Upload folder using huggingface_hub

12_head_baseline_lr_30e-4_head_dim_22_fixed/args.json

@@ -1 +1 @@
-{"hellaswag": true, "attention_kind": "selective", "log_dir": "logs/repro_selective_pattern_rankings/12_head_baseline_lr_30e-4_head_dim_22_fixed", "resume_checkpoint": null, "resume_optimizer": false, "add_a_head": false, "add_head_to_start": true, "new_head_init": "normal", "n_heads": 12, "n_embd": 264, "head_dim": 22, "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": 1341, "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": 64, "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": 0.003, "decay_lr": true, "readout_zero_init": false, "query_zero_init": false, "l1_loss": false, "debugpy": false, "key": "12_head_baseline_lr_30e-4_fixed"}
+{"hellaswag": true, "attention_kind": "selective", "log_dir": "logs/repro_selective_pattern_rankings/12_head_baseline_lr_30e-4_head_dim_22_fixed", "resume_checkpoint": null, "resume_optimizer": false, "add_a_head": false, "add_head_to_start": true, "new_head_init": "normal", "n_heads": 12, "n_embd": 264, "head_dim": 22, "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": 1340, "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": 64, "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": 0.003, "decay_lr": true, "readout_zero_init": false, "query_zero_init": false, "l1_loss": false, "debugpy": false, "key": "12_head_baseline_lr_30e-4_fixed"}

12_head_baseline_lr_30e-4_head_dim_22_fixed/log2.txt

@@ -1,529 +1,529 @@
 max_steps: 4375
-0 val loss 11.3009
-0 val perplexity 80893.6016
-0 train 11.308078 (lr=4.1958e-06) (hash(x)=83426457)
-10 train 9.967107 (lr=4.6154e-05) (hash(x)=78934557)
-20 train 9.448398 (lr=8.8112e-05) (hash(x)=77027105)
-30 train 8.687866 (lr=1.3007e-04) (hash(x)=71598963)
-40 train 8.070792 (lr=1.7203e-04) (hash(x)=82386827)
-50 train 7.851956 (lr=2.1399e-04) (hash(x)=85920934)
-60 train 7.647997 (lr=2.5594e-04) (hash(x)=81616761)
-70 train 7.565257 (lr=2.9790e-04) (hash(x)=86773189)
-80 train 7.327011 (lr=3.3986e-04) (hash(x)=71973470)
-90 train 7.220871 (lr=3.8182e-04) (hash(x)=83854524)
-100 val loss 7.0878
-100 val perplexity 1197.3223
-100 train 6.990294 (lr=4.2378e-04) (hash(x)=77188949)
-110 train 6.908389 (lr=4.6573e-04) (hash(x)=72138399)
-120 train 6.799723 (lr=5.0769e-04) (hash(x)=80173123)
-130 train 6.603529 (lr=5.4965e-04) (hash(x)=71981857)
-140 train 6.581169 (lr=5.9161e-04) (hash(x)=75014146)
-150 train 6.476323 (lr=6.3357e-04) (hash(x)=83329326)
-160 train 6.317798 (lr=6.7552e-04) (hash(x)=76522382)
-170 train 6.616613 (lr=7.1748e-04) (hash(x)=88508613)
-180 train 6.355318 (lr=7.5944e-04) (hash(x)=90018376)
-190 train 6.305059 (lr=8.0140e-04) (hash(x)=79854209)
-200 val loss 6.2276
-200 val perplexity 506.5309
-200 train 6.359780 (lr=8.4336e-04) (hash(x)=78560782)
-210 train 6.216138 (lr=8.8531e-04) (hash(x)=83733723)
-220 train 6.057124 (lr=9.2727e-04) (hash(x)=79271902)
-230 train 6.039199 (lr=9.6923e-04) (hash(x)=83461326)
-240 train 5.965477 (lr=1.0112e-03) (hash(x)=90489642)
-250 train 5.934767 (lr=1.0531e-03) (hash(x)=79614895)
-260 train 5.874289 (lr=1.0951e-03) (hash(x)=88646240)
-270 train 5.862950 (lr=1.1371e-03) (hash(x)=85620399)
-280 train 5.897366 (lr=1.1790e-03) (hash(x)=89054642)
-290 train 5.655605 (lr=1.2210e-03) (hash(x)=81160339)
-300 val loss 5.7660
-300 val perplexity 319.2502
-300 train 5.745425 (lr=1.2629e-03) (hash(x)=75259104)
-310 train 5.608951 (lr=1.3049e-03) (hash(x)=84448689)
-320 train 5.532045 (lr=1.3469e-03) (hash(x)=88463193)
-330 train 5.679371 (lr=1.3888e-03) (hash(x)=76313584)
-340 train 5.643176 (lr=1.4308e-03) (hash(x)=84598604)
-350 train 5.512202 (lr=1.4727e-03) (hash(x)=80551549)
-360 train 5.612100 (lr=1.5147e-03) (hash(x)=77926207)
-370 train 5.531306 (lr=1.5566e-03) (hash(x)=78142578)
-380 train 5.495136 (lr=1.5986e-03) (hash(x)=80853941)
-390 train 5.551072 (lr=1.6406e-03) (hash(x)=73606747)
-400 val loss 5.4746
-400 val perplexity 238.5595
-400 train 5.423995 (lr=1.6825e-03) (hash(x)=75283154)
-410 train 5.501243 (lr=1.7245e-03) (hash(x)=77829445)
-420 train 5.440038 (lr=1.7664e-03) (hash(x)=92661009)
-430 train 5.282501 (lr=1.8084e-03) (hash(x)=74159185)
-440 train 5.219159 (lr=1.8503e-03) (hash(x)=77656452)
-450 train 5.217721 (lr=1.8923e-03) (hash(x)=78633015)
-460 train 5.199747 (lr=1.9343e-03) (hash(x)=79484077)
-470 train 5.224099 (lr=1.9762e-03) (hash(x)=85308160)
-480 train 5.296034 (lr=2.0182e-03) (hash(x)=86733273)
-490 train 5.178659 (lr=2.0601e-03) (hash(x)=77806270)
-500 val loss 5.2423
-500 val perplexity 189.1110
-500 train 5.231162 (lr=2.1021e-03) (hash(x)=79154473)
-510 train 5.318253 (lr=2.1441e-03) (hash(x)=82252738)
-520 train 5.196646 (lr=2.1860e-03) (hash(x)=79224242)
-530 train 5.226891 (lr=2.2280e-03) (hash(x)=77695815)
-540 train 5.115963 (lr=2.2699e-03) (hash(x)=72435526)
-550 train 5.060951 (lr=2.3119e-03) (hash(x)=78722012)
-560 train 5.106854 (lr=2.3538e-03) (hash(x)=86379154)
-570 train 4.962247 (lr=2.3958e-03) (hash(x)=75571834)
-580 train 4.942504 (lr=2.4378e-03) (hash(x)=75212674)
-590 train 4.955385 (lr=2.4797e-03) (hash(x)=83039622)
-600 val loss 5.0358
-600 val perplexity 153.8284
-600 train 4.922983 (lr=2.5217e-03) (hash(x)=75453059)
-610 train 4.840949 (lr=2.5636e-03) (hash(x)=78156260)
-620 train 4.937030 (lr=2.6056e-03) (hash(x)=73224372)
-630 train 4.916430 (lr=2.6476e-03) (hash(x)=86669689)
-640 train 4.893358 (lr=2.6895e-03) (hash(x)=76838089)
-650 train 4.900596 (lr=2.7315e-03) (hash(x)=77816509)
-660 train 4.941143 (lr=2.7734e-03) (hash(x)=78714357)
-670 train 4.823511 (lr=2.8154e-03) (hash(x)=67108441)
-680 train 4.783498 (lr=2.8573e-03) (hash(x)=74297523)
-690 train 4.867505 (lr=2.8993e-03) (hash(x)=81807227)
-700 val loss 4.8612
-700 val perplexity 129.1827
-700 train 4.727699 (lr=2.9413e-03) (hash(x)=80680202)
-710 train 4.789280 (lr=2.9832e-03) (hash(x)=81685619)
-720 train 4.771536 (lr=3.0000e-03) (hash(x)=88293912)
-730 train 4.709539 (lr=2.9999e-03) (hash(x)=81156131)
-740 train 4.657020 (lr=2.9997e-03) (hash(x)=73015886)
-750 train 4.764834 (lr=2.9994e-03) (hash(x)=78392112)
-760 train 4.719293 (lr=2.9990e-03) (hash(x)=82475322)
-770 train 4.730413 (lr=2.9985e-03) (hash(x)=73069343)
-780 train 4.906968 (lr=2.9979e-03) (hash(x)=93027554)
-790 train 4.725779 (lr=2.9972e-03) (hash(x)=73364444)
-800 val loss 4.7368
-800 val perplexity 114.0640
-800 train 4.738317 (lr=2.9964e-03) (hash(x)=78801867)
-810 train 4.709892 (lr=2.9955e-03) (hash(x)=82551837)
-820 train 4.550749 (lr=2.9945e-03) (hash(x)=78287067)
-830 train 4.599868 (lr=2.9934e-03) (hash(x)=82837338)
-840 train 4.526814 (lr=2.9922e-03) (hash(x)=72494345)
-850 train 4.587938 (lr=2.9909e-03) (hash(x)=80840474)
-860 train 4.476433 (lr=2.9896e-03) (hash(x)=77298722)
-870 train 4.509427 (lr=2.9881e-03) (hash(x)=80296151)
-880 train 4.609691 (lr=2.9865e-03) (hash(x)=72779638)
-890 train 4.611323 (lr=2.9848e-03) (hash(x)=78473759)
-900 val loss 4.6745
-900 val perplexity 107.1826
-900 train 4.563942 (lr=2.9830e-03) (hash(x)=76625599)
-910 train 4.660621 (lr=2.9811e-03) (hash(x)=82707605)
-920 train 4.592109 (lr=2.9792e-03) (hash(x)=73411530)
-930 train 4.850684 (lr=2.9771e-03) (hash(x)=84116802)
-940 train 4.624489 (lr=2.9749e-03) (hash(x)=84662792)
-950 train 4.478994 (lr=2.9726e-03) (hash(x)=77520125)
-960 train 4.633759 (lr=2.9703e-03) (hash(x)=83412876)
-970 train 4.579736 (lr=2.9678e-03) (hash(x)=76679743)
-980 train 4.587318 (lr=2.9652e-03) (hash(x)=83578850)
-990 train 4.572951 (lr=2.9626e-03) (hash(x)=73508305)
-1000 val loss 4.5702
-1000 val perplexity 96.5622
-1000 train 4.582454 (lr=2.9598e-03) (hash(x)=73551731)
-1010 train 4.424799 (lr=2.9570e-03) (hash(x)=75472577)
-1020 train 4.474903 (lr=2.9540e-03) (hash(x)=74288773)
-1030 train 4.528303 (lr=2.9510e-03) (hash(x)=74358518)
-1040 train 4.477387 (lr=2.9478e-03) (hash(x)=81742445)
-1050 train 4.333405 (lr=2.9446e-03) (hash(x)=84334093)
-1060 train 4.329210 (lr=2.9412e-03) (hash(x)=83112407)
-1070 train 4.466598 (lr=2.9378e-03) (hash(x)=75361341)
-1080 train 4.339847 (lr=2.9343e-03) (hash(x)=80067829)
-1090 train 4.340133 (lr=2.9307e-03) (hash(x)=75084510)
-1100 val loss 4.5183
-1100 val perplexity 91.6806
-1100 train 4.504573 (lr=2.9270e-03) (hash(x)=81272082)
-1110 train 4.494093 (lr=2.9231e-03) (hash(x)=79636180)
-1120 train 4.456762 (lr=2.9192e-03) (hash(x)=106787061)
-1130 train 4.534008 (lr=2.9152e-03) (hash(x)=82157433)
-1140 train 4.488739 (lr=2.9112e-03) (hash(x)=74242006)
-1150 train 4.529262 (lr=2.9070e-03) (hash(x)=77502116)
-1160 train 4.475596 (lr=2.9027e-03) (hash(x)=78658526)
-1170 train 4.488718 (lr=2.8983e-03) (hash(x)=88605653)
-1180 train 4.425512 (lr=2.8939e-03) (hash(x)=75715872)
-1190 train 4.390070 (lr=2.8893e-03) (hash(x)=85140216)
-1200 val loss 4.4382
-1200 val perplexity 84.6258
-1200 train 4.433845 (lr=2.8847e-03) (hash(x)=76843977)
-1210 train 4.431321 (lr=2.8800e-03) (hash(x)=74668022)
-1220 train 4.364676 (lr=2.8751e-03) (hash(x)=73771388)
-1230 train 4.309374 (lr=2.8702e-03) (hash(x)=79549890)
-1240 train 4.411256 (lr=2.8652e-03) (hash(x)=80768426)
-1250 train 4.281186 (lr=2.8601e-03) (hash(x)=78485732)
-1260 train 4.223062 (lr=2.8550e-03) (hash(x)=75297547)
-1270 train 4.250462 (lr=2.8497e-03) (hash(x)=80488390)
-1280 train 4.557475 (lr=2.8443e-03) (hash(x)=83986686)
-1290 train 4.512302 (lr=2.8389e-03) (hash(x)=80260787)
-1300 val loss 4.4099
-1300 val perplexity 82.2619
-1300 train 4.532671 (lr=2.8333e-03) (hash(x)=91333241)
-1310 train 4.376293 (lr=2.8277e-03) (hash(x)=76196195)
-1320 train 4.351844 (lr=2.8220e-03) (hash(x)=89902604)
-1330 train 4.417183 (lr=2.8162e-03) (hash(x)=80055610)
-1340 train 4.402050 (lr=2.8103e-03) (hash(x)=75933940)
-1350 train 4.376811 (lr=2.8044e-03) (hash(x)=82880341)
-1360 train 4.365082 (lr=2.7983e-03) (hash(x)=74279731)
-1370 train 4.342416 (lr=2.7922e-03) (hash(x)=80381849)
-1380 train 4.428934 (lr=2.7860e-03) (hash(x)=80656732)
-1390 train 4.231915 (lr=2.7797e-03) (hash(x)=75892076)
-1400 val loss 4.3882
-1400 val perplexity 80.4990
-1400 train 4.292143 (lr=2.7733e-03) (hash(x)=85316269)
-1410 train 4.262203 (lr=2.7668e-03) (hash(x)=76176954)
-1420 train 4.106807 (lr=2.7603e-03) (hash(x)=82873585)
-1430 train 4.173633 (lr=2.7536e-03) (hash(x)=76423569)
-1440 train 4.238554 (lr=2.7469e-03) (hash(x)=77452162)
-1450 train 4.308330 (lr=2.7401e-03) (hash(x)=81479795)
-1460 train 4.076609 (lr=2.7333e-03) (hash(x)=83885401)
-1470 train 4.286392 (lr=2.7263e-03) (hash(x)=88229400)
-1480 train 4.406761 (lr=2.7193e-03) (hash(x)=80388539)
-1490 train 4.359925 (lr=2.7121e-03) (hash(x)=76421482)
-1500 val loss 4.3446
-1500 val perplexity 77.0579
-1500 train 4.304688 (lr=2.7050e-03) (hash(x)=82020652)
-1510 train 4.311339 (lr=2.6977e-03) (hash(x)=80450641)
-1520 train 4.246060 (lr=2.6903e-03) (hash(x)=80425758)
-1530 train 4.310731 (lr=2.6829e-03) (hash(x)=77928744)
-1540 train 4.269304 (lr=2.6754e-03) (hash(x)=81682588)
-1550 train 4.425693 (lr=2.6678e-03) (hash(x)=76218138)
-1560 train 4.254910 (lr=2.6602e-03) (hash(x)=82961366)
-1570 train 4.127914 (lr=2.6525e-03) (hash(x)=80344260)
-1580 train 4.218354 (lr=2.6447e-03) (hash(x)=81840773)
-1590 train 4.114624 (lr=2.6368e-03) (hash(x)=79596457)
-1600 val loss 4.3247
-1600 val perplexity 75.5454
-1600 train 4.160459 (lr=2.6289e-03) (hash(x)=78425415)
-1610 train 4.296721 (lr=2.6208e-03) (hash(x)=80735884)
-1620 train 4.253926 (lr=2.6128e-03) (hash(x)=82867340)
-1630 train 4.295040 (lr=2.6046e-03) (hash(x)=81152826)
-1640 train 4.326159 (lr=2.5964e-03) (hash(x)=79292243)
-1650 train 4.219516 (lr=2.5881e-03) (hash(x)=85633419)
-1660 train 4.227474 (lr=2.5797e-03) (hash(x)=56879214)
-1670 train 4.204163 (lr=2.5713e-03) (hash(x)=78580566)
-1680 train 4.140419 (lr=2.5628e-03) (hash(x)=73904044)
-1690 train 4.169278 (lr=2.5542e-03) (hash(x)=80456616)
-1700 val loss 4.2758
-1700 val perplexity 71.9406
-1700 train 4.166926 (lr=2.5455e-03) (hash(x)=83453850)
-1710 train 4.168399 (lr=2.5368e-03) (hash(x)=81246482)
-1720 train 4.167418 (lr=2.5281e-03) (hash(x)=74302931)
-1730 train 4.190975 (lr=2.5192e-03) (hash(x)=76366637)
-1740 train 4.121256 (lr=2.5103e-03) (hash(x)=113414000)
-1750 train 4.145040 (lr=2.5014e-03) (hash(x)=80537001)
-1760 train 4.263835 (lr=2.4924e-03) (hash(x)=80610857)
-1770 train 4.331417 (lr=2.4833e-03) (hash(x)=89782327)
-1780 train 4.255734 (lr=2.4741e-03) (hash(x)=82999075)
-1790 train 4.304556 (lr=2.4649e-03) (hash(x)=82486297)
-1800 val loss 4.2690
-1800 val perplexity 71.4484
-1800 train 4.265507 (lr=2.4556e-03) (hash(x)=79159575)
-1810 train 4.267162 (lr=2.4463e-03) (hash(x)=82428296)
-1820 train 4.247305 (lr=2.4369e-03) (hash(x)=72754716)
-1830 train 4.147896 (lr=2.4275e-03) (hash(x)=66172446)
-1840 train 4.221492 (lr=2.4180e-03) (hash(x)=79104767)
-1850 train 4.231563 (lr=2.4084e-03) (hash(x)=71073869)
-1860 train 4.115241 (lr=2.3988e-03) (hash(x)=76923769)
-1870 train 4.223158 (lr=2.3891e-03) (hash(x)=77628946)
-1880 train 4.122018 (lr=2.3794e-03) (hash(x)=87651837)
-1890 train 4.223927 (lr=2.3696e-03) (hash(x)=79373565)
-1900 val loss 4.2473
-1900 val perplexity 69.9140
-1900 train 4.290255 (lr=2.3598e-03) (hash(x)=86194345)
-1910 train 4.220722 (lr=2.3499e-03) (hash(x)=80865794)
-1920 train 4.187029 (lr=2.3400e-03) (hash(x)=79368476)
-1930 train 4.189737 (lr=2.3300e-03) (hash(x)=79247868)
-1940 train 4.136442 (lr=2.3200e-03) (hash(x)=72845520)
-1950 train 4.153883 (lr=2.3099e-03) (hash(x)=83254637)
-1960 train 4.095679 (lr=2.2998e-03) (hash(x)=81975531)
-1970 train 4.189250 (lr=2.2896e-03) (hash(x)=72939168)
-1980 train 4.081371 (lr=2.2793e-03) (hash(x)=82280940)
-1990 train 4.066816 (lr=2.2691e-03) (hash(x)=78119475)
-2000 val loss 4.2258
-2000 val perplexity 68.4315
-2000 train 4.100414 (lr=2.2588e-03) (hash(x)=83046712)
-2010 train 4.154689 (lr=2.2484e-03) (hash(x)=73468195)
-2020 train 4.094000 (lr=2.2380e-03) (hash(x)=74809817)
-2030 train 4.083929 (lr=2.2275e-03) (hash(x)=76037020)
-2040 train 4.312176 (lr=2.2170e-03) (hash(x)=79058361)
-2050 train 4.197538 (lr=2.2065e-03) (hash(x)=85014027)
-2060 train 4.182892 (lr=2.1959e-03) (hash(x)=78616227)
-2070 train 4.153384 (lr=2.1853e-03) (hash(x)=75928855)
-2080 train 4.247360 (lr=2.1746e-03) (hash(x)=87553274)
-2090 train 4.154814 (lr=2.1639e-03) (hash(x)=79805596)
-2100 val loss 4.2046
-2100 val perplexity 66.9959
-2100 train 4.023323 (lr=2.1532e-03) (hash(x)=71523881)
-2110 train 4.164555 (lr=2.1424e-03) (hash(x)=78431736)
-2120 train 4.059747 (lr=2.1316e-03) (hash(x)=76846294)
-2130 train 4.093466 (lr=2.1208e-03) (hash(x)=72808367)
-2140 train 3.909236 (lr=2.1099e-03) (hash(x)=71871519)
-2150 train 4.037587 (lr=2.0990e-03) (hash(x)=79598530)
-2160 train 4.098495 (lr=2.0881e-03) (hash(x)=81709468)
-2170 train 4.072194 (lr=2.0771e-03) (hash(x)=74075345)
-2180 train 4.226996 (lr=2.0661e-03) (hash(x)=79292886)
-2190 train 4.165350 (lr=2.0550e-03) (hash(x)=84316039)
-2200 val loss 4.1946
-2200 val perplexity 66.3280
-2200 train 4.236015 (lr=2.0440e-03) (hash(x)=84083188)
-2210 train 4.157927 (lr=2.0329e-03) (hash(x)=87482474)
-2220 train 4.252763 (lr=2.0217e-03) (hash(x)=73018507)
-2230 train 4.136269 (lr=2.0106e-03) (hash(x)=80194063)
-2240 train 4.042352 (lr=1.9994e-03) (hash(x)=70761174)
-2250 train 4.098277 (lr=1.9882e-03) (hash(x)=78296069)
-2260 train 4.069524 (lr=1.9770e-03) (hash(x)=71486863)
-2270 train 4.081946 (lr=1.9657e-03) (hash(x)=73106274)
-2280 train 4.033814 (lr=1.9544e-03) (hash(x)=65724504)
-2290 train 4.154279 (lr=1.9431e-03) (hash(x)=77464207)
-2300 val loss 4.1890
-2300 val perplexity 65.9539
-2300 train 4.297091 (lr=1.9318e-03) (hash(x)=79718531)
-2310 train 4.166912 (lr=1.9205e-03) (hash(x)=79578564)
-2320 train 4.088995 (lr=1.9091e-03) (hash(x)=77538251)
-2330 train 4.202308 (lr=1.8977e-03) (hash(x)=77684942)
-2340 train 4.166969 (lr=1.8863e-03) (hash(x)=74747888)
-2350 train 4.108972 (lr=1.8749e-03) (hash(x)=88812932)
-2360 train 4.098439 (lr=1.8635e-03) (hash(x)=77049145)
-2370 train 3.929385 (lr=1.8520e-03) (hash(x)=76636845)
-2380 train 4.004626 (lr=1.8406e-03) (hash(x)=79589860)
-2390 train 4.035613 (lr=1.8291e-03) (hash(x)=73694984)
-2400 val loss 4.1554
-2400 val perplexity 63.7763
-2400 train 4.084911 (lr=1.8176e-03) (hash(x)=76483417)
-2410 train 4.088894 (lr=1.8061e-03) (hash(x)=81534661)
-2420 train 4.170053 (lr=1.7946e-03) (hash(x)=90012063)
-2430 train 4.099506 (lr=1.7830e-03) (hash(x)=78359537)
-2440 train 4.204530 (lr=1.7715e-03) (hash(x)=79027074)
-2450 train 4.205441 (lr=1.7600e-03) (hash(x)=68888678)
-2460 train 4.064332 (lr=1.7484e-03) (hash(x)=78596234)
-2470 train 4.029956 (lr=1.7368e-03) (hash(x)=78240870)
-2480 train 4.107719 (lr=1.7253e-03) (hash(x)=78765106)
-2490 train 4.122521 (lr=1.7137e-03) (hash(x)=80888122)
-2500 val loss 4.1370
-2500 val perplexity 62.6158
-2500 train 3.868010 (lr=1.7021e-03) (hash(x)=77282980)
-2510 train 3.991797 (lr=1.6906e-03) (hash(x)=73932499)
-2520 train 3.893639 (lr=1.6790e-03) (hash(x)=78258721)
-2530 train 4.061619 (lr=1.6674e-03) (hash(x)=84237884)
-2540 train 3.980431 (lr=1.6558e-03) (hash(x)=82107056)
-2550 train 4.304190 (lr=1.6442e-03) (hash(x)=77739889)
-2560 train 4.305715 (lr=1.6326e-03) (hash(x)=74186552)
-2570 train 4.132262 (lr=1.6210e-03) (hash(x)=80473388)
-2580 train 4.048244 (lr=1.6094e-03) (hash(x)=82259419)
-2590 train 4.133770 (lr=1.5979e-03) (hash(x)=78413433)
-2600 val loss 4.1197
-2600 val perplexity 61.5383
-2600 train 4.045895 (lr=1.5863e-03) (hash(x)=74502607)
-2610 train 4.043626 (lr=1.5747e-03) (hash(x)=76291308)
-2620 train 4.007569 (lr=1.5632e-03) (hash(x)=89544859)
-2630 train 4.006664 (lr=1.5516e-03) (hash(x)=72230079)
-2640 train 4.032544 (lr=1.5400e-03) (hash(x)=80712391)
-2650 train 3.983006 (lr=1.5285e-03) (hash(x)=81110141)
-2660 train 4.059503 (lr=1.5170e-03) (hash(x)=73706547)
-2670 train 3.973346 (lr=1.5054e-03) (hash(x)=75978602)
-2680 train 4.048762 (lr=1.4939e-03) (hash(x)=77181602)
-2690 train 3.961176 (lr=1.4824e-03) (hash(x)=73503073)
-2700 val loss 4.1081
-2700 val perplexity 60.8335
-2700 train 4.015653 (lr=1.4709e-03) (hash(x)=76450620)
-2710 train 4.104738 (lr=1.4594e-03) (hash(x)=82384092)
-2720 train 4.094800 (lr=1.4480e-03) (hash(x)=78992460)
-2730 train 4.096491 (lr=1.4365e-03) (hash(x)=89640628)
-2740 train 4.068956 (lr=1.4251e-03) (hash(x)=78958333)
-2750 train 4.098755 (lr=1.4137e-03) (hash(x)=75231323)
-2760 train 4.049979 (lr=1.4023e-03) (hash(x)=80141789)
-2770 train 3.971216 (lr=1.3909e-03) (hash(x)=78041972)
-2780 train 4.094885 (lr=1.3795e-03) (hash(x)=78824636)
-2790 train 3.914609 (lr=1.3682e-03) (hash(x)=88495681)
-2800 val loss 4.0907
-2800 val perplexity 59.7837
-2800 train 3.968028 (lr=1.3569e-03) (hash(x)=70894085)
-2810 train 3.905442 (lr=1.3456e-03) (hash(x)=80942795)
-2820 train 3.923177 (lr=1.3343e-03) (hash(x)=80086512)
-2830 train 3.899423 (lr=1.3230e-03) (hash(x)=76162076)
-2840 train 4.040043 (lr=1.3118e-03) (hash(x)=92909437)
-2850 train 4.177738 (lr=1.3006e-03) (hash(x)=82192377)
-2860 train 4.001718 (lr=1.2894e-03) (hash(x)=79981204)
-2870 train 4.071570 (lr=1.2783e-03) (hash(x)=86642658)
-2880 train 4.016573 (lr=1.2671e-03) (hash(x)=73934349)
-2890 train 4.124953 (lr=1.2560e-03) (hash(x)=81104933)
-2900 val loss 4.0725
-2900 val perplexity 58.7020
-2900 train 4.000626 (lr=1.2450e-03) (hash(x)=83259335)
-2910 train 4.034023 (lr=1.2339e-03) (hash(x)=76528489)
-2920 train 4.024301 (lr=1.2229e-03) (hash(x)=82547590)
-2930 train 3.871829 (lr=1.2119e-03) (hash(x)=84692438)
-2940 train 3.898985 (lr=1.2010e-03) (hash(x)=76395062)
-2950 train 3.899982 (lr=1.1901e-03) (hash(x)=79351043)
-2960 train 3.981853 (lr=1.1792e-03) (hash(x)=75184745)
-2970 train 3.915520 (lr=1.1684e-03) (hash(x)=73888960)
-2980 train 4.110643 (lr=1.1576e-03) (hash(x)=80237171)
-2990 train 4.062630 (lr=1.1468e-03) (hash(x)=80090496)
-3000 val loss 4.0724
-3000 val perplexity 58.6951
-3000 train 4.115636 (lr=1.1361e-03) (hash(x)=99715255)
-3010 train 4.065180 (lr=1.1254e-03) (hash(x)=73302575)
-3020 train 3.964556 (lr=1.1147e-03) (hash(x)=81162327)
-3030 train 4.010807 (lr=1.1041e-03) (hash(x)=78686601)
-3040 train 4.030081 (lr=1.0935e-03) (hash(x)=76186508)
-3050 train 4.149834 (lr=1.0830e-03) (hash(x)=94420278)
-3060 train 4.042452 (lr=1.0725e-03) (hash(x)=73930563)
-3070 train 4.074058 (lr=1.0620e-03) (hash(x)=78074700)
-3080 train 3.920597 (lr=1.0516e-03) (hash(x)=85521020)
-3090 train 3.978158 (lr=1.0412e-03) (hash(x)=77429090)
-3100 val loss 4.0463
-3100 val perplexity 57.1875
-3100 train 3.955230 (lr=1.0309e-03) (hash(x)=74319016)
-3110 train 3.963254 (lr=1.0207e-03) (hash(x)=81856930)
-3120 train 3.976446 (lr=1.0104e-03) (hash(x)=78854668)
-3130 train 3.818309 (lr=1.0002e-03) (hash(x)=78558831)
-3140 train 3.765047 (lr=9.9011e-04) (hash(x)=75419854)
-3150 train 3.756888 (lr=9.8002e-04) (hash(x)=72645078)
-3160 train 3.931385 (lr=9.6999e-04) (hash(x)=71954229)
-3170 train 3.859103 (lr=9.6000e-04) (hash(x)=83944062)
-3180 train 4.154417 (lr=9.5007e-04) (hash(x)=81719676)
-3190 train 4.136951 (lr=9.4019e-04) (hash(x)=78117157)
-3200 val loss 4.0278
-3200 val perplexity 56.1352
-3200 train 4.104502 (lr=9.3036e-04) (hash(x)=95062223)
-3210 train 3.987017 (lr=9.2058e-04) (hash(x)=78243969)
-3220 train 3.998593 (lr=9.1085e-04) (hash(x)=80840117)
-3230 train 4.076189 (lr=9.0118e-04) (hash(x)=89040111)
-3240 train 3.949558 (lr=8.9157e-04) (hash(x)=78018689)
-3250 train 4.023832 (lr=8.8201e-04) (hash(x)=71514679)
-3260 train 3.960479 (lr=8.7251e-04) (hash(x)=72142220)
-3270 train 3.906598 (lr=8.6307e-04) (hash(x)=81981079)
-3280 train 4.044951 (lr=8.5368e-04) (hash(x)=81221688)
-3290 train 3.934371 (lr=8.4435e-04) (hash(x)=84425616)
-3300 val loss 4.0132
-3300 val perplexity 55.3244
-3300 train 3.951565 (lr=8.3508e-04) (hash(x)=85200306)
-3310 train 3.930038 (lr=8.2588e-04) (hash(x)=90538542)
-3320 train 3.799381 (lr=8.1673e-04) (hash(x)=75045669)
-3330 train 3.704250 (lr=8.0764e-04) (hash(x)=91874178)
-3340 train 3.724725 (lr=7.9862e-04) (hash(x)=93115709)
-3350 train 3.747840 (lr=7.8966e-04) (hash(x)=80614131)
-3360 train 3.776483 (lr=7.8076e-04) (hash(x)=75966518)
-3370 train 3.942496 (lr=7.7192e-04) (hash(x)=52504673)
-3380 train 3.995148 (lr=7.6315e-04) (hash(x)=89845158)
-3390 train 4.563801 (lr=7.5445e-04) (hash(x)=84328133)
-3400 val loss 4.0092
-3400 val perplexity 55.1030
-3400 train 4.059773 (lr=7.4581e-04) (hash(x)=84410966)
-3410 train 4.000085 (lr=7.3724e-04) (hash(x)=81040312)
-3420 train 4.123499 (lr=7.2874e-04) (hash(x)=80573120)
-3430 train 3.902299 (lr=7.2030e-04) (hash(x)=77676972)
-3440 train 3.966395 (lr=7.1193e-04) (hash(x)=86131899)
-3450 train 3.981323 (lr=7.0363e-04) (hash(x)=75051359)
-3460 train 3.920019 (lr=6.9541e-04) (hash(x)=78319212)
-3470 train 3.970574 (lr=6.8725e-04) (hash(x)=75514502)
-3480 train 3.834807 (lr=6.7916e-04) (hash(x)=84577433)
-3490 train 4.004914 (lr=6.7114e-04) (hash(x)=76740859)
+0 val loss 11.3281
+0 val perplexity 83128.3594
+0 train 11.329872 (lr=4.1958e-06) (hash(x)=92476531)
+10 train 9.885843 (lr=4.6154e-05) (hash(x)=80557183)
+20 train 9.346017 (lr=8.8112e-05) (hash(x)=77833840)
+30 train 8.609160 (lr=1.3007e-04) (hash(x)=80219380)
+40 train 7.899735 (lr=1.7203e-04) (hash(x)=81242310)
+50 train 7.627817 (lr=2.1399e-04) (hash(x)=82639280)
+60 train 7.547350 (lr=2.5594e-04) (hash(x)=75412499)
+70 train 7.563859 (lr=2.9790e-04) (hash(x)=87943396)
+80 train 7.461051 (lr=3.3986e-04) (hash(x)=85407935)
+90 train 7.217012 (lr=3.8182e-04) (hash(x)=79447147)
+100 val loss 7.1239
+100 val perplexity 1241.3134
+100 train 7.082138 (lr=4.2378e-04) (hash(x)=80787570)
+110 train 6.968734 (lr=4.6573e-04) (hash(x)=79791376)
+120 train 6.846392 (lr=5.0769e-04) (hash(x)=79296900)
+130 train 6.709699 (lr=5.4965e-04) (hash(x)=79174525)
+140 train 6.573351 (lr=5.9161e-04) (hash(x)=83957686)
+150 train 6.508861 (lr=6.3357e-04) (hash(x)=77592594)
+160 train 6.390038 (lr=6.7552e-04) (hash(x)=70564809)
+170 train 6.457429 (lr=7.1748e-04) (hash(x)=72584980)
+180 train 6.477027 (lr=7.5944e-04) (hash(x)=95506562)
+190 train 6.206078 (lr=8.0140e-04) (hash(x)=73554383)
+200 val loss 6.2165
+200 val perplexity 500.9544
+200 train 6.149079 (lr=8.4336e-04) (hash(x)=84992944)
+210 train 6.062893 (lr=8.8531e-04) (hash(x)=84224321)
+220 train 5.998404 (lr=9.2727e-04) (hash(x)=91374190)
+230 train 5.999579 (lr=9.6923e-04) (hash(x)=85276683)
+240 train 6.055002 (lr=1.0112e-03) (hash(x)=81723155)
+250 train 6.007201 (lr=1.0531e-03) (hash(x)=75382743)
+260 train 5.950223 (lr=1.0951e-03) (hash(x)=79107828)
+270 train 5.833077 (lr=1.1371e-03) (hash(x)=63566375)
+280 train 5.788705 (lr=1.1790e-03) (hash(x)=72509291)
+290 train 5.737301 (lr=1.2210e-03) (hash(x)=80854173)
+300 val loss 5.7537
+300 val perplexity 315.3483
+300 train 5.743345 (lr=1.2629e-03) (hash(x)=95549669)
+310 train 5.597397 (lr=1.3049e-03) (hash(x)=82435860)
+320 train 5.668325 (lr=1.3469e-03) (hash(x)=79528901)
+330 train 5.599624 (lr=1.3888e-03) (hash(x)=89988570)
+340 train 5.596200 (lr=1.4308e-03) (hash(x)=81102378)
+350 train 5.497179 (lr=1.4727e-03) (hash(x)=75258538)
+360 train 5.457752 (lr=1.5147e-03) (hash(x)=75080686)
+370 train 5.495667 (lr=1.5566e-03) (hash(x)=87436781)
+380 train 5.468419 (lr=1.5986e-03) (hash(x)=78319003)
+390 train 5.430259 (lr=1.6406e-03) (hash(x)=80258231)
+400 val loss 5.4636
+400 val perplexity 235.9404
+400 train 5.500962 (lr=1.6825e-03) (hash(x)=93677302)
+410 train 5.438931 (lr=1.7245e-03) (hash(x)=83320730)
+420 train 5.416919 (lr=1.7664e-03) (hash(x)=88090468)
+430 train 5.401352 (lr=1.8084e-03) (hash(x)=77168020)
+440 train 5.385225 (lr=1.8503e-03) (hash(x)=75302477)
+450 train 5.317851 (lr=1.8923e-03) (hash(x)=87839104)
+460 train 5.403702 (lr=1.9343e-03) (hash(x)=84600562)
+470 train 5.268123 (lr=1.9762e-03) (hash(x)=72223964)
+480 train 5.183892 (lr=2.0182e-03) (hash(x)=85410838)
+490 train 5.270505 (lr=2.0601e-03) (hash(x)=75686864)
+500 val loss 5.2662
+500 val perplexity 193.6834
+500 train 5.155718 (lr=2.1021e-03) (hash(x)=82209970)
+510 train 5.169087 (lr=2.1441e-03) (hash(x)=78903986)
+520 train 5.221563 (lr=2.1860e-03) (hash(x)=82788367)
+530 train 5.183284 (lr=2.2280e-03) (hash(x)=79014660)
+540 train 5.208410 (lr=2.2699e-03) (hash(x)=78203384)
+550 train 5.058145 (lr=2.3119e-03) (hash(x)=74749939)
+560 train 5.095774 (lr=2.3538e-03) (hash(x)=73863798)
+570 train 5.155106 (lr=2.3958e-03) (hash(x)=81655345)
+580 train 5.103041 (lr=2.4378e-03) (hash(x)=77989048)
+590 train 5.088838 (lr=2.4797e-03) (hash(x)=75240510)
+600 val loss 5.0562
+600 val perplexity 156.9969
+600 train 5.053777 (lr=2.5217e-03) (hash(x)=75393302)
+610 train 5.385088 (lr=2.5636e-03) (hash(x)=75805578)
+620 train 4.915320 (lr=2.6056e-03) (hash(x)=76469956)
+630 train 4.994261 (lr=2.6476e-03) (hash(x)=80862833)
+640 train 4.925799 (lr=2.6895e-03) (hash(x)=69614120)
+650 train 4.918322 (lr=2.7315e-03) (hash(x)=81753733)
+660 train 4.951305 (lr=2.7734e-03) (hash(x)=80638528)
+670 train 4.943464 (lr=2.8154e-03) (hash(x)=82087325)
+680 train 5.039653 (lr=2.8573e-03) (hash(x)=83366526)
+690 train 4.842476 (lr=2.8993e-03) (hash(x)=79702542)
+700 val loss 4.8776
+700 val perplexity 131.3186
+700 train 4.782776 (lr=2.9413e-03) (hash(x)=67691907)
+710 train 4.982585 (lr=2.9832e-03) (hash(x)=78194053)
+720 train 4.797255 (lr=3.0000e-03) (hash(x)=76369890)
+730 train 4.789774 (lr=2.9999e-03) (hash(x)=76486285)
+740 train 4.746025 (lr=2.9997e-03) (hash(x)=73198975)
+750 train 4.763426 (lr=2.9994e-03) (hash(x)=72993983)
+760 train 4.668073 (lr=2.9990e-03) (hash(x)=88306535)
+770 train 4.683599 (lr=2.9985e-03) (hash(x)=87649303)
+780 train 4.542311 (lr=2.9979e-03) (hash(x)=80261986)
+790 train 4.531882 (lr=2.9972e-03) (hash(x)=84518434)
+800 val loss 4.7750
+800 val perplexity 118.5102
+800 train 4.501464 (lr=2.9964e-03) (hash(x)=77880126)
+810 train 4.711572 (lr=2.9955e-03) (hash(x)=91800941)
+820 train 4.586892 (lr=2.9945e-03) (hash(x)=77438844)
+830 train 4.775039 (lr=2.9934e-03) (hash(x)=84295123)
+840 train 4.618631 (lr=2.9922e-03) (hash(x)=82587990)
+850 train 4.725470 (lr=2.9909e-03) (hash(x)=76145087)
+860 train 4.636567 (lr=2.9896e-03) (hash(x)=80380499)
+870 train 4.621845 (lr=2.9881e-03) (hash(x)=79852152)
+880 train 4.652238 (lr=2.9865e-03) (hash(x)=80506089)
+890 train 4.548776 (lr=2.9848e-03) (hash(x)=74969638)
+900 val loss 4.6333
+900 val perplexity 102.8558
+900 train 4.602500 (lr=2.9830e-03) (hash(x)=72563497)
+910 train 4.504281 (lr=2.9811e-03) (hash(x)=79240775)
+920 train 4.543964 (lr=2.9792e-03) (hash(x)=81010967)
+930 train 4.549799 (lr=2.9771e-03) (hash(x)=76027759)
+940 train 4.570710 (lr=2.9749e-03) (hash(x)=73716762)
+950 train 4.495275 (lr=2.9726e-03) (hash(x)=79940439)
+960 train 4.534607 (lr=2.9703e-03) (hash(x)=88318293)
+970 train 4.364140 (lr=2.9678e-03) (hash(x)=80583440)
+980 train 4.366107 (lr=2.9652e-03) (hash(x)=79458899)
+990 train 4.433765 (lr=2.9626e-03) (hash(x)=79258146)
+1000 val loss 4.5928
+1000 val perplexity 98.7741
+1000 train 4.799899 (lr=2.9598e-03) (hash(x)=78128159)
+1010 train 4.634326 (lr=2.9570e-03) (hash(x)=81340506)
+1020 train 4.652112 (lr=2.9540e-03) (hash(x)=94839111)
+1030 train 4.536103 (lr=2.9510e-03) (hash(x)=79238631)
+1040 train 4.525444 (lr=2.9478e-03) (hash(x)=87577788)
+1050 train 4.544426 (lr=2.9446e-03) (hash(x)=78297110)
+1060 train 4.476863 (lr=2.9412e-03) (hash(x)=72337518)
+1070 train 4.482951 (lr=2.9378e-03) (hash(x)=72814742)
+1080 train 4.511830 (lr=2.9343e-03) (hash(x)=75722002)
+1090 train 4.455788 (lr=2.9307e-03) (hash(x)=78305141)
+1100 val loss 4.5082
+1100 val perplexity 90.7581
+1100 train 4.409119 (lr=2.9270e-03) (hash(x)=79663017)
+1110 train 4.510953 (lr=2.9231e-03) (hash(x)=104050146)
+1120 train 4.467209 (lr=2.9192e-03) (hash(x)=67379529)
+1130 train 4.374723 (lr=2.9152e-03) (hash(x)=75589492)
+1140 train 4.357923 (lr=2.9112e-03) (hash(x)=77978407)
+1150 train 4.330865 (lr=2.9070e-03) (hash(x)=79982444)
+1160 train 4.202845 (lr=2.9027e-03) (hash(x)=77038480)
+1170 train 4.202691 (lr=2.8983e-03) (hash(x)=84041157)
+1180 train 4.201767 (lr=2.8939e-03) (hash(x)=85344564)
+1190 train 4.431473 (lr=2.8893e-03) (hash(x)=82018589)
+1200 val loss 4.4710
+1200 val perplexity 87.4456
+1200 train 4.476369 (lr=2.8847e-03) (hash(x)=83172541)
+1210 train 4.428150 (lr=2.8800e-03) (hash(x)=76704756)
+1220 train 4.315696 (lr=2.8751e-03) (hash(x)=80484892)
+1230 train 4.407027 (lr=2.8702e-03) (hash(x)=80724246)
+1240 train 4.513501 (lr=2.8652e-03) (hash(x)=82844155)
+1250 train 4.392875 (lr=2.8601e-03) (hash(x)=75005983)
+1260 train 4.374195 (lr=2.8550e-03) (hash(x)=76014668)
+1270 train 4.369526 (lr=2.8497e-03) (hash(x)=74062742)
+1280 train 4.465700 (lr=2.8443e-03) (hash(x)=83893278)
+1290 train 4.329505 (lr=2.8389e-03) (hash(x)=82235715)
+1300 val loss 4.4129
+1300 val perplexity 82.5089
+1300 train 4.374656 (lr=2.8333e-03) (hash(x)=79195156)
+1310 train 4.457879 (lr=2.8277e-03) (hash(x)=79036954)
+1320 train 4.262702 (lr=2.8220e-03) (hash(x)=85008455)
+1330 train 4.232652 (lr=2.8162e-03) (hash(x)=75114081)
+1340 train 4.130755 (lr=2.8103e-03) (hash(x)=82306787)
+1350 train 4.269059 (lr=2.8044e-03) (hash(x)=85112053)
+1360 train 4.127003 (lr=2.7983e-03) (hash(x)=82963845)
+1370 train 4.115549 (lr=2.7922e-03) (hash(x)=85030982)
+1380 train 4.364270 (lr=2.7860e-03) (hash(x)=93735902)
+1390 train 4.315973 (lr=2.7797e-03) (hash(x)=89011154)
+1400 val loss 4.3636
+1400 val perplexity 78.5408
+1400 train 4.302365 (lr=2.7733e-03) (hash(x)=82085703)
+1410 train 4.621972 (lr=2.7668e-03) (hash(x)=90334911)
+1420 train 4.387644 (lr=2.7603e-03) (hash(x)=83787797)
+1430 train 4.278757 (lr=2.7536e-03) (hash(x)=82084098)
+1440 train 4.295527 (lr=2.7469e-03) (hash(x)=77539484)
+1450 train 4.392406 (lr=2.7401e-03) (hash(x)=77887047)
+1460 train 4.342640 (lr=2.7333e-03) (hash(x)=81880305)
+1470 train 4.314835 (lr=2.7263e-03) (hash(x)=75176428)
+1480 train 4.316839 (lr=2.7193e-03) (hash(x)=83835098)
+1490 train 4.264648 (lr=2.7121e-03) (hash(x)=90482123)
+1500 val loss 4.3419
+1500 val perplexity 76.8517
+1500 train 4.230110 (lr=2.7050e-03) (hash(x)=78462195)
+1510 train 4.331771 (lr=2.6977e-03) (hash(x)=78628581)
+1520 train 4.120641 (lr=2.6903e-03) (hash(x)=75543035)
+1530 train 4.372730 (lr=2.6829e-03) (hash(x)=80858373)
+1540 train 4.324959 (lr=2.6754e-03) (hash(x)=75358272)
+1550 train 4.219049 (lr=2.6678e-03) (hash(x)=81074965)
+1560 train 4.251229 (lr=2.6602e-03) (hash(x)=75919017)
+1570 train 4.277203 (lr=2.6525e-03) (hash(x)=78449050)
+1580 train 4.238515 (lr=2.6447e-03) (hash(x)=80179401)
+1590 train 4.207901 (lr=2.6368e-03) (hash(x)=78201723)
+1600 val loss 4.3179
+1600 val perplexity 75.0308
+1600 train 4.215847 (lr=2.6289e-03) (hash(x)=82628634)
+1610 train 4.200640 (lr=2.6208e-03) (hash(x)=83088672)
+1620 train 4.113140 (lr=2.6128e-03) (hash(x)=86507787)
+1630 train 4.155645 (lr=2.6046e-03) (hash(x)=73078561)
+1640 train 4.099254 (lr=2.5964e-03) (hash(x)=82166841)
+1650 train 4.216335 (lr=2.5881e-03) (hash(x)=78193565)
+1660 train 4.321518 (lr=2.5797e-03) (hash(x)=85804701)
+1670 train 4.330482 (lr=2.5713e-03) (hash(x)=70345337)
+1680 train 4.270449 (lr=2.5628e-03) (hash(x)=80320580)
+1690 train 4.262575 (lr=2.5542e-03) (hash(x)=81513796)
+1700 val loss 4.2883
+1700 val perplexity 72.8438
+1700 train 4.175694 (lr=2.5455e-03) (hash(x)=82483619)
+1710 train 4.249171 (lr=2.5368e-03) (hash(x)=74499966)
+1720 train 4.156795 (lr=2.5281e-03) (hash(x)=82891474)
+1730 train 4.154605 (lr=2.5192e-03) (hash(x)=76311864)
+1740 train 4.300560 (lr=2.5103e-03) (hash(x)=78327998)
+1750 train 4.178199 (lr=2.5014e-03) (hash(x)=81274966)
+1760 train 4.182736 (lr=2.4924e-03) (hash(x)=77505483)
+1770 train 4.117759 (lr=2.4833e-03) (hash(x)=80847695)
+1780 train 4.092343 (lr=2.4741e-03) (hash(x)=70358393)
+1790 train 4.316379 (lr=2.4649e-03) (hash(x)=86191041)
+1800 val loss 4.2666
+1800 val perplexity 71.2794
+1800 train 4.327271 (lr=2.4556e-03) (hash(x)=77210920)
+1810 train 4.219414 (lr=2.4463e-03) (hash(x)=80523396)
+1820 train 4.365693 (lr=2.4369e-03) (hash(x)=87014389)
+1830 train 4.171864 (lr=2.4275e-03) (hash(x)=73990672)
+1840 train 4.276364 (lr=2.4180e-03) (hash(x)=91802507)
+1850 train 4.250656 (lr=2.4084e-03) (hash(x)=76955322)
+1860 train 4.079770 (lr=2.3988e-03) (hash(x)=77109229)
+1870 train 4.132666 (lr=2.3891e-03) (hash(x)=90135004)
+1880 train 4.101525 (lr=2.3794e-03) (hash(x)=73788366)
+1890 train 4.127709 (lr=2.3696e-03) (hash(x)=68877898)
+1900 val loss 4.2645
+1900 val perplexity 71.1268
+1900 train 4.117465 (lr=2.3598e-03) (hash(x)=78685171)
+1910 train 4.076835 (lr=2.3499e-03) (hash(x)=77401876)
+1920 train 4.287946 (lr=2.3400e-03) (hash(x)=76083249)
+1930 train 4.368454 (lr=2.3300e-03) (hash(x)=78497419)
+1940 train 4.194337 (lr=2.3200e-03) (hash(x)=77768007)
+1950 train 4.243146 (lr=2.3099e-03) (hash(x)=78166365)
+1960 train 4.040194 (lr=2.2998e-03) (hash(x)=76255691)
+1970 train 4.121019 (lr=2.2896e-03) (hash(x)=87599964)
+1980 train 4.101088 (lr=2.2793e-03) (hash(x)=76977354)
+1990 train 4.131729 (lr=2.2691e-03) (hash(x)=75418637)
+2000 val loss 4.2224
+2000 val perplexity 68.1952
+2000 train 4.162646 (lr=2.2588e-03) (hash(x)=74505485)
+2010 train 4.140091 (lr=2.2484e-03) (hash(x)=77968902)
+2020 train 3.998617 (lr=2.2380e-03) (hash(x)=74473597)
+2030 train 4.203417 (lr=2.2275e-03) (hash(x)=74915423)
+2040 train 4.151469 (lr=2.2170e-03) (hash(x)=80532849)
+2050 train 4.043805 (lr=2.2065e-03) (hash(x)=70871513)
+2060 train 4.315804 (lr=2.1959e-03) (hash(x)=83258099)
+2070 train 4.317490 (lr=2.1853e-03) (hash(x)=79318928)
+2080 train 4.115393 (lr=2.1746e-03) (hash(x)=80046106)
+2090 train 4.137668 (lr=2.1639e-03) (hash(x)=79574366)
+2100 val loss 4.2138
+2100 val perplexity 67.6150
+2100 train 4.156605 (lr=2.1532e-03) (hash(x)=78403637)
+2110 train 4.049765 (lr=2.1424e-03) (hash(x)=83330861)
+2120 train 4.138016 (lr=2.1316e-03) (hash(x)=82018746)
+2130 train 4.069549 (lr=2.1208e-03) (hash(x)=78997551)
+2140 train 4.035898 (lr=2.1099e-03) (hash(x)=73644529)
+2150 train 4.189118 (lr=2.0990e-03) (hash(x)=80653570)
+2160 train 4.088118 (lr=2.0881e-03) (hash(x)=73393914)
+2170 train 3.988484 (lr=2.0771e-03) (hash(x)=79883496)
+2180 train 4.043263 (lr=2.0661e-03) (hash(x)=78292168)
+2190 train 4.130346 (lr=2.0550e-03) (hash(x)=78366011)
+2200 val loss 4.1938
+2200 val perplexity 66.2713
+2200 train 4.215665 (lr=2.0440e-03) (hash(x)=87877942)
+2210 train 4.171733 (lr=2.0329e-03) (hash(x)=82277941)
+2220 train 4.124633 (lr=2.0217e-03) (hash(x)=76353795)
+2230 train 4.242550 (lr=2.0106e-03) (hash(x)=84692047)
+2240 train 4.177950 (lr=1.9994e-03) (hash(x)=86440918)
+2250 train 4.138790 (lr=1.9882e-03) (hash(x)=74843751)
+2260 train 4.193878 (lr=1.9770e-03) (hash(x)=76370726)
+2270 train 4.156634 (lr=1.9657e-03) (hash(x)=81088985)
+2280 train 4.002053 (lr=1.9544e-03) (hash(x)=82596808)
+2290 train 4.045807 (lr=1.9431e-03) (hash(x)=82101185)
+2300 val loss 4.1721
+2300 val perplexity 64.8498
+2300 train 4.140386 (lr=1.9318e-03) (hash(x)=76784929)
+2310 train 4.128248 (lr=1.9205e-03) (hash(x)=79063641)
+2320 train 4.092183 (lr=1.9091e-03) (hash(x)=72997210)
+2330 train 4.146320 (lr=1.8977e-03) (hash(x)=78364076)
+2340 train 4.119704 (lr=1.8863e-03) (hash(x)=80569869)
+2350 train 4.119130 (lr=1.8749e-03) (hash(x)=82997001)
+2360 train 4.186924 (lr=1.8635e-03) (hash(x)=80217398)
+2370 train 4.204122 (lr=1.8520e-03) (hash(x)=79078550)
+2380 train 4.129663 (lr=1.8406e-03) (hash(x)=80109951)
+2390 train 4.078945 (lr=1.8291e-03) (hash(x)=81684929)
+2400 val loss 4.1537
+2400 val perplexity 63.6688
+2400 train 4.053445 (lr=1.8176e-03) (hash(x)=68751146)
+2410 train 4.096542 (lr=1.8061e-03) (hash(x)=78732720)
+2420 train 3.936222 (lr=1.7946e-03) (hash(x)=71994630)
+2430 train 4.193006 (lr=1.7830e-03) (hash(x)=77169990)
+2440 train 4.068650 (lr=1.7715e-03) (hash(x)=79305139)
+2450 train 3.914380 (lr=1.7600e-03) (hash(x)=81449659)
+2460 train 4.036596 (lr=1.7484e-03) (hash(x)=77646193)
+2470 train 4.074513 (lr=1.7368e-03) (hash(x)=82360323)
+2480 train 4.093971 (lr=1.7253e-03) (hash(x)=77493945)
+2490 train 4.212600 (lr=1.7137e-03) (hash(x)=86536175)
+2500 val loss 4.1363
+2500 val perplexity 62.5702
+2500 train 4.281656 (lr=1.7021e-03) (hash(x)=85247781)
+2510 train 4.213260 (lr=1.6906e-03) (hash(x)=68014473)
+2520 train 3.997978 (lr=1.6790e-03) (hash(x)=76322450)
+2530 train 4.028214 (lr=1.6674e-03) (hash(x)=72040116)
+2540 train 4.088657 (lr=1.6558e-03) (hash(x)=72522060)
+2550 train 4.023317 (lr=1.6442e-03) (hash(x)=75822010)
+2560 train 3.967743 (lr=1.6326e-03) (hash(x)=80473391)
+2570 train 4.052541 (lr=1.6210e-03) (hash(x)=75327173)
+2580 train 3.904749 (lr=1.6094e-03) (hash(x)=79012888)
+2590 train 3.929135 (lr=1.5979e-03) (hash(x)=90222207)
+2600 val loss 4.1260
+2600 val perplexity 61.9321
+2600 train 3.928811 (lr=1.5863e-03) (hash(x)=72605303)
+2610 train 4.044061 (lr=1.5747e-03) (hash(x)=63545259)
+2620 train 4.175566 (lr=1.5632e-03) (hash(x)=75398896)
+2630 train 4.183388 (lr=1.5516e-03) (hash(x)=76569876)
+2640 train 4.267385 (lr=1.5400e-03) (hash(x)=81208719)
+2650 train 4.068154 (lr=1.5285e-03) (hash(x)=78825850)
+2660 train 4.076336 (lr=1.5170e-03) (hash(x)=71007296)
+2670 train 4.091203 (lr=1.5054e-03) (hash(x)=85702547)
+2680 train 4.008529 (lr=1.4939e-03) (hash(x)=84007025)
+2690 train 4.044486 (lr=1.4824e-03) (hash(x)=79114893)
+2700 val loss 4.1006
+2700 val perplexity 60.3752
+2700 train 4.075078 (lr=1.4709e-03) (hash(x)=76100226)
+2710 train 3.911572 (lr=1.4594e-03) (hash(x)=95412405)
+2720 train 3.984668 (lr=1.4480e-03) (hash(x)=77968369)
+2730 train 3.993602 (lr=1.4365e-03) (hash(x)=83745566)
+2740 train 3.912494 (lr=1.4251e-03) (hash(x)=71968605)
+2750 train 3.987499 (lr=1.4137e-03) (hash(x)=78030093)
+2760 train 4.199360 (lr=1.4023e-03) (hash(x)=82034862)
+2770 train 4.007241 (lr=1.3909e-03) (hash(x)=81426905)
+2780 train 4.126659 (lr=1.3795e-03) (hash(x)=80804180)
+2790 train 4.165523 (lr=1.3682e-03) (hash(x)=78769818)
+2800 val loss 4.0833
+2800 val perplexity 59.3416
+2800 train 3.941484 (lr=1.3569e-03) (hash(x)=81546487)
+2810 train 3.988103 (lr=1.3456e-03) (hash(x)=81942293)
+2820 train 3.949957 (lr=1.3343e-03) (hash(x)=79628664)
+2830 train 4.016489 (lr=1.3230e-03) (hash(x)=78405055)
+2840 train 3.977445 (lr=1.3118e-03) (hash(x)=74847727)
+2850 train 3.932621 (lr=1.3006e-03) (hash(x)=81869030)
+2860 train 3.907102 (lr=1.2894e-03) (hash(x)=100536172)
+2870 train 3.818714 (lr=1.2783e-03) (hash(x)=75411272)
+2880 train 3.969649 (lr=1.2671e-03) (hash(x)=81382860)
+2890 train 4.118701 (lr=1.2560e-03) (hash(x)=81157461)
+2900 val loss 4.0831
+2900 val perplexity 59.3269
+2900 train 4.061646 (lr=1.2450e-03) (hash(x)=76856547)
+2910 train 4.044005 (lr=1.2339e-03) (hash(x)=76535873)
+2920 train 4.073672 (lr=1.2229e-03) (hash(x)=77761442)
+2930 train 3.965763 (lr=1.2119e-03) (hash(x)=75488021)
+2940 train 3.951296 (lr=1.2010e-03) (hash(x)=85879517)
+2950 train 4.060417 (lr=1.1901e-03) (hash(x)=79599481)
+2960 train 3.939059 (lr=1.1792e-03) (hash(x)=76820375)
+2970 train 3.926475 (lr=1.1684e-03) (hash(x)=77945910)
+2980 train 3.963491 (lr=1.1576e-03) (hash(x)=83034701)
+2990 train 3.940017 (lr=1.1468e-03) (hash(x)=75166389)
+3000 val loss 4.0590
+3000 val perplexity 57.9142
+3000 train 3.985270 (lr=1.1361e-03) (hash(x)=73461176)
+3010 train 3.863402 (lr=1.1254e-03) (hash(x)=71826799)
+3020 train 3.807011 (lr=1.1147e-03) (hash(x)=90013784)
+3030 train 3.910875 (lr=1.1041e-03) (hash(x)=68244392)
+3040 train 4.074251 (lr=1.0935e-03) (hash(x)=81477870)
+3050 train 3.994513 (lr=1.0830e-03) (hash(x)=89071978)
+3060 train 4.190904 (lr=1.0725e-03) (hash(x)=78909964)
+3070 train 4.088107 (lr=1.0620e-03) (hash(x)=72927540)
+3080 train 4.011167 (lr=1.0516e-03) (hash(x)=78576606)
+3090 train 4.095238 (lr=1.0412e-03) (hash(x)=83549010)
+3100 val loss 4.0516
+3100 val perplexity 57.4911
+3100 train 4.123398 (lr=1.0309e-03) (hash(x)=88378940)
+3110 train 3.972738 (lr=1.0207e-03) (hash(x)=75361195)
+3120 train 3.964723 (lr=1.0104e-03) (hash(x)=81548673)
+3130 train 3.947186 (lr=1.0002e-03) (hash(x)=76723731)
+3140 train 4.014853 (lr=9.9011e-04) (hash(x)=78335834)
+3150 train 3.977509 (lr=9.8002e-04) (hash(x)=79512807)
+3160 train 3.997532 (lr=9.6999e-04) (hash(x)=78022958)
+3170 train 4.093410 (lr=9.6000e-04) (hash(x)=48780915)
+3180 train 4.057344 (lr=9.5007e-04) (hash(x)=72859953)
+3190 train 4.032578 (lr=9.4019e-04) (hash(x)=70441376)
+3200 val loss 4.0248
+3200 val perplexity 55.9691
+3200 train 4.043797 (lr=9.3036e-04) (hash(x)=85361027)
+3210 train 4.121223 (lr=9.2058e-04) (hash(x)=84671249)
+3220 train 4.055424 (lr=9.1085e-04) (hash(x)=77654883)
+3230 train 3.974788 (lr=9.0118e-04) (hash(x)=74937337)
+3240 train 4.132590 (lr=8.9157e-04) (hash(x)=77182900)
+3250 train 4.042948 (lr=8.8201e-04) (hash(x)=86214664)
+3260 train 3.916874 (lr=8.7251e-04) (hash(x)=83267879)
+3270 train 3.972420 (lr=8.6307e-04) (hash(x)=71561392)
+3280 train 3.996426 (lr=8.5368e-04) (hash(x)=83098376)
+3290 train 3.883922 (lr=8.4435e-04) (hash(x)=77431692)
+3300 val loss 4.0186
+3300 val perplexity 55.6248
+3300 train 4.037145 (lr=8.3508e-04) (hash(x)=82492818)
+3310 train 4.049354 (lr=8.2588e-04) (hash(x)=80149788)
+3320 train 4.000217 (lr=8.1673e-04) (hash(x)=78903528)
+3330 train 4.057049 (lr=8.0764e-04) (hash(x)=81042090)
+3340 train 3.999508 (lr=7.9862e-04) (hash(x)=76895955)
+3350 train 4.040513 (lr=7.8966e-04) (hash(x)=76790968)
+3360 train 4.060820 (lr=7.8076e-04) (hash(x)=85235065)
+3370 train 3.996822 (lr=7.7192e-04) (hash(x)=83413227)
+3380 train 3.975958 (lr=7.6315e-04) (hash(x)=94520262)
+3390 train 3.990027 (lr=7.5445e-04) (hash(x)=86213149)
+3400 val loss 4.0069
+3400 val perplexity 54.9769
+3400 train 3.946829 (lr=7.4581e-04) (hash(x)=77028659)
+3410 train 3.889719 (lr=7.3724e-04) (hash(x)=81165199)
+3420 train 4.003885 (lr=7.2874e-04) (hash(x)=78852885)
+3430 train 3.970803 (lr=7.2030e-04) (hash(x)=82364796)
+3440 train 4.178027 (lr=7.1193e-04) (hash(x)=66884329)
+3450 train 3.992281 (lr=7.0363e-04) (hash(x)=82196927)
+3460 train 4.047642 (lr=6.9541e-04) (hash(x)=83570482)
+3470 train 3.954385 (lr=6.8725e-04) (hash(x)=86984462)
+3480 train 4.041365 (lr=6.7916e-04) (hash(x)=80446942)
+3490 train 3.993014 (lr=6.7114e-04) (hash(x)=82386020)
 3500 val loss 3.9920
-3500 val perplexity 54.1643
-3500 train 3.964571 (lr=6.6320e-04) (hash(x)=85221122)
-3510 train 3.895456 (lr=6.5533e-04) (hash(x)=78373676)
-3520 train 3.784452 (lr=6.4753e-04) (hash(x)=76713204)
-3530 train 3.681569 (lr=6.3981e-04) (hash(x)=86885732)
-3540 train 3.815947 (lr=6.3216e-04) (hash(x)=80751876)
-3550 train 3.779734 (lr=6.2458e-04) (hash(x)=80092170)
-3560 train 3.962914 (lr=6.1708e-04) (hash(x)=85501243)
-3570 train 4.052373 (lr=6.0966e-04) (hash(x)=87853218)
-3580 train 3.999134 (lr=6.0231e-04) (hash(x)=88313614)
-3590 train 3.925564 (lr=5.9504e-04) (hash(x)=75521063)
-3600 val loss 3.9858
-3600 val perplexity 53.8265
-3600 train 4.007613 (lr=5.8785e-04) (hash(x)=83721651)
-3610 train 3.889624 (lr=5.8074e-04) (hash(x)=94293884)
-3620 train 3.943848 (lr=5.7370e-04) (hash(x)=77296763)
-3630 train 4.067281 (lr=5.6675e-04) (hash(x)=77338405)
-3640 train 3.919416 (lr=5.5987e-04) (hash(x)=81223547)
-3650 train 3.958375 (lr=5.5308e-04) (hash(x)=78952410)
-3660 train 3.807704 (lr=5.4636e-04) (hash(x)=79846386)
-3670 train 3.969377 (lr=5.3973e-04) (hash(x)=80481480)
-3680 train 3.862045 (lr=5.3318e-04) (hash(x)=83821130)
-3690 train 3.802406 (lr=5.2671e-04) (hash(x)=85303368)
-3700 val loss 3.9716
-3700 val perplexity 53.0684
-3700 train 3.838234 (lr=5.2033e-04) (hash(x)=70393870)
-3710 train 3.821644 (lr=5.1402e-04) (hash(x)=81014823)
-3720 train 3.829165 (lr=5.0780e-04) (hash(x)=79155367)
-3730 train 3.809116 (lr=5.0167e-04) (hash(x)=86979929)
-3740 train 3.763271 (lr=4.9562e-04) (hash(x)=88193263)
-3750 train 4.072014 (lr=4.8965e-04) (hash(x)=89853653)
-3760 train 3.979692 (lr=4.8377e-04) (hash(x)=77692484)
-3770 train 4.061010 (lr=4.7798e-04) (hash(x)=83300185)
-3780 train 3.975704 (lr=4.7227e-04) (hash(x)=82443218)
-3790 train 3.823617 (lr=4.6665e-04) (hash(x)=90032898)
-3800 val loss 3.9574
-3800 val perplexity 52.3231
-3800 train 3.932828 (lr=4.6112e-04) (hash(x)=77605593)
-3810 train 3.995205 (lr=4.5567e-04) (hash(x)=82532796)
-3820 train 3.950253 (lr=4.5031e-04) (hash(x)=89648204)
-3830 train 4.029373 (lr=4.4504e-04) (hash(x)=83708293)
-3840 train 3.968389 (lr=4.3986e-04) (hash(x)=84336512)
-3850 train 3.932840 (lr=4.3477e-04) (hash(x)=75025285)
-3860 train 3.813481 (lr=4.2977e-04) (hash(x)=79436536)
-3870 train 3.862545 (lr=4.2486e-04) (hash(x)=90467390)
-3880 train 3.938760 (lr=4.2004e-04) (hash(x)=77292020)
-3890 train 3.832536 (lr=4.1530e-04) (hash(x)=75568927)
-3900 val loss 3.9578
-3900 val perplexity 52.3433
-3900 train 3.879727 (lr=4.1066e-04) (hash(x)=79671781)
-3910 train 3.945248 (lr=4.0611e-04) (hash(x)=84280943)
-3920 train 3.970432 (lr=4.0166e-04) (hash(x)=80901294)
-3930 train 4.039755 (lr=3.9729e-04) (hash(x)=81437584)
-3940 train 3.944039 (lr=3.9302e-04) (hash(x)=73893778)
-3950 train 3.886929 (lr=3.8884e-04) (hash(x)=74987794)
-3960 train 3.935655 (lr=3.8475e-04) (hash(x)=93721374)
-3970 train 3.945935 (lr=3.8076e-04) (hash(x)=79149678)
-3980 train 3.896754 (lr=3.7685e-04) (hash(x)=83179387)
-3990 train 3.923690 (lr=3.7305e-04) (hash(x)=89009978)
-4000 val loss 3.9443
-4000 val perplexity 51.6416
-4000 train 3.938412 (lr=3.6933e-04) (hash(x)=76483735)
-4010 train 3.823814 (lr=3.6572e-04) (hash(x)=78638850)
-4020 train 3.898306 (lr=3.6219e-04) (hash(x)=90415206)
-4030 train 3.847905 (lr=3.5876e-04) (hash(x)=90819195)
-4040 train 3.963622 (lr=3.5543e-04) (hash(x)=85938852)
-4050 train 3.989661 (lr=3.5219e-04) (hash(x)=82436789)
-4060 train 4.145049 (lr=3.4905e-04) (hash(x)=63022149)
-4070 train 3.973024 (lr=3.4600e-04) (hash(x)=83501199)
-4080 train 3.969528 (lr=3.4305e-04) (hash(x)=77093326)
-4090 train 3.844126 (lr=3.4019e-04) (hash(x)=73437559)
-4100 val loss 3.9357
-4100 val perplexity 51.1959
-4100 train 3.945265 (lr=3.3744e-04) (hash(x)=84550388)
-4110 train 3.960652 (lr=3.3477e-04) (hash(x)=86184566)
-4120 train 3.951425 (lr=3.3221e-04) (hash(x)=85644922)
-4130 train 3.894052 (lr=3.2974e-04) (hash(x)=86179801)
-4140 train 3.867714 (lr=3.2737e-04) (hash(x)=82868303)
-4150 train 3.966802 (lr=3.2510e-04) (hash(x)=96780388)
-4160 train 3.934422 (lr=3.2292e-04) (hash(x)=87223122)
-4170 train 3.909276 (lr=3.2085e-04) (hash(x)=77292786)
-4180 train 3.920224 (lr=3.1887e-04) (hash(x)=88761375)
-4190 train 3.814977 (lr=3.1699e-04) (hash(x)=80956468)
-4200 val loss 3.9333
-4200 val perplexity 51.0761
-4200 train 3.982372 (lr=3.1520e-04) (hash(x)=84725462)
-4210 train 3.971297 (lr=3.1352e-04) (hash(x)=76098113)
-4220 train 3.973383 (lr=3.1193e-04) (hash(x)=84387685)
-4230 train 3.927773 (lr=3.1044e-04) (hash(x)=76120321)
-4240 train 3.896696 (lr=3.0905e-04) (hash(x)=85974065)
-4250 train 3.827776 (lr=3.0776e-04) (hash(x)=79340644)
-4260 train 3.926372 (lr=3.0657e-04) (hash(x)=88345025)
-4270 train 3.828650 (lr=3.0548e-04) (hash(x)=86406230)
-4280 train 3.822561 (lr=3.0449e-04) (hash(x)=80608975)
-4290 train 3.874141 (lr=3.0359e-04) (hash(x)=81635225)
-4300 val loss 3.9297
-4300 val perplexity 50.8908
-4300 train 3.743389 (lr=3.0280e-04) (hash(x)=82126376)
-4310 train 3.895471 (lr=3.0210e-04) (hash(x)=83737972)
-4320 train 3.823490 (lr=3.0150e-04) (hash(x)=81132345)
-4330 train 3.882552 (lr=3.0101e-04) (hash(x)=85850404)
-4340 train 3.887916 (lr=3.0061e-04) (hash(x)=83490640)
-4350 train 3.905009 (lr=3.0031e-04) (hash(x)=77545187)
-4360 train 3.932182 (lr=3.0011e-04) (hash(x)=86412685)
-4370 train 3.833520 (lr=3.0001e-04) (hash(x)=86429640)
-4374 val loss 3.9216
-4374 val perplexity 50.4819
+3500 val perplexity 54.1638
+3500 train 3.890259 (lr=6.6320e-04) (hash(x)=82573783)
+3510 train 3.998270 (lr=6.5533e-04) (hash(x)=86296100)
+3520 train 3.979122 (lr=6.4753e-04) (hash(x)=79959476)
+3530 train 4.001103 (lr=6.3981e-04) (hash(x)=77625970)
+3540 train 3.992567 (lr=6.3216e-04) (hash(x)=75218126)
+3550 train 3.918495 (lr=6.2458e-04) (hash(x)=85036598)
+3560 train 4.226938 (lr=6.1708e-04) (hash(x)=68457295)
+3570 train 3.910852 (lr=6.0966e-04) (hash(x)=74945892)
+3580 train 3.830779 (lr=6.0231e-04) (hash(x)=63411286)
+3590 train 3.966098 (lr=5.9504e-04) (hash(x)=81331248)
+3600 val loss 3.9810
+3600 val perplexity 53.5692
+3600 train 4.046678 (lr=5.8785e-04) (hash(x)=78629131)
+3610 train 4.046319 (lr=5.8074e-04) (hash(x)=88173496)
+3620 train 3.870049 (lr=5.7370e-04) (hash(x)=83159981)
+3630 train 4.033411 (lr=5.6675e-04) (hash(x)=73612411)
+3640 train 4.024232 (lr=5.5987e-04) (hash(x)=78561998)
+3650 train 4.083339 (lr=5.5308e-04) (hash(x)=81233526)
+3660 train 4.061245 (lr=5.4636e-04) (hash(x)=77460490)
+3670 train 3.886963 (lr=5.3973e-04) (hash(x)=79570240)
+3680 train 3.816802 (lr=5.3318e-04) (hash(x)=76916005)
+3690 train 4.045339 (lr=5.2671e-04) (hash(x)=74947748)
+3700 val loss 3.9753
+3700 val perplexity 53.2651
+3700 train 3.960517 (lr=5.2033e-04) (hash(x)=76363057)
+3710 train 3.907262 (lr=5.1402e-04) (hash(x)=74295889)
+3720 train 3.806081 (lr=5.0780e-04) (hash(x)=73447634)
+3730 train 3.910969 (lr=5.0167e-04) (hash(x)=85050263)
+3740 train 4.033413 (lr=4.9562e-04) (hash(x)=73977156)
+3750 train 3.947298 (lr=4.8965e-04) (hash(x)=79321142)
+3760 train 4.006605 (lr=4.8377e-04) (hash(x)=80904167)
+3770 train 3.921431 (lr=4.7798e-04) (hash(x)=82485963)
+3780 train 3.967444 (lr=4.7227e-04) (hash(x)=74274733)
+3790 train 3.971002 (lr=4.6665e-04) (hash(x)=95150488)
+3800 val loss 3.9583
+3800 val perplexity 52.3699
+3800 train 3.918595 (lr=4.6112e-04) (hash(x)=83818907)
+3810 train 3.957370 (lr=4.5567e-04) (hash(x)=77222344)
+3820 train 3.821163 (lr=4.5031e-04) (hash(x)=81087000)
+3830 train 3.905407 (lr=4.4504e-04) (hash(x)=81910357)
+3840 train 3.851204 (lr=4.3986e-04) (hash(x)=80293916)
+3850 train 3.957389 (lr=4.3477e-04) (hash(x)=70921646)
+3860 train 3.718121 (lr=4.2977e-04) (hash(x)=72790321)
+3870 train 3.682667 (lr=4.2486e-04) (hash(x)=78176722)
+3880 train 3.619627 (lr=4.2004e-04) (hash(x)=70542120)
+3890 train 3.730021 (lr=4.1530e-04) (hash(x)=79612492)
+3900 val loss 3.9657
+3900 val perplexity 52.7579
+3900 train 3.938238 (lr=4.1066e-04) (hash(x)=77006513)
+3910 train 4.003223 (lr=4.0611e-04) (hash(x)=85113117)
+3920 train 3.923265 (lr=4.0166e-04) (hash(x)=76822278)
+3930 train 3.935285 (lr=3.9729e-04) (hash(x)=84320111)
+3940 train 3.931844 (lr=3.9302e-04) (hash(x)=71475995)
+3950 train 3.976385 (lr=3.8884e-04) (hash(x)=79394971)
+3960 train 3.855446 (lr=3.8475e-04) (hash(x)=94369441)
+3970 train 3.974834 (lr=3.8076e-04) (hash(x)=77166806)
+3980 train 3.968989 (lr=3.7685e-04) (hash(x)=77834199)
+3990 train 3.853163 (lr=3.7305e-04) (hash(x)=77689578)
+4000 val loss 3.9432
+4000 val perplexity 51.5842
+4000 train 3.802305 (lr=3.6933e-04) (hash(x)=70746393)
+4010 train 3.824950 (lr=3.6572e-04) (hash(x)=85359486)
+4020 train 3.907485 (lr=3.6219e-04) (hash(x)=78457700)
+4030 train 3.859012 (lr=3.5876e-04) (hash(x)=78953976)
+4040 train 3.714244 (lr=3.5543e-04) (hash(x)=84086579)
+4050 train 3.660989 (lr=3.5219e-04) (hash(x)=84131543)
+4060 train 3.680241 (lr=3.4905e-04) (hash(x)=77221961)
+4070 train 3.580976 (lr=3.4600e-04) (hash(x)=87527687)
+4080 train 3.727655 (lr=3.4305e-04) (hash(x)=73788501)
+4090 train 3.920669 (lr=3.4019e-04) (hash(x)=87378361)
+4100 val loss 3.9425
+4100 val perplexity 51.5470
+4100 train 3.951674 (lr=3.3744e-04) (hash(x)=74282076)
+4110 train 3.930982 (lr=3.3477e-04) (hash(x)=86416721)
+4120 train 3.872025 (lr=3.3221e-04) (hash(x)=86325819)
+4130 train 3.892965 (lr=3.2974e-04) (hash(x)=77534282)
+4140 train 3.894976 (lr=3.2737e-04) (hash(x)=73179409)
+4150 train 3.944669 (lr=3.2510e-04) (hash(x)=91295398)
+4160 train 3.907704 (lr=3.2292e-04) (hash(x)=84833773)
+4170 train 3.910796 (lr=3.2085e-04) (hash(x)=74079521)
+4180 train 3.819831 (lr=3.1887e-04) (hash(x)=74585954)
+4190 train 3.848926 (lr=3.1699e-04) (hash(x)=74849974)
+4200 val loss 3.9353
+4200 val perplexity 51.1779
+4200 train 3.947260 (lr=3.1520e-04) (hash(x)=87782408)
+4210 train 3.998633 (lr=3.1352e-04) (hash(x)=71390249)
+4220 train 3.690743 (lr=3.1193e-04) (hash(x)=80184408)
+4230 train 3.704522 (lr=3.1044e-04) (hash(x)=78607457)
+4240 train 3.726249 (lr=3.0905e-04) (hash(x)=83027401)
+4250 train 3.584549 (lr=3.0776e-04) (hash(x)=80802729)
+4260 train 3.883817 (lr=3.0657e-04) (hash(x)=75252098)
+4270 train 3.932910 (lr=3.0548e-04) (hash(x)=79153989)
+4280 train 4.035439 (lr=3.0449e-04) (hash(x)=85833030)
+4290 train 3.976151 (lr=3.0359e-04) (hash(x)=84103039)
+4300 val loss 3.9287
+4300 val perplexity 50.8412
+4300 train 4.107261 (lr=3.0280e-04) (hash(x)=76802610)
+4310 train 4.010763 (lr=3.0210e-04) (hash(x)=77169206)
+4320 train 3.815757 (lr=3.0150e-04) (hash(x)=83226021)
+4330 train 3.811772 (lr=3.0101e-04) (hash(x)=80213024)
+4340 train 3.833462 (lr=3.0061e-04) (hash(x)=80732928)
+4350 train 3.886442 (lr=3.0031e-04) (hash(x)=81534899)
+4360 train 3.842135 (lr=3.0011e-04) (hash(x)=80854213)
+4370 train 3.916704 (lr=3.0001e-04) (hash(x)=76420919)
+4374 val loss 3.9299
+4374 val perplexity 50.9030

12_head_baseline_lr_30e-4_head_dim_22_fixed/model_04374.pt

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:2bb94add8725871cd7a33113d3733bc857499f0ddc1e7e191fcad7826d7a75aa
+oid sha256:18011adf0734fd8d510761290b009e475e98d78a4c84ed4ead39d20b2cea9b75
 size 96858242

12_head_baseline_lr_30e-4_head_dim_22_fixed/optimizer_04374.pt

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:d298e8ae000b9f9b2e93c7a14b7885a70066b11dcfe7a74a4ac01c26ed0805f8
+oid sha256:4ffd6439a763882d66220a8cf09822410a16975f15730dc2921a27645fe42696
 size 187435910