Upload folder using huggingface_hub

baseline/args.json

@@ -1 +1 @@
-{"hellaswag": true, "attention_kind": "selective", "log_dir": "logs/repro_selective_pattern_rankings/baseline", "resume_checkpoint": null, "resume_optimizer": false, "add_a_head": false, "add_head_to_start": true, "new_head_init": "normal", "n_heads": 4, "protect_bos_token": true, "prevent_from_masking_myself": true, "max_steps": 4375, "warmup_steps": 250, "group": "repro_selective_pattern_rankings", "use_wandb": true, "kill_self_after_run": false, "random_seed": 1339, "memory_penalty_epsilon": 0.1, "selection_head_linear_combo": "none", "selection_head_linear_combo_scale": 1.0, "protection_kind": "none", "leaky_relu_alpha": null, "leaky_relu_bias": null, "use_compile": true, "use_mini_model": false, "upload_to_hf": true, "seq_len": 256, "batch_size": 128, "total_batch_size": 131072, "protection_head_scaling_factor": 1.0, "protection_head_bias": 0.0, "n_sliced_masks": null, "n_latent_masks": null, "mask_layernorm": false, "residual_attention_masks": false, "compute_base_shapes": false, "base_shapes_savefile": null, "mup": true, "disable_selection": false, "mup_enable_coord_check_logging": false, "max_lr": 0.0011, "decay_lr": true, "readout_zero_init": false, "query_zero_init": false, "l1_loss": false, "debugpy": false, "key": "baseline", "n_embd": 256}
+{"hellaswag": true, "attention_kind": "selective", "log_dir": "allowing_more_selection_patterns/baseline", "resume_checkpoint": null, "resume_optimizer": false, "add_a_head": false, "add_head_to_start": true, "new_head_init": "normal", "n_heads": 12, "n_embd": 768, "head_dim": 64, "protect_bos_token": true, "prevent_from_masking_myself": true, "max_steps": null, "warmup_steps": null, "group": "allowing_more_selection_patterns", "use_wandb": true, "kill_self_after_run": false, "random_seed": 1337, "memory_penalty_epsilon": 0.1, "selection_head_linear_combo": "none", "selection_head_linear_combo_scale": 1.0, "protection_kind": "none", "leaky_relu_alpha": null, "leaky_relu_bias": null, "use_compile": true, "use_mini_model": false, "upload_to_hf": true, "seq_len": null, "batch_size": 4, "total_batch_size": null, "protection_head_scaling_factor": 1.0, "protection_head_bias": 0.0, "n_sliced_masks": null, "n_latent_masks": null, "mask_layernorm": false, "residual_attention_masks": false, "compute_base_shapes": false, "base_shapes_savefile": null, "mup": false, "disable_selection": false, "mup_enable_coord_check_logging": false, "max_lr": null, "decay_lr": true, "readout_zero_init": false, "query_zero_init": false, "l1_loss": false, "debugpy": false, "key": null}

baseline/log2.txt

@@ -1,529 +1,303 @@
-max_steps: 4375
-0 val loss 11.2574
-0 val perplexity 77450.9062
-0 train 11.263229 (lr=1.5385e-06) (hash(x)=161979327)
-10 train 10.274622 (lr=1.6923e-05) (hash(x)=165941070)
-20 train 9.825401 (lr=3.2308e-05) (hash(x)=152918033)
-30 train 9.548252 (lr=4.7692e-05) (hash(x)=169931607)
-40 train 9.087914 (lr=6.3077e-05) (hash(x)=174342675)
-50 train 8.611303 (lr=7.8462e-05) (hash(x)=163872554)
-60 train 8.261993 (lr=9.3846e-05) (hash(x)=161534908)
-70 train 7.767174 (lr=1.0923e-04) (hash(x)=156776712)
-80 train 7.681754 (lr=1.2462e-04) (hash(x)=164996929)
-90 train 7.553709 (lr=1.4000e-04) (hash(x)=152962169)
-100 val loss 7.4743
-100 val perplexity 1762.1432
-100 train 7.328951 (lr=1.5538e-04) (hash(x)=147111701)
-110 train 7.265400 (lr=1.7077e-04) (hash(x)=167595270)
-120 train 7.146955 (lr=1.8615e-04) (hash(x)=152205426)
-130 train 6.979732 (lr=2.0154e-04) (hash(x)=158427335)
-140 train 6.937882 (lr=2.1692e-04) (hash(x)=174588384)
-150 train 6.981928 (lr=2.3231e-04) (hash(x)=157067270)
-160 train 6.890918 (lr=2.4769e-04) (hash(x)=161859595)
-170 train 6.761150 (lr=2.6308e-04) (hash(x)=162739537)
-180 train 6.678267 (lr=2.7846e-04) (hash(x)=170785304)
-190 train 6.575134 (lr=2.9385e-04) (hash(x)=168658088)
-200 val loss 6.5186
-200 val perplexity 677.6003
-200 train 6.519105 (lr=3.0923e-04) (hash(x)=150749086)
-210 train 6.402273 (lr=3.2462e-04) (hash(x)=148789576)
-220 train 6.299723 (lr=3.4000e-04) (hash(x)=153164408)
-230 train 6.464761 (lr=3.5538e-04) (hash(x)=184594366)
-240 train 6.162462 (lr=3.7077e-04) (hash(x)=157817319)
-250 train 6.195898 (lr=3.8615e-04) (hash(x)=149451562)
-260 train 6.098128 (lr=4.0154e-04) (hash(x)=165738161)
-270 train 6.035585 (lr=4.1692e-04) (hash(x)=165724764)
-280 train 5.970767 (lr=4.3231e-04) (hash(x)=155414136)
-290 train 5.874425 (lr=4.4769e-04) (hash(x)=165090903)
-300 val loss 6.0340
-300 val perplexity 417.3607
-300 train 5.750409 (lr=4.6308e-04) (hash(x)=154445545)
-310 train 5.703163 (lr=4.7846e-04) (hash(x)=158949603)
-320 train 5.787862 (lr=4.9385e-04) (hash(x)=161412973)
-330 train 5.847661 (lr=5.0923e-04) (hash(x)=164572249)
-340 train 5.899548 (lr=5.2462e-04) (hash(x)=162202191)
-350 train 5.886857 (lr=5.4000e-04) (hash(x)=158135392)
-360 train 5.834284 (lr=5.5538e-04) (hash(x)=163752391)
-370 train 5.825579 (lr=5.7077e-04) (hash(x)=172758799)
-380 train 5.621739 (lr=5.8615e-04) (hash(x)=144569488)
-390 train 5.592220 (lr=6.0154e-04) (hash(x)=150429229)
-400 val loss 5.6536
-400 val perplexity 285.3069
-400 train 5.601220 (lr=6.1692e-04) (hash(x)=154261530)
-410 train 5.535435 (lr=6.3231e-04) (hash(x)=159172506)
-420 train 5.524165 (lr=6.4769e-04) (hash(x)=154419867)
-430 train 5.566013 (lr=6.6308e-04) (hash(x)=167337099)
-440 train 5.464113 (lr=6.7846e-04) (hash(x)=154535255)
-450 train 5.443037 (lr=6.9385e-04) (hash(x)=153077071)
-460 train 5.355242 (lr=7.0923e-04) (hash(x)=150853070)
-470 train 5.195047 (lr=7.2462e-04) (hash(x)=160701113)
-480 train 5.293690 (lr=7.4000e-04) (hash(x)=154690870)
-490 train 5.117364 (lr=7.5538e-04) (hash(x)=163834286)
-500 val loss 5.4535
-500 val perplexity 233.5737
-500 train 5.199679 (lr=7.7077e-04) (hash(x)=162651717)
-510 train 5.284278 (lr=7.8615e-04) (hash(x)=160448311)
-520 train 5.338402 (lr=8.0154e-04) (hash(x)=165741643)
-530 train 5.448352 (lr=8.1692e-04) (hash(x)=153201264)
-540 train 5.342171 (lr=8.3231e-04) (hash(x)=159027160)
-550 train 5.348803 (lr=8.4769e-04) (hash(x)=170769095)
-560 train 5.325680 (lr=8.6308e-04) (hash(x)=153606190)
-570 train 5.309877 (lr=8.7846e-04) (hash(x)=159708689)
-580 train 5.247071 (lr=8.9385e-04) (hash(x)=179108359)
-590 train 5.250077 (lr=9.0923e-04) (hash(x)=157134241)
-600 val loss 5.2541
-600 val perplexity 191.3425
-600 train 5.233368 (lr=9.2462e-04) (hash(x)=162178356)
-610 train 5.155111 (lr=9.4000e-04) (hash(x)=167646928)
-620 train 5.114224 (lr=9.5538e-04) (hash(x)=160979601)
-630 train 5.014015 (lr=9.7077e-04) (hash(x)=158349468)
-640 train 5.022974 (lr=9.8615e-04) (hash(x)=150370348)
-650 train 4.975046 (lr=1.0015e-03) (hash(x)=152810075)
-660 train 4.886074 (lr=1.0169e-03) (hash(x)=145786262)
-670 train 4.893470 (lr=1.0323e-03) (hash(x)=160779604)
-680 train 4.805331 (lr=1.0477e-03) (hash(x)=152343688)
-690 train 4.840962 (lr=1.0631e-03) (hash(x)=166507600)
-700 val loss 5.1230
-700 val perplexity 167.8409
-700 train 4.802319 (lr=1.0785e-03) (hash(x)=162623700)
-710 train 5.097156 (lr=1.0938e-03) (hash(x)=162293403)
-720 train 5.044741 (lr=1.1000e-03) (hash(x)=159564657)
-730 train 5.059063 (lr=1.1000e-03) (hash(x)=154209107)
-740 train 5.029888 (lr=1.0999e-03) (hash(x)=156485709)
-750 train 5.000640 (lr=1.0998e-03) (hash(x)=160834702)
-760 train 4.987276 (lr=1.0996e-03) (hash(x)=152240275)
-770 train 4.986840 (lr=1.0994e-03) (hash(x)=168903075)
-780 train 5.049429 (lr=1.0992e-03) (hash(x)=166843738)
-790 train 5.082653 (lr=1.0990e-03) (hash(x)=166358104)
-800 val loss 4.9533
-800 val perplexity 141.6453
-800 train 4.880929 (lr=1.0987e-03) (hash(x)=153477747)
-810 train 4.818970 (lr=1.0984e-03) (hash(x)=157954385)
-820 train 4.806408 (lr=1.0980e-03) (hash(x)=174383505)
-830 train 4.820668 (lr=1.0976e-03) (hash(x)=145226181)
-840 train 4.804605 (lr=1.0972e-03) (hash(x)=157665743)
-850 train 4.792170 (lr=1.0967e-03) (hash(x)=158565836)
-860 train 4.791689 (lr=1.0962e-03) (hash(x)=171168105)
-870 train 5.211392 (lr=1.0956e-03) (hash(x)=159017777)
-880 train 4.973852 (lr=1.0950e-03) (hash(x)=168156768)
-890 train 4.867979 (lr=1.0944e-03) (hash(x)=145645245)
-900 val loss 4.7977
-900 val perplexity 121.2296
-900 train 4.677175 (lr=1.0938e-03) (hash(x)=151245521)
-910 train 4.705009 (lr=1.0931e-03) (hash(x)=182614672)
-920 train 4.687078 (lr=1.0924e-03) (hash(x)=167377432)
-930 train 4.609860 (lr=1.0916e-03) (hash(x)=171347228)
-940 train 4.637469 (lr=1.0908e-03) (hash(x)=171588755)
-950 train 4.726392 (lr=1.0900e-03) (hash(x)=159560168)
-960 train 4.605119 (lr=1.0891e-03) (hash(x)=171801357)
-970 train 4.605552 (lr=1.0882e-03) (hash(x)=172513884)
-980 train 4.632417 (lr=1.0872e-03) (hash(x)=170659685)
-990 train 4.652870 (lr=1.0863e-03) (hash(x)=165436358)
-1000 val loss 4.6640
-1000 val perplexity 106.0595
-1000 train 4.717592 (lr=1.0853e-03) (hash(x)=167684690)
-1010 train 4.677588 (lr=1.0842e-03) (hash(x)=127225492)
-1020 train 4.681705 (lr=1.0831e-03) (hash(x)=167575323)
-1030 train 4.602217 (lr=1.0820e-03) (hash(x)=161266302)
-1040 train 4.581406 (lr=1.0809e-03) (hash(x)=155607661)
-1050 train 4.613191 (lr=1.0797e-03) (hash(x)=165685359)
-1060 train 4.468890 (lr=1.0785e-03) (hash(x)=160417824)
-1070 train 4.612445 (lr=1.0772e-03) (hash(x)=173313882)
-1080 train 4.580376 (lr=1.0759e-03) (hash(x)=165485047)
-1090 train 4.440218 (lr=1.0746e-03) (hash(x)=159848305)
-1100 val loss 4.5914
-1100 val perplexity 98.6331
-1100 train 4.613663 (lr=1.0732e-03) (hash(x)=170890601)
-1110 train 4.538614 (lr=1.0718e-03) (hash(x)=172689983)
-1120 train 4.569351 (lr=1.0704e-03) (hash(x)=154762859)
-1130 train 4.493426 (lr=1.0689e-03) (hash(x)=168196871)
-1140 train 4.514806 (lr=1.0674e-03) (hash(x)=156715577)
-1150 train 4.639327 (lr=1.0659e-03) (hash(x)=166381788)
-1160 train 4.554700 (lr=1.0643e-03) (hash(x)=162095130)
-1170 train 4.537933 (lr=1.0627e-03) (hash(x)=166094750)
-1180 train 4.471836 (lr=1.0611e-03) (hash(x)=154965753)
-1190 train 4.461709 (lr=1.0594e-03) (hash(x)=172766447)
-1200 val loss 4.5039
-1200 val perplexity 90.3710
-1200 train 4.382483 (lr=1.0577e-03) (hash(x)=161185334)
-1210 train 4.546267 (lr=1.0560e-03) (hash(x)=167224144)
-1220 train 4.384159 (lr=1.0542e-03) (hash(x)=167252866)
-1230 train 4.399038 (lr=1.0524e-03) (hash(x)=155856924)
-1240 train 4.396050 (lr=1.0506e-03) (hash(x)=155956446)
-1250 train 4.308051 (lr=1.0487e-03) (hash(x)=140794796)
-1260 train 4.419213 (lr=1.0468e-03) (hash(x)=166863869)
-1270 train 4.480976 (lr=1.0449e-03) (hash(x)=166021892)
-1280 train 4.487140 (lr=1.0429e-03) (hash(x)=170503648)
-1290 train 4.400294 (lr=1.0409e-03) (hash(x)=164012719)
-1300 val loss 4.4612
-1300 val perplexity 86.5942
-1300 train 4.469203 (lr=1.0389e-03) (hash(x)=156519834)
-1310 train 4.410668 (lr=1.0368e-03) (hash(x)=165146652)
-1320 train 4.442281 (lr=1.0347e-03) (hash(x)=174391990)
-1330 train 4.437904 (lr=1.0326e-03) (hash(x)=161131201)
-1340 train 4.422988 (lr=1.0305e-03) (hash(x)=162240912)
-1350 train 4.339061 (lr=1.0283e-03) (hash(x)=152462931)
-1360 train 4.343878 (lr=1.0261e-03) (hash(x)=150443278)
-1370 train 4.279283 (lr=1.0238e-03) (hash(x)=153721711)
-1380 train 4.420867 (lr=1.0215e-03) (hash(x)=158070628)
-1390 train 4.451275 (lr=1.0192e-03) (hash(x)=163318827)
-1400 val loss 4.4293
-1400 val perplexity 83.8754
-1400 train 4.350235 (lr=1.0169e-03) (hash(x)=154210336)
-1410 train 4.400532 (lr=1.0145e-03) (hash(x)=162393933)
-1420 train 4.404506 (lr=1.0121e-03) (hash(x)=156659308)
-1430 train 4.372114 (lr=1.0097e-03) (hash(x)=171243439)
-1440 train 4.510427 (lr=1.0072e-03) (hash(x)=178459521)
-1450 train 4.362426 (lr=1.0047e-03) (hash(x)=159166350)
-1460 train 4.288169 (lr=1.0022e-03) (hash(x)=173554569)
-1470 train 4.378874 (lr=9.9964e-04) (hash(x)=158362521)
-1480 train 4.346535 (lr=9.9706e-04) (hash(x)=165899207)
-1490 train 4.374932 (lr=9.9445e-04) (hash(x)=156470939)
-1500 val loss 4.3808
-1500 val perplexity 79.9019
-1500 train 4.342013 (lr=9.9182e-04) (hash(x)=164197780)
-1510 train 4.237641 (lr=9.8915e-04) (hash(x)=146709896)
-1520 train 4.252951 (lr=9.8646e-04) (hash(x)=151304838)
-1530 train 4.287971 (lr=9.8374e-04) (hash(x)=163187333)
-1540 train 4.240841 (lr=9.8099e-04) (hash(x)=153630537)
-1550 train 4.259836 (lr=9.7821e-04) (hash(x)=157299103)
-1560 train 4.170657 (lr=9.7540e-04) (hash(x)=153786477)
-1570 train 4.493762 (lr=9.7257e-04) (hash(x)=155037209)
-1580 train 4.368567 (lr=9.6971e-04) (hash(x)=146157308)
-1590 train 4.414457 (lr=9.6683e-04) (hash(x)=147823317)
-1600 val loss 4.3560
-1600 val perplexity 77.9452
-1600 train 4.400393 (lr=9.6391e-04) (hash(x)=169140159)
-1610 train 4.178614 (lr=9.6097e-04) (hash(x)=157811493)
-1620 train 4.270865 (lr=9.5801e-04) (hash(x)=168940851)
-1630 train 4.260715 (lr=9.5502e-04) (hash(x)=151149213)
-1640 train 4.360132 (lr=9.5200e-04) (hash(x)=161822009)
-1650 train 4.773558 (lr=9.4896e-04) (hash(x)=151056313)
-1660 train 4.279921 (lr=9.4589e-04) (hash(x)=156920306)
-1670 train 4.196922 (lr=9.4280e-04) (hash(x)=159695815)
-1680 train 4.208801 (lr=9.3968e-04) (hash(x)=160232907)
-1690 train 4.200097 (lr=9.3654e-04) (hash(x)=158099357)
-1700 val loss 4.3472
-1700 val perplexity 77.2584
-1700 train 4.250903 (lr=9.3337e-04) (hash(x)=164127381)
-1710 train 4.420138 (lr=9.3018e-04) (hash(x)=156592040)
-1720 train 4.346195 (lr=9.2696e-04) (hash(x)=160220265)
-1730 train 4.287699 (lr=9.2372e-04) (hash(x)=156162204)
-1740 train 4.280502 (lr=9.2046e-04) (hash(x)=152767893)
-1750 train 4.263510 (lr=9.1717e-04) (hash(x)=162707306)
-1760 train 4.284562 (lr=9.1386e-04) (hash(x)=154472004)
-1770 train 4.145760 (lr=9.1053e-04) (hash(x)=160374234)
-1780 train 4.306995 (lr=9.0718e-04) (hash(x)=159709225)
-1790 train 4.338753 (lr=9.0380e-04) (hash(x)=152095944)
-1800 val loss 4.3000
-1800 val perplexity 73.6977
-1800 train 4.308148 (lr=9.0040e-04) (hash(x)=157920255)
-1810 train 4.056387 (lr=8.9698e-04) (hash(x)=167237680)
-1820 train 4.168284 (lr=8.9354e-04) (hash(x)=161255774)
-1830 train 4.192098 (lr=8.9008e-04) (hash(x)=154275481)
-1840 train 4.139931 (lr=8.8660e-04) (hash(x)=152398178)
-1850 train 4.244508 (lr=8.8309e-04) (hash(x)=147646022)
-1860 train 4.268148 (lr=8.7957e-04) (hash(x)=158895559)
-1870 train 4.207963 (lr=8.7602e-04) (hash(x)=162742598)
-1880 train 4.209588 (lr=8.7245e-04) (hash(x)=169911029)
-1890 train 4.239878 (lr=8.6887e-04) (hash(x)=172068270)
-1900 val loss 4.3000
-1900 val perplexity 73.6996
-1900 train 4.331270 (lr=8.6527e-04) (hash(x)=159975457)
-1910 train 4.318810 (lr=8.6164e-04) (hash(x)=178775667)
-1920 train 4.191704 (lr=8.5800e-04) (hash(x)=151180831)
-1930 train 4.212752 (lr=8.5434e-04) (hash(x)=154239412)
-1940 train 4.222959 (lr=8.5066e-04) (hash(x)=139144465)
-1950 train 4.172968 (lr=8.4696e-04) (hash(x)=194482413)
-1960 train 4.190656 (lr=8.4324e-04) (hash(x)=155812344)
-1970 train 4.193388 (lr=8.3951e-04) (hash(x)=157577031)
-1980 train 4.229763 (lr=8.3576e-04) (hash(x)=152045027)
-1990 train 4.142585 (lr=8.3199e-04) (hash(x)=141742221)
-2000 val loss 4.2540
-2000 val perplexity 70.3849
-2000 train 4.260846 (lr=8.2821e-04) (hash(x)=160555812)
-2010 train 4.176174 (lr=8.2441e-04) (hash(x)=160371335)
-2020 train 4.178812 (lr=8.2059e-04) (hash(x)=167890767)
-2030 train 4.257946 (lr=8.1676e-04) (hash(x)=165127715)
-2040 train 4.199902 (lr=8.1291e-04) (hash(x)=151587302)
-2050 train 4.222305 (lr=8.0905e-04) (hash(x)=172347989)
-2060 train 4.293595 (lr=8.0517e-04) (hash(x)=161301919)
-2070 train 4.233062 (lr=8.0128e-04) (hash(x)=161472934)
-2080 train 4.317686 (lr=7.9737e-04) (hash(x)=159777322)
-2090 train 4.229338 (lr=7.9345e-04) (hash(x)=151149770)
-2100 val loss 4.2306
-2100 val perplexity 68.7611
-2100 train 4.057174 (lr=7.8951e-04) (hash(x)=160283520)
-2110 train 4.231931 (lr=7.8556e-04) (hash(x)=154222025)
-2120 train 4.170250 (lr=7.8160e-04) (hash(x)=155638178)
-2130 train 4.084243 (lr=7.7762e-04) (hash(x)=149125384)
-2140 train 4.139590 (lr=7.7363e-04) (hash(x)=159496074)
-2150 train 4.232738 (lr=7.6963e-04) (hash(x)=162019154)
-2160 train 4.377678 (lr=7.6562e-04) (hash(x)=155466928)
-2170 train 4.255927 (lr=7.6160e-04) (hash(x)=173469994)
-2180 train 4.366586 (lr=7.5756e-04) (hash(x)=163448753)
-2190 train 4.201189 (lr=7.5351e-04) (hash(x)=166595123)
-2200 val loss 4.2157
-2200 val perplexity 67.7392
-2200 train 4.084489 (lr=7.4945e-04) (hash(x)=165958875)
-2210 train 4.224067 (lr=7.4538e-04) (hash(x)=165019039)
-2220 train 4.200037 (lr=7.4130e-04) (hash(x)=167325225)
-2230 train 4.034694 (lr=7.3721e-04) (hash(x)=154287439)
-2240 train 4.104688 (lr=7.3312e-04) (hash(x)=158565817)
-2250 train 4.065577 (lr=7.2901e-04) (hash(x)=160592835)
-2260 train 4.109660 (lr=7.2489e-04) (hash(x)=144591648)
-2270 train 3.985849 (lr=7.2076e-04) (hash(x)=150486876)
-2280 train 4.181225 (lr=7.1663e-04) (hash(x)=159194788)
-2290 train 4.247094 (lr=7.1248e-04) (hash(x)=164315345)
-2300 val loss 4.1877
-2300 val perplexity 65.8734
-2300 train 4.180267 (lr=7.0833e-04) (hash(x)=170470665)
-2310 train 4.141272 (lr=7.0417e-04) (hash(x)=176539068)
-2320 train 4.125882 (lr=7.0001e-04) (hash(x)=161008474)
-2330 train 4.195934 (lr=6.9583e-04) (hash(x)=168474380)
-2340 train 4.196139 (lr=6.9165e-04) (hash(x)=156072131)
-2350 train 4.282218 (lr=6.8747e-04) (hash(x)=165127087)
-2360 train 4.277988 (lr=6.8327e-04) (hash(x)=159462801)
-2370 train 4.119334 (lr=6.7908e-04) (hash(x)=161698394)
-2380 train 4.044167 (lr=6.7487e-04) (hash(x)=165588078)
-2390 train 4.049417 (lr=6.7066e-04) (hash(x)=171673672)
-2400 val loss 4.1822
-2400 val perplexity 65.5089
-2400 train 4.110062 (lr=6.6645e-04) (hash(x)=157381516)
-2410 train 4.123432 (lr=6.6223e-04) (hash(x)=163238532)
-2420 train 4.072197 (lr=6.5801e-04) (hash(x)=156178237)
-2430 train 3.956477 (lr=6.5378e-04) (hash(x)=155080617)
-2440 train 4.065133 (lr=6.4955e-04) (hash(x)=167490538)
-2450 train 4.036334 (lr=6.4532e-04) (hash(x)=163757352)
-2460 train 3.995199 (lr=6.4108e-04) (hash(x)=156734811)
-2470 train 3.984332 (lr=6.3684e-04) (hash(x)=161522341)
-2480 train 4.251792 (lr=6.3260e-04) (hash(x)=174032060)
-2490 train 4.208078 (lr=6.2836e-04) (hash(x)=163065818)
-2500 val loss 4.1573
-2500 val perplexity 63.8964
-2500 train 4.266334 (lr=6.2412e-04) (hash(x)=126959780)
-2510 train 4.193432 (lr=6.1987e-04) (hash(x)=153844939)
-2520 train 4.174374 (lr=6.1562e-04) (hash(x)=174018174)
-2530 train 4.147284 (lr=6.1137e-04) (hash(x)=150568427)
-2540 train 4.150296 (lr=6.0712e-04) (hash(x)=161725206)
-2550 train 4.114000 (lr=6.0288e-04) (hash(x)=181346417)
-2560 train 4.102397 (lr=5.9863e-04) (hash(x)=168730011)
-2570 train 4.000432 (lr=5.9438e-04) (hash(x)=155900214)
-2580 train 4.055585 (lr=5.9013e-04) (hash(x)=160014621)
-2590 train 4.021368 (lr=5.8588e-04) (hash(x)=148996689)
-2600 val loss 4.1415
-2600 val perplexity 62.8973
-2600 train 4.102716 (lr=5.8164e-04) (hash(x)=145183968)
-2610 train 3.841917 (lr=5.7740e-04) (hash(x)=159420240)
-2620 train 3.829625 (lr=5.7316e-04) (hash(x)=151169903)
-2630 train 3.841263 (lr=5.6892e-04) (hash(x)=149454811)
-2640 train 3.903416 (lr=5.6468e-04) (hash(x)=155603914)
-2650 train 3.899846 (lr=5.6045e-04) (hash(x)=169545195)
-2660 train 4.246491 (lr=5.5622e-04) (hash(x)=168454079)
-2670 train 4.142575 (lr=5.5199e-04) (hash(x)=163110417)
-2680 train 4.095449 (lr=5.4777e-04) (hash(x)=166075603)
-2690 train 4.127481 (lr=5.4355e-04) (hash(x)=174132258)
-2700 val loss 4.1316
-2700 val perplexity 62.2747
-2700 train 4.132890 (lr=5.3934e-04) (hash(x)=169846940)
-2710 train 4.030401 (lr=5.3513e-04) (hash(x)=151553731)
-2720 train 4.130376 (lr=5.3092e-04) (hash(x)=164024255)
-2730 train 4.148735 (lr=5.2673e-04) (hash(x)=156458097)
-2740 train 4.084785 (lr=5.2253e-04) (hash(x)=154663748)
-2750 train 3.939831 (lr=5.1835e-04) (hash(x)=157548181)
-2760 train 4.056538 (lr=5.1417e-04) (hash(x)=151584783)
-2770 train 4.081723 (lr=5.0999e-04) (hash(x)=152692594)
-2780 train 4.055817 (lr=5.0583e-04) (hash(x)=165952919)
-2790 train 3.941807 (lr=5.0167e-04) (hash(x)=157484183)
-2800 val loss 4.1165
-2800 val perplexity 61.3440
-2800 train 4.016755 (lr=4.9752e-04) (hash(x)=154790034)
-2810 train 3.881252 (lr=4.9337e-04) (hash(x)=157856097)
-2820 train 3.861017 (lr=4.8924e-04) (hash(x)=168903939)
-2830 train 3.988133 (lr=4.8511e-04) (hash(x)=167001475)
-2840 train 3.817397 (lr=4.8099e-04) (hash(x)=158987150)
-2850 train 4.046437 (lr=4.7688e-04) (hash(x)=168695407)
-2860 train 4.184643 (lr=4.7279e-04) (hash(x)=164999034)
-2870 train 4.033503 (lr=4.6870e-04) (hash(x)=191197316)
-2880 train 3.989597 (lr=4.6462e-04) (hash(x)=159625200)
-2890 train 4.144570 (lr=4.6055e-04) (hash(x)=157756126)
-2900 val loss 4.0922
-2900 val perplexity 59.8697
-2900 train 3.998856 (lr=4.5649e-04) (hash(x)=164721377)
-2910 train 4.053391 (lr=4.5244e-04) (hash(x)=155873181)
-2920 train 4.001691 (lr=4.4840e-04) (hash(x)=156477082)
-2930 train 4.008209 (lr=4.4438e-04) (hash(x)=151261633)
-2940 train 3.944524 (lr=4.4037e-04) (hash(x)=160090385)
-2950 train 4.093969 (lr=4.3637e-04) (hash(x)=157300618)
-2960 train 4.002870 (lr=4.3238e-04) (hash(x)=163095555)
-2970 train 4.040195 (lr=4.2840e-04) (hash(x)=149649212)
-2980 train 3.845875 (lr=4.2444e-04) (hash(x)=174517675)
-2990 train 3.817628 (lr=4.2049e-04) (hash(x)=161344491)
-3000 val loss 4.0922
-3000 val perplexity 59.8704
-3000 train 3.840289 (lr=4.1655e-04) (hash(x)=163816905)
-3010 train 3.752749 (lr=4.1263e-04) (hash(x)=157420680)
-3020 train 3.859616 (lr=4.0872e-04) (hash(x)=153843035)
-3030 train 4.048740 (lr=4.0483e-04) (hash(x)=149770373)
-3040 train 4.119472 (lr=4.0095e-04) (hash(x)=165320741)
-3050 train 4.093215 (lr=3.9709e-04) (hash(x)=170181151)
-3060 train 4.008259 (lr=3.9324e-04) (hash(x)=163837230)
-3070 train 4.006531 (lr=3.8941e-04) (hash(x)=156812797)
-3080 train 3.981771 (lr=3.8559e-04) (hash(x)=143663627)
-3090 train 4.007534 (lr=3.8179e-04) (hash(x)=165052478)
-3100 val loss 4.0712
-3100 val perplexity 58.6272
-3100 train 3.940274 (lr=3.7801e-04) (hash(x)=151458705)
-3110 train 3.991632 (lr=3.7424e-04) (hash(x)=158228317)
-3120 train 4.163314 (lr=3.7049e-04) (hash(x)=149856801)
-3130 train 3.827750 (lr=3.6676e-04) (hash(x)=146297847)
-3140 train 3.928365 (lr=3.6304e-04) (hash(x)=172795167)
-3150 train 3.992523 (lr=3.5934e-04) (hash(x)=145936074)
-3160 train 4.076290 (lr=3.5566e-04) (hash(x)=157564303)
-3170 train 4.175789 (lr=3.5200e-04) (hash(x)=157451362)
-3180 train 4.090738 (lr=3.4836e-04) (hash(x)=156366043)
-3190 train 4.053452 (lr=3.4473e-04) (hash(x)=162344697)
-3200 val loss 4.0554
-3200 val perplexity 57.7058
-3200 train 4.075235 (lr=3.4113e-04) (hash(x)=164206298)
-3210 train 4.147161 (lr=3.3755e-04) (hash(x)=149468251)
-3220 train 4.080841 (lr=3.3398e-04) (hash(x)=157808606)
-3230 train 4.216505 (lr=3.3043e-04) (hash(x)=157078690)
-3240 train 4.016742 (lr=3.2691e-04) (hash(x)=165699696)
-3250 train 3.970005 (lr=3.2340e-04) (hash(x)=181352842)
-3260 train 4.065230 (lr=3.1992e-04) (hash(x)=142534383)
-3270 train 4.020547 (lr=3.1646e-04) (hash(x)=165297359)
-3280 train 4.069388 (lr=3.1302e-04) (hash(x)=160326815)
-3290 train 3.988149 (lr=3.0960e-04) (hash(x)=150043063)
-3300 val loss 4.0503
-3300 val perplexity 57.4167
-3300 train 3.946005 (lr=3.0620e-04) (hash(x)=164133091)
-3310 train 3.887840 (lr=3.0282e-04) (hash(x)=145803863)
-3320 train 3.912084 (lr=2.9947e-04) (hash(x)=157282009)
-3330 train 3.855647 (lr=2.9614e-04) (hash(x)=159126415)
-3340 train 4.055146 (lr=2.9283e-04) (hash(x)=164426675)
-3350 train 4.084033 (lr=2.8954e-04) (hash(x)=170424924)
-3360 train 4.029297 (lr=2.8628e-04) (hash(x)=163161621)
-3370 train 4.093150 (lr=2.8304e-04) (hash(x)=160786731)
-3380 train 4.046961 (lr=2.7982e-04) (hash(x)=165954688)
-3390 train 4.100743 (lr=2.7663e-04) (hash(x)=147371107)
-3400 val loss 4.0311
-3400 val perplexity 56.3249
-3400 train 4.038438 (lr=2.7346e-04) (hash(x)=153067604)
-3410 train 4.039258 (lr=2.7032e-04) (hash(x)=161748196)
-3420 train 4.049946 (lr=2.6720e-04) (hash(x)=168297963)
-3430 train 4.040197 (lr=2.6411e-04) (hash(x)=145086892)
-3440 train 3.946665 (lr=2.6104e-04) (hash(x)=173125711)
-3450 train 4.046565 (lr=2.5800e-04) (hash(x)=158080216)
-3460 train 3.867890 (lr=2.5498e-04) (hash(x)=158055750)
-3470 train 3.967304 (lr=2.5199e-04) (hash(x)=151381400)
-3480 train 3.907017 (lr=2.4903e-04) (hash(x)=147262346)
-3490 train 3.889940 (lr=2.4609e-04) (hash(x)=154740086)
-3500 val loss 4.0308
-3500 val perplexity 56.3046
-3500 train 3.808338 (lr=2.4317e-04) (hash(x)=145730968)
-3510 train 3.951792 (lr=2.4029e-04) (hash(x)=151355929)
-3520 train 3.943323 (lr=2.3743e-04) (hash(x)=165136835)
-3530 train 4.016521 (lr=2.3460e-04) (hash(x)=163869056)
-3540 train 3.957758 (lr=2.3179e-04) (hash(x)=161580134)
-3550 train 3.971200 (lr=2.2901e-04) (hash(x)=168595690)
-3560 train 4.012047 (lr=2.2626e-04) (hash(x)=158576466)
-3570 train 3.974293 (lr=2.2354e-04) (hash(x)=157298619)
-3580 train 4.061741 (lr=2.2085e-04) (hash(x)=160892091)
-3590 train 3.926429 (lr=2.1818e-04) (hash(x)=165168008)
-3600 val loss 4.0110
-3600 val perplexity 55.1995
-3600 train 4.029664 (lr=2.1555e-04) (hash(x)=144911638)
-3610 train 3.982770 (lr=2.1294e-04) (hash(x)=150614513)
-3620 train 3.977876 (lr=2.1036e-04) (hash(x)=167695016)
-3630 train 3.895358 (lr=2.0781e-04) (hash(x)=182708233)
-3640 train 3.998085 (lr=2.0529e-04) (hash(x)=154146274)
-3650 train 3.971681 (lr=2.0280e-04) (hash(x)=159129968)
-3660 train 3.882588 (lr=2.0033e-04) (hash(x)=146209497)
-3670 train 3.905688 (lr=1.9790e-04) (hash(x)=145578104)
-3680 train 3.893365 (lr=1.9550e-04) (hash(x)=150243521)
-3690 train 3.850880 (lr=1.9313e-04) (hash(x)=154328313)
-3700 val loss 4.0134
-3700 val perplexity 55.3345
-3700 train 3.824459 (lr=1.9079e-04) (hash(x)=159904582)
-3710 train 4.164941 (lr=1.8848e-04) (hash(x)=175986417)
-3720 train 4.081100 (lr=1.8619e-04) (hash(x)=158509479)
-3730 train 4.094745 (lr=1.8395e-04) (hash(x)=156087760)
-3740 train 4.090661 (lr=1.8173e-04) (hash(x)=177667170)
-3750 train 4.064246 (lr=1.7954e-04) (hash(x)=145256842)
-3760 train 4.028951 (lr=1.7738e-04) (hash(x)=149855635)
-3770 train 4.062577 (lr=1.7526e-04) (hash(x)=168798886)
-3780 train 3.976830 (lr=1.7317e-04) (hash(x)=156277179)
-3790 train 3.949924 (lr=1.7111e-04) (hash(x)=149698945)
-3800 val loss 3.9983
-3800 val perplexity 54.5053
-3800 train 3.904376 (lr=1.6908e-04) (hash(x)=168039654)
-3810 train 4.019974 (lr=1.6708e-04) (hash(x)=154595574)
-3820 train 4.069285 (lr=1.6511e-04) (hash(x)=158303284)
-3830 train 4.086666 (lr=1.6318e-04) (hash(x)=164128644)
-3840 train 4.034779 (lr=1.6128e-04) (hash(x)=166842997)
-3850 train 3.952650 (lr=1.5942e-04) (hash(x)=177681163)
-3860 train 3.927198 (lr=1.5758e-04) (hash(x)=177062575)
-3870 train 3.971887 (lr=1.5578e-04) (hash(x)=158957159)
-3880 train 3.902678 (lr=1.5401e-04) (hash(x)=151567740)
-3890 train 3.981099 (lr=1.5228e-04) (hash(x)=153261675)
-3900 val loss 3.9913
-3900 val perplexity 54.1231
-3900 train 4.249322 (lr=1.5058e-04) (hash(x)=160268683)
-3910 train 3.974450 (lr=1.4891e-04) (hash(x)=153745240)
-3920 train 4.024424 (lr=1.4727e-04) (hash(x)=164537824)
-3930 train 3.971678 (lr=1.4567e-04) (hash(x)=162062116)
-3940 train 3.934514 (lr=1.4411e-04) (hash(x)=168626742)
-3950 train 3.994434 (lr=1.4257e-04) (hash(x)=148753561)
-3960 train 3.924460 (lr=1.4108e-04) (hash(x)=163526242)
-3970 train 4.145557 (lr=1.3961e-04) (hash(x)=159329323)
-3980 train 4.067908 (lr=1.3818e-04) (hash(x)=146502339)
-3990 train 3.761982 (lr=1.3678e-04) (hash(x)=158409708)
-4000 val loss 3.9867
-4000 val perplexity 53.8755
-4000 train 3.938130 (lr=1.3542e-04) (hash(x)=154645137)
-4010 train 3.940663 (lr=1.3410e-04) (hash(x)=160044026)
-4020 train 3.922099 (lr=1.3280e-04) (hash(x)=155544686)
-4030 train 3.908717 (lr=1.3155e-04) (hash(x)=147589421)
-4040 train 4.015945 (lr=1.3032e-04) (hash(x)=158607990)
-4050 train 3.929790 (lr=1.2914e-04) (hash(x)=154108872)
-4060 train 4.011041 (lr=1.2798e-04) (hash(x)=167101465)
-4070 train 3.968780 (lr=1.2687e-04) (hash(x)=160323363)
-4080 train 3.958419 (lr=1.2578e-04) (hash(x)=150423692)
-4090 train 3.934869 (lr=1.2474e-04) (hash(x)=167461885)
-4100 val loss 3.9780
-4100 val perplexity 53.4092
-4100 train 3.968436 (lr=1.2373e-04) (hash(x)=151783377)
-4110 train 3.965554 (lr=1.2275e-04) (hash(x)=165679649)
-4120 train 3.862993 (lr=1.2181e-04) (hash(x)=153643039)
-4130 train 3.915146 (lr=1.2091e-04) (hash(x)=149648749)
-4140 train 3.952826 (lr=1.2004e-04) (hash(x)=162751418)
-4150 train 3.948545 (lr=1.1920e-04) (hash(x)=157047478)
-4160 train 3.885093 (lr=1.1841e-04) (hash(x)=156391812)
-4170 train 3.957154 (lr=1.1764e-04) (hash(x)=180926985)
-4180 train 3.999782 (lr=1.1692e-04) (hash(x)=168699059)
-4190 train 3.925567 (lr=1.1623e-04) (hash(x)=155410372)
-4200 val loss 3.9746
-4200 val perplexity 53.2314
-4200 train 3.931635 (lr=1.1557e-04) (hash(x)=152615252)
-4210 train 3.896113 (lr=1.1496e-04) (hash(x)=155679453)
-4220 train 4.012866 (lr=1.1437e-04) (hash(x)=152954399)
-4230 train 4.015634 (lr=1.1383e-04) (hash(x)=164608084)
-4240 train 4.040376 (lr=1.1332e-04) (hash(x)=162463867)
-4250 train 3.956368 (lr=1.1285e-04) (hash(x)=168832088)
-4260 train 3.928489 (lr=1.1241e-04) (hash(x)=162698882)
-4270 train 3.955882 (lr=1.1201e-04) (hash(x)=180861492)
-4280 train 3.845699 (lr=1.1164e-04) (hash(x)=160824300)
-4290 train 3.908987 (lr=1.1132e-04) (hash(x)=157750938)
-4300 val loss 3.9745
-4300 val perplexity 53.2245
-4300 train 3.861733 (lr=1.1103e-04) (hash(x)=157216833)
-4310 train 3.856105 (lr=1.1077e-04) (hash(x)=176184661)
-4320 train 4.024665 (lr=1.1055e-04) (hash(x)=161537117)
-4330 train 3.974107 (lr=1.1037e-04) (hash(x)=160700216)
-4340 train 3.925903 (lr=1.1022e-04) (hash(x)=175519903)
-4350 train 3.930590 (lr=1.1011e-04) (hash(x)=167988701)
-4360 train 3.879607 (lr=1.1004e-04) (hash(x)=162619084)
-4370 train 4.270770 (lr=1.1000e-04) (hash(x)=166292684)
-4374 val loss 3.9689
-4374 val perplexity 52.9279
+max_steps: 2500
+0 val loss 10.9737
+0 val perplexity 58317.8945
+0 train 10.973824 (lr=8.3916e-07) (hash(x)=23347629)
+10 train 10.111941 (lr=9.2308e-06) (hash(x)=19636655)
+20 train 9.612789 (lr=1.7622e-05) (hash(x)=24490248)
+30 train 9.418690 (lr=2.6014e-05) (hash(x)=21733558)
+40 train 9.161636 (lr=3.4406e-05) (hash(x)=31750616)
+50 train 8.849428 (lr=4.2797e-05) (hash(x)=20041895)
+60 train 8.489235 (lr=5.1189e-05) (hash(x)=21416280)
+70 train 8.164173 (lr=5.9580e-05) (hash(x)=18066780)
+80 train 7.865506 (lr=6.7972e-05) (hash(x)=21809427)
+90 train 7.656304 (lr=7.6364e-05) (hash(x)=22061595)
+100 val loss 7.4483
+100 val perplexity 1716.9163
+100 train 7.400749 (lr=8.4755e-05) (hash(x)=17468914)
+110 train 7.208503 (lr=9.3147e-05) (hash(x)=20421520)
+120 train 7.228409 (lr=1.0154e-04) (hash(x)=23140133)
+130 train 7.090940 (lr=1.0993e-04) (hash(x)=20006835)
+140 train 6.848573 (lr=1.1832e-04) (hash(x)=18628134)
+150 train 6.815238 (lr=1.2671e-04) (hash(x)=21913447)
+160 train 6.760431 (lr=1.3510e-04) (hash(x)=22997297)
+170 train 6.547211 (lr=1.4350e-04) (hash(x)=17904821)
+180 train 6.407768 (lr=1.5189e-04) (hash(x)=18826982)
+190 train 6.488183 (lr=1.6028e-04) (hash(x)=21387716)
+200 val loss 6.3713
+200 val perplexity 584.8389
+200 train 6.291077 (lr=1.6867e-04) (hash(x)=16236395)
+210 train 6.170132 (lr=1.7706e-04) (hash(x)=21428681)
+220 train 6.310472 (lr=1.8545e-04) (hash(x)=20461738)
+230 train 6.141428 (lr=1.9385e-04) (hash(x)=20394865)
+240 train 6.032245 (lr=2.0224e-04) (hash(x)=18028671)
+250 train 5.879856 (lr=2.1063e-04) (hash(x)=18122073)
+260 train 5.798477 (lr=2.1902e-04) (hash(x)=22994143)
+270 train 6.009269 (lr=2.2741e-04) (hash(x)=20646260)
+280 train 5.863047 (lr=2.3580e-04) (hash(x)=23204708)
+290 train 5.759854 (lr=2.4420e-04) (hash(x)=17900252)
+300 val loss 5.7958
+300 val perplexity 328.9242
+300 train 5.641637 (lr=2.5259e-04) (hash(x)=19073819)
+310 train 5.807188 (lr=2.6098e-04) (hash(x)=22378232)
+320 train 5.714730 (lr=2.6937e-04) (hash(x)=19910430)
+330 train 5.637177 (lr=2.7776e-04) (hash(x)=25168753)
+340 train 5.525199 (lr=2.8615e-04) (hash(x)=20665256)
+350 train 5.400629 (lr=2.9455e-04) (hash(x)=19951160)
+360 train 5.604391 (lr=3.0294e-04) (hash(x)=21178202)
+370 train 5.465904 (lr=3.1133e-04) (hash(x)=20435886)
+380 train 5.397383 (lr=3.1972e-04) (hash(x)=21255097)
+390 train 5.401009 (lr=3.2811e-04) (hash(x)=19596330)
+400 val loss 5.3366
+400 val perplexity 207.8033
+400 train 5.363309 (lr=3.3650e-04) (hash(x)=21789426)
+410 train 5.276180 (lr=3.4490e-04) (hash(x)=18366401)
+420 train 5.174843 (lr=3.5329e-04) (hash(x)=18614660)
+430 train 5.310909 (lr=3.6168e-04) (hash(x)=21649063)
+440 train 5.177170 (lr=3.7007e-04) (hash(x)=18505097)
+450 train 5.178087 (lr=3.7846e-04) (hash(x)=23434312)
+460 train 5.063648 (lr=3.8685e-04) (hash(x)=16584569)
+470 train 4.936549 (lr=3.9524e-04) (hash(x)=13447111)
+480 train 5.159476 (lr=4.0364e-04) (hash(x)=22645130)
+490 train 5.006290 (lr=4.1203e-04) (hash(x)=19362777)
+500 val loss 4.9944
+500 val perplexity 147.5845
+500 train 4.929747 (lr=4.2042e-04) (hash(x)=20921874)
+510 train 4.791442 (lr=4.2881e-04) (hash(x)=22123048)
+520 train 4.830326 (lr=4.3720e-04) (hash(x)=21215368)
+530 train 4.969443 (lr=4.4559e-04) (hash(x)=20717936)
+540 train 4.864767 (lr=4.5399e-04) (hash(x)=20118007)
+550 train 4.713600 (lr=4.6238e-04) (hash(x)=21402522)
+560 train 4.649039 (lr=4.7077e-04) (hash(x)=19057180)
+570 train 4.831183 (lr=4.7916e-04) (hash(x)=19360553)
+580 train 4.737230 (lr=4.8755e-04) (hash(x)=18194693)
+590 train 4.624774 (lr=4.9594e-04) (hash(x)=19863826)
+600 val loss 4.5948
+600 val perplexity 98.9706
+600 train 4.550488 (lr=5.0434e-04) (hash(x)=21794551)
+610 train 4.509958 (lr=5.1273e-04) (hash(x)=18685622)
+620 train 4.404768 (lr=5.2112e-04) (hash(x)=23473104)
+630 train 4.504869 (lr=5.2951e-04) (hash(x)=18671672)
+640 train 4.535685 (lr=5.3790e-04) (hash(x)=19234240)
+650 train 4.438543 (lr=5.4629e-04) (hash(x)=23183903)
+660 train 4.272037 (lr=5.5469e-04) (hash(x)=23665543)
+670 train 4.429008 (lr=5.6308e-04) (hash(x)=22055313)
+680 train 4.391896 (lr=5.7147e-04) (hash(x)=22321668)
+690 train 4.508237 (lr=5.7986e-04) (hash(x)=21568926)
+700 val loss 4.3439
+700 val perplexity 77.0095
+700 train 4.349285 (lr=5.8825e-04) (hash(x)=19783143)
+710 train 4.275688 (lr=5.9664e-04) (hash(x)=21565258)
+720 train 4.361594 (lr=5.9999e-04) (hash(x)=16127212)
+730 train 4.289129 (lr=5.9991e-04) (hash(x)=19658174)
+740 train 4.256943 (lr=5.9974e-04) (hash(x)=16454319)
+750 train 4.200531 (lr=5.9949e-04) (hash(x)=17993154)
+760 train 4.106220 (lr=5.9915e-04) (hash(x)=19256989)
+770 train 4.085679 (lr=5.9874e-04) (hash(x)=17351653)
+780 train 4.108175 (lr=5.9824e-04) (hash(x)=21626120)
+790 train 4.263186 (lr=5.9765e-04) (hash(x)=16367412)
+800 val loss 4.1683
+800 val perplexity 64.6086
+800 train 4.119504 (lr=5.9698e-04) (hash(x)=22321504)
+810 train 4.002688 (lr=5.9623e-04) (hash(x)=19869755)
+820 train 4.173855 (lr=5.9540e-04) (hash(x)=24183665)
+830 train 4.149780 (lr=5.9449e-04) (hash(x)=19194113)
+840 train 3.995206 (lr=5.9349e-04) (hash(x)=17710503)
+850 train 4.026993 (lr=5.9241e-04) (hash(x)=17231851)
+860 train 4.149008 (lr=5.9126e-04) (hash(x)=20584272)
+870 train 4.048472 (lr=5.9002e-04) (hash(x)=17468790)
+880 train 3.973584 (lr=5.8869e-04) (hash(x)=22304409)
+890 train 4.070874 (lr=5.8729e-04) (hash(x)=18871567)
+900 val loss 4.0364
+900 val perplexity 56.6249
+900 train 4.014625 (lr=5.8581e-04) (hash(x)=20924720)
+910 train 3.928895 (lr=5.8425e-04) (hash(x)=21163464)
+920 train 3.866852 (lr=5.8262e-04) (hash(x)=20120862)
+930 train 4.041391 (lr=5.8090e-04) (hash(x)=19565198)
+940 train 3.958992 (lr=5.7911e-04) (hash(x)=18517540)
+950 train 3.931305 (lr=5.7723e-04) (hash(x)=18300759)
+960 train 3.965863 (lr=5.7529e-04) (hash(x)=21443001)
+970 train 3.911733 (lr=5.7326e-04) (hash(x)=15210971)
+980 train 4.072883 (lr=5.7116e-04) (hash(x)=20874579)
+990 train 3.992456 (lr=5.6899e-04) (hash(x)=20690802)
+1000 val loss 3.9428
+1000 val perplexity 51.5621
+1000 train 3.864140 (lr=5.6674e-04) (hash(x)=18635877)
+1010 train 3.980101 (lr=5.6442e-04) (hash(x)=21141462)
+1020 train 3.996318 (lr=5.6202e-04) (hash(x)=19483400)
+1030 train 3.901821 (lr=5.5956e-04) (hash(x)=24792825)
+1040 train 3.879690 (lr=5.5702e-04) (hash(x)=16743928)
+1050 train 3.925042 (lr=5.5441e-04) (hash(x)=22540052)
+1060 train 3.943355 (lr=5.5174e-04) (hash(x)=20546637)
+1070 train 3.842065 (lr=5.4899e-04) (hash(x)=13739753)
+1080 train 3.921352 (lr=5.4618e-04) (hash(x)=20071304)
+1090 train 3.852434 (lr=5.4330e-04) (hash(x)=19339656)
+1100 val loss 3.8737
+1100 val perplexity 48.1209
+1100 train 3.840385 (lr=5.4035e-04) (hash(x)=20039628)
+1110 train 3.898952 (lr=5.3734e-04) (hash(x)=19249691)
+1120 train 3.916404 (lr=5.3426e-04) (hash(x)=19048809)
+1130 train 3.843654 (lr=5.3113e-04) (hash(x)=19309000)
+1140 train 3.801544 (lr=5.2792e-04) (hash(x)=21787927)
+1150 train 3.715839 (lr=5.2466e-04) (hash(x)=19998536)
+1160 train 3.865678 (lr=5.2134e-04) (hash(x)=23724694)
+1170 train 3.885109 (lr=5.1796e-04) (hash(x)=17148204)
+1180 train 3.797027 (lr=5.1452e-04) (hash(x)=19511852)
+1190 train 3.773531 (lr=5.1102e-04) (hash(x)=19352862)
+1200 val loss 3.8026
+1200 val perplexity 44.8180
+1200 train 3.714751 (lr=5.0746e-04) (hash(x)=19530144)
+1210 train 3.901145 (lr=5.0386e-04) (hash(x)=19897392)
+1220 train 3.802177 (lr=5.0019e-04) (hash(x)=24402963)
+1230 train 3.720290 (lr=4.9648e-04) (hash(x)=18589754)
+1240 train 3.728300 (lr=4.9271e-04) (hash(x)=21951750)
+1250 train 3.802309 (lr=4.8889e-04) (hash(x)=18611379)
+1260 train 3.824818 (lr=4.8503e-04) (hash(x)=20338521)
+1270 train 3.739093 (lr=4.8111e-04) (hash(x)=18308405)
+1280 train 3.740676 (lr=4.7715e-04) (hash(x)=20920945)
+1290 train 3.705351 (lr=4.7315e-04) (hash(x)=20094152)
+1300 val loss 3.7477
+1300 val perplexity 42.4247
+1300 train 3.821512 (lr=4.6909e-04) (hash(x)=19003947)
+1310 train 3.780432 (lr=4.6500e-04) (hash(x)=19356356)
+1320 train 3.734794 (lr=4.6086e-04) (hash(x)=19390343)
+1330 train 3.672205 (lr=4.5669e-04) (hash(x)=17535982)
+1340 train 3.695519 (lr=4.5247e-04) (hash(x)=21959104)
+1350 train 3.795491 (lr=4.4822e-04) (hash(x)=21982099)
+1360 train 3.755167 (lr=4.4393e-04) (hash(x)=17176243)
+1370 train 3.698761 (lr=4.3960e-04) (hash(x)=21223627)
+1380 train 3.716665 (lr=4.3524e-04) (hash(x)=22462931)
+1390 train 3.738315 (lr=4.3085e-04) (hash(x)=19153464)
+1400 val loss 3.6977
+1400 val perplexity 40.3550
+1400 train 3.629280 (lr=4.2643e-04) (hash(x)=20146496)
+1410 train 3.688986 (lr=4.2197e-04) (hash(x)=21781327)
+1420 train 3.761765 (lr=4.1749e-04) (hash(x)=19234051)
+1430 train 3.809854 (lr=4.1298e-04) (hash(x)=20769898)
+1440 train 3.657742 (lr=4.0845e-04) (hash(x)=22042430)
+1450 train 3.680538 (lr=4.0389e-04) (hash(x)=20527387)
+1460 train 3.667914 (lr=3.9931e-04) (hash(x)=18925766)
+1470 train 3.664164 (lr=3.9470e-04) (hash(x)=18198851)
+1480 train 3.738787 (lr=3.9008e-04) (hash(x)=21847281)
+1490 train 3.710153 (lr=3.8544e-04) (hash(x)=15645274)
+1500 val loss 3.6543
+1500 val perplexity 38.6424
+1500 train 3.578695 (lr=3.8078e-04) (hash(x)=14693370)
+1510 train 3.588135 (lr=3.7610e-04) (hash(x)=21810857)
+1520 train 3.684349 (lr=3.7142e-04) (hash(x)=18980052)
+1530 train 3.709998 (lr=3.6671e-04) (hash(x)=22868842)
+1540 train 3.621936 (lr=3.6200e-04) (hash(x)=17289166)
+1550 train 3.707079 (lr=3.5728e-04) (hash(x)=18109373)
+1560 train 3.740611 (lr=3.5255e-04) (hash(x)=22377619)
+1570 train 3.568432 (lr=3.4781e-04) (hash(x)=20307182)
+1580 train 3.715766 (lr=3.4306e-04) (hash(x)=20391291)
+1590 train 3.584409 (lr=3.3831e-04) (hash(x)=18690939)
+1600 val loss 3.6192
+1600 val perplexity 37.3083
+1600 train 3.597387 (lr=3.3356e-04) (hash(x)=16719886)
+1610 train 3.589051 (lr=3.2881e-04) (hash(x)=16040840)
+1620 train 3.664560 (lr=3.2406e-04) (hash(x)=25795140)
+1630 train 3.667998 (lr=3.1931e-04) (hash(x)=21936964)
+1640 train 3.566886 (lr=3.1456e-04) (hash(x)=11593027)
+1650 train 3.623425 (lr=3.0982e-04) (hash(x)=23256718)
+1660 train 3.584052 (lr=3.0509e-04) (hash(x)=20289140)
+1670 train 3.588777 (lr=3.0036e-04) (hash(x)=18382509)
+1680 train 3.655182 (lr=2.9564e-04) (hash(x)=26604329)
+1690 train 3.622003 (lr=2.9093e-04) (hash(x)=22122534)
+1700 val loss 3.5880
+1700 val perplexity 36.1632
+1700 train 3.638704 (lr=2.8624e-04) (hash(x)=22884812)
+1710 train 3.583168 (lr=2.8156e-04) (hash(x)=18598496)
+1720 train 3.603663 (lr=2.7689e-04) (hash(x)=22147447)
+1730 train 3.608187 (lr=2.7224e-04) (hash(x)=18912135)
+1740 train 3.445968 (lr=2.6761e-04) (hash(x)=21952974)
+1750 train 3.559899 (lr=2.6299e-04) (hash(x)=18390978)
+1760 train 3.568171 (lr=2.5840e-04) (hash(x)=17979288)
+1770 train 3.575947 (lr=2.5383e-04) (hash(x)=18979099)
+1780 train 3.541605 (lr=2.4928e-04) (hash(x)=20511922)
+1790 train 3.595030 (lr=2.4476e-04) (hash(x)=21814060)
+1800 val loss 3.5618
+1800 val perplexity 35.2250
+1800 train 3.578795 (lr=2.4026e-04) (hash(x)=18076243)
+1810 train 3.571294 (lr=2.3580e-04) (hash(x)=19721209)
+1820 train 3.516656 (lr=2.3136e-04) (hash(x)=18140847)
+1830 train 3.581553 (lr=2.2695e-04) (hash(x)=22065413)
+1840 train 3.479596 (lr=2.2257e-04) (hash(x)=21166111)
+1850 train 3.481246 (lr=2.1823e-04) (hash(x)=21904302)
+1860 train 3.546546 (lr=2.1392e-04) (hash(x)=18283828)
+1870 train 3.504316 (lr=2.0965e-04) (hash(x)=20892773)
+1880 train 3.533027 (lr=2.0542e-04) (hash(x)=17781264)
+1890 train 3.503898 (lr=2.0122e-04) (hash(x)=20612962)
+1900 val loss 3.5391
+1900 val perplexity 34.4369
+1900 train 3.594285 (lr=1.9706e-04) (hash(x)=20421058)
+1910 train 3.567181 (lr=1.9295e-04) (hash(x)=18554808)
+1920 train 3.524355 (lr=1.8887e-04) (hash(x)=18189279)
+1930 train 3.539775 (lr=1.8485e-04) (hash(x)=19923279)
+1940 train 3.506746 (lr=1.8086e-04) (hash(x)=21665511)
+1950 train 3.526539 (lr=1.7692e-04) (hash(x)=22038489)
+1960 train 3.499939 (lr=1.7303e-04) (hash(x)=18896956)
+1970 train 3.528859 (lr=1.6919e-04) (hash(x)=20408383)
+1980 train 3.510734 (lr=1.6540e-04) (hash(x)=23039940)
+1990 train 3.643764 (lr=1.6166e-04) (hash(x)=20631172)
+2000 val loss 3.5196
+2000 val perplexity 33.7720
+2000 train 3.457407 (lr=1.5797e-04) (hash(x)=20206106)
+2010 train 3.553501 (lr=1.5433e-04) (hash(x)=21864020)
+2020 train 3.537772 (lr=1.5075e-04) (hash(x)=21391128)
+2030 train 3.500875 (lr=1.4723e-04) (hash(x)=18676636)
+2040 train 3.548269 (lr=1.4376e-04) (hash(x)=15987134)
+2050 train 3.543647 (lr=1.4035e-04) (hash(x)=23940769)
+2060 train 3.518487 (lr=1.3699e-04) (hash(x)=17152840)
+2070 train 3.488767 (lr=1.3370e-04) (hash(x)=22958445)
+2080 train 3.541727 (lr=1.3047e-04) (hash(x)=23616724)
+2090 train 3.511062 (lr=1.2730e-04) (hash(x)=16571283)
+2100 val loss 3.4991
+2100 val perplexity 33.0870
+2100 train 3.507874 (lr=1.2419e-04) (hash(x)=19070327)
+2110 train 3.553292 (lr=1.2115e-04) (hash(x)=20758626)
+2120 train 3.497998 (lr=1.1817e-04) (hash(x)=18397810)
+2130 train 3.485493 (lr=1.1525e-04) (hash(x)=18045234)
+2140 train 3.590516 (lr=1.1241e-04) (hash(x)=18969565)
+2150 train 3.527048 (lr=1.0963e-04) (hash(x)=17149701)
+2160 train 3.446316 (lr=1.0692e-04) (hash(x)=23075443)
+2170 train 3.479116 (lr=1.0427e-04) (hash(x)=17977337)
+2180 train 3.541360 (lr=1.0170e-04) (hash(x)=20657407)
+2190 train 3.528511 (lr=9.9199e-05) (hash(x)=19985933)
+2200 val loss 3.4851
+2200 val perplexity 32.6242
+2200 train 3.482155 (lr=9.6769e-05) (hash(x)=19025387)
+2210 train 3.500153 (lr=9.4412e-05) (hash(x)=23318724)
+2220 train 3.571764 (lr=9.2127e-05) (hash(x)=19308039)
+2230 train 3.413027 (lr=8.9916e-05) (hash(x)=20733682)
+2240 train 3.511998 (lr=8.7779e-05) (hash(x)=18519560)
+2250 train 3.513827 (lr=8.5717e-05) (hash(x)=20299444)
+2260 train 3.472827 (lr=8.3731e-05) (hash(x)=18819615)
+2270 train 3.507519 (lr=8.1821e-05) (hash(x)=18743705)
+2280 train 3.491784 (lr=7.9988e-05) (hash(x)=17190793)
+2290 train 3.527252 (lr=7.8232e-05) (hash(x)=23126906)
+2300 val loss 3.4731
+2300 val perplexity 32.2360
+2300 train 3.462830 (lr=7.6555e-05) (hash(x)=24812267)
+2310 train 3.444661 (lr=7.4956e-05) (hash(x)=20147105)
+2320 train 3.561392 (lr=7.3436e-05) (hash(x)=24619936)
+2330 train 3.426275 (lr=7.1995e-05) (hash(x)=17730658)
+2340 train 3.404282 (lr=7.0635e-05) (hash(x)=19345731)
+2350 train 3.442647 (lr=6.9354e-05) (hash(x)=26415058)
+2360 train 3.475547 (lr=6.8155e-05) (hash(x)=18622999)
+2370 train 3.407006 (lr=6.7036e-05) (hash(x)=18848567)
+2380 train 3.492635 (lr=6.5999e-05) (hash(x)=21903360)
+2390 train 3.505845 (lr=6.5044e-05) (hash(x)=23479216)
+2400 val loss 3.4645
+2400 val perplexity 31.9606
+2400 train 3.501653 (lr=6.4171e-05) (hash(x)=19762013)
+2410 train 3.405435 (lr=6.3380e-05) (hash(x)=20622912)
+2420 train 3.525760 (lr=6.2672e-05) (hash(x)=17669226)
+2430 train 3.452360 (lr=6.2046e-05) (hash(x)=19536719)
+2440 train 3.394011 (lr=6.1504e-05) (hash(x)=19913725)
+2450 train 3.438047 (lr=6.1045e-05) (hash(x)=20782072)
+2460 train 3.405989 (lr=6.0669e-05) (hash(x)=19795002)
+2470 train 3.451315 (lr=6.0376e-05) (hash(x)=17003051)
+2480 train 3.421663 (lr=6.0167e-05) (hash(x)=21414190)
+2490 train 3.468951 (lr=6.0042e-05) (hash(x)=20943620)
+2499 val loss 3.4586
+2499 val perplexity 31.7722

baseline/model_02499.pt

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:8b35d0e7711f1050e2870cc8c101348da0fa14fe5a11d51400e028a1b6c73975
+oid sha256:e08b7d4fbe8be9101e4aeed21136757084a0745c436bcc4d14a1214e8a95f692
 size 548152706

baseline/optimizer_02499.pt

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:0de7cbb6520ede3edfcc16a0872f326e0931367cfce6321dfc89ffe8c61750b9
+oid sha256:c3adb411f872a91cc4c1e27535a25671288571635d92fe0f48c762e8750dcc9a
 size 995652870