Upload folder using huggingface_hub

attention_kindselective_n_heads4_seed1340/args.json

@@ -1 +1 @@
-{"hellaswag": true, "attention_kind": "selective", "log_dir": "wider_is_better_3/attention_kindselective_n_heads4_seed1340", "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": 10000, "warmup_steps": 200, "group": "wider_is_better_3", "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": 120, "total_batch_size": 30720, "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": 0.00015, "decay_lr": true, "readout_zero_init": false, "query_zero_init": false, "l1_loss": false, "debugpy": false, "key": "1.5e-4_30720_4_1340", "n_embd": 256}
+{"hellaswag": true, "attention_kind": "selective", "log_dir": "wider_is_better_4/attention_kindselective_n_heads4_seed1340", "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": 10000, "warmup_steps": 200, "group": "wider_is_better_4", "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": 120, "total_batch_size": 30720, "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": 0.00015, "decay_lr": true, "readout_zero_init": false, "query_zero_init": false, "l1_loss": false, "debugpy": false, "key": "1.5e-4_30720_4_1340", "n_embd": 256}

attention_kindselective_n_heads4_seed1340/log2.txt

@@ -2,302 +2,302 @@ max_steps: 10000
 0 val loss 11.2714
 0 val perplexity 78545.6094
 0 train 11.276529 (lr=7.5000e-07) (hash(x)=164406924)
-100 val loss 8.8150
-100 val perplexity 6734.2607
-100 train 9.008026 (lr=7.5750e-05) (hash(x)=177407419)
-200 val loss 7.5896
-200 val perplexity 1977.4829
-200 train 7.496974 (lr=1.5000e-04) (hash(x)=144903932)
-300 val loss 7.3940
-300 val perplexity 1626.2600
-300 train 7.754192 (lr=1.4997e-04) (hash(x)=173839165)
-400 val loss 7.2907
-400 val perplexity 1466.5542
-400 train 7.382928 (lr=1.4986e-04) (hash(x)=167734596)
-500 val loss 7.2478
-500 val perplexity 1404.9714
-500 train 7.273375 (lr=1.4969e-04) (hash(x)=153224076)
-600 val loss 7.1863
-600 val perplexity 1321.2104
-600 train 7.091864 (lr=1.4945e-04) (hash(x)=149619098)
-700 val loss 7.1670
-700 val perplexity 1295.9884
-700 train 7.079926 (lr=1.4913e-04) (hash(x)=146539909)
-800 val loss 7.0517
-800 val perplexity 1154.7979
-800 train 7.019431 (lr=1.4876e-04) (hash(x)=153710890)
-900 val loss 7.0787
-900 val perplexity 1186.3789
-900 train 7.007246 (lr=1.4831e-04) (hash(x)=155873620)
-1000 val loss 7.0019
-1000 val perplexity 1098.6797
-1000 train 6.921768 (lr=1.4779e-04) (hash(x)=145450636)
-1100 val loss 6.9365
-1100 val perplexity 1029.1884
-1100 train 7.008905 (lr=1.4721e-04) (hash(x)=154123388)
-1200 val loss 6.8752
-1200 val perplexity 967.9332
-1200 train 6.743853 (lr=1.4656e-04) (hash(x)=145249251)
-1300 val loss 6.8314
-1300 val perplexity 926.5063
-1300 train 6.742863 (lr=1.4585e-04) (hash(x)=148937127)
-1400 val loss 6.8084
-1400 val perplexity 905.3779
-1400 train 6.911214 (lr=1.4507e-04) (hash(x)=150475545)
-1500 val loss 6.7983
-1500 val perplexity 896.3302
-1500 train 6.726580 (lr=1.4422e-04) (hash(x)=154653428)
-1600 val loss 6.7646
-1600 val perplexity 866.6557
-1600 train 6.755625 (lr=1.4332e-04) (hash(x)=144483776)
-1700 val loss 6.7377
-1700 val perplexity 843.6576
-1700 train 6.890347 (lr=1.4235e-04) (hash(x)=157395496)
-1800 val loss 6.7153
-1800 val perplexity 824.9490
-1800 train 6.730232 (lr=1.4131e-04) (hash(x)=157916369)
-1900 val loss 6.6871
-1900 val perplexity 801.9601
-1900 train 6.886857 (lr=1.4022e-04) (hash(x)=166073923)
-2000 val loss 6.6997
-2000 val perplexity 812.1252
-2000 train 6.739951 (lr=1.3907e-04) (hash(x)=154856891)
-2100 val loss 6.6662
-2100 val perplexity 785.4100
-2100 train 6.646859 (lr=1.3786e-04) (hash(x)=151925203)
-2200 val loss 6.6335
-2200 val perplexity 760.1653
-2200 train 6.329777 (lr=1.3660e-04) (hash(x)=136191502)
-2300 val loss 6.6140
-2300 val perplexity 745.4225
-2300 train 6.765590 (lr=1.3527e-04) (hash(x)=153273362)
-2400 val loss 6.6332
-2400 val perplexity 759.9003
-2400 train 6.583218 (lr=1.3390e-04) (hash(x)=148021541)
-2500 val loss 6.5995
-2500 val perplexity 734.7002
-2500 train 6.545035 (lr=1.3247e-04) (hash(x)=141356608)
-2600 val loss 6.5794
-2600 val perplexity 720.0811
-2600 train 6.530703 (lr=1.3099e-04) (hash(x)=146005217)
-2700 val loss 6.5640
-2700 val perplexity 709.1022
-2700 train 6.412026 (lr=1.2946e-04) (hash(x)=144511718)
-2800 val loss 6.5768
-2800 val perplexity 718.2365
-2800 train 6.447824 (lr=1.2788e-04) (hash(x)=146019502)
-2900 val loss 6.5444
-2900 val perplexity 695.3627
-2900 train 6.449443 (lr=1.2626e-04) (hash(x)=146496200)
-3000 val loss 6.5497
-3000 val perplexity 699.0353
-3000 train 6.473566 (lr=1.2459e-04) (hash(x)=150127281)
-3100 val loss 6.5133
-3100 val perplexity 674.0693
-3100 train 6.502546 (lr=1.2287e-04) (hash(x)=142022255)
-3200 val loss 6.5119
-3200 val perplexity 673.0822
-3200 train 6.562176 (lr=1.2112e-04) (hash(x)=154120875)
-3300 val loss 6.5126
-3300 val perplexity 673.5568
-3300 train 6.590917 (lr=1.1932e-04) (hash(x)=153999717)
-3400 val loss 6.5166
-3400 val perplexity 676.2653
-3400 train 6.303998 (lr=1.1749e-04) (hash(x)=139694097)
-3500 val loss 6.4932
-3500 val perplexity 660.6465
-3500 train 6.682529 (lr=1.1562e-04) (hash(x)=162992732)
-3600 val loss 6.4755
-3600 val perplexity 649.0635
-3600 train 6.450060 (lr=1.1372e-04) (hash(x)=147574101)
-3700 val loss 6.4528
-3700 val perplexity 634.4659
-3700 train 6.548763 (lr=1.1179e-04) (hash(x)=157763099)
-3800 val loss 6.4608
-3800 val perplexity 639.5983
-3800 train 6.602811 (lr=1.0982e-04) (hash(x)=170800034)
-3900 val loss 6.4688
-3900 val perplexity 644.6869
-3900 train 6.549332 (lr=1.0783e-04) (hash(x)=164984528)
-4000 val loss 6.4359
-4000 val perplexity 623.8695
-4000 train 6.299426 (lr=1.0581e-04) (hash(x)=141743323)
-4100 val loss 6.4323
-4100 val perplexity 621.5958
-4100 train 6.459531 (lr=1.0377e-04) (hash(x)=153392872)
-4200 val loss 6.4372
-4200 val perplexity 624.6572
-4200 train 6.306248 (lr=1.0171e-04) (hash(x)=149074933)
-4300 val loss 6.4231
-4300 val perplexity 615.9368
-4300 train 6.761147 (lr=9.9622e-05) (hash(x)=167823423)
-4400 val loss 6.4058
-4400 val perplexity 605.3599
-4400 train 6.181880 (lr=9.7520e-05) (hash(x)=141203114)
-4500 val loss 6.3847
-4500 val perplexity 592.6842
-4500 train 6.381039 (lr=9.5403e-05) (hash(x)=146284780)
-4600 val loss 6.3872
-4600 val perplexity 594.1828
-4600 train 6.197599 (lr=9.3273e-05) (hash(x)=141126464)
-4700 val loss 6.3703
-4700 val perplexity 584.2452
-4700 train 6.402414 (lr=9.1132e-05) (hash(x)=154751926)
-4800 val loss 6.3828
-4800 val perplexity 591.6098
-4800 train 6.453866 (lr=8.8982e-05) (hash(x)=154793198)
-4900 val loss 6.3816
-4900 val perplexity 590.8748
-4900 train 6.118082 (lr=8.6825e-05) (hash(x)=139406392)
-5000 val loss 6.3803
-5000 val perplexity 590.1230
-5000 train 6.104129 (lr=8.4663e-05) (hash(x)=153548741)
-5100 val loss 6.3450
-5100 val perplexity 569.6191
-5100 train 6.405098 (lr=8.2500e-05) (hash(x)=160488568)
-5200 val loss 6.3360
-5200 val perplexity 564.5164
-5200 train 6.349102 (lr=8.0337e-05) (hash(x)=149645053)
-5300 val loss 6.3192
-5300 val perplexity 555.1131
-5300 train 6.399147 (lr=7.8175e-05) (hash(x)=155820556)
-5400 val loss 6.3165
-5400 val perplexity 553.6335
-5400 train 6.258524 (lr=7.6018e-05) (hash(x)=147538134)
-5500 val loss 6.3046
-5500 val perplexity 547.1053
-5500 train 6.505478 (lr=7.3868e-05) (hash(x)=166889307)
-5600 val loss 6.2934
-5600 val perplexity 541.0081
-5600 train 6.022371 (lr=7.1727e-05) (hash(x)=139516699)
-5700 val loss 6.2970
-5700 val perplexity 542.9310
-5700 train 5.996075 (lr=6.9597e-05) (hash(x)=140453511)
-5800 val loss 6.3013
-5800 val perplexity 545.3030
-5800 train 6.238603 (lr=6.7480e-05) (hash(x)=162964847)
-5900 val loss 6.2847
-5900 val perplexity 536.3085
-5900 train 6.292588 (lr=6.5378e-05) (hash(x)=150606634)
-6000 val loss 6.2641
-6000 val perplexity 525.3836
-6000 train 6.354014 (lr=6.3294e-05) (hash(x)=149890857)
-6100 val loss 6.2563
-6100 val perplexity 521.2612
-6100 train 6.352149 (lr=6.1230e-05) (hash(x)=173884145)
-6200 val loss 6.2540
-6200 val perplexity 520.1025
-6200 train 6.313532 (lr=5.9188e-05) (hash(x)=151987098)
-6300 val loss 6.2369
-6300 val perplexity 511.2752
-6300 train 6.189893 (lr=5.7169e-05) (hash(x)=148853562)
-6400 val loss 6.2421
-6400 val perplexity 513.9292
-6400 train 5.996774 (lr=5.5177e-05) (hash(x)=141530101)
-6500 val loss 6.2371
-6500 val perplexity 511.3586
-6500 train 6.104002 (lr=5.3213e-05) (hash(x)=142297809)
-6600 val loss 6.2206
-6600 val perplexity 502.9856
-6600 train 6.123561 (lr=5.1279e-05) (hash(x)=142447782)
-6700 val loss 6.2118
-6700 val perplexity 498.5954
-6700 train 6.167513 (lr=4.9377e-05) (hash(x)=147004686)
-6800 val loss 6.2079
-6800 val perplexity 496.6617
-6800 train 5.954415 (lr=4.7509e-05) (hash(x)=133438702)
-6900 val loss 6.2137
-6900 val perplexity 499.5584
-6900 train 6.254116 (lr=4.5676e-05) (hash(x)=157085143)
-7000 val loss 6.2020
-7000 val perplexity 493.7236
-7000 train 6.080952 (lr=4.3882e-05) (hash(x)=139437666)
-7100 val loss 6.1982
-7100 val perplexity 491.8668
-7100 train 6.245137 (lr=4.2128e-05) (hash(x)=159792986)
-7200 val loss 6.1857
-7200 val perplexity 485.7494
-7200 train 6.146164 (lr=4.0414e-05) (hash(x)=144930687)
-7300 val loss 6.1878
-7300 val perplexity 486.7650
-7300 train 6.239649 (lr=3.8745e-05) (hash(x)=156242690)
-7400 val loss 6.1793
-7400 val perplexity 482.6702
-7400 train 6.021890 (lr=3.7120e-05) (hash(x)=148183719)
-7500 val loss 6.1775
-7500 val perplexity 481.8043
-7500 train 6.192832 (lr=3.5541e-05) (hash(x)=152494758)
-7600 val loss 6.1712
-7600 val perplexity 478.7643
-7600 train 5.871748 (lr=3.4011e-05) (hash(x)=142485027)
-7700 val loss 6.1611
-7700 val perplexity 473.9326
-7700 train 6.010452 (lr=3.2531e-05) (hash(x)=147512165)
-7800 val loss 6.1533
-7800 val perplexity 470.2879
-7800 train 6.045797 (lr=3.1102e-05) (hash(x)=160346994)
-7900 val loss 6.1551
-7900 val perplexity 471.1004
-7900 train 6.012663 (lr=2.9726e-05) (hash(x)=144488254)
-8000 val loss 6.1499
-8000 val perplexity 468.6790
-8000 train 5.967876 (lr=2.8405e-05) (hash(x)=147637019)
-8100 val loss 6.1483
-8100 val perplexity 467.9341
-8100 train 6.004601 (lr=2.7138e-05) (hash(x)=147340534)
-8200 val loss 6.1459
-8200 val perplexity 466.7805
-8200 train 6.177722 (lr=2.5929e-05) (hash(x)=151630665)
-8300 val loss 6.1351
-8300 val perplexity 461.7711
-8300 train 6.219531 (lr=2.4778e-05) (hash(x)=149747064)
-8400 val loss 6.1336
-8400 val perplexity 461.0908
-8400 train 6.320711 (lr=2.3686e-05) (hash(x)=154245770)
-8500 val loss 6.1307
-8500 val perplexity 459.7752
-8500 train 6.014756 (lr=2.2655e-05) (hash(x)=152559100)
-8600 val loss 6.1278
-8600 val perplexity 458.4228
-8600 train 6.624016 (lr=2.1685e-05) (hash(x)=181365926)
-8700 val loss 6.1307
-8700 val perplexity 459.7448
-8700 train 5.961389 (lr=2.0777e-05) (hash(x)=154405991)
-8800 val loss 6.1234
-8800 val perplexity 456.4016
-8800 train 6.065418 (lr=1.9933e-05) (hash(x)=153755904)
-8900 val loss 6.1210
-8900 val perplexity 455.2997
-8900 train 6.049755 (lr=1.9153e-05) (hash(x)=152120568)
-9000 val loss 6.1119
-9000 val perplexity 451.1904
-9000 train 5.915850 (lr=1.8439e-05) (hash(x)=142797279)
-9100 val loss 6.1117
-9100 val perplexity 451.1129
-9100 train 5.974414 (lr=1.7790e-05) (hash(x)=143037503)
-9200 val loss 6.1116
-9200 val perplexity 451.0441
-9200 train 6.015065 (lr=1.7208e-05) (hash(x)=113690273)
-9300 val loss 6.1111
-9300 val perplexity 450.8334
-9300 train 6.035745 (lr=1.6692e-05) (hash(x)=158025077)
-9400 val loss 6.1085
-9400 val perplexity 449.6438
-9400 train 6.165466 (lr=1.6245e-05) (hash(x)=158251718)
-9500 val loss 6.1043
-9500 val perplexity 447.7663
-9500 train 6.131958 (lr=1.5865e-05) (hash(x)=154752610)
-9600 val loss 6.1010
-9600 val perplexity 446.3106
-9600 train 5.971730 (lr=1.5554e-05) (hash(x)=146889093)
-9700 val loss 6.0977
-9700 val perplexity 444.8401
-9700 train 6.104051 (lr=1.5312e-05) (hash(x)=156906516)
-9800 val loss 6.0975
-9800 val perplexity 444.7247
-9800 train 5.942422 (lr=1.5139e-05) (hash(x)=153841927)
-9900 val loss 6.0961
-9900 val perplexity 444.1263
-9900 train 6.294320 (lr=1.5035e-05) (hash(x)=163514334)
-9999 val loss 6.0919
-9999 val perplexity 442.2750
+100 val loss 9.0381
+100 val perplexity 8417.9463
+100 train 9.196967 (lr=7.5750e-05) (hash(x)=177407419)
+200 val loss 7.6693
+200 val perplexity 2141.5881
+200 train 7.579660 (lr=1.5000e-04) (hash(x)=144903932)
+300 val loss 7.5340
+300 val perplexity 1870.6281
+300 train 7.896007 (lr=1.4997e-04) (hash(x)=173839165)
+400 val loss 7.4474
+400 val perplexity 1715.3689
+400 train 7.538729 (lr=1.4986e-04) (hash(x)=167734596)
+500 val loss 7.3606
+500 val perplexity 1572.7972
+500 train 7.367005 (lr=1.4969e-04) (hash(x)=153224076)
+600 val loss 7.3408
+600 val perplexity 1541.9294
+600 train 7.236241 (lr=1.4945e-04) (hash(x)=149619098)
+700 val loss 7.3572
+700 val perplexity 1567.4814
+700 train 7.288768 (lr=1.4913e-04) (hash(x)=146539909)
+800 val loss 7.2903
+800 val perplexity 1466.0669
+800 train 7.251941 (lr=1.4876e-04) (hash(x)=153710890)
+900 val loss 7.2480
+900 val perplexity 1405.2601
+900 train 7.202537 (lr=1.4831e-04) (hash(x)=155873620)
+1000 val loss 7.1965
+1000 val perplexity 1334.7125
+1000 train 7.122793 (lr=1.4779e-04) (hash(x)=145450636)
+1100 val loss 7.1624
+1100 val perplexity 1289.9513
+1100 train 7.217214 (lr=1.4721e-04) (hash(x)=154123388)
+1200 val loss 7.1348
+1200 val perplexity 1254.8914
+1200 train 7.028561 (lr=1.4656e-04) (hash(x)=145249251)
+1300 val loss 7.1244
+1300 val perplexity 1241.8899
+1300 train 7.036378 (lr=1.4585e-04) (hash(x)=148937127)
+1400 val loss 7.1494
+1400 val perplexity 1273.3578
+1400 train 7.247538 (lr=1.4507e-04) (hash(x)=150475545)
+1500 val loss 7.1809
+1500 val perplexity 1314.0698
+1500 train 7.120584 (lr=1.4422e-04) (hash(x)=154653428)
+1600 val loss 7.2193
+1600 val perplexity 1365.5410
+1600 train 7.176501 (lr=1.4332e-04) (hash(x)=144483776)
+1700 val loss 7.0776
+1700 val perplexity 1185.1469
+1700 train 7.226075 (lr=1.4235e-04) (hash(x)=157395496)
+1800 val loss 7.0167
+1800 val perplexity 1115.0973
+1800 train 7.032615 (lr=1.4131e-04) (hash(x)=157916369)
+1900 val loss 6.9944
+1900 val perplexity 1090.5461
+1900 train 7.183726 (lr=1.4022e-04) (hash(x)=166073923)
+2000 val loss 6.9464
+2000 val perplexity 1039.3651
+2000 train 6.996574 (lr=1.3907e-04) (hash(x)=154856891)
+2100 val loss 6.9176
+2100 val perplexity 1009.8993
+2100 train 6.904967 (lr=1.3786e-04) (hash(x)=151925203)
+2200 val loss 6.8919
+2200 val perplexity 984.2449
+2200 train 6.614451 (lr=1.3660e-04) (hash(x)=136191502)
+2300 val loss 6.8251
+2300 val perplexity 920.6675
+2300 train 6.939867 (lr=1.3527e-04) (hash(x)=153273362)
+2400 val loss 6.8751
+2400 val perplexity 967.8990
+2400 train 6.812264 (lr=1.3390e-04) (hash(x)=148021541)
+2500 val loss 6.8260
+2500 val perplexity 921.5288
+2500 train 6.754780 (lr=1.3247e-04) (hash(x)=141356608)
+2600 val loss 6.8221
+2600 val perplexity 917.9173
+2600 train 6.776220 (lr=1.3099e-04) (hash(x)=146005217)
+2700 val loss 6.8013
+2700 val perplexity 898.9861
+2700 train 6.667194 (lr=1.2946e-04) (hash(x)=144511718)
+2800 val loss 6.7671
+2800 val perplexity 868.7567
+2800 train 6.655071 (lr=1.2788e-04) (hash(x)=146019502)
+2900 val loss 6.7840
+2900 val perplexity 883.6208
+2900 train 6.688586 (lr=1.2626e-04) (hash(x)=146496200)
+3000 val loss 6.7810
+3000 val perplexity 880.9838
+3000 train 6.711241 (lr=1.2459e-04) (hash(x)=150127281)
+3100 val loss 6.7099
+3100 val perplexity 820.5069
+3100 train 6.685182 (lr=1.2287e-04) (hash(x)=142022255)
+3200 val loss 6.6986
+3200 val perplexity 811.2672
+3200 train 6.748562 (lr=1.2112e-04) (hash(x)=154120875)
+3300 val loss 6.6998
+3300 val perplexity 812.2066
+3300 train 6.751196 (lr=1.1932e-04) (hash(x)=153999717)
+3400 val loss 6.6799
+3400 val perplexity 796.2400
+3400 train 6.459930 (lr=1.1749e-04) (hash(x)=139694097)
+3500 val loss 6.6282
+3500 val perplexity 756.0915
+3500 train 6.809334 (lr=1.1562e-04) (hash(x)=162992732)
+3600 val loss 6.6049
+3600 val perplexity 738.7014
+3600 train 6.586004 (lr=1.1372e-04) (hash(x)=147574101)
+3700 val loss 6.5722
+3700 val perplexity 714.9374
+3700 train 6.681458 (lr=1.1179e-04) (hash(x)=157763099)
+3800 val loss 6.5505
+3800 val perplexity 699.6288
+3800 train 6.692687 (lr=1.0982e-04) (hash(x)=170800034)
+3900 val loss 6.5508
+3900 val perplexity 699.7797
+3900 train 6.629467 (lr=1.0783e-04) (hash(x)=164984528)
+4000 val loss 6.5329
+4000 val perplexity 687.3677
+4000 train 6.404568 (lr=1.0581e-04) (hash(x)=141743323)
+4100 val loss 6.5432
+4100 val perplexity 694.5349
+4100 train 6.561630 (lr=1.0377e-04) (hash(x)=153392872)
+4200 val loss 6.5313
+4200 val perplexity 686.2911
+4200 train 6.403113 (lr=1.0171e-04) (hash(x)=149074933)
+4300 val loss 6.5191
+4300 val perplexity 677.9529
+4300 train 6.874164 (lr=9.9622e-05) (hash(x)=167823423)
+4400 val loss 6.4876
+4400 val perplexity 656.9694
+4400 train 6.258415 (lr=9.7520e-05) (hash(x)=141203114)
+4500 val loss 6.5010
+4500 val perplexity 665.8226
+4500 train 6.505226 (lr=9.5403e-05) (hash(x)=146284780)
+4600 val loss 6.4811
+4600 val perplexity 652.6643
+4600 train 6.300009 (lr=9.3273e-05) (hash(x)=141126464)
+4700 val loss 6.4968
+4700 val perplexity 663.0125
+4700 train 6.525679 (lr=9.1132e-05) (hash(x)=154751926)
+4800 val loss 6.4780
+4800 val perplexity 650.6808
+4800 train 6.554049 (lr=8.8982e-05) (hash(x)=154793198)
+4900 val loss 6.4792
+4900 val perplexity 651.4274
+4900 train 6.231535 (lr=8.6825e-05) (hash(x)=139406392)
+5000 val loss 6.4978
+5000 val perplexity 663.7080
+5000 train 6.241301 (lr=8.4663e-05) (hash(x)=153548741)
+5100 val loss 6.4781
+5100 val perplexity 650.7271
+5100 train 6.534034 (lr=8.2500e-05) (hash(x)=160488568)
+5200 val loss 6.4697
+5200 val perplexity 645.2586
+5200 train 6.485910 (lr=8.0337e-05) (hash(x)=149645053)
+5300 val loss 6.4504
+5300 val perplexity 632.9360
+5300 train 6.529132 (lr=7.8175e-05) (hash(x)=155820556)
+5400 val loss 6.4430
+5400 val perplexity 628.2618
+5400 train 6.396791 (lr=7.6018e-05) (hash(x)=147538134)
+5500 val loss 6.4343
+5500 val perplexity 622.8408
+5500 train 6.642034 (lr=7.3868e-05) (hash(x)=166889307)
+5600 val loss 6.4312
+5600 val perplexity 620.8967
+5600 train 6.160884 (lr=7.1727e-05) (hash(x)=139516699)
+5700 val loss 6.4238
+5700 val perplexity 616.3513
+5700 train 6.132991 (lr=6.9597e-05) (hash(x)=140453511)
+5800 val loss 6.4369
+5800 val perplexity 624.4707
+5800 train 6.383884 (lr=6.7480e-05) (hash(x)=162964847)
+5900 val loss 6.4048
+5900 val perplexity 604.7402
+5900 train 6.415184 (lr=6.5378e-05) (hash(x)=150606634)
+6000 val loss 6.3960
+6000 val perplexity 599.4601
+6000 train 6.499843 (lr=6.3294e-05) (hash(x)=149890857)
+6100 val loss 6.3879
+6100 val perplexity 594.5801
+6100 train 6.491496 (lr=6.1230e-05) (hash(x)=173884145)
+6200 val loss 6.3796
+6200 val perplexity 589.6749
+6200 train 6.443524 (lr=5.9188e-05) (hash(x)=151987098)
+6300 val loss 6.3793
+6300 val perplexity 589.4962
+6300 train 6.343131 (lr=5.7169e-05) (hash(x)=148853562)
+6400 val loss 6.3634
+6400 val perplexity 580.2146
+6400 train 6.131594 (lr=5.5177e-05) (hash(x)=141530101)
+6500 val loss 6.3555
+6500 val perplexity 575.6595
+6500 train 6.214462 (lr=5.3213e-05) (hash(x)=142297809)
+6600 val loss 6.3444
+6600 val perplexity 569.2677
+6600 train 6.239904 (lr=5.1279e-05) (hash(x)=142447782)
+6700 val loss 6.3507
+6700 val perplexity 572.8687
+6700 train 6.295353 (lr=4.9377e-05) (hash(x)=147004686)
+6800 val loss 6.3404
+6800 val perplexity 567.0021
+6800 train 6.077054 (lr=4.7509e-05) (hash(x)=133438702)
+6900 val loss 6.3437
+6900 val perplexity 568.9139
+6900 train 6.377135 (lr=4.5676e-05) (hash(x)=157085143)
+7000 val loss 6.3345
+7000 val perplexity 563.6675
+7000 train 6.220019 (lr=4.3882e-05) (hash(x)=139437666)
+7100 val loss 6.3297
+7100 val perplexity 560.9677
+7100 train 6.403553 (lr=4.2128e-05) (hash(x)=159792986)
+7200 val loss 6.3186
+7200 val perplexity 554.7760
+7200 train 6.285450 (lr=4.0414e-05) (hash(x)=144930687)
+7300 val loss 6.3137
+7300 val perplexity 552.0679
+7300 train 6.387995 (lr=3.8745e-05) (hash(x)=156242690)
+7400 val loss 6.3059
+7400 val perplexity 547.7939
+7400 train 6.166557 (lr=3.7120e-05) (hash(x)=148183719)
+7500 val loss 6.3071
+7500 val perplexity 548.4332
+7500 train 6.323625 (lr=3.5541e-05) (hash(x)=152494758)
+7600 val loss 6.3082
+7600 val perplexity 549.0623
+7600 train 6.021165 (lr=3.4011e-05) (hash(x)=142485027)
+7700 val loss 6.2966
+7700 val perplexity 542.7446
+7700 train 6.145420 (lr=3.2531e-05) (hash(x)=147512165)
+7800 val loss 6.2939
+7800 val perplexity 541.2756
+7800 train 6.209620 (lr=3.1102e-05) (hash(x)=160346994)
+7900 val loss 6.2867
+7900 val perplexity 537.3802
+7900 train 6.138093 (lr=2.9726e-05) (hash(x)=144488254)
+8000 val loss 6.2816
+8000 val perplexity 534.6220
+8000 train 6.089755 (lr=2.8405e-05) (hash(x)=147637019)
+8100 val loss 6.2795
+8100 val perplexity 533.5262
+8100 train 6.136616 (lr=2.7138e-05) (hash(x)=147340534)
+8200 val loss 6.2769
+8200 val perplexity 532.1191
+8200 train 6.305046 (lr=2.5929e-05) (hash(x)=151630665)
+8300 val loss 6.2704
+8300 val perplexity 528.6994
+8300 train 6.352071 (lr=2.4778e-05) (hash(x)=149747064)
+8400 val loss 6.2696
+8400 val perplexity 528.2690
+8400 train 6.457778 (lr=2.3686e-05) (hash(x)=154245770)
+8500 val loss 6.2686
+8500 val perplexity 527.7471
+8500 train 6.158002 (lr=2.2655e-05) (hash(x)=152559100)
+8600 val loss 6.2606
+8600 val perplexity 523.5390
+8600 train 6.811657 (lr=2.1685e-05) (hash(x)=181365926)
+8700 val loss 6.2652
+8700 val perplexity 525.9496
+8700 train 6.102303 (lr=2.0777e-05) (hash(x)=154405991)
+8800 val loss 6.2593
+8800 val perplexity 522.8442
+8800 train 6.197897 (lr=1.9933e-05) (hash(x)=153755904)
+8900 val loss 6.2577
+8900 val perplexity 522.0072
+8900 train 6.199480 (lr=1.9153e-05) (hash(x)=152120568)
+9000 val loss 6.2502
+9000 val perplexity 518.1027
+9000 train 6.055476 (lr=1.8439e-05) (hash(x)=142797279)
+9100 val loss 6.2482
+9100 val perplexity 517.1046
+9100 train 6.107931 (lr=1.7790e-05) (hash(x)=143037503)
+9200 val loss 6.2448
+9200 val perplexity 515.3453
+9200 train 6.117033 (lr=1.7208e-05) (hash(x)=113690273)
+9300 val loss 6.2433
+9300 val perplexity 514.5726
+9300 train 6.173688 (lr=1.6692e-05) (hash(x)=158025077)
+9400 val loss 6.2395
+9400 val perplexity 512.5978
+9400 train 6.295310 (lr=1.6245e-05) (hash(x)=158251718)
+9500 val loss 6.2356
+9500 val perplexity 510.6154
+9500 train 6.264369 (lr=1.5865e-05) (hash(x)=154752610)
+9600 val loss 6.2365
+9600 val perplexity 511.0636
+9600 train 6.110316 (lr=1.5554e-05) (hash(x)=146889093)
+9700 val loss 6.2315
+9700 val perplexity 508.5052
+9700 train 6.236925 (lr=1.5312e-05) (hash(x)=156906516)
+9800 val loss 6.2308
+9800 val perplexity 508.1559
+9800 train 6.083229 (lr=1.5139e-05) (hash(x)=153841927)
+9900 val loss 6.2303
+9900 val perplexity 507.8923
+9900 train 6.427679 (lr=1.5035e-05) (hash(x)=163514334)
+9999 val loss 6.2221
+9999 val perplexity 503.7782

attention_kindselective_n_heads4_seed1340/model_02500.pt

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:f77a0e86af837e96ddb73bcf1787b783ad3319e4b7779f945478abab4e76e4f9
+oid sha256:dc8ab622ad0bced5af47c95b062cb4142a172205f16a19af25bfeb37ae7e6ead
 size 92843394

attention_kindselective_n_heads4_seed1340/model_05000.pt

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:0e7e75dbf141b37b031861e58726b1c6227e8d371b6136bb3f597ea05775f710
+oid sha256:5101401041308ab5e3b94226fac88ca83468b023ae6cb4b47849d5c1a7d152b7
 size 92843394

attention_kindselective_n_heads4_seed1340/model_07500.pt

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:f8c999d26ff45c1412e6520045612b33258b6ff6547a6a4b6e054e2afd81f84b
+oid sha256:c0101f627a40dd410fab6f508a3a497c046dbbf1dbfe43ec81e6646b715a6e9b
 size 92843394

attention_kindselective_n_heads4_seed1340/model_09999.pt

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:505e431b92125e3353dbe2fe8f73f190410499e8fa0aae75dd9b4e6148abf833
+oid sha256:aa0f34e44744d28c8fc4936ee2d79509aae3e04975bced22d255115e2576cf62
 size 92843394

attention_kindselective_n_heads4_seed1340/optimizer_02500.pt

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:1839e8234b3588e61b65c80b0816353d4dbf9956962b809e16783bd005d8e8c7
+oid sha256:9d98a3486500cda9b7a78aff66d543fb4e921587f58675349d8d1a02e4977349
 size 179406214

attention_kindselective_n_heads4_seed1340/optimizer_05000.pt

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:c8c3254c5328b3a1eae82aa7bbcfae1269f5902ef429b7426fe60a02c5cdb78e
+oid sha256:9b0d77a77d8a5c48775bb7f2c0c19972e9a8fa816c04146058b96849cf820186
 size 179406214

attention_kindselective_n_heads4_seed1340/optimizer_07500.pt

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:da5e88ac26e1e0c3c72ce92eb60bd8311d31396b4c5509b1d6a424a6bec8127c
+oid sha256:7b64cf3e05ef6defb6d5634f73857bb471e6418aeab79e0d84177c0e7193b147
 size 179406214

attention_kindselective_n_heads4_seed1340/optimizer_09999.pt

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:5853170d79d7583bbccfae317f503078a42902ade708ab27df68638a7bbfd6f5
+oid sha256:28f88b813cbb7cebd77efdbeb49842210d132f1331d2a9518f163562c7fa3f8b
 size 179406214