": 0.4930795847750865, + "grad_norm": 6.65625, + "kd_loss": 0.10888671875, + "learning_rate": 4.1193580687269896e-06, + "loss": 0.1999, + "step": 1710, + "student_loss": 0.16470496356487274, + "teacher_loss": 0.0004161170218139887 + }, + { + "epoch": 0.4945213379469435, + "grad_norm": 6.21875, + "kd_loss": 0.10107421875, + "learning_rate": 4.092803323356357e-06, + "loss": 0.1825, + "step": 1715, + "student_loss": 0.0012359149986878037, + "teacher_loss": 0.0005305964150466025 + }, + { + "epoch": 0.49596309111880044, + "grad_norm": 6.53125, + "kd_loss": 0.11328125, + "learning_rate": 4.066275011455369e-06, + "loss": 0.156, + "step": 1720, + "student_loss": 0.0009467661147937179, + "teacher_loss": 0.0009835466044023633 + }, + { + "epoch": 0.49740484429065746, + "grad_norm": 6.1875, + "kd_loss": 0.1328125, + "learning_rate": 4.039773905993486e-06, + "loss": 0.1996, + "step": 1725, + "student_loss": 0.0016510486602783203, + "teacher_loss": 0.0012200974160805345 + }, + { + "epoch": 0.4988465974625144, + "grad_norm": 5.40625, + "kd_loss": 0.115234375, + "learning_rate": 4.013300779147445e-06, + "loss": 0.1658, + "step": 1730, + "student_loss": 0.006948298308998346, + "teacher_loss": 0.025476913899183273 + }, + { + "epoch": 0.5002883506343714, + "grad_norm": 5.625, + "kd_loss": 0.146484375, + "learning_rate": 3.98685640227875e-06, + "loss": 0.1611, + "step": 1735, + "student_loss": 0.020067384466528893, + "teacher_loss": 0.04975026100873947 + }, + { + "epoch": 0.5017301038062284, + "grad_norm": 4.0, + "kd_loss": 0.119140625, + "learning_rate": 3.960441545911205e-06, + "loss": 0.1927, + "step": 1740, + "student_loss": 0.01879638433456421, + "teacher_loss": 0.026825515553355217 + }, + { + "epoch": 0.5031718569780853, + "grad_norm": 3.8125, + "kd_loss": 0.11181640625, + "learning_rate": 3.934056979708456e-06, + "loss": 0.1586, + "step": 1745, + "student_loss": 0.09886375069618225, + "teacher_loss": 0.01867763325572014 + }, + { + "epoch": 0.5046136101499423, + "grad_norm": 8.9375, + "kd_loss": 0.10791015625, + "learning_rate": 3.907703472451574e-06, + "loss": 0.1918, + "step": 1750, + "student_loss": 0.011541678570210934, + "teacher_loss": 0.000535048427991569 + }, + { + "epoch": 0.5060553633217993, + "grad_norm": 9.5625, + "kd_loss": 0.1142578125, + "learning_rate": 3.881381792016645e-06, + "loss": 0.1977, + "step": 1755, + "student_loss": 0.0006449613138101995, + "teacher_loss": 0.0003036665730178356 + }, + { + "epoch": 0.5074971164936563, + "grad_norm": 6.84375, + "kd_loss": 0.130859375, + "learning_rate": 3.8550927053523994e-06, + "loss": 0.1702, + "step": 1760, + "student_loss": 0.0009996505687013268, + "teacher_loss": 0.04788116738200188 + }, + { + "epoch": 0.5089388696655133, + "grad_norm": 11.875, + "kd_loss": 0.1064453125, + "learning_rate": 3.828836978457868e-06, + "loss": 0.1905, + "step": 1765, + "student_loss": 0.1357133686542511, + "teacher_loss": 0.0223146490752697 + }, + { + "epoch": 0.5103806228373703, + "grad_norm": 4.75, + "kd_loss": 0.15625, + "learning_rate": 3.8026153763600603e-06, + "loss": 0.1733, + "step": 1770, + "student_loss": 0.060023948550224304, + "teacher_loss": 0.025619735941290855 + }, + { + "epoch": 0.5118223760092272, + "grad_norm": 5.59375, + "kd_loss": 0.1123046875, + "learning_rate": 3.7764286630916704e-06, + "loss": 0.1628, + "step": 1775, + "student_loss": 0.0012931956443935633, + "teacher_loss": 0.0008822145755402744 + }, + { + "epoch": 0.5132641291810842, + "grad_norm": 10.9375, + "kd_loss": 0.115234375, + "learning_rate": 3.7502776016688234e-06, + "loss": 0.1846, + "step": 1780, + "student_loss": 0.011938425712287426, + "teacher_loss": 0.017920546233654022 + }, + { + "epoch": 0.5147058823529411, + "grad_norm": 4.5, + "kd_loss": 0.1220703125, + "learning_rate": 3.724162954068835e-06, + "loss": 0.1777, + "step": 1785, + "student_loss": 0.16685064136981964, + "teacher_loss": 0.026212001219391823 + }, + { + "epoch": 0.5161476355247981, + "grad_norm": 5.53125, + "kd_loss": 0.09619140625, + "learning_rate": 3.6980854812080097e-06, + "loss": 0.1726, + "step": 1790, + "student_loss": 0.0010834241984412074, + "teacher_loss": 0.0006802008138038218 + }, + { + "epoch": 0.5175893886966552, + "grad_norm": 6.21875, + "kd_loss": 0.0986328125, + "learning_rate": 3.6720459429194743e-06, + "loss": 0.1899, + "step": 1795, + "student_loss": 0.18642298877239227, + "teacher_loss": 0.0003939522139262408 + }, + { + "epoch": 0.5190311418685121, + "grad_norm": 4.8125, + "kd_loss": 0.09326171875, + "learning_rate": 3.646045097931037e-06, + "loss": 0.1788, + "step": 1800, + "student_loss": 0.0009775545913726091, + "teacher_loss": 0.0007754040998406708 + }, + { + "epoch": 0.5204728950403691, + "grad_norm": 6.09375, + "kd_loss": 0.095703125, + "learning_rate": 3.620083703843077e-06, + "loss": 0.1816, + "step": 1805, + "student_loss": 0.018504267558455467, + "teacher_loss": 0.0006798621034249663 + }, + { + "epoch": 0.5219146482122261, + "grad_norm": 8.8125, + "kd_loss": 0.12451171875, + "learning_rate": 3.594162517106472e-06, + "loss": 0.1829, + "step": 1810, + "student_loss": 0.03579748421907425, + "teacher_loss": 0.005469038616865873 + }, + { + "epoch": 0.523356401384083, + "grad_norm": 4.78125, + "kd_loss": 0.1435546875, + "learning_rate": 3.5682822930005567e-06, + "loss": 0.162, + "step": 1815, + "student_loss": 0.1445358544588089, + "teacher_loss": 0.010837113484740257 + }, + { + "epoch": 0.52479815455594, + "grad_norm": 8.6875, + "kd_loss": 0.11279296875, + "learning_rate": 3.542443785611117e-06, + "loss": 0.1693, + "step": 1820, + "student_loss": 0.035146307200193405, + "teacher_loss": 0.002823705319315195 + }, + { + "epoch": 0.526239907727797, + "grad_norm": 6.90625, + "kd_loss": 0.1220703125, + "learning_rate": 3.516647747808417e-06, + "loss": 0.1674, + "step": 1825, + "student_loss": 0.0014678154839202762, + "teacher_loss": 0.003952109720557928 + }, + { + "epoch": 0.527681660899654, + "grad_norm": 8.5625, + "kd_loss": 0.1015625, + "learning_rate": 3.4908949312252593e-06, + "loss": 0.1654, + "step": 1830, + "student_loss": 0.000863194465637207, + "teacher_loss": 0.0006952537223696709 + }, + { + "epoch": 0.529123414071511, + "grad_norm": 6.0, + "kd_loss": 0.11669921875, + "learning_rate": 3.4651860862350893e-06, + "loss": 0.1592, + "step": 1835, + "student_loss": 0.0013549678260460496, + "teacher_loss": 0.0005926712183281779 + }, + { + "epoch": 0.5305651672433679, + "grad_norm": 7.9375, + "kd_loss": 0.09375, + "learning_rate": 3.4395219619301288e-06, + "loss": 0.1727, + "step": 1840, + "student_loss": 0.06812132149934769, + "teacher_loss": 0.00883288960903883 + }, + { + "epoch": 0.5320069204152249, + "grad_norm": 4.8125, + "kd_loss": 0.111328125, + "learning_rate": 3.4139033060995484e-06, + "loss": 0.1899, + "step": 1845, + "student_loss": 0.0008714240393601358, + "teacher_loss": 0.0008420728263445199 + }, + { + "epoch": 0.5334486735870819, + "grad_norm": 4.46875, + "kd_loss": 0.0927734375, + "learning_rate": 3.388330865207681e-06, + "loss": 0.1774, + "step": 1850, + "student_loss": 0.11201676726341248, + "teacher_loss": 0.012406986206769943 + }, + { + "epoch": 0.5348904267589388, + "grad_norm": 6.59375, + "kd_loss": 0.11083984375, + "learning_rate": 3.3628053843722674e-06, + "loss": 0.1887, + "step": 1855, + "student_loss": 0.0014650821685791016, + "teacher_loss": 0.0019600437954068184 + }, + { + "epoch": 0.5363321799307958, + "grad_norm": 7.0625, + "kd_loss": 0.0986328125, + "learning_rate": 3.337327607342753e-06, + "loss": 0.1661, + "step": 1860, + "student_loss": 0.0010325285838916898, + "teacher_loss": 0.0004530692531261593 + }, + { + "epoch": 0.5377739331026529, + "grad_norm": 6.625, + "kd_loss": 0.1240234375, + "learning_rate": 3.3118982764786055e-06, + "loss": 0.1899, + "step": 1865, + "student_loss": 0.20842736959457397, + "teacher_loss": 0.00036168971564620733 + }, + { + "epoch": 0.5392156862745098, + "grad_norm": 10.9375, + "kd_loss": 0.12353515625, + "learning_rate": 3.2865181327277007e-06, + "loss": 0.1776, + "step": 1870, + "student_loss": 0.35005033016204834, + "teacher_loss": 0.000831729150377214 + }, + { + "epoch": 0.5406574394463668, + "grad_norm": 4.4375, + "kd_loss": 0.10107421875, + "learning_rate": 3.2611879156047147e-06, + "loss": 0.1713, + "step": 1875, + "student_loss": 0.0016464912332594395, + "teacher_loss": 0.0013047648826614022 + }, + { + "epoch": 0.5420991926182238, + "grad_norm": 5.40625, + "kd_loss": 0.1181640625, + "learning_rate": 3.2359083631695897e-06, + "loss": 0.1531, + "step": 1880, + "student_loss": 0.0027816772926598787, + "teacher_loss": 0.0013133822940289974 + }, + { + "epoch": 0.5435409457900807, + "grad_norm": 6.09375, + "kd_loss": 0.1123046875, + "learning_rate": 3.2106802120060197e-06, + "loss": 0.1769, + "step": 1885, + "student_loss": 0.009803075343370438, + "teacher_loss": 0.0017775309970602393 + }, + { + "epoch": 0.5449826989619377, + "grad_norm": 3.890625, + "kd_loss": 0.1015625, + "learning_rate": 3.185504197199999e-06, + "loss": 0.1626, + "step": 1890, + "student_loss": 0.018362557515501976, + "teacher_loss": 0.0006177062750793993 + }, + { + "epoch": 0.5464244521337946, + "grad_norm": 5.375, + "kd_loss": 0.11767578125, + "learning_rate": 3.160381052318393e-06, + "loss": 0.1911, + "step": 1895, + "student_loss": 0.08161383122205734, + "teacher_loss": 0.004222067538648844 + }, + { + "epoch": 0.5478662053056517, + "grad_norm": 6.0, + "kd_loss": 0.10107421875, + "learning_rate": 3.1353115093875676e-06, + "loss": 0.192, + "step": 1900, + "student_loss": 0.004832001402974129, + "teacher_loss": 0.0006145957158878446 + }, + { + "epoch": 0.5493079584775087, + "grad_norm": 8.875, + "kd_loss": 0.10400390625, + "learning_rate": 3.1102962988720615e-06, + "loss": 0.1693, + "step": 1905, + "student_loss": 0.2723603844642639, + "teacher_loss": 0.0017128102481365204 + }, + { + "epoch": 0.5507497116493656, + "grad_norm": 5.8125, + "kd_loss": 0.11328125, + "learning_rate": 3.085336149653303e-06, + "loss": 0.1678, + "step": 1910, + "student_loss": 0.0009368823375552893, + "teacher_loss": 0.0007281577563844621 + }, + { + "epoch": 0.5521914648212226, + "grad_norm": 5.25, + "kd_loss": 0.11328125, + "learning_rate": 3.060431789008368e-06, + "loss": 0.1953, + "step": 1915, + "student_loss": 0.001682370901107788, + "teacher_loss": 0.002259776694700122 + }, + { + "epoch": 0.5536332179930796, + "grad_norm": 6.9375, + "kd_loss": 0.09130859375, + "learning_rate": 3.035583942588791e-06, + "loss": 0.1991, + "step": 1920, + "student_loss": 0.06771772354841232, + "teacher_loss": 0.00032804597867652774 + }, + { + "epoch": 0.5550749711649365, + "grad_norm": 3.984375, + "kd_loss": 0.11474609375, + "learning_rate": 3.0107933343994233e-06, + "loss": 0.1757, + "step": 1925, + "student_loss": 0.0011277003213763237, + "teacher_loss": 0.0006808692123740911 + }, + { + "epoch": 0.5565167243367936, + "grad_norm": 6.09375, + "kd_loss": 0.08740234375, + "learning_rate": 2.9860606867773323e-06, + "loss": 0.1636, + "step": 1930, + "student_loss": 0.18559186160564423, + "teacher_loss": 0.00035175459925085306 + }, + { + "epoch": 0.5579584775086506, + "grad_norm": 7.28125, + "kd_loss": 0.10986328125, + "learning_rate": 2.9613867203707627e-06, + "loss": 0.1817, + "step": 1935, + "student_loss": 0.14802882075309753, + "teacher_loss": 0.01818777061998844 + }, + { + "epoch": 0.5594002306805075, + "grad_norm": 4.21875, + "kd_loss": 0.1298828125, + "learning_rate": 2.936772154118129e-06, + "loss": 0.1777, + "step": 1940, + "student_loss": 0.006850242614746094, + "teacher_loss": 0.0013862320920452476 + }, + { + "epoch": 0.5608419838523645, + "grad_norm": 6.09375, + "kd_loss": 0.1162109375, + "learning_rate": 2.912217705227075e-06, + "loss": 0.1679, + "step": 1945, + "student_loss": 0.024072805419564247, + "teacher_loss": 0.027896327897906303 + }, + { + "epoch": 0.5622837370242214, + "grad_norm": 11.8125, + "kd_loss": 0.10986328125, + "learning_rate": 2.88772408915357e-06, + "loss": 0.1811, + "step": 1950, + "student_loss": 0.0006783945136703551, + "teacher_loss": 0.0007104156538844109 + }, + { + "epoch": 0.5637254901960784, + "grad_norm": 5.28125, + "kd_loss": 0.12451171875, + "learning_rate": 2.863292019581071e-06, + "loss": 0.1734, + "step": 1955, + "student_loss": 0.2896282970905304, + "teacher_loss": 0.00030293865711428225 + }, + { + "epoch": 0.5651672433679354, + "grad_norm": 5.9375, + "kd_loss": 0.11181640625, + "learning_rate": 2.838922208399712e-06, + "loss": 0.201, + "step": 1960, + "student_loss": 0.001777703408151865, + "teacher_loss": 0.001076407846994698 + }, + { + "epoch": 0.5666089965397924, + "grad_norm": 4.46875, + "kd_loss": 0.1044921875, + "learning_rate": 2.8146153656855858e-06, + "loss": 0.1773, + "step": 1965, + "student_loss": 0.0009869302157312632, + "teacher_loss": 0.000566221191547811 + }, + { + "epoch": 0.5680507497116494, + "grad_norm": 8.0625, + "kd_loss": 0.126953125, + "learning_rate": 2.7903721996800248e-06, + "loss": 0.1763, + "step": 1970, + "student_loss": 0.01651730388402939, + "teacher_loss": 0.0022809882648289204 + }, + { + "epoch": 0.5694925028835064, + "grad_norm": 5.46875, + "kd_loss": 0.140625, + "learning_rate": 2.7661934167689887e-06, + "loss": 0.1867, + "step": 1975, + "student_loss": 0.000893535849172622, + "teacher_loss": 0.0010718825506046414 + }, + { + "epoch": 0.5709342560553633, + "grad_norm": 4.5625, + "kd_loss": 0.10302734375, + "learning_rate": 2.742079721462471e-06, + "loss": 0.1941, + "step": 1980, + "student_loss": 0.055821552872657776, + "teacher_loss": 0.0015563679626211524 + }, + { + "epoch": 0.5723760092272203, + "grad_norm": 5.375, + "kd_loss": 0.10009765625, + "learning_rate": 2.7180318163739704e-06, + "loss": 0.1815, + "step": 1985, + "student_loss": 0.0019449080573394895, + "teacher_loss": 0.0008286093943752348 + }, + { + "epoch": 0.5738177623990772, + "grad_norm": 5.9375, + "kd_loss": 0.1474609375, + "learning_rate": 2.6940504022000248e-06, + "loss": 0.1757, + "step": 1990, + "student_loss": 0.0654199868440628, + "teacher_loss": 0.015855683013796806 + }, + { + "epoch": 0.5752595155709342, + "grad_norm": 10.125, + "kd_loss": 0.0966796875, + "learning_rate": 2.67013617769979e-06, + "loss": 0.178, + "step": 1995, + "student_loss": 0.0012741348473355174, + "teacher_loss": 0.0009025294566527009 + }, + { + "epoch": 0.5767012687427913, + "grad_norm": 3.640625, + "kd_loss": 0.10205078125, + "learning_rate": 2.6462898396746783e-06, + "loss": 0.1773, + "step": 2000, + "student_loss": 0.0014912512851879, + "teacher_loss": 0.0007635725778527558 + }, + { + "epoch": 0.5781430219146482, + "grad_norm": 4.84375, + "kd_loss": 0.10107421875, + "learning_rate": 2.622512082948063e-06, + "loss": 0.1896, + "step": 2005, + "student_loss": 0.0018232942093163729, + "teacher_loss": 0.0007667585741728544 + }, + { + "epoch": 0.5795847750865052, + "grad_norm": 11.8125, + "kd_loss": 0.1044921875, + "learning_rate": 2.598803600345021e-06, + "loss": 0.1879, + "step": 2010, + "student_loss": 0.01782207190990448, + "teacher_loss": 0.0021818995010107756 + }, + { + "epoch": 0.5810265282583622, + "grad_norm": 6.71875, + "kd_loss": 0.1064453125, + "learning_rate": 2.5751650826721584e-06, + "loss": 0.1743, + "step": 2015, + "student_loss": 0.04955288767814636, + "teacher_loss": 0.0005823059473186731 + }, + { + "epoch": 0.5824682814302191, + "grad_norm": 7.40625, + "kd_loss": 0.12109375, + "learning_rate": 2.551597218697476e-06, + "loss": 0.1657, + "step": 2020, + "student_loss": 0.47541993856430054, + "teacher_loss": 0.0016289220657199621 + }, + { + "epoch": 0.5839100346020761, + "grad_norm": 6.3125, + "kd_loss": 0.09765625, + "learning_rate": 2.5281006951302932e-06, + "loss": 0.1757, + "step": 2025, + "student_loss": 0.0066398135386407375, + "teacher_loss": 0.0004615852376446128 + }, + { + "epoch": 0.5853517877739332, + "grad_norm": 5.71875, + "kd_loss": 0.109375, + "learning_rate": 2.504676196601253e-06, + "loss": 0.1751, + "step": 2030, + "student_loss": 0.003715298604220152, + "teacher_loss": 0.0005154022946953773 + }, + { + "epoch": 0.5867935409457901, + "grad_norm": 7.25, + "kd_loss": 0.09716796875, + "learning_rate": 2.4813244056423692e-06, + "loss": 0.1758, + "step": 2035, + "student_loss": 0.0023021947126835585, + "teacher_loss": 0.0005425055860541761 + }, + { + "epoch": 0.5882352941176471, + "grad_norm": 5.4375, + "kd_loss": 0.1376953125, + "learning_rate": 2.4580460026671268e-06, + "loss": 0.1862, + "step": 2040, + "student_loss": 0.06547769904136658, + "teacher_loss": 0.012091674841940403 + }, + { + "epoch": 0.589677047289504, + "grad_norm": 6.15625, + "kd_loss": 0.11279296875, + "learning_rate": 2.4348416659506753e-06, + "loss": 0.1694, + "step": 2045, + "student_loss": 0.0671345591545105, + "teacher_loss": 0.0055143567733466625 + }, + { + "epoch": 0.591118800461361, + "grad_norm": 4.84375, + "kd_loss": 0.134765625, + "learning_rate": 2.4117120716100484e-06, + "loss": 0.174, + "step": 2050, + "student_loss": 0.0015601585619151592, + "teacher_loss": 0.01685117743909359 + }, + { + "epoch": 0.592560553633218, + "grad_norm": 6.34375, + "kd_loss": 0.107421875, + "learning_rate": 2.388657893584484e-06, + "loss": 0.1698, + "step": 2055, + "student_loss": 0.03631133586168289, + "teacher_loss": 0.0006107162334956229 + }, + { + "epoch": 0.5940023068050749, + "grad_norm": 4.9375, + "kd_loss": 0.1240234375, + "learning_rate": 2.365679803615761e-06, + "loss": 0.1729, + "step": 2060, + "student_loss": 0.0015836743405088782, + "teacher_loss": 0.0006585841765627265 + }, + { + "epoch": 0.595444059976932, + "grad_norm": 5.5, + "kd_loss": 0.1376953125, + "learning_rate": 2.342778471228648e-06, + "loss": 0.1936, + "step": 2065, + "student_loss": 0.0012043609749525785, + "teacher_loss": 0.011278706602752209 + }, + { + "epoch": 0.596885813148789, + "grad_norm": 6.3125, + "kd_loss": 0.12158203125, + "learning_rate": 2.3199545637113885e-06, + "loss": 0.1749, + "step": 2070, + "student_loss": 0.1277606338262558, + "teacher_loss": 0.00117887940723449 + }, + { + "epoch": 0.5983275663206459, + "grad_norm": 5.4375, + "kd_loss": 0.08837890625, + "learning_rate": 2.29720874609625e-06, + "loss": 0.1697, + "step": 2075, + "student_loss": 0.0008703434141352773, + "teacher_loss": 0.0050735012628138065 + }, + { + "epoch": 0.5997693194925029, + "grad_norm": 5.375, + "kd_loss": 0.119140625, + "learning_rate": 2.274541681140159e-06, + "loss": 0.1859, + "step": 2080, + "student_loss": 0.003949797246605158, + "teacher_loss": 0.00498243747279048 + }, + { + "epoch": 0.6012110726643599, + "grad_norm": 4.25, + "kd_loss": 0.115234375, + "learning_rate": 2.251954029305384e-06, + "loss": 0.1762, + "step": 2085, + "student_loss": 0.046131033450365067, + "teacher_loss": 0.00881399866193533 + }, + { + "epoch": 0.6026528258362168, + "grad_norm": 5.09375, + "kd_loss": 0.12890625, + "learning_rate": 2.2294464487402858e-06, + "loss": 0.1766, + "step": 2090, + "student_loss": 0.002719937125220895, + "teacher_loss": 0.02367260307073593 + }, + { + "epoch": 0.6040945790080738, + "grad_norm": 6.0625, + "kd_loss": 0.1005859375, + "learning_rate": 2.207019595260154e-06, + "loss": 0.1767, + "step": 2095, + "student_loss": 0.07975543290376663, + "teacher_loss": 0.000686533167026937 + }, + { + "epoch": 0.6055363321799307, + "grad_norm": 4.625, + "kd_loss": 0.1025390625, + "learning_rate": 2.1846741223280825e-06, + "loss": 0.1924, + "step": 2100, + "student_loss": 0.2042984813451767, + "teacher_loss": 0.02697836607694626 + }, + { + "epoch": 0.6069780853517878, + "grad_norm": 4.25, + "kd_loss": 0.09716796875, + "learning_rate": 2.1624106810359433e-06, + "loss": 0.1683, + "step": 2105, + "student_loss": 0.000995562644675374, + "teacher_loss": 0.0007496543694287539 + }, + { + "epoch": 0.6084198385236448, + "grad_norm": 4.34375, + "kd_loss": 0.11865234375, + "learning_rate": 2.140229920085409e-06, + "loss": 0.1943, + "step": 2110, + "student_loss": 0.10376831144094467, + "teacher_loss": 0.00259330659173429 + }, + { + "epoch": 0.6098615916955017, + "grad_norm": 6.34375, + "kd_loss": 0.09619140625, + "learning_rate": 2.1181324857690446e-06, + "loss": 0.1715, + "step": 2115, + "student_loss": 0.05130111053586006, + "teacher_loss": 0.0007214110228233039 + }, + { + "epoch": 0.6113033448673587, + "grad_norm": 3.84375, + "kd_loss": 0.1708984375, + "learning_rate": 2.096119021951491e-06, + "loss": 0.155, + "step": 2120, + "student_loss": 0.008221042342483997, + "teacher_loss": 0.0004744199104607105 + }, + { + "epoch": 0.6127450980392157, + "grad_norm": 6.625, + "kd_loss": 0.09912109375, + "learning_rate": 2.07419017005069e-06, + "loss": 0.1724, + "step": 2125, + "student_loss": 0.02029389515519142, + "teacher_loss": 0.0004736967384815216 + }, + { + "epoch": 0.6141868512110726, + "grad_norm": 5.46875, + "kd_loss": 0.10009765625, + "learning_rate": 2.0523465690192053e-06, + "loss": 0.168, + "step": 2130, + "student_loss": 0.0011598586570471525, + "teacher_loss": 0.00036264763912186027 + }, + { + "epoch": 0.6156286043829297, + "grad_norm": 6.15625, + "kd_loss": 0.10009765625, + "learning_rate": 2.0305888553256007e-06, + "loss": 0.2084, + "step": 2135, + "student_loss": 0.1590556800365448, + "teacher_loss": 0.0005438706139102578 + }, + { + "epoch": 0.6170703575547867, + "grad_norm": 4.875, + "kd_loss": 0.10693359375, + "learning_rate": 2.0089176629358904e-06, + "loss": 0.169, + "step": 2140, + "student_loss": 0.00457007298246026, + "teacher_loss": 0.0008019876549951732 + }, + { + "epoch": 0.6185121107266436, + "grad_norm": 6.21875, + "kd_loss": 0.103515625, + "learning_rate": 1.98733362329508e-06, + "loss": 0.1879, + "step": 2145, + "student_loss": 0.0019138423958793283, + "teacher_loss": 0.0007465453236363828 + }, + { + "epoch": 0.6199538638985006, + "grad_norm": 4.21875, + "kd_loss": 0.11083984375, + "learning_rate": 1.965837365308751e-06, + "loss": 0.1651, + "step": 2150, + "student_loss": 0.001106220530346036, + "teacher_loss": 0.010991056449711323 + }, + { + "epoch": 0.6213956170703575, + "grad_norm": 7.4375, + "kd_loss": 0.10107421875, + "learning_rate": 1.944429515324749e-06, + "loss": 0.1724, + "step": 2155, + "student_loss": 0.0006128507084213197, + "teacher_loss": 0.0003564523358363658 + }, + { + "epoch": 0.6228373702422145, + "grad_norm": 7.34375, + "kd_loss": 0.1083984375, + "learning_rate": 1.923110697114929e-06, + "loss": 0.1847, + "step": 2160, + "student_loss": 0.0005715858424082398, + "teacher_loss": 0.00028826954076066613 + }, + { + "epoch": 0.6242791234140715, + "grad_norm": 6.0, + "kd_loss": 0.103515625, + "learning_rate": 1.9018815318569756e-06, + "loss": 0.1812, + "step": 2165, + "student_loss": 0.08555524796247482, + "teacher_loss": 0.0004269266501069069 + }, + { + "epoch": 0.6257208765859285, + "grad_norm": 5.9375, + "kd_loss": 0.0986328125, + "learning_rate": 1.8807426381163151e-06, + "loss": 0.1848, + "step": 2170, + "student_loss": 0.0015966829378157854, + "teacher_loss": 0.0010793735273182392 + }, + { + "epoch": 0.6271626297577855, + "grad_norm": 5.4375, + "kd_loss": 0.11181640625, + "learning_rate": 1.8596946318280773e-06, + "loss": 0.1578, + "step": 2175, + "student_loss": 0.004371183924376965, + "teacher_loss": 0.0004928941489197314 + }, + { + "epoch": 0.6286043829296425, + "grad_norm": 4.75, + "kd_loss": 0.11328125, + "learning_rate": 1.838738126279161e-06, + "loss": 0.1683, + "step": 2180, + "student_loss": 0.0009796395897865295, + "teacher_loss": 0.00031363972811959684 + }, + { + "epoch": 0.6300461361014994, + "grad_norm": 5.875, + "kd_loss": 0.1396484375, + "learning_rate": 1.81787373209036e-06, + "loss": 0.1646, + "step": 2185, + "student_loss": 0.00304932938888669, + "teacher_loss": 0.032131947576999664 + }, + { + "epoch": 0.6314878892733564, + "grad_norm": 4.65625, + "kd_loss": 0.1005859375, + "learning_rate": 1.7971020571985664e-06, + "loss": 0.1607, + "step": 2190, + "student_loss": 0.013054043985903263, + "teacher_loss": 0.000551006116438657 + }, + { + "epoch": 0.6329296424452133, + "grad_norm": 7.125, + "kd_loss": 0.1005859375, + "learning_rate": 1.7764237068390643e-06, + "loss": 0.1795, + "step": 2195, + "student_loss": 0.09780588001012802, + "teacher_loss": 0.001486110733821988 + }, + { + "epoch": 0.6343713956170703, + "grad_norm": 3.953125, + "kd_loss": 0.1982421875, + "learning_rate": 1.755839283527893e-06, + "loss": 0.1655, + "step": 2200, + "student_loss": 0.0023929367307573557, + "teacher_loss": 0.004217212088406086 + }, + { + "epoch": 0.6358131487889274, + "grad_norm": 6.625, + "kd_loss": 0.166015625, + "learning_rate": 1.7353493870442844e-06, + "loss": 0.1835, + "step": 2205, + "student_loss": 0.14643274247646332, + "teacher_loss": 0.009425546042621136 + }, + { + "epoch": 0.6372549019607843, + "grad_norm": 5.375, + "kd_loss": 0.1328125, + "learning_rate": 1.714954614413198e-06, + "loss": 0.1897, + "step": 2210, + "student_loss": 0.02522740140557289, + "teacher_loss": 0.057045795023441315 + }, + { + "epoch": 0.6386966551326413, + "grad_norm": 4.96875, + "kd_loss": 0.140625, + "learning_rate": 1.6946555598879138e-06, + "loss": 0.1954, + "step": 2215, + "student_loss": 0.004027427174150944, + "teacher_loss": 0.0008342525688931346 + }, + { + "epoch": 0.6401384083044983, + "grad_norm": 3.5, + "kd_loss": 0.1171875, + "learning_rate": 1.6744528149327272e-06, + "loss": 0.1604, + "step": 2220, + "student_loss": 0.0016696155071258545, + "teacher_loss": 0.0018177750753238797 + }, + { + "epoch": 0.6415801614763552, + "grad_norm": 7.09375, + "kd_loss": 0.0947265625, + "learning_rate": 1.6543469682057105e-06, + "loss": 0.1548, + "step": 2225, + "student_loss": 0.004229382611811161, + "teacher_loss": 0.001329874387010932 + }, + { + "epoch": 0.6430219146482122, + "grad_norm": 4.5625, + "kd_loss": 0.1748046875, + "learning_rate": 1.6343386055415545e-06, + "loss": 0.1816, + "step": 2230, + "student_loss": 0.0028775471728295088, + "teacher_loss": 0.00235175178386271 + }, + { + "epoch": 0.6444636678200693, + "grad_norm": 6.25, + "kd_loss": 0.1064453125, + "learning_rate": 1.614428309934517e-06, + "loss": 0.1817, + "step": 2235, + "student_loss": 0.0013961255317553878, + "teacher_loss": 0.0007145421695895493 + }, + { + "epoch": 0.6459054209919262, + "grad_norm": 5.84375, + "kd_loss": 0.10302734375, + "learning_rate": 1.594616661521412e-06, + "loss": 0.1726, + "step": 2240, + "student_loss": 0.0009795799851417542, + "teacher_loss": 0.0012523789191618562 + }, + { + "epoch": 0.6473471741637832, + "grad_norm": 6.0625, + "kd_loss": 0.1123046875, + "learning_rate": 1.5749042375647261e-06, + "loss": 0.171, + "step": 2245, + "student_loss": 0.1344611793756485, + "teacher_loss": 0.0023756572045385838 + }, + { + "epoch": 0.6487889273356401, + "grad_norm": 6.84375, + "kd_loss": 0.10986328125, + "learning_rate": 1.5552916124357887e-06, + "loss": 0.1656, + "step": 2250, + "student_loss": 0.005796133074909449, + "teacher_loss": 0.0133487768471241 + }, + { + "epoch": 0.6502306805074971, + "grad_norm": 4.40625, + "kd_loss": 0.10986328125, + "learning_rate": 1.535779357598033e-06, + "loss": 0.186, + "step": 2255, + "student_loss": 0.007919806987047195, + "teacher_loss": 0.0009406052995473146 + }, + { + "epoch": 0.6516724336793541, + "grad_norm": 6.0, + "kd_loss": 0.09521484375, + "learning_rate": 1.516368041590358e-06, + "loss": 0.1706, + "step": 2260, + "student_loss": 0.0005208402872085571, + "teacher_loss": 0.0006157188327051699 + }, + { + "epoch": 0.653114186851211, + "grad_norm": 8.6875, + "kd_loss": 0.11572265625, + "learning_rate": 1.4970582300105467e-06, + "loss": 0.2178, + "step": 2265, + "student_loss": 0.49083903431892395, + "teacher_loss": 0.039741333574056625 + }, + { + "epoch": 0.654555940023068, + "grad_norm": 4.9375, + "kd_loss": 0.1103515625, + "learning_rate": 1.4778504854988003e-06, + "loss": 0.1796, + "step": 2270, + "student_loss": 0.0492987222969532, + "teacher_loss": 0.007138086948543787 + }, + { + "epoch": 0.6559976931949251, + "grad_norm": 6.8125, + "kd_loss": 0.1513671875, + "learning_rate": 1.4587453677213348e-06, + "loss": 0.176, + "step": 2275, + "student_loss": 0.002270741853863001, + "teacher_loss": 0.001979748485609889 + }, + { + "epoch": 0.657439446366782, + "grad_norm": 8.375, + "kd_loss": 0.11328125, + "learning_rate": 1.4397434333540755e-06, + "loss": 0.1915, + "step": 2280, + "student_loss": 0.2564370930194855, + "teacher_loss": 0.0026648673228919506 + }, + { + "epoch": 0.658881199538639, + "grad_norm": 8.875, + "kd_loss": 0.111328125, + "learning_rate": 1.420845236066441e-06, + "loss": 0.1967, + "step": 2285, + "student_loss": 0.03857283666729927, + "teacher_loss": 0.00034623852116055787 + }, + { + "epoch": 0.660322952710496, + "grad_norm": 3.265625, + "kd_loss": 0.0947265625, + "learning_rate": 1.4020513265052072e-06, + "loss": 0.182, + "step": 2290, + "student_loss": 0.007523265201598406, + "teacher_loss": 0.000504697731230408 + }, + { + "epoch": 0.6617647058823529, + "grad_norm": 7.78125, + "kd_loss": 0.1181640625, + "learning_rate": 1.3833622522784596e-06, + "loss": 0.1994, + "step": 2295, + "student_loss": 0.07745266705751419, + "teacher_loss": 0.0062800124287605286 + }, + { + "epoch": 0.6632064590542099, + "grad_norm": 6.78125, + "kd_loss": 0.1220703125, + "learning_rate": 1.3647785579396472e-06, + "loss": 0.1924, + "step": 2300, + "student_loss": 0.0019856293220072985, + "teacher_loss": 0.0006964774220250547 + }, + { + "epoch": 0.6646482122260668, + "grad_norm": 4.90625, + "kd_loss": 0.1083984375, + "learning_rate": 1.3463007849717035e-06, + "loss": 0.1605, + "step": 2305, + "student_loss": 0.023480596020817757, + "teacher_loss": 0.0013806336792185903 + }, + { + "epoch": 0.6660899653979239, + "grad_norm": 8.9375, + "kd_loss": 0.10888671875, + "learning_rate": 1.3279294717712787e-06, + "loss": 0.1938, + "step": 2310, + "student_loss": 0.13922151923179626, + "teacher_loss": 0.0086038988083601 + }, + { + "epoch": 0.6675317185697809, + "grad_norm": 5.1875, + "kd_loss": 0.1298828125, + "learning_rate": 1.309665153633049e-06, + "loss": 0.1708, + "step": 2315, + "student_loss": 0.09018564224243164, + "teacher_loss": 0.0006173376459628344 + }, + { + "epoch": 0.6689734717416378, + "grad_norm": 5.8125, + "kd_loss": 0.1259765625, + "learning_rate": 1.291508362734113e-06, + "loss": 0.1898, + "step": 2320, + "student_loss": 0.10888887941837311, + "teacher_loss": 0.0023176076356321573 + }, + { + "epoch": 0.6704152249134948, + "grad_norm": 4.34375, + "kd_loss": 0.11865234375, + "learning_rate": 1.2734596281185012e-06, + "loss": 0.1736, + "step": 2325, + "student_loss": 0.0012785395374521613, + "teacher_loss": 0.0016392101533710957 + }, + { + "epoch": 0.6718569780853518, + "grad_norm": 5.40625, + "kd_loss": 0.11376953125, + "learning_rate": 1.2555194756817407e-06, + "loss": 0.196, + "step": 2330, + "student_loss": 0.0013906876556575298, + "teacher_loss": 0.0013734942767769098 + }, + { + "epoch": 0.6732987312572087, + "grad_norm": 5.375, + "kd_loss": 0.1728515625, + "learning_rate": 1.2376884281555485e-06, + "loss": 0.1746, + "step": 2335, + "student_loss": 0.19898872077465057, + "teacher_loss": 0.00759601965546608 + }, + { + "epoch": 0.6747404844290658, + "grad_norm": 6.9375, + "kd_loss": 0.1455078125, + "learning_rate": 1.2199670050925932e-06, + "loss": 0.1775, + "step": 2340, + "student_loss": 0.18600302934646606, + "teacher_loss": 0.011311961337924004 + }, + { + "epoch": 0.6761822376009228, + "grad_norm": 6.375, + "kd_loss": 0.1044921875, + "learning_rate": 1.2023557228513543e-06, + "loss": 0.1793, + "step": 2345, + "student_loss": 0.11181621998548508, + "teacher_loss": 0.0005607152124866843 + }, + { + "epoch": 0.6776239907727797, + "grad_norm": 8.0625, + "kd_loss": 0.11181640625, + "learning_rate": 1.184855094581085e-06, + "loss": 0.1781, + "step": 2350, + "student_loss": 0.0018905261531472206, + "teacher_loss": 0.00048104205052368343 + }, + { + "epoch": 0.6790657439446367, + "grad_norm": 4.96875, + "kd_loss": 0.1259765625, + "learning_rate": 1.1674656302068487e-06, + "loss": 0.1868, + "step": 2355, + "student_loss": 0.012083466164767742, + "teacher_loss": 0.0036780540831387043 + }, + { + "epoch": 0.6805074971164936, + "grad_norm": 7.65625, + "kd_loss": 0.109375, + "learning_rate": 1.1501878364146734e-06, + "loss": 0.1719, + "step": 2360, + "student_loss": 0.001730799674987793, + "teacher_loss": 0.0004727965861093253 + }, + { + "epoch": 0.6819492502883506, + "grad_norm": 7.09375, + "kd_loss": 0.09765625, + "learning_rate": 1.133022216636781e-06, + "loss": 0.1644, + "step": 2365, + "student_loss": 0.010835912078619003, + "teacher_loss": 0.004248704761266708 + }, + { + "epoch": 0.6833910034602076, + "grad_norm": 8.9375, + "kd_loss": 0.10107421875, + "learning_rate": 1.1159692710369162e-06, + "loss": 0.178, + "step": 2370, + "student_loss": 0.07398489862680435, + "teacher_loss": 0.0004782878968399018 + }, + { + "epoch": 0.6848327566320646, + "grad_norm": 8.875, + "kd_loss": 0.11328125, + "learning_rate": 1.099029496495782e-06, + "loss": 0.1636, + "step": 2375, + "student_loss": 0.0014891972532495856, + "teacher_loss": 0.0017867174465209246 + }, + { + "epoch": 0.6862745098039216, + "grad_norm": 4.5625, + "kd_loss": 0.12255859375, + "learning_rate": 1.0822033865965503e-06, + "loss": 0.185, + "step": 2380, + "student_loss": 0.01883968897163868, + "teacher_loss": 0.0008310509147122502 + }, + { + "epoch": 0.6877162629757786, + "grad_norm": 8.5625, + "kd_loss": 0.103515625, + "learning_rate": 1.0654914316104913e-06, + "loss": 0.1854, + "step": 2385, + "student_loss": 0.0011299190809950233, + "teacher_loss": 0.0011506746523082256 + }, + { + "epoch": 0.6891580161476355, + "grad_norm": 7.21875, + "kd_loss": 0.10302734375, + "learning_rate": 1.0488941184826822e-06, + "loss": 0.1832, + "step": 2390, + "student_loss": 0.003757859580218792, + "teacher_loss": 0.0005284099606797099 + }, + { + "epoch": 0.6905997693194925, + "grad_norm": 4.8125, + "kd_loss": 0.1279296875, + "learning_rate": 1.0324119308178166e-06, + "loss": 0.1767, + "step": 2395, + "student_loss": 0.17005331814289093, + "teacher_loss": 0.0003893845423590392 + }, + { + "epoch": 0.6920415224913494, + "grad_norm": 5.25, + "kd_loss": 0.12109375, + "learning_rate": 1.0160453488661188e-06, + "loss": 0.1682, + "step": 2400, + "student_loss": 0.0006177967297844589, + "teacher_loss": 0.02288214862346649 + }, + { + "epoch": 0.6934832756632064, + "grad_norm": 5.0, + "kd_loss": 0.10888671875, + "learning_rate": 9.997948495093501e-07, + "loss": 0.1854, + "step": 2405, + "student_loss": 0.0014994695084169507, + "teacher_loss": 0.0013032156275585294 + }, + { + "epoch": 0.6949250288350635, + "grad_norm": 4.53125, + "kd_loss": 0.173828125, + "learning_rate": 9.836609062469066e-07, + "loss": 0.1746, + "step": 2410, + "student_loss": 0.018328603357076645, + "teacher_loss": 0.0008572032093070447 + }, + { + "epoch": 0.6963667820069204, + "grad_norm": 5.09375, + "kd_loss": 0.1064453125, + "learning_rate": 9.676439891820349e-07, + "loss": 0.1639, + "step": 2415, + "student_loss": 0.0007460613851435483, + "teacher_loss": 0.0011870411690324545 + }, + { + "epoch": 0.6978085351787774, + "grad_norm": 5.84375, + "kd_loss": 0.111328125, + "learning_rate": 9.517445650081203e-07, + "loss": 0.1828, + "step": 2420, + "student_loss": 0.007265963591635227, + "teacher_loss": 0.0006494870176538825 + }, + { + "epoch": 0.6992502883506344, + "grad_norm": 6.25, + "kd_loss": 0.107421875, + "learning_rate": 9.359630969951012e-07, + "loss": 0.1738, + "step": 2425, + "student_loss": 0.035345472395420074, + "teacher_loss": 0.0036039056722074747 + }, + { + "epoch": 0.7006920415224913, + "grad_norm": 4.5, + "kd_loss": 0.10986328125, + "learning_rate": 9.203000449759603e-07, + "loss": 0.203, + "step": 2430, + "student_loss": 0.005081401206552982, + "teacher_loss": 0.0009590298868715763 + }, + { + "epoch": 0.7021337946943483, + "grad_norm": 5.875, + "kd_loss": 0.09716796875, + "learning_rate": 9.047558653333333e-07, + "loss": 0.1761, + "step": 2435, + "student_loss": 0.0011679220478981733, + "teacher_loss": 0.0003915524866897613 + }, + { + "epoch": 0.7035755478662054, + "grad_norm": 5.90625, + "kd_loss": 0.1259765625, + "learning_rate": 8.893310109862102e-07, + "loss": 0.187, + "step": 2440, + "student_loss": 0.2089531272649765, + "teacher_loss": 0.002312681172043085 + }, + { + "epoch": 0.7050173010380623, + "grad_norm": 5.5625, + "kd_loss": 0.107421875, + "learning_rate": 8.740259313767318e-07, + "loss": 0.1734, + "step": 2445, + "student_loss": 0.2837088704109192, + "teacher_loss": 0.02171768806874752 + }, + { + "epoch": 0.7064590542099193, + "grad_norm": 5.71875, + "kd_loss": 0.109375, + "learning_rate": 8.588410724571039e-07, + "loss": 0.1644, + "step": 2450, + "student_loss": 0.10129602253437042, + "teacher_loss": 0.031316280364990234 + }, + { + "epoch": 0.7079008073817762, + "grad_norm": 5.84375, + "kd_loss": 0.1015625, + "learning_rate": 8.437768766765975e-07, + "loss": 0.178, + "step": 2455, + "student_loss": 0.0011241373140364885, + "teacher_loss": 0.0011174725368618965 + }, + { + "epoch": 0.7093425605536332, + "grad_norm": 5.84375, + "kd_loss": 0.12109375, + "learning_rate": 8.288337829686533e-07, + "loss": 0.1842, + "step": 2460, + "student_loss": 0.0021000599954277277, + "teacher_loss": 0.00038258079439401627 + }, + { + "epoch": 0.7107843137254902, + "grad_norm": 11.4375, + "kd_loss": 0.130859375, + "learning_rate": 8.14012226738104e-07, + "loss": 0.1823, + "step": 2465, + "student_loss": 0.23547177016735077, + "teacher_loss": 0.0010998636716976762 + }, + { + "epoch": 0.7122260668973471, + "grad_norm": 6.03125, + "kd_loss": 0.1171875, + "learning_rate": 7.993126398484741e-07, + "loss": 0.174, + "step": 2470, + "student_loss": 0.0014058814849704504, + "teacher_loss": 0.000562115223146975 + }, + { + "epoch": 0.7136678200692042, + "grad_norm": 5.59375, + "kd_loss": 0.12255859375, + "learning_rate": 7.847354506094068e-07, + "loss": 0.1694, + "step": 2475, + "student_loss": 0.06077536195516586, + "teacher_loss": 0.0004893479053862393 + }, + { + "epoch": 0.7151095732410612, + "grad_norm": 6.5625, + "kd_loss": 0.1220703125, + "learning_rate": 7.702810837641806e-07, + "loss": 0.1719, + "step": 2480, + "student_loss": 0.14460773766040802, + "teacher_loss": 0.000507499382365495 + }, + { + "epoch": 0.7165513264129181, + "grad_norm": 4.40625, + "kd_loss": 0.11279296875, + "learning_rate": 7.55949960477328e-07, + "loss": 0.1869, + "step": 2485, + "student_loss": 0.04027507081627846, + "teacher_loss": 0.0011340543860569596 + }, + { + "epoch": 0.7179930795847751, + "grad_norm": 3.921875, + "kd_loss": 0.0986328125, + "learning_rate": 7.417424983223742e-07, + "loss": 0.1618, + "step": 2490, + "student_loss": 0.0015134039567783475, + "teacher_loss": 0.0009958152659237385 + }, + { + "epoch": 0.7194348327566321, + "grad_norm": 5.25, + "kd_loss": 0.10986328125, + "learning_rate": 7.276591112696629e-07, + "loss": 0.1778, + "step": 2495, + "student_loss": 0.0006103800842538476, + "teacher_loss": 0.0004934420576319098 + }, + { + "epoch": 0.720876585928489, + "grad_norm": 16.875, + "kd_loss": 0.1337890625, + "learning_rate": 7.13700209674294e-07, + "loss": 0.1681, + "step": 2500, + "student_loss": 0.036151472479104996, + "teacher_loss": 0.0004605815338436514 + } + ], + "logging_steps": 5, + "max_steps": 3000, + "num_input_tokens_seen": 0, + "num_train_epochs": 1, + "save_steps": 500, + "stateful_callbacks": { + "TrainerControl": { + "args": { + "should_epoch_stop": false, + "should_evaluate": false, + "should_log": false, + "should_save": true, + "should_training_stop": false + }, + "attributes": {} + } + }, + "total_flos": 0.0, + "train_batch_size": 1, + "trial_name": null, + "trial_params": null +}