Upload folder using huggingface_hub

attention_kindselective_n_heads4_seed1338/args.json

@@ -1 +1 @@
-{"hellaswag": true, "attention_kind": "selective", "log_dir": "wider_is_better_3/attention_kindselective_n_heads4_seed1338", "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": 1338, "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_1338", "n_embd": 256}
+{"hellaswag": true, "attention_kind": "selective", "log_dir": "wider_is_better_4/attention_kindselective_n_heads4_seed1338", "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": 1338, "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_1338", "n_embd": 256}

attention_kindselective_n_heads4_seed1338/log2.txt

@@ -2,302 +2,302 @@ max_steps: 10000
 0 val loss 11.2663
 0 val perplexity 78147.0000
 0 train 11.263341 (lr=7.5000e-07) (hash(x)=150327452)
-100 val loss 8.6550
-100 val perplexity 5739.0356
-100 train 8.665512 (lr=7.5750e-05) (hash(x)=166441190)
-200 val loss 7.5622
-200 val perplexity 1924.0106
-200 train 7.762118 (lr=1.5000e-04) (hash(x)=166780046)
-300 val loss 7.4673
-300 val perplexity 1749.8013
-300 train 7.422103 (lr=1.4997e-04) (hash(x)=159835303)
-400 val loss 7.4181
-400 val perplexity 1665.9224
-400 train 7.179411 (lr=1.4986e-04) (hash(x)=155040610)
-500 val loss 7.3887
-500 val perplexity 1617.5394
-500 train 7.086969 (lr=1.4969e-04) (hash(x)=130190460)
-600 val loss 7.3440
-600 val perplexity 1546.8597
-600 train 7.368630 (lr=1.4945e-04) (hash(x)=155504036)
-700 val loss 7.3203
-700 val perplexity 1510.6406
-700 train 7.165936 (lr=1.4913e-04) (hash(x)=137347213)
-800 val loss 7.2788
-800 val perplexity 1449.2776
-800 train 7.137177 (lr=1.4876e-04) (hash(x)=143823248)
-900 val loss 7.2363
-900 val perplexity 1388.9182
-900 train 7.323705 (lr=1.4831e-04) (hash(x)=156260416)
-1000 val loss 7.2580
-1000 val perplexity 1419.4784
-1000 train 7.215821 (lr=1.4779e-04) (hash(x)=143734685)
-1100 val loss 7.3114
-1100 val perplexity 1497.2897
-1100 train 7.257675 (lr=1.4721e-04) (hash(x)=160013925)
-1200 val loss 7.3346
-1200 val perplexity 1532.4198
-1200 train 7.167082 (lr=1.4656e-04) (hash(x)=150678249)
-1300 val loss 7.2587
-1300 val perplexity 1420.4479
-1300 train 7.033679 (lr=1.4585e-04) (hash(x)=149073315)
-1400 val loss 7.2140
-1400 val perplexity 1358.2969
-1400 train 7.497233 (lr=1.4507e-04) (hash(x)=175802021)
-1500 val loss 7.1960
-1500 val perplexity 1334.0533
-1500 train 7.436237 (lr=1.4422e-04) (hash(x)=171034639)
-1600 val loss 7.1795
-1600 val perplexity 1312.1996
-1600 train 7.375320 (lr=1.4332e-04) (hash(x)=158681215)
-1700 val loss 7.1637
-1700 val perplexity 1291.6211
-1700 train 7.105666 (lr=1.4235e-04) (hash(x)=152116061)
-1800 val loss 7.1537
-1800 val perplexity 1278.7939
-1800 train 7.099223 (lr=1.4131e-04) (hash(x)=146108145)
-1900 val loss 7.1225
-1900 val perplexity 1239.5602
-1900 train 6.953999 (lr=1.4022e-04) (hash(x)=147598108)
-2000 val loss 7.1305
-2000 val perplexity 1249.5562
-2000 train 6.993223 (lr=1.3907e-04) (hash(x)=154996086)
-2100 val loss 7.1265
-2100 val perplexity 1244.4935
-2100 train 6.919074 (lr=1.3786e-04) (hash(x)=153396183)
-2200 val loss 7.0872
-2200 val perplexity 1196.5553
-2200 train 7.082876 (lr=1.3660e-04) (hash(x)=153885445)
-2300 val loss 7.0497
-2300 val perplexity 1152.4731
-2300 train 7.037508 (lr=1.3527e-04) (hash(x)=159666385)
-2400 val loss 7.0609
-2400 val perplexity 1165.5144
-2400 train 7.040409 (lr=1.3390e-04) (hash(x)=142353087)
-2500 val loss 7.0770
-2500 val perplexity 1184.3530
-2500 train 7.086487 (lr=1.3247e-04) (hash(x)=146491718)
-2600 val loss 7.0654
-2600 val perplexity 1170.7280
-2600 train 7.002256 (lr=1.3099e-04) (hash(x)=150750353)
-2700 val loss 7.0348
-2700 val perplexity 1135.4803
-2700 train 6.780224 (lr=1.2946e-04) (hash(x)=129849193)
-2800 val loss 7.0484
-2800 val perplexity 1151.0122
-2800 train 6.927810 (lr=1.2788e-04) (hash(x)=152767913)
-2900 val loss 7.0414
-2900 val perplexity 1143.0150
-2900 train 6.818239 (lr=1.2626e-04) (hash(x)=146531140)
-3000 val loss 7.0160
-3000 val perplexity 1114.3568
-3000 train 7.076974 (lr=1.2459e-04) (hash(x)=151562048)
-3100 val loss 6.9971
-3100 val perplexity 1093.4324
-3100 train 6.987829 (lr=1.2287e-04) (hash(x)=146001424)
-3200 val loss 6.9538
-3200 val perplexity 1047.0964
-3200 train 7.006474 (lr=1.2112e-04) (hash(x)=166486165)
-3300 val loss 6.9583
-3300 val perplexity 1051.8140
-3300 train 6.897943 (lr=1.1932e-04) (hash(x)=150866680)
-3400 val loss 6.9559
-3400 val perplexity 1049.2837
-3400 train 6.872110 (lr=1.1749e-04) (hash(x)=143900419)
-3500 val loss 6.9829
-3500 val perplexity 1078.0233
-3500 train 6.830639 (lr=1.1562e-04) (hash(x)=148845794)
-3600 val loss 6.9487
-3600 val perplexity 1041.7463
-3600 train 6.816872 (lr=1.1372e-04) (hash(x)=145667796)
-3700 val loss 6.9561
-3700 val perplexity 1049.5679
-3700 train 6.973476 (lr=1.1179e-04) (hash(x)=163563851)
-3800 val loss 6.9661
-3800 val perplexity 1060.0791
-3800 train 6.935971 (lr=1.0982e-04) (hash(x)=147488689)
-3900 val loss 6.9352
-3900 val perplexity 1027.8176
-3900 train 6.920702 (lr=1.0783e-04) (hash(x)=148186608)
-4000 val loss 6.8950
-4000 val perplexity 987.3676
-4000 train 6.763925 (lr=1.0581e-04) (hash(x)=142970187)
-4100 val loss 6.9069
-4100 val perplexity 999.1064
-4100 train 6.910237 (lr=1.0377e-04) (hash(x)=141584883)
-4200 val loss 6.8708
-4200 val perplexity 963.6940
-4200 train 6.701746 (lr=1.0171e-04) (hash(x)=145664585)
-4300 val loss 6.8682
-4300 val perplexity 961.1713
-4300 train 6.726182 (lr=9.9622e-05) (hash(x)=143736499)
-4400 val loss 6.8514
-4400 val perplexity 945.1613
-4400 train 6.674716 (lr=9.7520e-05) (hash(x)=151883322)
-4500 val loss 6.8442
-4500 val perplexity 938.4389
-4500 train 6.743720 (lr=9.5403e-05) (hash(x)=153904871)
-4600 val loss 6.8544
-4600 val perplexity 947.9967
-4600 train 6.910264 (lr=9.3273e-05) (hash(x)=154893521)
-4700 val loss 6.8130
-4700 val perplexity 909.5671
-4700 train 6.868094 (lr=9.1132e-05) (hash(x)=152323949)
-4800 val loss 6.7850
-4800 val perplexity 884.5167
-4800 train 6.726544 (lr=8.8982e-05) (hash(x)=154104619)
-4900 val loss 6.7750
-4900 val perplexity 875.6627
-4900 train 6.839969 (lr=8.6825e-05) (hash(x)=146311426)
-5000 val loss 6.7802
-5000 val perplexity 880.2397
-5000 train 6.811939 (lr=8.4663e-05) (hash(x)=156741847)
-5100 val loss 6.8186
-5100 val perplexity 914.6631
-5100 train 6.625038 (lr=8.2500e-05) (hash(x)=142086346)
-5200 val loss 6.8095
-5200 val perplexity 906.4241
-5200 train 6.609668 (lr=8.0337e-05) (hash(x)=150265428)
-5300 val loss 6.8070
-5300 val perplexity 904.1855
-5300 train 6.721405 (lr=7.8175e-05) (hash(x)=151339108)
-5400 val loss 6.7747
-5400 val perplexity 875.4039
-5400 train 6.778082 (lr=7.6018e-05) (hash(x)=154654372)
-5500 val loss 6.7680
-5500 val perplexity 869.5363
-5500 train 6.757663 (lr=7.3868e-05) (hash(x)=150575051)
-5600 val loss 6.7493
-5600 val perplexity 853.4575
-5600 train 6.652130 (lr=7.1727e-05) (hash(x)=140396423)
-5700 val loss 6.7432
-5700 val perplexity 848.2428
-5700 train 6.634809 (lr=6.9597e-05) (hash(x)=144678758)
-5800 val loss 6.7379
-5800 val perplexity 843.7819
-5800 train 6.839241 (lr=6.7480e-05) (hash(x)=151992743)
-5900 val loss 6.7451
-5900 val perplexity 849.8788
-5900 train 6.541656 (lr=6.5378e-05) (hash(x)=144396927)
-6000 val loss 6.7338
-6000 val perplexity 840.3497
-6000 train 6.688858 (lr=6.3294e-05) (hash(x)=165478625)
-6100 val loss 6.7536
-6100 val perplexity 857.1795
-6100 train 6.538942 (lr=6.1230e-05) (hash(x)=147088621)
-6200 val loss 6.7280
-6200 val perplexity 835.4413
-6200 train 6.533285 (lr=5.9188e-05) (hash(x)=140794994)
-6300 val loss 6.7068
-6300 val perplexity 817.9677
-6300 train 6.634907 (lr=5.7169e-05) (hash(x)=134780906)
-6400 val loss 6.6841
-6400 val perplexity 799.6194
-6400 train 6.686557 (lr=5.5177e-05) (hash(x)=149023655)
-6500 val loss 6.6806
-6500 val perplexity 796.8002
-6500 train 6.581760 (lr=5.3213e-05) (hash(x)=147497796)
-6600 val loss 6.6870
-6600 val perplexity 801.9154
-6600 train 6.556812 (lr=5.1279e-05) (hash(x)=152902689)
-6700 val loss 6.6708
-6700 val perplexity 789.0045
-6700 train 6.713040 (lr=4.9377e-05) (hash(x)=153846046)
-6800 val loss 6.6602
-6800 val perplexity 780.6896
-6800 train 6.773557 (lr=4.7509e-05) (hash(x)=158512738)
-6900 val loss 6.6533
-6900 val perplexity 775.3393
-6900 train 7.021071 (lr=4.5676e-05) (hash(x)=156849968)
-7000 val loss 6.6413
-7000 val perplexity 766.1014
-7000 train 6.475348 (lr=4.3882e-05) (hash(x)=142395855)
-7100 val loss 6.6401
-7100 val perplexity 765.1912
-7100 train 6.519809 (lr=4.2128e-05) (hash(x)=147114884)
-7200 val loss 6.6253
-7200 val perplexity 753.9130
-7200 train 6.653028 (lr=4.0414e-05) (hash(x)=156979839)
-7300 val loss 6.6312
-7300 val perplexity 758.3539
-7300 train 6.406144 (lr=3.8745e-05) (hash(x)=145584373)
-7400 val loss 6.6381
-7400 val perplexity 763.6647
-7400 train 6.395841 (lr=3.7120e-05) (hash(x)=141508204)
-7500 val loss 6.6150
-7500 val perplexity 746.1946
-7500 train 6.663067 (lr=3.5541e-05) (hash(x)=148803965)
-7600 val loss 6.6029
-7600 val perplexity 737.2593
-7600 train 6.661316 (lr=3.4011e-05) (hash(x)=151019676)
-7700 val loss 6.6007
-7700 val perplexity 735.5807
-7700 train 6.594604 (lr=3.2531e-05) (hash(x)=143155750)
-7800 val loss 6.5994
-7800 val perplexity 734.6207
-7800 train 6.602177 (lr=3.1102e-05) (hash(x)=152569653)
-7900 val loss 6.5959
-7900 val perplexity 732.0568
-7900 train 6.470634 (lr=2.9726e-05) (hash(x)=143519455)
-8000 val loss 6.5947
-8000 val perplexity 731.1951
-8000 train 6.702043 (lr=2.8405e-05) (hash(x)=161180944)
-8100 val loss 6.5933
-8100 val perplexity 730.1704
-8100 train 6.609785 (lr=2.7138e-05) (hash(x)=154107345)
-8200 val loss 6.5931
-8200 val perplexity 730.0322
-8200 train 6.603320 (lr=2.5929e-05) (hash(x)=152486517)
-8300 val loss 6.5839
-8300 val perplexity 723.3425
-8300 train 6.612710 (lr=2.4778e-05) (hash(x)=156167749)
-8400 val loss 6.5769
-8400 val perplexity 718.2834
-8400 train 6.650872 (lr=2.3686e-05) (hash(x)=149155006)
-8500 val loss 6.5718
-8500 val perplexity 714.6675
-8500 train 6.545753 (lr=2.2655e-05) (hash(x)=147844390)
-8600 val loss 6.5630
-8600 val perplexity 708.4192
-8600 train 6.785943 (lr=2.1685e-05) (hash(x)=165753320)
-8700 val loss 6.5601
-8700 val perplexity 706.3683
-8700 train 6.360015 (lr=2.0777e-05) (hash(x)=146079979)
-8800 val loss 6.5567
-8800 val perplexity 703.9554
-8800 train 6.748188 (lr=1.9933e-05) (hash(x)=172259509)
-8900 val loss 6.5590
-8900 val perplexity 705.5823
-8900 train 6.276247 (lr=1.9153e-05) (hash(x)=145148314)
-9000 val loss 6.5564
-9000 val perplexity 703.7161
-9000 train 6.482366 (lr=1.8439e-05) (hash(x)=144250633)
-9100 val loss 6.5477
-9100 val perplexity 697.6514
-9100 train 6.653626 (lr=1.7790e-05) (hash(x)=157219797)
-9200 val loss 6.5406
-9200 val perplexity 692.6686
-9200 train 6.510061 (lr=1.7208e-05) (hash(x)=142743778)
-9300 val loss 6.5375
-9300 val perplexity 690.5378
-9300 train 6.438917 (lr=1.6692e-05) (hash(x)=139669771)
-9400 val loss 6.5340
-9400 val perplexity 688.1193
-9400 train 6.500862 (lr=1.6245e-05) (hash(x)=145916843)
-9500 val loss 6.5310
-9500 val perplexity 686.0958
-9500 train 6.452840 (lr=1.5865e-05) (hash(x)=150196125)
-9600 val loss 6.5286
-9600 val perplexity 684.4724
-9600 train 6.808760 (lr=1.5554e-05) (hash(x)=160041419)
-9700 val loss 6.5288
-9700 val perplexity 684.5671
-9700 train 6.348433 (lr=1.5312e-05) (hash(x)=139931627)
-9800 val loss 6.5260
-9800 val perplexity 682.6550
-9800 train 6.475728 (lr=1.5139e-05) (hash(x)=150370792)
-9900 val loss 6.5226
-9900 val perplexity 680.3587
-9900 train 6.393685 (lr=1.5035e-05) (hash(x)=153014886)
-9999 val loss 6.5211
-9999 val perplexity 679.3424
+100 val loss 8.7777
+100 val perplexity 6488.2373
+100 train 8.785193 (lr=7.5750e-05) (hash(x)=166441190)
+200 val loss 7.5621
+200 val perplexity 1923.9125
+200 train 7.771656 (lr=1.5000e-04) (hash(x)=166780046)
+300 val loss 7.4551
+300 val perplexity 1728.7188
+300 train 7.411583 (lr=1.4997e-04) (hash(x)=159835303)
+400 val loss 7.4287
+400 val perplexity 1683.5997
+400 train 7.185664 (lr=1.4986e-04) (hash(x)=155040610)
+500 val loss 7.3802
+500 val perplexity 1603.9310
+500 train 7.082130 (lr=1.4969e-04) (hash(x)=130190460)
+600 val loss 7.2594
+600 val perplexity 1421.4446
+600 train 7.258858 (lr=1.4945e-04) (hash(x)=155504036)
+700 val loss 7.2145
+700 val perplexity 1359.0205
+700 train 7.076564 (lr=1.4913e-04) (hash(x)=137347213)
+800 val loss 7.1688
+800 val perplexity 1298.3121
+800 train 7.029332 (lr=1.4876e-04) (hash(x)=143823248)
+900 val loss 7.1819
+900 val perplexity 1315.4233
+900 train 7.241729 (lr=1.4831e-04) (hash(x)=156260416)
+1000 val loss 7.1672
+1000 val perplexity 1296.2437
+1000 train 7.118083 (lr=1.4779e-04) (hash(x)=143734685)
+1100 val loss 7.2244
+1100 val perplexity 1372.5431
+1100 train 7.164720 (lr=1.4721e-04) (hash(x)=160013925)
+1200 val loss 7.2210
+1200 val perplexity 1367.8552
+1200 train 6.965759 (lr=1.4656e-04) (hash(x)=150678249)
+1300 val loss 7.3149
+1300 val perplexity 1502.4619
+1300 train 7.109991 (lr=1.4585e-04) (hash(x)=149073315)
+1400 val loss 7.1382
+1400 val perplexity 1259.1836
+1400 train 7.414626 (lr=1.4507e-04) (hash(x)=175802021)
+1500 val loss 7.1499
+1500 val perplexity 1273.9269
+1500 train 7.380340 (lr=1.4422e-04) (hash(x)=171034639)
+1600 val loss 7.1045
+1600 val perplexity 1217.4829
+1600 train 7.302282 (lr=1.4332e-04) (hash(x)=158681215)
+1700 val loss 6.9913
+1700 val perplexity 1087.1879
+1700 train 6.945822 (lr=1.4235e-04) (hash(x)=152116061)
+1800 val loss 6.9835
+1800 val perplexity 1078.7386
+1800 train 6.930097 (lr=1.4131e-04) (hash(x)=146108145)
+1900 val loss 6.9365
+1900 val perplexity 1029.1526
+1900 train 6.772396 (lr=1.4022e-04) (hash(x)=147598108)
+2000 val loss 6.9423
+2000 val perplexity 1035.1649
+2000 train 6.790772 (lr=1.3907e-04) (hash(x)=154996086)
+2100 val loss 6.9431
+2100 val perplexity 1035.9935
+2100 train 6.721818 (lr=1.3786e-04) (hash(x)=153396183)
+2200 val loss 6.8870
+2200 val perplexity 979.4998
+2200 train 6.875124 (lr=1.3660e-04) (hash(x)=153885445)
+2300 val loss 6.8602
+2300 val perplexity 953.6027
+2300 train 6.850685 (lr=1.3527e-04) (hash(x)=159666385)
+2400 val loss 6.8501
+2400 val perplexity 943.9668
+2400 train 6.843330 (lr=1.3390e-04) (hash(x)=142353087)
+2500 val loss 6.8303
+2500 val perplexity 925.4475
+2500 train 6.832170 (lr=1.3247e-04) (hash(x)=146491718)
+2600 val loss 6.8423
+2600 val perplexity 936.5989
+2600 train 6.770065 (lr=1.3099e-04) (hash(x)=150750353)
+2700 val loss 6.8501
+2700 val perplexity 943.9425
+2700 train 6.593210 (lr=1.2946e-04) (hash(x)=129849193)
+2800 val loss 6.8524
+2800 val perplexity 946.1226
+2800 train 6.707880 (lr=1.2788e-04) (hash(x)=152767913)
+2900 val loss 6.8688
+2900 val perplexity 961.8150
+2900 train 6.658890 (lr=1.2626e-04) (hash(x)=146531140)
+3000 val loss 6.8300
+3000 val perplexity 925.2194
+3000 train 6.891155 (lr=1.2459e-04) (hash(x)=151562048)
+3100 val loss 6.7475
+3100 val perplexity 851.8889
+3100 train 6.743584 (lr=1.2287e-04) (hash(x)=146001424)
+3200 val loss 6.7215
+3200 val perplexity 830.0739
+3200 train 6.776397 (lr=1.2112e-04) (hash(x)=166486165)
+3300 val loss 6.7267
+3300 val perplexity 834.3564
+3300 train 6.665500 (lr=1.1932e-04) (hash(x)=150866680)
+3400 val loss 6.7039
+3400 val perplexity 815.5745
+3400 train 6.621673 (lr=1.1749e-04) (hash(x)=143900419)
+3500 val loss 6.6849
+3500 val perplexity 800.2004
+3500 train 6.524900 (lr=1.1562e-04) (hash(x)=148845794)
+3600 val loss 6.6943
+3600 val perplexity 807.8208
+3600 train 6.528556 (lr=1.1372e-04) (hash(x)=145667796)
+3700 val loss 6.7007
+3700 val perplexity 813.0059
+3700 train 6.712806 (lr=1.1179e-04) (hash(x)=163563851)
+3800 val loss 6.6819
+3800 val perplexity 797.8279
+3800 train 6.630010 (lr=1.0982e-04) (hash(x)=147488689)
+3900 val loss 6.6543
+3900 val perplexity 776.1143
+3900 train 6.648211 (lr=1.0783e-04) (hash(x)=148186608)
+4000 val loss 6.6528
+4000 val perplexity 774.9475
+4000 train 6.522067 (lr=1.0581e-04) (hash(x)=142970187)
+4100 val loss 6.6204
+4100 val perplexity 750.2690
+4100 train 6.626215 (lr=1.0377e-04) (hash(x)=141584883)
+4200 val loss 6.6039
+4200 val perplexity 737.9395
+4200 train 6.426653 (lr=1.0171e-04) (hash(x)=145664585)
+4300 val loss 6.6064
+4300 val perplexity 739.8495
+4300 train 6.458395 (lr=9.9622e-05) (hash(x)=143736499)
+4400 val loss 6.6145
+4400 val perplexity 745.8353
+4400 train 6.438884 (lr=9.7520e-05) (hash(x)=151883322)
+4500 val loss 6.6172
+4500 val perplexity 747.8121
+4500 train 6.509636 (lr=9.5403e-05) (hash(x)=153904871)
+4600 val loss 6.5999
+4600 val perplexity 735.0285
+4600 train 6.651087 (lr=9.3273e-05) (hash(x)=154893521)
+4700 val loss 6.5870
+4700 val perplexity 725.5793
+4700 train 6.638263 (lr=9.1132e-05) (hash(x)=152323949)
+4800 val loss 6.5637
+4800 val perplexity 708.8798
+4800 train 6.509049 (lr=8.8982e-05) (hash(x)=154104619)
+4900 val loss 6.5480
+4900 val perplexity 697.8463
+4900 train 6.612286 (lr=8.6825e-05) (hash(x)=146311426)
+5000 val loss 6.5250
+5000 val perplexity 682.0062
+5000 train 6.535800 (lr=8.4663e-05) (hash(x)=156741847)
+5100 val loss 6.5168
+5100 val perplexity 676.3936
+5100 train 6.322477 (lr=8.2500e-05) (hash(x)=142086346)
+5200 val loss 6.5240
+5200 val perplexity 681.3028
+5200 train 6.301110 (lr=8.0337e-05) (hash(x)=150265428)
+5300 val loss 6.5186
+5300 val perplexity 677.6261
+5300 train 6.414671 (lr=7.8175e-05) (hash(x)=151339108)
+5400 val loss 6.5056
+5400 val perplexity 668.8740
+5400 train 6.499859 (lr=7.6018e-05) (hash(x)=154654372)
+5500 val loss 6.4750
+5500 val perplexity 648.7399
+5500 train 6.470890 (lr=7.3868e-05) (hash(x)=150575051)
+5600 val loss 6.4751
+5600 val perplexity 648.7730
+5600 train 6.368444 (lr=7.1727e-05) (hash(x)=140396423)
+5700 val loss 6.4550
+5700 val perplexity 635.8600
+5700 train 6.341667 (lr=6.9597e-05) (hash(x)=144678758)
+5800 val loss 6.4481
+5800 val perplexity 631.4761
+5800 train 6.545341 (lr=6.7480e-05) (hash(x)=151992743)
+5900 val loss 6.4590
+5900 val perplexity 638.4401
+5900 train 6.260499 (lr=6.5378e-05) (hash(x)=144396927)
+6000 val loss 6.4424
+6000 val perplexity 627.9254
+6000 train 6.415303 (lr=6.3294e-05) (hash(x)=165478625)
+6100 val loss 6.4487
+6100 val perplexity 631.9029
+6100 train 6.227667 (lr=6.1230e-05) (hash(x)=147088621)
+6200 val loss 6.4394
+6200 val perplexity 626.0315
+6200 train 6.227491 (lr=5.9188e-05) (hash(x)=140794994)
+6300 val loss 6.4250
+6300 val perplexity 617.0739
+6300 train 6.354124 (lr=5.7169e-05) (hash(x)=134780906)
+6400 val loss 6.4077
+6400 val perplexity 606.5021
+6400 train 6.402531 (lr=5.5177e-05) (hash(x)=149023655)
+6500 val loss 6.4014
+6500 val perplexity 602.7130
+6500 train 6.293524 (lr=5.3213e-05) (hash(x)=147497796)
+6600 val loss 6.3992
+6600 val perplexity 601.3729
+6600 train 6.260563 (lr=5.1279e-05) (hash(x)=152902689)
+6700 val loss 6.4055
+6700 val perplexity 605.1715
+6700 train 6.442645 (lr=4.9377e-05) (hash(x)=153846046)
+6800 val loss 6.3892
+6800 val perplexity 595.3674
+6800 train 6.513471 (lr=4.7509e-05) (hash(x)=158512738)
+6900 val loss 6.3770
+6900 val perplexity 588.1899
+6900 train 6.826787 (lr=4.5676e-05) (hash(x)=156849968)
+7000 val loss 6.3729
+7000 val perplexity 585.7264
+7000 train 6.216920 (lr=4.3882e-05) (hash(x)=142395855)
+7100 val loss 6.3687
+7100 val perplexity 583.2773
+7100 train 6.247492 (lr=4.2128e-05) (hash(x)=147114884)
+7200 val loss 6.3600
+7200 val perplexity 578.2189
+7200 train 6.391733 (lr=4.0414e-05) (hash(x)=156979839)
+7300 val loss 6.3694
+7300 val perplexity 583.7050
+7300 train 6.152255 (lr=3.8745e-05) (hash(x)=145584373)
+7400 val loss 6.3727
+7400 val perplexity 585.6144
+7400 train 6.115252 (lr=3.7120e-05) (hash(x)=141508204)
+7500 val loss 6.3690
+7500 val perplexity 583.4774
+7500 train 6.418751 (lr=3.5541e-05) (hash(x)=148803965)
+7600 val loss 6.3510
+7600 val perplexity 573.0728
+7600 train 6.423841 (lr=3.4011e-05) (hash(x)=151019676)
+7700 val loss 6.3513
+7700 val perplexity 573.2105
+7700 train 6.371610 (lr=3.2531e-05) (hash(x)=143155750)
+7800 val loss 6.3405
+7800 val perplexity 567.0922
+7800 train 6.346223 (lr=3.1102e-05) (hash(x)=152569653)
+7900 val loss 6.3456
+7900 val perplexity 570.0016
+7900 train 6.216721 (lr=2.9726e-05) (hash(x)=143519455)
+8000 val loss 6.3375
+8000 val perplexity 565.3734
+8000 train 6.459105 (lr=2.8405e-05) (hash(x)=161180944)
+8100 val loss 6.3372
+8100 val perplexity 565.2216
+8100 train 6.351504 (lr=2.7138e-05) (hash(x)=154107345)
+8200 val loss 6.3397
+8200 val perplexity 566.6518
+8200 train 6.360279 (lr=2.5929e-05) (hash(x)=152486517)
+8300 val loss 6.3294
+8300 val perplexity 560.8088
+8300 train 6.357333 (lr=2.4778e-05) (hash(x)=156167749)
+8400 val loss 6.3236
+8400 val perplexity 557.5955
+8400 train 6.391531 (lr=2.3686e-05) (hash(x)=149155006)
+8500 val loss 6.3202
+8500 val perplexity 555.6780
+8500 train 6.275575 (lr=2.2655e-05) (hash(x)=147844390)
+8600 val loss 6.3152
+8600 val perplexity 552.9109
+8600 train 6.559919 (lr=2.1685e-05) (hash(x)=165753320)
+8700 val loss 6.3136
+8700 val perplexity 552.0432
+8700 train 6.108851 (lr=2.0777e-05) (hash(x)=146079979)
+8800 val loss 6.3150
+8800 val perplexity 552.8068
+8800 train 6.508952 (lr=1.9933e-05) (hash(x)=172259509)
+8900 val loss 6.3184
+8900 val perplexity 554.6667
+8900 train 6.030042 (lr=1.9153e-05) (hash(x)=145148314)
+9000 val loss 6.3184
+9000 val perplexity 554.6789
+9000 train 6.243028 (lr=1.8439e-05) (hash(x)=144250633)
+9100 val loss 6.3110
+9100 val perplexity 550.6165
+9100 train 6.421357 (lr=1.7790e-05) (hash(x)=157219797)
+9200 val loss 6.2984
+9200 val perplexity 543.7205
+9200 train 6.287928 (lr=1.7208e-05) (hash(x)=142743778)
+9300 val loss 6.2968
+9300 val perplexity 542.8534
+9300 train 6.204850 (lr=1.6692e-05) (hash(x)=139669771)
+9400 val loss 6.2941
+9400 val perplexity 541.3662
+9400 train 6.284011 (lr=1.6245e-05) (hash(x)=145916843)
+9500 val loss 6.2922
+9500 val perplexity 540.3427
+9500 train 6.226115 (lr=1.5865e-05) (hash(x)=150196125)
+9600 val loss 6.2924
+9600 val perplexity 540.4338
+9600 train 6.589438 (lr=1.5554e-05) (hash(x)=160041419)
+9700 val loss 6.2922
+9700 val perplexity 540.3143
+9700 train 6.096705 (lr=1.5312e-05) (hash(x)=139931627)
+9800 val loss 6.2895
+9800 val perplexity 538.9012
+9800 train 6.239325 (lr=1.5139e-05) (hash(x)=150370792)
+9900 val loss 6.2896
+9900 val perplexity 538.9210
+9900 train 6.155778 (lr=1.5035e-05) (hash(x)=153014886)
+9999 val loss 6.2886
+9999 val perplexity 538.4057

attention_kindselective_n_heads4_seed1338/model_02500.pt

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:d791b9df5a73ce7373c8c3633d7ef61a92efffb082c40753771a5af9686616b6
+oid sha256:d95e1fb93ae7eb339b0abe0762d19c1ada616bfdb4620397f8ab04ccf7f18393
 size 92843394

attention_kindselective_n_heads4_seed1338/model_05000.pt

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:ec252cc9736a9564ca93925735d67637349118511176e491ebd99ebc5ab280e2
+oid sha256:a6a87e17518a8ee6ac9c9e67f765ee942a503fcfcafd73a884926483d93e544a
 size 92843394

attention_kindselective_n_heads4_seed1338/model_07500.pt

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:e2c3687223cc276736023709fb477bf883ed750f52a3733196fbe90fb0c3d493
+oid sha256:fe708e276b43c181ed8d4484ae990f2df21dc02afeedda5ba49462fbfbc68fcf
 size 92843394

attention_kindselective_n_heads4_seed1338/model_09999.pt

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:d22cc128cf4ed809ae94894629d9b6cb11500983fdc2887db59483fc59d84406
+oid sha256:e9f0b09e98b5430893e91ea690dfb350667d0a5c5947ea2696e639d62776ecc9
 size 92843394

attention_kindselective_n_heads4_seed1338/optimizer_02500.pt

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:4957bc0a1124b62de2ad847f9bab34b5be4f4bf18c8d7e6e908770af55cbbff8
+oid sha256:9009f01dc7ff0e740d4cb3f16c47753a63d0b67ba1638f0d6d86fe8958087103
 size 179406214

attention_kindselective_n_heads4_seed1338/optimizer_05000.pt

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:b5ec9251ee7644d9c14036dded38a955e9e61897405fc03587dff8d4614fa33b
+oid sha256:993dd71322a0ba12777ebdfaea3bd660675b164837b455e94ea08d218041e7fe
 size 179406214

attention_kindselective_n_heads4_seed1338/optimizer_07500.pt

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:71e112b5d7a75337aac523f9377464968f888fd2fb95b1c001e11144064dad91
+oid sha256:b6baabd14ddfef800a950a6f88b02d5fca5ca757bc537ce6d93bba62b4d80eda
 size 179406214

attention_kindselective_n_heads4_seed1338/optimizer_09999.pt

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:d7af355bdcbdda09d35c074c218779f17c5de3a922e738ba2fb1ee7c2802edb5
+oid sha256:267f47f0cd7666df01048f45ff2935cc98edb5bc13e50682d842989caa528a21
 size 179406214