diff --git "a/checkpoint-18000/trainer_state.json" "b/checkpoint-18000/trainer_state.json" new file mode 100644--- /dev/null +++ "b/checkpoint-18000/trainer_state.json" @@ -0,0 +1,26854 @@ +{ + "best_global_step": 18000, + "best_metric": 2.546471447625592, + "best_model_checkpoint": "/gpfs/scratch/guoh/DNAFM/output/gencode_human_12.8k_12800/Mamba2_12.8K-100B/checkpoint-18000", + "epoch": 5.104957095241472, + "eval_steps": 100, + "global_step": 18000, + "is_hyper_param_search": false, + "is_local_process_zero": true, + "is_world_process_zero": true, + "log_history": [ + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 0.0028366782497695198, + "grad_norm": 1009.887451171875, + "loss": 320.7536, + "loss_ce": 118.95108795166016, + "loss_region": 0.0, + "loss_total": 118.95108795166016, + "lr": 1.8e-05, + "step": 10, + "tokens_trained": 0.03276544 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 0.0056733564995390395, + "grad_norm": 563.448486328125, + "loss": 30.7442, + "loss_ce": 10.381930351257324, + "loss_region": 0.0, + "loss_total": 10.381930351257324, + "lr": 3.8e-05, + "step": 20, + "tokens_trained": 0.06553088 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 0.00851003474930856, + "grad_norm": 675.3902587890625, + "loss": 9.8277, + "loss_ce": 13.00577449798584, + "loss_region": 0.0, + "loss_total": 13.00577449798584, + "lr": 5.800000000000001e-05, + "step": 30, + "tokens_trained": 0.09829632 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 0.011346712999078079, + "grad_norm": 602.3176879882812, + "loss": 13.6392, + "loss_ce": 11.880974769592285, + "loss_region": 0.0, + "loss_total": 11.880974769592285, + "lr": 7.8e-05, + "step": 40, + "tokens_trained": 0.13106176 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 0.014183391248847599, + "grad_norm": 168.45298767089844, + "loss": 9.743, + "loss_ce": 6.967355728149414, + "loss_region": 0.0, + "loss_total": 6.967355728149414, + "lr": 9.800000000000001e-05, + "step": 50, + "tokens_trained": 0.1638272 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 0.01702006949861712, + "grad_norm": 465.28070068359375, + "loss": 7.4841, + "loss_ce": 12.106856346130371, + "loss_region": 0.0, + "loss_total": 12.106856346130371, + "lr": 0.000118, + "step": 60, + "tokens_trained": 0.19659264 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 0.01985674774838664, + "grad_norm": 200.15444946289062, + "loss": 11.1839, + "loss_ce": 8.123963356018066, + "loss_region": 0.0, + "loss_total": 8.123963356018066, + "lr": 0.00013800000000000002, + "step": 70, + "tokens_trained": 0.22935808 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 0.022693425998156158, + "grad_norm": 303.4104309082031, + "loss": 11.8333, + "loss_ce": 8.580060958862305, + "loss_region": 0.0, + "loss_total": 8.580060958862305, + "lr": 0.000158, + "step": 80, + "tokens_trained": 0.26212192 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 0.025530104247925678, + "grad_norm": 334.0144348144531, + "loss": 15.386, + "loss_ce": 19.212779998779297, + "loss_region": 0.0, + "loss_total": 19.212779998779297, + "lr": 0.000178, + "step": 90, + "tokens_trained": 0.29488736 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 0.028366782497695198, + "grad_norm": 263.13104248046875, + "loss": 8.7856, + "loss_ce": 7.080021858215332, + "loss_region": 0.0, + "loss_total": 7.080021858215332, + "lr": 0.00019800000000000002, + "step": 100, + "tokens_trained": 0.3276528 + }, + { + "epoch": 0.028366782497695198, + "eval_ppl": 1228.7431785422737, + "eval_runtime": 1.4284, + "step": 100, + "tokens_trained": 0.3276528 + }, + { + "epoch": 0.028366782497695198, + "eval_ppl_cds": 3054.4316003279096, + "eval_ppl_dig": 522.5242173419364, + "eval_ppl_exon": 1207.22500677202, + "eval_ppl_intron": 995.6675256725446, + "eval_ppl_nig": 885.154185341024, + "eval_ppl_promoter": 2890.016614783514, + "eval_ppl_utr": 1227.2341056992152, + "step": 100, + "tokens_trained": 0.3276528 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 0.031203460747464717, + "grad_norm": 240.72030639648438, + "loss": 8.6465, + "loss_ce": 8.488709449768066, + "loss_region": 0.0, + "loss_total": 8.488709449768066, + "lr": 0.000218, + "step": 110, + "tokens_trained": 0.36041744 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 0.03404013899723424, + "grad_norm": 149.81283569335938, + "loss": 8.669, + "loss_ce": 6.995543956756592, + "loss_region": 0.0, + "loss_total": 6.995543956756592, + "lr": 0.00023799999999999998, + "step": 120, + "tokens_trained": 0.39318128 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 0.03687681724700376, + "grad_norm": 159.6621856689453, + "loss": 12.7164, + "loss_ce": 6.045105457305908, + "loss_region": 0.0, + "loss_total": 6.045105457305908, + "lr": 0.00025800000000000004, + "step": 130, + "tokens_trained": 0.42594672 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 0.03971349549677328, + "grad_norm": 283.01708984375, + "loss": 19.5001, + "loss_ce": 24.927230834960938, + "loss_region": 0.0, + "loss_total": 24.927230834960938, + "lr": 0.00027800000000000004, + "step": 140, + "tokens_trained": 0.458709112 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 0.042550173746542796, + "grad_norm": 142.15029907226562, + "loss": 13.6371, + "loss_ce": 12.840524673461914, + "loss_region": 0.0, + "loss_total": 12.840524673461914, + "lr": 0.000298, + "step": 150, + "tokens_trained": 0.491469992 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 0.045386851996312316, + "grad_norm": 169.8223114013672, + "loss": 10.7245, + "loss_ce": 12.956764221191406, + "loss_region": 0.0, + "loss_total": 12.956764221191406, + "lr": 0.00031800000000000003, + "step": 160, + "tokens_trained": 0.524234632 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 0.048223530246081836, + "grad_norm": 152.25840759277344, + "loss": 7.9437, + "loss_ce": 8.36349105834961, + "loss_region": 0.0, + "loss_total": 8.36349105834961, + "lr": 0.00033800000000000003, + "step": 170, + "tokens_trained": 0.556999272 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 0.051060208495851356, + "grad_norm": 152.3266143798828, + "loss": 11.4493, + "loss_ce": 14.225593566894531, + "loss_region": 0.0, + "loss_total": 14.225593566894531, + "lr": 0.000358, + "step": 180, + "tokens_trained": 0.589762952 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 0.053896886745620876, + "grad_norm": 137.1820526123047, + "loss": 11.294, + "loss_ce": 9.20856761932373, + "loss_region": 0.0, + "loss_total": 9.20856761932373, + "lr": 0.000378, + "step": 190, + "tokens_trained": 0.622527592 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 0.056733564995390395, + "grad_norm": 93.96635437011719, + "loss": 8.0307, + "loss_ce": 3.404432773590088, + "loss_region": 0.0, + "loss_total": 3.404432773590088, + "lr": 0.000398, + "step": 200, + "tokens_trained": 0.655293032 + }, + { + "epoch": 0.056733564995390395, + "eval_ppl": 1972.2048060236614, + "eval_runtime": 1.4101, + "step": 200, + "tokens_trained": 0.655293032 + }, + { + "epoch": 0.056733564995390395, + "eval_ppl_cds": 2864.110023772201, + "eval_ppl_dig": 1282.6537276346817, + "eval_ppl_exon": 2013.4369361082493, + "eval_ppl_intron": 1839.2661017019461, + "eval_ppl_nig": 1737.7572552442252, + "eval_ppl_promoter": 2648.6689395554804, + "eval_ppl_utr": 1938.5896840558044, + "step": 200, + "tokens_trained": 0.655293032 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 0.059570243245159915, + "grad_norm": 120.91366577148438, + "loss": 5.6139, + "loss_ce": 7.408278465270996, + "loss_region": 0.0, + "loss_total": 7.408278465270996, + "lr": 0.00041799999999999997, + "step": 210, + "tokens_trained": 0.688057672 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 0.062406921494929435, + "grad_norm": 116.25177764892578, + "loss": 5.3033, + "loss_ce": 3.9551126956939697, + "loss_region": 0.0, + "loss_total": 3.9551126956939697, + "lr": 0.000438, + "step": 220, + "tokens_trained": 0.720823112 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 0.06524359974469895, + "grad_norm": 38.49441146850586, + "loss": 3.4178, + "loss_ce": 1.59901762008667, + "loss_region": 0.0, + "loss_total": 1.59901762008667, + "lr": 0.000458, + "step": 230, + "tokens_trained": 0.753588552 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 0.06808027799446847, + "grad_norm": 126.20514678955078, + "loss": 4.3709, + "loss_ce": 4.5383148193359375, + "loss_region": 0.0, + "loss_total": 4.5383148193359375, + "lr": 0.00047799999999999996, + "step": 240, + "tokens_trained": 0.786353992 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 0.070916956244238, + "grad_norm": 119.89677429199219, + "loss": 3.4403, + "loss_ce": 3.9751737117767334, + "loss_region": 0.0, + "loss_total": 3.9751737117767334, + "lr": 0.000498, + "step": 250, + "tokens_trained": 0.819119432 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 0.07375363449400751, + "grad_norm": 143.06443786621094, + "loss": 5.0706, + "loss_ce": 5.964416027069092, + "loss_region": 0.0, + "loss_total": 5.964416027069092, + "lr": 0.000518, + "step": 260, + "tokens_trained": 0.851884072 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 0.07659031274377703, + "grad_norm": 112.6143569946289, + "loss": 5.0095, + "loss_ce": 5.09574031829834, + "loss_region": 0.0, + "loss_total": 5.09574031829834, + "lr": 0.0005380000000000001, + "step": 270, + "tokens_trained": 0.884649512 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 0.07942699099354655, + "grad_norm": 85.32650756835938, + "loss": 5.4793, + "loss_ce": 7.073756694793701, + "loss_region": 0.0, + "loss_total": 7.073756694793701, + "lr": 0.000558, + "step": 280, + "tokens_trained": 0.917414936 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 0.08226366924331607, + "grad_norm": 44.448856353759766, + "loss": 2.8812, + "loss_ce": 1.5789477825164795, + "loss_region": 0.0, + "loss_total": 1.5789477825164795, + "lr": 0.000578, + "step": 290, + "tokens_trained": 0.950180376 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 0.08510034749308559, + "grad_norm": 82.23748779296875, + "loss": 4.403, + "loss_ce": 4.2820024490356445, + "loss_region": 0.0, + "loss_total": 4.2820024490356445, + "lr": 0.000598, + "step": 300, + "tokens_trained": 0.982945816 + }, + { + "epoch": 0.08510034749308559, + "eval_ppl": 22.071452055640084, + "eval_runtime": 1.4075, + "step": 300, + "tokens_trained": 0.982945816 + }, + { + "epoch": 0.08510034749308559, + "eval_ppl_cds": 22.229556756039052, + "eval_ppl_dig": 26.57560788657472, + "eval_ppl_exon": 21.939773174986755, + "eval_ppl_intron": 22.2900456127021, + "eval_ppl_nig": 22.6453474879022, + "eval_ppl_promoter": 21.04298957613991, + "eval_ppl_utr": 21.656022620954808, + "step": 300, + "tokens_trained": 0.982945816 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 0.08793702574285511, + "grad_norm": 96.4254379272461, + "loss": 3.0312, + "loss_ce": 2.6277811527252197, + "loss_region": 0.0, + "loss_total": 2.6277811527252197, + "lr": 0.0006180000000000001, + "step": 310, + "tokens_trained": 1.015711256 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 0.09077370399262463, + "grad_norm": 122.59945678710938, + "loss": 3.4481, + "loss_ce": 5.398453712463379, + "loss_region": 0.0, + "loss_total": 5.398453712463379, + "lr": 0.000638, + "step": 320, + "tokens_trained": 1.048476696 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 0.09361038224239415, + "grad_norm": 34.828216552734375, + "loss": 3.5857, + "loss_ce": 1.6811786890029907, + "loss_region": 0.0, + "loss_total": 1.6811786890029907, + "lr": 0.0006580000000000001, + "step": 330, + "tokens_trained": 1.081242136 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 0.09644706049216367, + "grad_norm": 66.65411376953125, + "loss": 3.0143, + "loss_ce": 2.5892159938812256, + "loss_region": 0.0, + "loss_total": 2.5892159938812256, + "lr": 0.0006780000000000001, + "step": 340, + "tokens_trained": 1.114007576 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 0.09928373874193319, + "grad_norm": 48.7845344543457, + "loss": 2.0883, + "loss_ce": 1.5924757719039917, + "loss_region": 0.0, + "loss_total": 1.5924757719039917, + "lr": 0.0006979999999999999, + "step": 350, + "tokens_trained": 1.146773016 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 0.10212041699170271, + "grad_norm": 115.99755096435547, + "loss": 2.9957, + "loss_ce": 5.12968635559082, + "loss_region": 0.0, + "loss_total": 5.12968635559082, + "lr": 0.000718, + "step": 360, + "tokens_trained": 1.179538456 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 0.10495709524147223, + "grad_norm": 107.76811981201172, + "loss": 3.0654, + "loss_ce": 3.719006299972534, + "loss_region": 0.0, + "loss_total": 3.719006299972534, + "lr": 0.000738, + "step": 370, + "tokens_trained": 1.212303896 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 0.10779377349124175, + "grad_norm": 62.37234878540039, + "loss": 2.2783, + "loss_ce": 2.2487285137176514, + "loss_region": 0.0, + "loss_total": 2.2487285137176514, + "lr": 0.000758, + "step": 380, + "tokens_trained": 1.245068536 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 0.11063045174101127, + "grad_norm": 20.96576690673828, + "loss": 2.305, + "loss_ce": 1.43910551071167, + "loss_region": 0.0, + "loss_total": 1.43910551071167, + "lr": 0.000778, + "step": 390, + "tokens_trained": 1.277833176 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 0.11346712999078079, + "grad_norm": 95.008544921875, + "loss": 2.2551, + "loss_ce": 2.779327392578125, + "loss_region": 0.0, + "loss_total": 2.779327392578125, + "lr": 0.0007980000000000001, + "step": 400, + "tokens_trained": 1.310598616 + }, + { + "epoch": 0.11346712999078079, + "eval_ppl": 15.864071860715688, + "eval_runtime": 1.4126, + "step": 400, + "tokens_trained": 1.310598616 + }, + { + "epoch": 0.11346712999078079, + "eval_ppl_cds": 13.937295314647612, + "eval_ppl_dig": 19.429522164169192, + "eval_ppl_exon": 16.24043930094327, + "eval_ppl_intron": 16.569924627539425, + "eval_ppl_nig": 16.795222471822324, + "eval_ppl_promoter": 13.939916272492365, + "eval_ppl_utr": 15.448314726942886, + "step": 400, + "tokens_trained": 1.310598616 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 0.11630380824055031, + "grad_norm": 92.05335235595703, + "loss": 2.5682, + "loss_ce": 2.871081829071045, + "loss_region": 0.0, + "loss_total": 2.871081829071045, + "lr": 0.0008179999999999999, + "step": 410, + "tokens_trained": 1.343364056 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 0.11914048649031983, + "grad_norm": 83.41656494140625, + "loss": 2.8216, + "loss_ce": 4.226264953613281, + "loss_region": 0.0, + "loss_total": 4.226264953613281, + "lr": 0.000838, + "step": 420, + "tokens_trained": 1.376129496 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 0.12197716474008935, + "grad_norm": 84.8116226196289, + "loss": 3.1456, + "loss_ce": 2.328850030899048, + "loss_region": 0.0, + "loss_total": 2.328850030899048, + "lr": 0.000858, + "step": 430, + "tokens_trained": 1.408889864 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 0.12481384298985887, + "grad_norm": 77.26040649414062, + "loss": 2.7876, + "loss_ce": 2.19539475440979, + "loss_region": 0.0, + "loss_total": 2.19539475440979, + "lr": 0.000878, + "step": 440, + "tokens_trained": 1.441655304 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 0.1276505212396284, + "grad_norm": 42.24169921875, + "loss": 2.1762, + "loss_ce": 1.52322518825531, + "loss_region": 0.0, + "loss_total": 1.52322518825531, + "lr": 0.000898, + "step": 450, + "tokens_trained": 1.474420744 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 0.1304871994893979, + "grad_norm": 58.8355598449707, + "loss": 2.9465, + "loss_ce": 2.580343246459961, + "loss_region": 0.0, + "loss_total": 2.580343246459961, + "lr": 0.0009180000000000001, + "step": 460, + "tokens_trained": 1.507186184 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 0.13332387773916743, + "grad_norm": 73.76204681396484, + "loss": 2.5121, + "loss_ce": 2.1955363750457764, + "loss_region": 0.0, + "loss_total": 2.1955363750457764, + "lr": 0.0009379999999999999, + "step": 470, + "tokens_trained": 1.539950832 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 0.13616055598893695, + "grad_norm": 22.72353172302246, + "loss": 2.1327, + "loss_ce": 1.3411613702774048, + "loss_region": 0.0, + "loss_total": 1.3411613702774048, + "lr": 0.000958, + "step": 480, + "tokens_trained": 1.572715472 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 0.13899723423870647, + "grad_norm": 48.35820007324219, + "loss": 2.006, + "loss_ce": 2.022200584411621, + "loss_region": 0.0, + "loss_total": 2.022200584411621, + "lr": 0.000978, + "step": 490, + "tokens_trained": 1.605480912 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 0.141833912488476, + "grad_norm": 58.703765869140625, + "loss": 1.9284, + "loss_ce": 1.7221964597702026, + "loss_region": 0.0, + "loss_total": 1.7221964597702026, + "lr": 0.000998, + "step": 500, + "tokens_trained": 1.638244216 + }, + { + "epoch": 0.141833912488476, + "eval_ppl": 4.554795502338922, + "eval_runtime": 1.4105, + "step": 500, + "tokens_trained": 1.638244216 + }, + { + "epoch": 0.141833912488476, + "eval_ppl_cds": 5.576931549005295, + "eval_ppl_dig": 4.964244406566995, + "eval_ppl_exon": 4.862159525546744, + "eval_ppl_intron": 4.459020490585258, + "eval_ppl_nig": 4.467317204472241, + "eval_ppl_promoter": 5.051681021143016, + "eval_ppl_utr": 4.69584687078613, + "step": 500, + "tokens_trained": 1.638244216 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 0.1446705907382455, + "grad_norm": 46.24919128417969, + "loss": 1.5219, + "loss_ce": 1.5682318210601807, + "loss_region": 0.0, + "loss_total": 1.5682318210601807, + "lr": 0.0009997009966777408, + "step": 510, + "tokens_trained": 1.671005424 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 0.14750726898801503, + "grad_norm": 25.261741638183594, + "loss": 1.3809, + "loss_ce": 1.305256724357605, + "loss_region": 0.0, + "loss_total": 1.305256724357605, + "lr": 0.0009993687707641197, + "step": 520, + "tokens_trained": 1.703770864 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 0.15034394723778455, + "grad_norm": 36.93369674682617, + "loss": 1.3945, + "loss_ce": 1.5390609502792358, + "loss_region": 0.0, + "loss_total": 1.5390609502792358, + "lr": 0.0009990365448504983, + "step": 530, + "tokens_trained": 1.736536304 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 0.15318062548755407, + "grad_norm": 21.187707901000977, + "loss": 1.3156, + "loss_ce": 1.3369269371032715, + "loss_region": 0.0, + "loss_total": 1.3369269371032715, + "lr": 0.0009987043189368771, + "step": 540, + "tokens_trained": 1.769301744 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 0.1560173037373236, + "grad_norm": 16.391759872436523, + "loss": 1.2888, + "loss_ce": 1.2410407066345215, + "loss_region": 0.0, + "loss_total": 1.2410407066345215, + "lr": 0.0009983720930232557, + "step": 550, + "tokens_trained": 1.802067184 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 0.1588539819870931, + "grad_norm": 26.08888816833496, + "loss": 1.3874, + "loss_ce": 1.236966609954834, + "loss_region": 0.0, + "loss_total": 1.236966609954834, + "lr": 0.0009980398671096346, + "step": 560, + "tokens_trained": 1.834832624 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 0.16169066023686263, + "grad_norm": 71.1765365600586, + "loss": 1.5916, + "loss_ce": 1.9130433797836304, + "loss_region": 0.0, + "loss_total": 1.9130433797836304, + "lr": 0.0009977076411960134, + "step": 570, + "tokens_trained": 1.867598064 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 0.16452733848663215, + "grad_norm": 33.561527252197266, + "loss": 1.5963, + "loss_ce": 1.4402023553848267, + "loss_region": 0.0, + "loss_total": 1.4402023553848267, + "lr": 0.000997375415282392, + "step": 580, + "tokens_trained": 1.900363504 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 0.16736401673640167, + "grad_norm": 51.28582000732422, + "loss": 1.3928, + "loss_ce": 1.5661563873291016, + "loss_region": 0.0, + "loss_total": 1.5661563873291016, + "lr": 0.0009970431893687709, + "step": 590, + "tokens_trained": 1.933128944 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 0.17020069498617119, + "grad_norm": 54.031864166259766, + "loss": 1.6238, + "loss_ce": 1.843924880027771, + "loss_region": 0.0, + "loss_total": 1.843924880027771, + "lr": 0.0009967109634551495, + "step": 600, + "tokens_trained": 1.96589048 + }, + { + "epoch": 0.17020069498617119, + "eval_ppl": 5.004482639707367, + "eval_runtime": 1.4109, + "step": 600, + "tokens_trained": 1.96589048 + }, + { + "epoch": 0.17020069498617119, + "eval_ppl_cds": 6.2887597229877175, + "eval_ppl_dig": 5.389907758112067, + "eval_ppl_exon": 5.440573761181735, + "eval_ppl_intron": 4.924694148663745, + "eval_ppl_nig": 4.949271444767212, + "eval_ppl_promoter": 5.534675512278488, + "eval_ppl_utr": 5.253319543842166, + "step": 600, + "tokens_trained": 1.96589048 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 0.1730373732359407, + "grad_norm": 27.011795043945312, + "loss": 1.4576, + "loss_ce": 1.5860403776168823, + "loss_region": 0.0, + "loss_total": 1.5860403776168823, + "lr": 0.0009963787375415283, + "step": 610, + "tokens_trained": 1.99865592 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 0.17587405148571023, + "grad_norm": 25.85240364074707, + "loss": 1.297, + "loss_ce": 1.3282049894332886, + "loss_region": 0.0, + "loss_total": 1.3282049894332886, + "lr": 0.0009960465116279071, + "step": 620, + "tokens_trained": 2.03142136 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 0.17871072973547975, + "grad_norm": 37.66011047363281, + "loss": 1.3934, + "loss_ce": 1.6049169301986694, + "loss_region": 0.0, + "loss_total": 1.6049169301986694, + "lr": 0.0009957142857142858, + "step": 630, + "tokens_trained": 2.0641868 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 0.18154740798524927, + "grad_norm": 17.765363693237305, + "loss": 1.2926, + "loss_ce": 1.274528980255127, + "loss_region": 0.0, + "loss_total": 1.274528980255127, + "lr": 0.0009953820598006644, + "step": 640, + "tokens_trained": 2.09695224 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 0.18438408623501878, + "grad_norm": 20.909456253051758, + "loss": 1.4113, + "loss_ce": 1.3923135995864868, + "loss_region": 0.0, + "loss_total": 1.3923135995864868, + "lr": 0.0009950498338870432, + "step": 650, + "tokens_trained": 2.12971768 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 0.1872207644847883, + "grad_norm": 26.084135055541992, + "loss": 1.5733, + "loss_ce": 1.385710597038269, + "loss_region": 0.0, + "loss_total": 1.385710597038269, + "lr": 0.0009947176079734218, + "step": 660, + "tokens_trained": 2.16248312 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 0.19005744273455782, + "grad_norm": 17.783727645874023, + "loss": 1.4159, + "loss_ce": 1.3202787637710571, + "loss_region": 0.0, + "loss_total": 1.3202787637710571, + "lr": 0.0009943853820598007, + "step": 670, + "tokens_trained": 2.19524856 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 0.19289412098432734, + "grad_norm": 29.755754470825195, + "loss": 1.3776, + "loss_ce": 1.433687448501587, + "loss_region": 0.0, + "loss_total": 1.433687448501587, + "lr": 0.0009940531561461795, + "step": 680, + "tokens_trained": 2.228014 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 0.19573079923409686, + "grad_norm": 24.880355834960938, + "loss": 1.3955, + "loss_ce": 1.3981382846832275, + "loss_region": 0.0, + "loss_total": 1.3981382846832275, + "lr": 0.000993720930232558, + "step": 690, + "tokens_trained": 2.26077944 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 0.19856747748386638, + "grad_norm": 5.883652687072754, + "loss": 1.2893, + "loss_ce": 1.1893589496612549, + "loss_region": 0.0, + "loss_total": 1.1893589496612549, + "lr": 0.000993388704318937, + "step": 700, + "tokens_trained": 2.29354488 + }, + { + "epoch": 0.19856747748386638, + "eval_ppl": 3.695858356509464, + "eval_runtime": 1.4094, + "step": 700, + "tokens_trained": 2.29354488 + }, + { + "epoch": 0.19856747748386638, + "eval_ppl_cds": 4.201160190968239, + "eval_ppl_dig": 4.1659348465044745, + "eval_ppl_exon": 3.936117162961689, + "eval_ppl_intron": 3.6749621494047755, + "eval_ppl_nig": 3.706485050565789, + "eval_ppl_promoter": 3.8471985440321492, + "eval_ppl_utr": 3.8487199293970273, + "step": 700, + "tokens_trained": 2.29354488 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 0.2014041557336359, + "grad_norm": 23.792387008666992, + "loss": 1.3652, + "loss_ce": 1.405797004699707, + "loss_region": 0.0, + "loss_total": 1.405797004699707, + "lr": 0.0009930564784053156, + "step": 710, + "tokens_trained": 2.32631032 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 0.20424083398340542, + "grad_norm": 19.024282455444336, + "loss": 1.2726, + "loss_ce": 1.2662293910980225, + "loss_region": 0.0, + "loss_total": 1.2662293910980225, + "lr": 0.0009927242524916944, + "step": 720, + "tokens_trained": 2.35907576 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 0.20707751223317494, + "grad_norm": 14.736429214477539, + "loss": 1.3036, + "loss_ce": 1.2109957933425903, + "loss_region": 0.0, + "loss_total": 1.2109957933425903, + "lr": 0.0009923920265780732, + "step": 730, + "tokens_trained": 2.3918396 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 0.20991419048294446, + "grad_norm": 42.79386901855469, + "loss": 1.3053, + "loss_ce": 1.4333525896072388, + "loss_region": 0.0, + "loss_total": 1.4333525896072388, + "lr": 0.0009920598006644518, + "step": 740, + "tokens_trained": 2.424600048 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 0.21275086873271398, + "grad_norm": 6.748932838439941, + "loss": 1.3226, + "loss_ce": 1.2504080533981323, + "loss_region": 0.0, + "loss_total": 1.2504080533981323, + "lr": 0.0009917275747508305, + "step": 750, + "tokens_trained": 2.457364688 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 0.2155875469824835, + "grad_norm": 27.288007736206055, + "loss": 1.427, + "loss_ce": 1.4040110111236572, + "loss_region": 0.0, + "loss_total": 1.4040110111236572, + "lr": 0.0009913953488372093, + "step": 760, + "tokens_trained": 2.490130128 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 0.21842422523225302, + "grad_norm": 9.700417518615723, + "loss": 1.2838, + "loss_ce": 1.2128959894180298, + "loss_region": 0.0, + "loss_total": 1.2128959894180298, + "lr": 0.0009910631229235881, + "step": 770, + "tokens_trained": 2.522895568 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 0.22126090348202254, + "grad_norm": 23.441314697265625, + "loss": 1.2793, + "loss_ce": 1.435208797454834, + "loss_region": 0.0, + "loss_total": 1.435208797454834, + "lr": 0.0009907308970099667, + "step": 780, + "tokens_trained": 2.555659392 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 0.22409758173179206, + "grad_norm": 26.008195877075195, + "loss": 1.3653, + "loss_ce": 1.40993332862854, + "loss_region": 0.0, + "loss_total": 1.40993332862854, + "lr": 0.0009903986710963456, + "step": 790, + "tokens_trained": 2.588422136 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 0.22693425998156158, + "grad_norm": 10.16769790649414, + "loss": 1.2692, + "loss_ce": 1.266273856163025, + "loss_region": 0.0, + "loss_total": 1.266273856163025, + "lr": 0.0009900664451827242, + "step": 800, + "tokens_trained": 2.621187576 + }, + { + "epoch": 0.22693425998156158, + "eval_ppl": 3.5775458128434954, + "eval_runtime": 1.4024, + "step": 800, + "tokens_trained": 2.621187576 + }, + { + "epoch": 0.22693425998156158, + "eval_ppl_cds": 4.192082583120108, + "eval_ppl_dig": 2.2472346229495828, + "eval_ppl_exon": 3.8775077932630553, + "eval_ppl_intron": 3.5799924922543034, + "eval_ppl_nig": 3.579964841505043, + "eval_ppl_promoter": 3.8591878697574793, + "eval_ppl_utr": 3.8044907174219547, + "step": 800, + "tokens_trained": 2.621187576 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 0.2297709382313311, + "grad_norm": 22.04285430908203, + "loss": 1.2481, + "loss_ce": 1.3485249280929565, + "loss_region": 0.0, + "loss_total": 1.3485249280929565, + "lr": 0.000989734219269103, + "step": 810, + "tokens_trained": 2.653953016 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 0.23260761648110062, + "grad_norm": 16.790050506591797, + "loss": 1.3313, + "loss_ce": 1.291212558746338, + "loss_region": 0.0, + "loss_total": 1.291212558746338, + "lr": 0.0009894019933554819, + "step": 820, + "tokens_trained": 2.686718456 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 0.23544429473087014, + "grad_norm": 9.069281578063965, + "loss": 1.2359, + "loss_ce": 1.1202536821365356, + "loss_region": 0.0, + "loss_total": 1.1202536821365356, + "lr": 0.0009890697674418605, + "step": 830, + "tokens_trained": 2.71948036 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 0.23828097298063966, + "grad_norm": 16.948314666748047, + "loss": 1.2455, + "loss_ce": 1.3530311584472656, + "loss_region": 0.0, + "loss_total": 1.3530311584472656, + "lr": 0.0009887375415282393, + "step": 840, + "tokens_trained": 2.7522458 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 0.24111765123040918, + "grad_norm": 21.52397918701172, + "loss": 1.271, + "loss_ce": 1.2375919818878174, + "loss_region": 0.0, + "loss_total": 1.2375919818878174, + "lr": 0.000988405315614618, + "step": 850, + "tokens_trained": 2.78501124 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 0.2439543294801787, + "grad_norm": 17.04692840576172, + "loss": 1.2375, + "loss_ce": 1.2279222011566162, + "loss_region": 0.0, + "loss_total": 1.2279222011566162, + "lr": 0.0009880730897009968, + "step": 860, + "tokens_trained": 2.81777668 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 0.24679100772994822, + "grad_norm": 7.490538120269775, + "loss": 1.2344, + "loss_ce": 1.2513377666473389, + "loss_region": 0.0, + "loss_total": 1.2513377666473389, + "lr": 0.0009877408637873756, + "step": 870, + "tokens_trained": 2.85054212 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 0.24962768597971774, + "grad_norm": 11.91706371307373, + "loss": 1.1959, + "loss_ce": 1.2798902988433838, + "loss_region": 0.0, + "loss_total": 1.2798902988433838, + "lr": 0.0009874086378737542, + "step": 880, + "tokens_trained": 2.88330756 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 0.25246436422948726, + "grad_norm": 3.981492519378662, + "loss": 1.2037, + "loss_ce": 1.1099942922592163, + "loss_region": 0.0, + "loss_total": 1.1099942922592163, + "lr": 0.0009870764119601328, + "step": 890, + "tokens_trained": 2.916073 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 0.2553010424792568, + "grad_norm": 0.7608258724212646, + "loss": 1.1705, + "loss_ce": 1.1606467962265015, + "loss_region": 0.0, + "loss_total": 1.1606467962265015, + "lr": 0.0009867441860465116, + "step": 900, + "tokens_trained": 2.94883828 + }, + { + "epoch": 0.2553010424792568, + "eval_ppl": 3.2377567957373583, + "eval_runtime": 1.4146, + "step": 900, + "tokens_trained": 2.94883828 + }, + { + "epoch": 0.2553010424792568, + "eval_ppl_cds": 3.8787976724531545, + "eval_ppl_dig": 1.5590028489661645, + "eval_ppl_exon": 3.595619461289906, + "eval_ppl_intron": 3.268171899957025, + "eval_ppl_nig": 3.136674797709474, + "eval_ppl_promoter": 3.5228431930393285, + "eval_ppl_utr": 3.5013853599870175, + "step": 900, + "tokens_trained": 2.94883828 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 0.2581377207290263, + "grad_norm": 11.921713829040527, + "loss": 1.1685, + "loss_ce": 1.1965281963348389, + "loss_region": 0.0, + "loss_total": 1.1965281963348389, + "lr": 0.0009864119601328903, + "step": 910, + "tokens_trained": 2.981597288 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 0.2609743989787958, + "grad_norm": 10.19805908203125, + "loss": 1.183, + "loss_ce": 1.290944218635559, + "loss_region": 0.0, + "loss_total": 1.290944218635559, + "lr": 0.000986079734219269, + "step": 920, + "tokens_trained": 3.014362456 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 0.26381107722856534, + "grad_norm": 6.590117454528809, + "loss": 1.168, + "loss_ce": 1.1465442180633545, + "loss_region": 0.0, + "loss_total": 1.1465442180633545, + "lr": 0.000985747508305648, + "step": 930, + "tokens_trained": 3.047127096 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 0.26664775547833486, + "grad_norm": 8.28929615020752, + "loss": 1.1697, + "loss_ce": 1.1579564809799194, + "loss_region": 0.0, + "loss_total": 1.1579564809799194, + "lr": 0.0009854152823920265, + "step": 940, + "tokens_trained": 3.079892536 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 0.2694844337281044, + "grad_norm": 11.367734909057617, + "loss": 1.1633, + "loss_ce": 1.1996207237243652, + "loss_region": 0.0, + "loss_total": 1.1996207237243652, + "lr": 0.0009850830564784054, + "step": 950, + "tokens_trained": 3.11265336 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 0.2723211119778739, + "grad_norm": 4.552338123321533, + "loss": 1.1657, + "loss_ce": 1.2341638803482056, + "loss_region": 0.0, + "loss_total": 1.2341638803482056, + "lr": 0.000984750830564784, + "step": 960, + "tokens_trained": 3.1454188 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 0.2751577902276434, + "grad_norm": 4.104018211364746, + "loss": 1.1623, + "loss_ce": 1.1250998973846436, + "loss_region": 0.0, + "loss_total": 1.1250998973846436, + "lr": 0.0009844186046511628, + "step": 970, + "tokens_trained": 3.17818424 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 0.27799446847741294, + "grad_norm": 8.622637748718262, + "loss": 1.1876, + "loss_ce": 1.21863853931427, + "loss_region": 0.0, + "loss_total": 1.21863853931427, + "lr": 0.0009840863787375417, + "step": 980, + "tokens_trained": 3.21094968 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 0.28083114672718246, + "grad_norm": 6.942562580108643, + "loss": 1.1949, + "loss_ce": 1.2027949094772339, + "loss_region": 0.0, + "loss_total": 1.2027949094772339, + "lr": 0.0009837541528239203, + "step": 990, + "tokens_trained": 3.24371512 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 0.283667824976952, + "grad_norm": 17.273046493530273, + "loss": 1.2445, + "loss_ce": 1.274321436882019, + "loss_region": 0.0, + "loss_total": 1.274321436882019, + "lr": 0.000983421926910299, + "step": 1000, + "tokens_trained": 3.27648056 + }, + { + "epoch": 0.283667824976952, + "eval_ppl": 3.537316848516589, + "eval_runtime": 1.4107, + "step": 1000, + "tokens_trained": 3.27648056 + }, + { + "epoch": 0.283667824976952, + "eval_ppl_cds": 4.498508334110634, + "eval_ppl_dig": 1.3975029274713529, + "eval_ppl_exon": 3.990774522978742, + "eval_ppl_intron": 3.552393618865801, + "eval_ppl_nig": 3.3682938989717135, + "eval_ppl_promoter": 3.9961773459300214, + "eval_ppl_utr": 3.9240290873998758, + "step": 1000, + "tokens_trained": 3.27648056 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 0.2865045032267215, + "grad_norm": 18.324474334716797, + "loss": 1.2053, + "loss_ce": 1.377746343612671, + "loss_region": 0.0, + "loss_total": 1.377746343612671, + "lr": 0.0009830897009966777, + "step": 1010, + "tokens_trained": 3.309246 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 0.289341181476491, + "grad_norm": 14.222187995910645, + "loss": 1.1896, + "loss_ce": 1.2169245481491089, + "loss_region": 0.0, + "loss_total": 1.2169245481491089, + "lr": 0.0009827574750830566, + "step": 1020, + "tokens_trained": 3.34201144 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 0.29217785972626054, + "grad_norm": 3.756948709487915, + "loss": 1.1763, + "loss_ce": 1.093570351600647, + "loss_region": 0.0, + "loss_total": 1.093570351600647, + "lr": 0.0009824252491694352, + "step": 1030, + "tokens_trained": 3.37477688 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 0.29501453797603006, + "grad_norm": 7.29799222946167, + "loss": 1.1903, + "loss_ce": 1.0943608283996582, + "loss_region": 0.0, + "loss_total": 1.0943608283996582, + "lr": 0.000982093023255814, + "step": 1040, + "tokens_trained": 3.40754232 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 0.2978512162257996, + "grad_norm": 7.003857135772705, + "loss": 1.1602, + "loss_ce": 1.1793015003204346, + "loss_region": 0.0, + "loss_total": 1.1793015003204346, + "lr": 0.0009817607973421926, + "step": 1050, + "tokens_trained": 3.44030696 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 0.3006878944755691, + "grad_norm": 6.669435501098633, + "loss": 1.1397, + "loss_ce": 1.1221890449523926, + "loss_region": 0.0, + "loss_total": 1.1221890449523926, + "lr": 0.0009814285714285715, + "step": 1060, + "tokens_trained": 3.4730724 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 0.3035245727253386, + "grad_norm": 5.536543369293213, + "loss": 1.1422, + "loss_ce": 1.1252641677856445, + "loss_region": 0.0, + "loss_total": 1.1252641677856445, + "lr": 0.0009810963455149503, + "step": 1070, + "tokens_trained": 3.50583784 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 0.30636125097510813, + "grad_norm": 2.378337860107422, + "loss": 1.1352, + "loss_ce": 1.118593692779541, + "loss_region": 0.0, + "loss_total": 1.118593692779541, + "lr": 0.000980764119601329, + "step": 1080, + "tokens_trained": 3.53860328 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 0.30919792922487765, + "grad_norm": 13.784808158874512, + "loss": 1.1927, + "loss_ce": 1.2212815284729004, + "loss_region": 0.0, + "loss_total": 1.2212815284729004, + "lr": 0.0009804318936877077, + "step": 1090, + "tokens_trained": 3.57136872 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 0.3120346074746472, + "grad_norm": 6.65331506729126, + "loss": 1.1565, + "loss_ce": 1.0855413675308228, + "loss_region": 0.0, + "loss_total": 1.0855413675308228, + "lr": 0.0009800996677740864, + "step": 1100, + "tokens_trained": 3.60413416 + }, + { + "epoch": 0.3120346074746472, + "eval_ppl": 3.115542958140476, + "eval_runtime": 1.4179, + "step": 1100, + "tokens_trained": 3.60413416 + }, + { + "epoch": 0.3120346074746472, + "eval_ppl_cds": 3.850197039401023, + "eval_ppl_dig": 1.257814242987907, + "eval_ppl_exon": 3.503710933243971, + "eval_ppl_intron": 3.1508921197971795, + "eval_ppl_nig": 2.9694012661018383, + "eval_ppl_promoter": 3.4491209659674062, + "eval_ppl_utr": 3.426615851632063, + "step": 1100, + "tokens_trained": 3.60413416 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 0.3148712857244167, + "grad_norm": 6.421998023986816, + "loss": 1.1458, + "loss_ce": 1.1017615795135498, + "loss_region": 0.0, + "loss_total": 1.1017615795135498, + "lr": 0.0009797674418604652, + "step": 1110, + "tokens_trained": 3.6368996 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 0.3177079639741862, + "grad_norm": 11.737626075744629, + "loss": 1.1721, + "loss_ce": 1.2059839963912964, + "loss_region": 0.0, + "loss_total": 1.2059839963912964, + "lr": 0.000979435215946844, + "step": 1120, + "tokens_trained": 3.669661712 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 0.32054464222395573, + "grad_norm": 3.722311019897461, + "loss": 1.1549, + "loss_ce": 1.1525661945343018, + "loss_region": 0.0, + "loss_total": 1.1525661945343018, + "lr": 0.0009791029900332226, + "step": 1130, + "tokens_trained": 3.702426352 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 0.32338132047372525, + "grad_norm": 2.8785247802734375, + "loss": 1.1357, + "loss_ce": 1.1328191757202148, + "loss_region": 0.0, + "loss_total": 1.1328191757202148, + "lr": 0.0009787707641196013, + "step": 1140, + "tokens_trained": 3.735191792 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 0.3262179987234948, + "grad_norm": 6.521324157714844, + "loss": 1.1637, + "loss_ce": 1.096291422843933, + "loss_region": 0.0, + "loss_total": 1.096291422843933, + "lr": 0.00097843853820598, + "step": 1150, + "tokens_trained": 3.767957232 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 0.3290546769732643, + "grad_norm": 4.199342250823975, + "loss": 1.1409, + "loss_ce": 1.2215949296951294, + "loss_region": 0.0, + "loss_total": 1.2215949296951294, + "lr": 0.0009781063122923587, + "step": 1160, + "tokens_trained": 3.800722672 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 0.3318913552230338, + "grad_norm": 2.070523262023926, + "loss": 1.123, + "loss_ce": 1.064353585243225, + "loss_region": 0.0, + "loss_total": 1.064353585243225, + "lr": 0.0009777740863787375, + "step": 1170, + "tokens_trained": 3.833488112 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 0.33472803347280333, + "grad_norm": 10.165450096130371, + "loss": 1.1538, + "loss_ce": 1.185705542564392, + "loss_region": 0.0, + "loss_total": 1.185705542564392, + "lr": 0.0009774418604651164, + "step": 1180, + "tokens_trained": 3.866252752 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 0.33756471172257285, + "grad_norm": 8.56866455078125, + "loss": 1.1611, + "loss_ce": 1.165594458580017, + "loss_region": 0.0, + "loss_total": 1.165594458580017, + "lr": 0.000977109634551495, + "step": 1190, + "tokens_trained": 3.899018184 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 0.34040138997234237, + "grad_norm": 12.543554306030273, + "loss": 1.1515, + "loss_ce": 1.1477564573287964, + "loss_region": 0.0, + "loss_total": 1.1477564573287964, + "lr": 0.0009767774086378738, + "step": 1200, + "tokens_trained": 3.931783624 + }, + { + "epoch": 0.34040138997234237, + "eval_ppl": 3.3082192706615747, + "eval_runtime": 1.4127, + "step": 1200, + "tokens_trained": 3.931783624 + }, + { + "epoch": 0.34040138997234237, + "eval_ppl_cds": 4.112162862718435, + "eval_ppl_dig": 1.2686203421818523, + "eval_ppl_exon": 3.757932714632875, + "eval_ppl_intron": 3.3494948182387962, + "eval_ppl_nig": 3.1434499908625426, + "eval_ppl_promoter": 3.6902939278042086, + "eval_ppl_utr": 3.6861373831638296, + "step": 1200, + "tokens_trained": 3.931783624 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 0.3432380682221119, + "grad_norm": 15.498085975646973, + "loss": 1.1581, + "loss_ce": 1.2116421461105347, + "loss_region": 0.0, + "loss_total": 1.2116421461105347, + "lr": 0.0009764451827242526, + "step": 1210, + "tokens_trained": 3.964549064 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 0.3460747464718814, + "grad_norm": 4.98313045501709, + "loss": 1.143, + "loss_ce": 1.1006969213485718, + "loss_region": 0.0, + "loss_total": 1.1006969213485718, + "lr": 0.0009761129568106313, + "step": 1220, + "tokens_trained": 3.997311912 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 0.34891142472165093, + "grad_norm": 7.086497783660889, + "loss": 1.1413, + "loss_ce": 1.1618386507034302, + "loss_region": 0.0, + "loss_total": 1.1618386507034302, + "lr": 0.00097578073089701, + "step": 1230, + "tokens_trained": 4.030077352 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 0.35174810297142045, + "grad_norm": 8.35598087310791, + "loss": 1.13, + "loss_ce": 1.0692408084869385, + "loss_region": 0.0, + "loss_total": 1.0692408084869385, + "lr": 0.0009754485049833887, + "step": 1240, + "tokens_trained": 4.062842792 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 0.35458478122118997, + "grad_norm": 2.0300137996673584, + "loss": 1.1254, + "loss_ce": 1.1593506336212158, + "loss_region": 0.0, + "loss_total": 1.1593506336212158, + "lr": 0.0009751162790697675, + "step": 1250, + "tokens_trained": 4.095608232 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 0.3574214594709595, + "grad_norm": 5.932432651519775, + "loss": 1.135, + "loss_ce": 1.1723196506500244, + "loss_region": 0.0, + "loss_total": 1.1723196506500244, + "lr": 0.0009747840531561462, + "step": 1260, + "tokens_trained": 4.128373672 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 0.360258137720729, + "grad_norm": 2.3463633060455322, + "loss": 1.13, + "loss_ce": 1.1727961301803589, + "loss_region": 0.0, + "loss_total": 1.1727961301803589, + "lr": 0.0009744518272425249, + "step": 1270, + "tokens_trained": 4.161136768 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 0.36309481597049853, + "grad_norm": 4.4678144454956055, + "loss": 1.1302, + "loss_ce": 1.1143344640731812, + "loss_region": 0.0, + "loss_total": 1.1143344640731812, + "lr": 0.0009741196013289036, + "step": 1280, + "tokens_trained": 4.193902208 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 0.36593149422026805, + "grad_norm": 2.0101044178009033, + "loss": 1.1126, + "loss_ce": 1.1864662170410156, + "loss_region": 0.0, + "loss_total": 1.1864662170410156, + "lr": 0.0009737873754152823, + "step": 1290, + "tokens_trained": 4.226667648 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 0.36876817247003757, + "grad_norm": 3.1716599464416504, + "loss": 1.1262, + "loss_ce": 1.0644184350967407, + "loss_region": 0.0, + "loss_total": 1.0644184350967407, + "lr": 0.0009734551495016612, + "step": 1300, + "tokens_trained": 4.259424272 + }, + { + "epoch": 0.36876817247003757, + "eval_ppl": 3.0743957780059654, + "eval_runtime": 1.4149, + "step": 1300, + "tokens_trained": 4.259424272 + }, + { + "epoch": 0.36876817247003757, + "eval_ppl_cds": 3.835567124882846, + "eval_ppl_dig": 1.2022919695282421, + "eval_ppl_exon": 3.4830267609055667, + "eval_ppl_intron": 3.119053479641782, + "eval_ppl_nig": 2.9134178788018334, + "eval_ppl_promoter": 3.420036138437014, + "eval_ppl_utr": 3.402437783627456, + "step": 1300, + "tokens_trained": 4.259424272 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 0.3716048507198071, + "grad_norm": 3.236912965774536, + "loss": 1.1166, + "loss_ce": 1.045543909072876, + "loss_region": 0.0, + "loss_total": 1.045543909072876, + "lr": 0.0009731229235880399, + "step": 1310, + "tokens_trained": 4.292189712 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 0.3744415289695766, + "grad_norm": 3.8641388416290283, + "loss": 1.1145, + "loss_ce": 1.078147053718567, + "loss_region": 0.0, + "loss_total": 1.078147053718567, + "lr": 0.0009727906976744186, + "step": 1320, + "tokens_trained": 4.32495164 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 0.37727820721934613, + "grad_norm": 4.636021614074707, + "loss": 1.1104, + "loss_ce": 0.9784366488456726, + "loss_region": 0.0, + "loss_total": 0.9784366488456726, + "lr": 0.0009724584717607974, + "step": 1330, + "tokens_trained": 4.35771708 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 0.38011488546911565, + "grad_norm": 3.498225688934326, + "loss": 1.1071, + "loss_ce": 1.194345474243164, + "loss_region": 0.0, + "loss_total": 1.194345474243164, + "lr": 0.0009721262458471761, + "step": 1340, + "tokens_trained": 4.39048252 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 0.38295156371888517, + "grad_norm": 0.9551838040351868, + "loss": 1.0986, + "loss_ce": 0.983210027217865, + "loss_region": 0.0, + "loss_total": 0.983210027217865, + "lr": 0.0009717940199335549, + "step": 1350, + "tokens_trained": 4.42324796 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 0.3857882419686547, + "grad_norm": 3.231743335723877, + "loss": 1.1022, + "loss_ce": 1.1021300554275513, + "loss_region": 0.0, + "loss_total": 1.1021300554275513, + "lr": 0.0009714617940199336, + "step": 1360, + "tokens_trained": 4.4560134 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 0.3886249202184242, + "grad_norm": 4.28179407119751, + "loss": 1.1224, + "loss_ce": 1.090470790863037, + "loss_region": 0.0, + "loss_total": 1.090470790863037, + "lr": 0.0009711295681063124, + "step": 1370, + "tokens_trained": 4.48877884 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 0.3914615984681937, + "grad_norm": 4.509950637817383, + "loss": 1.1105, + "loss_ce": 1.1225405931472778, + "loss_region": 0.0, + "loss_total": 1.1225405931472778, + "lr": 0.000970797342192691, + "step": 1380, + "tokens_trained": 4.52154428 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 0.39429827671796325, + "grad_norm": 4.2598958015441895, + "loss": 1.1054, + "loss_ce": 1.143487811088562, + "loss_region": 0.0, + "loss_total": 1.143487811088562, + "lr": 0.0009704651162790697, + "step": 1390, + "tokens_trained": 4.55430972 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 0.39713495496773277, + "grad_norm": 1.7044597864151, + "loss": 1.1089, + "loss_ce": 0.9078133702278137, + "loss_region": 0.0, + "loss_total": 0.9078133702278137, + "lr": 0.0009701328903654485, + "step": 1400, + "tokens_trained": 4.58707516 + }, + { + "epoch": 0.39713495496773277, + "eval_ppl": 3.03798869655183, + "eval_runtime": 1.4289, + "step": 1400, + "tokens_trained": 4.58707516 + }, + { + "epoch": 0.39713495496773277, + "eval_ppl_cds": 3.8010728411330725, + "eval_ppl_dig": 1.1669456832273732, + "eval_ppl_exon": 3.4496569947897724, + "eval_ppl_intron": 3.0877591465921106, + "eval_ppl_nig": 2.875830646844691, + "eval_ppl_promoter": 3.3836497738805056, + "eval_ppl_utr": 3.3703454454773834, + "step": 1400, + "tokens_trained": 4.58707516 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 0.3999716332175023, + "grad_norm": 5.623720645904541, + "loss": 1.1099, + "loss_ce": 1.08753502368927, + "loss_region": 0.0, + "loss_total": 1.08753502368927, + "lr": 0.0009698006644518273, + "step": 1410, + "tokens_trained": 4.6198406 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 0.4028083114672718, + "grad_norm": 3.381288766860962, + "loss": 1.0982, + "loss_ce": 1.112361192703247, + "loss_region": 0.0, + "loss_total": 1.112361192703247, + "lr": 0.000969468438538206, + "step": 1420, + "tokens_trained": 4.652606024 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 0.4056449897170413, + "grad_norm": 2.889784812927246, + "loss": 1.1086, + "loss_ce": 1.0912803411483765, + "loss_region": 0.0, + "loss_total": 1.0912803411483765, + "lr": 0.0009691362126245847, + "step": 1430, + "tokens_trained": 4.685371464 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 0.40848166796681085, + "grad_norm": 5.499249458312988, + "loss": 1.1176, + "loss_ce": 1.1578763723373413, + "loss_region": 0.0, + "loss_total": 1.1578763723373413, + "lr": 0.0009688039867109634, + "step": 1440, + "tokens_trained": 4.718136904 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 0.41131834621658037, + "grad_norm": 2.569561243057251, + "loss": 1.1186, + "loss_ce": 1.1611396074295044, + "loss_region": 0.0, + "loss_total": 1.1611396074295044, + "lr": 0.0009684717607973423, + "step": 1450, + "tokens_trained": 4.750902344 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 0.4141550244663499, + "grad_norm": 4.719552040100098, + "loss": 1.1616, + "loss_ce": 1.2481640577316284, + "loss_region": 0.0, + "loss_total": 1.2481640577316284, + "lr": 0.000968139534883721, + "step": 1460, + "tokens_trained": 4.783666984 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 0.4169917027161194, + "grad_norm": 4.6903533935546875, + "loss": 1.2764, + "loss_ce": 1.3155672550201416, + "loss_region": 0.0, + "loss_total": 1.3155672550201416, + "lr": 0.0009678073089700997, + "step": 1470, + "tokens_trained": 4.816432424 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 0.4198283809658889, + "grad_norm": 5.891879081726074, + "loss": 1.3207, + "loss_ce": 1.2832636833190918, + "loss_region": 0.0, + "loss_total": 1.2832636833190918, + "lr": 0.0009674750830564784, + "step": 1480, + "tokens_trained": 4.849197864 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 0.42266505921565845, + "grad_norm": 4.008580207824707, + "loss": 1.2785, + "loss_ce": 1.2763237953186035, + "loss_region": 0.0, + "loss_total": 1.2763237953186035, + "lr": 0.0009671428571428572, + "step": 1490, + "tokens_trained": 4.881963248 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 0.42550173746542796, + "grad_norm": 5.416539669036865, + "loss": 1.24, + "loss_ce": 1.1884835958480835, + "loss_region": 0.0, + "loss_total": 1.1884835958480835, + "lr": 0.0009668106312292359, + "step": 1500, + "tokens_trained": 4.914728608 + }, + { + "epoch": 0.42550173746542796, + "eval_ppl": 3.4055522811824597, + "eval_runtime": 1.4372, + "step": 1500, + "tokens_trained": 4.914728608 + }, + { + "epoch": 0.42550173746542796, + "eval_ppl_cds": 3.859144679566995, + "eval_ppl_dig": 3.873088762461706, + "eval_ppl_exon": 3.595146405452722, + "eval_ppl_intron": 3.383269143050641, + "eval_ppl_nig": 3.4031621866685713, + "eval_ppl_promoter": 3.554832180034531, + "eval_ppl_utr": 3.5155789087689464, + "step": 1500, + "tokens_trained": 4.914728608 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 0.4283384157151975, + "grad_norm": 2.062234878540039, + "loss": 1.2155, + "loss_ce": 1.2394578456878662, + "loss_region": 0.0, + "loss_total": 1.2394578456878662, + "lr": 0.0009664784053156146, + "step": 1510, + "tokens_trained": 4.947494048 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 0.431175093964967, + "grad_norm": 4.506737232208252, + "loss": 1.2071, + "loss_ce": 1.2651498317718506, + "loss_region": 0.0, + "loss_total": 1.2651498317718506, + "lr": 0.0009661461794019933, + "step": 1520, + "tokens_trained": 4.980259488 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 0.4340117722147365, + "grad_norm": 4.304380416870117, + "loss": 1.1897, + "loss_ce": 1.1727701425552368, + "loss_region": 0.0, + "loss_total": 1.1727701425552368, + "lr": 0.0009658139534883721, + "step": 1530, + "tokens_trained": 5.013024928 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 0.43684845046450604, + "grad_norm": 5.058155536651611, + "loss": 1.1598, + "loss_ce": 1.1405799388885498, + "loss_region": 0.0, + "loss_total": 1.1405799388885498, + "lr": 0.0009654817275747508, + "step": 1540, + "tokens_trained": 5.04578704 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 0.43968512871427556, + "grad_norm": 2.9507944583892822, + "loss": 1.1429, + "loss_ce": 1.1713906526565552, + "loss_region": 0.0, + "loss_total": 1.1713906526565552, + "lr": 0.0009651495016611296, + "step": 1550, + "tokens_trained": 5.078551904 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 0.4425218069640451, + "grad_norm": 3.3541316986083984, + "loss": 1.1335, + "loss_ce": 1.078621506690979, + "loss_region": 0.0, + "loss_total": 1.078621506690979, + "lr": 0.0009648172757475084, + "step": 1560, + "tokens_trained": 5.111317344 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 0.4453584852138146, + "grad_norm": 7.192091464996338, + "loss": 1.1535, + "loss_ce": 1.241788625717163, + "loss_region": 0.0, + "loss_total": 1.241788625717163, + "lr": 0.0009644850498338871, + "step": 1570, + "tokens_trained": 5.144082784 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 0.4481951634635841, + "grad_norm": 4.517999172210693, + "loss": 1.1356, + "loss_ce": 1.1324427127838135, + "loss_region": 0.0, + "loss_total": 1.1324427127838135, + "lr": 0.0009641528239202658, + "step": 1580, + "tokens_trained": 5.176848224 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 0.45103184171335364, + "grad_norm": 1.4224352836608887, + "loss": 1.1316, + "loss_ce": 1.1069281101226807, + "loss_region": 0.0, + "loss_total": 1.1069281101226807, + "lr": 0.0009638205980066445, + "step": 1590, + "tokens_trained": 5.209613664 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 0.45386851996312316, + "grad_norm": 3.8900578022003174, + "loss": 1.1211, + "loss_ce": 1.1982207298278809, + "loss_region": 0.0, + "loss_total": 1.1982207298278809, + "lr": 0.0009634883720930234, + "step": 1600, + "tokens_trained": 5.242378304 + }, + { + "epoch": 0.45386851996312316, + "eval_ppl": 3.0572565090856805, + "eval_runtime": 1.4289, + "step": 1600, + "tokens_trained": 5.242378304 + }, + { + "epoch": 0.45386851996312316, + "eval_ppl_cds": 3.8094367767208666, + "eval_ppl_dig": 1.20537191204571, + "eval_ppl_exon": 3.45053348841887, + "eval_ppl_intron": 3.097392009256084, + "eval_ppl_nig": 2.9036356964183505, + "eval_ppl_promoter": 3.39861787155953, + "eval_ppl_utr": 3.367378161837914, + "step": 1600, + "tokens_trained": 5.242378304 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 0.4567051982128927, + "grad_norm": 4.516258716583252, + "loss": 1.1154, + "loss_ce": 1.0863256454467773, + "loss_region": 0.0, + "loss_total": 1.0863256454467773, + "lr": 0.0009631561461794021, + "step": 1610, + "tokens_trained": 5.275142944 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 0.4595418764626622, + "grad_norm": 3.7000138759613037, + "loss": 1.1195, + "loss_ce": 1.1473889350891113, + "loss_region": 0.0, + "loss_total": 1.1473889350891113, + "lr": 0.0009628239202657808, + "step": 1620, + "tokens_trained": 5.307906784 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 0.4623785547124317, + "grad_norm": 2.6858463287353516, + "loss": 1.1082, + "loss_ce": 1.0129168033599854, + "loss_region": 0.0, + "loss_total": 1.0129168033599854, + "lr": 0.0009624916943521594, + "step": 1630, + "tokens_trained": 5.340672224 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 0.46521523296220124, + "grad_norm": 2.1516175270080566, + "loss": 1.1098, + "loss_ce": 1.1181179285049438, + "loss_region": 0.0, + "loss_total": 1.1181179285049438, + "lr": 0.0009621594684385382, + "step": 1640, + "tokens_trained": 5.373436896 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 0.46805191121197076, + "grad_norm": 1.8694895505905151, + "loss": 1.1069, + "loss_ce": 1.1573594808578491, + "loss_region": 0.0, + "loss_total": 1.1573594808578491, + "lr": 0.000961827242524917, + "step": 1650, + "tokens_trained": 5.406202336 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 0.4708885894617403, + "grad_norm": 2.570172071456909, + "loss": 1.1039, + "loss_ce": 1.1458351612091064, + "loss_region": 0.0, + "loss_total": 1.1458351612091064, + "lr": 0.0009614950166112957, + "step": 1660, + "tokens_trained": 5.438967776 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 0.4737252677115098, + "grad_norm": 4.836528778076172, + "loss": 1.1044, + "loss_ce": 1.1123778820037842, + "loss_region": 0.0, + "loss_total": 1.1123778820037842, + "lr": 0.0009611627906976744, + "step": 1670, + "tokens_trained": 5.471733216 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 0.4765619459612793, + "grad_norm": 6.293850898742676, + "loss": 1.1025, + "loss_ce": 1.1383477449417114, + "loss_region": 0.0, + "loss_total": 1.1383477449417114, + "lr": 0.0009608305647840532, + "step": 1680, + "tokens_trained": 5.504498656 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 0.47939862421104884, + "grad_norm": 5.655012130737305, + "loss": 1.1037, + "loss_ce": 1.1046979427337646, + "loss_region": 0.0, + "loss_total": 1.1046979427337646, + "lr": 0.0009604983388704319, + "step": 1690, + "tokens_trained": 5.537264096 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 0.48223530246081836, + "grad_norm": 7.232522010803223, + "loss": 1.1028, + "loss_ce": 1.0931048393249512, + "loss_region": 0.0, + "loss_total": 1.0931048393249512, + "lr": 0.0009601661129568107, + "step": 1700, + "tokens_trained": 5.570029536 + }, + { + "epoch": 0.48223530246081836, + "eval_ppl": 3.0822528582502087, + "eval_runtime": 1.4128, + "step": 1700, + "tokens_trained": 5.570029536 + }, + { + "epoch": 0.48223530246081836, + "eval_ppl_cds": 3.9044991106698883, + "eval_ppl_dig": 1.15201642226196, + "eval_ppl_exon": 3.515847409787311, + "eval_ppl_intron": 3.136897299658799, + "eval_ppl_nig": 2.9085136227881985, + "eval_ppl_promoter": 3.4427753407285415, + "eval_ppl_utr": 3.425772875612898, + "step": 1700, + "tokens_trained": 5.570029536 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 0.4850719807105879, + "grad_norm": 4.651060581207275, + "loss": 1.1119, + "loss_ce": 1.0449148416519165, + "loss_region": 0.0, + "loss_total": 1.0449148416519165, + "lr": 0.0009598338870431894, + "step": 1710, + "tokens_trained": 5.602794976 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 0.4879086589603574, + "grad_norm": 3.397057294845581, + "loss": 1.0983, + "loss_ce": 1.073384404182434, + "loss_region": 0.0, + "loss_total": 1.073384404182434, + "lr": 0.0009595016611295682, + "step": 1720, + "tokens_trained": 5.635560416 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 0.4907453372101269, + "grad_norm": 2.0662901401519775, + "loss": 1.0973, + "loss_ce": 1.1481600999832153, + "loss_region": 0.0, + "loss_total": 1.1481600999832153, + "lr": 0.0009591694352159469, + "step": 1730, + "tokens_trained": 5.668325856 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 0.49358201545989644, + "grad_norm": 3.063814401626587, + "loss": 1.1037, + "loss_ce": 1.0913825035095215, + "loss_region": 0.0, + "loss_total": 1.0913825035095215, + "lr": 0.0009588372093023256, + "step": 1740, + "tokens_trained": 5.701091296 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 0.49641869370966596, + "grad_norm": 1.4234209060668945, + "loss": 1.0997, + "loss_ce": 1.1053032875061035, + "loss_region": 0.0, + "loss_total": 1.1053032875061035, + "lr": 0.0009585049833887043, + "step": 1750, + "tokens_trained": 5.733856736 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 0.4992553719594355, + "grad_norm": 2.636101722717285, + "loss": 1.098, + "loss_ce": 1.0762966871261597, + "loss_region": 0.0, + "loss_total": 1.0762966871261597, + "lr": 0.0009581727574750831, + "step": 1760, + "tokens_trained": 5.766622176 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 0.502092050209205, + "grad_norm": 2.4511797428131104, + "loss": 1.0954, + "loss_ce": 1.1882667541503906, + "loss_region": 0.0, + "loss_total": 1.1882667541503906, + "lr": 0.0009578405315614618, + "step": 1770, + "tokens_trained": 5.799387616 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 0.5049287284589745, + "grad_norm": 3.2402634620666504, + "loss": 1.1012, + "loss_ce": 0.9797148704528809, + "loss_region": 0.0, + "loss_total": 0.9797148704528809, + "lr": 0.0009575083056478405, + "step": 1780, + "tokens_trained": 5.832153056 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 0.507765406708744, + "grad_norm": 2.463937997817993, + "loss": 1.0914, + "loss_ce": 0.9584591388702393, + "loss_region": 0.0, + "loss_total": 0.9584591388702393, + "lr": 0.0009571760797342192, + "step": 1790, + "tokens_trained": 5.864918496 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 0.5106020849585136, + "grad_norm": 1.9751344919204712, + "loss": 1.0965, + "loss_ce": 1.0054190158843994, + "loss_region": 0.0, + "loss_total": 1.0054190158843994, + "lr": 0.0009568438538205981, + "step": 1800, + "tokens_trained": 5.897683936 + }, + { + "epoch": 0.5106020849585136, + "eval_ppl": 3.019369577481049, + "eval_runtime": 1.4225, + "step": 1800, + "tokens_trained": 5.897683936 + }, + { + "epoch": 0.5106020849585136, + "eval_ppl_cds": 3.758093371736748, + "eval_ppl_dig": 1.1337671288324114, + "eval_ppl_exon": 3.4437246178256418, + "eval_ppl_intron": 3.0769439394987623, + "eval_ppl_nig": 2.845541068535361, + "eval_ppl_promoter": 3.3619483150766083, + "eval_ppl_utr": 3.3584398152000876, + "step": 1800, + "tokens_trained": 5.897683936 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 0.5134387632082831, + "grad_norm": 2.742611885070801, + "loss": 1.0926, + "loss_ce": 1.0666062831878662, + "loss_region": 0.0, + "loss_total": 1.0666062831878662, + "lr": 0.0009565116279069768, + "step": 1810, + "tokens_trained": 5.930449376 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 0.5162754414580526, + "grad_norm": 2.8147523403167725, + "loss": 1.0898, + "loss_ce": 1.0232893228530884, + "loss_region": 0.0, + "loss_total": 1.0232893228530884, + "lr": 0.0009561794019933555, + "step": 1820, + "tokens_trained": 5.96321104 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 0.5191121197078221, + "grad_norm": 3.127530574798584, + "loss": 1.095, + "loss_ce": 1.1196261644363403, + "loss_region": 0.0, + "loss_total": 1.1196261644363403, + "lr": 0.0009558471760797342, + "step": 1830, + "tokens_trained": 5.99597648 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 0.5219487979575916, + "grad_norm": 2.1857750415802, + "loss": 1.0892, + "loss_ce": 1.0868723392486572, + "loss_region": 0.0, + "loss_total": 1.0868723392486572, + "lr": 0.000955514950166113, + "step": 1840, + "tokens_trained": 6.028741744 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 0.5247854762073612, + "grad_norm": 2.405391216278076, + "loss": 1.0885, + "loss_ce": 1.0642516613006592, + "loss_region": 0.0, + "loss_total": 1.0642516613006592, + "lr": 0.0009551827242524918, + "step": 1850, + "tokens_trained": 6.061507184 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 0.5276221544571307, + "grad_norm": 2.597893476486206, + "loss": 1.0954, + "loss_ce": 1.0345964431762695, + "loss_region": 0.0, + "loss_total": 1.0345964431762695, + "lr": 0.0009548504983388705, + "step": 1860, + "tokens_trained": 6.094268624 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 0.5304588327069002, + "grad_norm": 3.459991455078125, + "loss": 1.0814, + "loss_ce": 1.0414313077926636, + "loss_region": 0.0, + "loss_total": 1.0414313077926636, + "lr": 0.0009545182724252491, + "step": 1870, + "tokens_trained": 6.127034064 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 0.5332955109566697, + "grad_norm": 2.3480570316314697, + "loss": 1.0906, + "loss_ce": 1.1275650262832642, + "loss_region": 0.0, + "loss_total": 1.1275650262832642, + "lr": 0.0009541860465116279, + "step": 1880, + "tokens_trained": 6.159799504 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 0.5361321892064392, + "grad_norm": 3.644726514816284, + "loss": 1.0961, + "loss_ce": 1.0783389806747437, + "loss_region": 0.0, + "loss_total": 1.0783389806747437, + "lr": 0.0009538538205980066, + "step": 1890, + "tokens_trained": 6.192561072 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 0.5389688674562088, + "grad_norm": 3.054255485534668, + "loss": 1.0811, + "loss_ce": 1.1027147769927979, + "loss_region": 0.0, + "loss_total": 1.1027147769927979, + "lr": 0.0009535215946843853, + "step": 1900, + "tokens_trained": 6.225326512 + }, + { + "epoch": 0.5389688674562088, + "eval_ppl": 2.978250393692699, + "eval_runtime": 1.4195, + "step": 1900, + "tokens_trained": 6.225326512 + }, + { + "epoch": 0.5389688674562088, + "eval_ppl_cds": 3.7426133107218376, + "eval_ppl_dig": 1.1281127433791789, + "eval_ppl_exon": 3.3947373144691957, + "eval_ppl_intron": 3.0355446240629873, + "eval_ppl_nig": 2.800047886740616, + "eval_ppl_promoter": 3.330088453583318, + "eval_ppl_utr": 3.3119772364218916, + "step": 1900, + "tokens_trained": 6.225326512 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 0.5418055457059783, + "grad_norm": 3.7149195671081543, + "loss": 1.0993, + "loss_ce": 1.0938997268676758, + "loss_region": 0.0, + "loss_total": 1.0938997268676758, + "lr": 0.0009531893687707642, + "step": 1910, + "tokens_trained": 6.258091952 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 0.5446422239557478, + "grad_norm": 3.3361783027648926, + "loss": 1.0906, + "loss_ce": 1.1457546949386597, + "loss_region": 0.0, + "loss_total": 1.1457546949386597, + "lr": 0.0009528571428571429, + "step": 1920, + "tokens_trained": 6.290857392 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 0.5474789022055173, + "grad_norm": 0.24917560815811157, + "loss": 1.0825, + "loss_ce": 1.0943527221679688, + "loss_region": 0.0, + "loss_total": 1.0943527221679688, + "lr": 0.0009525249169435216, + "step": 1930, + "tokens_trained": 6.323622832 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 0.5503155804552868, + "grad_norm": 1.0417945384979248, + "loss": 1.0898, + "loss_ce": 1.1456317901611328, + "loss_region": 0.0, + "loss_total": 1.1456317901611328, + "lr": 0.0009521926910299003, + "step": 1940, + "tokens_trained": 6.356388272 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 0.5531522587050564, + "grad_norm": 3.4755916595458984, + "loss": 1.087, + "loss_ce": 1.2146564722061157, + "loss_region": 0.0, + "loss_total": 1.2146564722061157, + "lr": 0.000951860465116279, + "step": 1950, + "tokens_trained": 6.389153712 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 0.5559889369548259, + "grad_norm": 3.690385580062866, + "loss": 1.0976, + "loss_ce": 1.1537941694259644, + "loss_region": 0.0, + "loss_total": 1.1537941694259644, + "lr": 0.0009515282392026579, + "step": 1960, + "tokens_trained": 6.421919152 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 0.5588256152045954, + "grad_norm": 2.9730112552642822, + "loss": 1.0955, + "loss_ce": 1.154258370399475, + "loss_region": 0.0, + "loss_total": 1.154258370399475, + "lr": 0.0009511960132890366, + "step": 1970, + "tokens_trained": 6.454684592 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 0.5616622934543649, + "grad_norm": 3.8709256649017334, + "loss": 1.0917, + "loss_ce": 1.1360281705856323, + "loss_region": 0.0, + "loss_total": 1.1360281705856323, + "lr": 0.0009508637873754153, + "step": 1980, + "tokens_trained": 6.487450032 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 0.5644989717041344, + "grad_norm": 2.7163589000701904, + "loss": 1.0932, + "loss_ce": 1.1491780281066895, + "loss_region": 0.0, + "loss_total": 1.1491780281066895, + "lr": 0.0009505315614617941, + "step": 1990, + "tokens_trained": 6.520215472 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 0.567335649953904, + "grad_norm": 2.578860282897949, + "loss": 1.0888, + "loss_ce": 1.146824598312378, + "loss_region": 0.0, + "loss_total": 1.146824598312378, + "lr": 0.0009501993355481727, + "step": 2000, + "tokens_trained": 6.552980912 + }, + { + "epoch": 0.567335649953904, + "eval_ppl": 2.980583177445131, + "eval_runtime": 1.4234, + "step": 2000, + "tokens_trained": 6.552980912 + }, + { + "epoch": 0.567335649953904, + "eval_ppl_cds": 3.7316570872618735, + "eval_ppl_dig": 1.1280334819434104, + "eval_ppl_exon": 3.4095801918035527, + "eval_ppl_intron": 3.0412698629831234, + "eval_ppl_nig": 2.800380321625125, + "eval_ppl_promoter": 3.33147121202148, + "eval_ppl_utr": 3.323323299674321, + "step": 2000, + "tokens_trained": 6.552980912 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 0.5701723282036735, + "grad_norm": 2.27884578704834, + "loss": 1.0884, + "loss_ce": 1.1046146154403687, + "loss_region": 0.0, + "loss_total": 1.1046146154403687, + "lr": 0.0009498671096345515, + "step": 2010, + "tokens_trained": 6.585746352 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 0.573009006453443, + "grad_norm": 2.9008097648620605, + "loss": 1.0938, + "loss_ce": 1.152585506439209, + "loss_region": 0.0, + "loss_total": 1.152585506439209, + "lr": 0.0009495348837209302, + "step": 2020, + "tokens_trained": 6.618511792 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 0.5758456847032125, + "grad_norm": 3.6134374141693115, + "loss": 1.0898, + "loss_ce": 1.1322474479675293, + "loss_region": 0.0, + "loss_total": 1.1322474479675293, + "lr": 0.000949202657807309, + "step": 2030, + "tokens_trained": 6.651277232 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 0.578682362952982, + "grad_norm": 3.1809916496276855, + "loss": 1.09, + "loss_ce": 1.1081128120422363, + "loss_region": 0.0, + "loss_total": 1.1081128120422363, + "lr": 0.0009488704318936877, + "step": 2040, + "tokens_trained": 6.684041872 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 0.5815190412027516, + "grad_norm": 1.5797704458236694, + "loss": 1.0883, + "loss_ce": 1.1581259965896606, + "loss_region": 0.0, + "loss_total": 1.1581259965896606, + "lr": 0.0009485382059800664, + "step": 2050, + "tokens_trained": 6.716806512 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 0.5843557194525211, + "grad_norm": 2.6460490226745605, + "loss": 1.0903, + "loss_ce": 1.1154712438583374, + "loss_region": 0.0, + "loss_total": 1.1154712438583374, + "lr": 0.0009482059800664452, + "step": 2060, + "tokens_trained": 6.749571952 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 0.5871923977022906, + "grad_norm": 1.8731824159622192, + "loss": 1.0866, + "loss_ce": 1.040235161781311, + "loss_region": 0.0, + "loss_total": 1.040235161781311, + "lr": 0.000947873754152824, + "step": 2070, + "tokens_trained": 6.782337392 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 0.5900290759520601, + "grad_norm": 1.5950156450271606, + "loss": 1.0971, + "loss_ce": 1.1099926233291626, + "loss_region": 0.0, + "loss_total": 1.1099926233291626, + "lr": 0.0009475415282392027, + "step": 2080, + "tokens_trained": 6.815102832 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 0.5928657542018296, + "grad_norm": 1.787434458732605, + "loss": 1.0874, + "loss_ce": 1.0939483642578125, + "loss_region": 0.0, + "loss_total": 1.0939483642578125, + "lr": 0.0009472093023255814, + "step": 2090, + "tokens_trained": 6.847868272 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 0.5957024324515992, + "grad_norm": 3.179715156555176, + "loss": 1.0932, + "loss_ce": 1.2126123905181885, + "loss_region": 0.0, + "loss_total": 1.2126123905181885, + "lr": 0.0009468770764119601, + "step": 2100, + "tokens_trained": 6.880633712 + }, + { + "epoch": 0.5957024324515992, + "eval_ppl": 2.9830714078926, + "eval_runtime": 1.43, + "step": 2100, + "tokens_trained": 6.880633712 + }, + { + "epoch": 0.5957024324515992, + "eval_ppl_cds": 3.728204782590528, + "eval_ppl_dig": 1.1378440232827705, + "eval_ppl_exon": 3.4166593341234512, + "eval_ppl_intron": 3.044262455686869, + "eval_ppl_nig": 2.80005328927199, + "eval_ppl_promoter": 3.334691976240749, + "eval_ppl_utr": 3.3299644915488593, + "step": 2100, + "tokens_trained": 6.880633712 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 0.5985391107013687, + "grad_norm": 3.3825082778930664, + "loss": 1.0868, + "loss_ce": 1.109216570854187, + "loss_region": 0.0, + "loss_total": 1.109216570854187, + "lr": 0.000946544850498339, + "step": 2110, + "tokens_trained": 6.913397016 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 0.6013757889511382, + "grad_norm": 3.188910722732544, + "loss": 1.1193, + "loss_ce": 1.0802541971206665, + "loss_region": 0.0, + "loss_total": 1.0802541971206665, + "lr": 0.0009462126245847176, + "step": 2120, + "tokens_trained": 6.946162296 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 0.6042124672009077, + "grad_norm": 217.5547332763672, + "loss": 1.1262, + "loss_ce": 1.0244488716125488, + "loss_region": 0.0, + "loss_total": 1.0244488716125488, + "lr": 0.0009458803986710963, + "step": 2130, + "tokens_trained": 6.978927736 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 0.6070491454506772, + "grad_norm": 3.4189093112945557, + "loss": 1.1742, + "loss_ce": 1.115725040435791, + "loss_region": 0.0, + "loss_total": 1.115725040435791, + "lr": 0.000945548172757475, + "step": 2140, + "tokens_trained": 7.011693176 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 0.6098858237004467, + "grad_norm": 4.4236884117126465, + "loss": 1.1729, + "loss_ce": 1.175723910331726, + "loss_region": 0.0, + "loss_total": 1.175723910331726, + "lr": 0.0009452159468438538, + "step": 2150, + "tokens_trained": 7.044458616 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 0.6127225019502163, + "grad_norm": 2.3447749614715576, + "loss": 1.1618, + "loss_ce": 1.1432043313980103, + "loss_region": 0.0, + "loss_total": 1.1432043313980103, + "lr": 0.0009448837209302326, + "step": 2160, + "tokens_trained": 7.077224056 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 0.6155591801999858, + "grad_norm": 3.068040370941162, + "loss": 1.1306, + "loss_ce": 1.1193426847457886, + "loss_region": 0.0, + "loss_total": 1.1193426847457886, + "lr": 0.0009445514950166113, + "step": 2170, + "tokens_trained": 7.109989496 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 0.6183958584497553, + "grad_norm": 5.466696262359619, + "loss": 1.1397, + "loss_ce": 1.1701653003692627, + "loss_region": 0.0, + "loss_total": 1.1701653003692627, + "lr": 0.00094421926910299, + "step": 2180, + "tokens_trained": 7.142754936 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 0.6212325366995248, + "grad_norm": 4.5005621910095215, + "loss": 1.1339, + "loss_ce": 1.171733021736145, + "loss_region": 0.0, + "loss_total": 1.171733021736145, + "lr": 0.0009438870431893688, + "step": 2190, + "tokens_trained": 7.175520376 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 0.6240692149492943, + "grad_norm": 3.207864761352539, + "loss": 1.1129, + "loss_ce": 1.1302906274795532, + "loss_region": 0.0, + "loss_total": 1.1302906274795532, + "lr": 0.0009435548172757475, + "step": 2200, + "tokens_trained": 7.208285816 + }, + { + "epoch": 0.6240692149492943, + "eval_ppl": 3.0183597634348662, + "eval_runtime": 1.4191, + "step": 2200, + "tokens_trained": 7.208285816 + }, + { + "epoch": 0.6240692149492943, + "eval_ppl_cds": 3.739885513898486, + "eval_ppl_dig": 1.160543835046077, + "eval_ppl_exon": 3.4314628863756824, + "eval_ppl_intron": 3.0703161661522844, + "eval_ppl_nig": 2.8508472425765037, + "eval_ppl_promoter": 3.3584220153870468, + "eval_ppl_utr": 3.3553297923253633, + "step": 2200, + "tokens_trained": 7.208285816 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 0.6269058931990639, + "grad_norm": 2.8716187477111816, + "loss": 1.1065, + "loss_ce": 1.1441963911056519, + "loss_region": 0.0, + "loss_total": 1.1441963911056519, + "lr": 0.0009432225913621263, + "step": 2210, + "tokens_trained": 7.241051256 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 0.6297425714488334, + "grad_norm": 3.377782106399536, + "loss": 1.1095, + "loss_ce": 1.0713433027267456, + "loss_region": 0.0, + "loss_total": 1.0713433027267456, + "lr": 0.0009428903654485051, + "step": 2220, + "tokens_trained": 7.273816696 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 0.6325792496986029, + "grad_norm": 3.2118241786956787, + "loss": 1.1187, + "loss_ce": 1.1656153202056885, + "loss_region": 0.0, + "loss_total": 1.1656153202056885, + "lr": 0.0009425581395348838, + "step": 2230, + "tokens_trained": 7.306582136 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 0.6354159279483724, + "grad_norm": 2.515864610671997, + "loss": 1.1356, + "loss_ce": 1.0824023485183716, + "loss_region": 0.0, + "loss_total": 1.0824023485183716, + "lr": 0.0009422259136212625, + "step": 2240, + "tokens_trained": 7.339347576 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 0.638252606198142, + "grad_norm": 5.04034948348999, + "loss": 1.179, + "loss_ce": 1.2054611444473267, + "loss_region": 0.0, + "loss_total": 1.2054611444473267, + "lr": 0.0009418936877076411, + "step": 2250, + "tokens_trained": 7.372113016 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 0.6410892844479115, + "grad_norm": 71.43251037597656, + "loss": 1.2651, + "loss_ce": 1.2028636932373047, + "loss_region": 0.0, + "loss_total": 1.2028636932373047, + "lr": 0.00094156146179402, + "step": 2260, + "tokens_trained": 7.404878456 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 0.643925962697681, + "grad_norm": 201.0874786376953, + "loss": 1.3411, + "loss_ce": 1.359194278717041, + "loss_region": 0.0, + "loss_total": 1.359194278717041, + "lr": 0.0009412292358803987, + "step": 2270, + "tokens_trained": 7.437643896 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 0.6467626409474505, + "grad_norm": 141.78282165527344, + "loss": 1.3679, + "loss_ce": 1.3799196481704712, + "loss_region": 0.0, + "loss_total": 1.3799196481704712, + "lr": 0.0009408970099667774, + "step": 2280, + "tokens_trained": 7.470409336 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 0.64959931919722, + "grad_norm": 15.33470630645752, + "loss": 1.376, + "loss_ce": 1.3648813962936401, + "loss_region": 0.0, + "loss_total": 1.3648813962936401, + "lr": 0.0009405647840531561, + "step": 2290, + "tokens_trained": 7.503173472 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 0.6524359974469895, + "grad_norm": 21.829694747924805, + "loss": 1.3893, + "loss_ce": 1.3987280130386353, + "loss_region": 0.0, + "loss_total": 1.3987280130386353, + "lr": 0.0009402325581395349, + "step": 2300, + "tokens_trained": 7.535938912 + }, + { + "epoch": 0.6524359974469895, + "eval_ppl": 3.912632289090429, + "eval_runtime": 1.4168, + "step": 2300, + "tokens_trained": 7.535938912 + }, + { + "epoch": 0.6524359974469895, + "eval_ppl_cds": 4.008499374660205, + "eval_ppl_dig": 3.903481857187947, + "eval_ppl_exon": 3.923212343837822, + "eval_ppl_intron": 3.9012396038965833, + "eval_ppl_nig": 3.8973300936488267, + "eval_ppl_promoter": 3.94619457999195, + "eval_ppl_utr": 3.8998336771800752, + "step": 2300, + "tokens_trained": 7.535938912 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 0.6552726756967591, + "grad_norm": 12.745903015136719, + "loss": 1.3718, + "loss_ce": 1.3493191003799438, + "loss_region": 0.0, + "loss_total": 1.3493191003799438, + "lr": 0.0009399003322259137, + "step": 2310, + "tokens_trained": 7.568704352 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 0.6581093539465286, + "grad_norm": 23.916555404663086, + "loss": 1.3701, + "loss_ce": 1.4017199277877808, + "loss_region": 0.0, + "loss_total": 1.4017199277877808, + "lr": 0.0009395681063122924, + "step": 2320, + "tokens_trained": 7.601469792 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 0.6609460321962981, + "grad_norm": 18.63468360900879, + "loss": 1.3834, + "loss_ce": 1.4090750217437744, + "loss_region": 0.0, + "loss_total": 1.4090750217437744, + "lr": 0.0009392358803986711, + "step": 2330, + "tokens_trained": 7.634233608 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 0.6637827104460676, + "grad_norm": 18.90869140625, + "loss": 1.3664, + "loss_ce": 1.3705028295516968, + "loss_region": 0.0, + "loss_total": 1.3705028295516968, + "lr": 0.0009389036544850499, + "step": 2340, + "tokens_trained": 7.666998248 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 0.6666193886958371, + "grad_norm": 11.569411277770996, + "loss": 1.3783, + "loss_ce": 1.3593852519989014, + "loss_region": 0.0, + "loss_total": 1.3593852519989014, + "lr": 0.0009385714285714286, + "step": 2350, + "tokens_trained": 7.699763688 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 0.6694560669456067, + "grad_norm": 13.769268035888672, + "loss": 1.3715, + "loss_ce": 1.3537636995315552, + "loss_region": 0.0, + "loss_total": 1.3537636995315552, + "lr": 0.0009382392026578074, + "step": 2360, + "tokens_trained": 7.732529128 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 0.6722927451953762, + "grad_norm": 7.519562721252441, + "loss": 1.3794, + "loss_ce": 1.3774657249450684, + "loss_region": 0.0, + "loss_total": 1.3774657249450684, + "lr": 0.000937906976744186, + "step": 2370, + "tokens_trained": 7.765294568 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 0.6751294234451457, + "grad_norm": 25.69373321533203, + "loss": 1.4039, + "loss_ce": 1.376942753791809, + "loss_region": 0.0, + "loss_total": 1.376942753791809, + "lr": 0.0009375747508305648, + "step": 2380, + "tokens_trained": 7.798060008 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 0.6779661016949152, + "grad_norm": 26.833168029785156, + "loss": 1.4133, + "loss_ce": 1.4939353466033936, + "loss_region": 0.0, + "loss_total": 1.4939353466033936, + "lr": 0.0009372425249169435, + "step": 2390, + "tokens_trained": 7.830825448 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 0.6808027799446847, + "grad_norm": 85.41537475585938, + "loss": 1.4022, + "loss_ce": 1.352198839187622, + "loss_region": 0.0, + "loss_total": 1.352198839187622, + "lr": 0.0009369102990033222, + "step": 2400, + "tokens_trained": 7.863590888 + }, + { + "epoch": 0.6808027799446847, + "eval_ppl": 3.944675389152555, + "eval_runtime": 1.418, + "step": 2400, + "tokens_trained": 7.863590888 + }, + { + "epoch": 0.6808027799446847, + "eval_ppl_cds": 4.040480811612362, + "eval_ppl_dig": 3.891537910725513, + "eval_ppl_exon": 3.940749905900709, + "eval_ppl_intron": 3.931337409911382, + "eval_ppl_nig": 3.9180442166081586, + "eval_ppl_promoter": 4.000068866092487, + "eval_ppl_utr": 3.9241575601297174, + "step": 2400, + "tokens_trained": 7.863590888 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 0.6836394581944543, + "grad_norm": 8.55296516418457, + "loss": 1.3652, + "loss_ce": 1.365032434463501, + "loss_region": 0.0, + "loss_total": 1.365032434463501, + "lr": 0.000936578073089701, + "step": 2410, + "tokens_trained": 7.896356328 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 0.6864761364442238, + "grad_norm": 18.1574764251709, + "loss": 1.3818, + "loss_ce": 1.3817418813705444, + "loss_region": 0.0, + "loss_total": 1.3817418813705444, + "lr": 0.0009362458471760798, + "step": 2420, + "tokens_trained": 7.929121768 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 0.6893128146939933, + "grad_norm": 14.72149658203125, + "loss": 1.3844, + "loss_ce": 1.3504234552383423, + "loss_region": 0.0, + "loss_total": 1.3504234552383423, + "lr": 0.0009359136212624585, + "step": 2430, + "tokens_trained": 7.961887208 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 0.6921494929437628, + "grad_norm": 69.32920837402344, + "loss": 1.3583, + "loss_ce": 1.3607655763626099, + "loss_region": 0.0, + "loss_total": 1.3607655763626099, + "lr": 0.0009355813953488372, + "step": 2440, + "tokens_trained": 7.994651848 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 0.6949861711935323, + "grad_norm": 170.44522094726562, + "loss": 1.364, + "loss_ce": 1.3573222160339355, + "loss_region": 0.0, + "loss_total": 1.3573222160339355, + "lr": 0.0009352491694352159, + "step": 2450, + "tokens_trained": 8.027417288 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 0.6978228494433019, + "grad_norm": 14.767860412597656, + "loss": 1.355, + "loss_ce": 1.3655263185501099, + "loss_region": 0.0, + "loss_total": 1.3655263185501099, + "lr": 0.0009349169435215948, + "step": 2460, + "tokens_trained": 8.060182704 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 0.7006595276930714, + "grad_norm": 96.6727523803711, + "loss": 1.4019, + "loss_ce": 1.4026390314102173, + "loss_region": 0.0, + "loss_total": 1.4026390314102173, + "lr": 0.0009345847176079735, + "step": 2470, + "tokens_trained": 8.092948144 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 0.7034962059428409, + "grad_norm": 46.399166107177734, + "loss": 1.3902, + "loss_ce": 1.3999183177947998, + "loss_region": 0.0, + "loss_total": 1.3999183177947998, + "lr": 0.0009342524916943522, + "step": 2480, + "tokens_trained": 8.125713584 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 0.7063328841926104, + "grad_norm": 39.687950134277344, + "loss": 1.3773, + "loss_ce": 1.3881598711013794, + "loss_region": 0.0, + "loss_total": 1.3881598711013794, + "lr": 0.000933920265780731, + "step": 2490, + "tokens_trained": 8.158479016 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 0.7091695624423799, + "grad_norm": 23.020082473754883, + "loss": 1.3638, + "loss_ce": 1.3531800508499146, + "loss_region": 0.0, + "loss_total": 1.3531800508499146, + "lr": 0.0009335880398671096, + "step": 2500, + "tokens_trained": 8.191244456 + }, + { + "epoch": 0.7091695624423799, + "eval_ppl": 3.8538607443920245, + "eval_runtime": 1.4152, + "step": 2500, + "tokens_trained": 8.191244456 + }, + { + "epoch": 0.7091695624423799, + "eval_ppl_cds": 3.9287843303483947, + "eval_ppl_dig": 3.899126362815793, + "eval_ppl_exon": 3.8464106751903793, + "eval_ppl_intron": 3.8453944552574524, + "eval_ppl_nig": 3.8421122963060013, + "eval_ppl_promoter": 3.8824886149253857, + "eval_ppl_utr": 3.833312837496713, + "step": 2500, + "tokens_trained": 8.191244456 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 0.7120062406921495, + "grad_norm": 52.68931579589844, + "loss": 1.3533, + "loss_ce": 1.3587932586669922, + "loss_region": 0.0, + "loss_total": 1.3587932586669922, + "lr": 0.0009332558139534884, + "step": 2510, + "tokens_trained": 8.224009896 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 0.714842918941919, + "grad_norm": 19.883438110351562, + "loss": 1.3642, + "loss_ce": 1.3554009199142456, + "loss_region": 0.0, + "loss_total": 1.3554009199142456, + "lr": 0.0009329235880398671, + "step": 2520, + "tokens_trained": 8.256775336 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 0.7176795971916885, + "grad_norm": 53.332950592041016, + "loss": 1.3559, + "loss_ce": 1.3322890996932983, + "loss_region": 0.0, + "loss_total": 1.3322890996932983, + "lr": 0.0009325913621262458, + "step": 2530, + "tokens_trained": 8.289540776 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 0.720516275441458, + "grad_norm": 84.58511352539062, + "loss": 1.3482, + "loss_ce": 1.3485150337219238, + "loss_region": 0.0, + "loss_total": 1.3485150337219238, + "lr": 0.0009322591362126246, + "step": 2540, + "tokens_trained": 8.322306216 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 0.7233529536912275, + "grad_norm": 13.718729019165039, + "loss": 1.3445, + "loss_ce": 1.348463773727417, + "loss_region": 0.0, + "loss_total": 1.348463773727417, + "lr": 0.0009319269102990033, + "step": 2550, + "tokens_trained": 8.355071656 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 0.7261896319409971, + "grad_norm": 44.61623764038086, + "loss": 1.378, + "loss_ce": 1.3441184759140015, + "loss_region": 0.0, + "loss_total": 1.3441184759140015, + "lr": 0.0009315946843853821, + "step": 2560, + "tokens_trained": 8.387835072 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 0.7290263101907666, + "grad_norm": 37.00033950805664, + "loss": 1.3739, + "loss_ce": 1.3689384460449219, + "loss_region": 0.0, + "loss_total": 1.3689384460449219, + "lr": 0.0009312624584717609, + "step": 2570, + "tokens_trained": 8.420600512 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 0.7318629884405361, + "grad_norm": 6.434602737426758, + "loss": 1.3796, + "loss_ce": 1.3670883178710938, + "loss_region": 0.0, + "loss_total": 1.3670883178710938, + "lr": 0.0009309302325581396, + "step": 2580, + "tokens_trained": 8.453365928 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 0.7346996666903056, + "grad_norm": 16.95655059814453, + "loss": 1.3719, + "loss_ce": 1.3993109464645386, + "loss_region": 0.0, + "loss_total": 1.3993109464645386, + "lr": 0.0009305980066445183, + "step": 2590, + "tokens_trained": 8.486131368 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 0.7375363449400751, + "grad_norm": 7.510871410369873, + "loss": 1.3584, + "loss_ce": 1.346716046333313, + "loss_region": 0.0, + "loss_total": 1.346716046333313, + "lr": 0.000930265780730897, + "step": 2600, + "tokens_trained": 8.518896808 + }, + { + "epoch": 0.7375363449400751, + "eval_ppl": 3.89476075434051, + "eval_runtime": 1.4414, + "step": 2600, + "tokens_trained": 8.518896808 + }, + { + "epoch": 0.7375363449400751, + "eval_ppl_cds": 4.034369651087086, + "eval_ppl_dig": 3.886839255333321, + "eval_ppl_exon": 3.8940573588358847, + "eval_ppl_intron": 3.879074664115795, + "eval_ppl_nig": 3.8629925282180215, + "eval_ppl_promoter": 3.9560068162253033, + "eval_ppl_utr": 3.8646450083867756, + "step": 2600, + "tokens_trained": 8.518896808 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 0.7403730231898447, + "grad_norm": 6.940410614013672, + "loss": 1.3536, + "loss_ce": 1.3426551818847656, + "loss_region": 0.0, + "loss_total": 1.3426551818847656, + "lr": 0.0009299335548172759, + "step": 2610, + "tokens_trained": 8.551662248 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 0.7432097014396142, + "grad_norm": 4.374493598937988, + "loss": 1.3383, + "loss_ce": 1.3372553586959839, + "loss_region": 0.0, + "loss_total": 1.3372553586959839, + "lr": 0.0009296013289036545, + "step": 2620, + "tokens_trained": 8.584426888 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 0.7460463796893837, + "grad_norm": 5.0137763023376465, + "loss": 1.3339, + "loss_ce": 1.3266304731369019, + "loss_region": 0.0, + "loss_total": 1.3266304731369019, + "lr": 0.0009292691029900332, + "step": 2630, + "tokens_trained": 8.617192328 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 0.7488830579391532, + "grad_norm": 17.254764556884766, + "loss": 1.3319, + "loss_ce": 1.3472967147827148, + "loss_region": 0.0, + "loss_total": 1.3472967147827148, + "lr": 0.0009289368770764119, + "step": 2640, + "tokens_trained": 8.649951656 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 0.7517197361889227, + "grad_norm": 3.9123036861419678, + "loss": 1.3453, + "loss_ce": 1.3339078426361084, + "loss_region": 0.0, + "loss_total": 1.3339078426361084, + "lr": 0.0009286046511627907, + "step": 2650, + "tokens_trained": 8.682717096 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 0.7545564144386923, + "grad_norm": 1.3326530456542969, + "loss": 1.332, + "loss_ce": 1.3290140628814697, + "loss_region": 0.0, + "loss_total": 1.3290140628814697, + "lr": 0.0009282724252491695, + "step": 2660, + "tokens_trained": 8.715482536 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 0.7573930926884618, + "grad_norm": 3.0831685066223145, + "loss": 1.3464, + "loss_ce": 1.326156735420227, + "loss_region": 0.0, + "loss_total": 1.326156735420227, + "lr": 0.0009279401993355482, + "step": 2670, + "tokens_trained": 8.748247976 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 0.7602297709382313, + "grad_norm": 2.041679620742798, + "loss": 1.3359, + "loss_ce": 1.3283050060272217, + "loss_region": 0.0, + "loss_total": 1.3283050060272217, + "lr": 0.0009276079734219269, + "step": 2680, + "tokens_trained": 8.781013416 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 0.7630664491880008, + "grad_norm": 3.6898534297943115, + "loss": 1.3246, + "loss_ce": 1.3232933282852173, + "loss_region": 0.0, + "loss_total": 1.3232933282852173, + "lr": 0.0009272757475083057, + "step": 2690, + "tokens_trained": 8.813778696 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 0.7659031274377703, + "grad_norm": 4.392417907714844, + "loss": 1.3216, + "loss_ce": 1.3262367248535156, + "loss_region": 0.0, + "loss_total": 1.3262367248535156, + "lr": 0.0009269435215946844, + "step": 2700, + "tokens_trained": 8.846544136 + }, + { + "epoch": 0.7659031274377703, + "eval_ppl": 3.749328339657657, + "eval_runtime": 1.4343, + "step": 2700, + "tokens_trained": 8.846544136 + }, + { + "epoch": 0.7659031274377703, + "eval_ppl_cds": 3.875951664697824, + "eval_ppl_dig": 3.908276969955482, + "eval_ppl_exon": 3.756658238907496, + "eval_ppl_intron": 3.7376065932616234, + "eval_ppl_nig": 3.734867104137474, + "eval_ppl_promoter": 3.784194180093443, + "eval_ppl_utr": 3.74098477339548, + "step": 2700, + "tokens_trained": 8.846544136 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 0.7687398056875399, + "grad_norm": 2.882650136947632, + "loss": 1.32, + "loss_ce": 1.313364863395691, + "loss_region": 0.0, + "loss_total": 1.313364863395691, + "lr": 0.0009266112956810632, + "step": 2710, + "tokens_trained": 8.879309576 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 0.7715764839373094, + "grad_norm": 5.584534168243408, + "loss": 1.3205, + "loss_ce": 1.3270784616470337, + "loss_region": 0.0, + "loss_total": 1.3270784616470337, + "lr": 0.0009262790697674419, + "step": 2720, + "tokens_trained": 8.912075016 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 0.7744131621870789, + "grad_norm": 2.6551883220672607, + "loss": 1.318, + "loss_ce": 1.308359980583191, + "loss_region": 0.0, + "loss_total": 1.308359980583191, + "lr": 0.0009259468438538207, + "step": 2730, + "tokens_trained": 8.944840456 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 0.7772498404368484, + "grad_norm": 1.0403350591659546, + "loss": 1.3104, + "loss_ce": 1.2952345609664917, + "loss_region": 0.0, + "loss_total": 1.2952345609664917, + "lr": 0.0009256146179401993, + "step": 2740, + "tokens_trained": 8.977605896 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 0.7800865186866179, + "grad_norm": 4.945831298828125, + "loss": 1.3065, + "loss_ce": 1.3421580791473389, + "loss_region": 0.0, + "loss_total": 1.3421580791473389, + "lr": 0.000925282392026578, + "step": 2750, + "tokens_trained": 9.010371336 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 0.7829231969363875, + "grad_norm": 3.3751721382141113, + "loss": 1.3042, + "loss_ce": 1.3084043264389038, + "loss_region": 0.0, + "loss_total": 1.3084043264389038, + "lr": 0.0009249501661129568, + "step": 2760, + "tokens_trained": 9.043136776 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 0.785759875186157, + "grad_norm": 2.4980247020721436, + "loss": 1.2819, + "loss_ce": 1.291877031326294, + "loss_region": 0.0, + "loss_total": 1.291877031326294, + "lr": 0.0009246179401993356, + "step": 2770, + "tokens_trained": 9.075902216 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 0.7885965534359265, + "grad_norm": 1.511878490447998, + "loss": 1.2625, + "loss_ce": 1.236128807067871, + "loss_region": 0.0, + "loss_total": 1.236128807067871, + "lr": 0.0009242857142857143, + "step": 2780, + "tokens_trained": 9.108667656 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 0.791433231685696, + "grad_norm": 2.7031407356262207, + "loss": 1.2448, + "loss_ce": 1.255333662033081, + "loss_region": 0.0, + "loss_total": 1.255333662033081, + "lr": 0.000923953488372093, + "step": 2790, + "tokens_trained": 9.14143004 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 0.7942699099354655, + "grad_norm": 1.1955639123916626, + "loss": 1.2329, + "loss_ce": 1.2303880453109741, + "loss_region": 0.0, + "loss_total": 1.2303880453109741, + "lr": 0.0009236212624584717, + "step": 2800, + "tokens_trained": 9.17419548 + }, + { + "epoch": 0.7942699099354655, + "eval_ppl": 3.4107677521636, + "eval_runtime": 1.4342, + "step": 2800, + "tokens_trained": 9.17419548 + }, + { + "epoch": 0.7942699099354655, + "eval_ppl_cds": 3.8302389009168465, + "eval_ppl_dig": 3.783552356921221, + "eval_ppl_exon": 3.5896442191927225, + "eval_ppl_intron": 3.3909066180672185, + "eval_ppl_nig": 3.403709923115435, + "eval_ppl_promoter": 3.5545247298677314, + "eval_ppl_utr": 3.5221359660807963, + "step": 2800, + "tokens_trained": 9.17419548 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 0.797106588185235, + "grad_norm": 1.4083456993103027, + "loss": 1.2221, + "loss_ce": 1.2004368305206299, + "loss_region": 0.0, + "loss_total": 1.2004368305206299, + "lr": 0.0009232890365448506, + "step": 2810, + "tokens_trained": 9.20696092 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 0.7999432664350046, + "grad_norm": 1.30550217628479, + "loss": 1.2148, + "loss_ce": 1.171217918395996, + "loss_region": 0.0, + "loss_total": 1.171217918395996, + "lr": 0.0009229568106312293, + "step": 2820, + "tokens_trained": 9.23972636 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 0.8027799446847741, + "grad_norm": 0.8321244716644287, + "loss": 1.2117, + "loss_ce": 1.197412371635437, + "loss_region": 0.0, + "loss_total": 1.197412371635437, + "lr": 0.000922624584717608, + "step": 2830, + "tokens_trained": 9.2724918 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 0.8056166229345436, + "grad_norm": 1.6455938816070557, + "loss": 1.206, + "loss_ce": 1.225730538368225, + "loss_region": 0.0, + "loss_total": 1.225730538368225, + "lr": 0.0009222923588039868, + "step": 2840, + "tokens_trained": 9.30525692 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 0.8084533011843131, + "grad_norm": 1.1820716857910156, + "loss": 1.2045, + "loss_ce": 1.208493709564209, + "loss_region": 0.0, + "loss_total": 1.208493709564209, + "lr": 0.0009219601328903655, + "step": 2850, + "tokens_trained": 9.33802236 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 0.8112899794340827, + "grad_norm": 1.0212680101394653, + "loss": 1.2065, + "loss_ce": 1.1991009712219238, + "loss_region": 0.0, + "loss_total": 1.1991009712219238, + "lr": 0.0009216279069767443, + "step": 2860, + "tokens_trained": 9.3707878 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 0.8141266576838522, + "grad_norm": 1.439316987991333, + "loss": 1.2016, + "loss_ce": 1.2080260515213013, + "loss_region": 0.0, + "loss_total": 1.2080260515213013, + "lr": 0.0009212956810631229, + "step": 2870, + "tokens_trained": 9.403548272 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 0.8169633359336217, + "grad_norm": 1.4545512199401855, + "loss": 1.2, + "loss_ce": 1.1450130939483643, + "loss_region": 0.0, + "loss_total": 1.1450130939483643, + "lr": 0.0009209634551495017, + "step": 2880, + "tokens_trained": 9.436313712 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 0.8198000141833912, + "grad_norm": 1.202152967453003, + "loss": 1.1955, + "loss_ce": 1.2488645315170288, + "loss_region": 0.0, + "loss_total": 1.2488645315170288, + "lr": 0.0009206312292358804, + "step": 2890, + "tokens_trained": 9.469079152 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 0.8226366924331607, + "grad_norm": 1.8360052108764648, + "loss": 1.1974, + "loss_ce": 1.1611562967300415, + "loss_region": 0.0, + "loss_total": 1.1611562967300415, + "lr": 0.0009202990033222591, + "step": 2900, + "tokens_trained": 9.50184356 + }, + { + "epoch": 0.8226366924331607, + "eval_ppl": 3.2947074125284144, + "eval_runtime": 1.4216, + "step": 2900, + "tokens_trained": 9.50184356 + }, + { + "epoch": 0.8226366924331607, + "eval_ppl_cds": 3.809386386261858, + "eval_ppl_dig": 2.5893887228002894, + "eval_ppl_exon": 3.541010455768117, + "eval_ppl_intron": 3.293188997847933, + "eval_ppl_nig": 3.29005420231247, + "eval_ppl_promoter": 3.483675750001279, + "eval_ppl_utr": 3.46085088546961, + "step": 2900, + "tokens_trained": 9.50184356 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 0.8254733706829303, + "grad_norm": 1.3830058574676514, + "loss": 1.1891, + "loss_ce": 1.0914971828460693, + "loss_region": 0.0, + "loss_total": 1.0914971828460693, + "lr": 0.0009199667774086378, + "step": 2910, + "tokens_trained": 9.534608992 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 0.8283100489326998, + "grad_norm": 1.3291281461715698, + "loss": 1.1829, + "loss_ce": 1.2339963912963867, + "loss_region": 0.0, + "loss_total": 1.2339963912963867, + "lr": 0.0009196345514950167, + "step": 2920, + "tokens_trained": 9.567373632 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 0.8311467271824693, + "grad_norm": 0.6280083656311035, + "loss": 1.1807, + "loss_ce": 1.198470950126648, + "loss_region": 0.0, + "loss_total": 1.198470950126648, + "lr": 0.0009193023255813954, + "step": 2930, + "tokens_trained": 9.600139072 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 0.8339834054322388, + "grad_norm": 3.312791585922241, + "loss": 1.1745, + "loss_ce": 1.1420371532440186, + "loss_region": 0.0, + "loss_total": 1.1420371532440186, + "lr": 0.0009189700996677741, + "step": 2940, + "tokens_trained": 9.632904512 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 0.8368200836820083, + "grad_norm": 1.1626808643341064, + "loss": 1.1756, + "loss_ce": 1.0361547470092773, + "loss_region": 0.0, + "loss_total": 1.0361547470092773, + "lr": 0.0009186378737541528, + "step": 2950, + "tokens_trained": 9.665669952 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 0.8396567619317779, + "grad_norm": 2.0363259315490723, + "loss": 1.1694, + "loss_ce": 1.2019656896591187, + "loss_region": 0.0, + "loss_total": 1.2019656896591187, + "lr": 0.0009183056478405316, + "step": 2960, + "tokens_trained": 9.698432616 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 0.8424934401815474, + "grad_norm": 0.2531447410583496, + "loss": 1.1611, + "loss_ce": 1.194112777709961, + "loss_region": 0.0, + "loss_total": 1.194112777709961, + "lr": 0.0009179734219269104, + "step": 2970, + "tokens_trained": 9.731196464 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 0.8453301184313169, + "grad_norm": 0.4076802432537079, + "loss": 1.1671, + "loss_ce": 1.1872175931930542, + "loss_region": 0.0, + "loss_total": 1.1872175931930542, + "lr": 0.0009176411960132891, + "step": 2980, + "tokens_trained": 9.763955848 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 0.8481667966810864, + "grad_norm": 2.1375160217285156, + "loss": 1.1643, + "loss_ce": 1.1913644075393677, + "loss_region": 0.0, + "loss_total": 1.1913644075393677, + "lr": 0.0009173089700996677, + "step": 2990, + "tokens_trained": 9.796721288 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 0.8510034749308559, + "grad_norm": 0.7846505641937256, + "loss": 1.1642, + "loss_ce": 1.1912487745285034, + "loss_region": 0.0, + "loss_total": 1.1912487745285034, + "lr": 0.0009169767441860465, + "step": 3000, + "tokens_trained": 9.829486728 + }, + { + "epoch": 0.8510034749308559, + "eval_ppl": 3.189720605979384, + "eval_runtime": 1.4208, + "step": 3000, + "tokens_trained": 9.829486728 + }, + { + "epoch": 0.8510034749308559, + "eval_ppl_cds": 3.811933197586622, + "eval_ppl_dig": 1.29561654613828, + "eval_ppl_exon": 3.5369261867906636, + "eval_ppl_intron": 3.2197916654702157, + "eval_ppl_nig": 3.0858729844372834, + "eval_ppl_promoter": 3.477562882943513, + "eval_ppl_utr": 3.447236976646492, + "step": 3000, + "tokens_trained": 9.829486728 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 0.8538401531806254, + "grad_norm": 2.160705804824829, + "loss": 1.1495, + "loss_ce": 1.1959177255630493, + "loss_region": 0.0, + "loss_total": 1.1959177255630493, + "lr": 0.0009166445182724252, + "step": 3010, + "tokens_trained": 9.862252168 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 0.856676831430395, + "grad_norm": 0.6243520975112915, + "loss": 1.1502, + "loss_ce": 1.1305272579193115, + "loss_region": 0.0, + "loss_total": 1.1305272579193115, + "lr": 0.000916312292358804, + "step": 3020, + "tokens_trained": 9.895016808 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 0.8595135096801645, + "grad_norm": 2.0573174953460693, + "loss": 1.1502, + "loss_ce": 1.0190644264221191, + "loss_region": 0.0, + "loss_total": 1.0190644264221191, + "lr": 0.0009159800664451827, + "step": 3030, + "tokens_trained": 9.927782248 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 0.862350187929934, + "grad_norm": 1.931090235710144, + "loss": 1.1539, + "loss_ce": 1.1790469884872437, + "loss_region": 0.0, + "loss_total": 1.1790469884872437, + "lr": 0.0009156478405315615, + "step": 3040, + "tokens_trained": 9.960547688 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 0.8651868661797035, + "grad_norm": 1.7028014659881592, + "loss": 1.1539, + "loss_ce": 1.099159598350525, + "loss_region": 0.0, + "loss_total": 1.099159598350525, + "lr": 0.0009153156146179402, + "step": 3050, + "tokens_trained": 9.993313128 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 0.868023544429473, + "grad_norm": 0.8875393271446228, + "loss": 1.1526, + "loss_ce": 1.0859668254852295, + "loss_region": 0.0, + "loss_total": 1.0859668254852295, + "lr": 0.0009149833887043189, + "step": 3060, + "tokens_trained": 10.026078568 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 0.8708602226792426, + "grad_norm": 0.3323346972465515, + "loss": 1.1511, + "loss_ce": 1.1056842803955078, + "loss_region": 0.0, + "loss_total": 1.1056842803955078, + "lr": 0.0009146511627906977, + "step": 3070, + "tokens_trained": 10.058844008 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 0.8736969009290121, + "grad_norm": 2.532973289489746, + "loss": 1.1592, + "loss_ce": 1.144846796989441, + "loss_region": 0.0, + "loss_total": 1.144846796989441, + "lr": 0.0009143189368770765, + "step": 3080, + "tokens_trained": 10.091604144 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 0.8765335791787816, + "grad_norm": 2.339357376098633, + "loss": 1.1563, + "loss_ce": 1.1964597702026367, + "loss_region": 0.0, + "loss_total": 1.1964597702026367, + "lr": 0.0009139867109634552, + "step": 3090, + "tokens_trained": 10.124368784 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 0.8793702574285511, + "grad_norm": 2.280010461807251, + "loss": 1.1676, + "loss_ce": 1.139888048171997, + "loss_region": 0.0, + "loss_total": 1.139888048171997, + "lr": 0.0009136544850498339, + "step": 3100, + "tokens_trained": 10.157134224 + }, + { + "epoch": 0.8793702574285511, + "eval_ppl": 3.168095600810329, + "eval_runtime": 1.4396, + "step": 3100, + "tokens_trained": 10.157134224 + }, + { + "epoch": 0.8793702574285511, + "eval_ppl_cds": 3.82561026529932, + "eval_ppl_dig": 1.2539614454251742, + "eval_ppl_exon": 3.534617032442498, + "eval_ppl_intron": 3.204971999008107, + "eval_ppl_nig": 3.055955645295421, + "eval_ppl_promoter": 3.471874618455133, + "eval_ppl_utr": 3.452678269617012, + "step": 3100, + "tokens_trained": 10.157134224 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 0.8822069356783206, + "grad_norm": 1.3441884517669678, + "loss": 1.1494, + "loss_ce": 1.1858655214309692, + "loss_region": 0.0, + "loss_total": 1.1858655214309692, + "lr": 0.0009133222591362126, + "step": 3110, + "tokens_trained": 10.189899664 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 0.8850436139280902, + "grad_norm": 2.8617494106292725, + "loss": 1.1747, + "loss_ce": 1.191227674484253, + "loss_region": 0.0, + "loss_total": 1.191227674484253, + "lr": 0.0009129900332225914, + "step": 3120, + "tokens_trained": 10.222665104 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 0.8878802921778597, + "grad_norm": 2.730170726776123, + "loss": 1.1707, + "loss_ce": 1.120618224143982, + "loss_region": 0.0, + "loss_total": 1.120618224143982, + "lr": 0.0009126578073089701, + "step": 3130, + "tokens_trained": 10.255429744 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 0.8907169704276292, + "grad_norm": 1.5439730882644653, + "loss": 1.1652, + "loss_ce": 1.0898618698120117, + "loss_region": 0.0, + "loss_total": 1.0898618698120117, + "lr": 0.0009123255813953488, + "step": 3140, + "tokens_trained": 10.288194384 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 0.8935536486773987, + "grad_norm": 3.1067678928375244, + "loss": 1.1619, + "loss_ce": 1.188930630683899, + "loss_region": 0.0, + "loss_total": 1.188930630683899, + "lr": 0.0009119933554817275, + "step": 3150, + "tokens_trained": 10.320959664 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 0.8963903269271682, + "grad_norm": 1.5188531875610352, + "loss": 1.163, + "loss_ce": 1.1788606643676758, + "loss_region": 0.0, + "loss_total": 1.1788606643676758, + "lr": 0.0009116611295681063, + "step": 3160, + "tokens_trained": 10.353725104 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 0.8992270051769378, + "grad_norm": 1.484955906867981, + "loss": 1.1588, + "loss_ce": 1.1448098421096802, + "loss_region": 0.0, + "loss_total": 1.1448098421096802, + "lr": 0.0009113289036544851, + "step": 3170, + "tokens_trained": 10.386490544 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 0.9020636834267073, + "grad_norm": 3.5358922481536865, + "loss": 1.1626, + "loss_ce": 1.14382803440094, + "loss_region": 0.0, + "loss_total": 1.14382803440094, + "lr": 0.0009109966777408638, + "step": 3180, + "tokens_trained": 10.419255984 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 0.9049003616764768, + "grad_norm": 2.2361903190612793, + "loss": 1.1594, + "loss_ce": 1.1448397636413574, + "loss_region": 0.0, + "loss_total": 1.1448397636413574, + "lr": 0.0009106644518272426, + "step": 3190, + "tokens_trained": 10.452021424 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 0.9077370399262463, + "grad_norm": 4.083084583282471, + "loss": 1.1516, + "loss_ce": 1.215737223625183, + "loss_region": 0.0, + "loss_total": 1.215737223625183, + "lr": 0.0009103322259136213, + "step": 3200, + "tokens_trained": 10.484786064 + }, + { + "epoch": 0.9077370399262463, + "eval_ppl": 3.2049418657825886, + "eval_runtime": 1.4225, + "step": 3200, + "tokens_trained": 10.484786064 + }, + { + "epoch": 0.9077370399262463, + "eval_ppl_cds": 3.8675519075576, + "eval_ppl_dig": 1.2413221863526536, + "eval_ppl_exon": 3.613581470447166, + "eval_ppl_intron": 3.250271081014392, + "eval_ppl_nig": 3.0744965338928103, + "eval_ppl_promoter": 3.5155590235022354, + "eval_ppl_utr": 3.5140742683870148, + "step": 3200, + "tokens_trained": 10.484786064 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 0.9105737181760158, + "grad_norm": 1.6410759687423706, + "loss": 1.1502, + "loss_ce": 1.116316795349121, + "loss_region": 0.0, + "loss_total": 1.116316795349121, + "lr": 0.00091, + "step": 3210, + "tokens_trained": 10.517551504 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 0.9134103964257854, + "grad_norm": 3.838754415512085, + "loss": 1.1606, + "loss_ce": 1.1389206647872925, + "loss_region": 0.0, + "loss_total": 1.1389206647872925, + "lr": 0.0009096677740863788, + "step": 3220, + "tokens_trained": 10.550316944 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 0.9162470746755549, + "grad_norm": 2.1586124897003174, + "loss": 1.1548, + "loss_ce": 1.1771422624588013, + "loss_region": 0.0, + "loss_total": 1.1771422624588013, + "lr": 0.0009093355481727576, + "step": 3230, + "tokens_trained": 10.583082384 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 0.9190837529253244, + "grad_norm": 2.8425631523132324, + "loss": 1.1572, + "loss_ce": 1.2074710130691528, + "loss_region": 0.0, + "loss_total": 1.2074710130691528, + "lr": 0.0009090033222591362, + "step": 3240, + "tokens_trained": 10.615847824 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 0.9219204311750939, + "grad_norm": 3.649073839187622, + "loss": 1.1562, + "loss_ce": 1.193737506866455, + "loss_region": 0.0, + "loss_total": 1.193737506866455, + "lr": 0.0009086710963455149, + "step": 3250, + "tokens_trained": 10.648613248 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 0.9247571094248634, + "grad_norm": 2.413256883621216, + "loss": 1.1391, + "loss_ce": 0.9940924048423767, + "loss_region": 0.0, + "loss_total": 0.9940924048423767, + "lr": 0.0009083388704318936, + "step": 3260, + "tokens_trained": 10.681378688 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 0.927593787674633, + "grad_norm": 1.2481706142425537, + "loss": 1.1446, + "loss_ce": 1.0777127742767334, + "loss_region": 0.0, + "loss_total": 1.0777127742767334, + "lr": 0.0009080066445182725, + "step": 3270, + "tokens_trained": 10.714144128 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 0.9304304659244025, + "grad_norm": 3.8436150550842285, + "loss": 1.1462, + "loss_ce": 1.1017178297042847, + "loss_region": 0.0, + "loss_total": 1.1017178297042847, + "lr": 0.0009076744186046512, + "step": 3280, + "tokens_trained": 10.746909568 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 0.933267144174172, + "grad_norm": 1.962862253189087, + "loss": 1.1451, + "loss_ce": 1.122126579284668, + "loss_region": 0.0, + "loss_total": 1.122126579284668, + "lr": 0.0009073421926910299, + "step": 3290, + "tokens_trained": 10.779675008 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 0.9361038224239415, + "grad_norm": 2.8649613857269287, + "loss": 1.1484, + "loss_ce": 1.2155630588531494, + "loss_region": 0.0, + "loss_total": 1.2155630588531494, + "lr": 0.0009070099667774086, + "step": 3300, + "tokens_trained": 10.812440448 + }, + { + "epoch": 0.9361038224239415, + "eval_ppl": 3.1479659994317064, + "eval_runtime": 1.421, + "step": 3300, + "tokens_trained": 10.812440448 + }, + { + "epoch": 0.9361038224239415, + "eval_ppl_cds": 3.8508217546890036, + "eval_ppl_dig": 1.1859427582817685, + "eval_ppl_exon": 3.5443184655351816, + "eval_ppl_intron": 3.1881740263512133, + "eval_ppl_nig": 3.0129829272050754, + "eval_ppl_promoter": 3.472097730166916, + "eval_ppl_utr": 3.464450876809351, + "step": 3300, + "tokens_trained": 10.812440448 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 0.938940500673711, + "grad_norm": 2.5750465393066406, + "loss": 1.1462, + "loss_ce": 1.1143780946731567, + "loss_region": 0.0, + "loss_total": 1.1143780946731567, + "lr": 0.0009066777408637874, + "step": 3310, + "tokens_trained": 10.845202832 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 0.9417771789234806, + "grad_norm": 1.4010454416275024, + "loss": 1.1392, + "loss_ce": 1.2106729745864868, + "loss_region": 0.0, + "loss_total": 1.2106729745864868, + "lr": 0.0009063455149501662, + "step": 3320, + "tokens_trained": 10.877968272 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 0.9446138571732501, + "grad_norm": 2.7220399379730225, + "loss": 1.1497, + "loss_ce": 1.1590839624404907, + "loss_region": 0.0, + "loss_total": 1.1590839624404907, + "lr": 0.0009060132890365449, + "step": 3330, + "tokens_trained": 10.910733712 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 0.9474505354230196, + "grad_norm": 1.9293181896209717, + "loss": 1.1419, + "loss_ce": 1.077150821685791, + "loss_region": 0.0, + "loss_total": 1.077150821685791, + "lr": 0.0009056810631229236, + "step": 3340, + "tokens_trained": 10.943499152 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 0.9502872136727891, + "grad_norm": 0.7119876742362976, + "loss": 1.1673, + "loss_ce": 1.194408655166626, + "loss_region": 0.0, + "loss_total": 1.194408655166626, + "lr": 0.0009053488372093024, + "step": 3350, + "tokens_trained": 10.976264592 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 0.9531238919225586, + "grad_norm": 0.7917010188102722, + "loss": 1.1413, + "loss_ce": 1.132968544960022, + "loss_region": 0.0, + "loss_total": 1.132968544960022, + "lr": 0.0009050166112956811, + "step": 3360, + "tokens_trained": 11.009030032 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 0.9559605701723282, + "grad_norm": 1.6125084161758423, + "loss": 1.1417, + "loss_ce": 1.11857008934021, + "loss_region": 0.0, + "loss_total": 1.11857008934021, + "lr": 0.0009046843853820598, + "step": 3370, + "tokens_trained": 11.041794728 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 0.9587972484220977, + "grad_norm": 3.088146924972534, + "loss": 1.1549, + "loss_ce": 1.14041268825531, + "loss_region": 0.0, + "loss_total": 1.14041268825531, + "lr": 0.0009043521594684385, + "step": 3380, + "tokens_trained": 11.074559368 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 0.9616339266718672, + "grad_norm": 1.6197377443313599, + "loss": 1.1321, + "loss_ce": 1.132678747177124, + "loss_region": 0.0, + "loss_total": 1.132678747177124, + "lr": 0.0009040199335548173, + "step": 3390, + "tokens_trained": 11.107324808 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 0.9644706049216367, + "grad_norm": 2.6149978637695312, + "loss": 1.1533, + "loss_ce": 1.11184561252594, + "loss_region": 0.0, + "loss_total": 1.11184561252594, + "lr": 0.000903687707641196, + "step": 3400, + "tokens_trained": 11.140090248 + }, + { + "epoch": 0.9644706049216367, + "eval_ppl": 3.150496689547493, + "eval_runtime": 1.4191, + "step": 3400, + "tokens_trained": 11.140090248 + }, + { + "epoch": 0.9644706049216367, + "eval_ppl_cds": 3.8476080631609846, + "eval_ppl_dig": 1.1785163771599736, + "eval_ppl_exon": 3.555188263952797, + "eval_ppl_intron": 3.196135927934826, + "eval_ppl_nig": 3.006339682662659, + "eval_ppl_promoter": 3.4762817843939695, + "eval_ppl_utr": 3.4796094569312483, + "step": 3400, + "tokens_trained": 11.140090248 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 0.9673072831714062, + "grad_norm": 2.613956928253174, + "loss": 1.1413, + "loss_ce": 1.0743883848190308, + "loss_region": 0.0, + "loss_total": 1.0743883848190308, + "lr": 0.0009033554817275747, + "step": 3410, + "tokens_trained": 11.172855688 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 0.9701439614211758, + "grad_norm": 2.711869239807129, + "loss": 1.1313, + "loss_ce": 1.0758590698242188, + "loss_region": 0.0, + "loss_total": 1.0758590698242188, + "lr": 0.0009030232558139535, + "step": 3420, + "tokens_trained": 11.205621096 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 0.9729806396709453, + "grad_norm": 3.1511757373809814, + "loss": 1.1388, + "loss_ce": 1.0435189008712769, + "loss_region": 0.0, + "loss_total": 1.0435189008712769, + "lr": 0.0009026910299003323, + "step": 3430, + "tokens_trained": 11.238384136 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 0.9758173179207148, + "grad_norm": 3.15256667137146, + "loss": 1.141, + "loss_ce": 1.1389594078063965, + "loss_region": 0.0, + "loss_total": 1.1389594078063965, + "lr": 0.000902358803986711, + "step": 3440, + "tokens_trained": 11.271149576 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 0.9786539961704843, + "grad_norm": 1.0374544858932495, + "loss": 1.1307, + "loss_ce": 1.08757746219635, + "loss_region": 0.0, + "loss_total": 1.08757746219635, + "lr": 0.0009020265780730897, + "step": 3450, + "tokens_trained": 11.303915016 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 0.9814906744202538, + "grad_norm": 1.7981122732162476, + "loss": 1.1317, + "loss_ce": 1.0712560415267944, + "loss_region": 0.0, + "loss_total": 1.0712560415267944, + "lr": 0.0009016943521594684, + "step": 3460, + "tokens_trained": 11.336680456 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 0.9843273526700234, + "grad_norm": 1.8928661346435547, + "loss": 1.1353, + "loss_ce": 1.0297223329544067, + "loss_region": 0.0, + "loss_total": 1.0297223329544067, + "lr": 0.0009013621262458473, + "step": 3470, + "tokens_trained": 11.369444288 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 0.9871640309197929, + "grad_norm": 1.883483648300171, + "loss": 1.1209, + "loss_ce": 1.1508862972259521, + "loss_region": 0.0, + "loss_total": 1.1508862972259521, + "lr": 0.000901029900332226, + "step": 3480, + "tokens_trained": 11.402209728 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 0.9900007091695624, + "grad_norm": 3.058206558227539, + "loss": 1.1296, + "loss_ce": 1.1671435832977295, + "loss_region": 0.0, + "loss_total": 1.1671435832977295, + "lr": 0.0009006976744186046, + "step": 3490, + "tokens_trained": 11.434975168 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 0.9928373874193319, + "grad_norm": 1.8243311643600464, + "loss": 1.1186, + "loss_ce": 1.1452795267105103, + "loss_region": 0.0, + "loss_total": 1.1452795267105103, + "lr": 0.0009003654485049833, + "step": 3500, + "tokens_trained": 11.467740608 + }, + { + "epoch": 0.9928373874193319, + "eval_ppl": 3.066169061885205, + "eval_runtime": 1.4174, + "step": 3500, + "tokens_trained": 11.467740608 + }, + { + "epoch": 0.9928373874193319, + "eval_ppl_cds": 3.7891787836731368, + "eval_ppl_dig": 1.1645212783158296, + "eval_ppl_exon": 3.464243784965796, + "eval_ppl_intron": 3.1094084819227223, + "eval_ppl_nig": 2.917633199606573, + "eval_ppl_promoter": 3.399518234966468, + "eval_ppl_utr": 3.3901046874442486, + "step": 3500, + "tokens_trained": 11.467740608 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 0.9956740656691014, + "grad_norm": 6.722069263458252, + "loss": 1.1357, + "loss_ce": 1.1965770721435547, + "loss_region": 0.0, + "loss_total": 1.1965770721435547, + "lr": 0.0009000332225913621, + "step": 3510, + "tokens_trained": 11.500506048 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 0.998510743918871, + "grad_norm": 2.387634754180908, + "loss": 1.1435, + "loss_ce": 1.0625895261764526, + "loss_region": 0.0, + "loss_total": 1.0625895261764526, + "lr": 0.0008997009966777409, + "step": 3520, + "tokens_trained": 11.533271488 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 1.0011346712999079, + "grad_norm": 2.712785005569458, + "loss": 1.1235, + "loss_ce": 1.1449328660964966, + "loss_region": 0.0, + "loss_total": 1.1449328660964966, + "lr": 0.0008993687707641196, + "step": 3530, + "tokens_trained": 11.56357952 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 1.0039713495496774, + "grad_norm": 1.2178337574005127, + "loss": 1.1181, + "loss_ce": 1.0866310596466064, + "loss_region": 0.0, + "loss_total": 1.0866310596466064, + "lr": 0.0008990365448504984, + "step": 3540, + "tokens_trained": 11.59634496 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 1.006808027799447, + "grad_norm": 0.5407071709632874, + "loss": 1.1258, + "loss_ce": 1.0429565906524658, + "loss_region": 0.0, + "loss_total": 1.0429565906524658, + "lr": 0.0008987043189368771, + "step": 3550, + "tokens_trained": 11.6291104 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 1.0096447060492164, + "grad_norm": 2.4842042922973633, + "loss": 1.1212, + "loss_ce": 1.1752214431762695, + "loss_region": 0.0, + "loss_total": 1.1752214431762695, + "lr": 0.0008983720930232558, + "step": 3560, + "tokens_trained": 11.66187584 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 1.012481384298986, + "grad_norm": 2.785231828689575, + "loss": 1.1279, + "loss_ce": 1.0473734140396118, + "loss_region": 0.0, + "loss_total": 1.0473734140396118, + "lr": 0.0008980398671096346, + "step": 3570, + "tokens_trained": 11.69464128 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 1.0153180625487555, + "grad_norm": 4.229964733123779, + "loss": 1.1266, + "loss_ce": 1.1528112888336182, + "loss_region": 0.0, + "loss_total": 1.1528112888336182, + "lr": 0.0008977076411960134, + "step": 3580, + "tokens_trained": 11.72740592 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 1.018154740798525, + "grad_norm": 2.4812772274017334, + "loss": 1.1236, + "loss_ce": 1.1817506551742554, + "loss_region": 0.0, + "loss_total": 1.1817506551742554, + "lr": 0.0008973754152823921, + "step": 3590, + "tokens_trained": 11.76017136 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 1.0209914190482945, + "grad_norm": 2.572169065475464, + "loss": 1.1246, + "loss_ce": 1.0999400615692139, + "loss_region": 0.0, + "loss_total": 1.0999400615692139, + "lr": 0.0008970431893687708, + "step": 3600, + "tokens_trained": 11.7929352 + }, + { + "epoch": 1.0209914190482945, + "eval_ppl": 3.072980694149614, + "eval_runtime": 1.427, + "step": 3600, + "tokens_trained": 11.7929352 + }, + { + "epoch": 1.0209914190482945, + "eval_ppl_cds": 3.816520794498398, + "eval_ppl_dig": 1.1614647928900517, + "eval_ppl_exon": 3.485870116460774, + "eval_ppl_intron": 3.122053185929521, + "eval_ppl_nig": 2.919871378307291, + "eval_ppl_promoter": 3.412989289892484, + "eval_ppl_utr": 3.4059336828102116, + "step": 3600, + "tokens_trained": 11.7929352 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 1.023828097298064, + "grad_norm": 3.310690402984619, + "loss": 1.1218, + "loss_ce": 1.146215558052063, + "loss_region": 0.0, + "loss_total": 1.146215558052063, + "lr": 0.0008967109634551495, + "step": 3610, + "tokens_trained": 11.82570064 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 1.0266647755478335, + "grad_norm": 1.7069272994995117, + "loss": 1.1196, + "loss_ce": 1.0282905101776123, + "loss_region": 0.0, + "loss_total": 1.0282905101776123, + "lr": 0.0008963787375415283, + "step": 3620, + "tokens_trained": 11.85846448 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 1.029501453797603, + "grad_norm": 1.9944303035736084, + "loss": 1.1146, + "loss_ce": 1.1349163055419922, + "loss_region": 0.0, + "loss_total": 1.1349163055419922, + "lr": 0.000896046511627907, + "step": 3630, + "tokens_trained": 11.89122992 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 1.0323381320473726, + "grad_norm": 3.0683534145355225, + "loss": 1.1163, + "loss_ce": 1.115451693534851, + "loss_region": 0.0, + "loss_total": 1.115451693534851, + "lr": 0.0008957142857142857, + "step": 3640, + "tokens_trained": 11.92399536 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 1.035174810297142, + "grad_norm": 1.1857675313949585, + "loss": 1.1065, + "loss_ce": 1.1648534536361694, + "loss_region": 0.0, + "loss_total": 1.1648534536361694, + "lr": 0.0008953820598006644, + "step": 3650, + "tokens_trained": 11.9567608 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 1.0380114885469116, + "grad_norm": 2.9640052318573, + "loss": 1.1171, + "loss_ce": 1.1093553304672241, + "loss_region": 0.0, + "loss_total": 1.1093553304672241, + "lr": 0.0008950498338870432, + "step": 3660, + "tokens_trained": 11.98952624 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 1.0408481667966811, + "grad_norm": 1.0555254220962524, + "loss": 1.1114, + "loss_ce": 1.165023684501648, + "loss_region": 0.0, + "loss_total": 1.165023684501648, + "lr": 0.000894717607973422, + "step": 3670, + "tokens_trained": 12.02229168 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 1.0436848450464506, + "grad_norm": 0.7301769256591797, + "loss": 1.1169, + "loss_ce": 1.1221823692321777, + "loss_region": 0.0, + "loss_total": 1.1221823692321777, + "lr": 0.0008943853820598007, + "step": 3680, + "tokens_trained": 12.05505712 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 1.0465215232962202, + "grad_norm": 1.7822520732879639, + "loss": 1.1045, + "loss_ce": 1.1591402292251587, + "loss_region": 0.0, + "loss_total": 1.1591402292251587, + "lr": 0.0008940531561461794, + "step": 3690, + "tokens_trained": 12.08782256 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 1.0493582015459897, + "grad_norm": 2.0740301609039307, + "loss": 1.1105, + "loss_ce": 1.102281928062439, + "loss_region": 0.0, + "loss_total": 1.102281928062439, + "lr": 0.0008937209302325582, + "step": 3700, + "tokens_trained": 12.1205864 + }, + { + "epoch": 1.0493582015459897, + "eval_ppl": 3.0495794779291012, + "eval_runtime": 1.4282, + "step": 3700, + "tokens_trained": 12.1205864 + }, + { + "epoch": 1.0493582015459897, + "eval_ppl_cds": 3.7928346943430498, + "eval_ppl_dig": 1.1304920781740384, + "eval_ppl_exon": 3.4647563324137383, + "eval_ppl_intron": 3.1043868011292792, + "eval_ppl_nig": 2.8885740912274267, + "eval_ppl_promoter": 3.3882082198178005, + "eval_ppl_utr": 3.3858617175113457, + "step": 3700, + "tokens_trained": 12.1205864 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 1.0521948797957592, + "grad_norm": 1.7531991004943848, + "loss": 1.108, + "loss_ce": 1.1638976335525513, + "loss_region": 0.0, + "loss_total": 1.1638976335525513, + "lr": 0.0008933887043189369, + "step": 3710, + "tokens_trained": 12.15335184 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 1.0550315580455287, + "grad_norm": 2.461660146713257, + "loss": 1.116, + "loss_ce": 1.1260329484939575, + "loss_region": 0.0, + "loss_total": 1.1260329484939575, + "lr": 0.0008930564784053157, + "step": 3720, + "tokens_trained": 12.18611712 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 1.0578682362952982, + "grad_norm": 2.5355563163757324, + "loss": 1.109, + "loss_ce": 1.1065844297409058, + "loss_region": 0.0, + "loss_total": 1.1065844297409058, + "lr": 0.0008927242524916944, + "step": 3730, + "tokens_trained": 12.21888256 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 1.0607049145450678, + "grad_norm": 0.7454701066017151, + "loss": 1.1036, + "loss_ce": 1.1427736282348633, + "loss_region": 0.0, + "loss_total": 1.1427736282348633, + "lr": 0.0008923920265780731, + "step": 3740, + "tokens_trained": 12.251648 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 1.0635415927948373, + "grad_norm": 1.3640272617340088, + "loss": 1.11, + "loss_ce": 1.1260356903076172, + "loss_region": 0.0, + "loss_total": 1.1260356903076172, + "lr": 0.0008920598006644518, + "step": 3750, + "tokens_trained": 12.284409608 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 1.0663782710446068, + "grad_norm": 2.158928871154785, + "loss": 1.109, + "loss_ce": 1.1236199140548706, + "loss_region": 0.0, + "loss_total": 1.1236199140548706, + "lr": 0.0008917275747508305, + "step": 3760, + "tokens_trained": 12.317175048 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 1.0692149492943763, + "grad_norm": 2.835397481918335, + "loss": 1.0968, + "loss_ce": 1.048624038696289, + "loss_region": 0.0, + "loss_total": 1.048624038696289, + "lr": 0.0008913953488372093, + "step": 3770, + "tokens_trained": 12.349940488 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 1.0720516275441458, + "grad_norm": 2.2109668254852295, + "loss": 1.1087, + "loss_ce": 1.134214162826538, + "loss_region": 0.0, + "loss_total": 1.134214162826538, + "lr": 0.0008910631229235881, + "step": 3780, + "tokens_trained": 12.382705928 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 1.0748883057939154, + "grad_norm": 2.0000789165496826, + "loss": 1.1, + "loss_ce": 1.1183463335037231, + "loss_region": 0.0, + "loss_total": 1.1183463335037231, + "lr": 0.0008907308970099668, + "step": 3790, + "tokens_trained": 12.415470568 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 1.0777249840436849, + "grad_norm": 1.7075557708740234, + "loss": 1.1041, + "loss_ce": 1.1731345653533936, + "loss_region": 0.0, + "loss_total": 1.1731345653533936, + "lr": 0.0008903986710963455, + "step": 3800, + "tokens_trained": 12.448236008 + }, + { + "epoch": 1.0777249840436849, + "eval_ppl": 3.012991043787593, + "eval_runtime": 1.4275, + "step": 3800, + "tokens_trained": 12.448236008 + }, + { + "epoch": 1.0777249840436849, + "eval_ppl_cds": 3.7647737075387178, + "eval_ppl_dig": 1.1461106755590589, + "eval_ppl_exon": 3.4324894516044706, + "eval_ppl_intron": 3.069651274017756, + "eval_ppl_nig": 2.847952290398712, + "eval_ppl_promoter": 3.360163912951573, + "eval_ppl_utr": 3.3589429891320175, + "step": 3800, + "tokens_trained": 12.448236008 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 1.0805616622934544, + "grad_norm": 1.5322575569152832, + "loss": 1.102, + "loss_ce": 1.1474723815917969, + "loss_region": 0.0, + "loss_total": 1.1474723815917969, + "lr": 0.0008900664451827242, + "step": 3810, + "tokens_trained": 12.481001448 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 1.083398340543224, + "grad_norm": 1.9201018810272217, + "loss": 1.0932, + "loss_ce": 1.0961087942123413, + "loss_region": 0.0, + "loss_total": 1.0961087942123413, + "lr": 0.0008897342192691031, + "step": 3820, + "tokens_trained": 12.513766888 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 1.0862350187929934, + "grad_norm": 0.6309838891029358, + "loss": 1.1003, + "loss_ce": 1.17067551612854, + "loss_region": 0.0, + "loss_total": 1.17067551612854, + "lr": 0.0008894019933554818, + "step": 3830, + "tokens_trained": 12.546532328 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 1.089071697042763, + "grad_norm": 2.2249398231506348, + "loss": 1.1056, + "loss_ce": 1.0473724603652954, + "loss_region": 0.0, + "loss_total": 1.0473724603652954, + "lr": 0.0008890697674418605, + "step": 3840, + "tokens_trained": 12.57929404 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 1.0919083752925325, + "grad_norm": 1.9094284772872925, + "loss": 1.1021, + "loss_ce": 1.0956895351409912, + "loss_region": 0.0, + "loss_total": 1.0956895351409912, + "lr": 0.0008887375415282393, + "step": 3850, + "tokens_trained": 12.61205788 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 1.094745053542302, + "grad_norm": 1.414716362953186, + "loss": 1.0962, + "loss_ce": 1.1112486124038696, + "loss_region": 0.0, + "loss_total": 1.1112486124038696, + "lr": 0.0008884053156146179, + "step": 3860, + "tokens_trained": 12.64482332 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 1.0975817317920715, + "grad_norm": 1.6879093647003174, + "loss": 1.0897, + "loss_ce": 1.1716594696044922, + "loss_region": 0.0, + "loss_total": 1.1716594696044922, + "lr": 0.0008880730897009966, + "step": 3870, + "tokens_trained": 12.677587992 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 1.100418410041841, + "grad_norm": 2.3944084644317627, + "loss": 1.0957, + "loss_ce": 1.0909066200256348, + "loss_region": 0.0, + "loss_total": 1.0909066200256348, + "lr": 0.0008877408637873754, + "step": 3880, + "tokens_trained": 12.710352632 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 1.1032550882916106, + "grad_norm": 1.3609209060668945, + "loss": 1.0982, + "loss_ce": 1.0587270259857178, + "loss_region": 0.0, + "loss_total": 1.0587270259857178, + "lr": 0.0008874086378737542, + "step": 3890, + "tokens_trained": 12.743118072 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 1.10609176654138, + "grad_norm": 0.9393012523651123, + "loss": 1.093, + "loss_ce": 1.0574629306793213, + "loss_region": 0.0, + "loss_total": 1.0574629306793213, + "lr": 0.0008870764119601329, + "step": 3900, + "tokens_trained": 12.775882712 + }, + { + "epoch": 1.10609176654138, + "eval_ppl": 3.010736977156372, + "eval_runtime": 1.438, + "step": 3900, + "tokens_trained": 12.775882712 + }, + { + "epoch": 1.10609176654138, + "eval_ppl_cds": 3.781072911522729, + "eval_ppl_dig": 1.13292700966948, + "eval_ppl_exon": 3.4265111453165225, + "eval_ppl_intron": 3.0702147632944676, + "eval_ppl_nig": 2.839200394835176, + "eval_ppl_promoter": 3.36480075671075, + "eval_ppl_utr": 3.3449183430575085, + "step": 3900, + "tokens_trained": 12.775882712 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 1.1089284447911496, + "grad_norm": 2.0128800868988037, + "loss": 1.1188, + "loss_ce": 1.1648943424224854, + "loss_region": 0.0, + "loss_total": 1.1648943424224854, + "lr": 0.0008867441860465116, + "step": 3910, + "tokens_trained": 12.808648152 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 1.1117651230409191, + "grad_norm": 1.899614930152893, + "loss": 1.1065, + "loss_ce": 1.1452734470367432, + "loss_region": 0.0, + "loss_total": 1.1452734470367432, + "lr": 0.0008864119601328903, + "step": 3920, + "tokens_trained": 12.841413592 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 1.1146018012906886, + "grad_norm": 1.4031317234039307, + "loss": 1.1002, + "loss_ce": 1.0183168649673462, + "loss_region": 0.0, + "loss_total": 1.0183168649673462, + "lr": 0.0008860797342192692, + "step": 3930, + "tokens_trained": 12.874179032 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 1.1174384795404582, + "grad_norm": 2.6226985454559326, + "loss": 1.0946, + "loss_ce": 1.0670664310455322, + "loss_region": 0.0, + "loss_total": 1.0670664310455322, + "lr": 0.0008857475083056479, + "step": 3940, + "tokens_trained": 12.906944472 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 1.1202751577902277, + "grad_norm": 1.4078290462493896, + "loss": 1.0887, + "loss_ce": 1.0374524593353271, + "loss_region": 0.0, + "loss_total": 1.0374524593353271, + "lr": 0.0008854152823920266, + "step": 3950, + "tokens_trained": 12.939709912 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 1.1231118360399972, + "grad_norm": 0.9895715713500977, + "loss": 1.0819, + "loss_ce": 1.1141631603240967, + "loss_region": 0.0, + "loss_total": 1.1141631603240967, + "lr": 0.0008850830564784053, + "step": 3960, + "tokens_trained": 12.972475352 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 1.1259485142897667, + "grad_norm": 1.7113982439041138, + "loss": 1.0952, + "loss_ce": 1.1644362211227417, + "loss_region": 0.0, + "loss_total": 1.1644362211227417, + "lr": 0.0008847508305647841, + "step": 3970, + "tokens_trained": 13.005240792 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 1.1287851925395362, + "grad_norm": 1.4504528045654297, + "loss": 1.0889, + "loss_ce": 1.1063963174819946, + "loss_region": 0.0, + "loss_total": 1.1063963174819946, + "lr": 0.0008844186046511629, + "step": 3980, + "tokens_trained": 13.038005432 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 1.1316218707893058, + "grad_norm": 1.8971322774887085, + "loss": 1.0919, + "loss_ce": 1.1162850856781006, + "loss_region": 0.0, + "loss_total": 1.1162850856781006, + "lr": 0.0008840863787375415, + "step": 3990, + "tokens_trained": 13.070770872 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 1.1344585490390753, + "grad_norm": 3.883312702178955, + "loss": 1.0982, + "loss_ce": 1.0977280139923096, + "loss_region": 0.0, + "loss_total": 1.0977280139923096, + "lr": 0.0008837541528239202, + "step": 4000, + "tokens_trained": 13.103536312 + }, + { + "epoch": 1.1344585490390753, + "eval_ppl": 3.0343417985973162, + "eval_runtime": 1.4201, + "step": 4000, + "tokens_trained": 13.103536312 + }, + { + "epoch": 1.1344585490390753, + "eval_ppl_cds": 3.8138726635795637, + "eval_ppl_dig": 1.1319177062850718, + "eval_ppl_exon": 3.4849972429847624, + "eval_ppl_intron": 3.103300486755083, + "eval_ppl_nig": 2.849112021645167, + "eval_ppl_promoter": 3.398791128789424, + "eval_ppl_utr": 3.3841584643242992, + "step": 4000, + "tokens_trained": 13.103536312 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 1.1372952272888448, + "grad_norm": 1.364736795425415, + "loss": 1.0884, + "loss_ce": 1.1118701696395874, + "loss_region": 0.0, + "loss_total": 1.1118701696395874, + "lr": 0.000883421926910299, + "step": 4010, + "tokens_trained": 13.136300952 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 1.1401319055386143, + "grad_norm": 1.321040391921997, + "loss": 1.0956, + "loss_ce": 1.0324740409851074, + "loss_region": 0.0, + "loss_total": 1.0324740409851074, + "lr": 0.0008830897009966777, + "step": 4020, + "tokens_trained": 13.169065592 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 1.1429685837883838, + "grad_norm": 1.2876423597335815, + "loss": 1.0923, + "loss_ce": 1.0883049964904785, + "loss_region": 0.0, + "loss_total": 1.0883049964904785, + "lr": 0.0008827574750830565, + "step": 4030, + "tokens_trained": 13.201831032 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 1.1458052620381534, + "grad_norm": 1.3390523195266724, + "loss": 1.0866, + "loss_ce": 1.01687753200531, + "loss_region": 0.0, + "loss_total": 1.01687753200531, + "lr": 0.0008824252491694352, + "step": 4040, + "tokens_trained": 13.234596472 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 1.1486419402879229, + "grad_norm": 2.3616015911102295, + "loss": 1.092, + "loss_ce": 1.1318427324295044, + "loss_region": 0.0, + "loss_total": 1.1318427324295044, + "lr": 0.000882093023255814, + "step": 4050, + "tokens_trained": 13.267361888 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 1.1514786185376924, + "grad_norm": 2.2984602451324463, + "loss": 1.0951, + "loss_ce": 1.1276808977127075, + "loss_region": 0.0, + "loss_total": 1.1276808977127075, + "lr": 0.0008817607973421927, + "step": 4060, + "tokens_trained": 13.300127328 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 1.154315296787462, + "grad_norm": 1.9291963577270508, + "loss": 1.0949, + "loss_ce": 1.0479625463485718, + "loss_region": 0.0, + "loss_total": 1.0479625463485718, + "lr": 0.0008814285714285714, + "step": 4070, + "tokens_trained": 13.332892768 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 1.1571519750372314, + "grad_norm": 1.8348621129989624, + "loss": 1.0837, + "loss_ce": 1.1637519598007202, + "loss_region": 0.0, + "loss_total": 1.1637519598007202, + "lr": 0.0008810963455149503, + "step": 4080, + "tokens_trained": 13.365658208 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 1.159988653287001, + "grad_norm": 1.957354187965393, + "loss": 1.0868, + "loss_ce": 1.005294919013977, + "loss_region": 0.0, + "loss_total": 1.005294919013977, + "lr": 0.000880764119601329, + "step": 4090, + "tokens_trained": 13.398422568 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 1.1628253315367705, + "grad_norm": 1.4298845529556274, + "loss": 1.0871, + "loss_ce": 1.1631063222885132, + "loss_region": 0.0, + "loss_total": 1.1631063222885132, + "lr": 0.0008804318936877077, + "step": 4100, + "tokens_trained": 13.431185432 + }, + { + "epoch": 1.1628253315367705, + "eval_ppl": 2.9744982629037326, + "eval_runtime": 1.4292, + "step": 4100, + "tokens_trained": 13.431185432 + }, + { + "epoch": 1.1628253315367705, + "eval_ppl_cds": 3.697176902161677, + "eval_ppl_dig": 1.12236068719554, + "eval_ppl_exon": 3.406085874802647, + "eval_ppl_intron": 3.0410032961858406, + "eval_ppl_nig": 2.786823391770873, + "eval_ppl_promoter": 3.332150585502761, + "eval_ppl_utr": 3.3203950835055283, + "step": 4100, + "tokens_trained": 13.431185432 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 1.16566200978654, + "grad_norm": 1.258500337600708, + "loss": 1.0864, + "loss_ce": 1.1268681287765503, + "loss_region": 0.0, + "loss_total": 1.1268681287765503, + "lr": 0.0008800996677740863, + "step": 4110, + "tokens_trained": 13.463950872 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 1.1684986880363095, + "grad_norm": 1.2550866603851318, + "loss": 1.0828, + "loss_ce": 0.8582396507263184, + "loss_region": 0.0, + "loss_total": 0.8582396507263184, + "lr": 0.000879767441860465, + "step": 4120, + "tokens_trained": 13.496713976 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 1.171335366286079, + "grad_norm": 2.2830793857574463, + "loss": 1.0851, + "loss_ce": 1.1628100872039795, + "loss_region": 0.0, + "loss_total": 1.1628100872039795, + "lr": 0.0008794352159468439, + "step": 4130, + "tokens_trained": 13.529479416 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 1.1741720445358486, + "grad_norm": 2.4223439693450928, + "loss": 1.0938, + "loss_ce": 0.9137558341026306, + "loss_region": 0.0, + "loss_total": 0.9137558341026306, + "lr": 0.0008791029900332226, + "step": 4140, + "tokens_trained": 13.562244856 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 1.177008722785618, + "grad_norm": 1.8442988395690918, + "loss": 1.0864, + "loss_ce": 1.1363426446914673, + "loss_region": 0.0, + "loss_total": 1.1363426446914673, + "lr": 0.0008787707641196013, + "step": 4150, + "tokens_trained": 13.595010296 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 1.1798454010353876, + "grad_norm": 1.8979982137680054, + "loss": 1.0712, + "loss_ce": 1.0010813474655151, + "loss_region": 0.0, + "loss_total": 1.0010813474655151, + "lr": 0.00087843853820598, + "step": 4160, + "tokens_trained": 13.627775736 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 1.1826820792851571, + "grad_norm": 4.002573013305664, + "loss": 1.0864, + "loss_ce": 1.1242972612380981, + "loss_region": 0.0, + "loss_total": 1.1242972612380981, + "lr": 0.0008781063122923588, + "step": 4170, + "tokens_trained": 13.660541176 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 1.1855187575349266, + "grad_norm": 2.343280792236328, + "loss": 1.0846, + "loss_ce": 1.1182810068130493, + "loss_region": 0.0, + "loss_total": 1.1182810068130493, + "lr": 0.0008777740863787376, + "step": 4180, + "tokens_trained": 13.693306616 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 1.1883554357846962, + "grad_norm": 2.0315101146698, + "loss": 1.0807, + "loss_ce": 1.0237787961959839, + "loss_region": 0.0, + "loss_total": 1.0237787961959839, + "lr": 0.0008774418604651163, + "step": 4190, + "tokens_trained": 13.726072056 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 1.1911921140344657, + "grad_norm": 2.699315071105957, + "loss": 1.0908, + "loss_ce": 1.059959888458252, + "loss_region": 0.0, + "loss_total": 1.059959888458252, + "lr": 0.0008771096345514951, + "step": 4200, + "tokens_trained": 13.758837496 + }, + { + "epoch": 1.1911921140344657, + "eval_ppl": 2.965200992441565, + "eval_runtime": 1.4294, + "step": 4200, + "tokens_trained": 13.758837496 + }, + { + "epoch": 1.1911921140344657, + "eval_ppl_cds": 3.6857692518088268, + "eval_ppl_dig": 1.1262515443280499, + "eval_ppl_exon": 3.4015980099983603, + "eval_ppl_intron": 3.0309145180717567, + "eval_ppl_nig": 2.7697884525136023, + "eval_ppl_promoter": 3.327463809244272, + "eval_ppl_utr": 3.3167860404057192, + "step": 4200, + "tokens_trained": 13.758837496 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 1.1940287922842352, + "grad_norm": 0.9807295799255371, + "loss": 1.0836, + "loss_ce": 1.0797700881958008, + "loss_region": 0.0, + "loss_total": 1.0797700881958008, + "lr": 0.0008767774086378738, + "step": 4210, + "tokens_trained": 13.791602936 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 1.1968654705340047, + "grad_norm": 2.510774612426758, + "loss": 1.093, + "loss_ce": 1.0959842205047607, + "loss_region": 0.0, + "loss_total": 1.0959842205047607, + "lr": 0.0008764451827242525, + "step": 4220, + "tokens_trained": 13.824368376 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 1.1997021487837742, + "grad_norm": 1.4667102098464966, + "loss": 1.0805, + "loss_ce": 0.9582887291908264, + "loss_region": 0.0, + "loss_total": 0.9582887291908264, + "lr": 0.0008761129568106313, + "step": 4230, + "tokens_trained": 13.857133016 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 1.2025388270335438, + "grad_norm": 1.570622205734253, + "loss": 1.0751, + "loss_ce": 1.121809482574463, + "loss_region": 0.0, + "loss_total": 1.121809482574463, + "lr": 0.00087578073089701, + "step": 4240, + "tokens_trained": 13.889896856 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 1.2053755052833133, + "grad_norm": 2.9345927238464355, + "loss": 1.0813, + "loss_ce": 1.0893499851226807, + "loss_region": 0.0, + "loss_total": 1.0893499851226807, + "lr": 0.0008754485049833887, + "step": 4250, + "tokens_trained": 13.922662296 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 1.2082121835330828, + "grad_norm": 1.8342455625534058, + "loss": 1.0804, + "loss_ce": 1.0951244831085205, + "loss_region": 0.0, + "loss_total": 1.0951244831085205, + "lr": 0.0008751162790697674, + "step": 4260, + "tokens_trained": 13.955427736 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 1.2110488617828523, + "grad_norm": 1.9692314863204956, + "loss": 1.0768, + "loss_ce": 1.0899282693862915, + "loss_region": 0.0, + "loss_total": 1.0899282693862915, + "lr": 0.0008747840531561461, + "step": 4270, + "tokens_trained": 13.988193176 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 1.2138855400326218, + "grad_norm": 3.8746447563171387, + "loss": 1.0778, + "loss_ce": 1.1709401607513428, + "loss_region": 0.0, + "loss_total": 1.1709401607513428, + "lr": 0.000874451827242525, + "step": 4280, + "tokens_trained": 14.020958616 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 1.2167222182823914, + "grad_norm": 1.3762320280075073, + "loss": 1.0848, + "loss_ce": 1.0633325576782227, + "loss_region": 0.0, + "loss_total": 1.0633325576782227, + "lr": 0.0008741196013289037, + "step": 4290, + "tokens_trained": 14.053724056 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 1.2195588965321609, + "grad_norm": 1.2410640716552734, + "loss": 1.0826, + "loss_ce": 1.124458909034729, + "loss_region": 0.0, + "loss_total": 1.124458909034729, + "lr": 0.0008737873754152824, + "step": 4300, + "tokens_trained": 14.086488728 + }, + { + "epoch": 1.2195588965321609, + "eval_ppl": 2.951187993532025, + "eval_runtime": 1.418, + "step": 4300, + "tokens_trained": 14.086488728 + }, + { + "epoch": 1.2195588965321609, + "eval_ppl_cds": 3.616139386436684, + "eval_ppl_dig": 1.1229310394722243, + "eval_ppl_exon": 3.386197008726297, + "eval_ppl_intron": 3.021851954961441, + "eval_ppl_nig": 2.7497006798889134, + "eval_ppl_promoter": 3.3167881937888857, + "eval_ppl_utr": 3.307755212708026, + "step": 4300, + "tokens_trained": 14.086488728 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 1.2223955747819304, + "grad_norm": 3.1740565299987793, + "loss": 1.076, + "loss_ce": 1.0433340072631836, + "loss_region": 0.0, + "loss_total": 1.0433340072631836, + "lr": 0.0008734551495016611, + "step": 4310, + "tokens_trained": 14.119254168 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 1.2252322530317, + "grad_norm": 2.1822316646575928, + "loss": 1.0729, + "loss_ce": 1.0620695352554321, + "loss_region": 0.0, + "loss_total": 1.0620695352554321, + "lr": 0.0008731229235880399, + "step": 4320, + "tokens_trained": 14.152019608 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 1.2280689312814694, + "grad_norm": 1.640642523765564, + "loss": 1.0797, + "loss_ce": 1.0513261556625366, + "loss_region": 0.0, + "loss_total": 1.0513261556625366, + "lr": 0.0008727906976744187, + "step": 4330, + "tokens_trained": 14.184785048 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 1.230905609531239, + "grad_norm": 0.5384749174118042, + "loss": 1.0802, + "loss_ce": 1.0388718843460083, + "loss_region": 0.0, + "loss_total": 1.0388718843460083, + "lr": 0.0008724584717607974, + "step": 4340, + "tokens_trained": 14.217550488 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 1.2337422877810085, + "grad_norm": 2.1245718002319336, + "loss": 1.0705, + "loss_ce": 1.1147044897079468, + "loss_region": 0.0, + "loss_total": 1.1147044897079468, + "lr": 0.0008721262458471761, + "step": 4350, + "tokens_trained": 14.250315928 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 1.236578966030778, + "grad_norm": 2.869173049926758, + "loss": 1.0708, + "loss_ce": 1.1196682453155518, + "loss_region": 0.0, + "loss_total": 1.1196682453155518, + "lr": 0.0008717940199335548, + "step": 4360, + "tokens_trained": 14.283081368 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 1.2394156442805475, + "grad_norm": 1.38724946975708, + "loss": 1.0739, + "loss_ce": 1.1130313873291016, + "loss_region": 0.0, + "loss_total": 1.1130313873291016, + "lr": 0.0008714617940199335, + "step": 4370, + "tokens_trained": 14.315846808 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 1.242252322530317, + "grad_norm": 0.9248985052108765, + "loss": 1.0724, + "loss_ce": 1.1735085248947144, + "loss_region": 0.0, + "loss_total": 1.1735085248947144, + "lr": 0.0008711295681063123, + "step": 4380, + "tokens_trained": 14.348612232 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 1.2450890007800866, + "grad_norm": 2.6547763347625732, + "loss": 1.0832, + "loss_ce": 1.14912748336792, + "loss_region": 0.0, + "loss_total": 1.14912748336792, + "lr": 0.000870797342192691, + "step": 4390, + "tokens_trained": 14.381377672 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 1.247925679029856, + "grad_norm": 0.7941542267799377, + "loss": 1.0755, + "loss_ce": 1.0158019065856934, + "loss_region": 0.0, + "loss_total": 1.0158019065856934, + "lr": 0.0008704651162790698, + "step": 4400, + "tokens_trained": 14.414143112 + }, + { + "epoch": 1.247925679029856, + "eval_ppl": 2.9496977550809413, + "eval_runtime": 1.4276, + "step": 4400, + "tokens_trained": 14.414143112 + }, + { + "epoch": 1.247925679029856, + "eval_ppl_cds": 3.6556936362771273, + "eval_ppl_dig": 1.1193824356820266, + "eval_ppl_exon": 3.386813709659251, + "eval_ppl_intron": 3.02302864157853, + "eval_ppl_nig": 2.744096359670213, + "eval_ppl_promoter": 3.3203008818823005, + "eval_ppl_utr": 3.309144509919857, + "step": 4400, + "tokens_trained": 14.414143112 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 1.2507623572796256, + "grad_norm": 1.5432381629943848, + "loss": 1.0805, + "loss_ce": 1.0523535013198853, + "loss_region": 0.0, + "loss_total": 1.0523535013198853, + "lr": 0.0008701328903654485, + "step": 4410, + "tokens_trained": 14.446908552 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 1.2535990355293951, + "grad_norm": 1.2326829433441162, + "loss": 1.0757, + "loss_ce": 1.1546168327331543, + "loss_region": 0.0, + "loss_total": 1.1546168327331543, + "lr": 0.0008698006644518272, + "step": 4420, + "tokens_trained": 14.479673992 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 1.2564357137791646, + "grad_norm": 1.5566610097885132, + "loss": 1.0729, + "loss_ce": 1.0801700353622437, + "loss_region": 0.0, + "loss_total": 1.0801700353622437, + "lr": 0.000869468438538206, + "step": 4430, + "tokens_trained": 14.512439432 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 1.2592723920289342, + "grad_norm": 1.2869741916656494, + "loss": 1.0724, + "loss_ce": 1.1125671863555908, + "loss_region": 0.0, + "loss_total": 1.1125671863555908, + "lr": 0.0008691362126245848, + "step": 4440, + "tokens_trained": 14.545204872 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 1.2621090702787037, + "grad_norm": 2.487546920776367, + "loss": 1.0725, + "loss_ce": 0.951497495174408, + "loss_region": 0.0, + "loss_total": 0.951497495174408, + "lr": 0.0008688039867109635, + "step": 4450, + "tokens_trained": 14.577970312 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 1.2649457485284732, + "grad_norm": 1.4104713201522827, + "loss": 1.0727, + "loss_ce": 1.1030021905899048, + "loss_region": 0.0, + "loss_total": 1.1030021905899048, + "lr": 0.0008684717607973422, + "step": 4460, + "tokens_trained": 14.610735752 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 1.2677824267782427, + "grad_norm": 1.306612491607666, + "loss": 1.0784, + "loss_ce": 1.1074961423873901, + "loss_region": 0.0, + "loss_total": 1.1074961423873901, + "lr": 0.000868139534883721, + "step": 4470, + "tokens_trained": 14.643501192 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 1.2706191050280122, + "grad_norm": 1.6763625144958496, + "loss": 1.0793, + "loss_ce": 1.0440517663955688, + "loss_region": 0.0, + "loss_total": 1.0440517663955688, + "lr": 0.0008678073089700998, + "step": 4480, + "tokens_trained": 14.676266632 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 1.2734557832777817, + "grad_norm": 1.1120378971099854, + "loss": 1.0776, + "loss_ce": 0.9714234471321106, + "loss_region": 0.0, + "loss_total": 0.9714234471321106, + "lr": 0.0008674750830564784, + "step": 4490, + "tokens_trained": 14.709032072 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 1.2762924615275513, + "grad_norm": 1.8202154636383057, + "loss": 1.0634, + "loss_ce": 1.0794028043746948, + "loss_region": 0.0, + "loss_total": 1.0794028043746948, + "lr": 0.0008671428571428571, + "step": 4500, + "tokens_trained": 14.741797512 + }, + { + "epoch": 1.2762924615275513, + "eval_ppl": 2.956641844906708, + "eval_runtime": 1.4202, + "step": 4500, + "tokens_trained": 14.741797512 + }, + { + "epoch": 1.2762924615275513, + "eval_ppl_cds": 3.652418672322538, + "eval_ppl_dig": 1.1157348816652088, + "eval_ppl_exon": 3.4111893893482192, + "eval_ppl_intron": 3.0320334858159392, + "eval_ppl_nig": 2.7490435846348533, + "eval_ppl_promoter": 3.3412186722882393, + "eval_ppl_utr": 3.3389641964519283, + "step": 4500, + "tokens_trained": 14.741797512 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 1.2791291397773208, + "grad_norm": 0.7720796465873718, + "loss": 1.0754, + "loss_ce": 1.1711397171020508, + "loss_region": 0.0, + "loss_total": 1.1711397171020508, + "lr": 0.0008668106312292358, + "step": 4510, + "tokens_trained": 14.774562152 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 1.2819658180270903, + "grad_norm": 3.019979953765869, + "loss": 1.0704, + "loss_ce": 1.1320418119430542, + "loss_region": 0.0, + "loss_total": 1.1320418119430542, + "lr": 0.0008664784053156146, + "step": 4520, + "tokens_trained": 14.807327592 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 1.2848024962768598, + "grad_norm": 1.7778456211090088, + "loss": 1.0731, + "loss_ce": 1.0580512285232544, + "loss_region": 0.0, + "loss_total": 1.0580512285232544, + "lr": 0.0008661461794019934, + "step": 4530, + "tokens_trained": 14.840092232 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 1.2876391745266293, + "grad_norm": 3.0734405517578125, + "loss": 1.0821, + "loss_ce": 1.1059461832046509, + "loss_region": 0.0, + "loss_total": 1.1059461832046509, + "lr": 0.0008658139534883721, + "step": 4540, + "tokens_trained": 14.872854344 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 1.2904758527763989, + "grad_norm": 1.483434796333313, + "loss": 1.0764, + "loss_ce": 1.1566250324249268, + "loss_region": 0.0, + "loss_total": 1.1566250324249268, + "lr": 0.0008654817275747509, + "step": 4550, + "tokens_trained": 14.905619784 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 1.2933125310261684, + "grad_norm": 1.69507896900177, + "loss": 1.07, + "loss_ce": 1.1269878149032593, + "loss_region": 0.0, + "loss_total": 1.1269878149032593, + "lr": 0.0008651495016611296, + "step": 4560, + "tokens_trained": 14.938385224 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 1.296149209275938, + "grad_norm": 1.4877299070358276, + "loss": 1.0647, + "loss_ce": 1.05030357837677, + "loss_region": 0.0, + "loss_total": 1.05030357837677, + "lr": 0.0008648172757475083, + "step": 4570, + "tokens_trained": 14.971149864 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 1.2989858875257074, + "grad_norm": 0.6229726672172546, + "loss": 1.0688, + "loss_ce": 1.1508759260177612, + "loss_region": 0.0, + "loss_total": 1.1508759260177612, + "lr": 0.0008644850498338871, + "step": 4580, + "tokens_trained": 15.003915304 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 1.301822565775477, + "grad_norm": 1.969948172569275, + "loss": 1.0684, + "loss_ce": 1.034557819366455, + "loss_region": 0.0, + "loss_total": 1.034557819366455, + "lr": 0.0008641528239202659, + "step": 4590, + "tokens_trained": 15.036680744 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 1.3046592440252465, + "grad_norm": 0.9693886041641235, + "loss": 1.0699, + "loss_ce": 1.1122394800186157, + "loss_region": 0.0, + "loss_total": 1.1122394800186157, + "lr": 0.0008638205980066446, + "step": 4600, + "tokens_trained": 15.069442824 + }, + { + "epoch": 1.3046592440252465, + "eval_ppl": 2.9337285496347723, + "eval_runtime": 1.419, + "step": 4600, + "tokens_trained": 15.069442824 + }, + { + "epoch": 1.3046592440252465, + "eval_ppl_cds": 3.632139384263426, + "eval_ppl_dig": 1.1125116545018277, + "eval_ppl_exon": 3.388156997457177, + "eval_ppl_intron": 3.0095075376214013, + "eval_ppl_nig": 2.722866422291749, + "eval_ppl_promoter": 3.3131714047537537, + "eval_ppl_utr": 3.316873156442899, + "step": 4600, + "tokens_trained": 15.069442824 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 1.307495922275016, + "grad_norm": 1.0513139963150024, + "loss": 1.0651, + "loss_ce": 1.2082043886184692, + "loss_region": 0.0, + "loss_total": 1.2082043886184692, + "lr": 0.0008634883720930232, + "step": 4610, + "tokens_trained": 15.102208264 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 1.3103326005247855, + "grad_norm": 1.1261192560195923, + "loss": 1.068, + "loss_ce": 1.0544325113296509, + "loss_region": 0.0, + "loss_total": 1.0544325113296509, + "lr": 0.0008631561461794019, + "step": 4620, + "tokens_trained": 15.134972904 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 1.313169278774555, + "grad_norm": 1.5056110620498657, + "loss": 1.066, + "loss_ce": 1.0044147968292236, + "loss_region": 0.0, + "loss_total": 1.0044147968292236, + "lr": 0.0008628239202657808, + "step": 4630, + "tokens_trained": 15.167735632 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 1.3160059570243245, + "grad_norm": 1.9327380657196045, + "loss": 1.0713, + "loss_ce": 1.0666238069534302, + "loss_region": 0.0, + "loss_total": 1.0666238069534302, + "lr": 0.0008624916943521595, + "step": 4640, + "tokens_trained": 15.200501072 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 1.318842635274094, + "grad_norm": 2.023855209350586, + "loss": 1.0688, + "loss_ce": 1.0664957761764526, + "loss_region": 0.0, + "loss_total": 1.0664957761764526, + "lr": 0.0008621594684385382, + "step": 4650, + "tokens_trained": 15.233266512 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 1.3216793135238636, + "grad_norm": 1.4861963987350464, + "loss": 1.0627, + "loss_ce": 1.1512563228607178, + "loss_region": 0.0, + "loss_total": 1.1512563228607178, + "lr": 0.0008618272425249169, + "step": 4660, + "tokens_trained": 15.266028416 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 1.324515991773633, + "grad_norm": 2.027491569519043, + "loss": 1.0713, + "loss_ce": 1.1003071069717407, + "loss_region": 0.0, + "loss_total": 1.1003071069717407, + "lr": 0.0008614950166112957, + "step": 4670, + "tokens_trained": 15.298793832 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 1.3273526700234026, + "grad_norm": 1.2985600233078003, + "loss": 1.0644, + "loss_ce": 1.1905428171157837, + "loss_region": 0.0, + "loss_total": 1.1905428171157837, + "lr": 0.0008611627906976745, + "step": 4680, + "tokens_trained": 15.331559256 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 1.3301893482731721, + "grad_norm": 0.8797655701637268, + "loss": 1.0693, + "loss_ce": 1.0590300559997559, + "loss_region": 0.0, + "loss_total": 1.0590300559997559, + "lr": 0.0008608305647840532, + "step": 4690, + "tokens_trained": 15.364323896 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 1.3330260265229417, + "grad_norm": 1.4340611696243286, + "loss": 1.0682, + "loss_ce": 1.0440342426300049, + "loss_region": 0.0, + "loss_total": 1.0440342426300049, + "lr": 0.000860498338870432, + "step": 4700, + "tokens_trained": 15.397089336 + }, + { + "epoch": 1.3330260265229417, + "eval_ppl": 2.927314570775946, + "eval_runtime": 1.4206, + "step": 4700, + "tokens_trained": 15.397089336 + }, + { + "epoch": 1.3330260265229417, + "eval_ppl_cds": 3.6287376839304666, + "eval_ppl_dig": 1.1150079554071568, + "eval_ppl_exon": 3.383354334689886, + "eval_ppl_intron": 3.0061466394942973, + "eval_ppl_nig": 2.7199588420912537, + "eval_ppl_promoter": 3.3071915768535316, + "eval_ppl_utr": 3.304233169338821, + "step": 4700, + "tokens_trained": 15.397089336 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 1.3358627047727112, + "grad_norm": 1.4811822175979614, + "loss": 1.0594, + "loss_ce": 0.9994807839393616, + "loss_region": 0.0, + "loss_total": 0.9994807839393616, + "lr": 0.0008601661129568107, + "step": 4710, + "tokens_trained": 15.42985476 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 1.3386993830224807, + "grad_norm": 0.900145411491394, + "loss": 1.0688, + "loss_ce": 1.0909019708633423, + "loss_region": 0.0, + "loss_total": 1.0909019708633423, + "lr": 0.0008598338870431894, + "step": 4720, + "tokens_trained": 15.4626202 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 1.3415360612722502, + "grad_norm": 1.8021804094314575, + "loss": 1.0637, + "loss_ce": 0.9789250493049622, + "loss_region": 0.0, + "loss_total": 0.9789250493049622, + "lr": 0.0008595016611295681, + "step": 4730, + "tokens_trained": 15.49538484 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 1.3443727395220197, + "grad_norm": 1.7707029581069946, + "loss": 1.0586, + "loss_ce": 1.0249565839767456, + "loss_region": 0.0, + "loss_total": 1.0249565839767456, + "lr": 0.0008591694352159468, + "step": 4740, + "tokens_trained": 15.52815028 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 1.3472094177717893, + "grad_norm": 2.088066577911377, + "loss": 1.0617, + "loss_ce": 1.1340487003326416, + "loss_region": 0.0, + "loss_total": 1.1340487003326416, + "lr": 0.0008588372093023256, + "step": 4750, + "tokens_trained": 15.56091572 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 1.3500460960215588, + "grad_norm": 1.283279299736023, + "loss": 1.0684, + "loss_ce": 0.9195445775985718, + "loss_region": 0.0, + "loss_total": 0.9195445775985718, + "lr": 0.0008585049833887043, + "step": 4760, + "tokens_trained": 15.593678736 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 1.3528827742713283, + "grad_norm": 1.7129117250442505, + "loss": 1.0627, + "loss_ce": 1.0699961185455322, + "loss_region": 0.0, + "loss_total": 1.0699961185455322, + "lr": 0.000858172757475083, + "step": 4770, + "tokens_trained": 15.626444176 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 1.3557194525210978, + "grad_norm": 1.7075800895690918, + "loss": 1.0644, + "loss_ce": 1.0860153436660767, + "loss_region": 0.0, + "loss_total": 1.0860153436660767, + "lr": 0.0008578405315614619, + "step": 4780, + "tokens_trained": 15.659209616 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 1.3585561307708673, + "grad_norm": 1.3361881971359253, + "loss": 1.0658, + "loss_ce": 1.1329466104507446, + "loss_region": 0.0, + "loss_total": 1.1329466104507446, + "lr": 0.0008575083056478406, + "step": 4790, + "tokens_trained": 15.691975056 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 1.3613928090206369, + "grad_norm": 0.8998346328735352, + "loss": 1.068, + "loss_ce": 1.161861777305603, + "loss_region": 0.0, + "loss_total": 1.161861777305603, + "lr": 0.0008571760797342193, + "step": 4800, + "tokens_trained": 15.724740496 + }, + { + "epoch": 1.3613928090206369, + "eval_ppl": 2.9155986115291688, + "eval_runtime": 1.4187, + "step": 4800, + "tokens_trained": 15.724740496 + }, + { + "epoch": 1.3613928090206369, + "eval_ppl_cds": 3.5765631136705327, + "eval_ppl_dig": 1.1111050315798496, + "eval_ppl_exon": 3.3735210373367397, + "eval_ppl_intron": 2.9979755180823666, + "eval_ppl_nig": 2.7025535848418203, + "eval_ppl_promoter": 3.2877562515029997, + "eval_ppl_utr": 3.2984488220318973, + "step": 4800, + "tokens_trained": 15.724740496 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 1.3642294872704064, + "grad_norm": 1.6598438024520874, + "loss": 1.0624, + "loss_ce": 1.123659610748291, + "loss_region": 0.0, + "loss_total": 1.123659610748291, + "lr": 0.000856843853820598, + "step": 4810, + "tokens_trained": 15.757505936 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 1.367066165520176, + "grad_norm": 2.1694178581237793, + "loss": 1.0574, + "loss_ce": 1.0363062620162964, + "loss_region": 0.0, + "loss_total": 1.0363062620162964, + "lr": 0.0008565116279069768, + "step": 4820, + "tokens_trained": 15.790271376 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 1.3699028437699454, + "grad_norm": 0.4685896933078766, + "loss": 1.0678, + "loss_ce": 1.0816088914871216, + "loss_region": 0.0, + "loss_total": 1.0816088914871216, + "lr": 0.0008561794019933556, + "step": 4830, + "tokens_trained": 15.823036016 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 1.372739522019715, + "grad_norm": 1.0896044969558716, + "loss": 1.0613, + "loss_ce": 0.9804879426956177, + "loss_region": 0.0, + "loss_total": 0.9804879426956177, + "lr": 0.0008558471760797343, + "step": 4840, + "tokens_trained": 15.855801456 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 1.3755762002694845, + "grad_norm": 1.1067882776260376, + "loss": 1.0574, + "loss_ce": 0.9730198979377747, + "loss_region": 0.0, + "loss_total": 0.9730198979377747, + "lr": 0.000855514950166113, + "step": 4850, + "tokens_trained": 15.888566096 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 1.378412878519254, + "grad_norm": 2.1310548782348633, + "loss": 1.0589, + "loss_ce": 1.0982823371887207, + "loss_region": 0.0, + "loss_total": 1.0982823371887207, + "lr": 0.0008551827242524917, + "step": 4860, + "tokens_trained": 15.921331536 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 1.3812495567690235, + "grad_norm": 1.645630121231079, + "loss": 1.0581, + "loss_ce": 1.0913256406784058, + "loss_region": 0.0, + "loss_total": 1.0913256406784058, + "lr": 0.0008548504983388704, + "step": 4870, + "tokens_trained": 15.954096176 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 1.384086235018793, + "grad_norm": 1.103580117225647, + "loss": 1.0614, + "loss_ce": 1.1088241338729858, + "loss_region": 0.0, + "loss_total": 1.1088241338729858, + "lr": 0.0008545182724252492, + "step": 4880, + "tokens_trained": 15.9868564 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 1.3869229132685625, + "grad_norm": 2.0858447551727295, + "loss": 1.0568, + "loss_ce": 1.0553276538848877, + "loss_region": 0.0, + "loss_total": 1.0553276538848877, + "lr": 0.0008541860465116279, + "step": 4890, + "tokens_trained": 16.019621784 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 1.389759591518332, + "grad_norm": 1.5583046674728394, + "loss": 1.0621, + "loss_ce": 1.1190567016601562, + "loss_region": 0.0, + "loss_total": 1.1190567016601562, + "lr": 0.0008538538205980067, + "step": 4900, + "tokens_trained": 16.052387224 + }, + { + "epoch": 1.389759591518332, + "eval_ppl": 2.9063139387838666, + "eval_runtime": 1.4207, + "step": 4900, + "tokens_trained": 16.052387224 + }, + { + "epoch": 1.389759591518332, + "eval_ppl_cds": 3.5782616116630157, + "eval_ppl_dig": 1.112130312437596, + "eval_ppl_exon": 3.3613197755557667, + "eval_ppl_intron": 2.9875463324800204, + "eval_ppl_nig": 2.696276058535183, + "eval_ppl_promoter": 3.282297812744047, + "eval_ppl_utr": 3.2951855448902316, + "step": 4900, + "tokens_trained": 16.052387224 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 1.3925962697681016, + "grad_norm": 1.5331727266311646, + "loss": 1.058, + "loss_ce": 0.9149652123451233, + "loss_region": 0.0, + "loss_total": 0.9149652123451233, + "lr": 0.0008535215946843854, + "step": 4910, + "tokens_trained": 16.085152664 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 1.395432948017871, + "grad_norm": 1.3589428663253784, + "loss": 1.0627, + "loss_ce": 1.0038203001022339, + "loss_region": 0.0, + "loss_total": 1.0038203001022339, + "lr": 0.0008531893687707641, + "step": 4920, + "tokens_trained": 16.117918104 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 1.3982696262676406, + "grad_norm": 1.6175169944763184, + "loss": 1.0629, + "loss_ce": 1.0087686777114868, + "loss_region": 0.0, + "loss_total": 1.0087686777114868, + "lr": 0.0008528571428571428, + "step": 4930, + "tokens_trained": 16.150683544 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 1.4011063045174101, + "grad_norm": 1.5868284702301025, + "loss": 1.0582, + "loss_ce": 1.0160523653030396, + "loss_region": 0.0, + "loss_total": 1.0160523653030396, + "lr": 0.0008525249169435217, + "step": 4940, + "tokens_trained": 16.183448184 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 1.4039429827671797, + "grad_norm": 1.1419873237609863, + "loss": 1.0595, + "loss_ce": 1.0400149822235107, + "loss_region": 0.0, + "loss_total": 1.0400149822235107, + "lr": 0.0008521926910299004, + "step": 4950, + "tokens_trained": 16.216210576 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 1.4067796610169492, + "grad_norm": 1.48302161693573, + "loss": 1.0577, + "loss_ce": 1.1253412961959839, + "loss_region": 0.0, + "loss_total": 1.1253412961959839, + "lr": 0.0008518604651162791, + "step": 4960, + "tokens_trained": 16.248975216 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 1.4096163392667187, + "grad_norm": 0.5538215041160583, + "loss": 1.0589, + "loss_ce": 1.0566335916519165, + "loss_region": 0.0, + "loss_total": 1.0566335916519165, + "lr": 0.0008515282392026578, + "step": 4970, + "tokens_trained": 16.281739056 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 1.4124530175164882, + "grad_norm": 0.7753979563713074, + "loss": 1.0592, + "loss_ce": 1.0776869058609009, + "loss_region": 0.0, + "loss_total": 1.0776869058609009, + "lr": 0.0008511960132890365, + "step": 4980, + "tokens_trained": 16.314504496 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 1.4152896957662577, + "grad_norm": 0.2707841992378235, + "loss": 1.0581, + "loss_ce": 0.939789354801178, + "loss_region": 0.0, + "loss_total": 0.939789354801178, + "lr": 0.0008508637873754153, + "step": 4990, + "tokens_trained": 16.347269936 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 1.4181263740160273, + "grad_norm": 0.4657887816429138, + "loss": 1.0597, + "loss_ce": 1.099407434463501, + "loss_region": 0.0, + "loss_total": 1.099407434463501, + "lr": 0.000850531561461794, + "step": 5000, + "tokens_trained": 16.380034576 + }, + { + "epoch": 1.4181263740160273, + "eval_ppl": 2.917652757322049, + "eval_runtime": 1.4212, + "step": 5000, + "tokens_trained": 16.380034576 + }, + { + "epoch": 1.4181263740160273, + "eval_ppl_cds": 3.6315611784703834, + "eval_ppl_dig": 1.1034784136005515, + "eval_ppl_exon": 3.3838010857288023, + "eval_ppl_intron": 3.0001868222993995, + "eval_ppl_nig": 2.706846989775424, + "eval_ppl_promoter": 3.2977886112047248, + "eval_ppl_utr": 3.3139066059901396, + "step": 5000, + "tokens_trained": 16.380034576 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 1.4209630522657968, + "grad_norm": 2.072887420654297, + "loss": 1.0588, + "loss_ce": 1.0940048694610596, + "loss_region": 0.0, + "loss_total": 1.0940048694610596, + "lr": 0.0008501993355481727, + "step": 5010, + "tokens_trained": 16.412800016 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 1.4237997305155663, + "grad_norm": 1.8767870664596558, + "loss": 1.0612, + "loss_ce": 1.079982876777649, + "loss_region": 0.0, + "loss_total": 1.079982876777649, + "lr": 0.0008498671096345515, + "step": 5020, + "tokens_trained": 16.445565456 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 1.4266364087653358, + "grad_norm": 2.648505449295044, + "loss": 1.0527, + "loss_ce": 1.0782657861709595, + "loss_region": 0.0, + "loss_total": 1.0782657861709595, + "lr": 0.0008495348837209302, + "step": 5030, + "tokens_trained": 16.478329936 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 1.4294730870151053, + "grad_norm": 1.6128559112548828, + "loss": 1.0486, + "loss_ce": 1.1545112133026123, + "loss_region": 0.0, + "loss_total": 1.1545112133026123, + "lr": 0.000849202657807309, + "step": 5040, + "tokens_trained": 16.511090344 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 1.4323097652648749, + "grad_norm": 1.9148733615875244, + "loss": 1.0475, + "loss_ce": 0.9982237815856934, + "loss_region": 0.0, + "loss_total": 0.9982237815856934, + "lr": 0.0008488704318936877, + "step": 5050, + "tokens_trained": 16.543855784 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 1.4351464435146444, + "grad_norm": 1.6048266887664795, + "loss": 1.0625, + "loss_ce": 1.1319563388824463, + "loss_region": 0.0, + "loss_total": 1.1319563388824463, + "lr": 0.0008485382059800665, + "step": 5060, + "tokens_trained": 16.576621224 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 1.437983121764414, + "grad_norm": 2.0525214672088623, + "loss": 1.0573, + "loss_ce": 0.9687820076942444, + "loss_region": 0.0, + "loss_total": 0.9687820076942444, + "lr": 0.0008482059800664452, + "step": 5070, + "tokens_trained": 16.609386664 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 1.4408198000141834, + "grad_norm": 2.890258550643921, + "loss": 1.0609, + "loss_ce": 0.8600770831108093, + "loss_region": 0.0, + "loss_total": 0.8600770831108093, + "lr": 0.0008478737541528239, + "step": 5080, + "tokens_trained": 16.642152104 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 1.443656478263953, + "grad_norm": 0.9747489094734192, + "loss": 1.0551, + "loss_ce": 1.1045405864715576, + "loss_region": 0.0, + "loss_total": 1.1045405864715576, + "lr": 0.0008475415282392028, + "step": 5090, + "tokens_trained": 16.674914848 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 1.4464931565137225, + "grad_norm": 1.1794837713241577, + "loss": 1.0503, + "loss_ce": 1.143156886100769, + "loss_region": 0.0, + "loss_total": 1.143156886100769, + "lr": 0.0008472093023255815, + "step": 5100, + "tokens_trained": 16.707680128 + }, + { + "epoch": 1.4464931565137225, + "eval_ppl": 2.897575532050937, + "eval_runtime": 1.4192, + "step": 5100, + "tokens_trained": 16.707680128 + }, + { + "epoch": 1.4464931565137225, + "eval_ppl_cds": 3.5527688517387013, + "eval_ppl_dig": 1.1050475077199255, + "eval_ppl_exon": 3.3708244860185155, + "eval_ppl_intron": 2.9817903106266685, + "eval_ppl_nig": 2.696388405532854, + "eval_ppl_promoter": 3.2704429431257855, + "eval_ppl_utr": 3.296065096964438, + "step": 5100, + "tokens_trained": 16.707680128 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 1.449329834763492, + "grad_norm": 1.6240310668945312, + "loss": 1.0527, + "loss_ce": 1.1450821161270142, + "loss_region": 0.0, + "loss_total": 1.1450821161270142, + "lr": 0.0008468770764119601, + "step": 5110, + "tokens_trained": 16.740445568 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 1.4521665130132615, + "grad_norm": 3.8531057834625244, + "loss": 1.0555, + "loss_ce": 1.156142234802246, + "loss_region": 0.0, + "loss_total": 1.156142234802246, + "lr": 0.0008465448504983388, + "step": 5120, + "tokens_trained": 16.773211008 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 1.455003191263031, + "grad_norm": 1.4156396389007568, + "loss": 1.0762, + "loss_ce": 1.0218816995620728, + "loss_region": 0.0, + "loss_total": 1.0218816995620728, + "lr": 0.0008462126245847175, + "step": 5130, + "tokens_trained": 16.805971784 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 1.4578398695128005, + "grad_norm": 1.3686301708221436, + "loss": 1.0627, + "loss_ce": 1.0172085762023926, + "loss_region": 0.0, + "loss_total": 1.0172085762023926, + "lr": 0.0008458803986710964, + "step": 5140, + "tokens_trained": 16.838737224 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 1.46067654776257, + "grad_norm": 0.8231141567230225, + "loss": 1.0526, + "loss_ce": 1.078109622001648, + "loss_region": 0.0, + "loss_total": 1.078109622001648, + "lr": 0.0008455481727574751, + "step": 5150, + "tokens_trained": 16.871502664 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 1.4635132260123396, + "grad_norm": 1.5997247695922852, + "loss": 1.0497, + "loss_ce": 1.1104001998901367, + "loss_region": 0.0, + "loss_total": 1.1104001998901367, + "lr": 0.0008452159468438538, + "step": 5160, + "tokens_trained": 16.904268104 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 1.466349904262109, + "grad_norm": 2.271498203277588, + "loss": 1.0557, + "loss_ce": 1.0757694244384766, + "loss_region": 0.0, + "loss_total": 1.0757694244384766, + "lr": 0.0008448837209302326, + "step": 5170, + "tokens_trained": 16.937030128 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 1.4691865825118786, + "grad_norm": 1.2470595836639404, + "loss": 1.0565, + "loss_ce": 1.091043472290039, + "loss_region": 0.0, + "loss_total": 1.091043472290039, + "lr": 0.0008445514950166113, + "step": 5180, + "tokens_trained": 16.969795184 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 1.4720232607616481, + "grad_norm": 1.2891720533370972, + "loss": 1.0565, + "loss_ce": 1.0341897010803223, + "loss_region": 0.0, + "loss_total": 1.0341897010803223, + "lr": 0.0008442192691029901, + "step": 5190, + "tokens_trained": 17.002560624 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 1.4748599390114177, + "grad_norm": 1.3190512657165527, + "loss": 1.0596, + "loss_ce": 1.1108061075210571, + "loss_region": 0.0, + "loss_total": 1.1108061075210571, + "lr": 0.0008438870431893688, + "step": 5200, + "tokens_trained": 17.035326064 + }, + { + "epoch": 1.4748599390114177, + "eval_ppl": 2.8862166793921706, + "eval_runtime": 1.4211, + "step": 5200, + "tokens_trained": 17.035326064 + }, + { + "epoch": 1.4748599390114177, + "eval_ppl_cds": 3.5323927522436374, + "eval_ppl_dig": 1.1055318771688754, + "eval_ppl_exon": 3.3392595571861183, + "eval_ppl_intron": 2.9679053418375716, + "eval_ppl_nig": 2.6785872453463977, + "eval_ppl_promoter": 3.249890525312008, + "eval_ppl_utr": 3.2765661887525406, + "step": 5200, + "tokens_trained": 17.035326064 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 1.4776966172611872, + "grad_norm": 1.4953867197036743, + "loss": 1.0487, + "loss_ce": 1.003316879272461, + "loss_region": 0.0, + "loss_total": 1.003316879272461, + "lr": 0.0008435548172757476, + "step": 5210, + "tokens_trained": 17.068091504 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 1.4805332955109567, + "grad_norm": 1.14693021774292, + "loss": 1.0458, + "loss_ce": 1.015944242477417, + "loss_region": 0.0, + "loss_total": 1.015944242477417, + "lr": 0.0008432225913621263, + "step": 5220, + "tokens_trained": 17.100856944 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 1.4833699737607262, + "grad_norm": 1.691019892692566, + "loss": 1.0543, + "loss_ce": 1.0680232048034668, + "loss_region": 0.0, + "loss_total": 1.0680232048034668, + "lr": 0.0008428903654485049, + "step": 5230, + "tokens_trained": 17.133622384 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 1.4862066520104957, + "grad_norm": 2.2451138496398926, + "loss": 1.0454, + "loss_ce": 1.0888233184814453, + "loss_region": 0.0, + "loss_total": 1.0888233184814453, + "lr": 0.0008425581395348837, + "step": 5240, + "tokens_trained": 17.166385424 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 1.4890433302602653, + "grad_norm": 1.161699652671814, + "loss": 1.0542, + "loss_ce": 0.8528671860694885, + "loss_region": 0.0, + "loss_total": 0.8528671860694885, + "lr": 0.0008422259136212625, + "step": 5250, + "tokens_trained": 17.199150864 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 1.4918800085100348, + "grad_norm": 0.8473041653633118, + "loss": 1.0557, + "loss_ce": 1.0087968111038208, + "loss_region": 0.0, + "loss_total": 1.0087968111038208, + "lr": 0.0008418936877076412, + "step": 5260, + "tokens_trained": 17.231916304 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 1.4947166867598043, + "grad_norm": 2.06170654296875, + "loss": 1.0514, + "loss_ce": 1.0638319253921509, + "loss_region": 0.0, + "loss_total": 1.0638319253921509, + "lr": 0.0008415614617940199, + "step": 5270, + "tokens_trained": 17.264676432 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 1.4975533650095738, + "grad_norm": 1.3733689785003662, + "loss": 1.0471, + "loss_ce": 0.9597594738006592, + "loss_region": 0.0, + "loss_total": 0.9597594738006592, + "lr": 0.0008412292358803986, + "step": 5280, + "tokens_trained": 17.297440272 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 1.5003900432593433, + "grad_norm": 1.1995199918746948, + "loss": 1.0556, + "loss_ce": 1.1026781797409058, + "loss_region": 0.0, + "loss_total": 1.1026781797409058, + "lr": 0.0008408970099667775, + "step": 5290, + "tokens_trained": 17.330205712 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 1.5032267215091129, + "grad_norm": 1.8619412183761597, + "loss": 1.0507, + "loss_ce": 1.090835690498352, + "loss_region": 0.0, + "loss_total": 1.090835690498352, + "lr": 0.0008405647840531562, + "step": 5300, + "tokens_trained": 17.362971152 + }, + { + "epoch": 1.5032267215091129, + "eval_ppl": 2.886463383708472, + "eval_runtime": 1.4305, + "step": 5300, + "tokens_trained": 17.362971152 + }, + { + "epoch": 1.5032267215091129, + "eval_ppl_cds": 3.5554146161519977, + "eval_ppl_dig": 1.0980911626316492, + "eval_ppl_exon": 3.3588592627779037, + "eval_ppl_intron": 2.9742722701513933, + "eval_ppl_nig": 2.6757396148493595, + "eval_ppl_promoter": 3.2582604041509766, + "eval_ppl_utr": 3.2757323855415903, + "step": 5300, + "tokens_trained": 17.362971152 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 1.5060633997588824, + "grad_norm": 1.716063141822815, + "loss": 1.0536, + "loss_ce": 1.126861810684204, + "loss_region": 0.0, + "loss_total": 1.126861810684204, + "lr": 0.0008402325581395349, + "step": 5310, + "tokens_trained": 17.395736592 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 1.508900078008652, + "grad_norm": 1.3118699789047241, + "loss": 1.054, + "loss_ce": 1.0839499235153198, + "loss_region": 0.0, + "loss_total": 1.0839499235153198, + "lr": 0.0008399003322259136, + "step": 5320, + "tokens_trained": 17.428502032 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 1.5117367562584214, + "grad_norm": 2.032186508178711, + "loss": 1.0503, + "loss_ce": 1.0781562328338623, + "loss_region": 0.0, + "loss_total": 1.0781562328338623, + "lr": 0.0008395681063122924, + "step": 5330, + "tokens_trained": 17.461262816 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 1.514573434508191, + "grad_norm": 0.6521492004394531, + "loss": 1.0487, + "loss_ce": 1.0017558336257935, + "loss_region": 0.0, + "loss_total": 1.0017558336257935, + "lr": 0.0008392358803986712, + "step": 5340, + "tokens_trained": 17.494028256 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 1.5174101127579604, + "grad_norm": 1.7442623376846313, + "loss": 1.0591, + "loss_ce": 1.0969889163970947, + "loss_region": 0.0, + "loss_total": 1.0969889163970947, + "lr": 0.0008389036544850499, + "step": 5350, + "tokens_trained": 17.526793696 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 1.52024679100773, + "grad_norm": 1.4223484992980957, + "loss": 1.0525, + "loss_ce": 1.000662088394165, + "loss_region": 0.0, + "loss_total": 1.000662088394165, + "lr": 0.0008385714285714285, + "step": 5360, + "tokens_trained": 17.559559136 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 1.5230834692574995, + "grad_norm": 1.773373007774353, + "loss": 1.0469, + "loss_ce": 1.07241952419281, + "loss_region": 0.0, + "loss_total": 1.07241952419281, + "lr": 0.0008382392026578073, + "step": 5370, + "tokens_trained": 17.59232456 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 1.525920147507269, + "grad_norm": 1.1737433671951294, + "loss": 1.0524, + "loss_ce": 0.9929547309875488, + "loss_region": 0.0, + "loss_total": 0.9929547309875488, + "lr": 0.000837906976744186, + "step": 5380, + "tokens_trained": 17.6250892 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 1.5287568257570385, + "grad_norm": 1.2984753847122192, + "loss": 1.052, + "loss_ce": 1.0053417682647705, + "loss_region": 0.0, + "loss_total": 1.0053417682647705, + "lr": 0.0008375747508305648, + "step": 5390, + "tokens_trained": 17.65785464 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 1.531593504006808, + "grad_norm": 1.652105450630188, + "loss": 1.056, + "loss_ce": 1.1186614036560059, + "loss_region": 0.0, + "loss_total": 1.1186614036560059, + "lr": 0.0008372425249169435, + "step": 5400, + "tokens_trained": 17.69062008 + }, + { + "epoch": 1.531593504006808, + "eval_ppl": 2.866062614625813, + "eval_runtime": 1.4204, + "step": 5400, + "tokens_trained": 17.69062008 + }, + { + "epoch": 1.531593504006808, + "eval_ppl_cds": 3.4986651782131593, + "eval_ppl_dig": 1.0988525645268432, + "eval_ppl_exon": 3.3368998789772926, + "eval_ppl_intron": 2.95435590777396, + "eval_ppl_nig": 2.6642326516350754, + "eval_ppl_promoter": 3.228823861154718, + "eval_ppl_utr": 3.2577459536329574, + "step": 5400, + "tokens_trained": 17.69062008 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 1.5344301822565776, + "grad_norm": 1.452106237411499, + "loss": 1.047, + "loss_ce": 1.0881328582763672, + "loss_region": 0.0, + "loss_total": 1.0881328582763672, + "lr": 0.0008369102990033223, + "step": 5410, + "tokens_trained": 17.72338552 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 1.537266860506347, + "grad_norm": 1.549715280532837, + "loss": 1.0502, + "loss_ce": 1.174644947052002, + "loss_region": 0.0, + "loss_total": 1.174644947052002, + "lr": 0.000836578073089701, + "step": 5420, + "tokens_trained": 17.75615016 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 1.5401035387561166, + "grad_norm": 1.0829674005508423, + "loss": 1.0538, + "loss_ce": 1.0568816661834717, + "loss_region": 0.0, + "loss_total": 1.0568816661834717, + "lr": 0.0008362458471760797, + "step": 5430, + "tokens_trained": 17.78891464 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 1.5429402170058861, + "grad_norm": 0.8888135552406311, + "loss": 1.0464, + "loss_ce": 1.0331439971923828, + "loss_region": 0.0, + "loss_total": 1.0331439971923828, + "lr": 0.0008359136212624586, + "step": 5440, + "tokens_trained": 17.821677712 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 1.5457768952556556, + "grad_norm": 1.363316535949707, + "loss": 1.0439, + "loss_ce": 1.1340237855911255, + "loss_region": 0.0, + "loss_total": 1.1340237855911255, + "lr": 0.0008355813953488373, + "step": 5450, + "tokens_trained": 17.854443152 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 1.5486135735054252, + "grad_norm": 1.275102972984314, + "loss": 1.0507, + "loss_ce": 1.033170461654663, + "loss_region": 0.0, + "loss_total": 1.033170461654663, + "lr": 0.000835249169435216, + "step": 5460, + "tokens_trained": 17.88720588 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 1.5514502517551947, + "grad_norm": 1.6826164722442627, + "loss": 1.0468, + "loss_ce": 1.0219982862472534, + "loss_region": 0.0, + "loss_total": 1.0219982862472534, + "lr": 0.0008349169435215947, + "step": 5470, + "tokens_trained": 17.91997132 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 1.5542869300049642, + "grad_norm": 1.2667180299758911, + "loss": 1.0392, + "loss_ce": 0.8995307087898254, + "loss_region": 0.0, + "loss_total": 0.8995307087898254, + "lr": 0.0008345847176079733, + "step": 5480, + "tokens_trained": 17.952736472 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 1.5571236082547337, + "grad_norm": 2.5749096870422363, + "loss": 1.0496, + "loss_ce": 1.1163636445999146, + "loss_region": 0.0, + "loss_total": 1.1163636445999146, + "lr": 0.0008342524916943522, + "step": 5490, + "tokens_trained": 17.985498936 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 1.5599602865045032, + "grad_norm": 1.0826396942138672, + "loss": 1.0508, + "loss_ce": 1.090194582939148, + "loss_region": 0.0, + "loss_total": 1.090194582939148, + "lr": 0.0008339202657807309, + "step": 5500, + "tokens_trained": 18.018264376 + }, + { + "epoch": 1.5599602865045032, + "eval_ppl": 2.8618990823334367, + "eval_runtime": 1.4227, + "step": 5500, + "tokens_trained": 18.018264376 + }, + { + "epoch": 1.5599602865045032, + "eval_ppl_cds": 3.513708680948228, + "eval_ppl_dig": 1.0994976691520244, + "eval_ppl_exon": 3.3313078692611313, + "eval_ppl_intron": 2.9539192193358517, + "eval_ppl_nig": 2.66110168648567, + "eval_ppl_promoter": 3.2358122799677385, + "eval_ppl_utr": 3.2634029424645847, + "step": 5500, + "tokens_trained": 18.018264376 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 1.5627969647542728, + "grad_norm": 0.8231706619262695, + "loss": 1.0447, + "loss_ce": 1.0191712379455566, + "loss_region": 0.0, + "loss_total": 1.0191712379455566, + "lr": 0.0008335880398671096, + "step": 5510, + "tokens_trained": 18.051029816 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 1.5656336430040423, + "grad_norm": 1.9712238311767578, + "loss": 1.0447, + "loss_ce": 1.0317987203598022, + "loss_region": 0.0, + "loss_total": 1.0317987203598022, + "lr": 0.0008332558139534884, + "step": 5520, + "tokens_trained": 18.083794456 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 1.5684703212538118, + "grad_norm": 1.25809645652771, + "loss": 1.0399, + "loss_ce": 0.9980016350746155, + "loss_region": 0.0, + "loss_total": 0.9980016350746155, + "lr": 0.0008329235880398671, + "step": 5530, + "tokens_trained": 18.116559896 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 1.5713069995035813, + "grad_norm": 1.1979594230651855, + "loss": 1.0456, + "loss_ce": 1.1264092922210693, + "loss_region": 0.0, + "loss_total": 1.1264092922210693, + "lr": 0.0008325913621262459, + "step": 5540, + "tokens_trained": 18.149325336 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 1.5741436777533508, + "grad_norm": 1.372471570968628, + "loss": 1.0463, + "loss_ce": 1.0032322406768799, + "loss_region": 0.0, + "loss_total": 1.0032322406768799, + "lr": 0.0008322591362126246, + "step": 5550, + "tokens_trained": 18.182090776 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 1.5769803560031204, + "grad_norm": 0.895865797996521, + "loss": 1.0446, + "loss_ce": 0.8139809966087341, + "loss_region": 0.0, + "loss_total": 0.8139809966087341, + "lr": 0.0008319269102990034, + "step": 5560, + "tokens_trained": 18.214852704 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 1.5798170342528899, + "grad_norm": 1.7960448265075684, + "loss": 1.0549, + "loss_ce": 1.1213290691375732, + "loss_region": 0.0, + "loss_total": 1.1213290691375732, + "lr": 0.0008315946843853821, + "step": 5570, + "tokens_trained": 18.247618144 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 1.5826537125026594, + "grad_norm": 1.3914803266525269, + "loss": 1.0465, + "loss_ce": 0.9918665885925293, + "loss_region": 0.0, + "loss_total": 0.9918665885925293, + "lr": 0.0008312624584717608, + "step": 5580, + "tokens_trained": 18.280383584 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 1.585490390752429, + "grad_norm": 1.5150033235549927, + "loss": 1.0415, + "loss_ce": 0.920126736164093, + "loss_region": 0.0, + "loss_total": 0.920126736164093, + "lr": 0.0008309302325581396, + "step": 5590, + "tokens_trained": 18.313149024 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 1.5883270690021984, + "grad_norm": 1.222332239151001, + "loss": 1.0393, + "loss_ce": 1.0348728895187378, + "loss_region": 0.0, + "loss_total": 1.0348728895187378, + "lr": 0.0008305980066445184, + "step": 5600, + "tokens_trained": 18.345914464 + }, + { + "epoch": 1.5883270690021984, + "eval_ppl": 2.866112839275224, + "eval_runtime": 1.4159, + "step": 5600, + "tokens_trained": 18.345914464 + }, + { + "epoch": 1.5883270690021984, + "eval_ppl_cds": 3.5300214288497194, + "eval_ppl_dig": 1.0987145686882205, + "eval_ppl_exon": 3.3436452543843154, + "eval_ppl_intron": 2.960288360504182, + "eval_ppl_nig": 2.668121181702922, + "eval_ppl_promoter": 3.2286686678209757, + "eval_ppl_utr": 3.267289201879138, + "step": 5600, + "tokens_trained": 18.345914464 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 1.591163747251968, + "grad_norm": 1.3769258260726929, + "loss": 1.0448, + "loss_ce": 1.0562052726745605, + "loss_region": 0.0, + "loss_total": 1.0562052726745605, + "lr": 0.000830265780730897, + "step": 5610, + "tokens_trained": 18.378679104 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 1.5940004255017375, + "grad_norm": 1.2464054822921753, + "loss": 1.0371, + "loss_ce": 0.9268790483474731, + "loss_region": 0.0, + "loss_total": 0.9268790483474731, + "lr": 0.0008299335548172757, + "step": 5620, + "tokens_trained": 18.411444544 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 1.596837103751507, + "grad_norm": 1.2346057891845703, + "loss": 1.0431, + "loss_ce": 1.007016897201538, + "loss_region": 0.0, + "loss_total": 1.007016897201538, + "lr": 0.0008296013289036544, + "step": 5630, + "tokens_trained": 18.444209984 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 1.5996737820012765, + "grad_norm": 1.6421316862106323, + "loss": 1.0362, + "loss_ce": 0.9929165840148926, + "loss_region": 0.0, + "loss_total": 0.9929165840148926, + "lr": 0.0008292691029900333, + "step": 5640, + "tokens_trained": 18.476975424 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 1.602510460251046, + "grad_norm": 1.6476022005081177, + "loss": 1.043, + "loss_ce": 1.1608846187591553, + "loss_region": 0.0, + "loss_total": 1.1608846187591553, + "lr": 0.000828936877076412, + "step": 5650, + "tokens_trained": 18.509739264 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 1.6053471385008156, + "grad_norm": 1.0944863557815552, + "loss": 1.0422, + "loss_ce": 0.939422607421875, + "loss_region": 0.0, + "loss_total": 0.939422607421875, + "lr": 0.0008286046511627907, + "step": 5660, + "tokens_trained": 18.542503904 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 1.608183816750585, + "grad_norm": 1.5539265871047974, + "loss": 1.039, + "loss_ce": 1.0551543235778809, + "loss_region": 0.0, + "loss_total": 1.0551543235778809, + "lr": 0.0008282724252491694, + "step": 5670, + "tokens_trained": 18.575269344 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 1.6110204950003546, + "grad_norm": 1.5520743131637573, + "loss": 1.0358, + "loss_ce": 1.002668857574463, + "loss_region": 0.0, + "loss_total": 1.002668857574463, + "lr": 0.0008279401993355482, + "step": 5680, + "tokens_trained": 18.608031832 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 1.6138571732501241, + "grad_norm": 1.4361125230789185, + "loss": 1.0397, + "loss_ce": 1.1137359142303467, + "loss_region": 0.0, + "loss_total": 1.1137359142303467, + "lr": 0.000827607973421927, + "step": 5690, + "tokens_trained": 18.640797272 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 1.6166938514998936, + "grad_norm": 1.3013025522232056, + "loss": 1.0409, + "loss_ce": 0.9689463973045349, + "loss_region": 0.0, + "loss_total": 0.9689463973045349, + "lr": 0.0008272757475083057, + "step": 5700, + "tokens_trained": 18.673562712 + }, + { + "epoch": 1.6166938514998936, + "eval_ppl": 2.857052848003976, + "eval_runtime": 1.4202, + "step": 5700, + "tokens_trained": 18.673562712 + }, + { + "epoch": 1.6166938514998936, + "eval_ppl_cds": 3.505538790091151, + "eval_ppl_dig": 1.1021110142347539, + "eval_ppl_exon": 3.339895472270823, + "eval_ppl_intron": 2.95662890167907, + "eval_ppl_nig": 2.6468828951681234, + "eval_ppl_promoter": 3.1926960745592456, + "eval_ppl_utr": 3.2560670369463924, + "step": 5700, + "tokens_trained": 18.673562712 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 1.6195305297496632, + "grad_norm": 1.0005744695663452, + "loss": 1.0386, + "loss_ce": 1.0522420406341553, + "loss_region": 0.0, + "loss_total": 1.0522420406341553, + "lr": 0.0008269435215946845, + "step": 5710, + "tokens_trained": 18.706328152 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 1.6223672079994327, + "grad_norm": 1.77809739112854, + "loss": 1.0453, + "loss_ce": 1.1413697004318237, + "loss_region": 0.0, + "loss_total": 1.1413697004318237, + "lr": 0.0008266112956810632, + "step": 5720, + "tokens_trained": 18.739093568 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 1.6252038862492022, + "grad_norm": 1.0831490755081177, + "loss": 1.0425, + "loss_ce": 1.0019220113754272, + "loss_region": 0.0, + "loss_total": 1.0019220113754272, + "lr": 0.0008262790697674418, + "step": 5730, + "tokens_trained": 18.771858208 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 1.6280405644989717, + "grad_norm": 0.8199700117111206, + "loss": 1.0483, + "loss_ce": 1.1203182935714722, + "loss_region": 0.0, + "loss_total": 1.1203182935714722, + "lr": 0.0008259468438538206, + "step": 5740, + "tokens_trained": 18.804619848 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 1.6308772427487412, + "grad_norm": 1.3682235479354858, + "loss": 1.0371, + "loss_ce": 1.1180541515350342, + "loss_region": 0.0, + "loss_total": 1.1180541515350342, + "lr": 0.0008256146179401993, + "step": 5750, + "tokens_trained": 18.837385288 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 1.6337139209985108, + "grad_norm": 1.0964722633361816, + "loss": 1.0415, + "loss_ce": 1.1061787605285645, + "loss_region": 0.0, + "loss_total": 1.1061787605285645, + "lr": 0.0008252823920265781, + "step": 5760, + "tokens_trained": 18.870150648 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 1.6365505992482803, + "grad_norm": 0.9015624523162842, + "loss": 1.0353, + "loss_ce": 1.0320029258728027, + "loss_region": 0.0, + "loss_total": 1.0320029258728027, + "lr": 0.0008249501661129568, + "step": 5770, + "tokens_trained": 18.902916088 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 1.6393872774980498, + "grad_norm": 1.1412580013275146, + "loss": 1.0424, + "loss_ce": 1.0464942455291748, + "loss_region": 0.0, + "loss_total": 1.0464942455291748, + "lr": 0.0008246179401993355, + "step": 5780, + "tokens_trained": 18.935681528 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 1.6422239557478193, + "grad_norm": 1.981430172920227, + "loss": 1.0324, + "loss_ce": 0.9644071459770203, + "loss_region": 0.0, + "loss_total": 0.9644071459770203, + "lr": 0.0008242857142857144, + "step": 5790, + "tokens_trained": 18.968446968 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 1.6450606339975888, + "grad_norm": 1.7167972326278687, + "loss": 1.0357, + "loss_ce": 0.986993134021759, + "loss_region": 0.0, + "loss_total": 0.986993134021759, + "lr": 0.0008239534883720931, + "step": 5800, + "tokens_trained": 19.001210808 + }, + { + "epoch": 1.6450606339975888, + "eval_ppl": 2.8592419592570706, + "eval_runtime": 1.4223, + "step": 5800, + "tokens_trained": 19.001210808 + }, + { + "epoch": 1.6450606339975888, + "eval_ppl_cds": 3.479217190754078, + "eval_ppl_dig": 1.1014041936231727, + "eval_ppl_exon": 3.3507041928348507, + "eval_ppl_intron": 2.966781302600734, + "eval_ppl_nig": 2.6445048013225807, + "eval_ppl_promoter": 3.187040976775228, + "eval_ppl_utr": 3.2733342762282778, + "step": 5800, + "tokens_trained": 19.001210808 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 1.6478973122473584, + "grad_norm": 1.0692654848098755, + "loss": 1.0349, + "loss_ce": 1.0418235063552856, + "loss_region": 0.0, + "loss_total": 1.0418235063552856, + "lr": 0.0008236212624584718, + "step": 5810, + "tokens_trained": 19.033976248 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 1.6507339904971279, + "grad_norm": 2.4034342765808105, + "loss": 1.0389, + "loss_ce": 1.091537594795227, + "loss_region": 0.0, + "loss_total": 1.091537594795227, + "lr": 0.0008232890365448505, + "step": 5820, + "tokens_trained": 19.066741688 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 1.6535706687468974, + "grad_norm": 0.5223531723022461, + "loss": 1.0359, + "loss_ce": 1.078364610671997, + "loss_region": 0.0, + "loss_total": 1.078364610671997, + "lr": 0.0008229568106312293, + "step": 5830, + "tokens_trained": 19.099507128 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 1.656407346996667, + "grad_norm": 1.5824055671691895, + "loss": 1.0348, + "loss_ce": 1.0647660493850708, + "loss_region": 0.0, + "loss_total": 1.0647660493850708, + "lr": 0.0008226245847176081, + "step": 5840, + "tokens_trained": 19.132272568 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 1.6592440252464364, + "grad_norm": 1.6083046197891235, + "loss": 1.0392, + "loss_ce": 1.0626896619796753, + "loss_region": 0.0, + "loss_total": 1.0626896619796753, + "lr": 0.0008222923588039867, + "step": 5850, + "tokens_trained": 19.165038008 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 1.662080703496206, + "grad_norm": 0.8607194423675537, + "loss": 1.0344, + "loss_ce": 0.9903082847595215, + "loss_region": 0.0, + "loss_total": 0.9903082847595215, + "lr": 0.0008219601328903654, + "step": 5860, + "tokens_trained": 19.197803448 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 1.6649173817459755, + "grad_norm": 1.3313143253326416, + "loss": 1.0367, + "loss_ce": 1.115062952041626, + "loss_region": 0.0, + "loss_total": 1.115062952041626, + "lr": 0.0008216279069767442, + "step": 5870, + "tokens_trained": 19.230568888 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 1.667754059995745, + "grad_norm": 1.201103925704956, + "loss": 1.0352, + "loss_ce": 1.1068215370178223, + "loss_region": 0.0, + "loss_total": 1.1068215370178223, + "lr": 0.0008212956810631229, + "step": 5880, + "tokens_trained": 19.263333528 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 1.6705907382455145, + "grad_norm": 1.2985320091247559, + "loss": 1.0294, + "loss_ce": 1.0131157636642456, + "loss_region": 0.0, + "loss_total": 1.0131157636642456, + "lr": 0.0008209634551495017, + "step": 5890, + "tokens_trained": 19.296098168 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 1.673427416495284, + "grad_norm": 1.7572495937347412, + "loss": 1.0371, + "loss_ce": 0.9603492021560669, + "loss_region": 0.0, + "loss_total": 0.9603492021560669, + "lr": 0.0008206312292358804, + "step": 5900, + "tokens_trained": 19.328862688 + }, + { + "epoch": 1.673427416495284, + "eval_ppl": 2.8698756972354995, + "eval_runtime": 1.4203, + "step": 5900, + "tokens_trained": 19.328862688 + }, + { + "epoch": 1.673427416495284, + "eval_ppl_cds": 3.5329996555915133, + "eval_ppl_dig": 1.0998061187904777, + "eval_ppl_exon": 3.357298517109707, + "eval_ppl_intron": 2.9849768046068696, + "eval_ppl_nig": 2.6552757011991313, + "eval_ppl_promoter": 3.1839220351321247, + "eval_ppl_utr": 3.2720330390370593, + "step": 5900, + "tokens_trained": 19.328862688 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 1.6762640947450536, + "grad_norm": 1.4859322309494019, + "loss": 1.0431, + "loss_ce": 1.1363308429718018, + "loss_region": 0.0, + "loss_total": 1.1363308429718018, + "lr": 0.0008202990033222592, + "step": 5910, + "tokens_trained": 19.361627328 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 1.679100772994823, + "grad_norm": 1.13947331905365, + "loss": 1.0301, + "loss_ce": 0.8870135545730591, + "loss_region": 0.0, + "loss_total": 0.8870135545730591, + "lr": 0.0008199667774086379, + "step": 5920, + "tokens_trained": 19.394392768 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 1.6819374512445926, + "grad_norm": 1.6557886600494385, + "loss": 1.0382, + "loss_ce": 1.1340681314468384, + "loss_region": 0.0, + "loss_total": 1.1340681314468384, + "lr": 0.0008196345514950166, + "step": 5930, + "tokens_trained": 19.427158208 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 1.6847741294943621, + "grad_norm": 1.3010876178741455, + "loss": 1.0304, + "loss_ce": 1.028099536895752, + "loss_region": 0.0, + "loss_total": 1.028099536895752, + "lr": 0.0008193023255813954, + "step": 5940, + "tokens_trained": 19.459922848 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 1.6876108077441316, + "grad_norm": 1.3161413669586182, + "loss": 1.0357, + "loss_ce": 1.0652676820755005, + "loss_region": 0.0, + "loss_total": 1.0652676820755005, + "lr": 0.0008189700996677742, + "step": 5950, + "tokens_trained": 19.492688288 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 1.6904474859939012, + "grad_norm": 0.8242393732070923, + "loss": 1.0366, + "loss_ce": 1.0475338697433472, + "loss_region": 0.0, + "loss_total": 1.0475338697433472, + "lr": 0.0008186378737541529, + "step": 5960, + "tokens_trained": 19.525453728 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 1.6932841642436707, + "grad_norm": 1.064610481262207, + "loss": 1.0375, + "loss_ce": 0.9698154330253601, + "loss_region": 0.0, + "loss_total": 0.9698154330253601, + "lr": 0.0008183056478405316, + "step": 5970, + "tokens_trained": 19.558219168 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 1.6961208424934402, + "grad_norm": 2.1938300132751465, + "loss": 1.0269, + "loss_ce": 1.0361167192459106, + "loss_region": 0.0, + "loss_total": 1.0361167192459106, + "lr": 0.0008179734219269102, + "step": 5980, + "tokens_trained": 19.590983808 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 1.6989575207432097, + "grad_norm": 1.1251577138900757, + "loss": 1.0298, + "loss_ce": 1.0740998983383179, + "loss_region": 0.0, + "loss_total": 1.0740998983383179, + "lr": 0.000817641196013289, + "step": 5990, + "tokens_trained": 19.623749248 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 1.7017941989929792, + "grad_norm": 1.1355079412460327, + "loss": 1.0335, + "loss_ce": 1.0728849172592163, + "loss_region": 0.0, + "loss_total": 1.0728849172592163, + "lr": 0.0008173089700996678, + "step": 6000, + "tokens_trained": 19.65651468 + }, + { + "epoch": 1.7017941989929792, + "eval_ppl": 2.83538958101067, + "eval_runtime": 1.42, + "step": 6000, + "tokens_trained": 19.65651468 + }, + { + "epoch": 1.7017941989929792, + "eval_ppl_cds": 3.474531948270276, + "eval_ppl_dig": 1.0998519478813389, + "eval_ppl_exon": 3.3048611087901123, + "eval_ppl_intron": 2.9473441939130174, + "eval_ppl_nig": 2.6321982228164473, + "eval_ppl_promoter": 3.125453003069021, + "eval_ppl_utr": 3.2193681902624776, + "step": 6000, + "tokens_trained": 19.65651468 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 1.7046308772427488, + "grad_norm": 1.2556589841842651, + "loss": 1.0313, + "loss_ce": 1.116411805152893, + "loss_region": 0.0, + "loss_total": 1.116411805152893, + "lr": 0.0008169767441860465, + "step": 6010, + "tokens_trained": 19.68928012 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 1.7074675554925183, + "grad_norm": 1.389358639717102, + "loss": 1.0325, + "loss_ce": 1.0540366172790527, + "loss_region": 0.0, + "loss_total": 1.0540366172790527, + "lr": 0.0008166445182724252, + "step": 6020, + "tokens_trained": 19.722045536 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 1.7103042337422878, + "grad_norm": 1.335344672203064, + "loss": 1.0315, + "loss_ce": 1.0560168027877808, + "loss_region": 0.0, + "loss_total": 1.0560168027877808, + "lr": 0.000816312292358804, + "step": 6030, + "tokens_trained": 19.754810976 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 1.7131409119920573, + "grad_norm": 1.3902976512908936, + "loss": 1.0319, + "loss_ce": 1.1142243146896362, + "loss_region": 0.0, + "loss_total": 1.1142243146896362, + "lr": 0.0008159800664451827, + "step": 6040, + "tokens_trained": 19.787576416 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 1.7159775902418268, + "grad_norm": 0.9200903177261353, + "loss": 1.0276, + "loss_ce": 1.0004727840423584, + "loss_region": 0.0, + "loss_total": 1.0004727840423584, + "lr": 0.0008156478405315615, + "step": 6050, + "tokens_trained": 19.820338544 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 1.7188142684915964, + "grad_norm": 0.44678831100463867, + "loss": 1.0323, + "loss_ce": 1.11885666847229, + "loss_region": 0.0, + "loss_total": 1.11885666847229, + "lr": 0.0008153156146179403, + "step": 6060, + "tokens_trained": 19.853099816 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 1.7216509467413659, + "grad_norm": 1.2600339651107788, + "loss": 1.0322, + "loss_ce": 1.0328036546707153, + "loss_region": 0.0, + "loss_total": 1.0328036546707153, + "lr": 0.000814983388704319, + "step": 6070, + "tokens_trained": 19.885862632 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 1.7244876249911354, + "grad_norm": 1.3641071319580078, + "loss": 1.0376, + "loss_ce": 0.9443773627281189, + "loss_region": 0.0, + "loss_total": 0.9443773627281189, + "lr": 0.0008146511627906977, + "step": 6080, + "tokens_trained": 19.918628072 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 1.727324303240905, + "grad_norm": 0.8142416477203369, + "loss": 1.0323, + "loss_ce": 1.058523416519165, + "loss_region": 0.0, + "loss_total": 1.058523416519165, + "lr": 0.0008143189368770764, + "step": 6090, + "tokens_trained": 19.951393512 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 1.7301609814906744, + "grad_norm": 3.152662754058838, + "loss": 1.0272, + "loss_ce": 1.0129187107086182, + "loss_region": 0.0, + "loss_total": 1.0129187107086182, + "lr": 0.0008139867109634552, + "step": 6100, + "tokens_trained": 19.984154824 + }, + { + "epoch": 1.7301609814906744, + "eval_ppl": 2.823382107912903, + "eval_runtime": 1.4143, + "step": 6100, + "tokens_trained": 19.984154824 + }, + { + "epoch": 1.7301609814906744, + "eval_ppl_cds": 3.415610486374371, + "eval_ppl_dig": 1.0956763527953945, + "eval_ppl_exon": 3.3072060460219466, + "eval_ppl_intron": 2.9344026310298235, + "eval_ppl_nig": 2.608572195210171, + "eval_ppl_promoter": 3.0724819792052984, + "eval_ppl_utr": 3.183914864291981, + "step": 6100, + "tokens_trained": 19.984154824 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 1.732997659740444, + "grad_norm": 1.0291093587875366, + "loss": 1.0369, + "loss_ce": 1.040030598640442, + "loss_region": 0.0, + "loss_total": 1.040030598640442, + "lr": 0.0008136544850498339, + "step": 6110, + "tokens_trained": 20.016919464 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 1.7358343379902135, + "grad_norm": 1.5104340314865112, + "loss": 1.0247, + "loss_ce": 1.0479451417922974, + "loss_region": 0.0, + "loss_total": 1.0479451417922974, + "lr": 0.0008133222591362126, + "step": 6120, + "tokens_trained": 20.049684904 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 1.738671016239983, + "grad_norm": 0.6710711717605591, + "loss": 1.0312, + "loss_ce": 1.0656031370162964, + "loss_region": 0.0, + "loss_total": 1.0656031370162964, + "lr": 0.0008129900332225913, + "step": 6130, + "tokens_trained": 20.082450344 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 1.7415076944897525, + "grad_norm": 1.3666547536849976, + "loss": 1.0287, + "loss_ce": 1.0910308361053467, + "loss_region": 0.0, + "loss_total": 1.0910308361053467, + "lr": 0.00081265780730897, + "step": 6140, + "tokens_trained": 20.115215784 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 1.744344372739522, + "grad_norm": 1.0505642890930176, + "loss": 1.0345, + "loss_ce": 0.9948603510856628, + "loss_region": 0.0, + "loss_total": 0.9948603510856628, + "lr": 0.0008123255813953489, + "step": 6150, + "tokens_trained": 20.147981224 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 1.7471810509892916, + "grad_norm": 1.0794563293457031, + "loss": 1.0283, + "loss_ce": 0.9181787371635437, + "loss_region": 0.0, + "loss_total": 0.9181787371635437, + "lr": 0.0008119933554817276, + "step": 6160, + "tokens_trained": 20.180746664 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 1.750017729239061, + "grad_norm": 0.5335147380828857, + "loss": 1.0299, + "loss_ce": 0.9253838062286377, + "loss_region": 0.0, + "loss_total": 0.9253838062286377, + "lr": 0.0008116611295681063, + "step": 6170, + "tokens_trained": 20.213511184 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 1.7528544074888306, + "grad_norm": 1.860371470451355, + "loss": 1.0304, + "loss_ce": 1.010979413986206, + "loss_region": 0.0, + "loss_total": 1.010979413986206, + "lr": 0.0008113289036544851, + "step": 6180, + "tokens_trained": 20.246275824 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 1.7556910857386, + "grad_norm": 0.8605881929397583, + "loss": 1.0332, + "loss_ce": 1.040065884590149, + "loss_region": 0.0, + "loss_total": 1.040065884590149, + "lr": 0.0008109966777408638, + "step": 6190, + "tokens_trained": 20.279039664 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 1.7585277639883696, + "grad_norm": 0.5528202652931213, + "loss": 1.029, + "loss_ce": 0.8064289689064026, + "loss_region": 0.0, + "loss_total": 0.8064289689064026, + "lr": 0.0008106644518272426, + "step": 6200, + "tokens_trained": 20.311804304 + }, + { + "epoch": 1.7585277639883696, + "eval_ppl": 2.8173534562647404, + "eval_runtime": 1.4165, + "step": 6200, + "tokens_trained": 20.311804304 + }, + { + "epoch": 1.7585277639883696, + "eval_ppl_cds": 3.423139729033253, + "eval_ppl_dig": 1.0994585134593906, + "eval_ppl_exon": 3.3073815551375487, + "eval_ppl_intron": 2.9251090478964925, + "eval_ppl_nig": 2.6111381221444323, + "eval_ppl_promoter": 3.0705177396427787, + "eval_ppl_utr": 3.173056500438175, + "step": 6200, + "tokens_trained": 20.311804304 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 1.7613644422381391, + "grad_norm": 0.9401982426643372, + "loss": 1.0341, + "loss_ce": 1.0010727643966675, + "loss_region": 0.0, + "loss_total": 1.0010727643966675, + "lr": 0.0008103322259136213, + "step": 6210, + "tokens_trained": 20.344569744 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 1.7642011204879087, + "grad_norm": 1.6745414733886719, + "loss": 1.0255, + "loss_ce": 1.0673383474349976, + "loss_region": 0.0, + "loss_total": 1.0673383474349976, + "lr": 0.0008100000000000001, + "step": 6220, + "tokens_trained": 20.377335184 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 1.7670377987376782, + "grad_norm": 0.7196193337440491, + "loss": 1.0239, + "loss_ce": 1.0102505683898926, + "loss_region": 0.0, + "loss_total": 1.0102505683898926, + "lr": 0.0008096677740863787, + "step": 6230, + "tokens_trained": 20.410099824 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 1.7698744769874477, + "grad_norm": 0.970348060131073, + "loss": 1.0224, + "loss_ce": 0.9167762994766235, + "loss_region": 0.0, + "loss_total": 0.9167762994766235, + "lr": 0.0008093355481727574, + "step": 6240, + "tokens_trained": 20.442865264 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 1.7727111552372172, + "grad_norm": 0.8411285281181335, + "loss": 1.0209, + "loss_ce": 1.0111656188964844, + "loss_region": 0.0, + "loss_total": 1.0111656188964844, + "lr": 0.0008090033222591362, + "step": 6250, + "tokens_trained": 20.475630704 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 1.7755478334869867, + "grad_norm": 1.1691157817840576, + "loss": 1.0318, + "loss_ce": 0.9169570207595825, + "loss_region": 0.0, + "loss_total": 0.9169570207595825, + "lr": 0.000808671096345515, + "step": 6260, + "tokens_trained": 20.508396144 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 1.7783845117367563, + "grad_norm": 0.419515460729599, + "loss": 1.02, + "loss_ce": 0.9586695432662964, + "loss_region": 0.0, + "loss_total": 0.9586695432662964, + "lr": 0.0008083388704318937, + "step": 6270, + "tokens_trained": 20.541161584 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 1.7812211899865258, + "grad_norm": 0.6500226855278015, + "loss": 1.0314, + "loss_ce": 1.0502530336380005, + "loss_region": 0.0, + "loss_total": 1.0502530336380005, + "lr": 0.0008080066445182724, + "step": 6280, + "tokens_trained": 20.573927024 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 1.7840578682362953, + "grad_norm": 0.8042202591896057, + "loss": 1.0284, + "loss_ce": 1.0619235038757324, + "loss_region": 0.0, + "loss_total": 1.0619235038757324, + "lr": 0.0008076744186046511, + "step": 6290, + "tokens_trained": 20.606692464 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 1.7868945464860648, + "grad_norm": 0.9335095286369324, + "loss": 1.0309, + "loss_ce": 0.9953011870384216, + "loss_region": 0.0, + "loss_total": 0.9953011870384216, + "lr": 0.00080734219269103, + "step": 6300, + "tokens_trained": 20.639457104 + }, + { + "epoch": 1.7868945464860648, + "eval_ppl": 2.8104714802059565, + "eval_runtime": 1.4222, + "step": 6300, + "tokens_trained": 20.639457104 + }, + { + "epoch": 1.7868945464860648, + "eval_ppl_cds": 3.4044066818038208, + "eval_ppl_dig": 1.0933971780227265, + "eval_ppl_exon": 3.3155706513718983, + "eval_ppl_intron": 2.916534799297983, + "eval_ppl_nig": 2.6101643388926257, + "eval_ppl_promoter": 3.039577268401293, + "eval_ppl_utr": 3.1552800422924756, + "step": 6300, + "tokens_trained": 20.639457104 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 1.7897312247358343, + "grad_norm": 0.9687148332595825, + "loss": 1.0282, + "loss_ce": 1.0328952074050903, + "loss_region": 0.0, + "loss_total": 1.0328952074050903, + "lr": 0.0008070099667774087, + "step": 6310, + "tokens_trained": 20.672222544 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 1.7925679029856039, + "grad_norm": 2.1120142936706543, + "loss": 1.031, + "loss_ce": 1.028664469718933, + "loss_region": 0.0, + "loss_total": 1.028664469718933, + "lr": 0.0008066777408637874, + "step": 6320, + "tokens_trained": 20.704987984 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 1.7954045812353734, + "grad_norm": 0.962121307849884, + "loss": 1.0224, + "loss_ce": 1.0200809240341187, + "loss_region": 0.0, + "loss_total": 1.0200809240341187, + "lr": 0.0008063455149501661, + "step": 6330, + "tokens_trained": 20.737753424 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 1.798241259485143, + "grad_norm": 2.0929181575775146, + "loss": 1.0248, + "loss_ce": 1.1376185417175293, + "loss_region": 0.0, + "loss_total": 1.1376185417175293, + "lr": 0.0008060132890365449, + "step": 6340, + "tokens_trained": 20.770518864 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 1.8010779377349124, + "grad_norm": 0.45388558506965637, + "loss": 1.0295, + "loss_ce": 0.9641968607902527, + "loss_region": 0.0, + "loss_total": 0.9641968607902527, + "lr": 0.0008056810631229236, + "step": 6350, + "tokens_trained": 20.803284304 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 1.803914615984682, + "grad_norm": 1.2569286823272705, + "loss": 1.0231, + "loss_ce": 0.9915566444396973, + "loss_region": 0.0, + "loss_total": 0.9915566444396973, + "lr": 0.0008053488372093023, + "step": 6360, + "tokens_trained": 20.836048936 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 1.8067512942344515, + "grad_norm": 0.7926527261734009, + "loss": 1.0297, + "loss_ce": 1.0349805355072021, + "loss_region": 0.0, + "loss_total": 1.0349805355072021, + "lr": 0.000805016611295681, + "step": 6370, + "tokens_trained": 20.868814376 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 1.809587972484221, + "grad_norm": 0.6937260627746582, + "loss": 1.0308, + "loss_ce": 1.1288645267486572, + "loss_region": 0.0, + "loss_total": 1.1288645267486572, + "lr": 0.0008046843853820598, + "step": 6380, + "tokens_trained": 20.901579816 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 1.8124246507339905, + "grad_norm": 1.4966633319854736, + "loss": 1.0284, + "loss_ce": 1.040663242340088, + "loss_region": 0.0, + "loss_total": 1.040663242340088, + "lr": 0.0008043521594684385, + "step": 6390, + "tokens_trained": 20.934345256 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 1.81526132898376, + "grad_norm": 1.3519423007965088, + "loss": 1.0347, + "loss_ce": 0.9976351261138916, + "loss_region": 0.0, + "loss_total": 0.9976351261138916, + "lr": 0.0008040199335548173, + "step": 6400, + "tokens_trained": 20.967109088 + }, + { + "epoch": 1.81526132898376, + "eval_ppl": 2.8067139043264295, + "eval_runtime": 1.4492, + "step": 6400, + "tokens_trained": 20.967109088 + }, + { + "epoch": 1.81526132898376, + "eval_ppl_cds": 3.3503957797061243, + "eval_ppl_dig": 1.099062735467728, + "eval_ppl_exon": 3.302530769374913, + "eval_ppl_intron": 2.9166879901734863, + "eval_ppl_nig": 2.604666899404525, + "eval_ppl_promoter": 3.020606344999106, + "eval_ppl_utr": 3.139291609298409, + "step": 6400, + "tokens_trained": 20.967109088 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 1.8180980072335295, + "grad_norm": 1.0360265970230103, + "loss": 1.0207, + "loss_ce": 0.981937050819397, + "loss_region": 0.0, + "loss_total": 0.981937050819397, + "lr": 0.000803687707641196, + "step": 6410, + "tokens_trained": 20.999874528 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 1.820934685483299, + "grad_norm": 1.1245217323303223, + "loss": 1.023, + "loss_ce": 1.0317081212997437, + "loss_region": 0.0, + "loss_total": 1.0317081212997437, + "lr": 0.0008033554817275748, + "step": 6420, + "tokens_trained": 21.032639968 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 1.8237713637330686, + "grad_norm": 1.1308019161224365, + "loss": 1.0296, + "loss_ce": 0.9880747199058533, + "loss_region": 0.0, + "loss_total": 0.9880747199058533, + "lr": 0.0008030232558139535, + "step": 6430, + "tokens_trained": 21.065405408 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 1.826608041982838, + "grad_norm": 0.6195257306098938, + "loss": 1.0267, + "loss_ce": 1.0812627077102661, + "loss_region": 0.0, + "loss_total": 1.0812627077102661, + "lr": 0.0008026910299003322, + "step": 6440, + "tokens_trained": 21.098170848 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 1.8294447202326076, + "grad_norm": 1.0061508417129517, + "loss": 1.0171, + "loss_ce": 1.1000691652297974, + "loss_region": 0.0, + "loss_total": 1.1000691652297974, + "lr": 0.0008023588039867111, + "step": 6450, + "tokens_trained": 21.13093628 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 1.8322813984823771, + "grad_norm": 1.0187026262283325, + "loss": 1.0269, + "loss_ce": 1.019209384918213, + "loss_region": 0.0, + "loss_total": 1.019209384918213, + "lr": 0.0008020265780730898, + "step": 6460, + "tokens_trained": 21.163697856 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 1.8351180767321467, + "grad_norm": 0.9605537056922913, + "loss": 1.0204, + "loss_ce": 1.0015205144882202, + "loss_region": 0.0, + "loss_total": 1.0015205144882202, + "lr": 0.0008016943521594685, + "step": 6470, + "tokens_trained": 21.196463296 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 1.8379547549819162, + "grad_norm": 0.7531270980834961, + "loss": 1.0286, + "loss_ce": 1.0767279863357544, + "loss_region": 0.0, + "loss_total": 1.0767279863357544, + "lr": 0.0008013621262458471, + "step": 6480, + "tokens_trained": 21.229228736 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 1.8407914332316857, + "grad_norm": 1.0039128065109253, + "loss": 1.0191, + "loss_ce": 0.9908838868141174, + "loss_region": 0.0, + "loss_total": 0.9908838868141174, + "lr": 0.0008010299003322259, + "step": 6490, + "tokens_trained": 21.261994176 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 1.8436281114814552, + "grad_norm": 0.6186189651489258, + "loss": 1.0242, + "loss_ce": 1.1024316549301147, + "loss_region": 0.0, + "loss_total": 1.1024316549301147, + "lr": 0.0008006976744186047, + "step": 6500, + "tokens_trained": 21.294759616 + }, + { + "epoch": 1.8436281114814552, + "eval_ppl": 2.798049473519268, + "eval_runtime": 1.427, + "step": 6500, + "tokens_trained": 21.294759616 + }, + { + "epoch": 1.8436281114814552, + "eval_ppl_cds": 3.348097311722527, + "eval_ppl_dig": 1.0994443888102112, + "eval_ppl_exon": 3.297215419171105, + "eval_ppl_intron": 2.9146186047915252, + "eval_ppl_nig": 2.6052712752326994, + "eval_ppl_promoter": 3.018410491225804, + "eval_ppl_utr": 3.1394636643039115, + "step": 6500, + "tokens_trained": 21.294759616 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 1.8464647897312247, + "grad_norm": 1.8114935159683228, + "loss": 1.0185, + "loss_ce": 0.9125613570213318, + "loss_region": 0.0, + "loss_total": 0.9125613570213318, + "lr": 0.0008003654485049834, + "step": 6510, + "tokens_trained": 21.327524256 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 1.8493014679809943, + "grad_norm": 0.4037961959838867, + "loss": 1.0235, + "loss_ce": 1.0597658157348633, + "loss_region": 0.0, + "loss_total": 1.0597658157348633, + "lr": 0.0008000332225913621, + "step": 6520, + "tokens_trained": 21.360289696 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 1.8521381462307638, + "grad_norm": 0.7221969962120056, + "loss": 1.0187, + "loss_ce": 1.0339834690093994, + "loss_region": 0.0, + "loss_total": 1.0339834690093994, + "lr": 0.0007997009966777409, + "step": 6530, + "tokens_trained": 21.393055136 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 1.8549748244805333, + "grad_norm": 1.017179250717163, + "loss": 1.0205, + "loss_ce": 1.1257262229919434, + "loss_region": 0.0, + "loss_total": 1.1257262229919434, + "lr": 0.0007993687707641196, + "step": 6540, + "tokens_trained": 21.425820576 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 1.8578115027303028, + "grad_norm": 0.7812231779098511, + "loss": 1.0172, + "loss_ce": 1.032475233078003, + "loss_region": 0.0, + "loss_total": 1.032475233078003, + "lr": 0.0007990365448504984, + "step": 6550, + "tokens_trained": 21.458586016 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 1.8606481809800723, + "grad_norm": 1.1513891220092773, + "loss": 1.0271, + "loss_ce": 0.9168854355812073, + "loss_region": 0.0, + "loss_total": 0.9168854355812073, + "lr": 0.0007987043189368771, + "step": 6560, + "tokens_trained": 21.491351456 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 1.8634848592298419, + "grad_norm": 0.5324859619140625, + "loss": 1.0187, + "loss_ce": 1.0277824401855469, + "loss_region": 0.0, + "loss_total": 1.0277824401855469, + "lr": 0.0007983720930232559, + "step": 6570, + "tokens_trained": 21.524116896 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 1.8663215374796114, + "grad_norm": 0.8778380155563354, + "loss": 1.0198, + "loss_ce": 0.9187647104263306, + "loss_region": 0.0, + "loss_total": 0.9187647104263306, + "lr": 0.0007980398671096346, + "step": 6580, + "tokens_trained": 21.556882336 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 1.869158215729381, + "grad_norm": 0.9481834173202515, + "loss": 1.0238, + "loss_ce": 1.077518343925476, + "loss_region": 0.0, + "loss_total": 1.077518343925476, + "lr": 0.0007977076411960133, + "step": 6590, + "tokens_trained": 21.589647776 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 1.8719948939791504, + "grad_norm": 0.9477498531341553, + "loss": 1.0154, + "loss_ce": 0.9290522336959839, + "loss_region": 0.0, + "loss_total": 0.9290522336959839, + "lr": 0.000797375415282392, + "step": 6600, + "tokens_trained": 21.622413216 + }, + { + "epoch": 1.8719948939791504, + "eval_ppl": 2.794600656331191, + "eval_runtime": 1.4287, + "step": 6600, + "tokens_trained": 21.622413216 + }, + { + "epoch": 1.8719948939791504, + "eval_ppl_cds": 3.363298565370526, + "eval_ppl_dig": 1.100544609743424, + "eval_ppl_exon": 3.2929613313532737, + "eval_ppl_intron": 2.9212580449535603, + "eval_ppl_nig": 2.6044263717043816, + "eval_ppl_promoter": 3.046699277461652, + "eval_ppl_utr": 3.159782222829799, + "step": 6600, + "tokens_trained": 21.622413216 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 1.87483157222892, + "grad_norm": 1.1344876289367676, + "loss": 1.0218, + "loss_ce": 0.9838250279426575, + "loss_region": 0.0, + "loss_total": 0.9838250279426575, + "lr": 0.0007970431893687708, + "step": 6610, + "tokens_trained": 21.655178656 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 1.8776682504786895, + "grad_norm": 0.2379857748746872, + "loss": 1.0228, + "loss_ce": 0.9815813899040222, + "loss_region": 0.0, + "loss_total": 0.9815813899040222, + "lr": 0.0007967109634551495, + "step": 6620, + "tokens_trained": 21.687944096 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 1.880504928728459, + "grad_norm": 0.6573547124862671, + "loss": 1.0261, + "loss_ce": 1.0460282564163208, + "loss_region": 0.0, + "loss_total": 1.0460282564163208, + "lr": 0.0007963787375415282, + "step": 6630, + "tokens_trained": 21.720707792 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 1.8833416069782285, + "grad_norm": 1.0820962190628052, + "loss": 1.0238, + "loss_ce": 1.026914119720459, + "loss_region": 0.0, + "loss_total": 1.026914119720459, + "lr": 0.0007960465116279069, + "step": 6640, + "tokens_trained": 21.753473232 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 1.886178285227998, + "grad_norm": 0.9002099633216858, + "loss": 1.0173, + "loss_ce": 1.0607340335845947, + "loss_region": 0.0, + "loss_total": 1.0607340335845947, + "lr": 0.0007957142857142858, + "step": 6650, + "tokens_trained": 21.786238672 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 1.8890149634777675, + "grad_norm": 1.2160532474517822, + "loss": 1.0164, + "loss_ce": 1.0911822319030762, + "loss_region": 0.0, + "loss_total": 1.0911822319030762, + "lr": 0.0007953820598006645, + "step": 6660, + "tokens_trained": 21.819004112 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 1.891851641727537, + "grad_norm": 1.0074514150619507, + "loss": 1.0166, + "loss_ce": 0.9953575730323792, + "loss_region": 0.0, + "loss_total": 0.9953575730323792, + "lr": 0.0007950498338870432, + "step": 6670, + "tokens_trained": 21.851769552 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 1.8946883199773066, + "grad_norm": 0.6795614361763, + "loss": 1.0117, + "loss_ce": 1.1053005456924438, + "loss_region": 0.0, + "loss_total": 1.1053005456924438, + "lr": 0.000794717607973422, + "step": 6680, + "tokens_trained": 21.884534992 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 1.897524998227076, + "grad_norm": 1.3707964420318604, + "loss": 1.0155, + "loss_ce": 0.9704234004020691, + "loss_region": 0.0, + "loss_total": 0.9704234004020691, + "lr": 0.0007943853820598007, + "step": 6690, + "tokens_trained": 21.917300432 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 1.9003616764768456, + "grad_norm": 0.730884313583374, + "loss": 1.0104, + "loss_ce": 1.1746294498443604, + "loss_region": 0.0, + "loss_total": 1.1746294498443604, + "lr": 0.0007940531561461795, + "step": 6700, + "tokens_trained": 21.950065872 + }, + { + "epoch": 1.9003616764768456, + "eval_ppl": 2.78888080017487, + "eval_runtime": 1.4293, + "step": 6700, + "tokens_trained": 21.950065872 + }, + { + "epoch": 1.9003616764768456, + "eval_ppl_cds": 3.343644454609723, + "eval_ppl_dig": 1.1016898200170444, + "eval_ppl_exon": 3.2948630456331354, + "eval_ppl_intron": 2.919771453775612, + "eval_ppl_nig": 2.609909112443194, + "eval_ppl_promoter": 3.0294890842908133, + "eval_ppl_utr": 3.1435539472761547, + "step": 6700, + "tokens_trained": 21.950065872 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 1.9031983547266151, + "grad_norm": 1.1917855739593506, + "loss": 1.02, + "loss_ce": 1.0126667022705078, + "loss_region": 0.0, + "loss_total": 1.0126667022705078, + "lr": 0.0007937209302325582, + "step": 6710, + "tokens_trained": 21.982831312 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 1.9060350329763847, + "grad_norm": 0.6363831162452698, + "loss": 1.0226, + "loss_ce": 1.0006239414215088, + "loss_region": 0.0, + "loss_total": 1.0006239414215088, + "lr": 0.000793388704318937, + "step": 6720, + "tokens_trained": 22.015596752 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 1.9088717112261542, + "grad_norm": 1.1274362802505493, + "loss": 1.0106, + "loss_ce": 1.0669947862625122, + "loss_region": 0.0, + "loss_total": 1.0669947862625122, + "lr": 0.0007930564784053156, + "step": 6730, + "tokens_trained": 22.048362192 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 1.9117083894759237, + "grad_norm": 0.7051267623901367, + "loss": 1.0072, + "loss_ce": 1.149444818496704, + "loss_region": 0.0, + "loss_total": 1.149444818496704, + "lr": 0.0007927242524916943, + "step": 6740, + "tokens_trained": 22.081127632 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 1.9145450677256932, + "grad_norm": 0.9449176788330078, + "loss": 1.0155, + "loss_ce": 1.1274597644805908, + "loss_region": 0.0, + "loss_total": 1.1274597644805908, + "lr": 0.0007923920265780731, + "step": 6750, + "tokens_trained": 22.113893072 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 1.9173817459754627, + "grad_norm": 1.2375448942184448, + "loss": 1.0203, + "loss_ce": 0.9039602875709534, + "loss_region": 0.0, + "loss_total": 0.9039602875709534, + "lr": 0.0007920598006644519, + "step": 6760, + "tokens_trained": 22.146655448 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 1.9202184242252323, + "grad_norm": 0.8512572646141052, + "loss": 1.0172, + "loss_ce": 0.9630705118179321, + "loss_region": 0.0, + "loss_total": 0.9630705118179321, + "lr": 0.0007917275747508306, + "step": 6770, + "tokens_trained": 22.179420888 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 1.9230551024750018, + "grad_norm": 0.5015329718589783, + "loss": 1.0264, + "loss_ce": 1.0668619871139526, + "loss_region": 0.0, + "loss_total": 1.0668619871139526, + "lr": 0.0007913953488372093, + "step": 6780, + "tokens_trained": 22.212186328 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 1.9258917807247713, + "grad_norm": 0.7549113035202026, + "loss": 1.015, + "loss_ce": 1.105078101158142, + "loss_region": 0.0, + "loss_total": 1.105078101158142, + "lr": 0.000791063122923588, + "step": 6790, + "tokens_trained": 22.244951768 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 1.9287284589745408, + "grad_norm": 1.2461268901824951, + "loss": 1.0171, + "loss_ce": 1.018865942955017, + "loss_region": 0.0, + "loss_total": 1.018865942955017, + "lr": 0.0007907308970099669, + "step": 6800, + "tokens_trained": 22.277717208 + }, + { + "epoch": 1.9287284589745408, + "eval_ppl": 2.7845870569339306, + "eval_runtime": 1.4235, + "step": 6800, + "tokens_trained": 22.277717208 + }, + { + "epoch": 1.9287284589745408, + "eval_ppl_cds": 3.334744579211234, + "eval_ppl_dig": 1.1014356554381384, + "eval_ppl_exon": 3.283992864929886, + "eval_ppl_intron": 2.907521086042318, + "eval_ppl_nig": 2.5986421802567112, + "eval_ppl_promoter": 2.965322207832415, + "eval_ppl_utr": 3.1048967104696565, + "step": 6800, + "tokens_trained": 22.277717208 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 1.9315651372243103, + "grad_norm": 0.8074588775634766, + "loss": 1.0133, + "loss_ce": 1.0383925437927246, + "loss_region": 0.0, + "loss_total": 1.0383925437927246, + "lr": 0.0007903986710963456, + "step": 6810, + "tokens_trained": 22.310482648 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 1.9344018154740799, + "grad_norm": 0.5049921870231628, + "loss": 1.016, + "loss_ce": 1.1488726139068604, + "loss_region": 0.0, + "loss_total": 1.1488726139068604, + "lr": 0.0007900664451827243, + "step": 6820, + "tokens_trained": 22.343246488 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 1.9372384937238494, + "grad_norm": 1.78823721408844, + "loss": 1.0201, + "loss_ce": 1.0954252481460571, + "loss_region": 0.0, + "loss_total": 1.0954252481460571, + "lr": 0.000789734219269103, + "step": 6830, + "tokens_trained": 22.376011928 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 1.940075171973619, + "grad_norm": 0.5541342496871948, + "loss": 1.0199, + "loss_ce": 0.8927311897277832, + "loss_region": 0.0, + "loss_total": 0.8927311897277832, + "lr": 0.0007894019933554818, + "step": 6840, + "tokens_trained": 22.408777368 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 1.9429118502233884, + "grad_norm": 0.6893653869628906, + "loss": 1.0207, + "loss_ce": 1.09942626953125, + "loss_region": 0.0, + "loss_total": 1.09942626953125, + "lr": 0.0007890697674418605, + "step": 6850, + "tokens_trained": 22.441542808 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 1.945748528473158, + "grad_norm": 0.4325312674045563, + "loss": 1.0208, + "loss_ce": 0.8823619484901428, + "loss_region": 0.0, + "loss_total": 0.8823619484901428, + "lr": 0.0007887375415282392, + "step": 6860, + "tokens_trained": 22.474308248 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 1.9485852067229275, + "grad_norm": 0.8349773287773132, + "loss": 1.0077, + "loss_ce": 0.8815352320671082, + "loss_region": 0.0, + "loss_total": 0.8815352320671082, + "lr": 0.0007884053156146179, + "step": 6870, + "tokens_trained": 22.507073688 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 1.951421884972697, + "grad_norm": 0.6413218975067139, + "loss": 1.0138, + "loss_ce": 1.0080714225769043, + "loss_region": 0.0, + "loss_total": 1.0080714225769043, + "lr": 0.0007880730897009967, + "step": 6880, + "tokens_trained": 22.539837536 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 1.9542585632224665, + "grad_norm": 1.159388542175293, + "loss": 1.0222, + "loss_ce": 0.9643105268478394, + "loss_region": 0.0, + "loss_total": 0.9643105268478394, + "lr": 0.0007877408637873754, + "step": 6890, + "tokens_trained": 22.572602976 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 1.957095241472236, + "grad_norm": 0.9914594888687134, + "loss": 1.0171, + "loss_ce": 1.1230136156082153, + "loss_region": 0.0, + "loss_total": 1.1230136156082153, + "lr": 0.0007874086378737542, + "step": 6900, + "tokens_trained": 22.605367616 + }, + { + "epoch": 1.957095241472236, + "eval_ppl": 2.7852048818910418, + "eval_runtime": 1.4171, + "step": 6900, + "tokens_trained": 22.605367616 + }, + { + "epoch": 1.957095241472236, + "eval_ppl_cds": 3.329297997238432, + "eval_ppl_dig": 1.1023876031036097, + "eval_ppl_exon": 3.2882691264401553, + "eval_ppl_intron": 2.912112658862319, + "eval_ppl_nig": 2.597372195936058, + "eval_ppl_promoter": 2.992720939207019, + "eval_ppl_utr": 3.1111169058663237, + "step": 6900, + "tokens_trained": 22.605367616 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 1.9599319197220055, + "grad_norm": 1.6943984031677246, + "loss": 1.0103, + "loss_ce": 0.9212698340415955, + "loss_region": 0.0, + "loss_total": 0.9212698340415955, + "lr": 0.0007870764119601329, + "step": 6910, + "tokens_trained": 22.638133056 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 1.962768597971775, + "grad_norm": 1.2909159660339355, + "loss": 1.0114, + "loss_ce": 0.9539883732795715, + "loss_region": 0.0, + "loss_total": 0.9539883732795715, + "lr": 0.0007867441860465117, + "step": 6920, + "tokens_trained": 22.670898496 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 1.9656052762215446, + "grad_norm": 1.5697418451309204, + "loss": 1.0195, + "loss_ce": 1.1330108642578125, + "loss_region": 0.0, + "loss_total": 1.1330108642578125, + "lr": 0.0007864119601328904, + "step": 6930, + "tokens_trained": 22.703663936 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 1.968441954471314, + "grad_norm": 0.723041832447052, + "loss": 1.0108, + "loss_ce": 1.1327755451202393, + "loss_region": 0.0, + "loss_total": 1.1327755451202393, + "lr": 0.0007860797342192691, + "step": 6940, + "tokens_trained": 22.736429376 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 1.9712786327210836, + "grad_norm": 0.5307745933532715, + "loss": 1.0161, + "loss_ce": 1.0426133871078491, + "loss_region": 0.0, + "loss_total": 1.0426133871078491, + "lr": 0.000785747508305648, + "step": 6950, + "tokens_trained": 22.769194816 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 1.9741153109708531, + "grad_norm": 1.788638949394226, + "loss": 1.0184, + "loss_ce": 1.1597195863723755, + "loss_region": 0.0, + "loss_total": 1.1597195863723755, + "lr": 0.0007854152823920267, + "step": 6960, + "tokens_trained": 22.801960256 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 1.9769519892206227, + "grad_norm": 0.9060627818107605, + "loss": 1.0242, + "loss_ce": 0.9481877684593201, + "loss_region": 0.0, + "loss_total": 0.9481877684593201, + "lr": 0.0007850830564784053, + "step": 6970, + "tokens_trained": 22.834725696 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 1.9797886674703922, + "grad_norm": 1.095072627067566, + "loss": 1.0103, + "loss_ce": 1.0279875993728638, + "loss_region": 0.0, + "loss_total": 1.0279875993728638, + "lr": 0.000784750830564784, + "step": 6980, + "tokens_trained": 22.867491136 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 1.9826253457201617, + "grad_norm": 0.9623782634735107, + "loss": 1.0222, + "loss_ce": 0.864204466342926, + "loss_region": 0.0, + "loss_total": 0.864204466342926, + "lr": 0.0007844186046511627, + "step": 6990, + "tokens_trained": 22.900255448 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 1.9854620239699312, + "grad_norm": 0.9268852472305298, + "loss": 1.0118, + "loss_ce": 1.0299792289733887, + "loss_region": 0.0, + "loss_total": 1.0299792289733887, + "lr": 0.0007840863787375415, + "step": 7000, + "tokens_trained": 22.933020888 + }, + { + "epoch": 1.9854620239699312, + "eval_ppl": 2.7822554305263134, + "eval_runtime": 1.4245, + "step": 7000, + "tokens_trained": 22.933020888 + }, + { + "epoch": 1.9854620239699312, + "eval_ppl_cds": 3.3372268532942706, + "eval_ppl_dig": 1.1033711013852714, + "eval_ppl_exon": 3.290807039041509, + "eval_ppl_intron": 2.910610312552892, + "eval_ppl_nig": 2.600159934109166, + "eval_ppl_promoter": 2.9769039409901463, + "eval_ppl_utr": 3.1021206527216183, + "step": 7000, + "tokens_trained": 22.933020888 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 1.9882987022197007, + "grad_norm": 0.8806335926055908, + "loss": 1.0166, + "loss_ce": 1.0804855823516846, + "loss_region": 0.0, + "loss_total": 1.0804855823516846, + "lr": 0.0007837541528239203, + "step": 7010, + "tokens_trained": 22.965785528 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 1.9911353804694703, + "grad_norm": 1.1552938222885132, + "loss": 1.0197, + "loss_ce": 1.122083067893982, + "loss_region": 0.0, + "loss_total": 1.122083067893982, + "lr": 0.000783421926910299, + "step": 7020, + "tokens_trained": 22.998550968 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 1.9939720587192398, + "grad_norm": 0.6757546663284302, + "loss": 1.0128, + "loss_ce": 0.9135093092918396, + "loss_region": 0.0, + "loss_total": 0.9135093092918396, + "lr": 0.0007830897009966777, + "step": 7030, + "tokens_trained": 23.031316408 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 1.9968087369690093, + "grad_norm": 0.9053455591201782, + "loss": 1.0176, + "loss_ce": 0.9564316868782043, + "loss_region": 0.0, + "loss_total": 0.9564316868782043, + "lr": 0.0007827574750830565, + "step": 7040, + "tokens_trained": 23.064081848 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 1.9996454152187788, + "grad_norm": 1.3979105949401855, + "loss": 1.016, + "loss_ce": 0.9557048082351685, + "loss_region": 0.0, + "loss_total": 0.9557048082351685, + "lr": 0.0007824252491694352, + "step": 7050, + "tokens_trained": 23.096847288 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 2.0022693425998157, + "grad_norm": 0.843721330165863, + "loss": 1.0053, + "loss_ce": 0.9853013157844543, + "loss_region": 0.0, + "loss_total": 0.9853013157844543, + "lr": 0.000782093023255814, + "step": 7060, + "tokens_trained": 23.12715532 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 2.0051060208495852, + "grad_norm": 1.0112333297729492, + "loss": 1.0081, + "loss_ce": 0.9298333525657654, + "loss_region": 0.0, + "loss_total": 0.9298333525657654, + "lr": 0.0007817607973421928, + "step": 7070, + "tokens_trained": 23.15992076 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 2.0079426990993547, + "grad_norm": 1.142694354057312, + "loss": 1.0013, + "loss_ce": 1.0749361515045166, + "loss_region": 0.0, + "loss_total": 1.0749361515045166, + "lr": 0.0007814285714285715, + "step": 7080, + "tokens_trained": 23.1926862 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 2.0107793773491243, + "grad_norm": 0.7951752543449402, + "loss": 1.0087, + "loss_ce": 1.0061616897583008, + "loss_region": 0.0, + "loss_total": 1.0061616897583008, + "lr": 0.0007810963455149502, + "step": 7090, + "tokens_trained": 23.22545164 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 2.013616055598894, + "grad_norm": 0.9214019179344177, + "loss": 1.0103, + "loss_ce": 0.9987831711769104, + "loss_region": 0.0, + "loss_total": 0.9987831711769104, + "lr": 0.0007807641196013288, + "step": 7100, + "tokens_trained": 23.25821708 + }, + { + "epoch": 2.013616055598894, + "eval_ppl": 2.781281484208184, + "eval_runtime": 1.4313, + "step": 7100, + "tokens_trained": 23.25821708 + }, + { + "epoch": 2.013616055598894, + "eval_ppl_cds": 3.3136363165535254, + "eval_ppl_dig": 1.102758683889497, + "eval_ppl_exon": 3.3053467440592477, + "eval_ppl_intron": 2.913823844451076, + "eval_ppl_nig": 2.605358525114352, + "eval_ppl_promoter": 2.981512661077136, + "eval_ppl_utr": 3.1005994332266966, + "step": 7100, + "tokens_trained": 23.25821708 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 2.0164527338486633, + "grad_norm": 0.6453201770782471, + "loss": 1.0092, + "loss_ce": 0.9892328977584839, + "loss_region": 0.0, + "loss_total": 0.9892328977584839, + "lr": 0.0007804318936877077, + "step": 7110, + "tokens_trained": 23.29098252 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 2.019289412098433, + "grad_norm": 0.6038001179695129, + "loss": 1.008, + "loss_ce": 1.095076560974121, + "loss_region": 0.0, + "loss_total": 1.095076560974121, + "lr": 0.0007800996677740864, + "step": 7120, + "tokens_trained": 23.32374796 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 2.0221260903482023, + "grad_norm": 0.853272557258606, + "loss": 0.9989, + "loss_ce": 0.9461421966552734, + "loss_region": 0.0, + "loss_total": 0.9461421966552734, + "lr": 0.0007797674418604651, + "step": 7130, + "tokens_trained": 23.3565134 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 2.024962768597972, + "grad_norm": 1.1229009628295898, + "loss": 1.0114, + "loss_ce": 1.0299495458602905, + "loss_region": 0.0, + "loss_total": 1.0299495458602905, + "lr": 0.0007794352159468438, + "step": 7140, + "tokens_trained": 23.38927804 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 2.0277994468477414, + "grad_norm": 0.5124538540840149, + "loss": 0.9999, + "loss_ce": 0.94310462474823, + "loss_region": 0.0, + "loss_total": 0.94310462474823, + "lr": 0.0007791029900332226, + "step": 7150, + "tokens_trained": 23.42204348 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 2.030636125097511, + "grad_norm": 1.0442956686019897, + "loss": 0.9979, + "loss_ce": 1.0645129680633545, + "loss_region": 0.0, + "loss_total": 1.0645129680633545, + "lr": 0.0007787707641196014, + "step": 7160, + "tokens_trained": 23.45480892 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 2.0334728033472804, + "grad_norm": 0.5161722302436829, + "loss": 1.0026, + "loss_ce": 1.08507239818573, + "loss_region": 0.0, + "loss_total": 1.08507239818573, + "lr": 0.0007784385382059801, + "step": 7170, + "tokens_trained": 23.48757436 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 2.03630948159705, + "grad_norm": 0.3302663564682007, + "loss": 1.0012, + "loss_ce": 0.9567556381225586, + "loss_region": 0.0, + "loss_total": 0.9567556381225586, + "lr": 0.0007781063122923588, + "step": 7180, + "tokens_trained": 23.520339 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 2.0391461598468195, + "grad_norm": 1.1320247650146484, + "loss": 1.0015, + "loss_ce": 0.9663833379745483, + "loss_region": 0.0, + "loss_total": 0.9663833379745483, + "lr": 0.0007777740863787376, + "step": 7190, + "tokens_trained": 23.55310444 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 2.041982838096589, + "grad_norm": 0.9850210547447205, + "loss": 1.0151, + "loss_ce": 0.9994710683822632, + "loss_region": 0.0, + "loss_total": 0.9994710683822632, + "lr": 0.0007774418604651163, + "step": 7200, + "tokens_trained": 23.58586988 + }, + { + "epoch": 2.041982838096589, + "eval_ppl": 2.771127968147156, + "eval_runtime": 1.4313, + "step": 7200, + "tokens_trained": 23.58586988 + }, + { + "epoch": 2.041982838096589, + "eval_ppl_cds": 3.3032367330490224, + "eval_ppl_dig": 1.1023737562409253, + "eval_ppl_exon": 3.293251983523332, + "eval_ppl_intron": 2.9037172520056234, + "eval_ppl_nig": 2.5905965739701347, + "eval_ppl_promoter": 2.951108313969505, + "eval_ppl_utr": 3.0482862177687773, + "step": 7200, + "tokens_trained": 23.58586988 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 2.0448195163463585, + "grad_norm": 0.7062892913818359, + "loss": 1.0101, + "loss_ce": 1.0791970491409302, + "loss_region": 0.0, + "loss_total": 1.0791970491409302, + "lr": 0.0007771096345514951, + "step": 7210, + "tokens_trained": 23.61863532 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 2.047656194596128, + "grad_norm": 2.32240891456604, + "loss": 1.0101, + "loss_ce": 1.036105990409851, + "loss_region": 0.0, + "loss_total": 1.036105990409851, + "lr": 0.0007767774086378737, + "step": 7220, + "tokens_trained": 23.65140076 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 2.0504928728458975, + "grad_norm": 1.2545887231826782, + "loss": 1.0052, + "loss_ce": 1.117865800857544, + "loss_region": 0.0, + "loss_total": 1.117865800857544, + "lr": 0.0007764451827242525, + "step": 7230, + "tokens_trained": 23.6841662 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 2.053329551095667, + "grad_norm": 0.4743971526622772, + "loss": 1.0081, + "loss_ce": 0.8768163323402405, + "loss_region": 0.0, + "loss_total": 0.8768163323402405, + "lr": 0.0007761129568106312, + "step": 7240, + "tokens_trained": 23.71693004 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 2.0561662293454366, + "grad_norm": 1.2599022388458252, + "loss": 1.004, + "loss_ce": 1.09614896774292, + "loss_region": 0.0, + "loss_total": 1.09614896774292, + "lr": 0.0007757807308970099, + "step": 7250, + "tokens_trained": 23.74969548 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 2.059002907595206, + "grad_norm": 1.189494013786316, + "loss": 1.0019, + "loss_ce": 1.0309582948684692, + "loss_region": 0.0, + "loss_total": 1.0309582948684692, + "lr": 0.0007754485049833887, + "step": 7260, + "tokens_trained": 23.78246092 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 2.0618395858449756, + "grad_norm": 0.7742461562156677, + "loss": 0.9994, + "loss_ce": 1.1182734966278076, + "loss_region": 0.0, + "loss_total": 1.1182734966278076, + "lr": 0.0007751162790697675, + "step": 7270, + "tokens_trained": 23.81522636 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 2.064676264094745, + "grad_norm": 0.38940683007240295, + "loss": 1.001, + "loss_ce": 1.0810575485229492, + "loss_region": 0.0, + "loss_total": 1.0810575485229492, + "lr": 0.0007747840531561462, + "step": 7280, + "tokens_trained": 23.8479918 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 2.0675129423445147, + "grad_norm": 0.5152825117111206, + "loss": 1.0077, + "loss_ce": 0.9295870065689087, + "loss_region": 0.0, + "loss_total": 0.9295870065689087, + "lr": 0.0007744518272425249, + "step": 7290, + "tokens_trained": 23.88075724 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 2.070349620594284, + "grad_norm": 1.2133007049560547, + "loss": 1.0144, + "loss_ce": 1.0206096172332764, + "loss_region": 0.0, + "loss_total": 1.0206096172332764, + "lr": 0.0007741196013289036, + "step": 7300, + "tokens_trained": 23.91352268 + }, + { + "epoch": 2.070349620594284, + "eval_ppl": 2.766244503462139, + "eval_runtime": 1.425, + "step": 7300, + "tokens_trained": 23.91352268 + }, + { + "epoch": 2.070349620594284, + "eval_ppl_cds": 3.319210840498938, + "eval_ppl_dig": 1.0973421147058586, + "eval_ppl_exon": 3.2813848193139825, + "eval_ppl_intron": 2.9028043191892015, + "eval_ppl_nig": 2.5900302515294475, + "eval_ppl_promoter": 2.946759083447451, + "eval_ppl_utr": 3.0577907778414195, + "step": 7300, + "tokens_trained": 23.91352268 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 2.0731862988440537, + "grad_norm": 0.8950213193893433, + "loss": 1.0064, + "loss_ce": 1.0516961812973022, + "loss_region": 0.0, + "loss_total": 1.0516961812973022, + "lr": 0.0007737873754152825, + "step": 7310, + "tokens_trained": 23.94628812 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 2.076022977093823, + "grad_norm": 0.7184622287750244, + "loss": 1.001, + "loss_ce": 1.1180815696716309, + "loss_region": 0.0, + "loss_total": 1.1180815696716309, + "lr": 0.0007734551495016612, + "step": 7320, + "tokens_trained": 23.97905276 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 2.0788596553435927, + "grad_norm": 0.2874407470226288, + "loss": 0.9983, + "loss_ce": 0.8592913746833801, + "loss_region": 0.0, + "loss_total": 0.8592913746833801, + "lr": 0.0007731229235880399, + "step": 7330, + "tokens_trained": 24.0118182 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 2.0816963335933623, + "grad_norm": 0.6486414074897766, + "loss": 1.0056, + "loss_ce": 1.1516366004943848, + "loss_region": 0.0, + "loss_total": 1.1516366004943848, + "lr": 0.0007727906976744186, + "step": 7340, + "tokens_trained": 24.04458364 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 2.0845330118431318, + "grad_norm": 1.2657619714736938, + "loss": 0.9905, + "loss_ce": 0.8875536322593689, + "loss_region": 0.0, + "loss_total": 0.8875536322593689, + "lr": 0.0007724584717607973, + "step": 7350, + "tokens_trained": 24.07734908 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 2.0873696900929013, + "grad_norm": 0.7896116375923157, + "loss": 1.003, + "loss_ce": 0.8636258840560913, + "loss_region": 0.0, + "loss_total": 0.8636258840560913, + "lr": 0.0007721262458471761, + "step": 7360, + "tokens_trained": 24.110113704 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 2.090206368342671, + "grad_norm": 0.7823793888092041, + "loss": 0.9967, + "loss_ce": 0.9681412577629089, + "loss_region": 0.0, + "loss_total": 0.9681412577629089, + "lr": 0.0007717940199335548, + "step": 7370, + "tokens_trained": 24.142879144 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 2.0930430465924403, + "grad_norm": 2.050823926925659, + "loss": 1.0058, + "loss_ce": 1.0626552104949951, + "loss_region": 0.0, + "loss_total": 1.0626552104949951, + "lr": 0.0007714617940199335, + "step": 7380, + "tokens_trained": 24.175643784 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 2.09587972484221, + "grad_norm": 0.6461279988288879, + "loss": 1.0019, + "loss_ce": 1.0264571905136108, + "loss_region": 0.0, + "loss_total": 1.0264571905136108, + "lr": 0.0007711295681063123, + "step": 7390, + "tokens_trained": 24.208409224 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 2.0987164030919794, + "grad_norm": 0.5234412550926208, + "loss": 0.9916, + "loss_ce": 0.9820955991744995, + "loss_region": 0.0, + "loss_total": 0.9820955991744995, + "lr": 0.000770797342192691, + "step": 7400, + "tokens_trained": 24.241174664 + }, + { + "epoch": 2.0987164030919794, + "eval_ppl": 2.7561655958835027, + "eval_runtime": 1.421, + "step": 7400, + "tokens_trained": 24.241174664 + }, + { + "epoch": 2.0987164030919794, + "eval_ppl_cds": 3.3014326940897365, + "eval_ppl_dig": 1.0975911682928448, + "eval_ppl_exon": 3.2493993587420023, + "eval_ppl_intron": 2.892761054538143, + "eval_ppl_nig": 2.5886970310449002, + "eval_ppl_promoter": 2.916515491683284, + "eval_ppl_utr": 3.052616050039932, + "step": 7400, + "tokens_trained": 24.241174664 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 2.101553081341749, + "grad_norm": 1.2801110744476318, + "loss": 1.006, + "loss_ce": 1.049383521080017, + "loss_region": 0.0, + "loss_total": 1.049383521080017, + "lr": 0.0007704651162790698, + "step": 7410, + "tokens_trained": 24.273940104 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 2.1043897595915184, + "grad_norm": 0.6224086880683899, + "loss": 0.9924, + "loss_ce": 1.0817800760269165, + "loss_region": 0.0, + "loss_total": 1.0817800760269165, + "lr": 0.0007701328903654486, + "step": 7420, + "tokens_trained": 24.306705544 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 2.107226437841288, + "grad_norm": 0.9386166334152222, + "loss": 0.9992, + "loss_ce": 0.9773650169372559, + "loss_region": 0.0, + "loss_total": 0.9773650169372559, + "lr": 0.0007698006644518273, + "step": 7430, + "tokens_trained": 24.339470184 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 2.1100631160910575, + "grad_norm": 1.3666527271270752, + "loss": 1.007, + "loss_ce": 1.003891944885254, + "loss_region": 0.0, + "loss_total": 1.003891944885254, + "lr": 0.000769468438538206, + "step": 7440, + "tokens_trained": 24.372235624 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 2.112899794340827, + "grad_norm": 0.9148083925247192, + "loss": 0.9977, + "loss_ce": 0.9857218861579895, + "loss_region": 0.0, + "loss_total": 0.9857218861579895, + "lr": 0.0007691362126245847, + "step": 7450, + "tokens_trained": 24.404997 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 2.1157364725905965, + "grad_norm": 0.5491501688957214, + "loss": 0.9986, + "loss_ce": 1.0615127086639404, + "loss_region": 0.0, + "loss_total": 1.0615127086639404, + "lr": 0.0007688039867109636, + "step": 7460, + "tokens_trained": 24.43776244 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 2.118573150840366, + "grad_norm": 0.5097406506538391, + "loss": 1.0097, + "loss_ce": 0.9484491944313049, + "loss_region": 0.0, + "loss_total": 0.9484491944313049, + "lr": 0.0007684717607973422, + "step": 7470, + "tokens_trained": 24.47052788 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 2.1214098290901355, + "grad_norm": 0.5295173525810242, + "loss": 0.9952, + "loss_ce": 0.965541422367096, + "loss_region": 0.0, + "loss_total": 0.965541422367096, + "lr": 0.0007681395348837209, + "step": 7480, + "tokens_trained": 24.50329252 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 2.124246507339905, + "grad_norm": 1.267511010169983, + "loss": 1.0046, + "loss_ce": 1.0775152444839478, + "loss_region": 0.0, + "loss_total": 1.0775152444839478, + "lr": 0.0007678073089700996, + "step": 7490, + "tokens_trained": 24.53605796 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 2.1270831855896746, + "grad_norm": 0.7830173373222351, + "loss": 1.001, + "loss_ce": 1.0153032541275024, + "loss_region": 0.0, + "loss_total": 1.0153032541275024, + "lr": 0.0007674750830564784, + "step": 7500, + "tokens_trained": 24.5688234 + }, + { + "epoch": 2.1270831855896746, + "eval_ppl": 2.75957520451909, + "eval_runtime": 1.4223, + "step": 7500, + "tokens_trained": 24.5688234 + }, + { + "epoch": 2.1270831855896746, + "eval_ppl_cds": 3.3246460106601115, + "eval_ppl_dig": 1.0991563883076179, + "eval_ppl_exon": 3.263148911210653, + "eval_ppl_intron": 2.8943881029436564, + "eval_ppl_nig": 2.594665091716985, + "eval_ppl_promoter": 2.9257597323521742, + "eval_ppl_utr": 3.056925696538019, + "step": 7500, + "tokens_trained": 24.5688234 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 2.129919863839444, + "grad_norm": 1.1780282258987427, + "loss": 0.9998, + "loss_ce": 1.0088356733322144, + "loss_region": 0.0, + "loss_total": 1.0088356733322144, + "lr": 0.0007671428571428572, + "step": 7510, + "tokens_trained": 24.60158804 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 2.1327565420892136, + "grad_norm": 0.8023217916488647, + "loss": 0.9947, + "loss_ce": 1.0973060131072998, + "loss_region": 0.0, + "loss_total": 1.0973060131072998, + "lr": 0.0007668106312292359, + "step": 7520, + "tokens_trained": 24.634352656 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 2.135593220338983, + "grad_norm": 0.8709276914596558, + "loss": 0.999, + "loss_ce": 0.9893472790718079, + "loss_region": 0.0, + "loss_total": 0.9893472790718079, + "lr": 0.0007664784053156146, + "step": 7530, + "tokens_trained": 24.66711596 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 2.1384298985887527, + "grad_norm": 0.7187210321426392, + "loss": 1.0041, + "loss_ce": 1.018195629119873, + "loss_region": 0.0, + "loss_total": 1.018195629119873, + "lr": 0.0007661461794019934, + "step": 7540, + "tokens_trained": 24.6998814 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 2.141266576838522, + "grad_norm": 0.2996567487716675, + "loss": 1.0006, + "loss_ce": 1.0125586986541748, + "loss_region": 0.0, + "loss_total": 1.0125586986541748, + "lr": 0.0007658139534883721, + "step": 7550, + "tokens_trained": 24.732646072 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 2.1441032550882917, + "grad_norm": 1.3008689880371094, + "loss": 1.0006, + "loss_ce": 1.085386872291565, + "loss_region": 0.0, + "loss_total": 1.085386872291565, + "lr": 0.0007654817275747509, + "step": 7560, + "tokens_trained": 24.765403184 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 2.146939933338061, + "grad_norm": 0.5509862303733826, + "loss": 1.0023, + "loss_ce": 1.005602240562439, + "loss_region": 0.0, + "loss_total": 1.005602240562439, + "lr": 0.0007651495016611296, + "step": 7570, + "tokens_trained": 24.798168624 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 2.1497766115878307, + "grad_norm": 0.20807354152202606, + "loss": 1.0047, + "loss_ce": 1.0187366008758545, + "loss_region": 0.0, + "loss_total": 1.0187366008758545, + "lr": 0.0007648172757475084, + "step": 7580, + "tokens_trained": 24.830934064 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 2.1526132898376003, + "grad_norm": 1.0896281003952026, + "loss": 0.9837, + "loss_ce": 1.1182106733322144, + "loss_region": 0.0, + "loss_total": 1.1182106733322144, + "lr": 0.0007644850498338871, + "step": 7590, + "tokens_trained": 24.863699504 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 2.1554499680873698, + "grad_norm": 0.4763871729373932, + "loss": 0.9934, + "loss_ce": 0.9621040225028992, + "loss_region": 0.0, + "loss_total": 0.9621040225028992, + "lr": 0.0007641528239202657, + "step": 7600, + "tokens_trained": 24.896464928 + }, + { + "epoch": 2.1554499680873698, + "eval_ppl": 2.751357368561071, + "eval_runtime": 1.4251, + "step": 7600, + "tokens_trained": 24.896464928 + }, + { + "epoch": 2.1554499680873698, + "eval_ppl_cds": 3.312129240144391, + "eval_ppl_dig": 1.0982055864630271, + "eval_ppl_exon": 3.2575399589437315, + "eval_ppl_intron": 2.8878573073028426, + "eval_ppl_nig": 2.586817646255176, + "eval_ppl_promoter": 2.872156457493778, + "eval_ppl_utr": 3.034055020980737, + "step": 7600, + "tokens_trained": 24.896464928 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 2.1582866463371393, + "grad_norm": 0.3994500935077667, + "loss": 1.0016, + "loss_ce": 1.1178946495056152, + "loss_region": 0.0, + "loss_total": 1.1178946495056152, + "lr": 0.0007638205980066445, + "step": 7610, + "tokens_trained": 24.929227968 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 2.161123324586909, + "grad_norm": 1.0573714971542358, + "loss": 0.9987, + "loss_ce": 1.065189003944397, + "loss_region": 0.0, + "loss_total": 1.065189003944397, + "lr": 0.0007634883720930233, + "step": 7620, + "tokens_trained": 24.961993408 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 2.1639600028366783, + "grad_norm": 0.6987432837486267, + "loss": 1.0011, + "loss_ce": 1.0635952949523926, + "loss_region": 0.0, + "loss_total": 1.0635952949523926, + "lr": 0.000763156146179402, + "step": 7630, + "tokens_trained": 24.994758848 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 2.166796681086448, + "grad_norm": 1.3754652738571167, + "loss": 0.9928, + "loss_ce": 1.1473740339279175, + "loss_region": 0.0, + "loss_total": 1.1473740339279175, + "lr": 0.0007628239202657807, + "step": 7640, + "tokens_trained": 25.027524288 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 2.1696333593362174, + "grad_norm": 0.828959047794342, + "loss": 1.0047, + "loss_ce": 0.8789457082748413, + "loss_region": 0.0, + "loss_total": 0.8789457082748413, + "lr": 0.0007624916943521594, + "step": 7650, + "tokens_trained": 25.06028972 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 2.172470037585987, + "grad_norm": 0.5697711706161499, + "loss": 0.9936, + "loss_ce": 0.6560234427452087, + "loss_region": 0.0, + "loss_total": 0.6560234427452087, + "lr": 0.0007621594684385383, + "step": 7660, + "tokens_trained": 25.09305516 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 2.1753067158357564, + "grad_norm": 0.7546443343162537, + "loss": 0.9955, + "loss_ce": 1.1040114164352417, + "loss_region": 0.0, + "loss_total": 1.1040114164352417, + "lr": 0.000761827242524917, + "step": 7670, + "tokens_trained": 25.125817728 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 2.178143394085526, + "grad_norm": 0.7273686528205872, + "loss": 1.0012, + "loss_ce": 1.1376252174377441, + "loss_region": 0.0, + "loss_total": 1.1376252174377441, + "lr": 0.0007614950166112957, + "step": 7680, + "tokens_trained": 25.158583168 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 2.1809800723352954, + "grad_norm": 0.4702402353286743, + "loss": 0.9942, + "loss_ce": 1.0995597839355469, + "loss_region": 0.0, + "loss_total": 1.0995597839355469, + "lr": 0.0007611627906976745, + "step": 7690, + "tokens_trained": 25.191341672 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 2.183816750585065, + "grad_norm": 0.4278646409511566, + "loss": 0.994, + "loss_ce": 0.9860332608222961, + "loss_region": 0.0, + "loss_total": 0.9860332608222961, + "lr": 0.0007608305647840532, + "step": 7700, + "tokens_trained": 25.224107112 + }, + { + "epoch": 2.183816750585065, + "eval_ppl": 2.7500440927211516, + "eval_runtime": 1.4231, + "step": 7700, + "tokens_trained": 25.224107112 + }, + { + "epoch": 2.183816750585065, + "eval_ppl_cds": 3.2510708249395677, + "eval_ppl_dig": 1.0971984484068207, + "eval_ppl_exon": 3.250632633725076, + "eval_ppl_intron": 2.8857385834222042, + "eval_ppl_nig": 2.586582940113895, + "eval_ppl_promoter": 2.8462839606344303, + "eval_ppl_utr": 3.0155016934049317, + "step": 7700, + "tokens_trained": 25.224107112 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 2.1866534288348345, + "grad_norm": 0.4870636761188507, + "loss": 1.0012, + "loss_ce": 1.0289818048477173, + "loss_region": 0.0, + "loss_total": 1.0289818048477173, + "lr": 0.000760498338870432, + "step": 7710, + "tokens_trained": 25.256872552 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 2.189490107084604, + "grad_norm": 1.28205406665802, + "loss": 1.0004, + "loss_ce": 1.0419155359268188, + "loss_region": 0.0, + "loss_total": 1.0419155359268188, + "lr": 0.0007601661129568106, + "step": 7720, + "tokens_trained": 25.289637992 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 2.1923267853343735, + "grad_norm": 0.5651472806930542, + "loss": 0.9918, + "loss_ce": 1.0768500566482544, + "loss_region": 0.0, + "loss_total": 1.0768500566482544, + "lr": 0.0007598338870431894, + "step": 7730, + "tokens_trained": 25.322402632 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 2.195163463584143, + "grad_norm": 0.5187858939170837, + "loss": 0.9963, + "loss_ce": 0.9749172925949097, + "loss_region": 0.0, + "loss_total": 0.9749172925949097, + "lr": 0.0007595016611295681, + "step": 7740, + "tokens_trained": 25.355165584 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 2.1980001418339126, + "grad_norm": 0.3698139488697052, + "loss": 1.0002, + "loss_ce": 0.9338331818580627, + "loss_region": 0.0, + "loss_total": 0.9338331818580627, + "lr": 0.0007591694352159468, + "step": 7750, + "tokens_trained": 25.387931024 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 2.200836820083682, + "grad_norm": 1.170602560043335, + "loss": 0.9943, + "loss_ce": 0.9886414408683777, + "loss_region": 0.0, + "loss_total": 0.9886414408683777, + "lr": 0.0007588372093023256, + "step": 7760, + "tokens_trained": 25.420695664 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 2.2036734983334516, + "grad_norm": 0.5639888644218445, + "loss": 0.9981, + "loss_ce": 0.962650716304779, + "loss_region": 0.0, + "loss_total": 0.962650716304779, + "lr": 0.0007585049833887044, + "step": 7770, + "tokens_trained": 25.453461104 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 2.206510176583221, + "grad_norm": 0.24882017076015472, + "loss": 0.9959, + "loss_ce": 1.0465995073318481, + "loss_region": 0.0, + "loss_total": 1.0465995073318481, + "lr": 0.0007581727574750831, + "step": 7780, + "tokens_trained": 25.486226544 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 2.2093468548329906, + "grad_norm": 1.3218917846679688, + "loss": 1.0026, + "loss_ce": 0.8688437938690186, + "loss_region": 0.0, + "loss_total": 0.8688437938690186, + "lr": 0.0007578405315614618, + "step": 7790, + "tokens_trained": 25.518991984 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 2.21218353308276, + "grad_norm": 0.6549669504165649, + "loss": 0.9961, + "loss_ce": 0.8071460723876953, + "loss_region": 0.0, + "loss_total": 0.8071460723876953, + "lr": 0.0007575083056478405, + "step": 7800, + "tokens_trained": 25.551757424 + }, + { + "epoch": 2.21218353308276, + "eval_ppl": 2.744267560575186, + "eval_runtime": 1.4329, + "step": 7800, + "tokens_trained": 25.551757424 + }, + { + "epoch": 2.21218353308276, + "eval_ppl_cds": 3.24378347482224, + "eval_ppl_dig": 1.099771944895424, + "eval_ppl_exon": 3.2630807719543284, + "eval_ppl_intron": 2.8882382108116684, + "eval_ppl_nig": 2.588778580208235, + "eval_ppl_promoter": 2.879977174765759, + "eval_ppl_utr": 3.0124194046589974, + "step": 7800, + "tokens_trained": 25.551757424 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 2.2150202113325297, + "grad_norm": 0.5002632737159729, + "loss": 0.997, + "loss_ce": 0.9726791977882385, + "loss_region": 0.0, + "loss_total": 0.9726791977882385, + "lr": 0.0007571760797342194, + "step": 7810, + "tokens_trained": 25.584522064 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 2.217856889582299, + "grad_norm": 0.4766501486301422, + "loss": 1.0056, + "loss_ce": 1.095262885093689, + "loss_region": 0.0, + "loss_total": 1.095262885093689, + "lr": 0.0007568438538205981, + "step": 7820, + "tokens_trained": 25.617287344 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 2.2206935678320687, + "grad_norm": 0.9868918657302856, + "loss": 0.9952, + "loss_ce": 1.0100709199905396, + "loss_region": 0.0, + "loss_total": 1.0100709199905396, + "lr": 0.0007565116279069768, + "step": 7830, + "tokens_trained": 25.650052784 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 2.2235302460818382, + "grad_norm": 0.8106172680854797, + "loss": 0.9974, + "loss_ce": 0.8005494475364685, + "loss_region": 0.0, + "loss_total": 0.8005494475364685, + "lr": 0.0007561794019933554, + "step": 7840, + "tokens_trained": 25.682818224 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 2.2263669243316078, + "grad_norm": 0.2592508792877197, + "loss": 0.9965, + "loss_ce": 1.0648443698883057, + "loss_region": 0.0, + "loss_total": 1.0648443698883057, + "lr": 0.0007558471760797342, + "step": 7850, + "tokens_trained": 25.715583664 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 2.2292036025813773, + "grad_norm": 0.531985878944397, + "loss": 0.9989, + "loss_ce": 0.9530308246612549, + "loss_region": 0.0, + "loss_total": 0.9530308246612549, + "lr": 0.000755514950166113, + "step": 7860, + "tokens_trained": 25.748349104 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 2.232040280831147, + "grad_norm": 0.6074833273887634, + "loss": 0.9904, + "loss_ce": 0.9225383400917053, + "loss_region": 0.0, + "loss_total": 0.9225383400917053, + "lr": 0.0007551827242524917, + "step": 7870, + "tokens_trained": 25.781114544 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 2.2348769590809163, + "grad_norm": 0.9639485478401184, + "loss": 0.9879, + "loss_ce": 1.0513375997543335, + "loss_region": 0.0, + "loss_total": 1.0513375997543335, + "lr": 0.0007548504983388704, + "step": 7880, + "tokens_trained": 25.813879984 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 2.237713637330686, + "grad_norm": 0.303068608045578, + "loss": 0.9935, + "loss_ce": 0.9540935158729553, + "loss_region": 0.0, + "loss_total": 0.9540935158729553, + "lr": 0.0007545182724252492, + "step": 7890, + "tokens_trained": 25.846645424 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 2.2405503155804554, + "grad_norm": 0.5269011855125427, + "loss": 0.9928, + "loss_ce": 0.9369443655014038, + "loss_region": 0.0, + "loss_total": 0.9369443655014038, + "lr": 0.0007541860465116279, + "step": 7900, + "tokens_trained": 25.879410864 + }, + { + "epoch": 2.2405503155804554, + "eval_ppl": 2.7363542994192183, + "eval_runtime": 1.4493, + "step": 7900, + "tokens_trained": 25.879410864 + }, + { + "epoch": 2.2405503155804554, + "eval_ppl_cds": 3.1927041900030964, + "eval_ppl_dig": 1.1000094429003062, + "eval_ppl_exon": 3.269275596040741, + "eval_ppl_intron": 2.8798728538628975, + "eval_ppl_nig": 2.580984999359659, + "eval_ppl_promoter": 2.8511122908637665, + "eval_ppl_utr": 2.9835656182540795, + "step": 7900, + "tokens_trained": 25.879410864 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 2.243386993830225, + "grad_norm": 0.7621945142745972, + "loss": 0.9946, + "loss_ce": 1.022991418838501, + "loss_region": 0.0, + "loss_total": 1.022991418838501, + "lr": 0.0007538538205980067, + "step": 7910, + "tokens_trained": 25.912176304 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 2.2462236720799944, + "grad_norm": 0.5068185329437256, + "loss": 0.9896, + "loss_ce": 1.083464503288269, + "loss_region": 0.0, + "loss_total": 1.083464503288269, + "lr": 0.0007535215946843854, + "step": 7920, + "tokens_trained": 25.944941744 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 2.249060350329764, + "grad_norm": 0.9783192276954651, + "loss": 0.9902, + "loss_ce": 0.9971379637718201, + "loss_region": 0.0, + "loss_total": 0.9971379637718201, + "lr": 0.0007531893687707642, + "step": 7930, + "tokens_trained": 25.977707184 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 2.2518970285795334, + "grad_norm": 0.9254680275917053, + "loss": 0.9957, + "loss_ce": 1.0690739154815674, + "loss_region": 0.0, + "loss_total": 1.0690739154815674, + "lr": 0.0007528571428571429, + "step": 7940, + "tokens_trained": 26.010471824 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 2.254733706829303, + "grad_norm": 0.8929104208946228, + "loss": 0.9942, + "loss_ce": 0.9405961632728577, + "loss_region": 0.0, + "loss_total": 0.9405961632728577, + "lr": 0.0007525249169435216, + "step": 7950, + "tokens_trained": 26.043237264 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 2.2575703850790725, + "grad_norm": 0.5788987874984741, + "loss": 0.99, + "loss_ce": 1.0785738229751587, + "loss_region": 0.0, + "loss_total": 1.0785738229751587, + "lr": 0.0007521926910299005, + "step": 7960, + "tokens_trained": 26.076002704 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 2.260407063328842, + "grad_norm": 0.8546907305717468, + "loss": 0.9972, + "loss_ce": 0.9655200839042664, + "loss_region": 0.0, + "loss_total": 0.9655200839042664, + "lr": 0.0007518604651162791, + "step": 7970, + "tokens_trained": 26.108768144 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 2.2632437415786115, + "grad_norm": 0.7821748852729797, + "loss": 0.9929, + "loss_ce": 1.0527219772338867, + "loss_region": 0.0, + "loss_total": 1.0527219772338867, + "lr": 0.0007515282392026578, + "step": 7980, + "tokens_trained": 26.141533584 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 2.266080419828381, + "grad_norm": 0.8153959512710571, + "loss": 0.9969, + "loss_ce": 0.9742423295974731, + "loss_region": 0.0, + "loss_total": 0.9742423295974731, + "lr": 0.0007511960132890365, + "step": 7990, + "tokens_trained": 26.174299024 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 2.2689170980781506, + "grad_norm": 0.6376358270645142, + "loss": 0.9841, + "loss_ce": 0.9239227175712585, + "loss_region": 0.0, + "loss_total": 0.9239227175712585, + "lr": 0.0007508637873754152, + "step": 8000, + "tokens_trained": 26.207064464 + }, + { + "epoch": 2.2689170980781506, + "eval_ppl": 2.7408013899499353, + "eval_runtime": 1.4176, + "step": 8000, + "tokens_trained": 26.207064464 + }, + { + "epoch": 2.2689170980781506, + "eval_ppl_cds": 3.252823984422004, + "eval_ppl_dig": 1.1012084782510505, + "eval_ppl_exon": 3.2640467833836038, + "eval_ppl_intron": 2.890633888225694, + "eval_ppl_nig": 2.5965859872271553, + "eval_ppl_promoter": 2.873935424417248, + "eval_ppl_utr": 3.0151439105117035, + "step": 8000, + "tokens_trained": 26.207064464 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 2.27175377632792, + "grad_norm": 0.5925382971763611, + "loss": 0.9889, + "loss_ce": 0.9932094216346741, + "loss_region": 0.0, + "loss_total": 0.9932094216346741, + "lr": 0.000750531561461794, + "step": 8010, + "tokens_trained": 26.239829904 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 2.2745904545776896, + "grad_norm": 0.8806548714637756, + "loss": 0.9892, + "loss_ce": 1.1307108402252197, + "loss_region": 0.0, + "loss_total": 1.1307108402252197, + "lr": 0.0007501993355481728, + "step": 8020, + "tokens_trained": 26.272595344 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 2.277427132827459, + "grad_norm": 0.3244630694389343, + "loss": 0.9846, + "loss_ce": 0.9263905882835388, + "loss_region": 0.0, + "loss_total": 0.9263905882835388, + "lr": 0.0007498671096345515, + "step": 8030, + "tokens_trained": 26.305360784 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 2.2802638110772286, + "grad_norm": 0.49009421467781067, + "loss": 0.9913, + "loss_ce": 1.0677887201309204, + "loss_region": 0.0, + "loss_total": 1.0677887201309204, + "lr": 0.0007495348837209303, + "step": 8040, + "tokens_trained": 26.338126208 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 2.283100489326998, + "grad_norm": 0.9748938679695129, + "loss": 0.9973, + "loss_ce": 0.8806904554367065, + "loss_region": 0.0, + "loss_total": 0.8806904554367065, + "lr": 0.000749202657807309, + "step": 8050, + "tokens_trained": 26.370891648 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 2.2859371675767677, + "grad_norm": 0.5555410981178284, + "loss": 0.9938, + "loss_ce": 1.0121716260910034, + "loss_region": 0.0, + "loss_total": 1.0121716260910034, + "lr": 0.0007488704318936877, + "step": 8060, + "tokens_trained": 26.403656288 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 2.288773845826537, + "grad_norm": 0.7922304272651672, + "loss": 0.9909, + "loss_ce": 0.9498851299285889, + "loss_region": 0.0, + "loss_total": 0.9498851299285889, + "lr": 0.0007485382059800665, + "step": 8070, + "tokens_trained": 26.436421728 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 2.2916105240763067, + "grad_norm": 0.7461918592453003, + "loss": 0.9889, + "loss_ce": 1.0301202535629272, + "loss_region": 0.0, + "loss_total": 1.0301202535629272, + "lr": 0.0007482059800664453, + "step": 8080, + "tokens_trained": 26.469187168 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 2.2944472023260762, + "grad_norm": 0.6083531975746155, + "loss": 0.9841, + "loss_ce": 0.9082334637641907, + "loss_region": 0.0, + "loss_total": 0.9082334637641907, + "lr": 0.0007478737541528239, + "step": 8090, + "tokens_trained": 26.501951008 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 2.2972838805758458, + "grad_norm": 0.2979046702384949, + "loss": 1.0002, + "loss_ce": 1.0317283868789673, + "loss_region": 0.0, + "loss_total": 1.0317283868789673, + "lr": 0.0007475415282392026, + "step": 8100, + "tokens_trained": 26.534716448 + }, + { + "epoch": 2.2972838805758458, + "eval_ppl": 2.7330248577268583, + "eval_runtime": 1.4278, + "step": 8100, + "tokens_trained": 26.534716448 + }, + { + "epoch": 2.2972838805758458, + "eval_ppl_cds": 3.1783795973418347, + "eval_ppl_dig": 1.103372564137305, + "eval_ppl_exon": 3.2576595722109554, + "eval_ppl_intron": 2.8819958412645317, + "eval_ppl_nig": 2.592406325722491, + "eval_ppl_promoter": 2.855050196107403, + "eval_ppl_utr": 3.0024997171255943, + "step": 8100, + "tokens_trained": 26.534716448 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 2.3001205588256153, + "grad_norm": 0.617983877658844, + "loss": 0.988, + "loss_ce": 1.043581247329712, + "loss_region": 0.0, + "loss_total": 1.043581247329712, + "lr": 0.0007472093023255813, + "step": 8110, + "tokens_trained": 26.5674814 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 2.302957237075385, + "grad_norm": 0.5208008885383606, + "loss": 0.9972, + "loss_ce": 1.0981296300888062, + "loss_region": 0.0, + "loss_total": 1.0981296300888062, + "lr": 0.0007468770764119602, + "step": 8120, + "tokens_trained": 26.60024684 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 2.3057939153251543, + "grad_norm": 0.45728814601898193, + "loss": 0.9875, + "loss_ce": 1.0919119119644165, + "loss_region": 0.0, + "loss_total": 1.0919119119644165, + "lr": 0.0007465448504983389, + "step": 8130, + "tokens_trained": 26.633011272 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 2.308630593574924, + "grad_norm": 0.3884800672531128, + "loss": 0.9888, + "loss_ce": 0.9302235841751099, + "loss_region": 0.0, + "loss_total": 0.9302235841751099, + "lr": 0.0007462126245847176, + "step": 8140, + "tokens_trained": 26.665774 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 2.3114672718246934, + "grad_norm": 0.46373307704925537, + "loss": 0.994, + "loss_ce": 1.0090198516845703, + "loss_region": 0.0, + "loss_total": 1.0090198516845703, + "lr": 0.0007458803986710963, + "step": 8150, + "tokens_trained": 26.69853944 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 2.314303950074463, + "grad_norm": 0.8128484487533569, + "loss": 0.9904, + "loss_ce": 1.0319392681121826, + "loss_region": 0.0, + "loss_total": 1.0319392681121826, + "lr": 0.0007455481727574751, + "step": 8160, + "tokens_trained": 26.73130488 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 2.3171406283242324, + "grad_norm": 0.8561034202575684, + "loss": 0.995, + "loss_ce": 0.9631012082099915, + "loss_region": 0.0, + "loss_total": 0.9631012082099915, + "lr": 0.0007452159468438539, + "step": 8170, + "tokens_trained": 26.76407032 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 2.319977306574002, + "grad_norm": 0.565175473690033, + "loss": 0.9872, + "loss_ce": 1.0640523433685303, + "loss_region": 0.0, + "loss_total": 1.0640523433685303, + "lr": 0.0007448837209302326, + "step": 8180, + "tokens_trained": 26.79683576 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 2.3228139848237714, + "grad_norm": 0.4340415596961975, + "loss": 0.9953, + "loss_ce": 1.1132088899612427, + "loss_region": 0.0, + "loss_total": 1.1132088899612427, + "lr": 0.0007445514950166113, + "step": 8190, + "tokens_trained": 26.8296012 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 2.325650663073541, + "grad_norm": 0.5439237356185913, + "loss": 0.9916, + "loss_ce": 1.018317699432373, + "loss_region": 0.0, + "loss_total": 1.018317699432373, + "lr": 0.0007442192691029901, + "step": 8200, + "tokens_trained": 26.86236568 + }, + { + "epoch": 2.325650663073541, + "eval_ppl": 2.7298224731747918, + "eval_runtime": 1.4217, + "step": 8200, + "tokens_trained": 26.86236568 + }, + { + "epoch": 2.325650663073541, + "eval_ppl_cds": 3.196459500945774, + "eval_ppl_dig": 1.1011372508163217, + "eval_ppl_exon": 3.2542062715822544, + "eval_ppl_intron": 2.87515808899778, + "eval_ppl_nig": 2.5838539083329337, + "eval_ppl_promoter": 2.8139161739746297, + "eval_ppl_utr": 2.9957166628015064, + "step": 8200, + "tokens_trained": 26.86236568 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 2.3284873413233105, + "grad_norm": 0.17269718647003174, + "loss": 0.9879, + "loss_ce": 0.8390361070632935, + "loss_region": 0.0, + "loss_total": 0.8390361070632935, + "lr": 0.0007438870431893688, + "step": 8210, + "tokens_trained": 26.89513112 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 2.33132401957308, + "grad_norm": 1.373714804649353, + "loss": 0.9838, + "loss_ce": 0.9623157382011414, + "loss_region": 0.0, + "loss_total": 0.9623157382011414, + "lr": 0.0007435548172757475, + "step": 8220, + "tokens_trained": 26.92789576 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 2.3341606978228495, + "grad_norm": 0.5136083960533142, + "loss": 0.9903, + "loss_ce": 0.940282940864563, + "loss_region": 0.0, + "loss_total": 0.940282940864563, + "lr": 0.0007432225913621262, + "step": 8230, + "tokens_trained": 26.9606612 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 2.336997376072619, + "grad_norm": 0.2980974018573761, + "loss": 0.9887, + "loss_ce": 1.0230684280395508, + "loss_region": 0.0, + "loss_total": 1.0230684280395508, + "lr": 0.000742890365448505, + "step": 8240, + "tokens_trained": 26.99342664 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 2.3398340543223886, + "grad_norm": 1.053428292274475, + "loss": 0.9931, + "loss_ce": 0.8636865615844727, + "loss_region": 0.0, + "loss_total": 0.8636865615844727, + "lr": 0.0007425581395348837, + "step": 8250, + "tokens_trained": 27.02619208 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 2.342670732572158, + "grad_norm": 0.5025226473808289, + "loss": 0.9931, + "loss_ce": 0.8538046479225159, + "loss_region": 0.0, + "loss_total": 0.8538046479225159, + "lr": 0.0007422259136212624, + "step": 8260, + "tokens_trained": 27.05895752 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 2.3455074108219276, + "grad_norm": 0.16127973794937134, + "loss": 0.9844, + "loss_ce": 0.9715340733528137, + "loss_region": 0.0, + "loss_total": 0.9715340733528137, + "lr": 0.0007418936877076412, + "step": 8270, + "tokens_trained": 27.09172296 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 2.348344089071697, + "grad_norm": 0.9249735474586487, + "loss": 0.994, + "loss_ce": 0.922537088394165, + "loss_region": 0.0, + "loss_total": 0.922537088394165, + "lr": 0.00074156146179402, + "step": 8280, + "tokens_trained": 27.1244884 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 2.3511807673214666, + "grad_norm": 0.529431164264679, + "loss": 0.9804, + "loss_ce": 0.9556260704994202, + "loss_region": 0.0, + "loss_total": 0.9556260704994202, + "lr": 0.0007412292358803987, + "step": 8290, + "tokens_trained": 27.15725384 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 2.354017445571236, + "grad_norm": 0.7171990871429443, + "loss": 0.9846, + "loss_ce": 1.0559780597686768, + "loss_region": 0.0, + "loss_total": 1.0559780597686768, + "lr": 0.0007408970099667774, + "step": 8300, + "tokens_trained": 27.19001928 + }, + { + "epoch": 2.354017445571236, + "eval_ppl": 2.729238081061228, + "eval_runtime": 1.4282, + "step": 8300, + "tokens_trained": 27.19001928 + }, + { + "epoch": 2.354017445571236, + "eval_ppl_cds": 3.212145698892842, + "eval_ppl_dig": 1.0986551335364836, + "eval_ppl_exon": 3.2727153498701744, + "eval_ppl_intron": 2.8820906726501114, + "eval_ppl_nig": 2.5853522655397945, + "eval_ppl_promoter": 2.799596164557387, + "eval_ppl_utr": 2.9707948168664697, + "step": 8300, + "tokens_trained": 27.19001928 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 2.3568541238210057, + "grad_norm": 0.8374295234680176, + "loss": 0.9907, + "loss_ce": 1.050769329071045, + "loss_region": 0.0, + "loss_total": 1.050769329071045, + "lr": 0.0007405647840531561, + "step": 8310, + "tokens_trained": 27.22278472 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 2.359690802070775, + "grad_norm": 0.17253491282463074, + "loss": 0.9895, + "loss_ce": 0.9606397151947021, + "loss_region": 0.0, + "loss_total": 0.9606397151947021, + "lr": 0.000740232558139535, + "step": 8320, + "tokens_trained": 27.25555016 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 2.3625274803205447, + "grad_norm": 1.112982988357544, + "loss": 0.9839, + "loss_ce": 1.0603920221328735, + "loss_region": 0.0, + "loss_total": 1.0603920221328735, + "lr": 0.0007399003322259137, + "step": 8330, + "tokens_trained": 27.2883156 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 2.3653641585703142, + "grad_norm": 0.7953630685806274, + "loss": 0.9852, + "loss_ce": 0.9815343618392944, + "loss_region": 0.0, + "loss_total": 0.9815343618392944, + "lr": 0.0007395681063122923, + "step": 8340, + "tokens_trained": 27.32108104 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 2.3682008368200838, + "grad_norm": 0.4420929253101349, + "loss": 0.986, + "loss_ce": 1.0365334749221802, + "loss_region": 0.0, + "loss_total": 1.0365334749221802, + "lr": 0.000739235880398671, + "step": 8350, + "tokens_trained": 27.35384648 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 2.3710375150698533, + "grad_norm": 1.162212610244751, + "loss": 0.99, + "loss_ce": 0.949650228023529, + "loss_region": 0.0, + "loss_total": 0.949650228023529, + "lr": 0.0007389036544850498, + "step": 8360, + "tokens_trained": 27.386609784 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 2.373874193319623, + "grad_norm": 0.18821220099925995, + "loss": 0.9925, + "loss_ce": 0.9568367600440979, + "loss_region": 0.0, + "loss_total": 0.9568367600440979, + "lr": 0.0007385714285714286, + "step": 8370, + "tokens_trained": 27.419375224 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 2.3767108715693923, + "grad_norm": 1.0014206171035767, + "loss": 0.9997, + "loss_ce": 0.8971546292304993, + "loss_region": 0.0, + "loss_total": 0.8971546292304993, + "lr": 0.0007382392026578073, + "step": 8380, + "tokens_trained": 27.452140664 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 2.379547549819162, + "grad_norm": 0.7016841173171997, + "loss": 0.9878, + "loss_ce": 0.9658421277999878, + "loss_region": 0.0, + "loss_total": 0.9658421277999878, + "lr": 0.000737906976744186, + "step": 8390, + "tokens_trained": 27.484904504 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 2.3823842280689314, + "grad_norm": 0.6227653622627258, + "loss": 0.9951, + "loss_ce": 1.038636326789856, + "loss_region": 0.0, + "loss_total": 1.038636326789856, + "lr": 0.0007375747508305648, + "step": 8400, + "tokens_trained": 27.517669944 + }, + { + "epoch": 2.3823842280689314, + "eval_ppl": 2.7250322651806975, + "eval_runtime": 1.4197, + "step": 8400, + "tokens_trained": 27.517669944 + }, + { + "epoch": 2.3823842280689314, + "eval_ppl_cds": 3.233810770783425, + "eval_ppl_dig": 1.0911041895257327, + "eval_ppl_exon": 3.25688363047628, + "eval_ppl_intron": 2.8760081277679825, + "eval_ppl_nig": 2.569809037862364, + "eval_ppl_promoter": 2.788798826990937, + "eval_ppl_utr": 2.9661607956141096, + "step": 8400, + "tokens_trained": 27.517669944 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 2.385220906318701, + "grad_norm": 0.46367159485816956, + "loss": 0.9878, + "loss_ce": 0.9733189940452576, + "loss_region": 0.0, + "loss_total": 0.9733189940452576, + "lr": 0.0007372425249169435, + "step": 8410, + "tokens_trained": 27.550435384 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 2.3880575845684704, + "grad_norm": 0.6468173861503601, + "loss": 0.9906, + "loss_ce": 0.874672532081604, + "loss_region": 0.0, + "loss_total": 0.874672532081604, + "lr": 0.0007369102990033223, + "step": 8420, + "tokens_trained": 27.583198336 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 2.39089426281824, + "grad_norm": 0.3083384335041046, + "loss": 0.9858, + "loss_ce": 0.9261329770088196, + "loss_region": 0.0, + "loss_total": 0.9261329770088196, + "lr": 0.0007365780730897011, + "step": 8430, + "tokens_trained": 27.615963616 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 2.3937309410680094, + "grad_norm": 0.6261957883834839, + "loss": 0.9927, + "loss_ce": 1.0142236948013306, + "loss_region": 0.0, + "loss_total": 1.0142236948013306, + "lr": 0.0007362458471760798, + "step": 8440, + "tokens_trained": 27.648725192 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 2.396567619317779, + "grad_norm": 0.6377774477005005, + "loss": 0.9918, + "loss_ce": 0.9651938676834106, + "loss_region": 0.0, + "loss_total": 0.9651938676834106, + "lr": 0.0007359136212624585, + "step": 8450, + "tokens_trained": 27.681490632 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 2.3994042975675485, + "grad_norm": 0.42323675751686096, + "loss": 0.9806, + "loss_ce": 0.9034174084663391, + "loss_region": 0.0, + "loss_total": 0.9034174084663391, + "lr": 0.0007355813953488372, + "step": 8460, + "tokens_trained": 27.714256072 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 2.402240975817318, + "grad_norm": 0.34761300683021545, + "loss": 0.9908, + "loss_ce": 1.0787242650985718, + "loss_region": 0.0, + "loss_total": 1.0787242650985718, + "lr": 0.000735249169435216, + "step": 8470, + "tokens_trained": 27.747021512 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 2.4050776540670875, + "grad_norm": 0.37748709321022034, + "loss": 0.9795, + "loss_ce": 1.0569677352905273, + "loss_region": 0.0, + "loss_total": 1.0569677352905273, + "lr": 0.0007349169435215947, + "step": 8480, + "tokens_trained": 27.779786952 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 2.407914332316857, + "grad_norm": 0.7704441547393799, + "loss": 0.9906, + "loss_ce": 1.129168152809143, + "loss_region": 0.0, + "loss_total": 1.129168152809143, + "lr": 0.0007345847176079734, + "step": 8490, + "tokens_trained": 27.812552392 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 2.4107510105666266, + "grad_norm": 0.41767439246177673, + "loss": 0.9846, + "loss_ce": 0.931790828704834, + "loss_region": 0.0, + "loss_total": 0.931790828704834, + "lr": 0.0007342524916943521, + "step": 8500, + "tokens_trained": 27.845317832 + }, + { + "epoch": 2.4107510105666266, + "eval_ppl": 2.712576379109035, + "eval_runtime": 1.4125, + "step": 8500, + "tokens_trained": 27.845317832 + }, + { + "epoch": 2.4107510105666266, + "eval_ppl_cds": 3.1993995351165583, + "eval_ppl_dig": 1.0961448846601316, + "eval_ppl_exon": 3.2579177414868083, + "eval_ppl_intron": 2.870843289520874, + "eval_ppl_nig": 2.563238281363021, + "eval_ppl_promoter": 2.7821906159172727, + "eval_ppl_utr": 2.945183990454674, + "step": 8500, + "tokens_trained": 27.845317832 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 2.413587688816396, + "grad_norm": 0.5275272130966187, + "loss": 0.9872, + "loss_ce": 0.8734601140022278, + "loss_region": 0.0, + "loss_total": 0.8734601140022278, + "lr": 0.0007339202657807309, + "step": 8510, + "tokens_trained": 27.878081672 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 2.4164243670661656, + "grad_norm": 0.6070772409439087, + "loss": 0.9832, + "loss_ce": 1.031602144241333, + "loss_region": 0.0, + "loss_total": 1.031602144241333, + "lr": 0.0007335880398671097, + "step": 8520, + "tokens_trained": 27.910847112 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 2.419261045315935, + "grad_norm": 0.9143785834312439, + "loss": 0.9873, + "loss_ce": 0.8822223544120789, + "loss_region": 0.0, + "loss_total": 0.8822223544120789, + "lr": 0.0007332558139534884, + "step": 8530, + "tokens_trained": 27.943612552 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 2.4220977235657046, + "grad_norm": 0.5928965210914612, + "loss": 0.9868, + "loss_ce": 0.9988918304443359, + "loss_region": 0.0, + "loss_total": 0.9988918304443359, + "lr": 0.0007329235880398671, + "step": 8540, + "tokens_trained": 27.976377992 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 2.424934401815474, + "grad_norm": 0.8086594939231873, + "loss": 0.9872, + "loss_ce": 0.8709831833839417, + "loss_region": 0.0, + "loss_total": 0.8709831833839417, + "lr": 0.0007325913621262459, + "step": 8550, + "tokens_trained": 28.009141296 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 2.4277710800652437, + "grad_norm": 0.6255665421485901, + "loss": 0.9888, + "loss_ce": 1.0978755950927734, + "loss_region": 0.0, + "loss_total": 1.0978755950927734, + "lr": 0.0007322591362126246, + "step": 8560, + "tokens_trained": 28.041906736 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 2.430607758315013, + "grad_norm": 0.6477224826812744, + "loss": 0.9934, + "loss_ce": 1.0695774555206299, + "loss_region": 0.0, + "loss_total": 1.0695774555206299, + "lr": 0.0007319269102990034, + "step": 8570, + "tokens_trained": 28.074672176 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 2.4334444365647827, + "grad_norm": 1.0502874851226807, + "loss": 0.9831, + "loss_ce": 1.0647993087768555, + "loss_region": 0.0, + "loss_total": 1.0647993087768555, + "lr": 0.0007315946843853821, + "step": 8580, + "tokens_trained": 28.107436008 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 2.4362811148145522, + "grad_norm": 0.43670040369033813, + "loss": 0.9869, + "loss_ce": 0.9758757948875427, + "loss_region": 0.0, + "loss_total": 0.9758757948875427, + "lr": 0.0007312624584717608, + "step": 8590, + "tokens_trained": 28.140201448 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 2.4391177930643217, + "grad_norm": 0.8722648620605469, + "loss": 0.9829, + "loss_ce": 1.059212565422058, + "loss_region": 0.0, + "loss_total": 1.059212565422058, + "lr": 0.0007309302325581395, + "step": 8600, + "tokens_trained": 28.172966888 + }, + { + "epoch": 2.4391177930643217, + "eval_ppl": 2.7165529532838217, + "eval_runtime": 1.43, + "step": 8600, + "tokens_trained": 28.172966888 + }, + { + "epoch": 2.4391177930643217, + "eval_ppl_cds": 3.186705713380016, + "eval_ppl_dig": 1.0949231876490368, + "eval_ppl_exon": 3.2637623162578815, + "eval_ppl_intron": 2.87173197146853, + "eval_ppl_nig": 2.5561718721469573, + "eval_ppl_promoter": 2.750592022275602, + "eval_ppl_utr": 2.9442667842354875, + "step": 8600, + "tokens_trained": 28.172966888 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 2.4419544713140913, + "grad_norm": 0.5810690522193909, + "loss": 0.9899, + "loss_ce": 0.9366958141326904, + "loss_region": 0.0, + "loss_total": 0.9366958141326904, + "lr": 0.0007305980066445182, + "step": 8610, + "tokens_trained": 28.205730728 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 2.444791149563861, + "grad_norm": 0.7284494042396545, + "loss": 0.9884, + "loss_ce": 1.1194785833358765, + "loss_region": 0.0, + "loss_total": 1.1194785833358765, + "lr": 0.000730265780730897, + "step": 8620, + "tokens_trained": 28.238496168 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 2.4476278278136303, + "grad_norm": 0.5881525278091431, + "loss": 0.985, + "loss_ce": 0.9997726678848267, + "loss_region": 0.0, + "loss_total": 0.9997726678848267, + "lr": 0.0007299335548172758, + "step": 8630, + "tokens_trained": 28.271258408 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 2.4504645060634, + "grad_norm": 0.49066415429115295, + "loss": 0.9889, + "loss_ce": 1.1145786046981812, + "loss_region": 0.0, + "loss_total": 1.1145786046981812, + "lr": 0.0007296013289036545, + "step": 8640, + "tokens_trained": 28.304020264 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 2.4533011843131693, + "grad_norm": 0.7167149782180786, + "loss": 0.985, + "loss_ce": 1.065351963043213, + "loss_region": 0.0, + "loss_total": 1.065351963043213, + "lr": 0.0007292691029900332, + "step": 8650, + "tokens_trained": 28.336784904 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 2.456137862562939, + "grad_norm": 0.5837216377258301, + "loss": 0.9855, + "loss_ce": 0.9829132556915283, + "loss_region": 0.0, + "loss_total": 0.9829132556915283, + "lr": 0.000728936877076412, + "step": 8660, + "tokens_trained": 28.369550344 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 2.4589745408127084, + "grad_norm": 1.0803719758987427, + "loss": 0.9844, + "loss_ce": 1.0356309413909912, + "loss_region": 0.0, + "loss_total": 1.0356309413909912, + "lr": 0.0007286046511627908, + "step": 8670, + "tokens_trained": 28.402315784 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 2.461811219062478, + "grad_norm": 0.6330556273460388, + "loss": 0.9893, + "loss_ce": 0.9181745648384094, + "loss_region": 0.0, + "loss_total": 0.9181745648384094, + "lr": 0.0007282724252491695, + "step": 8680, + "tokens_trained": 28.435081224 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 2.4646478973122474, + "grad_norm": 1.051425814628601, + "loss": 0.9813, + "loss_ce": 1.0335696935653687, + "loss_region": 0.0, + "loss_total": 1.0335696935653687, + "lr": 0.0007279401993355482, + "step": 8690, + "tokens_trained": 28.467846664 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 2.467484575562017, + "grad_norm": 0.3882218897342682, + "loss": 0.9883, + "loss_ce": 1.0104987621307373, + "loss_region": 0.0, + "loss_total": 1.0104987621307373, + "lr": 0.000727607973421927, + "step": 8700, + "tokens_trained": 28.500612104 + }, + { + "epoch": 2.467484575562017, + "eval_ppl": 2.718115922766005, + "eval_runtime": 1.4225, + "step": 8700, + "tokens_trained": 28.500612104 + }, + { + "epoch": 2.467484575562017, + "eval_ppl_cds": 3.227773721173112, + "eval_ppl_dig": 1.0981689695397963, + "eval_ppl_exon": 3.2488164609128605, + "eval_ppl_intron": 2.8760421856580565, + "eval_ppl_nig": 2.565105046234904, + "eval_ppl_promoter": 2.7941826912077428, + "eval_ppl_utr": 2.9559965774199446, + "step": 8700, + "tokens_trained": 28.500612104 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 2.4703212538117865, + "grad_norm": 0.3562406301498413, + "loss": 0.9794, + "loss_ce": 1.0864207744598389, + "loss_region": 0.0, + "loss_total": 1.0864207744598389, + "lr": 0.0007272757475083057, + "step": 8710, + "tokens_trained": 28.533377544 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 2.473157932061556, + "grad_norm": 1.139136791229248, + "loss": 0.9904, + "loss_ce": 1.0949875116348267, + "loss_region": 0.0, + "loss_total": 1.0949875116348267, + "lr": 0.0007269435215946844, + "step": 8720, + "tokens_trained": 28.566141984 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 2.4759946103113255, + "grad_norm": 0.2430262416601181, + "loss": 0.9813, + "loss_ce": 1.0925880670547485, + "loss_region": 0.0, + "loss_total": 1.0925880670547485, + "lr": 0.0007266112956810631, + "step": 8730, + "tokens_trained": 28.598907424 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 2.478831288561095, + "grad_norm": 0.5169898867607117, + "loss": 0.9881, + "loss_ce": 0.9018744826316833, + "loss_region": 0.0, + "loss_total": 0.9018744826316833, + "lr": 0.0007262790697674419, + "step": 8740, + "tokens_trained": 28.631672864 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 2.4816679668108645, + "grad_norm": 0.9234143495559692, + "loss": 0.9961, + "loss_ce": 1.0438889265060425, + "loss_region": 0.0, + "loss_total": 1.0438889265060425, + "lr": 0.0007259468438538206, + "step": 8750, + "tokens_trained": 28.664438304 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 2.484504645060634, + "grad_norm": 0.7202972173690796, + "loss": 0.9735, + "loss_ce": 1.0292013883590698, + "loss_region": 0.0, + "loss_total": 1.0292013883590698, + "lr": 0.0007256146179401993, + "step": 8760, + "tokens_trained": 28.697203744 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 2.4873413233104036, + "grad_norm": 0.41513338685035706, + "loss": 0.9821, + "loss_ce": 0.9184842109680176, + "loss_region": 0.0, + "loss_total": 0.9184842109680176, + "lr": 0.0007252823920265781, + "step": 8770, + "tokens_trained": 28.729969184 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 2.490178001560173, + "grad_norm": 0.7200778722763062, + "loss": 0.9864, + "loss_ce": 1.067702054977417, + "loss_region": 0.0, + "loss_total": 1.067702054977417, + "lr": 0.0007249501661129569, + "step": 8780, + "tokens_trained": 28.762734624 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 2.4930146798099426, + "grad_norm": 0.6141319870948792, + "loss": 0.979, + "loss_ce": 0.9702039361000061, + "loss_region": 0.0, + "loss_total": 0.9702039361000061, + "lr": 0.0007246179401993356, + "step": 8790, + "tokens_trained": 28.795500064 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 2.495851358059712, + "grad_norm": 0.2716737985610962, + "loss": 0.9825, + "loss_ce": 0.9804413318634033, + "loss_region": 0.0, + "loss_total": 0.9804413318634033, + "lr": 0.0007242857142857143, + "step": 8800, + "tokens_trained": 28.828265504 + }, + { + "epoch": 2.495851358059712, + "eval_ppl": 2.7105924565533455, + "eval_runtime": 1.4152, + "step": 8800, + "tokens_trained": 28.828265504 + }, + { + "epoch": 2.495851358059712, + "eval_ppl_cds": 3.212807101000214, + "eval_ppl_dig": 1.0996599064420658, + "eval_ppl_exon": 3.2095252900542137, + "eval_ppl_intron": 2.8711764798646655, + "eval_ppl_nig": 2.5611015259475725, + "eval_ppl_promoter": 2.772829433677822, + "eval_ppl_utr": 2.9616635181614726, + "step": 8800, + "tokens_trained": 28.828265504 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 2.4986880363094817, + "grad_norm": 0.5031347274780273, + "loss": 0.9874, + "loss_ce": 0.6967055797576904, + "loss_region": 0.0, + "loss_total": 0.6967055797576904, + "lr": 0.000723953488372093, + "step": 8810, + "tokens_trained": 28.861030944 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 2.501524714559251, + "grad_norm": 0.5797333121299744, + "loss": 0.9865, + "loss_ce": 0.821042537689209, + "loss_region": 0.0, + "loss_total": 0.821042537689209, + "lr": 0.0007236212624584719, + "step": 8820, + "tokens_trained": 28.893796384 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 2.5043613928090207, + "grad_norm": 0.22722536325454712, + "loss": 0.9875, + "loss_ce": 1.0308706760406494, + "loss_region": 0.0, + "loss_total": 1.0308706760406494, + "lr": 0.0007232890365448506, + "step": 8830, + "tokens_trained": 28.926561824 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 2.5071980710587902, + "grad_norm": 0.6802752614021301, + "loss": 0.9931, + "loss_ce": 1.0496515035629272, + "loss_region": 0.0, + "loss_total": 1.0496515035629272, + "lr": 0.0007229568106312292, + "step": 8840, + "tokens_trained": 28.959327264 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 2.5100347493085597, + "grad_norm": 0.345431387424469, + "loss": 0.975, + "loss_ce": 0.9828302264213562, + "loss_region": 0.0, + "loss_total": 0.9828302264213562, + "lr": 0.0007226245847176079, + "step": 8850, + "tokens_trained": 28.992092704 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 2.5128714275583293, + "grad_norm": 0.5801659226417542, + "loss": 0.9895, + "loss_ce": 1.07368803024292, + "loss_region": 0.0, + "loss_total": 1.07368803024292, + "lr": 0.0007222923588039867, + "step": 8860, + "tokens_trained": 29.024858144 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 2.515708105808099, + "grad_norm": 0.6442795395851135, + "loss": 0.9871, + "loss_ce": 1.1036643981933594, + "loss_region": 0.0, + "loss_total": 1.1036643981933594, + "lr": 0.0007219601328903655, + "step": 8870, + "tokens_trained": 29.057623584 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 2.5185447840578683, + "grad_norm": 0.30478641390800476, + "loss": 0.9828, + "loss_ce": 0.8910502791404724, + "loss_region": 0.0, + "loss_total": 0.8910502791404724, + "lr": 0.0007216279069767442, + "step": 8880, + "tokens_trained": 29.090389024 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 2.521381462307638, + "grad_norm": 0.36072421073913574, + "loss": 0.9763, + "loss_ce": 1.126523494720459, + "loss_region": 0.0, + "loss_total": 1.126523494720459, + "lr": 0.0007212956810631229, + "step": 8890, + "tokens_trained": 29.123153664 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 2.5242181405574073, + "grad_norm": 0.5368815660476685, + "loss": 0.9784, + "loss_ce": 1.1199312210083008, + "loss_region": 0.0, + "loss_total": 1.1199312210083008, + "lr": 0.0007209634551495017, + "step": 8900, + "tokens_trained": 29.155917496 + }, + { + "epoch": 2.5242181405574073, + "eval_ppl": 2.698653887423433, + "eval_runtime": 1.4154, + "step": 8900, + "tokens_trained": 29.155917496 + }, + { + "epoch": 2.5242181405574073, + "eval_ppl_cds": 3.1826469214693325, + "eval_ppl_dig": 1.098563639866256, + "eval_ppl_exon": 3.219944193874653, + "eval_ppl_intron": 2.8529440916430384, + "eval_ppl_nig": 2.562223828817479, + "eval_ppl_promoter": 2.723857502377258, + "eval_ppl_utr": 2.960333212369637, + "step": 8900, + "tokens_trained": 29.155917496 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 2.527054818807177, + "grad_norm": 0.5763529539108276, + "loss": 0.98, + "loss_ce": 1.1149277687072754, + "loss_region": 0.0, + "loss_total": 1.1149277687072754, + "lr": 0.0007206312292358804, + "step": 8910, + "tokens_trained": 29.188682936 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 2.5298914970569464, + "grad_norm": 0.47801846265792847, + "loss": 0.9806, + "loss_ce": 0.8484393954277039, + "loss_region": 0.0, + "loss_total": 0.8484393954277039, + "lr": 0.0007202990033222592, + "step": 8920, + "tokens_trained": 29.221448376 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 2.532728175306716, + "grad_norm": 0.49632418155670166, + "loss": 0.9841, + "loss_ce": 1.0189871788024902, + "loss_region": 0.0, + "loss_total": 1.0189871788024902, + "lr": 0.000719966777408638, + "step": 8930, + "tokens_trained": 29.254213816 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 2.5355648535564854, + "grad_norm": 0.5294733643531799, + "loss": 0.9815, + "loss_ce": 1.0181454420089722, + "loss_region": 0.0, + "loss_total": 1.0181454420089722, + "lr": 0.0007196345514950167, + "step": 8940, + "tokens_trained": 29.286979256 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 2.538401531806255, + "grad_norm": 0.40740129351615906, + "loss": 0.9703, + "loss_ce": 0.929046094417572, + "loss_region": 0.0, + "loss_total": 0.929046094417572, + "lr": 0.0007193023255813954, + "step": 8950, + "tokens_trained": 29.319744696 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 2.5412382100560245, + "grad_norm": 0.2712377607822418, + "loss": 0.9823, + "loss_ce": 1.0443341732025146, + "loss_region": 0.0, + "loss_total": 1.0443341732025146, + "lr": 0.000718970099667774, + "step": 8960, + "tokens_trained": 29.352510136 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 2.544074888305794, + "grad_norm": 0.6240447163581848, + "loss": 0.9901, + "loss_ce": 0.9727276563644409, + "loss_region": 0.0, + "loss_total": 0.9727276563644409, + "lr": 0.0007186378737541527, + "step": 8970, + "tokens_trained": 29.385275576 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 2.5469115665555635, + "grad_norm": 0.5738998651504517, + "loss": 0.9701, + "loss_ce": 1.0515128374099731, + "loss_region": 0.0, + "loss_total": 1.0515128374099731, + "lr": 0.0007183056478405316, + "step": 8980, + "tokens_trained": 29.418036064 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 2.549748244805333, + "grad_norm": 0.21969926357269287, + "loss": 0.9812, + "loss_ce": 0.8952951431274414, + "loss_region": 0.0, + "loss_total": 0.8952951431274414, + "lr": 0.0007179734219269103, + "step": 8990, + "tokens_trained": 29.450801504 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 2.5525849230551025, + "grad_norm": 0.732857346534729, + "loss": 0.986, + "loss_ce": 0.991117000579834, + "loss_region": 0.0, + "loss_total": 0.991117000579834, + "lr": 0.000717641196013289, + "step": 9000, + "tokens_trained": 29.483566944 + }, + { + "epoch": 2.5525849230551025, + "eval_ppl": 2.7023366988464246, + "eval_runtime": 1.4383, + "step": 9000, + "tokens_trained": 29.483566944 + }, + { + "epoch": 2.5525849230551025, + "eval_ppl_cds": 3.183629490043268, + "eval_ppl_dig": 1.096421837616465, + "eval_ppl_exon": 3.245302048329688, + "eval_ppl_intron": 2.854814087494366, + "eval_ppl_nig": 2.5684067780814765, + "eval_ppl_promoter": 2.741789576724992, + "eval_ppl_utr": 2.977924794396983, + "step": 9000, + "tokens_trained": 29.483566944 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 2.555421601304872, + "grad_norm": 0.7816854119300842, + "loss": 0.9855, + "loss_ce": 0.9761223793029785, + "loss_region": 0.0, + "loss_total": 0.9761223793029785, + "lr": 0.0007173089700996677, + "step": 9010, + "tokens_trained": 29.516332384 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 2.5582582795546416, + "grad_norm": 0.615638792514801, + "loss": 0.9814, + "loss_ce": 0.8662219643592834, + "loss_region": 0.0, + "loss_total": 0.8662219643592834, + "lr": 0.0007169767441860465, + "step": 9020, + "tokens_trained": 29.549097824 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 2.561094957804411, + "grad_norm": 0.8233470320701599, + "loss": 0.9867, + "loss_ce": 1.0241360664367676, + "loss_region": 0.0, + "loss_total": 1.0241360664367676, + "lr": 0.0007166445182724253, + "step": 9030, + "tokens_trained": 29.581861424 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 2.5639316360541806, + "grad_norm": 0.11227558553218842, + "loss": 0.9794, + "loss_ce": 0.9534710049629211, + "loss_region": 0.0, + "loss_total": 0.9534710049629211, + "lr": 0.000716312292358804, + "step": 9040, + "tokens_trained": 29.614626864 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 2.56676831430395, + "grad_norm": 0.5573281645774841, + "loss": 0.9753, + "loss_ce": 1.0324139595031738, + "loss_region": 0.0, + "loss_total": 1.0324139595031738, + "lr": 0.0007159800664451828, + "step": 9050, + "tokens_trained": 29.647392304 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 2.5696049925537197, + "grad_norm": 0.4533429741859436, + "loss": 0.9748, + "loss_ce": 0.869403064250946, + "loss_region": 0.0, + "loss_total": 0.869403064250946, + "lr": 0.0007156478405315615, + "step": 9060, + "tokens_trained": 29.680156944 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 2.572441670803489, + "grad_norm": 0.4390071928501129, + "loss": 0.9701, + "loss_ce": 1.0792540311813354, + "loss_region": 0.0, + "loss_total": 1.0792540311813354, + "lr": 0.0007153156146179402, + "step": 9070, + "tokens_trained": 29.712922384 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 2.5752783490532587, + "grad_norm": 0.3482957184314728, + "loss": 0.9822, + "loss_ce": 1.044465184211731, + "loss_region": 0.0, + "loss_total": 1.044465184211731, + "lr": 0.000714983388704319, + "step": 9080, + "tokens_trained": 29.745687824 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 2.578115027303028, + "grad_norm": 0.5302751064300537, + "loss": 0.9806, + "loss_ce": 0.960172712802887, + "loss_region": 0.0, + "loss_total": 0.960172712802887, + "lr": 0.0007146511627906977, + "step": 9090, + "tokens_trained": 29.778453264 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 2.5809517055527977, + "grad_norm": 0.5905822515487671, + "loss": 0.977, + "loss_ce": 1.0715593099594116, + "loss_region": 0.0, + "loss_total": 1.0715593099594116, + "lr": 0.0007143189368770764, + "step": 9100, + "tokens_trained": 29.811218704 + }, + { + "epoch": 2.5809517055527977, + "eval_ppl": 2.705026808802284, + "eval_runtime": 1.4358, + "step": 9100, + "tokens_trained": 29.811218704 + }, + { + "epoch": 2.5809517055527977, + "eval_ppl_cds": 3.1762044030291126, + "eval_ppl_dig": 1.0955061355126097, + "eval_ppl_exon": 3.248699099234216, + "eval_ppl_intron": 2.8578732986082858, + "eval_ppl_nig": 2.5591080884362705, + "eval_ppl_promoter": 2.7472209535589376, + "eval_ppl_utr": 2.9209666247504087, + "step": 9100, + "tokens_trained": 29.811218704 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 2.5837883838025673, + "grad_norm": 0.22313964366912842, + "loss": 0.9777, + "loss_ce": 0.9785915017127991, + "loss_region": 0.0, + "loss_total": 0.9785915017127991, + "lr": 0.0007139867109634551, + "step": 9110, + "tokens_trained": 29.843984144 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 2.5866250620523368, + "grad_norm": 0.46888267993927, + "loss": 0.9803, + "loss_ce": 0.9646623134613037, + "loss_region": 0.0, + "loss_total": 0.9646623134613037, + "lr": 0.0007136544850498338, + "step": 9120, + "tokens_trained": 29.876749584 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 2.5894617403021063, + "grad_norm": 0.7590495944023132, + "loss": 0.9786, + "loss_ce": 0.9680768847465515, + "loss_region": 0.0, + "loss_total": 0.9680768847465515, + "lr": 0.0007133222591362127, + "step": 9130, + "tokens_trained": 29.909511696 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 2.592298418551876, + "grad_norm": 0.4459081292152405, + "loss": 0.9809, + "loss_ce": 0.9997375011444092, + "loss_region": 0.0, + "loss_total": 0.9997375011444092, + "lr": 0.0007129900332225914, + "step": 9140, + "tokens_trained": 29.942277136 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 2.5951350968016453, + "grad_norm": 0.2955224812030792, + "loss": 0.9834, + "loss_ce": 1.075130820274353, + "loss_region": 0.0, + "loss_total": 1.075130820274353, + "lr": 0.0007126578073089701, + "step": 9150, + "tokens_trained": 29.975041776 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 2.597971775051415, + "grad_norm": 0.5466042757034302, + "loss": 0.9801, + "loss_ce": 1.0575356483459473, + "loss_region": 0.0, + "loss_total": 1.0575356483459473, + "lr": 0.0007123255813953488, + "step": 9160, + "tokens_trained": 30.007807216 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 2.6008084533011844, + "grad_norm": 0.571448028087616, + "loss": 0.9774, + "loss_ce": 1.0882505178451538, + "loss_region": 0.0, + "loss_total": 1.0882505178451538, + "lr": 0.0007119933554817276, + "step": 9170, + "tokens_trained": 30.040572656 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 2.603645131550954, + "grad_norm": 0.24392850697040558, + "loss": 0.9719, + "loss_ce": 1.0596762895584106, + "loss_region": 0.0, + "loss_total": 1.0596762895584106, + "lr": 0.0007116611295681064, + "step": 9180, + "tokens_trained": 30.073338096 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 2.6064818098007234, + "grad_norm": 0.6270098090171814, + "loss": 0.9792, + "loss_ce": 0.877307653427124, + "loss_region": 0.0, + "loss_total": 0.877307653427124, + "lr": 0.0007113289036544851, + "step": 9190, + "tokens_trained": 30.106103536 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 2.609318488050493, + "grad_norm": 0.47577545046806335, + "loss": 0.9762, + "loss_ce": 1.055944561958313, + "loss_region": 0.0, + "loss_total": 1.055944561958313, + "lr": 0.0007109966777408638, + "step": 9200, + "tokens_trained": 30.138868248 + }, + { + "epoch": 2.609318488050493, + "eval_ppl": 2.692761393285462, + "eval_runtime": 1.4216, + "step": 9200, + "tokens_trained": 30.138868248 + }, + { + "epoch": 2.609318488050493, + "eval_ppl_cds": 3.14027474031236, + "eval_ppl_dig": 1.099149521838848, + "eval_ppl_exon": 3.210568457058113, + "eval_ppl_intron": 2.8443774192943865, + "eval_ppl_nig": 2.548816915148588, + "eval_ppl_promoter": 2.6918392614015567, + "eval_ppl_utr": 2.8653053317236705, + "step": 9200, + "tokens_trained": 30.138868248 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 2.6121551663002625, + "grad_norm": 0.5589614510536194, + "loss": 0.9787, + "loss_ce": 0.9909216165542603, + "loss_region": 0.0, + "loss_total": 0.9909216165542603, + "lr": 0.0007106644518272425, + "step": 9210, + "tokens_trained": 30.171632888 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 2.614991844550032, + "grad_norm": 0.3331902325153351, + "loss": 0.9775, + "loss_ce": 1.007265329360962, + "loss_region": 0.0, + "loss_total": 1.007265329360962, + "lr": 0.0007103322259136212, + "step": 9220, + "tokens_trained": 30.204398328 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 2.6178285227998015, + "grad_norm": 0.564941942691803, + "loss": 0.9903, + "loss_ce": 0.8603456616401672, + "loss_region": 0.0, + "loss_total": 0.8603456616401672, + "lr": 0.00071, + "step": 9230, + "tokens_trained": 30.237163608 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 2.620665201049571, + "grad_norm": 0.26822522282600403, + "loss": 0.973, + "loss_ce": 0.9406780004501343, + "loss_region": 0.0, + "loss_total": 0.9406780004501343, + "lr": 0.0007096677740863787, + "step": 9240, + "tokens_trained": 30.269929048 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 2.6235018792993405, + "grad_norm": 0.13524872064590454, + "loss": 0.9804, + "loss_ce": 1.0971295833587646, + "loss_region": 0.0, + "loss_total": 1.0971295833587646, + "lr": 0.0007093355481727575, + "step": 9250, + "tokens_trained": 30.302694488 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 2.62633855754911, + "grad_norm": 0.4664212465286255, + "loss": 0.9773, + "loss_ce": 1.0394096374511719, + "loss_region": 0.0, + "loss_total": 1.0394096374511719, + "lr": 0.0007090033222591362, + "step": 9260, + "tokens_trained": 30.335459928 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 2.6291752357988796, + "grad_norm": 0.6285772323608398, + "loss": 0.9788, + "loss_ce": 1.0686041116714478, + "loss_region": 0.0, + "loss_total": 1.0686041116714478, + "lr": 0.0007086710963455149, + "step": 9270, + "tokens_trained": 30.368225368 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 2.632011914048649, + "grad_norm": 0.35884857177734375, + "loss": 0.978, + "loss_ce": 1.110680341720581, + "loss_region": 0.0, + "loss_total": 1.110680341720581, + "lr": 0.0007083388704318938, + "step": 9280, + "tokens_trained": 30.400990008 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 2.6348485922984186, + "grad_norm": 0.43512535095214844, + "loss": 0.9793, + "loss_ce": 0.9251062870025635, + "loss_region": 0.0, + "loss_total": 0.9251062870025635, + "lr": 0.0007080066445182725, + "step": 9290, + "tokens_trained": 30.433755448 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 2.637685270548188, + "grad_norm": 0.5349493026733398, + "loss": 0.9749, + "loss_ce": 1.0520955324172974, + "loss_region": 0.0, + "loss_total": 1.0520955324172974, + "lr": 0.0007076744186046512, + "step": 9300, + "tokens_trained": 30.466520888 + }, + { + "epoch": 2.637685270548188, + "eval_ppl": 2.6924842221752585, + "eval_runtime": 1.4198, + "step": 9300, + "tokens_trained": 30.466520888 + }, + { + "epoch": 2.637685270548188, + "eval_ppl_cds": 3.1483320584654484, + "eval_ppl_dig": 1.0989434219468293, + "eval_ppl_exon": 3.1920188570174597, + "eval_ppl_intron": 2.843277100119342, + "eval_ppl_nig": 2.5450647671922155, + "eval_ppl_promoter": 2.7029581774207188, + "eval_ppl_utr": 2.8917651499537604, + "step": 9300, + "tokens_trained": 30.466520888 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 2.6405219487979577, + "grad_norm": 0.28184249997138977, + "loss": 0.9739, + "loss_ce": 0.8384683728218079, + "loss_region": 0.0, + "loss_total": 0.8384683728218079, + "lr": 0.0007073421926910299, + "step": 9310, + "tokens_trained": 30.499286328 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 2.643358627047727, + "grad_norm": 0.8277426958084106, + "loss": 0.9862, + "loss_ce": 0.9730556607246399, + "loss_region": 0.0, + "loss_total": 0.9730556607246399, + "lr": 0.0007070099667774087, + "step": 9320, + "tokens_trained": 30.532051768 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 2.6461953052974967, + "grad_norm": 0.53327476978302, + "loss": 0.98, + "loss_ce": 1.0288070440292358, + "loss_region": 0.0, + "loss_total": 1.0288070440292358, + "lr": 0.0007066777408637875, + "step": 9330, + "tokens_trained": 30.564817208 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 2.649031983547266, + "grad_norm": 0.5355494022369385, + "loss": 0.9802, + "loss_ce": 0.9263233542442322, + "loss_region": 0.0, + "loss_total": 0.9263233542442322, + "lr": 0.0007063455149501661, + "step": 9340, + "tokens_trained": 30.597582648 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 2.6518686617970357, + "grad_norm": 0.7099814414978027, + "loss": 0.9768, + "loss_ce": 0.9702062606811523, + "loss_region": 0.0, + "loss_total": 0.9702062606811523, + "lr": 0.0007060132890365448, + "step": 9350, + "tokens_trained": 30.630342424 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 2.6547053400468053, + "grad_norm": 0.6038627624511719, + "loss": 0.9803, + "loss_ce": 0.9583320021629333, + "loss_region": 0.0, + "loss_total": 0.9583320021629333, + "lr": 0.0007056810631229235, + "step": 9360, + "tokens_trained": 30.663107864 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 2.6575420182965748, + "grad_norm": 0.3464861810207367, + "loss": 0.9712, + "loss_ce": 0.8309147953987122, + "loss_region": 0.0, + "loss_total": 0.8309147953987122, + "lr": 0.0007053488372093023, + "step": 9370, + "tokens_trained": 30.695873304 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 2.6603786965463443, + "grad_norm": 0.7632194757461548, + "loss": 0.9845, + "loss_ce": 0.9533660411834717, + "loss_region": 0.0, + "loss_total": 0.9533660411834717, + "lr": 0.0007050166112956811, + "step": 9380, + "tokens_trained": 30.728638744 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 2.663215374796114, + "grad_norm": 0.4814608097076416, + "loss": 0.9756, + "loss_ce": 0.9621901512145996, + "loss_region": 0.0, + "loss_total": 0.9621901512145996, + "lr": 0.0007046843853820598, + "step": 9390, + "tokens_trained": 30.761404184 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 2.6660520530458833, + "grad_norm": 0.7259153723716736, + "loss": 0.9804, + "loss_ce": 1.0596530437469482, + "loss_region": 0.0, + "loss_total": 1.0596530437469482, + "lr": 0.0007043521594684386, + "step": 9400, + "tokens_trained": 30.794167 + }, + { + "epoch": 2.6660520530458833, + "eval_ppl": 2.6882993121287684, + "eval_runtime": 1.4225, + "step": 9400, + "tokens_trained": 30.794167 + }, + { + "epoch": 2.6660520530458833, + "eval_ppl_cds": 3.1606309967764448, + "eval_ppl_dig": 1.0991153578632042, + "eval_ppl_exon": 3.183608010573729, + "eval_ppl_intron": 2.848181698995431, + "eval_ppl_nig": 2.552747173318592, + "eval_ppl_promoter": 2.724502606064903, + "eval_ppl_utr": 2.9305105137049385, + "step": 9400, + "tokens_trained": 30.794167 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 2.668888731295653, + "grad_norm": 0.29261067509651184, + "loss": 0.9698, + "loss_ce": 0.9443883299827576, + "loss_region": 0.0, + "loss_total": 0.9443883299827576, + "lr": 0.0007040199335548173, + "step": 9410, + "tokens_trained": 30.82693244 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 2.6717254095454224, + "grad_norm": 0.4178066551685333, + "loss": 0.9781, + "loss_ce": 0.8742818236351013, + "loss_region": 0.0, + "loss_total": 0.8742818236351013, + "lr": 0.000703687707641196, + "step": 9420, + "tokens_trained": 30.85969788 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 2.674562087795192, + "grad_norm": 1.2577335834503174, + "loss": 0.9835, + "loss_ce": 0.9298757910728455, + "loss_region": 0.0, + "loss_total": 0.9298757910728455, + "lr": 0.0007033554817275748, + "step": 9430, + "tokens_trained": 30.89246252 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 2.6773987660449614, + "grad_norm": 0.7232295274734497, + "loss": 0.9811, + "loss_ce": 0.9959359169006348, + "loss_region": 0.0, + "loss_total": 0.9959359169006348, + "lr": 0.0007030232558139536, + "step": 9440, + "tokens_trained": 30.92522796 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 2.680235444294731, + "grad_norm": 0.23632271587848663, + "loss": 0.9785, + "loss_ce": 0.8528959155082703, + "loss_region": 0.0, + "loss_total": 0.8528959155082703, + "lr": 0.0007026910299003323, + "step": 9450, + "tokens_trained": 30.9579934 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 2.6830721225445004, + "grad_norm": 0.6304146647453308, + "loss": 0.981, + "loss_ce": 0.973279595375061, + "loss_region": 0.0, + "loss_total": 0.973279595375061, + "lr": 0.0007023588039867109, + "step": 9460, + "tokens_trained": 30.99075884 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 2.68590880079427, + "grad_norm": 0.3138343393802643, + "loss": 0.9839, + "loss_ce": 1.035901665687561, + "loss_region": 0.0, + "loss_total": 1.035901665687561, + "lr": 0.0007020265780730896, + "step": 9470, + "tokens_trained": 31.023521856 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 2.6887454790440395, + "grad_norm": 0.6773609519004822, + "loss": 0.9694, + "loss_ce": 0.8665898442268372, + "loss_region": 0.0, + "loss_total": 0.8665898442268372, + "lr": 0.0007016943521594685, + "step": 9480, + "tokens_trained": 31.056286496 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 2.691582157293809, + "grad_norm": 0.6268877387046814, + "loss": 0.9753, + "loss_ce": 1.1145713329315186, + "loss_region": 0.0, + "loss_total": 1.1145713329315186, + "lr": 0.0007013621262458472, + "step": 9490, + "tokens_trained": 31.089051936 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 2.6944188355435785, + "grad_norm": 0.19423076510429382, + "loss": 0.9744, + "loss_ce": 1.0642563104629517, + "loss_region": 0.0, + "loss_total": 1.0642563104629517, + "lr": 0.0007010299003322259, + "step": 9500, + "tokens_trained": 31.121814016 + }, + { + "epoch": 2.6944188355435785, + "eval_ppl": 2.687520682200247, + "eval_runtime": 1.4204, + "step": 9500, + "tokens_trained": 31.121814016 + }, + { + "epoch": 2.6944188355435785, + "eval_ppl_cds": 3.174116889514191, + "eval_ppl_dig": 1.1013579847269759, + "eval_ppl_exon": 3.192533291810655, + "eval_ppl_intron": 2.847678101289237, + "eval_ppl_nig": 2.558837384577747, + "eval_ppl_promoter": 2.73530080715119, + "eval_ppl_utr": 2.9570194166422827, + "step": 9500, + "tokens_trained": 31.121814016 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 2.697255513793348, + "grad_norm": 0.31077682971954346, + "loss": 0.98, + "loss_ce": 1.0547945499420166, + "loss_region": 0.0, + "loss_total": 1.0547945499420166, + "lr": 0.0007006976744186046, + "step": 9510, + "tokens_trained": 31.154577856 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 2.7000921920431176, + "grad_norm": 0.47652798891067505, + "loss": 0.976, + "loss_ce": 0.9187541604042053, + "loss_region": 0.0, + "loss_total": 0.9187541604042053, + "lr": 0.0007003654485049834, + "step": 9520, + "tokens_trained": 31.187343296 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 2.702928870292887, + "grad_norm": 0.46409115195274353, + "loss": 0.9763, + "loss_ce": 0.9762084484100342, + "loss_region": 0.0, + "loss_total": 0.9762084484100342, + "lr": 0.0007000332225913622, + "step": 9530, + "tokens_trained": 31.220108736 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 2.7057655485426566, + "grad_norm": 0.4413476288318634, + "loss": 0.9781, + "loss_ce": 1.0225728750228882, + "loss_region": 0.0, + "loss_total": 1.0225728750228882, + "lr": 0.0006997009966777409, + "step": 9540, + "tokens_trained": 31.252874176 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 2.708602226792426, + "grad_norm": 0.7439092397689819, + "loss": 0.9739, + "loss_ce": 1.0426510572433472, + "loss_region": 0.0, + "loss_total": 1.0426510572433472, + "lr": 0.0006993687707641196, + "step": 9550, + "tokens_trained": 31.285639616 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 2.7114389050421956, + "grad_norm": 0.3502562940120697, + "loss": 0.9784, + "loss_ce": 0.9957247376441956, + "loss_region": 0.0, + "loss_total": 0.9957247376441956, + "lr": 0.0006990365448504984, + "step": 9560, + "tokens_trained": 31.318405056 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 2.714275583291965, + "grad_norm": 0.6744600534439087, + "loss": 0.9799, + "loss_ce": 1.048980712890625, + "loss_region": 0.0, + "loss_total": 1.048980712890625, + "lr": 0.0006987043189368771, + "step": 9570, + "tokens_trained": 31.351170496 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 2.7171122615417347, + "grad_norm": 0.43942025303840637, + "loss": 0.9742, + "loss_ce": 0.8412730693817139, + "loss_region": 0.0, + "loss_total": 0.8412730693817139, + "lr": 0.0006983720930232559, + "step": 9580, + "tokens_trained": 31.383935936 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 2.719948939791504, + "grad_norm": 0.21825721859931946, + "loss": 0.9817, + "loss_ce": 1.0638011693954468, + "loss_region": 0.0, + "loss_total": 1.0638011693954468, + "lr": 0.0006980398671096345, + "step": 9590, + "tokens_trained": 31.41669796 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 2.7227856180412737, + "grad_norm": 0.5426555275917053, + "loss": 0.9718, + "loss_ce": 1.0312381982803345, + "loss_region": 0.0, + "loss_total": 1.0312381982803345, + "lr": 0.0006977076411960133, + "step": 9600, + "tokens_trained": 31.4494626 + }, + { + "epoch": 2.7227856180412737, + "eval_ppl": 2.6875996563973588, + "eval_runtime": 1.4173, + "step": 9600, + "tokens_trained": 31.4494626 + }, + { + "epoch": 2.7227856180412737, + "eval_ppl_cds": 3.180167187323816, + "eval_ppl_dig": 1.0971699225033706, + "eval_ppl_exon": 3.217577384038602, + "eval_ppl_intron": 2.855248119757852, + "eval_ppl_nig": 2.5574452955098486, + "eval_ppl_promoter": 2.7442666894288554, + "eval_ppl_utr": 2.9229413122777967, + "step": 9600, + "tokens_trained": 31.4494626 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 2.7256222962910432, + "grad_norm": 0.5038434863090515, + "loss": 0.9707, + "loss_ce": 0.9080817699432373, + "loss_region": 0.0, + "loss_total": 0.9080817699432373, + "lr": 0.000697375415282392, + "step": 9610, + "tokens_trained": 31.48222804 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 2.7284589745408128, + "grad_norm": 0.3576555550098419, + "loss": 0.9687, + "loss_ce": 1.084140419960022, + "loss_region": 0.0, + "loss_total": 1.084140419960022, + "lr": 0.0006970431893687707, + "step": 9620, + "tokens_trained": 31.51499348 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 2.7312956527905823, + "grad_norm": 0.5124894976615906, + "loss": 0.9757, + "loss_ce": 1.0853725671768188, + "loss_region": 0.0, + "loss_total": 1.0853725671768188, + "lr": 0.0006967109634551496, + "step": 9630, + "tokens_trained": 31.54775892 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 2.734132331040352, + "grad_norm": 0.3995230495929718, + "loss": 0.9858, + "loss_ce": 1.1460939645767212, + "loss_region": 0.0, + "loss_total": 1.1460939645767212, + "lr": 0.0006963787375415283, + "step": 9640, + "tokens_trained": 31.58052436 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 2.7369690092901213, + "grad_norm": 0.4509827196598053, + "loss": 0.9725, + "loss_ce": 1.0097097158432007, + "loss_region": 0.0, + "loss_total": 1.0097097158432007, + "lr": 0.000696046511627907, + "step": 9650, + "tokens_trained": 31.6132898 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 2.739805687539891, + "grad_norm": 0.35714709758758545, + "loss": 0.975, + "loss_ce": 1.0049821138381958, + "loss_region": 0.0, + "loss_total": 1.0049821138381958, + "lr": 0.0006957142857142857, + "step": 9660, + "tokens_trained": 31.646054392 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 2.7426423657896604, + "grad_norm": 0.5600670576095581, + "loss": 0.9744, + "loss_ce": 1.028450608253479, + "loss_region": 0.0, + "loss_total": 1.028450608253479, + "lr": 0.0006953820598006645, + "step": 9670, + "tokens_trained": 31.678819032 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 2.74547904403943, + "grad_norm": 0.2500137686729431, + "loss": 0.9812, + "loss_ce": 1.0120899677276611, + "loss_region": 0.0, + "loss_total": 1.0120899677276611, + "lr": 0.0006950498338870433, + "step": 9680, + "tokens_trained": 31.711584472 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 2.7483157222891994, + "grad_norm": 0.403161883354187, + "loss": 0.979, + "loss_ce": 0.9016726016998291, + "loss_region": 0.0, + "loss_total": 0.9016726016998291, + "lr": 0.000694717607973422, + "step": 9690, + "tokens_trained": 31.744349912 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 2.751152400538969, + "grad_norm": 0.6087920069694519, + "loss": 0.9772, + "loss_ce": 0.9027508497238159, + "loss_region": 0.0, + "loss_total": 0.9027508497238159, + "lr": 0.0006943853820598007, + "step": 9700, + "tokens_trained": 31.777114552 + }, + { + "epoch": 2.751152400538969, + "eval_ppl": 2.6909213957608173, + "eval_runtime": 1.426, + "step": 9700, + "tokens_trained": 31.777114552 + }, + { + "epoch": 2.751152400538969, + "eval_ppl_cds": 3.131436743485574, + "eval_ppl_dig": 1.0988436353530977, + "eval_ppl_exon": 3.1920788102108317, + "eval_ppl_intron": 2.8508236185669187, + "eval_ppl_nig": 2.564471957085892, + "eval_ppl_promoter": 2.7179774202245013, + "eval_ppl_utr": 2.9280073675204745, + "step": 9700, + "tokens_trained": 31.777114552 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 2.7539890787887384, + "grad_norm": 0.5997163653373718, + "loss": 0.9655, + "loss_ce": 0.9887319207191467, + "loss_region": 0.0, + "loss_total": 0.9887319207191467, + "lr": 0.0006940531561461794, + "step": 9710, + "tokens_trained": 31.809879992 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 2.756825757038508, + "grad_norm": 0.4034530520439148, + "loss": 0.975, + "loss_ce": 1.0225152969360352, + "loss_region": 0.0, + "loss_total": 1.0225152969360352, + "lr": 0.0006937209302325581, + "step": 9720, + "tokens_trained": 31.842645432 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 2.7596624352882775, + "grad_norm": 0.2323325127363205, + "loss": 0.9736, + "loss_ce": 1.016234278678894, + "loss_region": 0.0, + "loss_total": 1.016234278678894, + "lr": 0.0006933887043189369, + "step": 9730, + "tokens_trained": 31.875410872 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 2.762499113538047, + "grad_norm": 0.7340709567070007, + "loss": 0.9839, + "loss_ce": 0.8361617922782898, + "loss_region": 0.0, + "loss_total": 0.8361617922782898, + "lr": 0.0006930564784053156, + "step": 9740, + "tokens_trained": 31.908176312 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 2.7653357917878165, + "grad_norm": 0.4039103090763092, + "loss": 0.9783, + "loss_ce": 0.9872311353683472, + "loss_region": 0.0, + "loss_total": 0.9872311353683472, + "lr": 0.0006927242524916944, + "step": 9750, + "tokens_trained": 31.940939304 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 2.768172470037586, + "grad_norm": 0.6232449412345886, + "loss": 0.9645, + "loss_ce": 0.9803999066352844, + "loss_region": 0.0, + "loss_total": 0.9803999066352844, + "lr": 0.0006923920265780731, + "step": 9760, + "tokens_trained": 31.973704744 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 2.7710091482873556, + "grad_norm": 0.2904205918312073, + "loss": 0.9769, + "loss_ce": 0.9185511469841003, + "loss_region": 0.0, + "loss_total": 0.9185511469841003, + "lr": 0.0006920598006644518, + "step": 9770, + "tokens_trained": 32.006470184 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 2.773845826537125, + "grad_norm": 0.264252245426178, + "loss": 0.9814, + "loss_ce": 0.8619754910469055, + "loss_region": 0.0, + "loss_total": 0.8619754910469055, + "lr": 0.0006917275747508306, + "step": 9780, + "tokens_trained": 32.039235624 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 2.7766825047868946, + "grad_norm": 0.29041656851768494, + "loss": 0.9853, + "loss_ce": 0.606002151966095, + "loss_region": 0.0, + "loss_total": 0.606002151966095, + "lr": 0.0006913953488372094, + "step": 9790, + "tokens_trained": 32.071999472 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 2.779519183036664, + "grad_norm": 0.40187138319015503, + "loss": 0.9646, + "loss_ce": 0.7136797904968262, + "loss_region": 0.0, + "loss_total": 0.7136797904968262, + "lr": 0.0006910631229235881, + "step": 9800, + "tokens_trained": 32.104764112 + }, + { + "epoch": 2.779519183036664, + "eval_ppl": 2.679721379518401, + "eval_runtime": 1.4303, + "step": 9800, + "tokens_trained": 32.104764112 + }, + { + "epoch": 2.779519183036664, + "eval_ppl_cds": 3.133120327376721, + "eval_ppl_dig": 1.0954998142628052, + "eval_ppl_exon": 3.1714094826749477, + "eval_ppl_intron": 2.844246531112501, + "eval_ppl_nig": 2.552784998886415, + "eval_ppl_promoter": 2.6969957736615724, + "eval_ppl_utr": 2.9059909676859363, + "step": 9800, + "tokens_trained": 32.104764112 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 2.7823558612864336, + "grad_norm": 0.4792974591255188, + "loss": 0.9636, + "loss_ce": 0.9207782745361328, + "loss_region": 0.0, + "loss_total": 0.9207782745361328, + "lr": 0.0006907308970099668, + "step": 9810, + "tokens_trained": 32.137528752 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 2.785192539536203, + "grad_norm": 0.703385055065155, + "loss": 0.9691, + "loss_ce": 0.8748564720153809, + "loss_region": 0.0, + "loss_total": 0.8748564720153809, + "lr": 0.0006903986710963455, + "step": 9820, + "tokens_trained": 32.170294192 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 2.7880292177859727, + "grad_norm": 0.34845930337905884, + "loss": 0.9692, + "loss_ce": 1.0528037548065186, + "loss_region": 0.0, + "loss_total": 1.0528037548065186, + "lr": 0.0006900664451827243, + "step": 9830, + "tokens_trained": 32.203055736 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 2.790865896035742, + "grad_norm": 0.4675162434577942, + "loss": 0.9791, + "loss_ce": 0.8465257883071899, + "loss_region": 0.0, + "loss_total": 0.8465257883071899, + "lr": 0.000689734219269103, + "step": 9840, + "tokens_trained": 32.235821176 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 2.7937025742855117, + "grad_norm": 0.21671399474143982, + "loss": 0.9801, + "loss_ce": 0.8629037141799927, + "loss_region": 0.0, + "loss_total": 0.8629037141799927, + "lr": 0.0006894019933554817, + "step": 9850, + "tokens_trained": 32.268585856 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 2.7965392525352812, + "grad_norm": 0.7356233596801758, + "loss": 0.9737, + "loss_ce": 0.874710738658905, + "loss_region": 0.0, + "loss_total": 0.874710738658905, + "lr": 0.0006890697674418604, + "step": 9860, + "tokens_trained": 32.301351296 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 2.7993759307850508, + "grad_norm": 0.5242382884025574, + "loss": 0.9842, + "loss_ce": 1.0356239080429077, + "loss_region": 0.0, + "loss_total": 1.0356239080429077, + "lr": 0.0006887375415282392, + "step": 9870, + "tokens_trained": 32.334115936 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 2.8022126090348203, + "grad_norm": 0.698110044002533, + "loss": 0.9725, + "loss_ce": 0.9825286865234375, + "loss_region": 0.0, + "loss_total": 0.9825286865234375, + "lr": 0.000688405315614618, + "step": 9880, + "tokens_trained": 32.366879776 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 2.80504928728459, + "grad_norm": 0.4742681086063385, + "loss": 0.969, + "loss_ce": 0.9589372873306274, + "loss_region": 0.0, + "loss_total": 0.9589372873306274, + "lr": 0.0006880730897009967, + "step": 9890, + "tokens_trained": 32.399645216 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 2.8078859655343593, + "grad_norm": 0.28677329421043396, + "loss": 0.9703, + "loss_ce": 0.7044788599014282, + "loss_region": 0.0, + "loss_total": 0.7044788599014282, + "lr": 0.0006877408637873754, + "step": 9900, + "tokens_trained": 32.432410656 + }, + { + "epoch": 2.8078859655343593, + "eval_ppl": 2.6766956832941218, + "eval_runtime": 1.4188, + "step": 9900, + "tokens_trained": 32.432410656 + }, + { + "epoch": 2.8078859655343593, + "eval_ppl_cds": 3.158302620850563, + "eval_ppl_dig": 1.0992373005550335, + "eval_ppl_exon": 3.11422579621016, + "eval_ppl_intron": 2.844787226710901, + "eval_ppl_nig": 2.5567131137265253, + "eval_ppl_promoter": 2.703905748880135, + "eval_ppl_utr": 2.9228059033799245, + "step": 9900, + "tokens_trained": 32.432410656 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 2.810722643784129, + "grad_norm": 0.34656521677970886, + "loss": 0.9748, + "loss_ce": 1.0430089235305786, + "loss_region": 0.0, + "loss_total": 1.0430089235305786, + "lr": 0.0006874086378737542, + "step": 9910, + "tokens_trained": 32.465176096 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 2.8135593220338984, + "grad_norm": 0.3758755922317505, + "loss": 0.9702, + "loss_ce": 1.0023715496063232, + "loss_region": 0.0, + "loss_total": 1.0023715496063232, + "lr": 0.0006870764119601329, + "step": 9920, + "tokens_trained": 32.497940736 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 2.816396000283668, + "grad_norm": 0.430412620306015, + "loss": 0.9707, + "loss_ce": 0.9781486988067627, + "loss_region": 0.0, + "loss_total": 0.9781486988067627, + "lr": 0.0006867441860465117, + "step": 9930, + "tokens_trained": 32.530706176 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 2.8192326785334374, + "grad_norm": 0.5404825210571289, + "loss": 0.9722, + "loss_ce": 0.9729744791984558, + "loss_region": 0.0, + "loss_total": 0.9729744791984558, + "lr": 0.0006864119601328905, + "step": 9940, + "tokens_trained": 32.563471616 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 2.822069356783207, + "grad_norm": 0.8061553835868835, + "loss": 0.9689, + "loss_ce": 1.0052988529205322, + "loss_region": 0.0, + "loss_total": 1.0052988529205322, + "lr": 0.0006860797342192692, + "step": 9950, + "tokens_trained": 32.596237056 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 2.8249060350329764, + "grad_norm": 0.5862414240837097, + "loss": 0.9652, + "loss_ce": 1.0880645513534546, + "loss_region": 0.0, + "loss_total": 1.0880645513534546, + "lr": 0.0006857475083056478, + "step": 9960, + "tokens_trained": 32.629002496 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 2.827742713282746, + "grad_norm": 0.18819700181484222, + "loss": 0.9737, + "loss_ce": 0.98984295129776, + "loss_region": 0.0, + "loss_total": 0.98984295129776, + "lr": 0.0006854152823920265, + "step": 9970, + "tokens_trained": 32.661767936 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 2.8305793915325155, + "grad_norm": 0.28081241250038147, + "loss": 0.9667, + "loss_ce": 0.9262552857398987, + "loss_region": 0.0, + "loss_total": 0.9262552857398987, + "lr": 0.0006850830564784052, + "step": 9980, + "tokens_trained": 32.694533376 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 2.833416069782285, + "grad_norm": 0.5774762034416199, + "loss": 0.9716, + "loss_ce": 1.0517221689224243, + "loss_region": 0.0, + "loss_total": 1.0517221689224243, + "lr": 0.0006847508305647841, + "step": 9990, + "tokens_trained": 32.727298816 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 2.8362527480320545, + "grad_norm": 0.5392642021179199, + "loss": 0.9739, + "loss_ce": 0.9199241995811462, + "loss_region": 0.0, + "loss_total": 0.9199241995811462, + "lr": 0.0006844186046511628, + "step": 10000, + "tokens_trained": 32.760063936 + }, + { + "epoch": 2.8362527480320545, + "eval_ppl": 2.6745911750806437, + "eval_runtime": 1.4161, + "step": 10000, + "tokens_trained": 32.760063936 + }, + { + "epoch": 2.8362527480320545, + "eval_ppl_cds": 3.1572251645538882, + "eval_ppl_dig": 1.0998734312528822, + "eval_ppl_exon": 3.1641416234888786, + "eval_ppl_intron": 2.842673704129833, + "eval_ppl_nig": 2.549722507592597, + "eval_ppl_promoter": 2.702699259643083, + "eval_ppl_utr": 2.9357384855402073, + "step": 10000, + "tokens_trained": 32.760063936 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 2.839089426281824, + "grad_norm": 0.5304338932037354, + "loss": 0.9696, + "loss_ce": 1.052014946937561, + "loss_region": 0.0, + "loss_total": 1.052014946937561, + "lr": 0.0006840863787375415, + "step": 10010, + "tokens_trained": 32.792829376 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 2.8419261045315936, + "grad_norm": 0.6559067368507385, + "loss": 0.9756, + "loss_ce": 1.0114803314208984, + "loss_region": 0.0, + "loss_total": 1.0114803314208984, + "lr": 0.0006837541528239203, + "step": 10020, + "tokens_trained": 32.825594792 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 2.844762782781363, + "grad_norm": 0.4310396611690521, + "loss": 0.9683, + "loss_ce": 0.8235009908676147, + "loss_region": 0.0, + "loss_total": 0.8235009908676147, + "lr": 0.000683421926910299, + "step": 10030, + "tokens_trained": 32.858360232 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 2.8475994610311326, + "grad_norm": 0.6877245306968689, + "loss": 0.9765, + "loss_ce": 1.0717467069625854, + "loss_region": 0.0, + "loss_total": 1.0717467069625854, + "lr": 0.0006830897009966778, + "step": 10040, + "tokens_trained": 32.89112352 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 2.850436139280902, + "grad_norm": 0.21539609134197235, + "loss": 0.972, + "loss_ce": 1.0462064743041992, + "loss_region": 0.0, + "loss_total": 1.0462064743041992, + "lr": 0.0006827574750830565, + "step": 10050, + "tokens_trained": 32.92388896 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 2.8532728175306716, + "grad_norm": 1.1161195039749146, + "loss": 0.9811, + "loss_ce": 1.101271629333496, + "loss_region": 0.0, + "loss_total": 1.101271629333496, + "lr": 0.0006824252491694353, + "step": 10060, + "tokens_trained": 32.9566544 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 2.856109495780441, + "grad_norm": 0.47272589802742004, + "loss": 0.9712, + "loss_ce": 1.155585527420044, + "loss_region": 0.0, + "loss_total": 1.155585527420044, + "lr": 0.000682093023255814, + "step": 10070, + "tokens_trained": 32.98941984 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 2.8589461740302107, + "grad_norm": 0.41197991371154785, + "loss": 0.9691, + "loss_ce": 0.9902181029319763, + "loss_region": 0.0, + "loss_total": 0.9902181029319763, + "lr": 0.0006817607973421926, + "step": 10080, + "tokens_trained": 33.02218368 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 2.86178285227998, + "grad_norm": 0.23536767065525055, + "loss": 0.9646, + "loss_ce": 1.0358691215515137, + "loss_region": 0.0, + "loss_total": 1.0358691215515137, + "lr": 0.0006814285714285714, + "step": 10090, + "tokens_trained": 33.05494816 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 2.8646195305297497, + "grad_norm": 0.45352667570114136, + "loss": 0.9654, + "loss_ce": 0.9513305425643921, + "loss_region": 0.0, + "loss_total": 0.9513305425643921, + "lr": 0.0006810963455149502, + "step": 10100, + "tokens_trained": 33.0877128 + }, + { + "epoch": 2.8646195305297497, + "eval_ppl": 2.671029280054436, + "eval_runtime": 1.4157, + "step": 10100, + "tokens_trained": 33.0877128 + }, + { + "epoch": 2.8646195305297497, + "eval_ppl_cds": 3.1451981770180613, + "eval_ppl_dig": 1.1014287011966604, + "eval_ppl_exon": 3.1913012217227075, + "eval_ppl_intron": 2.825108548013478, + "eval_ppl_nig": 2.539974423508268, + "eval_ppl_promoter": 2.664604753967414, + "eval_ppl_utr": 2.9001518599403004, + "step": 10100, + "tokens_trained": 33.0877128 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 2.8674562087795192, + "grad_norm": 0.36726754903793335, + "loss": 0.9741, + "loss_ce": 0.9453221559524536, + "loss_region": 0.0, + "loss_total": 0.9453221559524536, + "lr": 0.0006807641196013289, + "step": 10110, + "tokens_trained": 33.12047824 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 2.8702928870292888, + "grad_norm": 0.3920325040817261, + "loss": 0.9716, + "loss_ce": 0.9319329261779785, + "loss_region": 0.0, + "loss_total": 0.9319329261779785, + "lr": 0.0006804318936877076, + "step": 10120, + "tokens_trained": 33.15324368 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 2.8731295652790583, + "grad_norm": 0.23158857226371765, + "loss": 0.9622, + "loss_ce": 0.9810049533843994, + "loss_region": 0.0, + "loss_total": 0.9810049533843994, + "lr": 0.0006800996677740863, + "step": 10130, + "tokens_trained": 33.18600912 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 2.875966243528828, + "grad_norm": 0.3975411057472229, + "loss": 0.979, + "loss_ce": 1.098279595375061, + "loss_region": 0.0, + "loss_total": 1.098279595375061, + "lr": 0.0006797674418604652, + "step": 10140, + "tokens_trained": 33.21877456 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 2.8788029217785973, + "grad_norm": 0.4859318435192108, + "loss": 0.9697, + "loss_ce": 0.9473326206207275, + "loss_region": 0.0, + "loss_total": 0.9473326206207275, + "lr": 0.0006794352159468439, + "step": 10150, + "tokens_trained": 33.2515392 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 2.881639600028367, + "grad_norm": 0.41784191131591797, + "loss": 0.9722, + "loss_ce": 0.8371819257736206, + "loss_region": 0.0, + "loss_total": 0.8371819257736206, + "lr": 0.0006791029900332226, + "step": 10160, + "tokens_trained": 33.28430464 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 2.8844762782781364, + "grad_norm": 0.8353554010391235, + "loss": 0.9643, + "loss_ce": 1.0610942840576172, + "loss_region": 0.0, + "loss_total": 1.0610942840576172, + "lr": 0.0006787707641196013, + "step": 10170, + "tokens_trained": 33.31707008 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 2.887312956527906, + "grad_norm": 0.6107811331748962, + "loss": 0.9616, + "loss_ce": 0.974677562713623, + "loss_region": 0.0, + "loss_total": 0.974677562713623, + "lr": 0.0006784385382059801, + "step": 10180, + "tokens_trained": 33.34983472 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 2.8901496347776754, + "grad_norm": 0.401248037815094, + "loss": 0.9754, + "loss_ce": 0.7586034536361694, + "loss_region": 0.0, + "loss_total": 0.7586034536361694, + "lr": 0.0006781063122923589, + "step": 10190, + "tokens_trained": 33.38259936 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 2.892986313027445, + "grad_norm": 0.41112038493156433, + "loss": 0.969, + "loss_ce": 0.9582545757293701, + "loss_region": 0.0, + "loss_total": 0.9582545757293701, + "lr": 0.0006777740863787376, + "step": 10200, + "tokens_trained": 33.415364 + }, + { + "epoch": 2.892986313027445, + "eval_ppl": 2.669924461490042, + "eval_runtime": 1.4186, + "step": 10200, + "tokens_trained": 33.415364 + }, + { + "epoch": 2.892986313027445, + "eval_ppl_cds": 3.1167759570103257, + "eval_ppl_dig": 1.1024885632393808, + "eval_ppl_exon": 3.2096913724872933, + "eval_ppl_intron": 2.8239524518133194, + "eval_ppl_nig": 2.5390552878713746, + "eval_ppl_promoter": 2.6739690479876157, + "eval_ppl_utr": 2.8885875671741, + "step": 10200, + "tokens_trained": 33.415364 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 2.8958229912772144, + "grad_norm": 0.340558797121048, + "loss": 0.968, + "loss_ce": 0.9316603541374207, + "loss_region": 0.0, + "loss_total": 0.9316603541374207, + "lr": 0.0006774418604651162, + "step": 10210, + "tokens_trained": 33.44812944 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 2.898659669526984, + "grad_norm": 0.39050695300102234, + "loss": 0.9677, + "loss_ce": 1.0607205629348755, + "loss_region": 0.0, + "loss_total": 1.0607205629348755, + "lr": 0.000677109634551495, + "step": 10220, + "tokens_trained": 33.48089488 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 2.9014963477767535, + "grad_norm": 0.5038371682167053, + "loss": 0.9755, + "loss_ce": 1.0975452661514282, + "loss_region": 0.0, + "loss_total": 1.0975452661514282, + "lr": 0.0006767774086378737, + "step": 10230, + "tokens_trained": 33.51366032 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 2.904333026026523, + "grad_norm": 0.3648936152458191, + "loss": 0.9713, + "loss_ce": 1.0211875438690186, + "loss_region": 0.0, + "loss_total": 1.0211875438690186, + "lr": 0.0006764451827242525, + "step": 10240, + "tokens_trained": 33.54642576 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 2.9071697042762925, + "grad_norm": 0.3357242941856384, + "loss": 0.9718, + "loss_ce": 0.9881623387336731, + "loss_region": 0.0, + "loss_total": 0.9881623387336731, + "lr": 0.0006761129568106312, + "step": 10250, + "tokens_trained": 33.5791912 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 2.910006382526062, + "grad_norm": 0.4346785247325897, + "loss": 0.9842, + "loss_ce": 0.9385141730308533, + "loss_region": 0.0, + "loss_total": 0.9385141730308533, + "lr": 0.00067578073089701, + "step": 10260, + "tokens_trained": 33.61195664 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 2.9128430607758315, + "grad_norm": 0.43080833554267883, + "loss": 0.9755, + "loss_ce": 1.0151162147521973, + "loss_region": 0.0, + "loss_total": 1.0151162147521973, + "lr": 0.0006754485049833887, + "step": 10270, + "tokens_trained": 33.64472208 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 2.915679739025601, + "grad_norm": 0.16799435019493103, + "loss": 0.9679, + "loss_ce": 0.9646403193473816, + "loss_region": 0.0, + "loss_total": 0.9646403193473816, + "lr": 0.0006751162790697674, + "step": 10280, + "tokens_trained": 33.67748752 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 2.9185164172753706, + "grad_norm": 0.9791780710220337, + "loss": 0.9692, + "loss_ce": 1.024813175201416, + "loss_region": 0.0, + "loss_total": 1.024813175201416, + "lr": 0.0006747840531561463, + "step": 10290, + "tokens_trained": 33.71025296 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 2.92135309552514, + "grad_norm": 0.4908826947212219, + "loss": 0.965, + "loss_ce": 0.9707331657409668, + "loss_region": 0.0, + "loss_total": 0.9707331657409668, + "lr": 0.000674451827242525, + "step": 10300, + "tokens_trained": 33.743018392 + }, + { + "epoch": 2.92135309552514, + "eval_ppl": 2.6627964099301122, + "eval_runtime": 1.4297, + "step": 10300, + "tokens_trained": 33.743018392 + }, + { + "epoch": 2.92135309552514, + "eval_ppl_cds": 3.110869074839191, + "eval_ppl_dig": 1.100043624175283, + "eval_ppl_exon": 3.2096483390761965, + "eval_ppl_intron": 2.812861954875765, + "eval_ppl_nig": 2.5392253593118554, + "eval_ppl_promoter": 2.64006329088895, + "eval_ppl_utr": 2.8844069748647194, + "step": 10300, + "tokens_trained": 33.743018392 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 2.9241897737749096, + "grad_norm": 0.33514222502708435, + "loss": 0.9615, + "loss_ce": 1.0495747327804565, + "loss_region": 0.0, + "loss_total": 1.0495747327804565, + "lr": 0.0006741196013289037, + "step": 10310, + "tokens_trained": 33.775783832 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 2.927026452024679, + "grad_norm": 0.7031782865524292, + "loss": 0.9686, + "loss_ce": 1.1198009252548218, + "loss_region": 0.0, + "loss_total": 1.1198009252548218, + "lr": 0.0006737873754152824, + "step": 10320, + "tokens_trained": 33.808547592 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 2.9298631302744487, + "grad_norm": 0.28139829635620117, + "loss": 0.9736, + "loss_ce": 0.9068629145622253, + "loss_region": 0.0, + "loss_total": 0.9068629145622253, + "lr": 0.000673455149501661, + "step": 10330, + "tokens_trained": 33.841313032 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 2.932699808524218, + "grad_norm": 0.5585699677467346, + "loss": 0.9705, + "loss_ce": 1.0591907501220703, + "loss_region": 0.0, + "loss_total": 1.0591907501220703, + "lr": 0.0006731229235880399, + "step": 10340, + "tokens_trained": 33.874078472 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 2.9355364867739877, + "grad_norm": 0.5135933756828308, + "loss": 0.9695, + "loss_ce": 0.8969708681106567, + "loss_region": 0.0, + "loss_total": 0.8969708681106567, + "lr": 0.0006727906976744186, + "step": 10350, + "tokens_trained": 33.906843912 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 2.9383731650237572, + "grad_norm": 0.34856677055358887, + "loss": 0.9723, + "loss_ce": 1.1170352697372437, + "loss_region": 0.0, + "loss_total": 1.1170352697372437, + "lr": 0.0006724584717607973, + "step": 10360, + "tokens_trained": 33.9396068 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 2.9412098432735267, + "grad_norm": 0.4313170611858368, + "loss": 0.9664, + "loss_ce": 1.0496752262115479, + "loss_region": 0.0, + "loss_total": 1.0496752262115479, + "lr": 0.000672126245847176, + "step": 10370, + "tokens_trained": 33.97237224 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 2.9440465215232963, + "grad_norm": 0.5418225526809692, + "loss": 0.9784, + "loss_ce": 1.1155195236206055, + "loss_region": 0.0, + "loss_total": 1.1155195236206055, + "lr": 0.0006717940199335548, + "step": 10380, + "tokens_trained": 34.00513768 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 2.946883199773066, + "grad_norm": 0.38688722252845764, + "loss": 0.9745, + "loss_ce": 0.8803756237030029, + "loss_region": 0.0, + "loss_total": 0.8803756237030029, + "lr": 0.0006714617940199336, + "step": 10390, + "tokens_trained": 34.03790312 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 2.9497198780228353, + "grad_norm": 0.0761430487036705, + "loss": 0.9725, + "loss_ce": 0.9497737288475037, + "loss_region": 0.0, + "loss_total": 0.9497737288475037, + "lr": 0.0006711295681063123, + "step": 10400, + "tokens_trained": 34.07066856 + }, + { + "epoch": 2.9497198780228353, + "eval_ppl": 2.664611140870884, + "eval_runtime": 1.4195, + "step": 10400, + "tokens_trained": 34.07066856 + }, + { + "epoch": 2.9497198780228353, + "eval_ppl_cds": 3.1384757723253838, + "eval_ppl_dig": 1.1001558696071505, + "eval_ppl_exon": 3.2179209820244172, + "eval_ppl_intron": 2.8231409971039105, + "eval_ppl_nig": 2.5489621356153043, + "eval_ppl_promoter": 2.6736548627843018, + "eval_ppl_utr": 2.9201580297025154, + "step": 10400, + "tokens_trained": 34.07066856 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 2.952556556272605, + "grad_norm": 0.3361152708530426, + "loss": 0.9665, + "loss_ce": 1.0357940196990967, + "loss_region": 0.0, + "loss_total": 1.0357940196990967, + "lr": 0.0006707973421926911, + "step": 10410, + "tokens_trained": 34.103434 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 2.9553932345223743, + "grad_norm": 0.35342198610305786, + "loss": 0.9774, + "loss_ce": 0.8822641372680664, + "loss_region": 0.0, + "loss_total": 0.8822641372680664, + "lr": 0.0006704651162790698, + "step": 10420, + "tokens_trained": 34.13619944 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 2.958229912772144, + "grad_norm": 0.9376862049102783, + "loss": 0.9605, + "loss_ce": 1.1419216394424438, + "loss_region": 0.0, + "loss_total": 1.1419216394424438, + "lr": 0.0006701328903654485, + "step": 10430, + "tokens_trained": 34.16896488 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 2.9610665910219134, + "grad_norm": 0.3814321756362915, + "loss": 0.9798, + "loss_ce": 0.890766441822052, + "loss_region": 0.0, + "loss_total": 0.890766441822052, + "lr": 0.0006698006644518273, + "step": 10440, + "tokens_trained": 34.20173032 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 2.963903269271683, + "grad_norm": 0.2588978707790375, + "loss": 0.9664, + "loss_ce": 0.9431135058403015, + "loss_region": 0.0, + "loss_total": 0.9431135058403015, + "lr": 0.0006694684385382061, + "step": 10450, + "tokens_trained": 34.23449576 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 2.9667399475214524, + "grad_norm": 0.44445472955703735, + "loss": 0.9662, + "loss_ce": 1.0568519830703735, + "loss_region": 0.0, + "loss_total": 1.0568519830703735, + "lr": 0.0006691362126245847, + "step": 10460, + "tokens_trained": 34.2672604 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 2.969576625771222, + "grad_norm": 0.30835816264152527, + "loss": 0.9596, + "loss_ce": 0.9751994013786316, + "loss_region": 0.0, + "loss_total": 0.9751994013786316, + "lr": 0.0006688039867109634, + "step": 10470, + "tokens_trained": 34.30002504 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 2.9724133040209915, + "grad_norm": 0.19293968379497528, + "loss": 0.9605, + "loss_ce": 0.9859769940376282, + "loss_region": 0.0, + "loss_total": 0.9859769940376282, + "lr": 0.0006684717607973421, + "step": 10480, + "tokens_trained": 34.33279048 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 2.975249982270761, + "grad_norm": 0.5898815393447876, + "loss": 0.9693, + "loss_ce": 1.0070358514785767, + "loss_region": 0.0, + "loss_total": 1.0070358514785767, + "lr": 0.000668139534883721, + "step": 10490, + "tokens_trained": 34.365555152 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 2.9780866605205305, + "grad_norm": 0.423366904258728, + "loss": 0.9677, + "loss_ce": 0.9775635600090027, + "loss_region": 0.0, + "loss_total": 0.9775635600090027, + "lr": 0.0006678073089700997, + "step": 10500, + "tokens_trained": 34.398320592 + }, + { + "epoch": 2.9780866605205305, + "eval_ppl": 2.660031601958519, + "eval_runtime": 1.422, + "step": 10500, + "tokens_trained": 34.398320592 + }, + { + "epoch": 2.9780866605205305, + "eval_ppl_cds": 3.1216368998761292, + "eval_ppl_dig": 1.0996887278258098, + "eval_ppl_exon": 3.188729695756402, + "eval_ppl_intron": 2.812506630352669, + "eval_ppl_nig": 2.5479919561685245, + "eval_ppl_promoter": 2.648986429427844, + "eval_ppl_utr": 2.8907703586615345, + "step": 10500, + "tokens_trained": 34.398320592 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 2.9809233387703, + "grad_norm": 0.46533504128456116, + "loss": 0.9623, + "loss_ce": 1.1161397695541382, + "loss_region": 0.0, + "loss_total": 1.1161397695541382, + "lr": 0.0006674750830564784, + "step": 10510, + "tokens_trained": 34.431081872 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 2.9837600170200695, + "grad_norm": 0.3514612019062042, + "loss": 0.9676, + "loss_ce": 0.8101162314414978, + "loss_region": 0.0, + "loss_total": 0.8101162314414978, + "lr": 0.0006671428571428571, + "step": 10520, + "tokens_trained": 34.463842376 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 2.986596695269839, + "grad_norm": 0.3695121109485626, + "loss": 0.9625, + "loss_ce": 1.0984371900558472, + "loss_region": 0.0, + "loss_total": 1.0984371900558472, + "lr": 0.0006668106312292359, + "step": 10530, + "tokens_trained": 34.496607816 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 2.9894333735196086, + "grad_norm": 0.45197412371635437, + "loss": 0.9775, + "loss_ce": 1.095833420753479, + "loss_region": 0.0, + "loss_total": 1.095833420753479, + "lr": 0.0006664784053156147, + "step": 10540, + "tokens_trained": 34.529373256 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 2.992270051769378, + "grad_norm": 0.5059137344360352, + "loss": 0.9743, + "loss_ce": 0.8169457316398621, + "loss_region": 0.0, + "loss_total": 0.8169457316398621, + "lr": 0.0006661461794019934, + "step": 10550, + "tokens_trained": 34.562137896 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 2.9951067300191476, + "grad_norm": 0.691680908203125, + "loss": 0.9613, + "loss_ce": 0.9091237187385559, + "loss_region": 0.0, + "loss_total": 0.9091237187385559, + "lr": 0.0006658139534883722, + "step": 10560, + "tokens_trained": 34.594903176 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 2.997943408268917, + "grad_norm": 0.4358217418193817, + "loss": 0.9667, + "loss_ce": 0.9528109431266785, + "loss_region": 0.0, + "loss_total": 0.9528109431266785, + "lr": 0.0006654817275747509, + "step": 10570, + "tokens_trained": 34.627668616 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 3.000567335649954, + "grad_norm": 0.479568749666214, + "loss": 0.9673, + "loss_ce": 1.0996983051300049, + "loss_region": 0.0, + "loss_total": 1.0996983051300049, + "lr": 0.0006651495016611295, + "step": 10580, + "tokens_trained": 34.657976648 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 3.0034040138997233, + "grad_norm": 0.26386234164237976, + "loss": 0.9613, + "loss_ce": 0.8043960928916931, + "loss_region": 0.0, + "loss_total": 0.8043960928916931, + "lr": 0.0006648172757475083, + "step": 10590, + "tokens_trained": 34.690742088 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 3.006240692149493, + "grad_norm": 0.2861813008785248, + "loss": 0.9669, + "loss_ce": 0.8900136947631836, + "loss_region": 0.0, + "loss_total": 0.8900136947631836, + "lr": 0.000664485049833887, + "step": 10600, + "tokens_trained": 34.723505928 + }, + { + "epoch": 3.006240692149493, + "eval_ppl": 2.6591587675767063, + "eval_runtime": 1.4137, + "step": 10600, + "tokens_trained": 34.723505928 + }, + { + "epoch": 3.006240692149493, + "eval_ppl_cds": 3.1006087066236305, + "eval_ppl_dig": 1.099249441051081, + "eval_ppl_exon": 3.156277733778931, + "eval_ppl_intron": 2.812089139951921, + "eval_ppl_nig": 2.537047966989843, + "eval_ppl_promoter": 2.625440962723904, + "eval_ppl_utr": 2.8782475547573196, + "step": 10600, + "tokens_trained": 34.723505928 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 3.0090773703992624, + "grad_norm": 0.07822912186384201, + "loss": 0.9599, + "loss_ce": 1.1005885601043701, + "loss_region": 0.0, + "loss_total": 1.1005885601043701, + "lr": 0.0006641528239202658, + "step": 10610, + "tokens_trained": 34.756271368 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 3.011914048649032, + "grad_norm": 0.38587436079978943, + "loss": 0.961, + "loss_ce": 0.9035478234291077, + "loss_region": 0.0, + "loss_total": 0.9035478234291077, + "lr": 0.0006638205980066445, + "step": 10620, + "tokens_trained": 34.789036808 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 3.0147507268988014, + "grad_norm": 0.515742838382721, + "loss": 0.9691, + "loss_ce": 1.0287610292434692, + "loss_region": 0.0, + "loss_total": 1.0287610292434692, + "lr": 0.0006634883720930232, + "step": 10630, + "tokens_trained": 34.82180148 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 3.017587405148571, + "grad_norm": 0.2521349787712097, + "loss": 0.9613, + "loss_ce": 0.8735443353652954, + "loss_region": 0.0, + "loss_total": 0.8735443353652954, + "lr": 0.0006631561461794021, + "step": 10640, + "tokens_trained": 34.85456432 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 3.0204240833983405, + "grad_norm": 0.35564473271369934, + "loss": 0.9601, + "loss_ce": 0.9267745614051819, + "loss_region": 0.0, + "loss_total": 0.9267745614051819, + "lr": 0.0006628239202657808, + "step": 10650, + "tokens_trained": 34.88732976 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 3.02326076164811, + "grad_norm": 0.5696294903755188, + "loss": 0.9647, + "loss_ce": 1.0435744524002075, + "loss_region": 0.0, + "loss_total": 1.0435744524002075, + "lr": 0.0006624916943521595, + "step": 10660, + "tokens_trained": 34.9200952 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 3.0260974398978795, + "grad_norm": 0.29344141483306885, + "loss": 0.9636, + "loss_ce": 0.9749945402145386, + "loss_region": 0.0, + "loss_total": 0.9749945402145386, + "lr": 0.0006621594684385382, + "step": 10670, + "tokens_trained": 34.95286064 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 3.028934118147649, + "grad_norm": 0.35718098282814026, + "loss": 0.9585, + "loss_ce": 0.9447165727615356, + "loss_region": 0.0, + "loss_total": 0.9447165727615356, + "lr": 0.000661827242524917, + "step": 10680, + "tokens_trained": 34.98562528 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 3.0317707963974185, + "grad_norm": 0.5475853681564331, + "loss": 0.9582, + "loss_ce": 0.8818969130516052, + "loss_region": 0.0, + "loss_total": 0.8818969130516052, + "lr": 0.0006614950166112958, + "step": 10690, + "tokens_trained": 35.01838912 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 3.034607474647188, + "grad_norm": 0.34258919954299927, + "loss": 0.959, + "loss_ce": 0.9844611883163452, + "loss_region": 0.0, + "loss_total": 0.9844611883163452, + "lr": 0.0006611627906976745, + "step": 10700, + "tokens_trained": 35.051152152 + }, + { + "epoch": 3.034607474647188, + "eval_ppl": 2.6615598308862873, + "eval_runtime": 1.4117, + "step": 10700, + "tokens_trained": 35.051152152 + }, + { + "epoch": 3.034607474647188, + "eval_ppl_cds": 3.120084328108816, + "eval_ppl_dig": 1.1008279448907312, + "eval_ppl_exon": 3.153381157516803, + "eval_ppl_intron": 2.8146827969861725, + "eval_ppl_nig": 2.5411697372991586, + "eval_ppl_promoter": 2.6343662525026605, + "eval_ppl_utr": 2.8898194369544017, + "step": 10700, + "tokens_trained": 35.051152152 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 3.0374441528969576, + "grad_norm": 0.32405561208724976, + "loss": 0.9669, + "loss_ce": 0.810122013092041, + "loss_region": 0.0, + "loss_total": 0.810122013092041, + "lr": 0.0006608305647840531, + "step": 10710, + "tokens_trained": 35.083917592 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 3.040280831146727, + "grad_norm": 0.29983675479888916, + "loss": 0.9581, + "loss_ce": 0.9554064869880676, + "loss_region": 0.0, + "loss_total": 0.9554064869880676, + "lr": 0.0006604983388704319, + "step": 10720, + "tokens_trained": 35.116683032 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 3.0431175093964966, + "grad_norm": 0.21896053850650787, + "loss": 0.9623, + "loss_ce": 1.0718941688537598, + "loss_region": 0.0, + "loss_total": 1.0718941688537598, + "lr": 0.0006601661129568106, + "step": 10730, + "tokens_trained": 35.149447672 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 3.045954187646266, + "grad_norm": 0.42219772934913635, + "loss": 0.9549, + "loss_ce": 1.014609456062317, + "loss_region": 0.0, + "loss_total": 1.014609456062317, + "lr": 0.0006598338870431894, + "step": 10740, + "tokens_trained": 35.182213032 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 3.0487908658960357, + "grad_norm": 0.4437616467475891, + "loss": 0.9634, + "loss_ce": 1.0749627351760864, + "loss_region": 0.0, + "loss_total": 1.0749627351760864, + "lr": 0.0006595016611295681, + "step": 10750, + "tokens_trained": 35.214976872 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 3.051627544145805, + "grad_norm": 0.19731970131397247, + "loss": 0.9551, + "loss_ce": 0.9427595138549805, + "loss_region": 0.0, + "loss_total": 0.9427595138549805, + "lr": 0.0006591694352159469, + "step": 10760, + "tokens_trained": 35.247742312 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 3.0544642223955747, + "grad_norm": 0.5920313000679016, + "loss": 0.9588, + "loss_ce": 0.8994665145874023, + "loss_region": 0.0, + "loss_total": 0.8994665145874023, + "lr": 0.0006588372093023256, + "step": 10770, + "tokens_trained": 35.280507752 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 3.057300900645344, + "grad_norm": 0.4098859429359436, + "loss": 0.9584, + "loss_ce": 1.0081416368484497, + "loss_region": 0.0, + "loss_total": 1.0081416368484497, + "lr": 0.0006585049833887043, + "step": 10780, + "tokens_trained": 35.313273192 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 3.0601375788951137, + "grad_norm": 0.21337725222110748, + "loss": 0.9621, + "loss_ce": 0.9820064902305603, + "loss_region": 0.0, + "loss_total": 0.9820064902305603, + "lr": 0.0006581727574750831, + "step": 10790, + "tokens_trained": 35.346038312 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 3.0629742571448832, + "grad_norm": 0.22523583471775055, + "loss": 0.963, + "loss_ce": 0.8709080219268799, + "loss_region": 0.0, + "loss_total": 0.8709080219268799, + "lr": 0.0006578405315614619, + "step": 10800, + "tokens_trained": 35.378803752 + }, + { + "epoch": 3.0629742571448832, + "eval_ppl": 2.6574961657958935, + "eval_runtime": 1.4261, + "step": 10800, + "tokens_trained": 35.378803752 + }, + { + "epoch": 3.0629742571448832, + "eval_ppl_cds": 3.1307457318904097, + "eval_ppl_dig": 1.0912533649657, + "eval_ppl_exon": 3.1721557778280345, + "eval_ppl_intron": 2.819056626526808, + "eval_ppl_nig": 2.546545374233694, + "eval_ppl_promoter": 2.658417253880614, + "eval_ppl_utr": 2.895326113435088, + "step": 10800, + "tokens_trained": 35.378803752 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 3.0658109353946528, + "grad_norm": 0.3578243553638458, + "loss": 0.9602, + "loss_ce": 1.0572514533996582, + "loss_region": 0.0, + "loss_total": 1.0572514533996582, + "lr": 0.0006575083056478406, + "step": 10810, + "tokens_trained": 35.411569192 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 3.0686476136444223, + "grad_norm": 0.3954513669013977, + "loss": 0.9552, + "loss_ce": 1.0732814073562622, + "loss_region": 0.0, + "loss_total": 1.0732814073562622, + "lr": 0.0006571760797342193, + "step": 10820, + "tokens_trained": 35.444330768 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 3.071484291894192, + "grad_norm": 0.3284713327884674, + "loss": 0.9611, + "loss_ce": 0.924236536026001, + "loss_region": 0.0, + "loss_total": 0.924236536026001, + "lr": 0.0006568438538205979, + "step": 10830, + "tokens_trained": 35.477096208 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 3.0743209701439613, + "grad_norm": 0.36103978753089905, + "loss": 0.9718, + "loss_ce": 0.930971622467041, + "loss_region": 0.0, + "loss_total": 0.930971622467041, + "lr": 0.0006565116279069768, + "step": 10840, + "tokens_trained": 35.509861648 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 3.077157648393731, + "grad_norm": 0.16559931635856628, + "loss": 0.9604, + "loss_ce": 1.0561270713806152, + "loss_region": 0.0, + "loss_total": 1.0561270713806152, + "lr": 0.0006561794019933555, + "step": 10850, + "tokens_trained": 35.542625488 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 3.0799943266435004, + "grad_norm": 0.4456230700016022, + "loss": 0.9564, + "loss_ce": 1.0505366325378418, + "loss_region": 0.0, + "loss_total": 1.0505366325378418, + "lr": 0.0006558471760797342, + "step": 10860, + "tokens_trained": 35.575390928 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 3.08283100489327, + "grad_norm": 0.4156387150287628, + "loss": 0.9613, + "loss_ce": 0.9476015567779541, + "loss_region": 0.0, + "loss_total": 0.9476015567779541, + "lr": 0.0006555149501661129, + "step": 10870, + "tokens_trained": 35.608155752 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 3.0856676831430394, + "grad_norm": 0.2523500919342041, + "loss": 0.9602, + "loss_ce": 0.9598459601402283, + "loss_region": 0.0, + "loss_total": 0.9598459601402283, + "lr": 0.0006551827242524917, + "step": 10880, + "tokens_trained": 35.640921192 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 3.088504361392809, + "grad_norm": 0.2711198925971985, + "loss": 0.949, + "loss_ce": 1.0145546197891235, + "loss_region": 0.0, + "loss_total": 1.0145546197891235, + "lr": 0.0006548504983388705, + "step": 10890, + "tokens_trained": 35.673686632 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 3.0913410396425784, + "grad_norm": 0.2912087142467499, + "loss": 0.9611, + "loss_ce": 0.7664705514907837, + "loss_region": 0.0, + "loss_total": 0.7664705514907837, + "lr": 0.0006545182724252492, + "step": 10900, + "tokens_trained": 35.706452072 + }, + { + "epoch": 3.0913410396425784, + "eval_ppl": 2.6501240609957253, + "eval_runtime": 1.4264, + "step": 10900, + "tokens_trained": 35.706452072 + }, + { + "epoch": 3.0913410396425784, + "eval_ppl_cds": 3.120734917915377, + "eval_ppl_dig": 1.0957686197333776, + "eval_ppl_exon": 3.183292429099265, + "eval_ppl_intron": 2.8162739825003, + "eval_ppl_nig": 2.5480068769262654, + "eval_ppl_promoter": 2.63656391134655, + "eval_ppl_utr": 2.9020772957658982, + "step": 10900, + "tokens_trained": 35.706452072 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 3.094177717892348, + "grad_norm": 0.5137304067611694, + "loss": 0.9559, + "loss_ce": 1.0300456285476685, + "loss_region": 0.0, + "loss_total": 1.0300456285476685, + "lr": 0.000654186046511628, + "step": 10910, + "tokens_trained": 35.739217512 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 3.0970143961421175, + "grad_norm": 0.4634801149368286, + "loss": 0.9472, + "loss_ce": 0.87733393907547, + "loss_region": 0.0, + "loss_total": 0.87733393907547, + "lr": 0.0006538538205980067, + "step": 10920, + "tokens_trained": 35.771982952 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 3.099851074391887, + "grad_norm": 0.4145699739456177, + "loss": 0.9637, + "loss_ce": 1.129374384880066, + "loss_region": 0.0, + "loss_total": 1.129374384880066, + "lr": 0.0006535215946843854, + "step": 10930, + "tokens_trained": 35.804748392 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 3.1026877526416565, + "grad_norm": 0.34833163022994995, + "loss": 0.9594, + "loss_ce": 0.892644464969635, + "loss_region": 0.0, + "loss_total": 0.892644464969635, + "lr": 0.0006531893687707642, + "step": 10940, + "tokens_trained": 35.837513832 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 3.105524430891426, + "grad_norm": 0.6839308738708496, + "loss": 0.9717, + "loss_ce": 1.108083963394165, + "loss_region": 0.0, + "loss_total": 1.108083963394165, + "lr": 0.0006528571428571429, + "step": 10950, + "tokens_trained": 35.870278472 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 3.1083611091411956, + "grad_norm": 0.460269957780838, + "loss": 0.9624, + "loss_ce": 0.7921956777572632, + "loss_region": 0.0, + "loss_total": 0.7921956777572632, + "lr": 0.0006525249169435216, + "step": 10960, + "tokens_trained": 35.903043752 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 3.111197787390965, + "grad_norm": 0.2990061044692993, + "loss": 0.9493, + "loss_ce": 0.992203414440155, + "loss_region": 0.0, + "loss_total": 0.992203414440155, + "lr": 0.0006521926910299003, + "step": 10970, + "tokens_trained": 35.935809192 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 3.1140344656407346, + "grad_norm": 0.2857515811920166, + "loss": 0.9623, + "loss_ce": 1.0307515859603882, + "loss_region": 0.0, + "loss_total": 1.0307515859603882, + "lr": 0.000651860465116279, + "step": 10980, + "tokens_trained": 35.968574632 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 3.116871143890504, + "grad_norm": 0.2124212086200714, + "loss": 0.963, + "loss_ce": 0.9193806052207947, + "loss_region": 0.0, + "loss_total": 0.9193806052207947, + "lr": 0.0006515282392026577, + "step": 10990, + "tokens_trained": 36.001340048 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 3.1197078221402736, + "grad_norm": 0.6648218035697937, + "loss": 0.9663, + "loss_ce": 1.085507869720459, + "loss_region": 0.0, + "loss_total": 1.085507869720459, + "lr": 0.0006511960132890366, + "step": 11000, + "tokens_trained": 36.034105488 + }, + { + "epoch": 3.1197078221402736, + "eval_ppl": 2.6517564452297684, + "eval_runtime": 1.4176, + "step": 11000, + "tokens_trained": 36.034105488 + }, + { + "epoch": 3.1197078221402736, + "eval_ppl_cds": 3.137716020876085, + "eval_ppl_dig": 1.0987450000597891, + "eval_ppl_exon": 3.1985379982966267, + "eval_ppl_intron": 2.8191475887209747, + "eval_ppl_nig": 2.5503362428596663, + "eval_ppl_promoter": 2.649670514617122, + "eval_ppl_utr": 2.9090702110550146, + "step": 11000, + "tokens_trained": 36.034105488 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 3.122544500390043, + "grad_norm": 0.3351917564868927, + "loss": 0.963, + "loss_ce": 0.9537840485572815, + "loss_region": 0.0, + "loss_total": 0.9537840485572815, + "lr": 0.0006508637873754153, + "step": 11010, + "tokens_trained": 36.066870928 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 3.1253811786398127, + "grad_norm": 0.47707200050354004, + "loss": 0.9573, + "loss_ce": 0.9620508551597595, + "loss_region": 0.0, + "loss_total": 0.9620508551597595, + "lr": 0.000650531561461794, + "step": 11020, + "tokens_trained": 36.099636368 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 3.128217856889582, + "grad_norm": 0.09057100117206573, + "loss": 0.9617, + "loss_ce": 0.9611527919769287, + "loss_region": 0.0, + "loss_total": 0.9611527919769287, + "lr": 0.0006501993355481728, + "step": 11030, + "tokens_trained": 36.132401648 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 3.1310545351393517, + "grad_norm": 0.32626935839653015, + "loss": 0.9663, + "loss_ce": 0.8963630199432373, + "loss_region": 0.0, + "loss_total": 0.8963630199432373, + "lr": 0.0006498671096345515, + "step": 11040, + "tokens_trained": 36.165167088 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 3.1338912133891212, + "grad_norm": 0.2244894653558731, + "loss": 0.9645, + "loss_ce": 0.9216999411582947, + "loss_region": 0.0, + "loss_total": 0.9216999411582947, + "lr": 0.0006495348837209303, + "step": 11050, + "tokens_trained": 36.197931728 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 3.1367278916388908, + "grad_norm": 0.384146124124527, + "loss": 0.9586, + "loss_ce": 1.056295394897461, + "loss_region": 0.0, + "loss_total": 1.056295394897461, + "lr": 0.000649202657807309, + "step": 11060, + "tokens_trained": 36.230697168 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 3.1395645698886603, + "grad_norm": 0.34388700127601624, + "loss": 0.9626, + "loss_ce": 0.9692562222480774, + "loss_region": 0.0, + "loss_total": 0.9692562222480774, + "lr": 0.0006488704318936878, + "step": 11070, + "tokens_trained": 36.263462608 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 3.14240124813843, + "grad_norm": 0.4322400987148285, + "loss": 0.9552, + "loss_ce": 1.0747169256210327, + "loss_region": 0.0, + "loss_total": 1.0747169256210327, + "lr": 0.0006485382059800664, + "step": 11080, + "tokens_trained": 36.296228048 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 3.1452379263881993, + "grad_norm": 0.24304872751235962, + "loss": 0.9538, + "loss_ce": 1.0415270328521729, + "loss_region": 0.0, + "loss_total": 1.0415270328521729, + "lr": 0.0006482059800664451, + "step": 11090, + "tokens_trained": 36.328993488 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 3.148074604637969, + "grad_norm": 0.4047473669052124, + "loss": 0.9512, + "loss_ce": 0.9804026484489441, + "loss_region": 0.0, + "loss_total": 0.9804026484489441, + "lr": 0.0006478737541528239, + "step": 11100, + "tokens_trained": 36.361758928 + }, + { + "epoch": 3.148074604637969, + "eval_ppl": 2.6446964446878916, + "eval_runtime": 1.4304, + "step": 11100, + "tokens_trained": 36.361758928 + }, + { + "epoch": 3.148074604637969, + "eval_ppl_cds": 3.0950770655505777, + "eval_ppl_dig": 1.0969397123337123, + "eval_ppl_exon": 3.1429744328305484, + "eval_ppl_intron": 2.8126805256485823, + "eval_ppl_nig": 2.54189006282619, + "eval_ppl_promoter": 2.619355843777506, + "eval_ppl_utr": 2.8563938817591974, + "step": 11100, + "tokens_trained": 36.361758928 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 3.1509112828877384, + "grad_norm": 0.3351514935493469, + "loss": 0.9588, + "loss_ce": 1.0315817594528198, + "loss_region": 0.0, + "loss_total": 1.0315817594528198, + "lr": 0.0006475415282392027, + "step": 11110, + "tokens_trained": 36.394524368 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 3.153747961137508, + "grad_norm": 0.18956217169761658, + "loss": 0.9538, + "loss_ce": 1.0245298147201538, + "loss_region": 0.0, + "loss_total": 1.0245298147201538, + "lr": 0.0006472093023255814, + "step": 11120, + "tokens_trained": 36.427287248 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 3.1565846393872774, + "grad_norm": 0.4612583816051483, + "loss": 0.959, + "loss_ce": 0.8369152545928955, + "loss_region": 0.0, + "loss_total": 0.8369152545928955, + "lr": 0.0006468770764119601, + "step": 11130, + "tokens_trained": 36.460052528 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 3.159421317637047, + "grad_norm": 0.27334514260292053, + "loss": 0.956, + "loss_ce": 0.9203778505325317, + "loss_region": 0.0, + "loss_total": 0.9203778505325317, + "lr": 0.0006465448504983388, + "step": 11140, + "tokens_trained": 36.492817168 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 3.1622579958868164, + "grad_norm": 0.07751717418432236, + "loss": 0.9474, + "loss_ce": 1.0421650409698486, + "loss_region": 0.0, + "loss_total": 1.0421650409698486, + "lr": 0.0006462126245847177, + "step": 11150, + "tokens_trained": 36.525582608 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 3.165094674136586, + "grad_norm": 0.45421579480171204, + "loss": 0.9578, + "loss_ce": 0.7321340441703796, + "loss_region": 0.0, + "loss_total": 0.7321340441703796, + "lr": 0.0006458803986710964, + "step": 11160, + "tokens_trained": 36.558348048 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 3.1679313523863555, + "grad_norm": 0.3423018753528595, + "loss": 0.9569, + "loss_ce": 0.8933833837509155, + "loss_region": 0.0, + "loss_total": 0.8933833837509155, + "lr": 0.0006455481727574751, + "step": 11170, + "tokens_trained": 36.591111888 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 3.170768030636125, + "grad_norm": 0.3392423391342163, + "loss": 0.946, + "loss_ce": 1.0609477758407593, + "loss_region": 0.0, + "loss_total": 1.0609477758407593, + "lr": 0.0006452159468438538, + "step": 11180, + "tokens_trained": 36.623877328 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 3.1736047088858945, + "grad_norm": 0.2937402129173279, + "loss": 0.9512, + "loss_ce": 0.9885709285736084, + "loss_region": 0.0, + "loss_total": 0.9885709285736084, + "lr": 0.0006448837209302326, + "step": 11190, + "tokens_trained": 36.656642768 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 3.176441387135664, + "grad_norm": 0.39738303422927856, + "loss": 0.9577, + "loss_ce": 0.9088342785835266, + "loss_region": 0.0, + "loss_total": 0.9088342785835266, + "lr": 0.0006445514950166113, + "step": 11200, + "tokens_trained": 36.689408208 + }, + { + "epoch": 3.176441387135664, + "eval_ppl": 2.645768114955475, + "eval_runtime": 1.4315, + "step": 11200, + "tokens_trained": 36.689408208 + }, + { + "epoch": 3.176441387135664, + "eval_ppl_cds": 3.1173614731606065, + "eval_ppl_dig": 1.0981211892237515, + "eval_ppl_exon": 3.133767374902806, + "eval_ppl_intron": 2.8178745461949912, + "eval_ppl_nig": 2.542699716098563, + "eval_ppl_promoter": 2.635296075204437, + "eval_ppl_utr": 2.857980276149635, + "step": 11200, + "tokens_trained": 36.689408208 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 3.1792780653854336, + "grad_norm": 0.48402807116508484, + "loss": 0.9536, + "loss_ce": 1.1045215129852295, + "loss_region": 0.0, + "loss_total": 1.1045215129852295, + "lr": 0.00064421926910299, + "step": 11210, + "tokens_trained": 36.722172848 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 3.182114743635203, + "grad_norm": 0.2937231957912445, + "loss": 0.9565, + "loss_ce": 1.0653750896453857, + "loss_region": 0.0, + "loss_total": 1.0653750896453857, + "lr": 0.0006438870431893687, + "step": 11220, + "tokens_trained": 36.754938288 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 3.1849514218849726, + "grad_norm": 0.3180597424507141, + "loss": 0.9533, + "loss_ce": 0.8676323294639587, + "loss_region": 0.0, + "loss_total": 0.8676323294639587, + "lr": 0.0006435548172757475, + "step": 11230, + "tokens_trained": 36.787703728 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 3.187788100134742, + "grad_norm": 0.13840338587760925, + "loss": 0.9562, + "loss_ce": 0.7355803847312927, + "loss_region": 0.0, + "loss_total": 0.7355803847312927, + "lr": 0.0006432225913621262, + "step": 11240, + "tokens_trained": 36.820469168 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 3.1906247783845116, + "grad_norm": 0.4488730728626251, + "loss": 0.956, + "loss_ce": 1.0945569276809692, + "loss_region": 0.0, + "loss_total": 1.0945569276809692, + "lr": 0.000642890365448505, + "step": 11250, + "tokens_trained": 36.853232016 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 3.193461456634281, + "grad_norm": 0.4940653443336487, + "loss": 0.9676, + "loss_ce": 0.9472804069519043, + "loss_region": 0.0, + "loss_total": 0.9472804069519043, + "lr": 0.0006425581395348838, + "step": 11260, + "tokens_trained": 36.885997456 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 3.1962981348840507, + "grad_norm": 0.35126930475234985, + "loss": 0.9496, + "loss_ce": 0.8703694939613342, + "loss_region": 0.0, + "loss_total": 0.8703694939613342, + "lr": 0.0006422259136212625, + "step": 11270, + "tokens_trained": 36.918762896 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 3.19913481313382, + "grad_norm": 0.398426353931427, + "loss": 0.9566, + "loss_ce": 0.970214307308197, + "loss_region": 0.0, + "loss_total": 0.970214307308197, + "lr": 0.0006418936877076412, + "step": 11280, + "tokens_trained": 36.951528336 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 3.2019714913835897, + "grad_norm": 0.299973726272583, + "loss": 0.9421, + "loss_ce": 1.1430424451828003, + "loss_region": 0.0, + "loss_total": 1.1430424451828003, + "lr": 0.0006415614617940199, + "step": 11290, + "tokens_trained": 36.984293776 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 3.2048081696333592, + "grad_norm": 0.46328336000442505, + "loss": 0.9623, + "loss_ce": 0.9646133780479431, + "loss_region": 0.0, + "loss_total": 0.9646133780479431, + "lr": 0.0006412292358803988, + "step": 11300, + "tokens_trained": 37.017059216 + }, + { + "epoch": 3.2048081696333592, + "eval_ppl": 2.643104010580077, + "eval_runtime": 1.4148, + "step": 11300, + "tokens_trained": 37.017059216 + }, + { + "epoch": 3.2048081696333592, + "eval_ppl_cds": 3.083482747391328, + "eval_ppl_dig": 1.0944491028768348, + "eval_ppl_exon": 3.1480875264073513, + "eval_ppl_intron": 2.816356318846718, + "eval_ppl_nig": 2.539995896092654, + "eval_ppl_promoter": 2.6167339437689288, + "eval_ppl_utr": 2.8547133367987114, + "step": 11300, + "tokens_trained": 37.017059216 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 3.2076448478831288, + "grad_norm": 0.24404732882976532, + "loss": 0.9589, + "loss_ce": 1.051213264465332, + "loss_region": 0.0, + "loss_total": 1.051213264465332, + "lr": 0.0006408970099667775, + "step": 11310, + "tokens_trained": 37.049824656 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 3.2104815261328983, + "grad_norm": 0.20346620678901672, + "loss": 0.9631, + "loss_ce": 0.8799743056297302, + "loss_region": 0.0, + "loss_total": 0.8799743056297302, + "lr": 0.0006405647840531562, + "step": 11320, + "tokens_trained": 37.082590096 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 3.213318204382668, + "grad_norm": 0.23345786333084106, + "loss": 0.9619, + "loss_ce": 0.629621684551239, + "loss_region": 0.0, + "loss_total": 0.629621684551239, + "lr": 0.0006402325581395348, + "step": 11330, + "tokens_trained": 37.115355536 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 3.2161548826324373, + "grad_norm": 0.25960221886634827, + "loss": 0.9597, + "loss_ce": 1.0486069917678833, + "loss_region": 0.0, + "loss_total": 1.0486069917678833, + "lr": 0.0006399003322259136, + "step": 11340, + "tokens_trained": 37.148120976 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 3.218991560882207, + "grad_norm": 0.7528862357139587, + "loss": 0.9448, + "loss_ce": 0.9782311916351318, + "loss_region": 0.0, + "loss_total": 0.9782311916351318, + "lr": 0.0006395681063122924, + "step": 11350, + "tokens_trained": 37.180886416 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 3.2218282391319764, + "grad_norm": 0.4925711452960968, + "loss": 0.9605, + "loss_ce": 1.0627838373184204, + "loss_region": 0.0, + "loss_total": 1.0627838373184204, + "lr": 0.0006392358803986711, + "step": 11360, + "tokens_trained": 37.213651856 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 3.224664917381746, + "grad_norm": 0.2448003888130188, + "loss": 0.9522, + "loss_ce": 0.7761248350143433, + "loss_region": 0.0, + "loss_total": 0.7761248350143433, + "lr": 0.0006389036544850498, + "step": 11370, + "tokens_trained": 37.246417296 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 3.2275015956315154, + "grad_norm": 0.15379244089126587, + "loss": 0.9479, + "loss_ce": 0.9516382217407227, + "loss_region": 0.0, + "loss_total": 0.9516382217407227, + "lr": 0.0006385714285714286, + "step": 11380, + "tokens_trained": 37.279182576 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 3.230338273881285, + "grad_norm": 0.29391154646873474, + "loss": 0.9554, + "loss_ce": 1.013104796409607, + "loss_region": 0.0, + "loss_total": 1.013104796409607, + "lr": 0.0006382392026578073, + "step": 11390, + "tokens_trained": 37.311948016 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 3.2331749521310544, + "grad_norm": 0.25678935647010803, + "loss": 0.9462, + "loss_ce": 1.1340473890304565, + "loss_region": 0.0, + "loss_total": 1.1340473890304565, + "lr": 0.0006379069767441861, + "step": 11400, + "tokens_trained": 37.344713456 + }, + { + "epoch": 3.2331749521310544, + "eval_ppl": 2.637687470176251, + "eval_runtime": 1.4177, + "step": 11400, + "tokens_trained": 37.344713456 + }, + { + "epoch": 3.2331749521310544, + "eval_ppl_cds": 3.06778139460225, + "eval_ppl_dig": 1.0984742910572258, + "eval_ppl_exon": 3.1680637826054094, + "eval_ppl_intron": 2.8100742631368822, + "eval_ppl_nig": 2.530391030495758, + "eval_ppl_promoter": 2.595410689722999, + "eval_ppl_utr": 2.8104915468344625, + "step": 11400, + "tokens_trained": 37.344713456 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 3.236011630380824, + "grad_norm": 0.09828726947307587, + "loss": 0.9565, + "loss_ce": 0.9265450239181519, + "loss_region": 0.0, + "loss_total": 0.9265450239181519, + "lr": 0.0006375747508305648, + "step": 11410, + "tokens_trained": 37.377478896 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 3.2388483086305935, + "grad_norm": 0.34177708625793457, + "loss": 0.9453, + "loss_ce": 0.9972662329673767, + "loss_region": 0.0, + "loss_total": 0.9972662329673767, + "lr": 0.0006372425249169436, + "step": 11420, + "tokens_trained": 37.410244336 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 3.241684986880363, + "grad_norm": 0.3409394919872284, + "loss": 0.9579, + "loss_ce": 1.090147614479065, + "loss_region": 0.0, + "loss_total": 1.090147614479065, + "lr": 0.0006369102990033223, + "step": 11430, + "tokens_trained": 37.443009776 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 3.2445216651301325, + "grad_norm": 0.42206576466560364, + "loss": 0.9608, + "loss_ce": 0.9614535570144653, + "loss_region": 0.0, + "loss_total": 0.9614535570144653, + "lr": 0.000636578073089701, + "step": 11440, + "tokens_trained": 37.4757752 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 3.247358343379902, + "grad_norm": 0.1503220945596695, + "loss": 0.9533, + "loss_ce": 0.9195599555969238, + "loss_region": 0.0, + "loss_total": 0.9195599555969238, + "lr": 0.0006362458471760797, + "step": 11450, + "tokens_trained": 37.50854064 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 3.2501950216296716, + "grad_norm": 0.3709371089935303, + "loss": 0.9551, + "loss_ce": 1.01471745967865, + "loss_region": 0.0, + "loss_total": 1.01471745967865, + "lr": 0.0006359136212624585, + "step": 11460, + "tokens_trained": 37.54130608 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 3.253031699879441, + "grad_norm": 0.24377739429473877, + "loss": 0.956, + "loss_ce": 0.8894580006599426, + "loss_region": 0.0, + "loss_total": 0.8894580006599426, + "lr": 0.0006355813953488372, + "step": 11470, + "tokens_trained": 37.57407152 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 3.2558683781292106, + "grad_norm": 0.20686297118663788, + "loss": 0.9503, + "loss_ce": 0.8176604509353638, + "loss_region": 0.0, + "loss_total": 0.8176604509353638, + "lr": 0.0006352491694352159, + "step": 11480, + "tokens_trained": 37.60683696 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 3.25870505637898, + "grad_norm": 0.1525667905807495, + "loss": 0.9469, + "loss_ce": 0.9944685697555542, + "loss_region": 0.0, + "loss_total": 0.9944685697555542, + "lr": 0.0006349169435215946, + "step": 11490, + "tokens_trained": 37.6396024 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 3.2615417346287496, + "grad_norm": 0.2898004651069641, + "loss": 0.9582, + "loss_ce": 1.0103609561920166, + "loss_region": 0.0, + "loss_total": 1.0103609561920166, + "lr": 0.0006345847176079735, + "step": 11500, + "tokens_trained": 37.67236692 + }, + { + "epoch": 3.2615417346287496, + "eval_ppl": 2.6401298879792736, + "eval_runtime": 1.4127, + "step": 11500, + "tokens_trained": 37.67236692 + }, + { + "epoch": 3.2615417346287496, + "eval_ppl_cds": 3.082339638404223, + "eval_ppl_dig": 1.0945182483303293, + "eval_ppl_exon": 3.1318781449757696, + "eval_ppl_intron": 2.814528376949657, + "eval_ppl_nig": 2.5310385827473016, + "eval_ppl_promoter": 2.6044154782511355, + "eval_ppl_utr": 2.801120064653944, + "step": 11500, + "tokens_trained": 37.67236692 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 3.264378412878519, + "grad_norm": 0.1535561978816986, + "loss": 0.9647, + "loss_ce": 0.6815728545188904, + "loss_region": 0.0, + "loss_total": 0.6815728545188904, + "lr": 0.0006342524916943522, + "step": 11510, + "tokens_trained": 37.70513236 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 3.2672150911282887, + "grad_norm": 0.6132323145866394, + "loss": 0.9562, + "loss_ce": 1.0589679479599, + "loss_region": 0.0, + "loss_total": 1.0589679479599, + "lr": 0.0006339202657807309, + "step": 11520, + "tokens_trained": 37.7378978 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 3.270051769378058, + "grad_norm": 0.4236352741718292, + "loss": 0.9677, + "loss_ce": 0.9900020360946655, + "loss_region": 0.0, + "loss_total": 0.9900020360946655, + "lr": 0.0006335880398671096, + "step": 11530, + "tokens_trained": 37.77066324 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 3.2728884476278277, + "grad_norm": 0.19762250781059265, + "loss": 0.9559, + "loss_ce": 0.9672777056694031, + "loss_region": 0.0, + "loss_total": 0.9672777056694031, + "lr": 0.0006332558139534884, + "step": 11540, + "tokens_trained": 37.80342868 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 3.2757251258775972, + "grad_norm": 0.14947308599948883, + "loss": 0.9522, + "loss_ce": 1.0436310768127441, + "loss_region": 0.0, + "loss_total": 1.0436310768127441, + "lr": 0.0006329235880398672, + "step": 11550, + "tokens_trained": 37.83619412 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 3.2785618041273668, + "grad_norm": 0.1421384960412979, + "loss": 0.9487, + "loss_ce": 0.9980403184890747, + "loss_region": 0.0, + "loss_total": 0.9980403184890747, + "lr": 0.0006325913621262459, + "step": 11560, + "tokens_trained": 37.86895796 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 3.2813984823771363, + "grad_norm": 0.3175358474254608, + "loss": 0.949, + "loss_ce": 0.7462145686149597, + "loss_region": 0.0, + "loss_total": 0.7462145686149597, + "lr": 0.0006322591362126247, + "step": 11570, + "tokens_trained": 37.9017234 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 3.284235160626906, + "grad_norm": 0.2967318892478943, + "loss": 0.9601, + "loss_ce": 1.0201125144958496, + "loss_region": 0.0, + "loss_total": 1.0201125144958496, + "lr": 0.0006319269102990033, + "step": 11580, + "tokens_trained": 37.93448884 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 3.2870718388766753, + "grad_norm": 0.23879355192184448, + "loss": 0.956, + "loss_ce": 0.9494261145591736, + "loss_region": 0.0, + "loss_total": 0.9494261145591736, + "lr": 0.000631594684385382, + "step": 11590, + "tokens_trained": 37.96725348 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 3.289908517126445, + "grad_norm": 0.21428297460079193, + "loss": 0.9641, + "loss_ce": 1.1193504333496094, + "loss_region": 0.0, + "loss_total": 1.1193504333496094, + "lr": 0.0006312624584717608, + "step": 11600, + "tokens_trained": 38.00001892 + }, + { + "epoch": 3.289908517126445, + "eval_ppl": 2.6352069444384054, + "eval_runtime": 1.4168, + "step": 11600, + "tokens_trained": 38.00001892 + }, + { + "epoch": 3.289908517126445, + "eval_ppl_cds": 3.1018347286739245, + "eval_ppl_dig": 1.0964726603510015, + "eval_ppl_exon": 3.163117912899122, + "eval_ppl_intron": 2.805083222519743, + "eval_ppl_nig": 2.5301270968505123, + "eval_ppl_promoter": 2.5901339354675206, + "eval_ppl_utr": 2.8356796769228736, + "step": 11600, + "tokens_trained": 38.00001892 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 3.2927451953762144, + "grad_norm": 0.3368994891643524, + "loss": 0.9408, + "loss_ce": 0.8660275936126709, + "loss_region": 0.0, + "loss_total": 0.8660275936126709, + "lr": 0.0006309302325581396, + "step": 11610, + "tokens_trained": 38.03278436 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 3.295581873625984, + "grad_norm": 0.5153021812438965, + "loss": 0.955, + "loss_ce": 0.9584696292877197, + "loss_region": 0.0, + "loss_total": 0.9584696292877197, + "lr": 0.0006305980066445183, + "step": 11620, + "tokens_trained": 38.0655498 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 3.2984185518757534, + "grad_norm": 0.27426061034202576, + "loss": 0.9557, + "loss_ce": 0.9420018792152405, + "loss_region": 0.0, + "loss_total": 0.9420018792152405, + "lr": 0.000630265780730897, + "step": 11630, + "tokens_trained": 38.09831524 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 3.301255230125523, + "grad_norm": 0.18444307148456573, + "loss": 0.9489, + "loss_ce": 0.8412085771560669, + "loss_region": 0.0, + "loss_total": 0.8412085771560669, + "lr": 0.0006299335548172757, + "step": 11640, + "tokens_trained": 38.13108068 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 3.3040919083752924, + "grad_norm": 0.3821467459201813, + "loss": 0.9584, + "loss_ce": 1.036531925201416, + "loss_region": 0.0, + "loss_total": 1.036531925201416, + "lr": 0.0006296013289036546, + "step": 11650, + "tokens_trained": 38.16384612 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 3.306928586625062, + "grad_norm": 0.27587890625, + "loss": 0.9466, + "loss_ce": 0.9778302311897278, + "loss_region": 0.0, + "loss_total": 0.9778302311897278, + "lr": 0.0006292691029900333, + "step": 11660, + "tokens_trained": 38.19661148 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 3.3097652648748315, + "grad_norm": 0.34393054246902466, + "loss": 0.9583, + "loss_ce": 0.976923406124115, + "loss_region": 0.0, + "loss_total": 0.976923406124115, + "lr": 0.000628936877076412, + "step": 11670, + "tokens_trained": 38.22937692 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 3.312601943124601, + "grad_norm": 0.24539527297019958, + "loss": 0.9454, + "loss_ce": 0.7070435881614685, + "loss_region": 0.0, + "loss_total": 0.7070435881614685, + "lr": 0.0006286046511627907, + "step": 11680, + "tokens_trained": 38.26214156 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 3.3154386213743705, + "grad_norm": 0.22384734451770782, + "loss": 0.949, + "loss_ce": 0.9038400053977966, + "loss_region": 0.0, + "loss_total": 0.9038400053977966, + "lr": 0.0006282724252491695, + "step": 11690, + "tokens_trained": 38.294907 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 3.31827529962414, + "grad_norm": 0.1247464194893837, + "loss": 0.9473, + "loss_ce": 1.0343053340911865, + "loss_region": 0.0, + "loss_total": 1.0343053340911865, + "lr": 0.0006279401993355482, + "step": 11700, + "tokens_trained": 38.32767244 + }, + { + "epoch": 3.31827529962414, + "eval_ppl": 2.6317262052742976, + "eval_runtime": 1.4171, + "step": 11700, + "tokens_trained": 38.32767244 + }, + { + "epoch": 3.31827529962414, + "eval_ppl_cds": 3.105489056146539, + "eval_ppl_dig": 1.0959944551743939, + "eval_ppl_exon": 3.1265602621561706, + "eval_ppl_intron": 2.7974548782281157, + "eval_ppl_nig": 2.5360796935326664, + "eval_ppl_promoter": 2.597383710741534, + "eval_ppl_utr": 2.8571163574592853, + "step": 11700, + "tokens_trained": 38.32767244 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 3.3211119778739095, + "grad_norm": 0.18657512962818146, + "loss": 0.947, + "loss_ce": 1.0242297649383545, + "loss_region": 0.0, + "loss_total": 1.0242297649383545, + "lr": 0.0006276079734219269, + "step": 11710, + "tokens_trained": 38.36043708 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 3.323948656123679, + "grad_norm": 0.442084401845932, + "loss": 0.9438, + "loss_ce": 1.0823557376861572, + "loss_region": 0.0, + "loss_total": 1.0823557376861572, + "lr": 0.0006272757475083056, + "step": 11720, + "tokens_trained": 38.39320252 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 3.3267853343734486, + "grad_norm": 0.3311280310153961, + "loss": 0.9452, + "loss_ce": 1.0395894050598145, + "loss_region": 0.0, + "loss_total": 1.0395894050598145, + "lr": 0.0006269435215946844, + "step": 11730, + "tokens_trained": 38.42596796 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 3.329622012623218, + "grad_norm": 0.3949175477027893, + "loss": 0.9565, + "loss_ce": 1.0577032566070557, + "loss_region": 0.0, + "loss_total": 1.0577032566070557, + "lr": 0.0006266112956810631, + "step": 11740, + "tokens_trained": 38.4587334 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 3.3324586908729876, + "grad_norm": 0.3584371507167816, + "loss": 0.9486, + "loss_ce": 0.9534963369369507, + "loss_region": 0.0, + "loss_total": 0.9534963369369507, + "lr": 0.0006262790697674419, + "step": 11750, + "tokens_trained": 38.49149884 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 3.335295369122757, + "grad_norm": 0.42444124817848206, + "loss": 0.9571, + "loss_ce": 1.0538257360458374, + "loss_region": 0.0, + "loss_total": 1.0538257360458374, + "lr": 0.0006259468438538206, + "step": 11760, + "tokens_trained": 38.524259664 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 3.3381320473725267, + "grad_norm": 0.22080418467521667, + "loss": 0.9599, + "loss_ce": 1.0074657201766968, + "loss_region": 0.0, + "loss_total": 1.0074657201766968, + "lr": 0.0006256146179401994, + "step": 11770, + "tokens_trained": 38.557025104 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 3.340968725622296, + "grad_norm": 0.32095688581466675, + "loss": 0.9646, + "loss_ce": 1.002951741218567, + "loss_region": 0.0, + "loss_total": 1.002951741218567, + "lr": 0.0006252823920265781, + "step": 11780, + "tokens_trained": 38.589790544 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 3.3438054038720657, + "grad_norm": 0.20919004082679749, + "loss": 0.9508, + "loss_ce": 0.9820741415023804, + "loss_region": 0.0, + "loss_total": 0.9820741415023804, + "lr": 0.0006249501661129568, + "step": 11790, + "tokens_trained": 38.622555984 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 3.3466420821218352, + "grad_norm": 0.43626120686531067, + "loss": 0.9556, + "loss_ce": 1.0207399129867554, + "loss_region": 0.0, + "loss_total": 1.0207399129867554, + "lr": 0.0006246179401993356, + "step": 11800, + "tokens_trained": 38.655321424 + }, + { + "epoch": 3.3466420821218352, + "eval_ppl": 2.634525817400975, + "eval_runtime": 1.4223, + "step": 11800, + "tokens_trained": 38.655321424 + }, + { + "epoch": 3.3466420821218352, + "eval_ppl_cds": 3.112501350974926, + "eval_ppl_dig": 1.0957834914370603, + "eval_ppl_exon": 3.121987380183722, + "eval_ppl_intron": 2.8060205478728824, + "eval_ppl_nig": 2.5307335019866364, + "eval_ppl_promoter": 2.6138685241662936, + "eval_ppl_utr": 2.8667201037568346, + "step": 11800, + "tokens_trained": 38.655321424 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 3.3494787603716047, + "grad_norm": 0.13183672726154327, + "loss": 0.9509, + "loss_ce": 1.0804741382598877, + "loss_region": 0.0, + "loss_total": 1.0804741382598877, + "lr": 0.0006242857142857144, + "step": 11810, + "tokens_trained": 38.688086864 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 3.3523154386213743, + "grad_norm": 0.20997831225395203, + "loss": 0.9563, + "loss_ce": 0.8840482234954834, + "loss_region": 0.0, + "loss_total": 0.8840482234954834, + "lr": 0.000623953488372093, + "step": 11820, + "tokens_trained": 38.720852304 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 3.355152116871144, + "grad_norm": 0.2296973615884781, + "loss": 0.9471, + "loss_ce": 0.9842191934585571, + "loss_region": 0.0, + "loss_total": 0.9842191934585571, + "lr": 0.0006236212624584717, + "step": 11830, + "tokens_trained": 38.753610568 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 3.3579887951209133, + "grad_norm": 0.2567266821861267, + "loss": 0.9523, + "loss_ce": 1.099923014640808, + "loss_region": 0.0, + "loss_total": 1.099923014640808, + "lr": 0.0006232890365448504, + "step": 11840, + "tokens_trained": 38.786376008 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 3.360825473370683, + "grad_norm": 0.5349225997924805, + "loss": 0.9547, + "loss_ce": 1.0821436643600464, + "loss_region": 0.0, + "loss_total": 1.0821436643600464, + "lr": 0.0006229568106312293, + "step": 11850, + "tokens_trained": 38.819136512 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 3.3636621516204523, + "grad_norm": 0.17492428421974182, + "loss": 0.9559, + "loss_ce": 0.8973754048347473, + "loss_region": 0.0, + "loss_total": 0.8973754048347473, + "lr": 0.000622624584717608, + "step": 11860, + "tokens_trained": 38.851901952 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 3.366498829870222, + "grad_norm": 0.14793908596038818, + "loss": 0.9444, + "loss_ce": 1.0035542249679565, + "loss_region": 0.0, + "loss_total": 1.0035542249679565, + "lr": 0.0006222923588039867, + "step": 11870, + "tokens_trained": 38.884667392 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 3.3693355081199914, + "grad_norm": 0.33121436834335327, + "loss": 0.9428, + "loss_ce": 0.910064160823822, + "loss_region": 0.0, + "loss_total": 0.910064160823822, + "lr": 0.0006219601328903654, + "step": 11880, + "tokens_trained": 38.917431224 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 3.372172186369761, + "grad_norm": 0.17440633475780487, + "loss": 0.9497, + "loss_ce": 0.7135308980941772, + "loss_region": 0.0, + "loss_total": 0.7135308980941772, + "lr": 0.0006216279069767442, + "step": 11890, + "tokens_trained": 38.950196664 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 3.3750088646195304, + "grad_norm": 0.2654556930065155, + "loss": 0.9493, + "loss_ce": 1.073326826095581, + "loss_region": 0.0, + "loss_total": 1.073326826095581, + "lr": 0.000621295681063123, + "step": 11900, + "tokens_trained": 38.982962096 + }, + { + "epoch": 3.3750088646195304, + "eval_ppl": 2.6256356810829447, + "eval_runtime": 1.421, + "step": 11900, + "tokens_trained": 38.982962096 + }, + { + "epoch": 3.3750088646195304, + "eval_ppl_cds": 3.091921741544279, + "eval_ppl_dig": 1.098476030188425, + "eval_ppl_exon": 3.1434096291749656, + "eval_ppl_intron": 2.7906067264573378, + "eval_ppl_nig": 2.521197642470606, + "eval_ppl_promoter": 2.577043063937752, + "eval_ppl_utr": 2.874760926068736, + "step": 11900, + "tokens_trained": 38.982962096 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 3.3778455428693, + "grad_norm": 0.3252089321613312, + "loss": 0.9567, + "loss_ce": 0.8784661293029785, + "loss_region": 0.0, + "loss_total": 0.8784661293029785, + "lr": 0.0006209634551495017, + "step": 11910, + "tokens_trained": 39.015725136 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 3.3806822211190695, + "grad_norm": 0.2935715317726135, + "loss": 0.9541, + "loss_ce": 1.051597237586975, + "loss_region": 0.0, + "loss_total": 1.051597237586975, + "lr": 0.0006206312292358805, + "step": 11920, + "tokens_trained": 39.048489776 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 3.383518899368839, + "grad_norm": 0.09987371414899826, + "loss": 0.9508, + "loss_ce": 1.0675004720687866, + "loss_region": 0.0, + "loss_total": 1.0675004720687866, + "lr": 0.0006202990033222592, + "step": 11930, + "tokens_trained": 39.081255216 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 3.3863555776186085, + "grad_norm": 0.292390912771225, + "loss": 0.9637, + "loss_ce": 0.9413557052612305, + "loss_region": 0.0, + "loss_total": 0.9413557052612305, + "lr": 0.0006199667774086379, + "step": 11940, + "tokens_trained": 39.114017592 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 3.389192255868378, + "grad_norm": 0.20926284790039062, + "loss": 0.9533, + "loss_ce": 0.9888216853141785, + "loss_region": 0.0, + "loss_total": 0.9888216853141785, + "lr": 0.0006196345514950165, + "step": 11950, + "tokens_trained": 39.146783032 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 3.3920289341181475, + "grad_norm": 0.19041214883327484, + "loss": 0.9496, + "loss_ce": 0.7865795493125916, + "loss_region": 0.0, + "loss_total": 0.7865795493125916, + "lr": 0.0006193023255813954, + "step": 11960, + "tokens_trained": 39.179548472 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 3.394865612367917, + "grad_norm": 0.31935620307922363, + "loss": 0.9465, + "loss_ce": 0.9386882781982422, + "loss_region": 0.0, + "loss_total": 0.9386882781982422, + "lr": 0.0006189700996677741, + "step": 11970, + "tokens_trained": 39.212313896 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 3.3977022906176866, + "grad_norm": 0.18576759099960327, + "loss": 0.9526, + "loss_ce": 0.8237330317497253, + "loss_region": 0.0, + "loss_total": 0.8237330317497253, + "lr": 0.0006186378737541528, + "step": 11980, + "tokens_trained": 39.245079336 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 3.400538968867456, + "grad_norm": 0.2894119322299957, + "loss": 0.9477, + "loss_ce": 1.0361120700836182, + "loss_region": 0.0, + "loss_total": 1.0361120700836182, + "lr": 0.0006183056478405315, + "step": 11990, + "tokens_trained": 39.277844616 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 3.4033756471172256, + "grad_norm": 0.15425419807434082, + "loss": 0.9528, + "loss_ce": 1.0357000827789307, + "loss_region": 0.0, + "loss_total": 1.0357000827789307, + "lr": 0.0006179734219269103, + "step": 12000, + "tokens_trained": 39.310610056 + }, + { + "epoch": 3.4033756471172256, + "eval_ppl": 2.6239206875206817, + "eval_runtime": 1.4232, + "step": 12000, + "tokens_trained": 39.310610056 + }, + { + "epoch": 3.4033756471172256, + "eval_ppl_cds": 3.097244012907773, + "eval_ppl_dig": 1.0992741942394952, + "eval_ppl_exon": 3.16630135144949, + "eval_ppl_intron": 2.7963019799612177, + "eval_ppl_nig": 2.529718594838619, + "eval_ppl_promoter": 2.582653589088057, + "eval_ppl_utr": 2.876505977275087, + "step": 12000, + "tokens_trained": 39.310610056 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 3.406212325366995, + "grad_norm": 0.06246515363454819, + "loss": 0.9508, + "loss_ce": 0.9340513944625854, + "loss_region": 0.0, + "loss_total": 0.9340513944625854, + "lr": 0.0006176411960132891, + "step": 12010, + "tokens_trained": 39.343375496 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 3.4090490036167647, + "grad_norm": 0.2621397078037262, + "loss": 0.9393, + "loss_ce": 1.0959811210632324, + "loss_region": 0.0, + "loss_total": 1.0959811210632324, + "lr": 0.0006173089700996678, + "step": 12020, + "tokens_trained": 39.376140936 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 3.411885681866534, + "grad_norm": 0.39492836594581604, + "loss": 0.9445, + "loss_ce": 1.124451994895935, + "loss_region": 0.0, + "loss_total": 1.124451994895935, + "lr": 0.0006169767441860465, + "step": 12030, + "tokens_trained": 39.408906376 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 3.4147223601163037, + "grad_norm": 0.3433813750743866, + "loss": 0.9549, + "loss_ce": 0.8775874972343445, + "loss_region": 0.0, + "loss_total": 0.8775874972343445, + "lr": 0.0006166445182724253, + "step": 12040, + "tokens_trained": 39.441671816 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 3.417559038366073, + "grad_norm": 0.23307295143604279, + "loss": 0.9465, + "loss_ce": 0.8307032585144043, + "loss_region": 0.0, + "loss_total": 0.8307032585144043, + "lr": 0.000616312292358804, + "step": 12050, + "tokens_trained": 39.474437256 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 3.4203957166158427, + "grad_norm": 0.34740176796913147, + "loss": 0.9546, + "loss_ce": 1.015007734298706, + "loss_region": 0.0, + "loss_total": 1.015007734298706, + "lr": 0.0006159800664451828, + "step": 12060, + "tokens_trained": 39.507202696 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 3.4232323948656123, + "grad_norm": 0.1054476723074913, + "loss": 0.9484, + "loss_ce": 1.0444495677947998, + "loss_region": 0.0, + "loss_total": 1.0444495677947998, + "lr": 0.0006156478405315614, + "step": 12070, + "tokens_trained": 39.539968136 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 3.426069073115382, + "grad_norm": 0.18345382809638977, + "loss": 0.9545, + "loss_ce": 0.7878419756889343, + "loss_region": 0.0, + "loss_total": 0.7878419756889343, + "lr": 0.0006153156146179402, + "step": 12080, + "tokens_trained": 39.572733576 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 3.4289057513651513, + "grad_norm": 0.21209219098091125, + "loss": 0.9537, + "loss_ce": 0.8085374236106873, + "loss_region": 0.0, + "loss_total": 0.8085374236106873, + "lr": 0.0006149833887043189, + "step": 12090, + "tokens_trained": 39.605498192 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 3.431742429614921, + "grad_norm": 0.26699861884117126, + "loss": 0.9548, + "loss_ce": 1.060858130455017, + "loss_region": 0.0, + "loss_total": 1.060858130455017, + "lr": 0.0006146511627906976, + "step": 12100, + "tokens_trained": 39.638259296 + }, + { + "epoch": 3.431742429614921, + "eval_ppl": 2.6229063307154985, + "eval_runtime": 1.4241, + "step": 12100, + "tokens_trained": 39.638259296 + }, + { + "epoch": 3.431742429614921, + "eval_ppl_cds": 3.107777877284338, + "eval_ppl_dig": 1.0983957082489293, + "eval_ppl_exon": 3.008535105241024, + "eval_ppl_intron": 2.7931422957270806, + "eval_ppl_nig": 2.520619015292159, + "eval_ppl_promoter": 2.5747879012999637, + "eval_ppl_utr": 2.879902280143258, + "step": 12100, + "tokens_trained": 39.638259296 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 3.4345791078646903, + "grad_norm": 0.12325490266084671, + "loss": 0.9522, + "loss_ce": 0.8527242541313171, + "loss_region": 0.0, + "loss_total": 0.8527242541313171, + "lr": 0.0006143189368770764, + "step": 12110, + "tokens_trained": 39.671023016 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 3.43741578611446, + "grad_norm": 0.3549254834651947, + "loss": 0.9429, + "loss_ce": 0.9839443564414978, + "loss_region": 0.0, + "loss_total": 0.9839443564414978, + "lr": 0.0006139867109634552, + "step": 12120, + "tokens_trained": 39.703787648 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 3.4402524643642294, + "grad_norm": 0.24043598771095276, + "loss": 0.9474, + "loss_ce": 0.7967652678489685, + "loss_region": 0.0, + "loss_total": 0.7967652678489685, + "lr": 0.0006136544850498339, + "step": 12130, + "tokens_trained": 39.73654832 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 3.443089142613999, + "grad_norm": 0.14909416437149048, + "loss": 0.9441, + "loss_ce": 0.9117237329483032, + "loss_region": 0.0, + "loss_total": 0.9117237329483032, + "lr": 0.0006133222591362126, + "step": 12140, + "tokens_trained": 39.76931376 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 3.4459258208637684, + "grad_norm": 0.19614329934120178, + "loss": 0.9468, + "loss_ce": 0.9683312773704529, + "loss_region": 0.0, + "loss_total": 0.9683312773704529, + "lr": 0.0006129900332225913, + "step": 12150, + "tokens_trained": 39.8020792 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 3.448762499113538, + "grad_norm": 0.3456745743751526, + "loss": 0.9506, + "loss_ce": 0.8271016478538513, + "loss_region": 0.0, + "loss_total": 0.8271016478538513, + "lr": 0.0006126578073089702, + "step": 12160, + "tokens_trained": 39.83484464 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 3.4515991773633075, + "grad_norm": 0.27946969866752625, + "loss": 0.9548, + "loss_ce": 0.9036428332328796, + "loss_region": 0.0, + "loss_total": 0.9036428332328796, + "lr": 0.0006123255813953489, + "step": 12170, + "tokens_trained": 39.86761008 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 3.454435855613077, + "grad_norm": 0.33434823155403137, + "loss": 0.9442, + "loss_ce": 0.8776450157165527, + "loss_region": 0.0, + "loss_total": 0.8776450157165527, + "lr": 0.0006119933554817276, + "step": 12180, + "tokens_trained": 39.90037552 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 3.4572725338628465, + "grad_norm": 0.10921714454889297, + "loss": 0.9538, + "loss_ce": 0.8836456537246704, + "loss_region": 0.0, + "loss_total": 0.8836456537246704, + "lr": 0.0006116611295681063, + "step": 12190, + "tokens_trained": 39.93314096 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 3.460109212112616, + "grad_norm": 0.36999791860580444, + "loss": 0.947, + "loss_ce": 0.8808584213256836, + "loss_region": 0.0, + "loss_total": 0.8808584213256836, + "lr": 0.000611328903654485, + "step": 12200, + "tokens_trained": 39.9659056 + }, + { + "epoch": 3.460109212112616, + "eval_ppl": 2.62441542058786, + "eval_runtime": 1.4341, + "step": 12200, + "tokens_trained": 39.9659056 + }, + { + "epoch": 3.460109212112616, + "eval_ppl_cds": 3.0848922411575614, + "eval_ppl_dig": 1.1003734472529256, + "eval_ppl_exon": 3.0873201595833604, + "eval_ppl_intron": 2.7891826958350605, + "eval_ppl_nig": 2.5201012246688315, + "eval_ppl_promoter": 2.5641570506625873, + "eval_ppl_utr": 2.8410405505317855, + "step": 12200, + "tokens_trained": 39.9659056 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 3.4629458903623855, + "grad_norm": 0.19262050092220306, + "loss": 0.9577, + "loss_ce": 0.9619672298431396, + "loss_region": 0.0, + "loss_total": 0.9619672298431396, + "lr": 0.0006109966777408638, + "step": 12210, + "tokens_trained": 39.99867104 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 3.465782568612155, + "grad_norm": 0.2488001435995102, + "loss": 0.9527, + "loss_ce": 0.8716427087783813, + "loss_region": 0.0, + "loss_total": 0.8716427087783813, + "lr": 0.0006106644518272425, + "step": 12220, + "tokens_trained": 40.03143648 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 3.4686192468619246, + "grad_norm": 0.41935738921165466, + "loss": 0.9524, + "loss_ce": 0.9753369688987732, + "loss_region": 0.0, + "loss_total": 0.9753369688987732, + "lr": 0.0006103322259136212, + "step": 12230, + "tokens_trained": 40.06420192 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 3.471455925111694, + "grad_norm": 0.26525360345840454, + "loss": 0.9536, + "loss_ce": 0.8010275363922119, + "loss_region": 0.0, + "loss_total": 0.8010275363922119, + "lr": 0.00061, + "step": 12240, + "tokens_trained": 40.09696312 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 3.4742926033614636, + "grad_norm": 0.21974614262580872, + "loss": 0.9479, + "loss_ce": 0.8798851370811462, + "loss_region": 0.0, + "loss_total": 0.8798851370811462, + "lr": 0.0006096677740863787, + "step": 12250, + "tokens_trained": 40.12972856 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 3.477129281611233, + "grad_norm": 0.4037272334098816, + "loss": 0.9441, + "loss_ce": 1.0752309560775757, + "loss_region": 0.0, + "loss_total": 1.0752309560775757, + "lr": 0.0006093355481727575, + "step": 12260, + "tokens_trained": 40.162494 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 3.4799659598610027, + "grad_norm": 0.24140764772891998, + "loss": 0.9508, + "loss_ce": 0.9578425288200378, + "loss_region": 0.0, + "loss_total": 0.9578425288200378, + "lr": 0.0006090033222591363, + "step": 12270, + "tokens_trained": 40.19525944 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 3.482802638110772, + "grad_norm": 0.20029568672180176, + "loss": 0.9515, + "loss_ce": 0.8523944616317749, + "loss_region": 0.0, + "loss_total": 0.8523944616317749, + "lr": 0.000608671096345515, + "step": 12280, + "tokens_trained": 40.228023272 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 3.4856393163605417, + "grad_norm": 0.09978004544973373, + "loss": 0.9485, + "loss_ce": 0.9497880935668945, + "loss_region": 0.0, + "loss_total": 0.9497880935668945, + "lr": 0.0006083388704318937, + "step": 12290, + "tokens_trained": 40.260787112 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 3.488475994610311, + "grad_norm": 0.29346227645874023, + "loss": 0.9569, + "loss_ce": 1.0877289772033691, + "loss_region": 0.0, + "loss_total": 1.0877289772033691, + "lr": 0.0006080066445182724, + "step": 12300, + "tokens_trained": 40.293551752 + }, + { + "epoch": 3.488475994610311, + "eval_ppl": 2.6252042458065, + "eval_runtime": 1.4208, + "step": 12300, + "tokens_trained": 40.293551752 + }, + { + "epoch": 3.488475994610311, + "eval_ppl_cds": 3.0692543269676316, + "eval_ppl_dig": 1.0991344996224006, + "eval_ppl_exon": 3.1165449843848436, + "eval_ppl_intron": 2.792120395189008, + "eval_ppl_nig": 2.5242255607061663, + "eval_ppl_promoter": 2.5657106175190076, + "eval_ppl_utr": 2.8590819110362795, + "step": 12300, + "tokens_trained": 40.293551752 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 3.4913126728600807, + "grad_norm": 0.30922287702560425, + "loss": 0.9448, + "loss_ce": 0.9467347860336304, + "loss_region": 0.0, + "loss_total": 0.9467347860336304, + "lr": 0.0006076744186046513, + "step": 12310, + "tokens_trained": 40.326317192 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 3.4941493511098503, + "grad_norm": 0.32134050130844116, + "loss": 0.9465, + "loss_ce": 1.0034700632095337, + "loss_region": 0.0, + "loss_total": 1.0034700632095337, + "lr": 0.0006073421926910299, + "step": 12320, + "tokens_trained": 40.359082632 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 3.4969860293596198, + "grad_norm": 0.43840786814689636, + "loss": 0.9527, + "loss_ce": 1.1029853820800781, + "loss_region": 0.0, + "loss_total": 1.1029853820800781, + "lr": 0.0006070099667774086, + "step": 12330, + "tokens_trained": 40.391848072 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 3.4998227076093893, + "grad_norm": 0.2021530270576477, + "loss": 0.9466, + "loss_ce": 1.0491234064102173, + "loss_region": 0.0, + "loss_total": 1.0491234064102173, + "lr": 0.0006066777408637873, + "step": 12340, + "tokens_trained": 40.424613512 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 3.502659385859159, + "grad_norm": 0.21949490904808044, + "loss": 0.9477, + "loss_ce": 1.0171594619750977, + "loss_region": 0.0, + "loss_total": 1.0171594619750977, + "lr": 0.0006063455149501661, + "step": 12350, + "tokens_trained": 40.457378952 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 3.5054960641089283, + "grad_norm": 0.3294987976551056, + "loss": 0.9393, + "loss_ce": 0.9809642434120178, + "loss_region": 0.0, + "loss_total": 0.9809642434120178, + "lr": 0.0006060132890365449, + "step": 12360, + "tokens_trained": 40.490144392 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 3.508332742358698, + "grad_norm": 0.28642022609710693, + "loss": 0.9469, + "loss_ce": 1.041870355606079, + "loss_region": 0.0, + "loss_total": 1.041870355606079, + "lr": 0.0006056810631229236, + "step": 12370, + "tokens_trained": 40.522909032 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 3.5111694206084674, + "grad_norm": 0.22936759889125824, + "loss": 0.9485, + "loss_ce": 1.0164532661437988, + "loss_region": 0.0, + "loss_total": 1.0164532661437988, + "lr": 0.0006053488372093023, + "step": 12380, + "tokens_trained": 40.555674472 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 3.514006098858237, + "grad_norm": 0.12295112013816833, + "loss": 0.9422, + "loss_ce": 0.72328782081604, + "loss_region": 0.0, + "loss_total": 0.72328782081604, + "lr": 0.0006050166112956811, + "step": 12390, + "tokens_trained": 40.588439912 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 3.5168427771080064, + "grad_norm": 0.3339201807975769, + "loss": 0.9477, + "loss_ce": 1.013103723526001, + "loss_region": 0.0, + "loss_total": 1.013103723526001, + "lr": 0.0006046843853820598, + "step": 12400, + "tokens_trained": 40.621205352 + }, + { + "epoch": 3.5168427771080064, + "eval_ppl": 2.622030362230799, + "eval_runtime": 1.4226, + "step": 12400, + "tokens_trained": 40.621205352 + }, + { + "epoch": 3.5168427771080064, + "eval_ppl_cds": 3.0917120124034394, + "eval_ppl_dig": 1.0988334171975207, + "eval_ppl_exon": 3.028897889284253, + "eval_ppl_intron": 2.793585913372453, + "eval_ppl_nig": 2.5189190304986684, + "eval_ppl_promoter": 2.570601489830848, + "eval_ppl_utr": 2.874916079918353, + "step": 12400, + "tokens_trained": 40.621205352 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 3.519679455357776, + "grad_norm": 0.14271755516529083, + "loss": 0.9502, + "loss_ce": 1.0582314729690552, + "loss_region": 0.0, + "loss_total": 1.0582314729690552, + "lr": 0.0006043521594684386, + "step": 12410, + "tokens_trained": 40.653970792 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 3.5225161336075455, + "grad_norm": 0.20899970829486847, + "loss": 0.9446, + "loss_ce": 0.960830807685852, + "loss_region": 0.0, + "loss_total": 0.960830807685852, + "lr": 0.0006040199335548173, + "step": 12420, + "tokens_trained": 40.686736232 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 3.525352811857315, + "grad_norm": 0.2057674378156662, + "loss": 0.9488, + "loss_ce": 1.0503498315811157, + "loss_region": 0.0, + "loss_total": 1.0503498315811157, + "lr": 0.0006036877076411961, + "step": 12430, + "tokens_trained": 40.719501672 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 3.5281894901070845, + "grad_norm": 0.3475935459136963, + "loss": 0.9488, + "loss_ce": 1.1495763063430786, + "loss_region": 0.0, + "loss_total": 1.1495763063430786, + "lr": 0.0006033554817275748, + "step": 12440, + "tokens_trained": 40.752267112 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 3.531026168356854, + "grad_norm": 0.16497746109962463, + "loss": 0.9521, + "loss_ce": 1.0105419158935547, + "loss_region": 0.0, + "loss_total": 1.0105419158935547, + "lr": 0.0006030232558139534, + "step": 12450, + "tokens_trained": 40.785032552 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 3.5338628466066235, + "grad_norm": 0.2899499535560608, + "loss": 0.9552, + "loss_ce": 1.0910053253173828, + "loss_region": 0.0, + "loss_total": 1.0910053253173828, + "lr": 0.0006026910299003322, + "step": 12460, + "tokens_trained": 40.817797992 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 3.536699524856393, + "grad_norm": 0.1915474236011505, + "loss": 0.9464, + "loss_ce": 1.0994977951049805, + "loss_region": 0.0, + "loss_total": 1.0994977951049805, + "lr": 0.000602358803986711, + "step": 12470, + "tokens_trained": 40.850563432 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 3.5395362031061626, + "grad_norm": 0.24397313594818115, + "loss": 0.9471, + "loss_ce": 1.0004920959472656, + "loss_region": 0.0, + "loss_total": 1.0004920959472656, + "lr": 0.0006020265780730897, + "step": 12480, + "tokens_trained": 40.883328872 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 3.542372881355932, + "grad_norm": 0.26217418909072876, + "loss": 0.9458, + "loss_ce": 0.712237536907196, + "loss_region": 0.0, + "loss_total": 0.712237536907196, + "lr": 0.0006016943521594684, + "step": 12490, + "tokens_trained": 40.916091784 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 3.5452095596057016, + "grad_norm": 0.1370612531900406, + "loss": 0.9445, + "loss_ce": 1.037952184677124, + "loss_region": 0.0, + "loss_total": 1.037952184677124, + "lr": 0.0006013621262458471, + "step": 12500, + "tokens_trained": 40.948854648 + }, + { + "epoch": 3.5452095596057016, + "eval_ppl": 2.6181752991078673, + "eval_runtime": 1.4167, + "step": 12500, + "tokens_trained": 40.948854648 + }, + { + "epoch": 3.5452095596057016, + "eval_ppl_cds": 3.091121804495149, + "eval_ppl_dig": 1.1004770357932234, + "eval_ppl_exon": 3.074601074557108, + "eval_ppl_intron": 2.787356732899468, + "eval_ppl_nig": 2.5216610811601172, + "eval_ppl_promoter": 2.5538148463737396, + "eval_ppl_utr": 2.8720237326587137, + "step": 12500, + "tokens_trained": 40.948854648 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 3.548046237855471, + "grad_norm": 0.18424645066261292, + "loss": 0.9479, + "loss_ce": 0.7886044383049011, + "loss_region": 0.0, + "loss_total": 0.7886044383049011, + "lr": 0.000601029900332226, + "step": 12510, + "tokens_trained": 40.981619512 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 3.5508829161052406, + "grad_norm": 0.40130680799484253, + "loss": 0.9593, + "loss_ce": 0.985082745552063, + "loss_region": 0.0, + "loss_total": 0.985082745552063, + "lr": 0.0006006976744186047, + "step": 12520, + "tokens_trained": 41.014384952 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 3.55371959435501, + "grad_norm": 0.2267717868089676, + "loss": 0.957, + "loss_ce": 0.9643515348434448, + "loss_region": 0.0, + "loss_total": 0.9643515348434448, + "lr": 0.0006003654485049834, + "step": 12530, + "tokens_trained": 41.047150392 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 3.5565562726047797, + "grad_norm": 0.14739426970481873, + "loss": 0.9473, + "loss_ce": 0.9906076192855835, + "loss_region": 0.0, + "loss_total": 0.9906076192855835, + "lr": 0.0006000332225913622, + "step": 12540, + "tokens_trained": 41.079910968 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 3.559392950854549, + "grad_norm": 0.28577426075935364, + "loss": 0.9466, + "loss_ce": 0.8776283860206604, + "loss_region": 0.0, + "loss_total": 0.8776283860206604, + "lr": 0.0005997009966777409, + "step": 12550, + "tokens_trained": 41.112676408 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 3.5622296291043187, + "grad_norm": 0.37244006991386414, + "loss": 0.9493, + "loss_ce": 1.0123413801193237, + "loss_region": 0.0, + "loss_total": 1.0123413801193237, + "lr": 0.0005993687707641197, + "step": 12560, + "tokens_trained": 41.145441848 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 3.5650663073540882, + "grad_norm": 0.13532061874866486, + "loss": 0.9537, + "loss_ce": 0.8353666067123413, + "loss_region": 0.0, + "loss_total": 0.8353666067123413, + "lr": 0.0005990365448504983, + "step": 12570, + "tokens_trained": 41.178207288 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 3.5679029856038578, + "grad_norm": 0.1363706886768341, + "loss": 0.9497, + "loss_ce": 0.8574619293212891, + "loss_region": 0.0, + "loss_total": 0.8574619293212891, + "lr": 0.000598704318936877, + "step": 12580, + "tokens_trained": 41.210972728 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 3.5707396638536273, + "grad_norm": 0.17625047266483307, + "loss": 0.9406, + "loss_ce": 0.9787061214447021, + "loss_region": 0.0, + "loss_total": 0.9787061214447021, + "lr": 0.0005983720930232558, + "step": 12590, + "tokens_trained": 41.243734016 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 3.573576342103397, + "grad_norm": 0.21298667788505554, + "loss": 0.945, + "loss_ce": 1.0493332147598267, + "loss_region": 0.0, + "loss_total": 1.0493332147598267, + "lr": 0.0005980398671096345, + "step": 12600, + "tokens_trained": 41.276498656 + }, + { + "epoch": 3.573576342103397, + "eval_ppl": 2.612269810117274, + "eval_runtime": 1.4223, + "step": 12600, + "tokens_trained": 41.276498656 + }, + { + "epoch": 3.573576342103397, + "eval_ppl_cds": 3.084323499791903, + "eval_ppl_dig": 1.099748376225578, + "eval_ppl_exon": 3.0537313272906754, + "eval_ppl_intron": 2.7807535740814617, + "eval_ppl_nig": 2.512105545418687, + "eval_ppl_promoter": 2.5412587553664916, + "eval_ppl_utr": 2.8436674398526893, + "step": 12600, + "tokens_trained": 41.276498656 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 3.5764130203531663, + "grad_norm": 0.33634713292121887, + "loss": 0.9503, + "loss_ce": 0.8533303737640381, + "loss_region": 0.0, + "loss_total": 0.8533303737640381, + "lr": 0.0005977076411960133, + "step": 12610, + "tokens_trained": 41.309264096 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 3.579249698602936, + "grad_norm": 0.14293387532234192, + "loss": 0.9438, + "loss_ce": 0.9591403007507324, + "loss_region": 0.0, + "loss_total": 0.9591403007507324, + "lr": 0.0005973754152823921, + "step": 12620, + "tokens_trained": 41.342029536 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 3.5820863768527054, + "grad_norm": 0.4260941445827484, + "loss": 0.9458, + "loss_ce": 0.9232190251350403, + "loss_region": 0.0, + "loss_total": 0.9232190251350403, + "lr": 0.0005970431893687708, + "step": 12630, + "tokens_trained": 41.374794176 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 3.584923055102475, + "grad_norm": 0.17284148931503296, + "loss": 0.9489, + "loss_ce": 0.8899317383766174, + "loss_region": 0.0, + "loss_total": 0.8899317383766174, + "lr": 0.0005967109634551495, + "step": 12640, + "tokens_trained": 41.407559616 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 3.5877597333522444, + "grad_norm": 0.30073729157447815, + "loss": 0.9461, + "loss_ce": 0.9793792366981506, + "loss_region": 0.0, + "loss_total": 0.9793792366981506, + "lr": 0.0005963787375415282, + "step": 12650, + "tokens_trained": 41.440325056 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 3.590596411602014, + "grad_norm": 0.25127941370010376, + "loss": 0.9437, + "loss_ce": 1.0668412446975708, + "loss_region": 0.0, + "loss_total": 1.0668412446975708, + "lr": 0.0005960465116279071, + "step": 12660, + "tokens_trained": 41.47308888 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 3.5934330898517834, + "grad_norm": 0.2902870178222656, + "loss": 0.9487, + "loss_ce": 1.0375736951828003, + "loss_region": 0.0, + "loss_total": 1.0375736951828003, + "lr": 0.0005957142857142858, + "step": 12670, + "tokens_trained": 41.50585432 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 3.596269768101553, + "grad_norm": 0.13351333141326904, + "loss": 0.9489, + "loss_ce": 0.8356289863586426, + "loss_region": 0.0, + "loss_total": 0.8356289863586426, + "lr": 0.0005953820598006645, + "step": 12680, + "tokens_trained": 41.53861896 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 3.5991064463513225, + "grad_norm": 0.1618579626083374, + "loss": 0.9399, + "loss_ce": 1.0330883264541626, + "loss_region": 0.0, + "loss_total": 1.0330883264541626, + "lr": 0.0005950498338870432, + "step": 12690, + "tokens_trained": 41.5713836 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 3.601943124601092, + "grad_norm": 0.3169199824333191, + "loss": 0.9469, + "loss_ce": 0.9600179195404053, + "loss_region": 0.0, + "loss_total": 0.9600179195404053, + "lr": 0.0005947176079734219, + "step": 12700, + "tokens_trained": 41.60414904 + }, + { + "epoch": 3.601943124601092, + "eval_ppl": 2.6146456783644227, + "eval_runtime": 1.4191, + "step": 12700, + "tokens_trained": 41.60414904 + }, + { + "epoch": 3.601943124601092, + "eval_ppl_cds": 3.1140516612186855, + "eval_ppl_dig": 1.101163191268421, + "eval_ppl_exon": 3.091921516744756, + "eval_ppl_intron": 2.789362854763435, + "eval_ppl_nig": 2.51530833928123, + "eval_ppl_promoter": 2.569751674188886, + "eval_ppl_utr": 2.8722517370726353, + "step": 12700, + "tokens_trained": 41.60414904 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 3.6047798028508615, + "grad_norm": 0.13512879610061646, + "loss": 0.9478, + "loss_ce": 0.8617331981658936, + "loss_region": 0.0, + "loss_total": 0.8617331981658936, + "lr": 0.0005943853820598007, + "step": 12710, + "tokens_trained": 41.636914416 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 3.607616481100631, + "grad_norm": 0.18713897466659546, + "loss": 0.939, + "loss_ce": 0.981195867061615, + "loss_region": 0.0, + "loss_total": 0.981195867061615, + "lr": 0.0005940531561461794, + "step": 12720, + "tokens_trained": 41.669679856 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 3.6104531593504006, + "grad_norm": 0.2848551869392395, + "loss": 0.9433, + "loss_ce": 1.0167688131332397, + "loss_region": 0.0, + "loss_total": 1.0167688131332397, + "lr": 0.0005937209302325581, + "step": 12730, + "tokens_trained": 41.702445296 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 3.61328983760017, + "grad_norm": 0.31109315156936646, + "loss": 0.9477, + "loss_ce": 0.9585043787956238, + "loss_region": 0.0, + "loss_total": 0.9585043787956238, + "lr": 0.0005933887043189369, + "step": 12740, + "tokens_trained": 41.73520896 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 3.6161265158499396, + "grad_norm": 0.11634766310453415, + "loss": 0.9415, + "loss_ce": 0.778812050819397, + "loss_region": 0.0, + "loss_total": 0.778812050819397, + "lr": 0.0005930564784053156, + "step": 12750, + "tokens_trained": 41.7679744 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 3.618963194099709, + "grad_norm": 0.11680868268013, + "loss": 0.9474, + "loss_ce": 1.1161885261535645, + "loss_region": 0.0, + "loss_total": 1.1161885261535645, + "lr": 0.0005927242524916944, + "step": 12760, + "tokens_trained": 41.80073984 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 3.6217998723494786, + "grad_norm": 0.31807950139045715, + "loss": 0.9497, + "loss_ce": 1.0034286975860596, + "loss_region": 0.0, + "loss_total": 1.0034286975860596, + "lr": 0.0005923920265780731, + "step": 12770, + "tokens_trained": 41.83350528 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 3.624636550599248, + "grad_norm": 0.17009148001670837, + "loss": 0.9415, + "loss_ce": 1.0413172245025635, + "loss_region": 0.0, + "loss_total": 1.0413172245025635, + "lr": 0.0005920598006644519, + "step": 12780, + "tokens_trained": 41.86627072 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 3.6274732288490177, + "grad_norm": 0.3388350307941437, + "loss": 0.9493, + "loss_ce": 0.7774043679237366, + "loss_region": 0.0, + "loss_total": 0.7774043679237366, + "lr": 0.0005917275747508306, + "step": 12790, + "tokens_trained": 41.89903616 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 3.630309907098787, + "grad_norm": 0.2695292830467224, + "loss": 0.9414, + "loss_ce": 1.03560209274292, + "loss_region": 0.0, + "loss_total": 1.03560209274292, + "lr": 0.0005913953488372093, + "step": 12800, + "tokens_trained": 41.9318016 + }, + { + "epoch": 3.630309907098787, + "eval_ppl": 2.6112381673549523, + "eval_runtime": 1.4298, + "step": 12800, + "tokens_trained": 41.9318016 + }, + { + "epoch": 3.630309907098787, + "eval_ppl_cds": 3.091192668413387, + "eval_ppl_dig": 1.0999046814960758, + "eval_ppl_exon": 3.1120980524578585, + "eval_ppl_intron": 2.781984232405517, + "eval_ppl_nig": 2.518258232911911, + "eval_ppl_promoter": 2.555930570415866, + "eval_ppl_utr": 2.8494429368783676, + "step": 12800, + "tokens_trained": 41.9318016 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 3.6331465853485567, + "grad_norm": 0.1865565925836563, + "loss": 0.9443, + "loss_ce": 0.8540247678756714, + "loss_region": 0.0, + "loss_total": 0.8540247678756714, + "lr": 0.0005910631229235882, + "step": 12810, + "tokens_trained": 41.96456704 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 3.6359832635983262, + "grad_norm": 0.11242073774337769, + "loss": 0.9399, + "loss_ce": 0.716280460357666, + "loss_region": 0.0, + "loss_total": 0.716280460357666, + "lr": 0.0005907308970099668, + "step": 12820, + "tokens_trained": 41.997332472 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 3.6388199418480958, + "grad_norm": 0.2709301710128784, + "loss": 0.9429, + "loss_ce": 1.0540114641189575, + "loss_region": 0.0, + "loss_total": 1.0540114641189575, + "lr": 0.0005903986710963455, + "step": 12830, + "tokens_trained": 42.030097912 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 3.6416566200978653, + "grad_norm": 0.22692221403121948, + "loss": 0.949, + "loss_ce": 0.9170104265213013, + "loss_region": 0.0, + "loss_total": 0.9170104265213013, + "lr": 0.0005900664451827242, + "step": 12840, + "tokens_trained": 42.062863352 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 3.644493298347635, + "grad_norm": 0.30304521322250366, + "loss": 0.9446, + "loss_ce": 1.020004391670227, + "loss_region": 0.0, + "loss_total": 1.020004391670227, + "lr": 0.0005897342192691029, + "step": 12850, + "tokens_trained": 42.09562296 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 3.6473299765974043, + "grad_norm": 0.13292884826660156, + "loss": 0.949, + "loss_ce": 0.9863396883010864, + "loss_region": 0.0, + "loss_total": 0.9863396883010864, + "lr": 0.0005894019933554818, + "step": 12860, + "tokens_trained": 42.1283884 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 3.650166654847174, + "grad_norm": 0.23564019799232483, + "loss": 0.9421, + "loss_ce": 0.9964564442634583, + "loss_region": 0.0, + "loss_total": 0.9964564442634583, + "lr": 0.0005890697674418605, + "step": 12870, + "tokens_trained": 42.16115384 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 3.6530033330969434, + "grad_norm": 0.15399199724197388, + "loss": 0.9514, + "loss_ce": 1.0713781118392944, + "loss_region": 0.0, + "loss_total": 1.0713781118392944, + "lr": 0.0005887375415282392, + "step": 12880, + "tokens_trained": 42.19391928 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 3.655840011346713, + "grad_norm": 0.2553921639919281, + "loss": 0.9567, + "loss_ce": 0.7739346623420715, + "loss_region": 0.0, + "loss_total": 0.7739346623420715, + "lr": 0.000588405315614618, + "step": 12890, + "tokens_trained": 42.22668472 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 3.6586766895964824, + "grad_norm": 0.17485478520393372, + "loss": 0.9358, + "loss_ce": 1.0984776020050049, + "loss_region": 0.0, + "loss_total": 1.0984776020050049, + "lr": 0.0005880730897009967, + "step": 12900, + "tokens_trained": 42.25944936 + }, + { + "epoch": 3.6586766895964824, + "eval_ppl": 2.6086164510199463, + "eval_runtime": 1.423, + "step": 12900, + "tokens_trained": 42.25944936 + }, + { + "epoch": 3.6586766895964824, + "eval_ppl_cds": 3.0695194319658303, + "eval_ppl_dig": 1.100538122967987, + "eval_ppl_exon": 3.1211441034202743, + "eval_ppl_intron": 2.785688581009607, + "eval_ppl_nig": 2.524085565245664, + "eval_ppl_promoter": 2.5551337992360277, + "eval_ppl_utr": 2.8611951768845443, + "step": 12900, + "tokens_trained": 42.25944936 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 3.661513367846252, + "grad_norm": 0.09438804537057877, + "loss": 0.9476, + "loss_ce": 0.8398815989494324, + "loss_region": 0.0, + "loss_total": 0.8398815989494324, + "lr": 0.0005877408637873755, + "step": 12910, + "tokens_trained": 42.292212712 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 3.6643500460960214, + "grad_norm": 0.33867543935775757, + "loss": 0.9476, + "loss_ce": 0.8592687845230103, + "loss_region": 0.0, + "loss_total": 0.8592687845230103, + "lr": 0.0005874086378737542, + "step": 12920, + "tokens_trained": 42.324978152 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 3.667186724345791, + "grad_norm": 0.24801434576511383, + "loss": 0.9335, + "loss_ce": 1.0160895586013794, + "loss_region": 0.0, + "loss_total": 1.0160895586013794, + "lr": 0.000587076411960133, + "step": 12930, + "tokens_trained": 42.357741992 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 3.6700234025955605, + "grad_norm": 0.20852713286876678, + "loss": 0.943, + "loss_ce": 1.1322470903396606, + "loss_region": 0.0, + "loss_total": 1.1322470903396606, + "lr": 0.0005867441860465116, + "step": 12940, + "tokens_trained": 42.390507432 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 3.67286008084533, + "grad_norm": 0.3159957230091095, + "loss": 0.9514, + "loss_ce": 0.9900782704353333, + "loss_region": 0.0, + "loss_total": 0.9900782704353333, + "lr": 0.0005864119601328903, + "step": 12950, + "tokens_trained": 42.423272872 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 3.6756967590950995, + "grad_norm": 0.15550002455711365, + "loss": 0.9382, + "loss_ce": 0.6979476809501648, + "loss_region": 0.0, + "loss_total": 0.6979476809501648, + "lr": 0.000586079734219269, + "step": 12960, + "tokens_trained": 42.456038312 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 3.678533437344869, + "grad_norm": 0.2368851900100708, + "loss": 0.9521, + "loss_ce": 0.8380557298660278, + "loss_region": 0.0, + "loss_total": 0.8380557298660278, + "lr": 0.0005857475083056479, + "step": 12970, + "tokens_trained": 42.488803752 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 3.6813701155946386, + "grad_norm": 0.18330933153629303, + "loss": 0.9469, + "loss_ce": 0.8510177135467529, + "loss_region": 0.0, + "loss_total": 0.8510177135467529, + "lr": 0.0005854152823920266, + "step": 12980, + "tokens_trained": 42.521569192 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 3.684206793844408, + "grad_norm": 0.3295491933822632, + "loss": 0.9361, + "loss_ce": 1.0580601692199707, + "loss_region": 0.0, + "loss_total": 1.0580601692199707, + "lr": 0.0005850830564784053, + "step": 12990, + "tokens_trained": 42.554334632 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 3.6870434720941776, + "grad_norm": 0.16922013461589813, + "loss": 0.9446, + "loss_ce": 0.924680233001709, + "loss_region": 0.0, + "loss_total": 0.924680233001709, + "lr": 0.000584750830564784, + "step": 13000, + "tokens_trained": 42.587100056 + }, + { + "epoch": 3.6870434720941776, + "eval_ppl": 2.6072690494453243, + "eval_runtime": 1.4159, + "step": 13000, + "tokens_trained": 42.587100056 + }, + { + "epoch": 3.6870434720941776, + "eval_ppl_cds": 3.1000104400144624, + "eval_ppl_dig": 1.0998357097520117, + "eval_ppl_exon": 3.1309960685217058, + "eval_ppl_intron": 2.7877596739420247, + "eval_ppl_nig": 2.5248483437474682, + "eval_ppl_promoter": 2.5514354918018727, + "eval_ppl_utr": 2.8444304832239125, + "step": 13000, + "tokens_trained": 42.587100056 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 3.689880150343947, + "grad_norm": 0.26088443398475647, + "loss": 0.9478, + "loss_ce": 0.8114020228385925, + "loss_region": 0.0, + "loss_total": 0.8114020228385925, + "lr": 0.0005844186046511628, + "step": 13010, + "tokens_trained": 42.619862168 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 3.6927168285937166, + "grad_norm": 0.22251196205615997, + "loss": 0.9493, + "loss_ce": 0.9486069679260254, + "loss_region": 0.0, + "loss_total": 0.9486069679260254, + "lr": 0.0005840863787375416, + "step": 13020, + "tokens_trained": 42.652627608 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 3.695553506843486, + "grad_norm": 0.19181273877620697, + "loss": 0.9383, + "loss_ce": 1.028370976448059, + "loss_region": 0.0, + "loss_total": 1.028370976448059, + "lr": 0.0005837541528239203, + "step": 13030, + "tokens_trained": 42.685392248 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 3.6983901850932557, + "grad_norm": 0.36815327405929565, + "loss": 0.9502, + "loss_ce": 1.0881353616714478, + "loss_region": 0.0, + "loss_total": 1.0881353616714478, + "lr": 0.000583421926910299, + "step": 13040, + "tokens_trained": 42.718156072 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 3.701226863343025, + "grad_norm": 0.34447288513183594, + "loss": 0.958, + "loss_ce": 0.9004485011100769, + "loss_region": 0.0, + "loss_total": 0.9004485011100769, + "lr": 0.0005830897009966778, + "step": 13050, + "tokens_trained": 42.750920712 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 3.7040635415927947, + "grad_norm": 0.24551263451576233, + "loss": 0.9549, + "loss_ce": 0.9561207294464111, + "loss_region": 0.0, + "loss_total": 0.9561207294464111, + "lr": 0.0005827574750830565, + "step": 13060, + "tokens_trained": 42.783686152 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 3.7069002198425642, + "grad_norm": 0.1799919605255127, + "loss": 0.9338, + "loss_ce": 1.113955020904541, + "loss_region": 0.0, + "loss_total": 1.113955020904541, + "lr": 0.0005824252491694352, + "step": 13070, + "tokens_trained": 42.816449192 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 3.7097368980923338, + "grad_norm": 0.16731490194797516, + "loss": 0.9471, + "loss_ce": 1.0577377080917358, + "loss_region": 0.0, + "loss_total": 1.0577377080917358, + "lr": 0.0005820930232558139, + "step": 13080, + "tokens_trained": 42.849214632 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 3.7125735763421033, + "grad_norm": 0.207774817943573, + "loss": 0.9438, + "loss_ce": 0.9111372828483582, + "loss_region": 0.0, + "loss_total": 0.9111372828483582, + "lr": 0.0005817607973421927, + "step": 13090, + "tokens_trained": 42.881978504 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 3.715410254591873, + "grad_norm": 0.27425700426101685, + "loss": 0.942, + "loss_ce": 0.9448755979537964, + "loss_region": 0.0, + "loss_total": 0.9448755979537964, + "lr": 0.0005814285714285714, + "step": 13100, + "tokens_trained": 42.914743784 + }, + { + "epoch": 3.715410254591873, + "eval_ppl": 2.6085279811813975, + "eval_runtime": 1.4146, + "step": 13100, + "tokens_trained": 42.914743784 + }, + { + "epoch": 3.715410254591873, + "eval_ppl_cds": 3.056955116582625, + "eval_ppl_dig": 1.0948919014160883, + "eval_ppl_exon": 3.1392225420073694, + "eval_ppl_intron": 2.7835429616637186, + "eval_ppl_nig": 2.5169821564222956, + "eval_ppl_promoter": 2.5272943545666204, + "eval_ppl_utr": 2.8360288262679068, + "step": 13100, + "tokens_trained": 42.914743784 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 3.7182469328416423, + "grad_norm": 0.12981437146663666, + "loss": 0.9418, + "loss_ce": 0.9193331599235535, + "loss_region": 0.0, + "loss_total": 0.9193331599235535, + "lr": 0.0005810963455149501, + "step": 13110, + "tokens_trained": 42.947506512 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 3.721083611091412, + "grad_norm": 0.18780262768268585, + "loss": 0.9321, + "loss_ce": 0.9367557764053345, + "loss_region": 0.0, + "loss_total": 0.9367557764053345, + "lr": 0.000580764119601329, + "step": 13120, + "tokens_trained": 42.980271952 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 3.7239202893411814, + "grad_norm": 0.23737166821956635, + "loss": 0.9453, + "loss_ce": 0.9894770979881287, + "loss_region": 0.0, + "loss_total": 0.9894770979881287, + "lr": 0.0005804318936877077, + "step": 13130, + "tokens_trained": 43.013037392 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 3.726756967590951, + "grad_norm": 0.13469639420509338, + "loss": 0.9433, + "loss_ce": 1.0237447023391724, + "loss_region": 0.0, + "loss_total": 1.0237447023391724, + "lr": 0.0005800996677740864, + "step": 13140, + "tokens_trained": 43.045802832 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 3.7295936458407204, + "grad_norm": 0.19024422764778137, + "loss": 0.9469, + "loss_ce": 1.0295419692993164, + "loss_region": 0.0, + "loss_total": 1.0295419692993164, + "lr": 0.0005797674418604651, + "step": 13150, + "tokens_trained": 43.078568272 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 3.73243032409049, + "grad_norm": 0.25045034289360046, + "loss": 0.9453, + "loss_ce": 0.8829964995384216, + "loss_region": 0.0, + "loss_total": 0.8829964995384216, + "lr": 0.0005794352159468438, + "step": 13160, + "tokens_trained": 43.111332912 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 3.7352670023402594, + "grad_norm": 0.25939199328422546, + "loss": 0.9534, + "loss_ce": 1.0155121088027954, + "loss_region": 0.0, + "loss_total": 1.0155121088027954, + "lr": 0.0005791029900332227, + "step": 13170, + "tokens_trained": 43.144098192 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 3.738103680590029, + "grad_norm": 0.22645778954029083, + "loss": 0.9479, + "loss_ce": 1.0778868198394775, + "loss_region": 0.0, + "loss_total": 1.0778868198394775, + "lr": 0.0005787707641196014, + "step": 13180, + "tokens_trained": 43.176863632 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 3.7409403588397985, + "grad_norm": 0.134956493973732, + "loss": 0.9413, + "loss_ce": 0.8266254663467407, + "loss_region": 0.0, + "loss_total": 0.8266254663467407, + "lr": 0.00057843853820598, + "step": 13190, + "tokens_trained": 43.209629072 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 3.743777037089568, + "grad_norm": 0.15215621888637543, + "loss": 0.9397, + "loss_ce": 0.8877963423728943, + "loss_region": 0.0, + "loss_total": 0.8877963423728943, + "lr": 0.0005781063122923587, + "step": 13200, + "tokens_trained": 43.242394496 + }, + { + "epoch": 3.743777037089568, + "eval_ppl": 2.6090261879013603, + "eval_runtime": 1.4194, + "step": 13200, + "tokens_trained": 43.242394496 + }, + { + "epoch": 3.743777037089568, + "eval_ppl_cds": 3.0346852565171067, + "eval_ppl_dig": 1.0991188800895955, + "eval_ppl_exon": 3.075480049932689, + "eval_ppl_intron": 2.7797357291977987, + "eval_ppl_nig": 2.51411870153755, + "eval_ppl_promoter": 2.5211204945022554, + "eval_ppl_utr": 2.8258784897952043, + "step": 13200, + "tokens_trained": 43.242394496 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 3.7466137153393375, + "grad_norm": 0.3049950897693634, + "loss": 0.9553, + "loss_ce": 0.8566665053367615, + "loss_region": 0.0, + "loss_total": 0.8566665053367615, + "lr": 0.0005777740863787375, + "step": 13210, + "tokens_trained": 43.275159936 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 3.749450393589107, + "grad_norm": 0.17330819368362427, + "loss": 0.9383, + "loss_ce": 0.888554573059082, + "loss_region": 0.0, + "loss_total": 0.888554573059082, + "lr": 0.0005774418604651163, + "step": 13220, + "tokens_trained": 43.307924568 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 3.7522870718388766, + "grad_norm": 0.19180597364902496, + "loss": 0.9477, + "loss_ce": 1.0416995286941528, + "loss_region": 0.0, + "loss_total": 1.0416995286941528, + "lr": 0.000577109634551495, + "step": 13230, + "tokens_trained": 43.340690008 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 3.755123750088646, + "grad_norm": 0.22475208342075348, + "loss": 0.9393, + "loss_ce": 0.8465211987495422, + "loss_region": 0.0, + "loss_total": 0.8465211987495422, + "lr": 0.0005767774086378738, + "step": 13240, + "tokens_trained": 43.373452136 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 3.7579604283384156, + "grad_norm": 0.14102283120155334, + "loss": 0.9452, + "loss_ce": 0.9070911407470703, + "loss_region": 0.0, + "loss_total": 0.9070911407470703, + "lr": 0.0005764451827242525, + "step": 13250, + "tokens_trained": 43.406217576 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 3.760797106588185, + "grad_norm": 0.1408279538154602, + "loss": 0.9409, + "loss_ce": 0.935166597366333, + "loss_region": 0.0, + "loss_total": 0.935166597366333, + "lr": 0.0005761129568106312, + "step": 13260, + "tokens_trained": 43.438983016 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 3.7636337848379546, + "grad_norm": 0.20332716405391693, + "loss": 0.9481, + "loss_ce": 1.068856954574585, + "loss_region": 0.0, + "loss_total": 1.068856954574585, + "lr": 0.00057578073089701, + "step": 13270, + "tokens_trained": 43.471748456 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 3.766470463087724, + "grad_norm": 0.1404893696308136, + "loss": 0.9561, + "loss_ce": 1.0914115905761719, + "loss_region": 0.0, + "loss_total": 1.0914115905761719, + "lr": 0.0005754485049833888, + "step": 13280, + "tokens_trained": 43.504513896 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 3.7693071413374937, + "grad_norm": 0.205766499042511, + "loss": 0.9355, + "loss_ce": 0.8861879706382751, + "loss_region": 0.0, + "loss_total": 0.8861879706382751, + "lr": 0.0005751162790697675, + "step": 13290, + "tokens_trained": 43.537279336 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 3.772143819587263, + "grad_norm": 0.08984963595867157, + "loss": 0.9436, + "loss_ce": 0.8197200298309326, + "loss_region": 0.0, + "loss_total": 0.8197200298309326, + "lr": 0.0005747840531561462, + "step": 13300, + "tokens_trained": 43.570044776 + }, + { + "epoch": 3.772143819587263, + "eval_ppl": 2.605949836513458, + "eval_runtime": 1.4191, + "step": 13300, + "tokens_trained": 43.570044776 + }, + { + "epoch": 3.772143819587263, + "eval_ppl_cds": 3.0383826838032895, + "eval_ppl_dig": 1.0980829308596707, + "eval_ppl_exon": 3.11121321104423, + "eval_ppl_intron": 2.7756843988262907, + "eval_ppl_nig": 2.5097469920133726, + "eval_ppl_promoter": 2.5188318427061533, + "eval_ppl_utr": 2.8109071526818936, + "step": 13300, + "tokens_trained": 43.570044776 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 3.7749804978370327, + "grad_norm": 0.21549661457538605, + "loss": 0.9427, + "loss_ce": 0.9878718256950378, + "loss_region": 0.0, + "loss_total": 0.9878718256950378, + "lr": 0.0005744518272425249, + "step": 13310, + "tokens_trained": 43.602810216 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 3.7778171760868022, + "grad_norm": 0.16967253386974335, + "loss": 0.9423, + "loss_ce": 0.8229381442070007, + "loss_region": 0.0, + "loss_total": 0.8229381442070007, + "lr": 0.0005741196013289037, + "step": 13320, + "tokens_trained": 43.635575656 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 3.7806538543365718, + "grad_norm": 0.23282964527606964, + "loss": 0.9586, + "loss_ce": 0.9853363633155823, + "loss_region": 0.0, + "loss_total": 0.9853363633155823, + "lr": 0.0005737873754152824, + "step": 13330, + "tokens_trained": 43.668339976 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 3.7834905325863413, + "grad_norm": 0.27506309747695923, + "loss": 0.936, + "loss_ce": 0.9311396479606628, + "loss_region": 0.0, + "loss_total": 0.9311396479606628, + "lr": 0.0005734551495016611, + "step": 13340, + "tokens_trained": 43.701105416 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 3.786327210836111, + "grad_norm": 0.1773795634508133, + "loss": 0.9403, + "loss_ce": 0.7607808113098145, + "loss_region": 0.0, + "loss_total": 0.7607808113098145, + "lr": 0.0005731229235880398, + "step": 13350, + "tokens_trained": 43.73386788 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 3.7891638890858803, + "grad_norm": 0.23445580899715424, + "loss": 0.9456, + "loss_ce": 0.8150612115859985, + "loss_region": 0.0, + "loss_total": 0.8150612115859985, + "lr": 0.0005727906976744186, + "step": 13360, + "tokens_trained": 43.76663332 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 3.79200056733565, + "grad_norm": 0.38194015622138977, + "loss": 0.9471, + "loss_ce": 0.7512699365615845, + "loss_region": 0.0, + "loss_total": 0.7512699365615845, + "lr": 0.0005724584717607974, + "step": 13370, + "tokens_trained": 43.79939876 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 3.7948372455854193, + "grad_norm": 0.1635681837797165, + "loss": 0.9401, + "loss_ce": 0.9618135094642639, + "loss_region": 0.0, + "loss_total": 0.9618135094642639, + "lr": 0.0005721262458471761, + "step": 13380, + "tokens_trained": 43.8321642 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 3.797673923835189, + "grad_norm": 0.36317092180252075, + "loss": 0.9388, + "loss_ce": 0.9944338202476501, + "loss_region": 0.0, + "loss_total": 0.9944338202476501, + "lr": 0.0005717940199335548, + "step": 13390, + "tokens_trained": 43.86492964 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 3.8005106020849584, + "grad_norm": 0.16642428934574127, + "loss": 0.943, + "loss_ce": 0.7773996591567993, + "loss_region": 0.0, + "loss_total": 0.7773996591567993, + "lr": 0.0005714617940199336, + "step": 13400, + "tokens_trained": 43.89769508 + }, + { + "epoch": 3.8005106020849584, + "eval_ppl": 2.6031725226201807, + "eval_runtime": 1.4263, + "step": 13400, + "tokens_trained": 43.89769508 + }, + { + "epoch": 3.8005106020849584, + "eval_ppl_cds": 3.056358699272338, + "eval_ppl_dig": 1.0998155487906889, + "eval_ppl_exon": 3.1385334319756972, + "eval_ppl_intron": 2.7813740901713793, + "eval_ppl_nig": 2.505663566974777, + "eval_ppl_promoter": 2.528178155349829, + "eval_ppl_utr": 2.8300342055302834, + "step": 13400, + "tokens_trained": 43.89769508 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 3.803347280334728, + "grad_norm": 0.18868686258792877, + "loss": 0.9473, + "loss_ce": 0.9924156069755554, + "loss_region": 0.0, + "loss_total": 0.9924156069755554, + "lr": 0.0005711295681063123, + "step": 13410, + "tokens_trained": 43.930457968 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 3.8061839585844974, + "grad_norm": 0.0958852469921112, + "loss": 0.9456, + "loss_ce": 1.0558315515518188, + "loss_region": 0.0, + "loss_total": 1.0558315515518188, + "lr": 0.0005707973421926911, + "step": 13420, + "tokens_trained": 43.963223408 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 3.809020636834267, + "grad_norm": 0.2677806615829468, + "loss": 0.935, + "loss_ce": 0.9139768481254578, + "loss_region": 0.0, + "loss_total": 0.9139768481254578, + "lr": 0.0005704651162790698, + "step": 13430, + "tokens_trained": 43.995988848 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 3.8118573150840365, + "grad_norm": 0.08105498552322388, + "loss": 0.9496, + "loss_ce": 0.9813448190689087, + "loss_region": 0.0, + "loss_total": 0.9813448190689087, + "lr": 0.0005701328903654485, + "step": 13440, + "tokens_trained": 44.028750544 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 3.814693993333806, + "grad_norm": 0.30944448709487915, + "loss": 0.9438, + "loss_ce": 0.9254712462425232, + "loss_region": 0.0, + "loss_total": 0.9254712462425232, + "lr": 0.0005698006644518272, + "step": 13450, + "tokens_trained": 44.061515184 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 3.8175306715835755, + "grad_norm": 0.19011414051055908, + "loss": 0.9474, + "loss_ce": 0.7497115135192871, + "loss_region": 0.0, + "loss_total": 0.7497115135192871, + "lr": 0.0005694684385382059, + "step": 13460, + "tokens_trained": 44.094280624 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 3.820367349833345, + "grad_norm": 0.1864045113325119, + "loss": 0.9449, + "loss_ce": 1.0778380632400513, + "loss_region": 0.0, + "loss_total": 1.0778380632400513, + "lr": 0.0005691362126245847, + "step": 13470, + "tokens_trained": 44.127043664 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 3.8232040280831145, + "grad_norm": 0.14197470247745514, + "loss": 0.9501, + "loss_ce": 0.9996015429496765, + "loss_region": 0.0, + "loss_total": 0.9996015429496765, + "lr": 0.0005688039867109635, + "step": 13480, + "tokens_trained": 44.159809104 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 3.826040706332884, + "grad_norm": 0.2649572491645813, + "loss": 0.9465, + "loss_ce": 0.723941445350647, + "loss_region": 0.0, + "loss_total": 0.723941445350647, + "lr": 0.0005684717607973422, + "step": 13490, + "tokens_trained": 44.192574544 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 3.8288773845826536, + "grad_norm": 0.24132545292377472, + "loss": 0.9442, + "loss_ce": 0.9906350374221802, + "loss_region": 0.0, + "loss_total": 0.9906350374221802, + "lr": 0.0005681395348837209, + "step": 13500, + "tokens_trained": 44.225339984 + }, + { + "epoch": 3.8288773845826536, + "eval_ppl": 2.6026256369376894, + "eval_runtime": 1.4349, + "step": 13500, + "tokens_trained": 44.225339984 + }, + { + "epoch": 3.8288773845826536, + "eval_ppl_cds": 3.0784665394726844, + "eval_ppl_dig": 1.098586711646851, + "eval_ppl_exon": 3.0349732077598075, + "eval_ppl_intron": 2.7795911864061273, + "eval_ppl_nig": 2.5085631009344094, + "eval_ppl_promoter": 2.5295712243058177, + "eval_ppl_utr": 2.850062747907543, + "step": 13500, + "tokens_trained": 44.225339984 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 3.831714062832423, + "grad_norm": 0.10767563432455063, + "loss": 0.941, + "loss_ce": 0.9765941500663757, + "loss_region": 0.0, + "loss_total": 0.9765941500663757, + "lr": 0.0005678073089700996, + "step": 13510, + "tokens_trained": 44.258105424 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 3.8345507410821926, + "grad_norm": 0.18366152048110962, + "loss": 0.939, + "loss_ce": 0.859066367149353, + "loss_region": 0.0, + "loss_total": 0.859066367149353, + "lr": 0.0005674750830564785, + "step": 13520, + "tokens_trained": 44.290870864 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 3.837387419331962, + "grad_norm": 0.15339790284633636, + "loss": 0.9373, + "loss_ce": 0.8971619606018066, + "loss_region": 0.0, + "loss_total": 0.8971619606018066, + "lr": 0.0005671428571428572, + "step": 13530, + "tokens_trained": 44.323635304 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 3.8402240975817317, + "grad_norm": 0.2694911062717438, + "loss": 0.9266, + "loss_ce": 1.016292929649353, + "loss_region": 0.0, + "loss_total": 1.016292929649353, + "lr": 0.0005668106312292359, + "step": 13540, + "tokens_trained": 44.356400744 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 3.843060775831501, + "grad_norm": 0.18835560977458954, + "loss": 0.9576, + "loss_ce": 1.1038981676101685, + "loss_region": 0.0, + "loss_total": 1.1038981676101685, + "lr": 0.0005664784053156147, + "step": 13550, + "tokens_trained": 44.389166184 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 3.8458974540812707, + "grad_norm": 0.40701913833618164, + "loss": 0.9446, + "loss_ce": 1.0079967975616455, + "loss_region": 0.0, + "loss_total": 1.0079967975616455, + "lr": 0.0005661461794019934, + "step": 13560, + "tokens_trained": 44.421931624 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 3.8487341323310402, + "grad_norm": 0.22316637635231018, + "loss": 0.9416, + "loss_ce": 0.9914200305938721, + "loss_region": 0.0, + "loss_total": 0.9914200305938721, + "lr": 0.0005658139534883721, + "step": 13570, + "tokens_trained": 44.454697064 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 3.8515708105808097, + "grad_norm": 0.1183377355337143, + "loss": 0.9352, + "loss_ce": 0.7328471541404724, + "loss_region": 0.0, + "loss_total": 0.7328471541404724, + "lr": 0.0005654817275747508, + "step": 13580, + "tokens_trained": 44.487462504 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 3.8544074888305793, + "grad_norm": 0.26154854893684387, + "loss": 0.9431, + "loss_ce": 0.8241196274757385, + "loss_region": 0.0, + "loss_total": 0.8241196274757385, + "lr": 0.0005651495016611296, + "step": 13590, + "tokens_trained": 44.520227944 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 3.857244167080349, + "grad_norm": 0.1439606100320816, + "loss": 0.9532, + "loss_ce": 1.0603502988815308, + "loss_region": 0.0, + "loss_total": 1.0603502988815308, + "lr": 0.0005648172757475083, + "step": 13600, + "tokens_trained": 44.552993384 + }, + { + "epoch": 3.857244167080349, + "eval_ppl": 2.600134999107408, + "eval_runtime": 1.4372, + "step": 13600, + "tokens_trained": 44.552993384 + }, + { + "epoch": 3.857244167080349, + "eval_ppl_cds": 3.0462448839126552, + "eval_ppl_dig": 1.0957597678734592, + "eval_ppl_exon": 3.07119398046086, + "eval_ppl_intron": 2.777814918167956, + "eval_ppl_nig": 2.510970078786593, + "eval_ppl_promoter": 2.520990938070408, + "eval_ppl_utr": 2.843975712643797, + "step": 13600, + "tokens_trained": 44.552993384 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 3.8600808453301183, + "grad_norm": 0.12047246098518372, + "loss": 0.9401, + "loss_ce": 0.9858757853507996, + "loss_region": 0.0, + "loss_total": 0.9858757853507996, + "lr": 0.000564485049833887, + "step": 13610, + "tokens_trained": 44.585756064 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 3.862917523579888, + "grad_norm": 0.23453688621520996, + "loss": 0.9486, + "loss_ce": 0.9512929320335388, + "loss_region": 0.0, + "loss_total": 0.9512929320335388, + "lr": 0.0005641528239202658, + "step": 13620, + "tokens_trained": 44.618518448 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 3.8657542018296573, + "grad_norm": 0.3163595497608185, + "loss": 0.9441, + "loss_ce": 0.9028114676475525, + "loss_region": 0.0, + "loss_total": 0.9028114676475525, + "lr": 0.0005638205980066446, + "step": 13630, + "tokens_trained": 44.651283888 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 3.868590880079427, + "grad_norm": 0.20938563346862793, + "loss": 0.9436, + "loss_ce": 1.0595521926879883, + "loss_region": 0.0, + "loss_total": 1.0595521926879883, + "lr": 0.0005634883720930233, + "step": 13640, + "tokens_trained": 44.684049328 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 3.8714275583291964, + "grad_norm": 0.16795805096626282, + "loss": 0.9385, + "loss_ce": 1.062254786491394, + "loss_region": 0.0, + "loss_total": 1.062254786491394, + "lr": 0.000563156146179402, + "step": 13650, + "tokens_trained": 44.716814768 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 3.874264236578966, + "grad_norm": 0.3279435336589813, + "loss": 0.9365, + "loss_ce": 0.9496954679489136, + "loss_region": 0.0, + "loss_total": 0.9496954679489136, + "lr": 0.0005628239202657807, + "step": 13660, + "tokens_trained": 44.749580208 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 3.8771009148287354, + "grad_norm": 0.23187603056430817, + "loss": 0.9321, + "loss_ce": 0.9736423492431641, + "loss_region": 0.0, + "loss_total": 0.9736423492431641, + "lr": 0.0005624916943521596, + "step": 13670, + "tokens_trained": 44.782344848 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 3.879937593078505, + "grad_norm": 0.19851872324943542, + "loss": 0.9405, + "loss_ce": 0.8860262632369995, + "loss_region": 0.0, + "loss_total": 0.8860262632369995, + "lr": 0.0005621594684385383, + "step": 13680, + "tokens_trained": 44.815110288 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 3.8827742713282745, + "grad_norm": 0.13353969156742096, + "loss": 0.9397, + "loss_ce": 0.9871578216552734, + "loss_region": 0.0, + "loss_total": 0.9871578216552734, + "lr": 0.0005618272425249169, + "step": 13690, + "tokens_trained": 44.847875728 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 3.885610949578044, + "grad_norm": 0.20620061457157135, + "loss": 0.9378, + "loss_ce": 0.9713705778121948, + "loss_region": 0.0, + "loss_total": 0.9713705778121948, + "lr": 0.0005614950166112956, + "step": 13700, + "tokens_trained": 44.880640368 + }, + { + "epoch": 3.885610949578044, + "eval_ppl": 2.5953898134599833, + "eval_runtime": 1.4175, + "step": 13700, + "tokens_trained": 44.880640368 + }, + { + "epoch": 3.885610949578044, + "eval_ppl_cds": 3.0228053232499166, + "eval_ppl_dig": 1.0984327413076589, + "eval_ppl_exon": 3.089195869641836, + "eval_ppl_intron": 2.7793805667411555, + "eval_ppl_nig": 2.5125802508636483, + "eval_ppl_promoter": 2.5189983287465436, + "eval_ppl_utr": 2.813546807115016, + "step": 13700, + "tokens_trained": 44.880640368 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 3.8884476278278135, + "grad_norm": 0.13635562360286713, + "loss": 0.9445, + "loss_ce": 0.9753177762031555, + "loss_region": 0.0, + "loss_total": 0.9753177762031555, + "lr": 0.0005611627906976744, + "step": 13710, + "tokens_trained": 44.913405808 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 3.891284306077583, + "grad_norm": 0.47737714648246765, + "loss": 0.9366, + "loss_ce": 0.9709941744804382, + "loss_region": 0.0, + "loss_total": 0.9709941744804382, + "lr": 0.0005608305647840532, + "step": 13720, + "tokens_trained": 44.94616792 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 3.8941209843273525, + "grad_norm": 0.20591862499713898, + "loss": 0.941, + "loss_ce": 0.9817864298820496, + "loss_region": 0.0, + "loss_total": 0.9817864298820496, + "lr": 0.0005604983388704319, + "step": 13730, + "tokens_trained": 44.97893336 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 3.896957662577122, + "grad_norm": 0.23804260790348053, + "loss": 0.9497, + "loss_ce": 0.9329025149345398, + "loss_region": 0.0, + "loss_total": 0.9329025149345398, + "lr": 0.0005601661129568106, + "step": 13740, + "tokens_trained": 45.0116988 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 3.8997943408268916, + "grad_norm": 0.12682615220546722, + "loss": 0.9436, + "loss_ce": 0.9289522767066956, + "loss_region": 0.0, + "loss_total": 0.9289522767066956, + "lr": 0.0005598338870431894, + "step": 13750, + "tokens_trained": 45.04446424 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 3.902631019076661, + "grad_norm": 0.14412567019462585, + "loss": 0.9516, + "loss_ce": 0.9099929332733154, + "loss_region": 0.0, + "loss_total": 0.9099929332733154, + "lr": 0.0005595016611295681, + "step": 13760, + "tokens_trained": 45.07722968 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 3.9054676973264306, + "grad_norm": 0.24995571374893188, + "loss": 0.9383, + "loss_ce": 1.014732003211975, + "loss_region": 0.0, + "loss_total": 1.014732003211975, + "lr": 0.0005591694352159469, + "step": 13770, + "tokens_trained": 45.10999512 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 3.9083043755762, + "grad_norm": 0.21862512826919556, + "loss": 0.9392, + "loss_ce": 0.9254577159881592, + "loss_region": 0.0, + "loss_total": 0.9254577159881592, + "lr": 0.0005588372093023257, + "step": 13780, + "tokens_trained": 45.14276056 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 3.9111410538259697, + "grad_norm": 0.1596546471118927, + "loss": 0.9479, + "loss_ce": 1.0092334747314453, + "loss_region": 0.0, + "loss_total": 1.0092334747314453, + "lr": 0.0005585049833887044, + "step": 13790, + "tokens_trained": 45.175526 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 3.913977732075739, + "grad_norm": 0.08991825580596924, + "loss": 0.946, + "loss_ce": 0.9505770206451416, + "loss_region": 0.0, + "loss_total": 0.9505770206451416, + "lr": 0.0005581727574750831, + "step": 13800, + "tokens_trained": 45.20829144 + }, + { + "epoch": 3.913977732075739, + "eval_ppl": 2.5942156173431092, + "eval_runtime": 1.4164, + "step": 13800, + "tokens_trained": 45.20829144 + }, + { + "epoch": 3.913977732075739, + "eval_ppl_cds": 3.0182581461111226, + "eval_ppl_dig": 1.091091566643531, + "eval_ppl_exon": 3.1066995203543626, + "eval_ppl_intron": 2.7684667304200206, + "eval_ppl_nig": 2.5022969634986443, + "eval_ppl_promoter": 2.497705235448826, + "eval_ppl_utr": 2.8125641355743287, + "step": 13800, + "tokens_trained": 45.20829144 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 3.9168144103255087, + "grad_norm": 0.1786630004644394, + "loss": 0.9436, + "loss_ce": 1.049873948097229, + "loss_region": 0.0, + "loss_total": 1.049873948097229, + "lr": 0.0005578405315614617, + "step": 13810, + "tokens_trained": 45.24105688 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 3.919651088575278, + "grad_norm": 0.15669773519039154, + "loss": 0.947, + "loss_ce": 0.7171390056610107, + "loss_region": 0.0, + "loss_total": 0.7171390056610107, + "lr": 0.0005575083056478405, + "step": 13820, + "tokens_trained": 45.27382152 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 3.9224877668250477, + "grad_norm": 0.12358835339546204, + "loss": 0.939, + "loss_ce": 1.031872272491455, + "loss_region": 0.0, + "loss_total": 1.031872272491455, + "lr": 0.0005571760797342193, + "step": 13830, + "tokens_trained": 45.30658696 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 3.9253244450748173, + "grad_norm": 0.1952536553144455, + "loss": 0.9467, + "loss_ce": 0.7888644933700562, + "loss_region": 0.0, + "loss_total": 0.7888644933700562, + "lr": 0.000556843853820598, + "step": 13840, + "tokens_trained": 45.3393524 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 3.928161123324587, + "grad_norm": 0.23049363493919373, + "loss": 0.9468, + "loss_ce": 0.8587316274642944, + "loss_region": 0.0, + "loss_total": 0.8587316274642944, + "lr": 0.0005565116279069767, + "step": 13850, + "tokens_trained": 45.37211624 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 3.9309978015743563, + "grad_norm": 0.1925053894519806, + "loss": 0.9372, + "loss_ce": 0.991087019443512, + "loss_region": 0.0, + "loss_total": 0.991087019443512, + "lr": 0.0005561794019933554, + "step": 13860, + "tokens_trained": 45.40488168 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 3.933834479824126, + "grad_norm": 0.255811482667923, + "loss": 0.952, + "loss_ce": 0.9305105805397034, + "loss_region": 0.0, + "loss_total": 0.9305105805397034, + "lr": 0.0005558471760797343, + "step": 13870, + "tokens_trained": 45.43764712 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 3.9366711580738953, + "grad_norm": 0.08352957665920258, + "loss": 0.9424, + "loss_ce": 0.9220941066741943, + "loss_region": 0.0, + "loss_total": 0.9220941066741943, + "lr": 0.000555514950166113, + "step": 13880, + "tokens_trained": 45.470411832 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 3.939507836323665, + "grad_norm": 0.21067480742931366, + "loss": 0.9402, + "loss_ce": 0.9089932441711426, + "loss_region": 0.0, + "loss_total": 0.9089932441711426, + "lr": 0.0005551827242524917, + "step": 13890, + "tokens_trained": 45.503177272 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 3.9423445145734344, + "grad_norm": 0.1847335696220398, + "loss": 0.9335, + "loss_ce": 1.035788893699646, + "loss_region": 0.0, + "loss_total": 1.035788893699646, + "lr": 0.0005548504983388705, + "step": 13900, + "tokens_trained": 45.535942712 + }, + { + "epoch": 3.9423445145734344, + "eval_ppl": 2.594609019021004, + "eval_runtime": 1.4303, + "step": 13900, + "tokens_trained": 45.535942712 + }, + { + "epoch": 3.9423445145734344, + "eval_ppl_cds": 3.0270684446437564, + "eval_ppl_dig": 1.0955217037048515, + "eval_ppl_exon": 3.0316235343057176, + "eval_ppl_intron": 2.766654238095024, + "eval_ppl_nig": 2.5089935816447126, + "eval_ppl_promoter": 2.4905795308803294, + "eval_ppl_utr": 2.8105278048842277, + "step": 13900, + "tokens_trained": 45.535942712 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 3.945181192823204, + "grad_norm": 0.1813579946756363, + "loss": 0.9395, + "loss_ce": 0.9378644227981567, + "loss_region": 0.0, + "loss_total": 0.9378644227981567, + "lr": 0.0005545182724252492, + "step": 13910, + "tokens_trained": 45.568708152 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 3.9480178710729734, + "grad_norm": 0.14234627783298492, + "loss": 0.9291, + "loss_ce": 0.9929129481315613, + "loss_region": 0.0, + "loss_total": 0.9929129481315613, + "lr": 0.000554186046511628, + "step": 13920, + "tokens_trained": 45.601472792 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 3.950854549322743, + "grad_norm": 0.22427336871623993, + "loss": 0.9432, + "loss_ce": 0.95787113904953, + "loss_region": 0.0, + "loss_total": 0.95787113904953, + "lr": 0.0005538538205980067, + "step": 13930, + "tokens_trained": 45.634238176 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 3.9536912275725125, + "grad_norm": 0.10548944771289825, + "loss": 0.9453, + "loss_ce": 0.9141303300857544, + "loss_region": 0.0, + "loss_total": 0.9141303300857544, + "lr": 0.0005535215946843854, + "step": 13940, + "tokens_trained": 45.667002816 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 3.956527905822282, + "grad_norm": 0.32432252168655396, + "loss": 0.9348, + "loss_ce": 1.0894911289215088, + "loss_region": 0.0, + "loss_total": 1.0894911289215088, + "lr": 0.0005531893687707641, + "step": 13950, + "tokens_trained": 45.699768256 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 3.9593645840720515, + "grad_norm": 0.18692389130592346, + "loss": 0.9441, + "loss_ce": 0.9264270663261414, + "loss_region": 0.0, + "loss_total": 0.9264270663261414, + "lr": 0.0005528571428571428, + "step": 13960, + "tokens_trained": 45.732533696 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 3.962201262321821, + "grad_norm": 0.17212215065956116, + "loss": 0.9363, + "loss_ce": 1.0768640041351318, + "loss_region": 0.0, + "loss_total": 1.0768640041351318, + "lr": 0.0005525249169435215, + "step": 13970, + "tokens_trained": 45.765299136 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 3.9650379405715905, + "grad_norm": 0.16986428201198578, + "loss": 0.9332, + "loss_ce": 1.005123496055603, + "loss_region": 0.0, + "loss_total": 1.005123496055603, + "lr": 0.0005521926910299004, + "step": 13980, + "tokens_trained": 45.798064576 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 3.96787461882136, + "grad_norm": 0.18781083822250366, + "loss": 0.9407, + "loss_ce": 0.8882608413696289, + "loss_region": 0.0, + "loss_total": 0.8882608413696289, + "lr": 0.0005518604651162791, + "step": 13990, + "tokens_trained": 45.830830016 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 3.9707112970711296, + "grad_norm": 0.21958835422992706, + "loss": 0.9426, + "loss_ce": 0.9133145213127136, + "loss_region": 0.0, + "loss_total": 0.9133145213127136, + "lr": 0.0005515282392026578, + "step": 14000, + "tokens_trained": 45.863592288 + }, + { + "epoch": 3.9707112970711296, + "eval_ppl": 2.5944030324005523, + "eval_runtime": 1.421, + "step": 14000, + "tokens_trained": 45.863592288 + }, + { + "epoch": 3.9707112970711296, + "eval_ppl_cds": 3.0306027628367462, + "eval_ppl_dig": 1.0941921109043133, + "eval_ppl_exon": 3.0064781426872624, + "eval_ppl_intron": 2.765199351144961, + "eval_ppl_nig": 2.503585674899626, + "eval_ppl_promoter": 2.489192188373034, + "eval_ppl_utr": 2.7677694773317185, + "step": 14000, + "tokens_trained": 45.863592288 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 3.973547975320899, + "grad_norm": 0.1750277578830719, + "loss": 0.9435, + "loss_ce": 0.8454325795173645, + "loss_region": 0.0, + "loss_total": 0.8454325795173645, + "lr": 0.0005511960132890365, + "step": 14010, + "tokens_trained": 45.896357728 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 3.9763846535706686, + "grad_norm": 0.24825716018676758, + "loss": 0.9363, + "loss_ce": 1.0020930767059326, + "loss_region": 0.0, + "loss_total": 1.0020930767059326, + "lr": 0.0005508637873754153, + "step": 14020, + "tokens_trained": 45.929123168 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 3.979221331820438, + "grad_norm": 0.10889997333288193, + "loss": 0.9428, + "loss_ce": 0.82785564661026, + "loss_region": 0.0, + "loss_total": 0.82785564661026, + "lr": 0.0005505315614617941, + "step": 14030, + "tokens_trained": 45.96188528 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 3.9820580100702077, + "grad_norm": 0.2572927474975586, + "loss": 0.9435, + "loss_ce": 0.9821985363960266, + "loss_region": 0.0, + "loss_total": 0.9821985363960266, + "lr": 0.0005501993355481728, + "step": 14040, + "tokens_trained": 45.99465072 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 3.984894688319977, + "grad_norm": 0.23487243056297302, + "loss": 0.9436, + "loss_ce": 0.883675754070282, + "loss_region": 0.0, + "loss_total": 0.883675754070282, + "lr": 0.0005498671096345515, + "step": 14050, + "tokens_trained": 46.02741616 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 3.9877313665697467, + "grad_norm": 0.190826416015625, + "loss": 0.9413, + "loss_ce": 0.9926276803016663, + "loss_region": 0.0, + "loss_total": 0.9926276803016663, + "lr": 0.0005495348837209302, + "step": 14060, + "tokens_trained": 46.060177768 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 3.990568044819516, + "grad_norm": 0.23745839297771454, + "loss": 0.9415, + "loss_ce": 0.9895360469818115, + "loss_region": 0.0, + "loss_total": 0.9895360469818115, + "lr": 0.0005492026578073089, + "step": 14070, + "tokens_trained": 46.092942408 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 3.9934047230692857, + "grad_norm": 0.1988748461008072, + "loss": 0.9415, + "loss_ce": 1.1041780710220337, + "loss_region": 0.0, + "loss_total": 1.1041780710220337, + "lr": 0.0005488704318936877, + "step": 14080, + "tokens_trained": 46.125707848 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 3.9962414013190553, + "grad_norm": 0.15050694346427917, + "loss": 0.9287, + "loss_ce": 1.1100118160247803, + "loss_region": 0.0, + "loss_total": 1.1100118160247803, + "lr": 0.0005485382059800664, + "step": 14090, + "tokens_trained": 46.1584696 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 3.9990780795688248, + "grad_norm": 0.1549491435289383, + "loss": 0.9429, + "loss_ce": 0.6208906173706055, + "loss_region": 0.0, + "loss_total": 0.6208906173706055, + "lr": 0.0005482059800664452, + "step": 14100, + "tokens_trained": 46.191235016 + }, + { + "epoch": 3.9990780795688248, + "eval_ppl": 2.5948193525644685, + "eval_runtime": 1.4339, + "step": 14100, + "tokens_trained": 46.191235016 + }, + { + "epoch": 3.9990780795688248, + "eval_ppl_cds": 3.049817568492804, + "eval_ppl_dig": 1.0960263478459746, + "eval_ppl_exon": 3.0416204713171804, + "eval_ppl_intron": 2.7718118291068756, + "eval_ppl_nig": 2.5085394516372683, + "eval_ppl_promoter": 2.5071368880712326, + "eval_ppl_utr": 2.804425911518042, + "step": 14100, + "tokens_trained": 46.191235016 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 4.001702006949862, + "grad_norm": 0.2325429618358612, + "loss": 0.9418, + "loss_ce": 0.9895012974739075, + "loss_region": 0.0, + "loss_total": 0.9895012974739075, + "lr": 0.0005478737541528239, + "step": 14110, + "tokens_trained": 46.221543048 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 4.004538685199631, + "grad_norm": 0.2070443332195282, + "loss": 0.9313, + "loss_ce": 0.891310453414917, + "loss_region": 0.0, + "loss_total": 0.891310453414917, + "lr": 0.0005475415282392026, + "step": 14120, + "tokens_trained": 46.254308488 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 4.007375363449401, + "grad_norm": 0.20442824065685272, + "loss": 0.9306, + "loss_ce": 0.8435589075088501, + "loss_region": 0.0, + "loss_total": 0.8435589075088501, + "lr": 0.0005472093023255815, + "step": 14130, + "tokens_trained": 46.287073928 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 4.0102120416991704, + "grad_norm": 0.18940719962120056, + "loss": 0.9288, + "loss_ce": 0.9312524199485779, + "loss_region": 0.0, + "loss_total": 0.9312524199485779, + "lr": 0.0005468770764119602, + "step": 14140, + "tokens_trained": 46.319839368 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 4.01304871994894, + "grad_norm": 0.27701181173324585, + "loss": 0.9368, + "loss_ce": 0.7133271098136902, + "loss_region": 0.0, + "loss_total": 0.7133271098136902, + "lr": 0.0005465448504983389, + "step": 14150, + "tokens_trained": 46.352604808 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 4.0158853981987095, + "grad_norm": 0.1410902887582779, + "loss": 0.9417, + "loss_ce": 1.0395240783691406, + "loss_region": 0.0, + "loss_total": 1.0395240783691406, + "lr": 0.0005462126245847176, + "step": 14160, + "tokens_trained": 46.385370248 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 4.018722076448479, + "grad_norm": 0.15153512358665466, + "loss": 0.9489, + "loss_ce": 1.0305821895599365, + "loss_region": 0.0, + "loss_total": 1.0305821895599365, + "lr": 0.0005458803986710964, + "step": 14170, + "tokens_trained": 46.418135688 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 4.0215587546982485, + "grad_norm": 0.2822026014328003, + "loss": 0.9392, + "loss_ce": 0.988848090171814, + "loss_region": 0.0, + "loss_total": 0.988848090171814, + "lr": 0.0005455481727574752, + "step": 14180, + "tokens_trained": 46.450900928 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 4.024395432948018, + "grad_norm": 0.21233919262886047, + "loss": 0.9352, + "loss_ce": 0.8462297320365906, + "loss_region": 0.0, + "loss_total": 0.8462297320365906, + "lr": 0.0005452159468438538, + "step": 14190, + "tokens_trained": 46.483664336 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 4.027232111197788, + "grad_norm": 0.17563201487064362, + "loss": 0.9392, + "loss_ce": 0.9329037070274353, + "loss_region": 0.0, + "loss_total": 0.9329037070274353, + "lr": 0.0005448837209302325, + "step": 14200, + "tokens_trained": 46.516429776 + }, + { + "epoch": 4.027232111197788, + "eval_ppl": 2.5924621181335383, + "eval_runtime": 1.4237, + "step": 14200, + "tokens_trained": 46.516429776 + }, + { + "epoch": 4.027232111197788, + "eval_ppl_cds": 3.0403016497596176, + "eval_ppl_dig": 1.0963033231325885, + "eval_ppl_exon": 3.0114048050438083, + "eval_ppl_intron": 2.770337564728709, + "eval_ppl_nig": 2.502542066684816, + "eval_ppl_promoter": 2.508019804392938, + "eval_ppl_utr": 2.782150209251606, + "step": 14200, + "tokens_trained": 46.516429776 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 4.030068789447557, + "grad_norm": 0.13513405621051788, + "loss": 0.9431, + "loss_ce": 0.9928596019744873, + "loss_region": 0.0, + "loss_total": 0.9928596019744873, + "lr": 0.0005445514950166112, + "step": 14210, + "tokens_trained": 46.549195216 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 4.032905467697327, + "grad_norm": 0.10843551903963089, + "loss": 0.9336, + "loss_ce": 0.7746082544326782, + "loss_region": 0.0, + "loss_total": 0.7746082544326782, + "lr": 0.00054421926910299, + "step": 14220, + "tokens_trained": 46.581960656 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 4.035742145947096, + "grad_norm": 0.18599511682987213, + "loss": 0.9349, + "loss_ce": 0.941452145576477, + "loss_region": 0.0, + "loss_total": 0.941452145576477, + "lr": 0.0005438870431893688, + "step": 14230, + "tokens_trained": 46.614726072 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 4.038578824196866, + "grad_norm": 0.19950851798057556, + "loss": 0.9299, + "loss_ce": 1.0581520795822144, + "loss_region": 0.0, + "loss_total": 1.0581520795822144, + "lr": 0.0005435548172757475, + "step": 14240, + "tokens_trained": 46.647491512 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 4.041415502446635, + "grad_norm": 0.16132177412509918, + "loss": 0.938, + "loss_ce": 0.9467852115631104, + "loss_region": 0.0, + "loss_total": 0.9467852115631104, + "lr": 0.0005432225913621263, + "step": 14250, + "tokens_trained": 46.680256952 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 4.044252180696405, + "grad_norm": 0.2297714501619339, + "loss": 0.934, + "loss_ce": 0.8540815114974976, + "loss_region": 0.0, + "loss_total": 0.8540815114974976, + "lr": 0.000542890365448505, + "step": 14260, + "tokens_trained": 46.713021592 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 4.047088858946174, + "grad_norm": 0.200074702501297, + "loss": 0.9341, + "loss_ce": 1.0087862014770508, + "loss_region": 0.0, + "loss_total": 1.0087862014770508, + "lr": 0.0005425581395348837, + "step": 14270, + "tokens_trained": 46.745787032 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 4.049925537195944, + "grad_norm": 0.2723829746246338, + "loss": 0.9422, + "loss_ce": 1.046331763267517, + "loss_region": 0.0, + "loss_total": 1.046331763267517, + "lr": 0.0005422259136212625, + "step": 14280, + "tokens_trained": 46.778552472 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 4.052762215445713, + "grad_norm": 0.1683189868927002, + "loss": 0.9354, + "loss_ce": 1.0137995481491089, + "loss_region": 0.0, + "loss_total": 1.0137995481491089, + "lr": 0.0005418936877076413, + "step": 14290, + "tokens_trained": 46.811317912 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 4.055598893695483, + "grad_norm": 0.1514330506324768, + "loss": 0.9359, + "loss_ce": 1.1184216737747192, + "loss_region": 0.0, + "loss_total": 1.1184216737747192, + "lr": 0.00054156146179402, + "step": 14300, + "tokens_trained": 46.844083352 + }, + { + "epoch": 4.055598893695483, + "eval_ppl": 2.589649939128754, + "eval_runtime": 1.4289, + "step": 14300, + "tokens_trained": 46.844083352 + }, + { + "epoch": 4.055598893695483, + "eval_ppl_cds": 3.0152947874732874, + "eval_ppl_dig": 1.094072700052671, + "eval_ppl_exon": 2.975900110376241, + "eval_ppl_intron": 2.772770152060655, + "eval_ppl_nig": 2.5051838330709746, + "eval_ppl_promoter": 2.4973853405975324, + "eval_ppl_utr": 2.7503691177340945, + "step": 14300, + "tokens_trained": 46.844083352 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 4.058435571945252, + "grad_norm": 0.18604254722595215, + "loss": 0.9252, + "loss_ce": 0.9513590335845947, + "loss_region": 0.0, + "loss_total": 0.9513590335845947, + "lr": 0.0005412292358803986, + "step": 14310, + "tokens_trained": 46.876848792 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 4.061272250195022, + "grad_norm": 0.2508116662502289, + "loss": 0.9483, + "loss_ce": 0.9133107662200928, + "loss_region": 0.0, + "loss_total": 0.9133107662200928, + "lr": 0.0005408970099667773, + "step": 14320, + "tokens_trained": 46.909614232 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 4.064108928444791, + "grad_norm": 0.13024458289146423, + "loss": 0.9392, + "loss_ce": 0.9563823342323303, + "loss_region": 0.0, + "loss_total": 0.9563823342323303, + "lr": 0.0005405647840531562, + "step": 14330, + "tokens_trained": 46.942379672 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 4.066945606694561, + "grad_norm": 0.10867466777563095, + "loss": 0.9398, + "loss_ce": 0.958825945854187, + "loss_region": 0.0, + "loss_total": 0.958825945854187, + "lr": 0.0005402325581395349, + "step": 14340, + "tokens_trained": 46.975145112 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 4.06978228494433, + "grad_norm": 0.19304095208644867, + "loss": 0.9276, + "loss_ce": 0.9885596036911011, + "loss_region": 0.0, + "loss_total": 0.9885596036911011, + "lr": 0.0005399003322259136, + "step": 14350, + "tokens_trained": 47.007910552 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 4.0726189631941, + "grad_norm": 0.1235615462064743, + "loss": 0.9399, + "loss_ce": 0.5473272800445557, + "loss_region": 0.0, + "loss_total": 0.5473272800445557, + "lr": 0.0005395681063122923, + "step": 14360, + "tokens_trained": 47.040675992 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 4.075455641443869, + "grad_norm": 0.2799862027168274, + "loss": 0.9378, + "loss_ce": 0.8747535347938538, + "loss_region": 0.0, + "loss_total": 0.8747535347938538, + "lr": 0.0005392358803986711, + "step": 14370, + "tokens_trained": 47.073441432 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 4.078292319693639, + "grad_norm": 0.10949348658323288, + "loss": 0.939, + "loss_ce": 1.0397464036941528, + "loss_region": 0.0, + "loss_total": 1.0397464036941528, + "lr": 0.0005389036544850499, + "step": 14380, + "tokens_trained": 47.106203704 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 4.081128997943408, + "grad_norm": 0.36191606521606445, + "loss": 0.943, + "loss_ce": 1.0997977256774902, + "loss_region": 0.0, + "loss_total": 1.0997977256774902, + "lr": 0.0005385714285714286, + "step": 14390, + "tokens_trained": 47.138968344 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 4.083965676193178, + "grad_norm": 0.17926238477230072, + "loss": 0.9265, + "loss_ce": 0.8269014358520508, + "loss_region": 0.0, + "loss_total": 0.8269014358520508, + "lr": 0.0005382392026578073, + "step": 14400, + "tokens_trained": 47.171733784 + }, + { + "epoch": 4.083965676193178, + "eval_ppl": 2.589119783412371, + "eval_runtime": 1.4185, + "step": 14400, + "tokens_trained": 47.171733784 + }, + { + "epoch": 4.083965676193178, + "eval_ppl_cds": 3.014421577434182, + "eval_ppl_dig": 1.0964278295571124, + "eval_ppl_exon": 2.9856616005631094, + "eval_ppl_intron": 2.7718336914018398, + "eval_ppl_nig": 2.4981672035132787, + "eval_ppl_promoter": 2.4925113844584397, + "eval_ppl_utr": 2.7157205152543624, + "step": 14400, + "tokens_trained": 47.171733784 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 4.0868023544429475, + "grad_norm": 0.1681579351425171, + "loss": 0.9309, + "loss_ce": 0.8968380689620972, + "loss_region": 0.0, + "loss_total": 0.8968380689620972, + "lr": 0.0005379069767441861, + "step": 14410, + "tokens_trained": 47.204499224 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 4.089639032692717, + "grad_norm": 0.24388742446899414, + "loss": 0.9375, + "loss_ce": 1.0679600238800049, + "loss_region": 0.0, + "loss_total": 1.0679600238800049, + "lr": 0.0005375747508305648, + "step": 14420, + "tokens_trained": 47.237264664 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 4.0924757109424865, + "grad_norm": 0.2286989837884903, + "loss": 0.9421, + "loss_ce": 0.8518466353416443, + "loss_region": 0.0, + "loss_total": 0.8518466353416443, + "lr": 0.0005372425249169436, + "step": 14430, + "tokens_trained": 47.270030104 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 4.095312389192256, + "grad_norm": 0.21489913761615753, + "loss": 0.9369, + "loss_ce": 0.9005475044250488, + "loss_region": 0.0, + "loss_total": 0.9005475044250488, + "lr": 0.0005369102990033222, + "step": 14440, + "tokens_trained": 47.302795544 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 4.098149067442026, + "grad_norm": 0.1362367868423462, + "loss": 0.9407, + "loss_ce": 0.9853972792625427, + "loss_region": 0.0, + "loss_total": 0.9853972792625427, + "lr": 0.000536578073089701, + "step": 14450, + "tokens_trained": 47.335558272 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 4.100985745691795, + "grad_norm": 0.1515919417142868, + "loss": 0.9449, + "loss_ce": 0.8820052146911621, + "loss_region": 0.0, + "loss_total": 0.8820052146911621, + "lr": 0.0005362458471760797, + "step": 14460, + "tokens_trained": 47.368323712 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 4.103822423941565, + "grad_norm": 0.1323785036802292, + "loss": 0.94, + "loss_ce": 0.7193222641944885, + "loss_region": 0.0, + "loss_total": 0.7193222641944885, + "lr": 0.0005359136212624584, + "step": 14470, + "tokens_trained": 47.401089152 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 4.106659102191334, + "grad_norm": 0.26915857195854187, + "loss": 0.9415, + "loss_ce": 0.8298725485801697, + "loss_region": 0.0, + "loss_total": 0.8298725485801697, + "lr": 0.0005355813953488373, + "step": 14480, + "tokens_trained": 47.433851264 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 4.109495780441104, + "grad_norm": 0.1973770558834076, + "loss": 0.9275, + "loss_ce": 0.8649148344993591, + "loss_region": 0.0, + "loss_total": 0.8649148344993591, + "lr": 0.000535249169435216, + "step": 14490, + "tokens_trained": 47.466616704 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 4.112332458690873, + "grad_norm": 0.14872097969055176, + "loss": 0.932, + "loss_ce": 1.0018435716629028, + "loss_region": 0.0, + "loss_total": 1.0018435716629028, + "lr": 0.0005349169435215947, + "step": 14500, + "tokens_trained": 47.499382144 + }, + { + "epoch": 4.112332458690873, + "eval_ppl": 2.5871035598722076, + "eval_runtime": 1.4164, + "step": 14500, + "tokens_trained": 47.499382144 + }, + { + "epoch": 4.112332458690873, + "eval_ppl_cds": 3.0089095565687787, + "eval_ppl_dig": 1.095604145672208, + "eval_ppl_exon": 3.0274631591135575, + "eval_ppl_intron": 2.770320808199072, + "eval_ppl_nig": 2.498604814542771, + "eval_ppl_promoter": 2.4816142204225624, + "eval_ppl_utr": 2.7269036230117356, + "step": 14500, + "tokens_trained": 47.499382144 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 4.115169136940643, + "grad_norm": 0.15225590765476227, + "loss": 0.9369, + "loss_ce": 0.9728121161460876, + "loss_region": 0.0, + "loss_total": 0.9728121161460876, + "lr": 0.0005345847176079734, + "step": 14510, + "tokens_trained": 47.532147576 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 4.118005815190412, + "grad_norm": 0.06935209780931473, + "loss": 0.9419, + "loss_ce": 1.1027604341506958, + "loss_region": 0.0, + "loss_total": 1.1027604341506958, + "lr": 0.0005342524916943522, + "step": 14520, + "tokens_trained": 47.564912216 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 4.120842493440182, + "grad_norm": 0.1499994993209839, + "loss": 0.9337, + "loss_ce": 0.8296639323234558, + "loss_region": 0.0, + "loss_total": 0.8296639323234558, + "lr": 0.000533920265780731, + "step": 14530, + "tokens_trained": 47.597677656 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 4.123679171689951, + "grad_norm": 0.2594814896583557, + "loss": 0.9309, + "loss_ce": 1.0892819166183472, + "loss_region": 0.0, + "loss_total": 1.0892819166183472, + "lr": 0.0005335880398671097, + "step": 14540, + "tokens_trained": 47.630443096 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 4.126515849939721, + "grad_norm": 0.22025279700756073, + "loss": 0.9313, + "loss_ce": 0.8121228218078613, + "loss_region": 0.0, + "loss_total": 0.8121228218078613, + "lr": 0.0005332558139534884, + "step": 14550, + "tokens_trained": 47.663208536 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 4.12935252818949, + "grad_norm": 0.18731170892715454, + "loss": 0.9305, + "loss_ce": 0.7753037810325623, + "loss_region": 0.0, + "loss_total": 0.7753037810325623, + "lr": 0.000532923588039867, + "step": 14560, + "tokens_trained": 47.695973976 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 4.13218920643926, + "grad_norm": 0.17981189489364624, + "loss": 0.9517, + "loss_ce": 0.962387204170227, + "loss_region": 0.0, + "loss_total": 0.962387204170227, + "lr": 0.0005325913621262458, + "step": 14570, + "tokens_trained": 47.728739416 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 4.135025884689029, + "grad_norm": 0.07995792478322983, + "loss": 0.937, + "loss_ce": 0.7757516503334045, + "loss_region": 0.0, + "loss_total": 0.7757516503334045, + "lr": 0.0005322591362126246, + "step": 14580, + "tokens_trained": 47.761504856 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 4.137862562938799, + "grad_norm": 0.16591013967990875, + "loss": 0.9328, + "loss_ce": 1.003652811050415, + "loss_region": 0.0, + "loss_total": 1.003652811050415, + "lr": 0.0005319269102990033, + "step": 14590, + "tokens_trained": 47.794270296 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 4.140699241188568, + "grad_norm": 0.1347040832042694, + "loss": 0.933, + "loss_ce": 0.9665950536727905, + "loss_region": 0.0, + "loss_total": 0.9665950536727905, + "lr": 0.0005315946843853821, + "step": 14600, + "tokens_trained": 47.827035736 + }, + { + "epoch": 4.140699241188568, + "eval_ppl": 2.588139859937515, + "eval_runtime": 1.4212, + "step": 14600, + "tokens_trained": 47.827035736 + }, + { + "epoch": 4.140699241188568, + "eval_ppl_cds": 3.001972899444411, + "eval_ppl_dig": 1.0970813985597443, + "eval_ppl_exon": 2.9599491567741762, + "eval_ppl_intron": 2.7652306371326976, + "eval_ppl_nig": 2.504245684770931, + "eval_ppl_promoter": 2.4705281069449874, + "eval_ppl_utr": 2.6948219937052325, + "step": 14600, + "tokens_trained": 47.827035736 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 4.143535919438338, + "grad_norm": 0.1914171725511551, + "loss": 0.9306, + "loss_ce": 0.8242806792259216, + "loss_region": 0.0, + "loss_total": 0.8242806792259216, + "lr": 0.0005312624584717608, + "step": 14610, + "tokens_trained": 47.859801176 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 4.146372597688107, + "grad_norm": 0.16787536442279816, + "loss": 0.931, + "loss_ce": 0.8013043403625488, + "loss_region": 0.0, + "loss_total": 0.8013043403625488, + "lr": 0.0005309302325581395, + "step": 14620, + "tokens_trained": 47.892565824 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 4.149209275937877, + "grad_norm": 0.2180580049753189, + "loss": 0.9351, + "loss_ce": 0.8189687728881836, + "loss_region": 0.0, + "loss_total": 0.8189687728881836, + "lr": 0.0005305980066445183, + "step": 14630, + "tokens_trained": 47.925331264 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 4.152045954187646, + "grad_norm": 0.10411303490400314, + "loss": 0.9393, + "loss_ce": 0.8919556736946106, + "loss_region": 0.0, + "loss_total": 0.8919556736946106, + "lr": 0.0005302657807308971, + "step": 14640, + "tokens_trained": 47.958094152 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 4.154882632437416, + "grad_norm": 0.09382779151201248, + "loss": 0.9335, + "loss_ce": 0.98001629114151, + "loss_region": 0.0, + "loss_total": 0.98001629114151, + "lr": 0.0005299335548172758, + "step": 14650, + "tokens_trained": 47.990858792 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 4.1577193106871855, + "grad_norm": 0.12049087136983871, + "loss": 0.9379, + "loss_ce": 1.1302889585494995, + "loss_region": 0.0, + "loss_total": 1.1302889585494995, + "lr": 0.0005296013289036545, + "step": 14660, + "tokens_trained": 48.023624232 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 4.160555988936955, + "grad_norm": 0.102897047996521, + "loss": 0.9271, + "loss_ce": 0.9518076181411743, + "loss_region": 0.0, + "loss_total": 0.9518076181411743, + "lr": 0.0005292691029900332, + "step": 14670, + "tokens_trained": 48.056389672 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 4.1633926671867245, + "grad_norm": 0.1314830482006073, + "loss": 0.927, + "loss_ce": 0.8603493571281433, + "loss_region": 0.0, + "loss_total": 0.8603493571281433, + "lr": 0.0005289368770764121, + "step": 14680, + "tokens_trained": 48.089155112 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 4.166229345436494, + "grad_norm": 0.2593640089035034, + "loss": 0.9325, + "loss_ce": 1.0744990110397339, + "loss_region": 0.0, + "loss_total": 1.0744990110397339, + "lr": 0.0005286046511627907, + "step": 14690, + "tokens_trained": 48.121920552 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 4.1690660236862636, + "grad_norm": 0.12625226378440857, + "loss": 0.934, + "loss_ce": 0.9418997168540955, + "loss_region": 0.0, + "loss_total": 0.9418997168540955, + "lr": 0.0005282724252491694, + "step": 14700, + "tokens_trained": 48.154685192 + }, + { + "epoch": 4.1690660236862636, + "eval_ppl": 2.5855541257812495, + "eval_runtime": 1.4185, + "step": 14700, + "tokens_trained": 48.154685192 + }, + { + "epoch": 4.1690660236862636, + "eval_ppl_cds": 3.035082707035292, + "eval_ppl_dig": 1.0955182661453848, + "eval_ppl_exon": 2.9651632904933583, + "eval_ppl_intron": 2.769701576233253, + "eval_ppl_nig": 2.4997965928031243, + "eval_ppl_promoter": 2.4900980380863853, + "eval_ppl_utr": 2.746065823003088, + "step": 14700, + "tokens_trained": 48.154685192 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 4.171902701936033, + "grad_norm": 0.2351093739271164, + "loss": 0.9407, + "loss_ce": 1.038684606552124, + "loss_region": 0.0, + "loss_total": 1.038684606552124, + "lr": 0.0005279401993355481, + "step": 14710, + "tokens_trained": 48.187450632 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 4.174739380185803, + "grad_norm": 0.1942693442106247, + "loss": 0.9234, + "loss_ce": 0.9025586247444153, + "loss_region": 0.0, + "loss_total": 0.9025586247444153, + "lr": 0.0005276079734219269, + "step": 14720, + "tokens_trained": 48.220216072 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 4.177576058435572, + "grad_norm": 0.12884767353534698, + "loss": 0.9352, + "loss_ce": 0.791648268699646, + "loss_region": 0.0, + "loss_total": 0.791648268699646, + "lr": 0.0005272757475083057, + "step": 14730, + "tokens_trained": 48.252981512 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 4.180412736685342, + "grad_norm": 0.26337212324142456, + "loss": 0.9318, + "loss_ce": 1.003935694694519, + "loss_region": 0.0, + "loss_total": 1.003935694694519, + "lr": 0.0005269435215946844, + "step": 14740, + "tokens_trained": 48.285746952 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 4.183249414935111, + "grad_norm": 0.15729133784770966, + "loss": 0.9394, + "loss_ce": 0.9131348133087158, + "loss_region": 0.0, + "loss_total": 0.9131348133087158, + "lr": 0.0005266112956810631, + "step": 14750, + "tokens_trained": 48.318512392 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 4.186086093184881, + "grad_norm": 0.23509542644023895, + "loss": 0.9375, + "loss_ce": 0.7503214478492737, + "loss_region": 0.0, + "loss_total": 0.7503214478492737, + "lr": 0.0005262790697674419, + "step": 14760, + "tokens_trained": 48.351275144 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 4.18892277143465, + "grad_norm": 0.18812300264835358, + "loss": 0.9344, + "loss_ce": 0.8815061450004578, + "loss_region": 0.0, + "loss_total": 0.8815061450004578, + "lr": 0.0005259468438538206, + "step": 14770, + "tokens_trained": 48.384040584 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 4.19175944968442, + "grad_norm": 0.12919968366622925, + "loss": 0.93, + "loss_ce": 0.8708346486091614, + "loss_region": 0.0, + "loss_total": 0.8708346486091614, + "lr": 0.0005256146179401994, + "step": 14780, + "tokens_trained": 48.416806024 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 4.194596127934189, + "grad_norm": 0.17369531095027924, + "loss": 0.9143, + "loss_ce": 1.0542610883712769, + "loss_region": 0.0, + "loss_total": 1.0542610883712769, + "lr": 0.0005252823920265782, + "step": 14790, + "tokens_trained": 48.449571464 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 4.197432806183959, + "grad_norm": 0.11261310428380966, + "loss": 0.9405, + "loss_ce": 1.046539068222046, + "loss_region": 0.0, + "loss_total": 1.046539068222046, + "lr": 0.0005249501661129569, + "step": 14800, + "tokens_trained": 48.482335304 + }, + { + "epoch": 4.197432806183959, + "eval_ppl": 2.5827723779738494, + "eval_runtime": 1.4286, + "step": 14800, + "tokens_trained": 48.482335304 + }, + { + "epoch": 4.197432806183959, + "eval_ppl_cds": 3.0051464894350377, + "eval_ppl_dig": 1.097509477009342, + "eval_ppl_exon": 2.933674448983607, + "eval_ppl_intron": 2.764039313050795, + "eval_ppl_nig": 2.4990666052957433, + "eval_ppl_promoter": 2.488141834578141, + "eval_ppl_utr": 2.7534613508109533, + "step": 14800, + "tokens_trained": 48.482335304 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 4.200269484433728, + "grad_norm": 0.1631338745355606, + "loss": 0.9368, + "loss_ce": 0.9397408366203308, + "loss_region": 0.0, + "loss_total": 0.9397408366203308, + "lr": 0.0005246179401993355, + "step": 14810, + "tokens_trained": 48.515100744 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 4.203106162683498, + "grad_norm": 0.07354038208723068, + "loss": 0.9319, + "loss_ce": 0.8120086193084717, + "loss_region": 0.0, + "loss_total": 0.8120086193084717, + "lr": 0.0005242857142857142, + "step": 14820, + "tokens_trained": 48.547866176 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 4.205942840933267, + "grad_norm": 0.16728396713733673, + "loss": 0.9356, + "loss_ce": 0.8453701138496399, + "loss_region": 0.0, + "loss_total": 0.8453701138496399, + "lr": 0.000523953488372093, + "step": 14830, + "tokens_trained": 48.580631616 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 4.208779519183037, + "grad_norm": 0.2008795291185379, + "loss": 0.9377, + "loss_ce": 0.9129871726036072, + "loss_region": 0.0, + "loss_total": 0.9129871726036072, + "lr": 0.0005236212624584718, + "step": 14840, + "tokens_trained": 48.613397056 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 4.211616197432806, + "grad_norm": 0.13080498576164246, + "loss": 0.9375, + "loss_ce": 0.9644050598144531, + "loss_region": 0.0, + "loss_total": 0.9644050598144531, + "lr": 0.0005232890365448505, + "step": 14850, + "tokens_trained": 48.646162496 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 4.214452875682576, + "grad_norm": 0.11413059383630753, + "loss": 0.9212, + "loss_ce": 0.99642413854599, + "loss_region": 0.0, + "loss_total": 0.99642413854599, + "lr": 0.0005229568106312292, + "step": 14860, + "tokens_trained": 48.678927936 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 4.217289553932345, + "grad_norm": 0.2572321593761444, + "loss": 0.9282, + "loss_ce": 0.878646194934845, + "loss_region": 0.0, + "loss_total": 0.878646194934845, + "lr": 0.000522624584717608, + "step": 14870, + "tokens_trained": 48.711693376 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 4.220126232182115, + "grad_norm": 0.139862060546875, + "loss": 0.9282, + "loss_ce": 0.908257007598877, + "loss_region": 0.0, + "loss_total": 0.908257007598877, + "lr": 0.0005222923588039868, + "step": 14880, + "tokens_trained": 48.744458016 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 4.222962910431884, + "grad_norm": 0.2697189450263977, + "loss": 0.9257, + "loss_ce": 0.8285881280899048, + "loss_region": 0.0, + "loss_total": 0.8285881280899048, + "lr": 0.0005219601328903655, + "step": 14890, + "tokens_trained": 48.777223456 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 4.225799588681654, + "grad_norm": 0.13424909114837646, + "loss": 0.9418, + "loss_ce": 0.9720458388328552, + "loss_region": 0.0, + "loss_total": 0.9720458388328552, + "lr": 0.0005216279069767442, + "step": 14900, + "tokens_trained": 48.809988096 + }, + { + "epoch": 4.225799588681654, + "eval_ppl": 2.581936898425045, + "eval_runtime": 1.4164, + "step": 14900, + "tokens_trained": 48.809988096 + }, + { + "epoch": 4.225799588681654, + "eval_ppl_cds": 3.0110678151185555, + "eval_ppl_dig": 1.0944437062913703, + "eval_ppl_exon": 2.9905737049508336, + "eval_ppl_intron": 2.7547914159792266, + "eval_ppl_nig": 2.494870890180993, + "eval_ppl_promoter": 2.4573303634289645, + "eval_ppl_utr": 2.766112617710405, + "step": 14900, + "tokens_trained": 48.809988096 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 4.2286362669314235, + "grad_norm": 0.17796799540519714, + "loss": 0.9317, + "loss_ce": 0.8936126828193665, + "loss_region": 0.0, + "loss_total": 0.8936126828193665, + "lr": 0.000521295681063123, + "step": 14910, + "tokens_trained": 48.842753536 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 4.231472945181193, + "grad_norm": 0.14166666567325592, + "loss": 0.9295, + "loss_ce": 0.9019049406051636, + "loss_region": 0.0, + "loss_total": 0.9019049406051636, + "lr": 0.0005209634551495017, + "step": 14920, + "tokens_trained": 48.875518976 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 4.2343096234309625, + "grad_norm": 0.19761599600315094, + "loss": 0.9276, + "loss_ce": 0.7904037833213806, + "loss_region": 0.0, + "loss_total": 0.7904037833213806, + "lr": 0.0005206312292358803, + "step": 14930, + "tokens_trained": 48.908284416 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 4.237146301680732, + "grad_norm": 0.08856824040412903, + "loss": 0.9352, + "loss_ce": 0.9264026284217834, + "loss_region": 0.0, + "loss_total": 0.9264026284217834, + "lr": 0.0005202990033222591, + "step": 14940, + "tokens_trained": 48.941049856 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 4.2399829799305015, + "grad_norm": 0.15115244686603546, + "loss": 0.9248, + "loss_ce": 1.1383808851242065, + "loss_region": 0.0, + "loss_total": 1.1383808851242065, + "lr": 0.0005199667774086379, + "step": 14950, + "tokens_trained": 48.97381528 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 4.242819658180271, + "grad_norm": 0.18591441214084625, + "loss": 0.9301, + "loss_ce": 0.8672671318054199, + "loss_region": 0.0, + "loss_total": 0.8672671318054199, + "lr": 0.0005196345514950166, + "step": 14960, + "tokens_trained": 49.00658072 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 4.245656336430041, + "grad_norm": 0.10978300869464874, + "loss": 0.9306, + "loss_ce": 0.9068782329559326, + "loss_region": 0.0, + "loss_total": 0.9068782329559326, + "lr": 0.0005193023255813953, + "step": 14970, + "tokens_trained": 49.03934536 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 4.24849301467981, + "grad_norm": 0.1324789971113205, + "loss": 0.9381, + "loss_ce": 0.8899797797203064, + "loss_region": 0.0, + "loss_total": 0.8899797797203064, + "lr": 0.000518970099667774, + "step": 14980, + "tokens_trained": 49.0721108 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 4.25132969292958, + "grad_norm": 0.16805671155452728, + "loss": 0.9318, + "loss_ce": 1.0589925050735474, + "loss_region": 0.0, + "loss_total": 1.0589925050735474, + "lr": 0.0005186378737541529, + "step": 14990, + "tokens_trained": 49.10487624 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 4.254166371179349, + "grad_norm": 0.17664922773838043, + "loss": 0.9348, + "loss_ce": 0.9828522205352783, + "loss_region": 0.0, + "loss_total": 0.9828522205352783, + "lr": 0.0005183056478405316, + "step": 15000, + "tokens_trained": 49.13764168 + }, + { + "epoch": 4.254166371179349, + "eval_ppl": 2.5810306134272096, + "eval_runtime": 1.4317, + "step": 15000, + "tokens_trained": 49.13764168 + }, + { + "epoch": 4.254166371179349, + "eval_ppl_cds": 3.016936286387769, + "eval_ppl_dig": 1.0962482215022684, + "eval_ppl_exon": 2.987409541723818, + "eval_ppl_intron": 2.757232669685202, + "eval_ppl_nig": 2.4928610164750276, + "eval_ppl_promoter": 2.4664301311971455, + "eval_ppl_utr": 2.7253495680079363, + "step": 15000, + "tokens_trained": 49.13764168 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 4.257003049429119, + "grad_norm": 0.20196615159511566, + "loss": 0.9281, + "loss_ce": 1.0060920715332031, + "loss_region": 0.0, + "loss_total": 1.0060920715332031, + "lr": 0.0005179734219269103, + "step": 15010, + "tokens_trained": 49.17040712 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 4.259839727678888, + "grad_norm": 0.07102058082818985, + "loss": 0.9322, + "loss_ce": 0.8237578868865967, + "loss_region": 0.0, + "loss_total": 0.8237578868865967, + "lr": 0.000517641196013289, + "step": 15020, + "tokens_trained": 49.20317256 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 4.262676405928658, + "grad_norm": 0.28695979714393616, + "loss": 0.9272, + "loss_ce": 1.077439308166504, + "loss_region": 0.0, + "loss_total": 1.077439308166504, + "lr": 0.0005173089700996678, + "step": 15030, + "tokens_trained": 49.235938 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 4.265513084178427, + "grad_norm": 0.11359913647174835, + "loss": 0.9295, + "loss_ce": 0.8940212726593018, + "loss_region": 0.0, + "loss_total": 0.8940212726593018, + "lr": 0.0005169767441860466, + "step": 15040, + "tokens_trained": 49.26870344 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 4.268349762428197, + "grad_norm": 0.30057230591773987, + "loss": 0.9326, + "loss_ce": 1.0402922630310059, + "loss_region": 0.0, + "loss_total": 1.0402922630310059, + "lr": 0.0005166445182724253, + "step": 15050, + "tokens_trained": 49.30146888 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 4.271186440677966, + "grad_norm": 0.14278599619865417, + "loss": 0.9318, + "loss_ce": 1.0237839221954346, + "loss_region": 0.0, + "loss_total": 1.0237839221954346, + "lr": 0.0005163122923588039, + "step": 15060, + "tokens_trained": 49.33423432 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 4.274023118927736, + "grad_norm": 0.16903148591518402, + "loss": 0.9299, + "loss_ce": 0.9481615424156189, + "loss_region": 0.0, + "loss_total": 0.9481615424156189, + "lr": 0.0005159800664451827, + "step": 15070, + "tokens_trained": 49.36699976 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 4.276859797177505, + "grad_norm": 0.1531340330839157, + "loss": 0.9296, + "loss_ce": 0.7572094202041626, + "loss_region": 0.0, + "loss_total": 0.7572094202041626, + "lr": 0.0005156478405315614, + "step": 15080, + "tokens_trained": 49.3997644 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 4.279696475427275, + "grad_norm": 0.16908736526966095, + "loss": 0.9262, + "loss_ce": 0.8261103630065918, + "loss_region": 0.0, + "loss_total": 0.8261103630065918, + "lr": 0.0005153156146179402, + "step": 15090, + "tokens_trained": 49.43252984 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 4.282533153677044, + "grad_norm": 0.21589568257331848, + "loss": 0.9369, + "loss_ce": 1.086761713027954, + "loss_region": 0.0, + "loss_total": 1.086761713027954, + "lr": 0.000514983388704319, + "step": 15100, + "tokens_trained": 49.46529512 + }, + { + "epoch": 4.282533153677044, + "eval_ppl": 2.5793619737190747, + "eval_runtime": 1.4273, + "step": 15100, + "tokens_trained": 49.46529512 + }, + { + "epoch": 4.282533153677044, + "eval_ppl_cds": 3.0141054791359854, + "eval_ppl_dig": 1.0976779117467883, + "eval_ppl_exon": 2.900732441151229, + "eval_ppl_intron": 2.7555828855780766, + "eval_ppl_nig": 2.4928874755046335, + "eval_ppl_promoter": 2.4553534489020614, + "eval_ppl_utr": 2.7256760291813715, + "step": 15100, + "tokens_trained": 49.46529512 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 4.285369831926814, + "grad_norm": 0.23912525177001953, + "loss": 0.9317, + "loss_ce": 1.0376296043395996, + "loss_region": 0.0, + "loss_total": 1.0376296043395996, + "lr": 0.0005146511627906977, + "step": 15110, + "tokens_trained": 49.49806056 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 4.288206510176583, + "grad_norm": 0.11948299407958984, + "loss": 0.9343, + "loss_ce": 0.8722971677780151, + "loss_region": 0.0, + "loss_total": 0.8722971677780151, + "lr": 0.0005143189368770764, + "step": 15120, + "tokens_trained": 49.530826 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 4.291043188426353, + "grad_norm": 0.052003879100084305, + "loss": 0.9301, + "loss_ce": 0.9829298853874207, + "loss_region": 0.0, + "loss_total": 0.9829298853874207, + "lr": 0.0005139867109634551, + "step": 15130, + "tokens_trained": 49.56359144 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 4.293879866676122, + "grad_norm": 0.15866383910179138, + "loss": 0.9338, + "loss_ce": 0.7746691107749939, + "loss_region": 0.0, + "loss_total": 0.7746691107749939, + "lr": 0.000513654485049834, + "step": 15140, + "tokens_trained": 49.59635688 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 4.296716544925892, + "grad_norm": 0.1146237701177597, + "loss": 0.9305, + "loss_ce": 0.8104406595230103, + "loss_region": 0.0, + "loss_total": 0.8104406595230103, + "lr": 0.0005133222591362127, + "step": 15150, + "tokens_trained": 49.62912232 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 4.2995532231756615, + "grad_norm": 0.13325276970863342, + "loss": 0.9289, + "loss_ce": 1.107825756072998, + "loss_region": 0.0, + "loss_total": 1.107825756072998, + "lr": 0.0005129900332225914, + "step": 15160, + "tokens_trained": 49.66188776 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 4.302389901425431, + "grad_norm": 0.19458481669425964, + "loss": 0.9166, + "loss_ce": 0.8240382671356201, + "loss_region": 0.0, + "loss_total": 0.8240382671356201, + "lr": 0.0005126578073089701, + "step": 15170, + "tokens_trained": 49.6946532 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 4.3052265796752005, + "grad_norm": 0.15703192353248596, + "loss": 0.9293, + "loss_ce": 1.0412712097167969, + "loss_region": 0.0, + "loss_total": 1.0412712097167969, + "lr": 0.0005123255813953487, + "step": 15180, + "tokens_trained": 49.72741588 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 4.30806325792497, + "grad_norm": 0.18391850590705872, + "loss": 0.9277, + "loss_ce": 0.8091780543327332, + "loss_region": 0.0, + "loss_total": 0.8091780543327332, + "lr": 0.0005119933554817276, + "step": 15190, + "tokens_trained": 49.76018132 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 4.3108999361747395, + "grad_norm": 0.19432410597801208, + "loss": 0.9286, + "loss_ce": 1.029596209526062, + "loss_region": 0.0, + "loss_total": 1.029596209526062, + "lr": 0.0005116611295681063, + "step": 15200, + "tokens_trained": 49.79294676 + }, + { + "epoch": 4.3108999361747395, + "eval_ppl": 2.579773112467937, + "eval_runtime": 1.4434, + "step": 15200, + "tokens_trained": 49.79294676 + }, + { + "epoch": 4.3108999361747395, + "eval_ppl_cds": 3.0171151550255124, + "eval_ppl_dig": 1.0978474194040977, + "eval_ppl_exon": 2.8749387209146735, + "eval_ppl_intron": 2.758648118168563, + "eval_ppl_nig": 2.498188033652964, + "eval_ppl_promoter": 2.451649096542282, + "eval_ppl_utr": 2.7087323676109007, + "step": 15200, + "tokens_trained": 49.79294676 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 4.313736614424509, + "grad_norm": 0.17559577524662018, + "loss": 0.9271, + "loss_ce": 1.0286651849746704, + "loss_region": 0.0, + "loss_total": 1.0286651849746704, + "lr": 0.000511328903654485, + "step": 15210, + "tokens_trained": 49.8257114 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 4.316573292674279, + "grad_norm": 0.06335200369358063, + "loss": 0.9302, + "loss_ce": 0.9148672819137573, + "loss_region": 0.0, + "loss_total": 0.9148672819137573, + "lr": 0.0005109966777408638, + "step": 15220, + "tokens_trained": 49.85847684 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 4.319409970924048, + "grad_norm": 0.22458826005458832, + "loss": 0.9291, + "loss_ce": 0.8881521224975586, + "loss_region": 0.0, + "loss_total": 0.8881521224975586, + "lr": 0.0005106644518272425, + "step": 15230, + "tokens_trained": 49.89124148 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 4.322246649173818, + "grad_norm": 0.19339519739151, + "loss": 0.9359, + "loss_ce": 0.7688725590705872, + "loss_region": 0.0, + "loss_total": 0.7688725590705872, + "lr": 0.0005103322259136213, + "step": 15240, + "tokens_trained": 49.92400692 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 4.325083327423587, + "grad_norm": 0.2637823224067688, + "loss": 0.9289, + "loss_ce": 0.9978100061416626, + "loss_region": 0.0, + "loss_total": 0.9978100061416626, + "lr": 0.00051, + "step": 15250, + "tokens_trained": 49.95677236 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 4.327920005673357, + "grad_norm": 0.1992368847131729, + "loss": 0.9345, + "loss_ce": 0.9657806754112244, + "loss_region": 0.0, + "loss_total": 0.9657806754112244, + "lr": 0.0005096677740863788, + "step": 15260, + "tokens_trained": 49.9895378 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 4.330756683923126, + "grad_norm": 0.1954343616962433, + "loss": 0.9356, + "loss_ce": 0.9019122123718262, + "loss_region": 0.0, + "loss_total": 0.9019122123718262, + "lr": 0.0005093355481727575, + "step": 15270, + "tokens_trained": 50.02230324 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 4.333593362172896, + "grad_norm": 0.1778547763824463, + "loss": 0.9207, + "loss_ce": 1.0396413803100586, + "loss_region": 0.0, + "loss_total": 1.0396413803100586, + "lr": 0.0005090033222591362, + "step": 15280, + "tokens_trained": 50.05506868 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 4.336430040422665, + "grad_norm": 0.14214618504047394, + "loss": 0.9307, + "loss_ce": 0.9619539380073547, + "loss_region": 0.0, + "loss_total": 0.9619539380073547, + "lr": 0.000508671096345515, + "step": 15290, + "tokens_trained": 50.08783412 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 4.339266718672435, + "grad_norm": 0.17765429615974426, + "loss": 0.9258, + "loss_ce": 0.9866986870765686, + "loss_region": 0.0, + "loss_total": 0.9866986870765686, + "lr": 0.0005083388704318938, + "step": 15300, + "tokens_trained": 50.12059956 + }, + { + "epoch": 4.339266718672435, + "eval_ppl": 2.5761959117087754, + "eval_runtime": 1.4267, + "step": 15300, + "tokens_trained": 50.12059956 + }, + { + "epoch": 4.339266718672435, + "eval_ppl_cds": 2.99756606120586, + "eval_ppl_dig": 1.0934701052962241, + "eval_ppl_exon": 2.9267731357281392, + "eval_ppl_intron": 2.756604946390797, + "eval_ppl_nig": 2.4966945346313363, + "eval_ppl_promoter": 2.468308610354462, + "eval_ppl_utr": 2.706221702565527, + "step": 15300, + "tokens_trained": 50.12059956 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 4.342103396922204, + "grad_norm": 0.13617455959320068, + "loss": 0.9343, + "loss_ce": 0.9448725581169128, + "loss_region": 0.0, + "loss_total": 0.9448725581169128, + "lr": 0.0005080066445182724, + "step": 15310, + "tokens_trained": 50.1533642 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 4.344940075171974, + "grad_norm": 0.14304068684577942, + "loss": 0.9281, + "loss_ce": 0.89931720495224, + "loss_region": 0.0, + "loss_total": 0.89931720495224, + "lr": 0.0005076744186046511, + "step": 15320, + "tokens_trained": 50.18612964 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 4.347776753421743, + "grad_norm": 0.1982458382844925, + "loss": 0.9308, + "loss_ce": 1.138089656829834, + "loss_region": 0.0, + "loss_total": 1.138089656829834, + "lr": 0.0005073421926910298, + "step": 15330, + "tokens_trained": 50.21889508 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 4.350613431671513, + "grad_norm": 0.14487256109714508, + "loss": 0.9344, + "loss_ce": 0.9443150162696838, + "loss_region": 0.0, + "loss_total": 0.9443150162696838, + "lr": 0.0005070099667774087, + "step": 15340, + "tokens_trained": 50.25165972 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 4.353450109921282, + "grad_norm": 0.08038835227489471, + "loss": 0.9405, + "loss_ce": 1.0328091382980347, + "loss_region": 0.0, + "loss_total": 1.0328091382980347, + "lr": 0.0005066777408637874, + "step": 15350, + "tokens_trained": 50.28442516 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 4.356286788171052, + "grad_norm": 0.07886890321969986, + "loss": 0.9239, + "loss_ce": 0.7864028811454773, + "loss_region": 0.0, + "loss_total": 0.7864028811454773, + "lr": 0.0005063455149501661, + "step": 15360, + "tokens_trained": 50.3171906 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 4.359123466420821, + "grad_norm": 0.21300162374973297, + "loss": 0.9349, + "loss_ce": 0.748558521270752, + "loss_region": 0.0, + "loss_total": 0.748558521270752, + "lr": 0.0005060132890365448, + "step": 15370, + "tokens_trained": 50.34995604 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 4.361960144670591, + "grad_norm": 0.13175861537456512, + "loss": 0.9329, + "loss_ce": 0.8637915849685669, + "loss_region": 0.0, + "loss_total": 0.8637915849685669, + "lr": 0.0005056810631229236, + "step": 15380, + "tokens_trained": 50.38272148 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 4.36479682292036, + "grad_norm": 0.12641306221485138, + "loss": 0.9247, + "loss_ce": 1.0295052528381348, + "loss_region": 0.0, + "loss_total": 1.0295052528381348, + "lr": 0.0005053488372093024, + "step": 15390, + "tokens_trained": 50.41548692 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 4.36763350117013, + "grad_norm": 0.18316739797592163, + "loss": 0.9325, + "loss_ce": 1.0001800060272217, + "loss_region": 0.0, + "loss_total": 1.0001800060272217, + "lr": 0.0005050166112956811, + "step": 15400, + "tokens_trained": 50.44825236 + }, + { + "epoch": 4.36763350117013, + "eval_ppl": 2.5743998116253097, + "eval_runtime": 1.4153, + "step": 15400, + "tokens_trained": 50.44825236 + }, + { + "epoch": 4.36763350117013, + "eval_ppl_cds": 2.986392841064382, + "eval_ppl_dig": 1.0957227622299064, + "eval_ppl_exon": 2.9764032389607884, + "eval_ppl_intron": 2.747988682373357, + "eval_ppl_nig": 2.4943605520917127, + "eval_ppl_promoter": 2.436089681230409, + "eval_ppl_utr": 2.729301147095051, + "step": 15400, + "tokens_trained": 50.44825236 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 4.3704701794198995, + "grad_norm": 0.2070658951997757, + "loss": 0.934, + "loss_ce": 1.0496149063110352, + "loss_region": 0.0, + "loss_total": 1.0496149063110352, + "lr": 0.0005046843853820599, + "step": 15410, + "tokens_trained": 50.481017 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 4.373306857669669, + "grad_norm": 0.2235785573720932, + "loss": 0.9203, + "loss_ce": 1.0714854001998901, + "loss_region": 0.0, + "loss_total": 1.0714854001998901, + "lr": 0.0005043521594684386, + "step": 15420, + "tokens_trained": 50.51378244 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 4.3761435359194385, + "grad_norm": 0.10541270673274994, + "loss": 0.9271, + "loss_ce": 0.9930810332298279, + "loss_region": 0.0, + "loss_total": 0.9930810332298279, + "lr": 0.0005040199335548172, + "step": 15430, + "tokens_trained": 50.54654788 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 4.378980214169208, + "grad_norm": 0.18978799879550934, + "loss": 0.919, + "loss_ce": 0.8183229565620422, + "loss_region": 0.0, + "loss_total": 0.8183229565620422, + "lr": 0.000503687707641196, + "step": 15440, + "tokens_trained": 50.57931332 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 4.3818168924189775, + "grad_norm": 0.25611600279808044, + "loss": 0.9347, + "loss_ce": 0.9658758640289307, + "loss_region": 0.0, + "loss_total": 0.9658758640289307, + "lr": 0.0005033554817275747, + "step": 15450, + "tokens_trained": 50.61207876 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 4.384653570668747, + "grad_norm": 0.12053602933883667, + "loss": 0.9183, + "loss_ce": 0.9227601885795593, + "loss_region": 0.0, + "loss_total": 0.9227601885795593, + "lr": 0.0005030232558139535, + "step": 15460, + "tokens_trained": 50.6448442 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 4.387490248918517, + "grad_norm": 0.26709434390068054, + "loss": 0.9207, + "loss_ce": 1.0827765464782715, + "loss_region": 0.0, + "loss_total": 1.0827765464782715, + "lr": 0.0005026910299003322, + "step": 15470, + "tokens_trained": 50.67760964 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 4.390326927168286, + "grad_norm": 0.08132723718881607, + "loss": 0.9266, + "loss_ce": 0.9786382913589478, + "loss_region": 0.0, + "loss_total": 0.9786382913589478, + "lr": 0.0005023588039867109, + "step": 15480, + "tokens_trained": 50.71037508 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 4.393163605418056, + "grad_norm": 0.11918062716722488, + "loss": 0.9256, + "loss_ce": 1.0620453357696533, + "loss_region": 0.0, + "loss_total": 1.0620453357696533, + "lr": 0.0005020265780730898, + "step": 15490, + "tokens_trained": 50.74314052 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 4.396000283667825, + "grad_norm": 0.43450960516929626, + "loss": 0.9436, + "loss_ce": 0.9900060892105103, + "loss_region": 0.0, + "loss_total": 0.9900060892105103, + "lr": 0.0005016943521594685, + "step": 15500, + "tokens_trained": 50.77590596 + }, + { + "epoch": 4.396000283667825, + "eval_ppl": 2.5737713727631495, + "eval_runtime": 1.4173, + "step": 15500, + "tokens_trained": 50.77590596 + }, + { + "epoch": 4.396000283667825, + "eval_ppl_cds": 2.995901359356415, + "eval_ppl_dig": 1.0980625611548562, + "eval_ppl_exon": 2.9983486218145172, + "eval_ppl_intron": 2.7573470299180043, + "eval_ppl_nig": 2.5018042728258205, + "eval_ppl_promoter": 2.449795500824944, + "eval_ppl_utr": 2.7504203955735034, + "step": 15500, + "tokens_trained": 50.77590596 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 4.398836961917595, + "grad_norm": 0.13150915503501892, + "loss": 0.921, + "loss_ce": 0.9996090531349182, + "loss_region": 0.0, + "loss_total": 0.9996090531349182, + "lr": 0.0005013621262458472, + "step": 15510, + "tokens_trained": 50.8086706 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 4.401673640167364, + "grad_norm": 0.19343218207359314, + "loss": 0.9294, + "loss_ce": 0.9878471493721008, + "loss_region": 0.0, + "loss_total": 0.9878471493721008, + "lr": 0.0005010299003322259, + "step": 15520, + "tokens_trained": 50.841434264 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 4.404510318417134, + "grad_norm": 0.14430224895477295, + "loss": 0.9326, + "loss_ce": 0.7629892826080322, + "loss_region": 0.0, + "loss_total": 0.7629892826080322, + "lr": 0.0005006976744186047, + "step": 15530, + "tokens_trained": 50.874199704 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 4.407346996666903, + "grad_norm": 0.10339469462633133, + "loss": 0.938, + "loss_ce": 1.1004698276519775, + "loss_region": 0.0, + "loss_total": 1.1004698276519775, + "lr": 0.0005003654485049835, + "step": 15540, + "tokens_trained": 50.90696476 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 4.410183674916673, + "grad_norm": 0.24511145055294037, + "loss": 0.9297, + "loss_ce": 1.0108002424240112, + "loss_region": 0.0, + "loss_total": 1.0108002424240112, + "lr": 0.0005000332225913622, + "step": 15550, + "tokens_trained": 50.9397294 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 4.413020353166442, + "grad_norm": 0.1939845234155655, + "loss": 0.9348, + "loss_ce": 1.1254174709320068, + "loss_region": 0.0, + "loss_total": 1.1254174709320068, + "lr": 0.0004997009966777408, + "step": 15560, + "tokens_trained": 50.97249484 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 4.415857031416212, + "grad_norm": 0.17064185440540314, + "loss": 0.9328, + "loss_ce": 1.0714191198349, + "loss_region": 0.0, + "loss_total": 1.0714191198349, + "lr": 0.0004993687707641197, + "step": 15570, + "tokens_trained": 51.00526028 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 4.418693709665981, + "grad_norm": 0.0969436764717102, + "loss": 0.9364, + "loss_ce": 0.7118410468101501, + "loss_region": 0.0, + "loss_total": 0.7118410468101501, + "lr": 0.0004990365448504984, + "step": 15580, + "tokens_trained": 51.03802464 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 4.421530387915751, + "grad_norm": 0.1562354862689972, + "loss": 0.9423, + "loss_ce": 0.9648487567901611, + "loss_region": 0.0, + "loss_total": 0.9648487567901611, + "lr": 0.0004987043189368771, + "step": 15590, + "tokens_trained": 51.07079008 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 4.42436706616552, + "grad_norm": 0.1782713532447815, + "loss": 0.9459, + "loss_ce": 0.9606878161430359, + "loss_region": 0.0, + "loss_total": 0.9606878161430359, + "lr": 0.0004983720930232558, + "step": 15600, + "tokens_trained": 51.10355552 + }, + { + "epoch": 4.42436706616552, + "eval_ppl": 2.5758843710192796, + "eval_runtime": 1.4297, + "step": 15600, + "tokens_trained": 51.10355552 + }, + { + "epoch": 4.42436706616552, + "eval_ppl_cds": 3.004282506656525, + "eval_ppl_dig": 1.098586271581108, + "eval_ppl_exon": 2.953104478381982, + "eval_ppl_intron": 2.7612088829259402, + "eval_ppl_nig": 2.501551183319815, + "eval_ppl_promoter": 2.4612039311753677, + "eval_ppl_utr": 2.76213690119518, + "step": 15600, + "tokens_trained": 51.10355552 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 4.42720374441529, + "grad_norm": 0.11753146350383759, + "loss": 0.9325, + "loss_ce": 0.9402565956115723, + "loss_region": 0.0, + "loss_total": 0.9402565956115723, + "lr": 0.0004980398671096346, + "step": 15610, + "tokens_trained": 51.13632096 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 4.430040422665059, + "grad_norm": 0.17790931463241577, + "loss": 0.9284, + "loss_ce": 0.8795666098594666, + "loss_region": 0.0, + "loss_total": 0.8795666098594666, + "lr": 0.0004977076411960133, + "step": 15620, + "tokens_trained": 51.1690856 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 4.432877100914829, + "grad_norm": 0.08770544081926346, + "loss": 0.9192, + "loss_ce": 0.9186168313026428, + "loss_region": 0.0, + "loss_total": 0.9186168313026428, + "lr": 0.000497375415282392, + "step": 15630, + "tokens_trained": 51.20185104 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 4.435713779164598, + "grad_norm": 0.11992325633764267, + "loss": 0.9397, + "loss_ce": 1.0753116607666016, + "loss_region": 0.0, + "loss_total": 1.0753116607666016, + "lr": 0.0004970431893687708, + "step": 15640, + "tokens_trained": 51.23461648 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 4.438550457414368, + "grad_norm": 0.1275496929883957, + "loss": 0.9227, + "loss_ce": 0.9865513443946838, + "loss_region": 0.0, + "loss_total": 0.9865513443946838, + "lr": 0.0004967109634551495, + "step": 15650, + "tokens_trained": 51.26738112 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 4.4413871356641375, + "grad_norm": 0.10570922493934631, + "loss": 0.9319, + "loss_ce": 0.8681632280349731, + "loss_region": 0.0, + "loss_total": 0.8681632280349731, + "lr": 0.0004963787375415282, + "step": 15660, + "tokens_trained": 51.30014656 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 4.444223813913907, + "grad_norm": 0.32851770520210266, + "loss": 0.9282, + "loss_ce": 0.8650022745132446, + "loss_region": 0.0, + "loss_total": 0.8650022745132446, + "lr": 0.000496046511627907, + "step": 15670, + "tokens_trained": 51.332912 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 4.4470604921636765, + "grad_norm": 0.12826789915561676, + "loss": 0.9239, + "loss_ce": 0.8721836805343628, + "loss_region": 0.0, + "loss_total": 0.8721836805343628, + "lr": 0.0004957142857142857, + "step": 15680, + "tokens_trained": 51.36567744 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 4.449897170413446, + "grad_norm": 0.18927614390850067, + "loss": 0.9262, + "loss_ce": 0.9267632365226746, + "loss_region": 0.0, + "loss_total": 0.9267632365226746, + "lr": 0.0004953820598006645, + "step": 15690, + "tokens_trained": 51.39844208 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 4.4527338486632155, + "grad_norm": 0.17039854824543, + "loss": 0.9291, + "loss_ce": 0.5154620409011841, + "loss_region": 0.0, + "loss_total": 0.5154620409011841, + "lr": 0.0004950498338870432, + "step": 15700, + "tokens_trained": 51.431207512 + }, + { + "epoch": 4.4527338486632155, + "eval_ppl": 2.5742386981713996, + "eval_runtime": 1.4188, + "step": 15700, + "tokens_trained": 51.431207512 + }, + { + "epoch": 4.4527338486632155, + "eval_ppl_cds": 3.0144675394264744, + "eval_ppl_dig": 1.0944766394467502, + "eval_ppl_exon": 2.988564358848134, + "eval_ppl_intron": 2.759357468294496, + "eval_ppl_nig": 2.5027258831106285, + "eval_ppl_promoter": 2.4654970470274877, + "eval_ppl_utr": 2.776436368143082, + "step": 15700, + "tokens_trained": 51.431207512 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 4.455570526912985, + "grad_norm": 0.13729164004325867, + "loss": 0.9288, + "loss_ce": 0.8835110664367676, + "loss_region": 0.0, + "loss_total": 0.8835110664367676, + "lr": 0.0004947176079734219, + "step": 15710, + "tokens_trained": 51.463972952 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 4.458407205162755, + "grad_norm": 0.09196799248456955, + "loss": 0.9264, + "loss_ce": 0.9608463644981384, + "loss_region": 0.0, + "loss_total": 0.9608463644981384, + "lr": 0.0004943853820598006, + "step": 15720, + "tokens_trained": 51.496738392 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 4.461243883412524, + "grad_norm": 0.10296088457107544, + "loss": 0.9371, + "loss_ce": 0.9518950581550598, + "loss_region": 0.0, + "loss_total": 0.9518950581550598, + "lr": 0.0004940531561461794, + "step": 15730, + "tokens_trained": 51.529503832 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 4.464080561662294, + "grad_norm": 0.1622326672077179, + "loss": 0.9404, + "loss_ce": 0.8943182229995728, + "loss_region": 0.0, + "loss_total": 0.8943182229995728, + "lr": 0.0004937209302325582, + "step": 15740, + "tokens_trained": 51.562268472 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 4.466917239912063, + "grad_norm": 0.15834426879882812, + "loss": 0.9262, + "loss_ce": 0.9945397973060608, + "loss_region": 0.0, + "loss_total": 0.9945397973060608, + "lr": 0.0004933887043189369, + "step": 15750, + "tokens_trained": 51.595033912 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 4.469753918161833, + "grad_norm": 0.12652069330215454, + "loss": 0.9293, + "loss_ce": 0.9882761240005493, + "loss_region": 0.0, + "loss_total": 0.9882761240005493, + "lr": 0.0004930564784053157, + "step": 15760, + "tokens_trained": 51.627799352 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 4.472590596411602, + "grad_norm": 0.14466051757335663, + "loss": 0.9357, + "loss_ce": 0.9301459193229675, + "loss_region": 0.0, + "loss_total": 0.9301459193229675, + "lr": 0.0004927242524916944, + "step": 15770, + "tokens_trained": 51.660564792 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 4.475427274661372, + "grad_norm": 0.18799051642417908, + "loss": 0.9275, + "loss_ce": 0.7369033098220825, + "loss_region": 0.0, + "loss_total": 0.7369033098220825, + "lr": 0.0004923920265780731, + "step": 15780, + "tokens_trained": 51.693329432 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 4.478263952911141, + "grad_norm": 0.06953403353691101, + "loss": 0.9291, + "loss_ce": 0.9005758762359619, + "loss_region": 0.0, + "loss_total": 0.9005758762359619, + "lr": 0.0004920598006644518, + "step": 15790, + "tokens_trained": 51.726094072 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 4.481100631160911, + "grad_norm": 0.15121448040008545, + "loss": 0.9304, + "loss_ce": 1.0478205680847168, + "loss_region": 0.0, + "loss_total": 1.0478205680847168, + "lr": 0.0004917275747508306, + "step": 15800, + "tokens_trained": 51.758857112 + }, + { + "epoch": 4.481100631160911, + "eval_ppl": 2.5758467552990694, + "eval_runtime": 1.4306, + "step": 15800, + "tokens_trained": 51.758857112 + }, + { + "epoch": 4.481100631160911, + "eval_ppl_cds": 3.013906424842157, + "eval_ppl_dig": 1.0949424548160496, + "eval_ppl_exon": 3.0084914675280365, + "eval_ppl_intron": 2.7586568919194323, + "eval_ppl_nig": 2.5002717424912118, + "eval_ppl_promoter": 2.4532246099296966, + "eval_ppl_utr": 2.783884390730955, + "step": 15800, + "tokens_trained": 51.758857112 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 4.48393730941068, + "grad_norm": 0.1478106528520584, + "loss": 0.928, + "loss_ce": 0.8704930543899536, + "loss_region": 0.0, + "loss_total": 0.8704930543899536, + "lr": 0.0004913953488372093, + "step": 15810, + "tokens_trained": 51.791622552 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 4.48677398766045, + "grad_norm": 0.20750539004802704, + "loss": 0.9423, + "loss_ce": 0.9225895404815674, + "loss_region": 0.0, + "loss_total": 0.9225895404815674, + "lr": 0.0004910631229235881, + "step": 15820, + "tokens_trained": 51.824387992 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 4.489610665910219, + "grad_norm": 0.11356493830680847, + "loss": 0.9311, + "loss_ce": 0.9070982933044434, + "loss_region": 0.0, + "loss_total": 0.9070982933044434, + "lr": 0.0004907308970099668, + "step": 15830, + "tokens_trained": 51.857153432 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 4.492447344159989, + "grad_norm": 0.18913142383098602, + "loss": 0.9383, + "loss_ce": 1.0711545944213867, + "loss_region": 0.0, + "loss_total": 1.0711545944213867, + "lr": 0.0004903986710963456, + "step": 15840, + "tokens_trained": 51.889918072 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 4.495284022409758, + "grad_norm": 0.17819862067699432, + "loss": 0.9219, + "loss_ce": 1.0315732955932617, + "loss_region": 0.0, + "loss_total": 1.0315732955932617, + "lr": 0.0004900664451827243, + "step": 15850, + "tokens_trained": 51.922683512 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 4.498120700659528, + "grad_norm": 0.1523796170949936, + "loss": 0.9362, + "loss_ce": 0.7382262945175171, + "loss_region": 0.0, + "loss_total": 0.7382262945175171, + "lr": 0.000489734219269103, + "step": 15860, + "tokens_trained": 51.955448952 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 4.500957378909297, + "grad_norm": 0.21345892548561096, + "loss": 0.9344, + "loss_ce": 0.6751561760902405, + "loss_region": 0.0, + "loss_total": 0.6751561760902405, + "lr": 0.0004894019933554817, + "step": 15870, + "tokens_trained": 51.988214392 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 4.503794057159067, + "grad_norm": 0.14908160269260406, + "loss": 0.9286, + "loss_ce": 1.0536459684371948, + "loss_region": 0.0, + "loss_total": 1.0536459684371948, + "lr": 0.0004890697674418605, + "step": 15880, + "tokens_trained": 52.020979032 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 4.506630735408836, + "grad_norm": 0.14660301804542542, + "loss": 0.9325, + "loss_ce": 0.8004583120346069, + "loss_region": 0.0, + "loss_total": 0.8004583120346069, + "lr": 0.0004887375415282393, + "step": 15890, + "tokens_trained": 52.053744472 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 4.509467413658606, + "grad_norm": 0.09811637550592422, + "loss": 0.9317, + "loss_ce": 0.8321784138679504, + "loss_region": 0.0, + "loss_total": 0.8321784138679504, + "lr": 0.0004884053156146179, + "step": 15900, + "tokens_trained": 52.086508312 + }, + { + "epoch": 4.509467413658606, + "eval_ppl": 2.56893480360411, + "eval_runtime": 1.4296, + "step": 15900, + "tokens_trained": 52.086508312 + }, + { + "epoch": 4.509467413658606, + "eval_ppl_cds": 3.016316558065203, + "eval_ppl_dig": 1.0968727572594354, + "eval_ppl_exon": 3.01710630604572, + "eval_ppl_intron": 2.759232009648304, + "eval_ppl_nig": 2.502205452081182, + "eval_ppl_promoter": 2.461293897300532, + "eval_ppl_utr": 2.814303158872567, + "step": 15900, + "tokens_trained": 52.086508312 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 4.5123040919083754, + "grad_norm": 0.30091455578804016, + "loss": 0.9248, + "loss_ce": 1.0530736446380615, + "loss_region": 0.0, + "loss_total": 1.0530736446380615, + "lr": 0.0004880730897009967, + "step": 15910, + "tokens_trained": 52.119273752 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 4.515140770158145, + "grad_norm": 0.09840469807386398, + "loss": 0.9231, + "loss_ce": 0.9508024454116821, + "loss_region": 0.0, + "loss_total": 0.9508024454116821, + "lr": 0.0004877408637873754, + "step": 15920, + "tokens_trained": 52.152039192 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 4.5179774484079145, + "grad_norm": 0.13455446064472198, + "loss": 0.9237, + "loss_ce": 1.0692213773727417, + "loss_region": 0.0, + "loss_total": 1.0692213773727417, + "lr": 0.0004874086378737542, + "step": 15930, + "tokens_trained": 52.184804632 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 4.520814126657684, + "grad_norm": 0.08803252875804901, + "loss": 0.9303, + "loss_ce": 0.9568328261375427, + "loss_region": 0.0, + "loss_total": 0.9568328261375427, + "lr": 0.0004870764119601329, + "step": 15940, + "tokens_trained": 52.217570072 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 4.5236508049074535, + "grad_norm": 0.20695538818836212, + "loss": 0.9319, + "loss_ce": 0.8277720212936401, + "loss_region": 0.0, + "loss_total": 0.8277720212936401, + "lr": 0.00048674418604651164, + "step": 15950, + "tokens_trained": 52.250335512 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 4.526487483157223, + "grad_norm": 0.11362215876579285, + "loss": 0.9236, + "loss_ce": 0.8344137668609619, + "loss_region": 0.0, + "loss_total": 0.8344137668609619, + "lr": 0.00048641196013289036, + "step": 15960, + "tokens_trained": 52.283100952 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 4.529324161406993, + "grad_norm": 0.2659211754798889, + "loss": 0.925, + "loss_ce": 0.9331261515617371, + "loss_region": 0.0, + "loss_total": 0.9331261515617371, + "lr": 0.0004860797342192691, + "step": 15970, + "tokens_trained": 52.315866392 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 4.532160839656762, + "grad_norm": 0.16167588531970978, + "loss": 0.9185, + "loss_ce": 0.9107192754745483, + "loss_region": 0.0, + "loss_total": 0.9107192754745483, + "lr": 0.00048574750830564787, + "step": 15980, + "tokens_trained": 52.348631832 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 4.534997517906532, + "grad_norm": 0.12289244681596756, + "loss": 0.9306, + "loss_ce": 0.9790875315666199, + "loss_region": 0.0, + "loss_total": 0.9790875315666199, + "lr": 0.0004854152823920266, + "step": 15990, + "tokens_trained": 52.381397272 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 4.537834196156301, + "grad_norm": 0.12660706043243408, + "loss": 0.9257, + "loss_ce": 0.9312461614608765, + "loss_region": 0.0, + "loss_total": 0.9312461614608765, + "lr": 0.0004850830564784053, + "step": 16000, + "tokens_trained": 52.414162712 + }, + { + "epoch": 4.537834196156301, + "eval_ppl": 2.5676368233582214, + "eval_runtime": 1.4242, + "step": 16000, + "tokens_trained": 52.414162712 + }, + { + "epoch": 4.537834196156301, + "eval_ppl_cds": 2.9919811728647456, + "eval_ppl_dig": 1.0984822114601522, + "eval_ppl_exon": 2.958704259110702, + "eval_ppl_intron": 2.7535587057033575, + "eval_ppl_nig": 2.496332929788112, + "eval_ppl_promoter": 2.44096141946489, + "eval_ppl_utr": 2.7935482482294667, + "step": 16000, + "tokens_trained": 52.414162712 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 4.540670874406071, + "grad_norm": 0.08722657710313797, + "loss": 0.9296, + "loss_ce": 0.9602150917053223, + "loss_region": 0.0, + "loss_total": 0.9602150917053223, + "lr": 0.0004847508305647841, + "step": 16010, + "tokens_trained": 52.446928152 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 4.54350755265584, + "grad_norm": 0.22613491117954254, + "loss": 0.9253, + "loss_ce": 1.0069479942321777, + "loss_region": 0.0, + "loss_total": 1.0069479942321777, + "lr": 0.00048441860465116277, + "step": 16020, + "tokens_trained": 52.479693592 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 4.54634423090561, + "grad_norm": 0.15536433458328247, + "loss": 0.9305, + "loss_ce": 1.0051676034927368, + "loss_region": 0.0, + "loss_total": 1.0051676034927368, + "lr": 0.00048408637873754155, + "step": 16030, + "tokens_trained": 52.512459032 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 4.549180909155379, + "grad_norm": 0.12669970095157623, + "loss": 0.926, + "loss_ce": 0.6448575258255005, + "loss_region": 0.0, + "loss_total": 0.6448575258255005, + "lr": 0.00048375415282392027, + "step": 16040, + "tokens_trained": 52.545224472 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 4.552017587405149, + "grad_norm": 0.14614376425743103, + "loss": 0.916, + "loss_ce": 0.8516849279403687, + "loss_region": 0.0, + "loss_total": 0.8516849279403687, + "lr": 0.000483421926910299, + "step": 16050, + "tokens_trained": 52.577989112 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 4.554854265654918, + "grad_norm": 0.16248199343681335, + "loss": 0.9249, + "loss_ce": 0.9332535862922668, + "loss_region": 0.0, + "loss_total": 0.9332535862922668, + "lr": 0.0004830897009966778, + "step": 16060, + "tokens_trained": 52.610751224 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 4.557690943904688, + "grad_norm": 0.048990897834300995, + "loss": 0.9231, + "loss_ce": 0.9490761160850525, + "loss_region": 0.0, + "loss_total": 0.9490761160850525, + "lr": 0.0004827574750830565, + "step": 16070, + "tokens_trained": 52.643516664 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 4.560527622154457, + "grad_norm": 0.31222665309906006, + "loss": 0.9194, + "loss_ce": 1.0577529668807983, + "loss_region": 0.0, + "loss_total": 1.0577529668807983, + "lr": 0.0004824252491694352, + "step": 16080, + "tokens_trained": 52.676281304 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 4.563364300404227, + "grad_norm": 0.13834363222122192, + "loss": 0.9226, + "loss_ce": 1.0636241436004639, + "loss_region": 0.0, + "loss_total": 1.0636241436004639, + "lr": 0.00048209302325581395, + "step": 16090, + "tokens_trained": 52.709046744 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 4.566200978653996, + "grad_norm": 0.10975273698568344, + "loss": 0.9131, + "loss_ce": 1.0472058057785034, + "loss_region": 0.0, + "loss_total": 1.0472058057785034, + "lr": 0.0004817607973421927, + "step": 16100, + "tokens_trained": 52.741812184 + }, + { + "epoch": 4.566200978653996, + "eval_ppl": 2.56892837255341, + "eval_runtime": 1.4345, + "step": 16100, + "tokens_trained": 52.741812184 + }, + { + "epoch": 4.566200978653996, + "eval_ppl_cds": 3.0037928744804847, + "eval_ppl_dig": 1.0972039318276103, + "eval_ppl_exon": 2.994856306167148, + "eval_ppl_intron": 2.7515818566290324, + "eval_ppl_nig": 2.5017525643281924, + "eval_ppl_promoter": 2.4422033445300673, + "eval_ppl_utr": 2.8066433550111407, + "step": 16100, + "tokens_trained": 52.741812184 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 4.569037656903766, + "grad_norm": 0.11331555992364883, + "loss": 0.9242, + "loss_ce": 0.9629917144775391, + "loss_region": 0.0, + "loss_total": 0.9629917144775391, + "lr": 0.00048142857142857145, + "step": 16110, + "tokens_trained": 52.774577624 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 4.571874335153535, + "grad_norm": 0.19050516188144684, + "loss": 0.9314, + "loss_ce": 0.9416550993919373, + "loss_region": 0.0, + "loss_total": 0.9416550993919373, + "lr": 0.0004810963455149502, + "step": 16120, + "tokens_trained": 52.807342264 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 4.574711013403305, + "grad_norm": 0.12779501080513, + "loss": 0.9296, + "loss_ce": 0.8560206294059753, + "loss_region": 0.0, + "loss_total": 0.8560206294059753, + "lr": 0.00048076411960132896, + "step": 16130, + "tokens_trained": 52.840106104 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 4.577547691653074, + "grad_norm": 0.1036221832036972, + "loss": 0.9255, + "loss_ce": 0.9927167892456055, + "loss_region": 0.0, + "loss_total": 0.9927167892456055, + "lr": 0.0004804318936877077, + "step": 16140, + "tokens_trained": 52.872871544 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 4.580384369902844, + "grad_norm": 0.17344117164611816, + "loss": 0.9233, + "loss_ce": 0.9769418239593506, + "loss_region": 0.0, + "loss_total": 0.9769418239593506, + "lr": 0.00048009966777408635, + "step": 16150, + "tokens_trained": 52.905633752 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 4.583221048152613, + "grad_norm": 0.1484891027212143, + "loss": 0.9302, + "loss_ce": 0.8622235655784607, + "loss_region": 0.0, + "loss_total": 0.8622235655784607, + "lr": 0.00047976744186046513, + "step": 16160, + "tokens_trained": 52.938399192 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 4.586057726402383, + "grad_norm": 0.13628415763378143, + "loss": 0.9282, + "loss_ce": 0.798586368560791, + "loss_region": 0.0, + "loss_total": 0.798586368560791, + "lr": 0.00047943521594684386, + "step": 16170, + "tokens_trained": 52.971164632 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 4.5888944046521525, + "grad_norm": 0.09560734778642654, + "loss": 0.9241, + "loss_ce": 1.022865891456604, + "loss_region": 0.0, + "loss_total": 1.022865891456604, + "lr": 0.00047910299003322263, + "step": 16180, + "tokens_trained": 53.003930072 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 4.591731082901922, + "grad_norm": 0.17656567692756653, + "loss": 0.934, + "loss_ce": 1.075649380683899, + "loss_region": 0.0, + "loss_total": 1.075649380683899, + "lr": 0.00047877076411960136, + "step": 16190, + "tokens_trained": 53.036695512 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 4.5945677611516915, + "grad_norm": 0.11458529531955719, + "loss": 0.9303, + "loss_ce": 0.8241126537322998, + "loss_region": 0.0, + "loss_total": 0.8241126537322998, + "lr": 0.0004784385382059801, + "step": 16200, + "tokens_trained": 53.069460952 + }, + { + "epoch": 4.5945677611516915, + "eval_ppl": 2.5693912962088197, + "eval_runtime": 1.4205, + "step": 16200, + "tokens_trained": 53.069460952 + }, + { + "epoch": 4.5945677611516915, + "eval_ppl_cds": 2.9933804379461106, + "eval_ppl_dig": 1.0894713901583546, + "eval_ppl_exon": 2.937377558125047, + "eval_ppl_intron": 2.753558251473607, + "eval_ppl_nig": 2.5000895466512505, + "eval_ppl_promoter": 2.444341635053944, + "eval_ppl_utr": 2.803990675661993, + "step": 16200, + "tokens_trained": 53.069460952 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 4.597404439401461, + "grad_norm": 0.18281662464141846, + "loss": 0.9182, + "loss_ce": 0.9379709362983704, + "loss_region": 0.0, + "loss_total": 0.9379709362983704, + "lr": 0.0004781063122923588, + "step": 16210, + "tokens_trained": 53.102226392 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 4.600241117651231, + "grad_norm": 0.15752702951431274, + "loss": 0.9213, + "loss_ce": 0.9892854690551758, + "loss_region": 0.0, + "loss_total": 0.9892854690551758, + "lr": 0.00047777408637873753, + "step": 16220, + "tokens_trained": 53.134988 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 4.603077795901, + "grad_norm": 0.09169012308120728, + "loss": 0.9254, + "loss_ce": 0.999997079372406, + "loss_region": 0.0, + "loss_total": 0.999997079372406, + "lr": 0.00047744186046511626, + "step": 16230, + "tokens_trained": 53.16775344 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 4.60591447415077, + "grad_norm": 0.10625771433115005, + "loss": 0.9229, + "loss_ce": 0.9337685108184814, + "loss_region": 0.0, + "loss_total": 0.9337685108184814, + "lr": 0.00047710963455149504, + "step": 16240, + "tokens_trained": 53.200515552 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 4.608751152400539, + "grad_norm": 0.08325252681970596, + "loss": 0.931, + "loss_ce": 1.0336110591888428, + "loss_region": 0.0, + "loss_total": 1.0336110591888428, + "lr": 0.00047677740863787376, + "step": 16250, + "tokens_trained": 53.233280992 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 4.611587830650309, + "grad_norm": 0.13937987387180328, + "loss": 0.9373, + "loss_ce": 1.0750398635864258, + "loss_region": 0.0, + "loss_total": 1.0750398635864258, + "lr": 0.00047644518272425254, + "step": 16260, + "tokens_trained": 53.266046432 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 4.614424508900078, + "grad_norm": 0.17220363020896912, + "loss": 0.9319, + "loss_ce": 1.0887336730957031, + "loss_region": 0.0, + "loss_total": 1.0887336730957031, + "lr": 0.0004761129568106312, + "step": 16270, + "tokens_trained": 53.298811864 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 4.617261187149848, + "grad_norm": 0.1152011975646019, + "loss": 0.9304, + "loss_ce": 0.9676727056503296, + "loss_region": 0.0, + "loss_total": 0.9676727056503296, + "lr": 0.00047578073089700994, + "step": 16280, + "tokens_trained": 53.331576504 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 4.620097865399617, + "grad_norm": 0.11433112621307373, + "loss": 0.9134, + "loss_ce": 0.9733253121376038, + "loss_region": 0.0, + "loss_total": 0.9733253121376038, + "lr": 0.0004754485049833887, + "step": 16290, + "tokens_trained": 53.364341944 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 4.622934543649387, + "grad_norm": 0.19164679944515228, + "loss": 0.9134, + "loss_ce": 0.9794634580612183, + "loss_region": 0.0, + "loss_total": 0.9794634580612183, + "lr": 0.00047511627906976744, + "step": 16300, + "tokens_trained": 53.397106584 + }, + { + "epoch": 4.622934543649387, + "eval_ppl": 2.5640879135290966, + "eval_runtime": 1.4171, + "step": 16300, + "tokens_trained": 53.397106584 + }, + { + "epoch": 4.622934543649387, + "eval_ppl_cds": 2.996289910703147, + "eval_ppl_dig": 1.0942589127836253, + "eval_ppl_exon": 2.9814079445165285, + "eval_ppl_intron": 2.7508917856487187, + "eval_ppl_nig": 2.496621728733375, + "eval_ppl_promoter": 2.4398726012191156, + "eval_ppl_utr": 2.797715360198447, + "step": 16300, + "tokens_trained": 53.397106584 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 4.625771221899156, + "grad_norm": 0.11532744020223618, + "loss": 0.918, + "loss_ce": 0.803875207901001, + "loss_region": 0.0, + "loss_total": 0.803875207901001, + "lr": 0.0004747840531561462, + "step": 16310, + "tokens_trained": 53.429872024 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 4.628607900148926, + "grad_norm": 0.1509167104959488, + "loss": 0.9274, + "loss_ce": 0.7876687049865723, + "loss_region": 0.0, + "loss_total": 0.7876687049865723, + "lr": 0.00047445182724252494, + "step": 16320, + "tokens_trained": 53.462637464 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 4.631444578398695, + "grad_norm": 0.08353590220212936, + "loss": 0.9244, + "loss_ce": 1.0277942419052124, + "loss_region": 0.0, + "loss_total": 1.0277942419052124, + "lr": 0.0004741196013289036, + "step": 16330, + "tokens_trained": 53.495402904 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 4.634281256648465, + "grad_norm": 0.21260319650173187, + "loss": 0.9195, + "loss_ce": 1.058595895767212, + "loss_region": 0.0, + "loss_total": 1.058595895767212, + "lr": 0.0004737873754152824, + "step": 16340, + "tokens_trained": 53.528168288 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 4.637117934898234, + "grad_norm": 0.1452857255935669, + "loss": 0.9284, + "loss_ce": 1.0082448720932007, + "loss_region": 0.0, + "loss_total": 1.0082448720932007, + "lr": 0.0004734551495016611, + "step": 16350, + "tokens_trained": 53.560933728 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 4.639954613148004, + "grad_norm": 0.11499320715665817, + "loss": 0.9333, + "loss_ce": 0.9481572508811951, + "loss_region": 0.0, + "loss_total": 0.9481572508811951, + "lr": 0.0004731229235880399, + "step": 16360, + "tokens_trained": 53.593696768 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 4.642791291397773, + "grad_norm": 0.13798771798610687, + "loss": 0.9159, + "loss_ce": 1.0439023971557617, + "loss_region": 0.0, + "loss_total": 1.0439023971557617, + "lr": 0.0004727906976744186, + "step": 16370, + "tokens_trained": 53.626462208 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 4.645627969647543, + "grad_norm": 0.12522056698799133, + "loss": 0.9229, + "loss_ce": 1.0254292488098145, + "loss_region": 0.0, + "loss_total": 1.0254292488098145, + "lr": 0.00047245847176079735, + "step": 16380, + "tokens_trained": 53.659227648 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 4.648464647897312, + "grad_norm": 0.20186886191368103, + "loss": 0.9117, + "loss_ce": 0.9411741495132446, + "loss_region": 0.0, + "loss_total": 0.9411741495132446, + "lr": 0.0004721262458471761, + "step": 16390, + "tokens_trained": 53.691993088 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 4.651301326147082, + "grad_norm": 0.1198345199227333, + "loss": 0.9346, + "loss_ce": 0.940863847732544, + "loss_region": 0.0, + "loss_total": 0.940863847732544, + "lr": 0.0004717940199335548, + "step": 16400, + "tokens_trained": 53.724758528 + }, + { + "epoch": 4.651301326147082, + "eval_ppl": 2.564423095022431, + "eval_runtime": 1.4399, + "step": 16400, + "tokens_trained": 53.724758528 + }, + { + "epoch": 4.651301326147082, + "eval_ppl_cds": 3.00002029720599, + "eval_ppl_dig": 1.0952364543097932, + "eval_ppl_exon": 3.0233413428279285, + "eval_ppl_intron": 2.753812292672633, + "eval_ppl_nig": 2.4955212532454656, + "eval_ppl_promoter": 2.444013442258783, + "eval_ppl_utr": 2.8092867118025313, + "step": 16400, + "tokens_trained": 53.724758528 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 4.654138004396851, + "grad_norm": 0.15076102316379547, + "loss": 0.9271, + "loss_ce": 0.8207283020019531, + "loss_region": 0.0, + "loss_total": 0.8207283020019531, + "lr": 0.0004714617940199336, + "step": 16410, + "tokens_trained": 53.757523968 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 4.656974682646621, + "grad_norm": 0.17216069996356964, + "loss": 0.9201, + "loss_ce": 1.0633656978607178, + "loss_region": 0.0, + "loss_total": 1.0633656978607178, + "lr": 0.0004711295681063123, + "step": 16420, + "tokens_trained": 53.790289408 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 4.6598113608963905, + "grad_norm": 0.11065241694450378, + "loss": 0.9311, + "loss_ce": 0.9911098480224609, + "loss_region": 0.0, + "loss_total": 0.9911098480224609, + "lr": 0.000470797342192691, + "step": 16430, + "tokens_trained": 53.823054848 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 4.66264803914616, + "grad_norm": 0.1397082805633545, + "loss": 0.9292, + "loss_ce": 0.9798691868782043, + "loss_region": 0.0, + "loss_total": 0.9798691868782043, + "lr": 0.0004704651162790698, + "step": 16440, + "tokens_trained": 53.855819488 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 4.6654847173959295, + "grad_norm": 0.16542689502239227, + "loss": 0.9267, + "loss_ce": 1.046899437904358, + "loss_region": 0.0, + "loss_total": 1.046899437904358, + "lr": 0.00047013289036544853, + "step": 16450, + "tokens_trained": 53.888584928 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 4.668321395645699, + "grad_norm": 0.1302492320537567, + "loss": 0.9266, + "loss_ce": 0.9942110180854797, + "loss_region": 0.0, + "loss_total": 0.9942110180854797, + "lr": 0.00046980066445182725, + "step": 16460, + "tokens_trained": 53.921350368 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 4.6711580738954686, + "grad_norm": 0.1455739438533783, + "loss": 0.9293, + "loss_ce": 1.026310920715332, + "loss_region": 0.0, + "loss_total": 1.026310920715332, + "lr": 0.000469468438538206, + "step": 16470, + "tokens_trained": 53.954111656 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 4.673994752145238, + "grad_norm": 0.1151554137468338, + "loss": 0.9245, + "loss_ce": 0.9114605188369751, + "loss_region": 0.0, + "loss_total": 0.9114605188369751, + "lr": 0.0004691362126245847, + "step": 16480, + "tokens_trained": 53.986876296 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 4.676831430395008, + "grad_norm": 0.17493972182273865, + "loss": 0.9163, + "loss_ce": 0.9554585218429565, + "loss_region": 0.0, + "loss_total": 0.9554585218429565, + "lr": 0.0004688039867109635, + "step": 16490, + "tokens_trained": 54.019641736 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 4.679668108644777, + "grad_norm": 0.12935680150985718, + "loss": 0.9355, + "loss_ce": 0.9278181195259094, + "loss_region": 0.0, + "loss_total": 0.9278181195259094, + "lr": 0.0004684717607973422, + "step": 16500, + "tokens_trained": 54.052407176 + }, + { + "epoch": 4.679668108644777, + "eval_ppl": 2.562874718129587, + "eval_runtime": 1.4152, + "step": 16500, + "tokens_trained": 54.052407176 + }, + { + "epoch": 4.679668108644777, + "eval_ppl_cds": 2.982272209055691, + "eval_ppl_dig": 1.0945610697612413, + "eval_ppl_exon": 2.9432895001775408, + "eval_ppl_intron": 2.7540889519282343, + "eval_ppl_nig": 2.4986607377205305, + "eval_ppl_promoter": 2.444851121615766, + "eval_ppl_utr": 2.8120986954873284, + "step": 16500, + "tokens_trained": 54.052407176 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 4.682504786894547, + "grad_norm": 0.16720305383205414, + "loss": 0.9356, + "loss_ce": 0.7489759922027588, + "loss_region": 0.0, + "loss_total": 0.7489759922027588, + "lr": 0.000468139534883721, + "step": 16510, + "tokens_trained": 54.085172616 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 4.685341465144316, + "grad_norm": 0.18737174570560455, + "loss": 0.9304, + "loss_ce": 0.9273119568824768, + "loss_region": 0.0, + "loss_total": 0.9273119568824768, + "lr": 0.00046780730897009966, + "step": 16520, + "tokens_trained": 54.117938056 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 4.688178143394086, + "grad_norm": 0.1397438794374466, + "loss": 0.9158, + "loss_ce": 0.9287741780281067, + "loss_region": 0.0, + "loss_total": 0.9287741780281067, + "lr": 0.0004674750830564784, + "step": 16530, + "tokens_trained": 54.150703496 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 4.691014821643855, + "grad_norm": 0.10008392482995987, + "loss": 0.9212, + "loss_ce": 1.0415300130844116, + "loss_region": 0.0, + "loss_total": 1.0415300130844116, + "lr": 0.00046714285714285716, + "step": 16540, + "tokens_trained": 54.183468936 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 4.693851499893625, + "grad_norm": 0.07797539234161377, + "loss": 0.9327, + "loss_ce": 0.9918743371963501, + "loss_region": 0.0, + "loss_total": 0.9918743371963501, + "lr": 0.0004668106312292359, + "step": 16550, + "tokens_trained": 54.216233576 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 4.696688178143394, + "grad_norm": 0.17613288760185242, + "loss": 0.9245, + "loss_ce": 1.0036059617996216, + "loss_region": 0.0, + "loss_total": 1.0036059617996216, + "lr": 0.00046647840531561466, + "step": 16560, + "tokens_trained": 54.248999016 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 4.699524856393164, + "grad_norm": 0.13874471187591553, + "loss": 0.9256, + "loss_ce": 0.9179160594940186, + "loss_region": 0.0, + "loss_total": 0.9179160594940186, + "lr": 0.0004661461794019934, + "step": 16570, + "tokens_trained": 54.281764456 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 4.702361534642933, + "grad_norm": 0.08599690347909927, + "loss": 0.9284, + "loss_ce": 1.0132784843444824, + "loss_region": 0.0, + "loss_total": 1.0132784843444824, + "lr": 0.00046581395348837206, + "step": 16580, + "tokens_trained": 54.314524944 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 4.705198212892703, + "grad_norm": 0.14794325828552246, + "loss": 0.9244, + "loss_ce": 0.940870463848114, + "loss_region": 0.0, + "loss_total": 0.940870463848114, + "lr": 0.00046548172757475084, + "step": 16590, + "tokens_trained": 54.347289584 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 4.708034891142472, + "grad_norm": 0.2000809758901596, + "loss": 0.9267, + "loss_ce": 0.9168792963027954, + "loss_region": 0.0, + "loss_total": 0.9168792963027954, + "lr": 0.00046514950166112956, + "step": 16600, + "tokens_trained": 54.380055024 + }, + { + "epoch": 4.708034891142472, + "eval_ppl": 2.5655721844743353, + "eval_runtime": 1.4158, + "step": 16600, + "tokens_trained": 54.380055024 + }, + { + "epoch": 4.708034891142472, + "eval_ppl_cds": 2.988831674586221, + "eval_ppl_dig": 1.0962969868269614, + "eval_ppl_exon": 2.987989905612387, + "eval_ppl_intron": 2.7531774031715623, + "eval_ppl_nig": 2.491392611083825, + "eval_ppl_promoter": 2.4306593458203065, + "eval_ppl_utr": 2.7661131233526253, + "step": 16600, + "tokens_trained": 54.380055024 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 4.710871569392242, + "grad_norm": 0.11255978792905807, + "loss": 0.9279, + "loss_ce": 0.6783397197723389, + "loss_region": 0.0, + "loss_total": 0.6783397197723389, + "lr": 0.00046481727574750834, + "step": 16610, + "tokens_trained": 54.412820464 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 4.713708247642011, + "grad_norm": 0.1477881819009781, + "loss": 0.9212, + "loss_ce": 0.9619972109794617, + "loss_region": 0.0, + "loss_total": 0.9619972109794617, + "lr": 0.00046448504983388707, + "step": 16620, + "tokens_trained": 54.445584304 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 4.716544925891781, + "grad_norm": 0.14732776582241058, + "loss": 0.928, + "loss_ce": 0.7544255256652832, + "loss_region": 0.0, + "loss_total": 0.7544255256652832, + "lr": 0.0004641528239202658, + "step": 16630, + "tokens_trained": 54.478349744 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 4.71938160414155, + "grad_norm": 0.12068971991539001, + "loss": 0.9178, + "loss_ce": 0.8668636083602905, + "loss_region": 0.0, + "loss_total": 0.8668636083602905, + "lr": 0.0004638205980066445, + "step": 16640, + "tokens_trained": 54.511115184 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 4.72221828239132, + "grad_norm": 0.14151553809642792, + "loss": 0.9317, + "loss_ce": 0.8737510442733765, + "loss_region": 0.0, + "loss_total": 0.8737510442733765, + "lr": 0.00046348837209302324, + "step": 16650, + "tokens_trained": 54.543875696 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 4.725054960641089, + "grad_norm": 0.08409151434898376, + "loss": 0.9266, + "loss_ce": 0.9186882376670837, + "loss_region": 0.0, + "loss_total": 0.9186882376670837, + "lr": 0.000463156146179402, + "step": 16660, + "tokens_trained": 54.576641136 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 4.727891638890859, + "grad_norm": 0.1738700568675995, + "loss": 0.9257, + "loss_ce": 0.8904275298118591, + "loss_region": 0.0, + "loss_total": 0.8904275298118591, + "lr": 0.00046282392026578074, + "step": 16670, + "tokens_trained": 54.609406576 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 4.7307283171406285, + "grad_norm": 0.1327747404575348, + "loss": 0.9194, + "loss_ce": 0.888667106628418, + "loss_region": 0.0, + "loss_total": 0.888667106628418, + "lr": 0.00046249169435215947, + "step": 16680, + "tokens_trained": 54.642171216 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 4.733564995390398, + "grad_norm": 0.15241555869579315, + "loss": 0.9277, + "loss_ce": 0.9113573431968689, + "loss_region": 0.0, + "loss_total": 0.9113573431968689, + "lr": 0.00046215946843853825, + "step": 16690, + "tokens_trained": 54.674933096 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 4.7364016736401675, + "grad_norm": 0.10262667387723923, + "loss": 0.9234, + "loss_ce": 0.887509822845459, + "loss_region": 0.0, + "loss_total": 0.887509822845459, + "lr": 0.000461827242524917, + "step": 16700, + "tokens_trained": 54.707697344 + }, + { + "epoch": 4.7364016736401675, + "eval_ppl": 2.5612630041319444, + "eval_runtime": 1.4337, + "step": 16700, + "tokens_trained": 54.707697344 + }, + { + "epoch": 4.7364016736401675, + "eval_ppl_cds": 3.0003969769632226, + "eval_ppl_dig": 1.0972838421912197, + "eval_ppl_exon": 2.9452678754755888, + "eval_ppl_intron": 2.748108328848305, + "eval_ppl_nig": 2.4962009081294023, + "eval_ppl_promoter": 2.413485429246324, + "eval_ppl_utr": 2.7340122098020396, + "step": 16700, + "tokens_trained": 54.707697344 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 4.739238351889937, + "grad_norm": 0.24678035080432892, + "loss": 0.9271, + "loss_ce": 0.8693206906318665, + "loss_region": 0.0, + "loss_total": 0.8693206906318665, + "lr": 0.00046149501661129564, + "step": 16710, + "tokens_trained": 54.740459456 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 4.7420750301397065, + "grad_norm": 0.14169371128082275, + "loss": 0.9191, + "loss_ce": 0.6505361795425415, + "loss_region": 0.0, + "loss_total": 0.6505361795425415, + "lr": 0.0004611627906976744, + "step": 16720, + "tokens_trained": 54.773224896 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 4.744911708389476, + "grad_norm": 0.1344275176525116, + "loss": 0.9183, + "loss_ce": 0.8193193078041077, + "loss_region": 0.0, + "loss_total": 0.8193193078041077, + "lr": 0.00046083056478405315, + "step": 16730, + "tokens_trained": 54.805990336 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 4.747748386639246, + "grad_norm": 0.12071927636861801, + "loss": 0.9116, + "loss_ce": 1.0378565788269043, + "loss_region": 0.0, + "loss_total": 1.0378565788269043, + "lr": 0.0004604983388704319, + "step": 16740, + "tokens_trained": 54.838755776 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 4.750585064889015, + "grad_norm": 0.1157970279455185, + "loss": 0.9294, + "loss_ce": 0.9684221148490906, + "loss_region": 0.0, + "loss_total": 0.9684221148490906, + "lr": 0.00046016611295681065, + "step": 16750, + "tokens_trained": 54.871521216 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 4.753421743138785, + "grad_norm": 0.11791826039552689, + "loss": 0.9207, + "loss_ce": 1.1009392738342285, + "loss_region": 0.0, + "loss_total": 1.1009392738342285, + "lr": 0.0004598338870431894, + "step": 16760, + "tokens_trained": 54.904286656 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 4.756258421388554, + "grad_norm": 0.14532431960105896, + "loss": 0.9344, + "loss_ce": 0.8781588077545166, + "loss_region": 0.0, + "loss_total": 0.8781588077545166, + "lr": 0.0004595016611295681, + "step": 16770, + "tokens_trained": 54.937051296 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 4.759095099638324, + "grad_norm": 0.0992966741323471, + "loss": 0.9322, + "loss_ce": 0.7950173020362854, + "loss_region": 0.0, + "loss_total": 0.7950173020362854, + "lr": 0.0004591694352159468, + "step": 16780, + "tokens_trained": 54.969816736 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 4.761931777888093, + "grad_norm": 0.13158787786960602, + "loss": 0.928, + "loss_ce": 1.0366183519363403, + "loss_region": 0.0, + "loss_total": 1.0366183519363403, + "lr": 0.0004588372093023256, + "step": 16790, + "tokens_trained": 55.002582176 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 4.764768456137863, + "grad_norm": 0.10292531549930573, + "loss": 0.9159, + "loss_ce": 0.8145430088043213, + "loss_region": 0.0, + "loss_total": 0.8145430088043213, + "lr": 0.00045850498338870433, + "step": 16800, + "tokens_trained": 55.035347616 + }, + { + "epoch": 4.764768456137863, + "eval_ppl": 2.560284620117812, + "eval_runtime": 1.4186, + "step": 16800, + "tokens_trained": 55.035347616 + }, + { + "epoch": 4.764768456137863, + "eval_ppl_cds": 3.00534931286103, + "eval_ppl_dig": 1.0910893501096273, + "eval_ppl_exon": 2.9673971990885146, + "eval_ppl_intron": 2.7461516405723225, + "eval_ppl_nig": 2.4922480972467898, + "eval_ppl_promoter": 2.4155004906663655, + "eval_ppl_utr": 2.7645493315741585, + "step": 16800, + "tokens_trained": 55.035347616 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 4.767605134387632, + "grad_norm": 0.14795951545238495, + "loss": 0.9149, + "loss_ce": 1.0385814905166626, + "loss_region": 0.0, + "loss_total": 1.0385814905166626, + "lr": 0.00045817275747508305, + "step": 16810, + "tokens_trained": 55.068112256 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 4.770441812637402, + "grad_norm": 0.08617939054965973, + "loss": 0.9186, + "loss_ce": 0.8196008205413818, + "loss_region": 0.0, + "loss_total": 0.8196008205413818, + "lr": 0.00045784053156146183, + "step": 16820, + "tokens_trained": 55.100873096 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 4.773278490887171, + "grad_norm": 0.15386250615119934, + "loss": 0.9231, + "loss_ce": 1.032848596572876, + "loss_region": 0.0, + "loss_total": 1.032848596572876, + "lr": 0.0004575083056478405, + "step": 16830, + "tokens_trained": 55.133638536 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 4.776115169136941, + "grad_norm": 0.17445272207260132, + "loss": 0.9263, + "loss_ce": 1.0597662925720215, + "loss_region": 0.0, + "loss_total": 1.0597662925720215, + "lr": 0.0004571760797342193, + "step": 16840, + "tokens_trained": 55.166403976 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 4.77895184738671, + "grad_norm": 0.20155097544193268, + "loss": 0.9211, + "loss_ce": 0.8742157220840454, + "loss_region": 0.0, + "loss_total": 0.8742157220840454, + "lr": 0.000456843853820598, + "step": 16850, + "tokens_trained": 55.199163296 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 4.78178852563648, + "grad_norm": 0.12918783724308014, + "loss": 0.9347, + "loss_ce": 0.9175214767456055, + "loss_region": 0.0, + "loss_total": 0.9175214767456055, + "lr": 0.00045651162790697673, + "step": 16860, + "tokens_trained": 55.231927936 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 4.784625203886249, + "grad_norm": 0.11244601011276245, + "loss": 0.9275, + "loss_ce": 0.8316949009895325, + "loss_region": 0.0, + "loss_total": 0.8316949009895325, + "lr": 0.0004561794019933555, + "step": 16870, + "tokens_trained": 55.264693376 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 4.787461882136019, + "grad_norm": 0.18249911069869995, + "loss": 0.9237, + "loss_ce": 1.0908030271530151, + "loss_region": 0.0, + "loss_total": 1.0908030271530151, + "lr": 0.00045584717607973424, + "step": 16880, + "tokens_trained": 55.297458768 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 4.790298560385788, + "grad_norm": 0.1671723872423172, + "loss": 0.9253, + "loss_ce": 0.9975810647010803, + "loss_region": 0.0, + "loss_total": 0.9975810647010803, + "lr": 0.00045551495016611296, + "step": 16890, + "tokens_trained": 55.330224208 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 4.793135238635558, + "grad_norm": 0.2093230038881302, + "loss": 0.9189, + "loss_ce": 1.0263930559158325, + "loss_region": 0.0, + "loss_total": 1.0263930559158325, + "lr": 0.0004551827242524917, + "step": 16900, + "tokens_trained": 55.362989648 + }, + { + "epoch": 4.793135238635558, + "eval_ppl": 2.5602008414229993, + "eval_runtime": 1.4262, + "step": 16900, + "tokens_trained": 55.362989648 + }, + { + "epoch": 4.793135238635558, + "eval_ppl_cds": 2.983541390956707, + "eval_ppl_dig": 1.0941681400322212, + "eval_ppl_exon": 2.9987409779556673, + "eval_ppl_intron": 2.745652471227272, + "eval_ppl_nig": 2.4949824466624406, + "eval_ppl_promoter": 2.416334110891088, + "eval_ppl_utr": 2.7824931945340956, + "step": 16900, + "tokens_trained": 55.362989648 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 4.795971916885327, + "grad_norm": 0.1462746113538742, + "loss": 0.929, + "loss_ce": 1.0181397199630737, + "loss_region": 0.0, + "loss_total": 1.0181397199630737, + "lr": 0.0004548504983388704, + "step": 16910, + "tokens_trained": 55.395755088 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 4.798808595135097, + "grad_norm": 0.12140128016471863, + "loss": 0.9393, + "loss_ce": 0.8151614665985107, + "loss_region": 0.0, + "loss_total": 0.8151614665985107, + "lr": 0.0004545182724252492, + "step": 16920, + "tokens_trained": 55.428520528 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 4.8016452733848665, + "grad_norm": 0.18710795044898987, + "loss": 0.9202, + "loss_ce": 0.7153810262680054, + "loss_region": 0.0, + "loss_total": 0.7153810262680054, + "lr": 0.0004541860465116279, + "step": 16930, + "tokens_trained": 55.461285968 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 4.804481951634636, + "grad_norm": 0.1329832375049591, + "loss": 0.9308, + "loss_ce": 0.8823424577713013, + "loss_region": 0.0, + "loss_total": 0.8823424577713013, + "lr": 0.0004538538205980067, + "step": 16940, + "tokens_trained": 55.494051408 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 4.8073186298844055, + "grad_norm": 0.14468584954738617, + "loss": 0.9209, + "loss_ce": 0.9444326162338257, + "loss_region": 0.0, + "loss_total": 0.9444326162338257, + "lr": 0.0004535215946843854, + "step": 16950, + "tokens_trained": 55.526816664 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 4.810155308134175, + "grad_norm": 0.21904921531677246, + "loss": 0.9239, + "loss_ce": 0.7863251566886902, + "loss_region": 0.0, + "loss_total": 0.7863251566886902, + "lr": 0.0004531893687707641, + "step": 16960, + "tokens_trained": 55.559581304 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 4.8129919863839445, + "grad_norm": 0.15534481406211853, + "loss": 0.9162, + "loss_ce": 0.6812537908554077, + "loss_region": 0.0, + "loss_total": 0.6812537908554077, + "lr": 0.00045285714285714287, + "step": 16970, + "tokens_trained": 55.592345944 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 4.815828664633714, + "grad_norm": 0.24342893064022064, + "loss": 0.9215, + "loss_ce": 0.8051536679267883, + "loss_region": 0.0, + "loss_total": 0.8051536679267883, + "lr": 0.0004525249169435216, + "step": 16980, + "tokens_trained": 55.625111384 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 4.818665342883484, + "grad_norm": 0.12493342161178589, + "loss": 0.915, + "loss_ce": 0.9643319249153137, + "loss_region": 0.0, + "loss_total": 0.9643319249153137, + "lr": 0.00045219269102990037, + "step": 16990, + "tokens_trained": 55.657876824 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 4.821502021133253, + "grad_norm": 0.15463975071907043, + "loss": 0.9263, + "loss_ce": 0.9251692295074463, + "loss_region": 0.0, + "loss_total": 0.9251692295074463, + "lr": 0.0004518604651162791, + "step": 17000, + "tokens_trained": 55.690642264 + }, + { + "epoch": 4.821502021133253, + "eval_ppl": 2.5543542544695534, + "eval_runtime": 1.4172, + "step": 17000, + "tokens_trained": 55.690642264 + }, + { + "epoch": 4.821502021133253, + "eval_ppl_cds": 2.966744542363627, + "eval_ppl_dig": 1.0945298565098078, + "eval_ppl_exon": 2.9951709315607458, + "eval_ppl_intron": 2.7447188122299764, + "eval_ppl_nig": 2.4872695750505946, + "eval_ppl_promoter": 2.41737183930774, + "eval_ppl_utr": 2.668047172624076, + "step": 17000, + "tokens_trained": 55.690642264 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 4.824338699383023, + "grad_norm": 0.08117688447237015, + "loss": 0.9297, + "loss_ce": 0.9578638076782227, + "loss_region": 0.0, + "loss_total": 0.9578638076782227, + "lr": 0.0004515282392026578, + "step": 17010, + "tokens_trained": 55.723404032 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 4.827175377632792, + "grad_norm": 0.1341707706451416, + "loss": 0.9221, + "loss_ce": 1.0973695516586304, + "loss_region": 0.0, + "loss_total": 1.0973695516586304, + "lr": 0.00045119601328903655, + "step": 17020, + "tokens_trained": 55.756168672 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 4.830012055882562, + "grad_norm": 0.1514272540807724, + "loss": 0.924, + "loss_ce": 0.9449554681777954, + "loss_region": 0.0, + "loss_total": 0.9449554681777954, + "lr": 0.00045086378737541527, + "step": 17030, + "tokens_trained": 55.788934112 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 4.832848734132331, + "grad_norm": 0.15431512892246246, + "loss": 0.9072, + "loss_ce": 0.837026834487915, + "loss_region": 0.0, + "loss_total": 0.837026834487915, + "lr": 0.00045053156146179405, + "step": 17040, + "tokens_trained": 55.821699552 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 4.835685412382101, + "grad_norm": 0.1585170030593872, + "loss": 0.9238, + "loss_ce": 0.9261703491210938, + "loss_region": 0.0, + "loss_total": 0.9261703491210938, + "lr": 0.0004501993355481728, + "step": 17050, + "tokens_trained": 55.854464992 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 4.83852209063187, + "grad_norm": 0.12394976615905762, + "loss": 0.9274, + "loss_ce": 0.9132611751556396, + "loss_region": 0.0, + "loss_total": 0.9132611751556396, + "lr": 0.0004498671096345515, + "step": 17060, + "tokens_trained": 55.887229632 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 4.84135876888164, + "grad_norm": 0.14210300147533417, + "loss": 0.9177, + "loss_ce": 1.0789587497711182, + "loss_region": 0.0, + "loss_total": 1.0789587497711182, + "lr": 0.0004495348837209303, + "step": 17070, + "tokens_trained": 55.919995072 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 4.844195447131409, + "grad_norm": 0.10092613101005554, + "loss": 0.9239, + "loss_ce": 0.9087733626365662, + "loss_region": 0.0, + "loss_total": 0.9087733626365662, + "lr": 0.00044920265780730895, + "step": 17080, + "tokens_trained": 55.952760128 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 4.847032125381179, + "grad_norm": 0.1452142894268036, + "loss": 0.9258, + "loss_ce": 0.9867022633552551, + "loss_region": 0.0, + "loss_total": 0.9867022633552551, + "lr": 0.00044887043189368773, + "step": 17090, + "tokens_trained": 55.985525088 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 4.849868803630948, + "grad_norm": 0.0988725796341896, + "loss": 0.9106, + "loss_ce": 0.8421356081962585, + "loss_region": 0.0, + "loss_total": 0.8421356081962585, + "lr": 0.00044853820598006645, + "step": 17100, + "tokens_trained": 56.018288752 + }, + { + "epoch": 4.849868803630948, + "eval_ppl": 2.5533748569133716, + "eval_runtime": 1.4189, + "step": 17100, + "tokens_trained": 56.018288752 + }, + { + "epoch": 4.849868803630948, + "eval_ppl_cds": 2.991375050473792, + "eval_ppl_dig": 1.096464597574787, + "eval_ppl_exon": 3.012026281337517, + "eval_ppl_intron": 2.7517499392709506, + "eval_ppl_nig": 2.488675039835682, + "eval_ppl_promoter": 2.4388968274747116, + "eval_ppl_utr": 2.6691359493900473, + "step": 17100, + "tokens_trained": 56.018288752 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 4.852705481880718, + "grad_norm": 0.17017412185668945, + "loss": 0.9249, + "loss_ce": 1.0002654790878296, + "loss_region": 0.0, + "loss_total": 1.0002654790878296, + "lr": 0.0004482059800664452, + "step": 17110, + "tokens_trained": 56.051054192 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 4.855542160130487, + "grad_norm": 0.16871517896652222, + "loss": 0.9348, + "loss_ce": 0.8369864821434021, + "loss_region": 0.0, + "loss_total": 0.8369864821434021, + "lr": 0.00044787375415282396, + "step": 17120, + "tokens_trained": 56.083819632 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 4.858378838380257, + "grad_norm": 0.10200805217027664, + "loss": 0.9229, + "loss_ce": 0.8621363639831543, + "loss_region": 0.0, + "loss_total": 0.8621363639831543, + "lr": 0.0004475415282392027, + "step": 17130, + "tokens_trained": 56.116585072 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 4.861215516630026, + "grad_norm": 0.16511406004428864, + "loss": 0.9234, + "loss_ce": 0.9281312823295593, + "loss_region": 0.0, + "loss_total": 0.9281312823295593, + "lr": 0.0004472093023255814, + "step": 17140, + "tokens_trained": 56.149350488 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 4.864052194879796, + "grad_norm": 0.13002271950244904, + "loss": 0.9314, + "loss_ce": 0.9951198697090149, + "loss_region": 0.0, + "loss_total": 0.9951198697090149, + "lr": 0.00044687707641196013, + "step": 17150, + "tokens_trained": 56.182115928 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 4.866888873129565, + "grad_norm": 0.16862201690673828, + "loss": 0.9235, + "loss_ce": 0.9303140640258789, + "loss_region": 0.0, + "loss_total": 0.9303140640258789, + "lr": 0.00044654485049833886, + "step": 17160, + "tokens_trained": 56.214881368 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 4.869725551379335, + "grad_norm": 0.11526070535182953, + "loss": 0.915, + "loss_ce": 1.0060145854949951, + "loss_region": 0.0, + "loss_total": 1.0060145854949951, + "lr": 0.00044621262458471763, + "step": 17170, + "tokens_trained": 56.247646808 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 4.8725622296291045, + "grad_norm": 0.13752709329128265, + "loss": 0.9251, + "loss_ce": 0.9934242963790894, + "loss_region": 0.0, + "loss_total": 0.9934242963790894, + "lr": 0.00044588039867109636, + "step": 17180, + "tokens_trained": 56.28041144 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 4.875398907878874, + "grad_norm": 0.15185503661632538, + "loss": 0.9062, + "loss_ce": 0.8598689436912537, + "loss_region": 0.0, + "loss_total": 0.8598689436912537, + "lr": 0.00044554817275747514, + "step": 17190, + "tokens_trained": 56.31317688 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 4.8782355861286435, + "grad_norm": 0.10225552320480347, + "loss": 0.9223, + "loss_ce": 0.9809095859527588, + "loss_region": 0.0, + "loss_total": 0.9809095859527588, + "lr": 0.0004452159468438538, + "step": 17200, + "tokens_trained": 56.34594152 + }, + { + "epoch": 4.8782355861286435, + "eval_ppl": 2.5519262384291057, + "eval_runtime": 1.4218, + "step": 17200, + "tokens_trained": 56.34594152 + }, + { + "epoch": 4.8782355861286435, + "eval_ppl_cds": 2.9941382920868156, + "eval_ppl_dig": 1.0977670910797026, + "eval_ppl_exon": 2.9177052740359213, + "eval_ppl_intron": 2.7525471159109522, + "eval_ppl_nig": 2.490268647074542, + "eval_ppl_promoter": 2.4247875878259775, + "eval_ppl_utr": 2.6852051205417125, + "step": 17200, + "tokens_trained": 56.34594152 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 4.881072264378413, + "grad_norm": 0.081932932138443, + "loss": 0.9104, + "loss_ce": 1.016195297241211, + "loss_region": 0.0, + "loss_total": 1.016195297241211, + "lr": 0.00044488372093023253, + "step": 17210, + "tokens_trained": 56.378706216 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 4.8839089426281825, + "grad_norm": 0.1594044715166092, + "loss": 0.9228, + "loss_ce": 1.068606972694397, + "loss_region": 0.0, + "loss_total": 1.068606972694397, + "lr": 0.0004445514950166113, + "step": 17220, + "tokens_trained": 56.411471656 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 4.886745620877952, + "grad_norm": 0.07568998634815216, + "loss": 0.9255, + "loss_ce": 1.0680323839187622, + "loss_region": 0.0, + "loss_total": 1.0680323839187622, + "lr": 0.00044421926910299004, + "step": 17230, + "tokens_trained": 56.444237096 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 4.889582299127722, + "grad_norm": 0.10386646538972855, + "loss": 0.921, + "loss_ce": 0.8561453819274902, + "loss_region": 0.0, + "loss_total": 0.8561453819274902, + "lr": 0.00044388704318936876, + "step": 17240, + "tokens_trained": 56.477002536 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 4.892418977377491, + "grad_norm": 0.14083078503608704, + "loss": 0.917, + "loss_ce": 1.0136077404022217, + "loss_region": 0.0, + "loss_total": 1.0136077404022217, + "lr": 0.00044355481727574754, + "step": 17250, + "tokens_trained": 56.509767976 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 4.895255655627261, + "grad_norm": 0.11162248253822327, + "loss": 0.9214, + "loss_ce": 0.7712157964706421, + "loss_region": 0.0, + "loss_total": 0.7712157964706421, + "lr": 0.00044322259136212627, + "step": 17260, + "tokens_trained": 56.542533416 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 4.89809233387703, + "grad_norm": 0.0875403881072998, + "loss": 0.912, + "loss_ce": 0.8822762370109558, + "loss_region": 0.0, + "loss_total": 0.8822762370109558, + "lr": 0.000442890365448505, + "step": 17270, + "tokens_trained": 56.575296344 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 4.9009290121268, + "grad_norm": 0.16741853952407837, + "loss": 0.9145, + "loss_ce": 0.8550717234611511, + "loss_region": 0.0, + "loss_total": 0.8550717234611511, + "lr": 0.0004425581395348837, + "step": 17280, + "tokens_trained": 56.60806104 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 4.903765690376569, + "grad_norm": 0.11295992881059647, + "loss": 0.9217, + "loss_ce": 0.7185900807380676, + "loss_region": 0.0, + "loss_total": 0.7185900807380676, + "lr": 0.00044222591362126244, + "step": 17290, + "tokens_trained": 56.64082648 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 4.906602368626339, + "grad_norm": 0.16369926929473877, + "loss": 0.9277, + "loss_ce": 0.8218139410018921, + "loss_region": 0.0, + "loss_total": 0.8218139410018921, + "lr": 0.0004418936877076412, + "step": 17300, + "tokens_trained": 56.67359192 + }, + { + "epoch": 4.906602368626339, + "eval_ppl": 2.5559623827666273, + "eval_runtime": 1.4141, + "step": 17300, + "tokens_trained": 56.67359192 + }, + { + "epoch": 4.906602368626339, + "eval_ppl_cds": 2.9972472461532904, + "eval_ppl_dig": 1.0968119153801337, + "eval_ppl_exon": 2.960768517952172, + "eval_ppl_intron": 2.7497360442360366, + "eval_ppl_nig": 2.491241426517302, + "eval_ppl_promoter": 2.4272108094434026, + "eval_ppl_utr": 2.7206400351163227, + "step": 17300, + "tokens_trained": 56.67359192 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 4.909439046876108, + "grad_norm": 0.11913631856441498, + "loss": 0.9293, + "loss_ce": 0.7647722363471985, + "loss_region": 0.0, + "loss_total": 0.7647722363471985, + "lr": 0.00044156146179401994, + "step": 17310, + "tokens_trained": 56.70635736 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 4.912275725125878, + "grad_norm": 0.17421816289424896, + "loss": 0.9206, + "loss_ce": 1.0365341901779175, + "loss_region": 0.0, + "loss_total": 1.0365341901779175, + "lr": 0.0004412292358803987, + "step": 17320, + "tokens_trained": 56.739119576 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 4.915112403375647, + "grad_norm": 0.0745103731751442, + "loss": 0.9303, + "loss_ce": 0.9841645359992981, + "loss_region": 0.0, + "loss_total": 0.9841645359992981, + "lr": 0.0004408970099667774, + "step": 17330, + "tokens_trained": 56.771885016 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 4.917949081625417, + "grad_norm": 0.16869230568408966, + "loss": 0.9278, + "loss_ce": 0.7607373595237732, + "loss_region": 0.0, + "loss_total": 0.7607373595237732, + "lr": 0.0004405647840531561, + "step": 17340, + "tokens_trained": 56.80464964 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 4.920785759875186, + "grad_norm": 0.07334867119789124, + "loss": 0.9086, + "loss_ce": 0.9545066952705383, + "loss_region": 0.0, + "loss_total": 0.9545066952705383, + "lr": 0.0004402325581395349, + "step": 17350, + "tokens_trained": 56.83741508 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 4.923622438124956, + "grad_norm": 0.13579054176807404, + "loss": 0.9231, + "loss_ce": 0.7216113805770874, + "loss_region": 0.0, + "loss_total": 0.7216113805770874, + "lr": 0.0004399003322259136, + "step": 17360, + "tokens_trained": 56.87017972 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 4.926459116374725, + "grad_norm": 0.09995526075363159, + "loss": 0.9254, + "loss_ce": 0.991142988204956, + "loss_region": 0.0, + "loss_total": 0.991142988204956, + "lr": 0.0004395681063122924, + "step": 17370, + "tokens_trained": 56.90294356 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 4.929295794624495, + "grad_norm": 0.17563554644584656, + "loss": 0.9308, + "loss_ce": 0.9403387904167175, + "loss_region": 0.0, + "loss_total": 0.9403387904167175, + "lr": 0.0004392358803986711, + "step": 17380, + "tokens_trained": 56.935708256 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 4.932132472874264, + "grad_norm": 0.14683009684085846, + "loss": 0.9195, + "loss_ce": 1.0081899166107178, + "loss_region": 0.0, + "loss_total": 1.0081899166107178, + "lr": 0.0004389036544850498, + "step": 17390, + "tokens_trained": 56.968473696 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 4.934969151124034, + "grad_norm": 0.12380640208721161, + "loss": 0.9223, + "loss_ce": 0.8103255033493042, + "loss_region": 0.0, + "loss_total": 0.8103255033493042, + "lr": 0.0004385714285714286, + "step": 17400, + "tokens_trained": 57.001239136 + }, + { + "epoch": 4.934969151124034, + "eval_ppl": 2.5549994250218355, + "eval_runtime": 1.4175, + "step": 17400, + "tokens_trained": 57.001239136 + }, + { + "epoch": 4.934969151124034, + "eval_ppl_cds": 2.9858539389382415, + "eval_ppl_dig": 1.0956163609856806, + "eval_ppl_exon": 2.9286858658863193, + "eval_ppl_intron": 2.7510083939599332, + "eval_ppl_nig": 2.4918487717795714, + "eval_ppl_promoter": 2.419724396989759, + "eval_ppl_utr": 2.7218166774945343, + "step": 17400, + "tokens_trained": 57.001239136 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 4.937805829373803, + "grad_norm": 0.1066407859325409, + "loss": 0.9263, + "loss_ce": 0.9555703997612, + "loss_region": 0.0, + "loss_total": 0.9555703997612, + "lr": 0.0004382392026578073, + "step": 17410, + "tokens_trained": 57.034000768 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 4.940642507623573, + "grad_norm": 0.1391371637582779, + "loss": 0.9129, + "loss_ce": 0.8579071164131165, + "loss_region": 0.0, + "loss_total": 0.8579071164131165, + "lr": 0.0004379069767441861, + "step": 17420, + "tokens_trained": 57.066766208 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 4.9434791858733425, + "grad_norm": 0.10166085511445999, + "loss": 0.9136, + "loss_ce": 0.8815908432006836, + "loss_region": 0.0, + "loss_total": 0.8815908432006836, + "lr": 0.0004375747508305648, + "step": 17430, + "tokens_trained": 57.099531648 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 4.946315864123112, + "grad_norm": 0.12692035734653473, + "loss": 0.9326, + "loss_ce": 1.0164276361465454, + "loss_region": 0.0, + "loss_total": 1.0164276361465454, + "lr": 0.00043724252491694353, + "step": 17440, + "tokens_trained": 57.132297088 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 4.9491525423728815, + "grad_norm": 0.12383224815130234, + "loss": 0.9204, + "loss_ce": 1.0124059915542603, + "loss_region": 0.0, + "loss_total": 1.0124059915542603, + "lr": 0.00043691029900332225, + "step": 17450, + "tokens_trained": 57.165062528 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 4.951989220622651, + "grad_norm": 0.16749532520771027, + "loss": 0.9222, + "loss_ce": 0.8913025259971619, + "loss_region": 0.0, + "loss_total": 0.8913025259971619, + "lr": 0.000436578073089701, + "step": 17460, + "tokens_trained": 57.197827968 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 4.9548258988724205, + "grad_norm": 0.17603330314159393, + "loss": 0.9195, + "loss_ce": 0.9197914004325867, + "loss_region": 0.0, + "loss_total": 0.9197914004325867, + "lr": 0.00043624584717607976, + "step": 17470, + "tokens_trained": 57.230588992 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 4.95766257712219, + "grad_norm": 0.09146023541688919, + "loss": 0.9171, + "loss_ce": 0.8358765244483948, + "loss_region": 0.0, + "loss_total": 0.8358765244483948, + "lr": 0.0004359136212624585, + "step": 17480, + "tokens_trained": 57.263354424 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 4.96049925537196, + "grad_norm": 0.1036766991019249, + "loss": 0.9362, + "loss_ce": 0.982742190361023, + "loss_region": 0.0, + "loss_total": 0.982742190361023, + "lr": 0.0004355813953488372, + "step": 17490, + "tokens_trained": 57.296118088 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 4.963335933621729, + "grad_norm": 0.10255859792232513, + "loss": 0.9222, + "loss_ce": 0.9505459666252136, + "loss_region": 0.0, + "loss_total": 0.9505459666252136, + "lr": 0.000435249169435216, + "step": 17500, + "tokens_trained": 57.32888352 + }, + { + "epoch": 4.963335933621729, + "eval_ppl": 2.552856388003165, + "eval_runtime": 1.412, + "step": 17500, + "tokens_trained": 57.32888352 + }, + { + "epoch": 4.963335933621729, + "eval_ppl_cds": 2.993046867628731, + "eval_ppl_dig": 1.0968725898768712, + "eval_ppl_exon": 2.900238423158983, + "eval_ppl_intron": 2.7416014828842123, + "eval_ppl_nig": 2.493602432393268, + "eval_ppl_promoter": 2.4030043946862376, + "eval_ppl_utr": 2.741809632374564, + "step": 17500, + "tokens_trained": 57.32888352 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 4.966172611871499, + "grad_norm": 0.18555358052253723, + "loss": 0.9199, + "loss_ce": 1.0070841312408447, + "loss_region": 0.0, + "loss_total": 1.0070841312408447, + "lr": 0.00043491694352159466, + "step": 17510, + "tokens_trained": 57.36164896 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 4.969009290121268, + "grad_norm": 0.10226114094257355, + "loss": 0.9183, + "loss_ce": 0.7328879237174988, + "loss_region": 0.0, + "loss_total": 0.7328879237174988, + "lr": 0.00043458471760797344, + "step": 17520, + "tokens_trained": 57.39441328 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 4.971845968371038, + "grad_norm": 0.11692332476377487, + "loss": 0.9156, + "loss_ce": 0.9549495577812195, + "loss_region": 0.0, + "loss_total": 0.9549495577812195, + "lr": 0.00043425249169435216, + "step": 17530, + "tokens_trained": 57.42717872 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 4.974682646620807, + "grad_norm": 0.08691312372684479, + "loss": 0.9263, + "loss_ce": 1.0711389780044556, + "loss_region": 0.0, + "loss_total": 1.0711389780044556, + "lr": 0.0004339202657807309, + "step": 17540, + "tokens_trained": 57.45994416 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 4.977519324870577, + "grad_norm": 0.150173157453537, + "loss": 0.9267, + "loss_ce": 0.9745522737503052, + "loss_region": 0.0, + "loss_total": 0.9745522737503052, + "lr": 0.00043358803986710966, + "step": 17550, + "tokens_trained": 57.4927096 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 4.980356003120346, + "grad_norm": 0.10016833245754242, + "loss": 0.9288, + "loss_ce": 0.9500195980072021, + "loss_region": 0.0, + "loss_total": 0.9500195980072021, + "lr": 0.0004332558139534884, + "step": 17560, + "tokens_trained": 57.52547424 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 4.983192681370116, + "grad_norm": 0.10020744800567627, + "loss": 0.9156, + "loss_ce": 0.8748807311058044, + "loss_region": 0.0, + "loss_total": 0.8748807311058044, + "lr": 0.00043292358803986717, + "step": 17570, + "tokens_trained": 57.55823888 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 4.986029359619885, + "grad_norm": 0.16984371840953827, + "loss": 0.9221, + "loss_ce": 1.0221600532531738, + "loss_region": 0.0, + "loss_total": 1.0221600532531738, + "lr": 0.00043259136212624584, + "step": 17580, + "tokens_trained": 57.59100432 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 4.988866037869655, + "grad_norm": 0.1290518343448639, + "loss": 0.9218, + "loss_ce": 1.0232229232788086, + "loss_region": 0.0, + "loss_total": 1.0232229232788086, + "lr": 0.00043225913621262456, + "step": 17590, + "tokens_trained": 57.62376896 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 4.991702716119424, + "grad_norm": 0.10437120497226715, + "loss": 0.9156, + "loss_ce": 0.8477900624275208, + "loss_region": 0.0, + "loss_total": 0.8477900624275208, + "lr": 0.00043192691029900334, + "step": 17600, + "tokens_trained": 57.6565344 + }, + { + "epoch": 4.991702716119424, + "eval_ppl": 2.5482740356072044, + "eval_runtime": 1.4195, + "step": 17600, + "tokens_trained": 57.6565344 + }, + { + "epoch": 4.991702716119424, + "eval_ppl_cds": 3.00128055487882, + "eval_ppl_dig": 1.0969372432927735, + "eval_ppl_exon": 2.9482006582511158, + "eval_ppl_intron": 2.741570723529436, + "eval_ppl_nig": 2.496384764676252, + "eval_ppl_promoter": 2.40873448280811, + "eval_ppl_utr": 2.7549954362311095, + "step": 17600, + "tokens_trained": 57.6565344 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 4.994539394369194, + "grad_norm": 0.16440729796886444, + "loss": 0.9219, + "loss_ce": 0.9165452718734741, + "loss_region": 0.0, + "loss_total": 0.9165452718734741, + "lr": 0.00043159468438538207, + "step": 17610, + "tokens_trained": 57.68929984 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 4.997376072618963, + "grad_norm": 0.12417641282081604, + "loss": 0.9243, + "loss_ce": 0.9157454967498779, + "loss_region": 0.0, + "loss_total": 0.9157454967498779, + "lr": 0.00043126245847176085, + "step": 17620, + "tokens_trained": 57.72206448 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 5.0, + "grad_norm": 0.11461741477251053, + "loss": 0.921, + "loss_ce": 1.0491862297058105, + "loss_region": 0.0, + "loss_total": 1.0491862297058105, + "lr": 0.00043093023255813957, + "step": 17630, + "tokens_trained": 57.752370912 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 5.0028366782497695, + "grad_norm": 0.1601700335741043, + "loss": 0.923, + "loss_ce": 0.9205655455589294, + "loss_region": 0.0, + "loss_total": 0.9205655455589294, + "lr": 0.00043059800664451824, + "step": 17640, + "tokens_trained": 57.785135552 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 5.005673356499539, + "grad_norm": 0.0878598764538765, + "loss": 0.9253, + "loss_ce": 0.893688440322876, + "loss_region": 0.0, + "loss_total": 0.893688440322876, + "lr": 0.000430265780730897, + "step": 17650, + "tokens_trained": 57.817900704 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 5.008510034749309, + "grad_norm": 0.19085650146007538, + "loss": 0.9122, + "loss_ce": 0.9708080291748047, + "loss_region": 0.0, + "loss_total": 0.9708080291748047, + "lr": 0.00042993355481727575, + "step": 17660, + "tokens_trained": 57.850666144 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 5.011346712999078, + "grad_norm": 0.16817298531532288, + "loss": 0.9182, + "loss_ce": 0.9039763808250427, + "loss_region": 0.0, + "loss_total": 0.9039763808250427, + "lr": 0.0004296013289036545, + "step": 17670, + "tokens_trained": 57.883430784 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 5.014183391248848, + "grad_norm": 0.14587271213531494, + "loss": 0.917, + "loss_ce": 1.119781732559204, + "loss_region": 0.0, + "loss_total": 1.119781732559204, + "lr": 0.00042926910299003325, + "step": 17680, + "tokens_trained": 57.916195424 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 5.017020069498617, + "grad_norm": 0.14200352132320404, + "loss": 0.9081, + "loss_ce": 1.0582093000411987, + "loss_region": 0.0, + "loss_total": 1.0582093000411987, + "lr": 0.000428936877076412, + "step": 17690, + "tokens_trained": 57.948956888 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 5.019856747748387, + "grad_norm": 0.18842604756355286, + "loss": 0.9213, + "loss_ce": 0.9292584657669067, + "loss_region": 0.0, + "loss_total": 0.9292584657669067, + "lr": 0.0004286046511627907, + "step": 17700, + "tokens_trained": 57.981722328 + }, + { + "epoch": 5.019856747748387, + "eval_ppl": 2.552533520449594, + "eval_runtime": 1.434, + "step": 17700, + "tokens_trained": 57.981722328 + }, + { + "epoch": 5.019856747748387, + "eval_ppl_cds": 2.9967982481091497, + "eval_ppl_dig": 1.0977534593014642, + "eval_ppl_exon": 2.9857646647149703, + "eval_ppl_intron": 2.744189054240852, + "eval_ppl_nig": 2.4920916886486717, + "eval_ppl_promoter": 2.4142839472008317, + "eval_ppl_utr": 2.7290681780421036, + "step": 17700, + "tokens_trained": 57.981722328 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 5.022693425998156, + "grad_norm": 0.13920627534389496, + "loss": 0.9184, + "loss_ce": 1.0729138851165771, + "loss_region": 0.0, + "loss_total": 1.0729138851165771, + "lr": 0.0004282724252491694, + "step": 17710, + "tokens_trained": 58.014487768 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 5.025530104247926, + "grad_norm": 0.12319612503051758, + "loss": 0.9183, + "loss_ce": 0.8163343071937561, + "loss_region": 0.0, + "loss_total": 0.8163343071937561, + "lr": 0.00042794019933554815, + "step": 17720, + "tokens_trained": 58.047252408 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 5.028366782497695, + "grad_norm": 0.10207246989011765, + "loss": 0.9236, + "loss_ce": 0.8462365865707397, + "loss_region": 0.0, + "loss_total": 0.8462365865707397, + "lr": 0.00042760797342192693, + "step": 17730, + "tokens_trained": 58.080017848 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 5.031203460747465, + "grad_norm": 0.17635266482830048, + "loss": 0.9216, + "loss_ce": 0.8869784474372864, + "loss_region": 0.0, + "loss_total": 0.8869784474372864, + "lr": 0.00042727574750830565, + "step": 17740, + "tokens_trained": 58.112783288 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 5.034040138997234, + "grad_norm": 0.07229283452033997, + "loss": 0.9107, + "loss_ce": 1.0851471424102783, + "loss_region": 0.0, + "loss_total": 1.0851471424102783, + "lr": 0.00042694352159468443, + "step": 17750, + "tokens_trained": 58.145548728 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 5.036876817247004, + "grad_norm": 0.16474567353725433, + "loss": 0.9233, + "loss_ce": 0.9593318104743958, + "loss_region": 0.0, + "loss_total": 0.9593318104743958, + "lr": 0.0004266112956810631, + "step": 17760, + "tokens_trained": 58.178314168 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 5.039713495496773, + "grad_norm": 0.12190917879343033, + "loss": 0.9216, + "loss_ce": 1.1222518682479858, + "loss_region": 0.0, + "loss_total": 1.1222518682479858, + "lr": 0.0004262790697674418, + "step": 17770, + "tokens_trained": 58.211079608 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 5.042550173746543, + "grad_norm": 0.11338907480239868, + "loss": 0.9211, + "loss_ce": 1.090105414390564, + "loss_region": 0.0, + "loss_total": 1.090105414390564, + "lr": 0.0004259468438538206, + "step": 17780, + "tokens_trained": 58.243844248 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 5.045386851996312, + "grad_norm": 0.1324814110994339, + "loss": 0.915, + "loss_ce": 0.9279975295066833, + "loss_region": 0.0, + "loss_total": 0.9279975295066833, + "lr": 0.00042561461794019933, + "step": 17790, + "tokens_trained": 58.276609688 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 5.048223530246082, + "grad_norm": 0.12237859517335892, + "loss": 0.9036, + "loss_ce": 0.6891218423843384, + "loss_region": 0.0, + "loss_total": 0.6891218423843384, + "lr": 0.0004252823920265781, + "step": 17800, + "tokens_trained": 58.309375128 + }, + { + "epoch": 5.048223530246082, + "eval_ppl": 2.548590591878222, + "eval_runtime": 1.438, + "step": 17800, + "tokens_trained": 58.309375128 + }, + { + "epoch": 5.048223530246082, + "eval_ppl_cds": 2.9924767410421467, + "eval_ppl_dig": 1.0991033865709765, + "eval_ppl_exon": 2.987712974921064, + "eval_ppl_intron": 2.7387782288314444, + "eval_ppl_nig": 2.4898046466403816, + "eval_ppl_promoter": 2.4087348723947044, + "eval_ppl_utr": 2.7103691192608883, + "step": 17800, + "tokens_trained": 58.309375128 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 5.051060208495851, + "grad_norm": 0.15965960919857025, + "loss": 0.9222, + "loss_ce": 0.9903594851493835, + "loss_region": 0.0, + "loss_total": 0.9903594851493835, + "lr": 0.00042495016611295683, + "step": 17810, + "tokens_trained": 58.342140568 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 5.053896886745621, + "grad_norm": 0.13134394586086273, + "loss": 0.9193, + "loss_ce": 0.9138078689575195, + "loss_region": 0.0, + "loss_total": 0.9138078689575195, + "lr": 0.00042461794019933556, + "step": 17820, + "tokens_trained": 58.374905992 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 5.05673356499539, + "grad_norm": 0.15448595583438873, + "loss": 0.9133, + "loss_ce": 1.0023940801620483, + "loss_region": 0.0, + "loss_total": 1.0023940801620483, + "lr": 0.0004242857142857143, + "step": 17830, + "tokens_trained": 58.407671432 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 5.05957024324516, + "grad_norm": 0.15147879719734192, + "loss": 0.9103, + "loss_ce": 0.731934130191803, + "loss_region": 0.0, + "loss_total": 0.731934130191803, + "lr": 0.000423953488372093, + "step": 17840, + "tokens_trained": 58.440436872 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 5.062406921494929, + "grad_norm": 0.14379438757896423, + "loss": 0.9261, + "loss_ce": 1.0318130254745483, + "loss_region": 0.0, + "loss_total": 1.0318130254745483, + "lr": 0.0004236212624584718, + "step": 17850, + "tokens_trained": 58.473202312 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 5.065243599744699, + "grad_norm": 0.1465359330177307, + "loss": 0.9172, + "loss_ce": 0.704222559928894, + "loss_region": 0.0, + "loss_total": 0.704222559928894, + "lr": 0.0004232890365448505, + "step": 17860, + "tokens_trained": 58.505967752 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 5.0680802779944685, + "grad_norm": 0.1104859784245491, + "loss": 0.9124, + "loss_ce": 0.8247342109680176, + "loss_region": 0.0, + "loss_total": 0.8247342109680176, + "lr": 0.00042295681063122924, + "step": 17870, + "tokens_trained": 58.538733192 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 5.070916956244238, + "grad_norm": 0.20892663300037384, + "loss": 0.9169, + "loss_ce": 0.736333966255188, + "loss_region": 0.0, + "loss_total": 0.736333966255188, + "lr": 0.000422624584717608, + "step": 17880, + "tokens_trained": 58.571498632 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 5.0737536344940075, + "grad_norm": 0.13189846277236938, + "loss": 0.9114, + "loss_ce": 0.8729366660118103, + "loss_region": 0.0, + "loss_total": 0.8729366660118103, + "lr": 0.0004222923588039867, + "step": 17890, + "tokens_trained": 58.604264072 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 5.076590312743777, + "grad_norm": 0.10594058781862259, + "loss": 0.9189, + "loss_ce": 1.0024880170822144, + "loss_region": 0.0, + "loss_total": 1.0024880170822144, + "lr": 0.00042196013289036547, + "step": 17900, + "tokens_trained": 58.637029352 + }, + { + "epoch": 5.076590312743777, + "eval_ppl": 2.5466752983740184, + "eval_runtime": 1.4315, + "step": 17900, + "tokens_trained": 58.637029352 + }, + { + "epoch": 5.076590312743777, + "eval_ppl_cds": 3.000335534139723, + "eval_ppl_dig": 1.0981901267664276, + "eval_ppl_exon": 2.9302861509876, + "eval_ppl_intron": 2.7418163617686595, + "eval_ppl_nig": 2.4961325974730784, + "eval_ppl_promoter": 2.4141852227323093, + "eval_ppl_utr": 2.7282629800031786, + "step": 17900, + "tokens_trained": 58.637029352 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 5.0794269909935466, + "grad_norm": 0.06331738084554672, + "loss": 0.9205, + "loss_ce": 0.7057142853736877, + "loss_region": 0.0, + "loss_total": 0.7057142853736877, + "lr": 0.0004216279069767442, + "step": 17910, + "tokens_trained": 58.669794792 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 5.082263669243316, + "grad_norm": 0.12811894714832306, + "loss": 0.9243, + "loss_ce": 0.9566288590431213, + "loss_region": 0.0, + "loss_total": 0.9566288590431213, + "lr": 0.0004212956810631229, + "step": 17920, + "tokens_trained": 58.702558632 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 5.085100347493086, + "grad_norm": 0.10277832299470901, + "loss": 0.9143, + "loss_ce": 0.7163805961608887, + "loss_region": 0.0, + "loss_total": 0.7163805961608887, + "lr": 0.0004209634551495017, + "step": 17930, + "tokens_trained": 58.735324072 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 5.087937025742855, + "grad_norm": 0.1347387284040451, + "loss": 0.9232, + "loss_ce": 1.0738403797149658, + "loss_region": 0.0, + "loss_total": 1.0738403797149658, + "lr": 0.0004206312292358804, + "step": 17940, + "tokens_trained": 58.768089512 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 5.090773703992625, + "grad_norm": 0.09530395269393921, + "loss": 0.9191, + "loss_ce": 0.7366634607315063, + "loss_region": 0.0, + "loss_total": 0.7366634607315063, + "lr": 0.00042029900332225914, + "step": 17950, + "tokens_trained": 58.800854952 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 5.093610382242394, + "grad_norm": 0.12268706411123276, + "loss": 0.9164, + "loss_ce": 0.9690417647361755, + "loss_region": 0.0, + "loss_total": 0.9690417647361755, + "lr": 0.00041996677740863787, + "step": 17960, + "tokens_trained": 58.833619592 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 5.096447060492164, + "grad_norm": 0.0977795198559761, + "loss": 0.9256, + "loss_ce": 0.9048771858215332, + "loss_region": 0.0, + "loss_total": 0.9048771858215332, + "lr": 0.0004196345514950166, + "step": 17970, + "tokens_trained": 58.866385032 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 5.099283738741933, + "grad_norm": 0.13332830369472504, + "loss": 0.9145, + "loss_ce": 0.9937964081764221, + "loss_region": 0.0, + "loss_total": 0.9937964081764221, + "lr": 0.00041930232558139537, + "step": 17980, + "tokens_trained": 58.899149672 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 5.102120416991703, + "grad_norm": 0.12063479423522949, + "loss": 0.9171, + "loss_ce": 1.0338681936264038, + "loss_region": 0.0, + "loss_total": 1.0338681936264038, + "lr": 0.0004189700996677741, + "step": 17990, + "tokens_trained": 58.931915112 + }, + { + "comp/rl_weight": 0.03, + "comp/strictness": 0.0, + "epoch": 5.104957095241472, + "grad_norm": 0.17523616552352905, + "loss": 0.909, + "loss_ce": 0.7492321133613586, + "loss_region": 0.0, + "loss_total": 0.7492321133613586, + "lr": 0.0004186378737541529, + "step": 18000, + "tokens_trained": 58.964680552 + }, + { + "epoch": 5.104957095241472, + "eval_ppl": 2.546471447625592, + "eval_runtime": 1.4203, + "step": 18000, + "tokens_trained": 58.964680552 + }, + { + "epoch": 5.104957095241472, + "eval_ppl_cds": 2.989548984548807, + "eval_ppl_dig": 1.097845283381969, + "eval_ppl_exon": 2.9555996020895883, + "eval_ppl_intron": 2.740102599475998, + "eval_ppl_nig": 2.4895622581121977, + "eval_ppl_promoter": 2.399853200152211, + "eval_ppl_utr": 2.697212844887806, + "step": 18000, + "tokens_trained": 58.964680552 + } + ], + "logging_steps": 10, + "max_steps": 30600, + "num_input_tokens_seen": 0, + "num_train_epochs": 9, + "save_steps": 3000, + "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": 8, + "trial_name": null, + "trial_params": null +}