Upload folder using huggingface_hub

attention_kindselective_n_heads2_seed1338/args.json

@@ -1 +1 @@
-{"hellaswag": true, "attention_kind": "selective", "log_dir": "wider_is_better_4/attention_kindselective_n_heads2_seed1338", "resume_checkpoint": null, "resume_optimizer": false, "add_a_head": false, "add_head_to_start": true, "new_head_init": "normal", "n_heads": 2, "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": 5e-05, "decay_lr": true, "readout_zero_init": false, "query_zero_init": false, "l1_loss": false, "debugpy": false, "key": "0.5e-4_30720_2_1338", "n_embd": 128}
+{"hellaswag": true, "attention_kind": "selective", "log_dir": "wider_is_better_4/attention_kindselective_n_heads2_seed1338", "resume_checkpoint": null, "resume_optimizer": false, "add_a_head": false, "add_head_to_start": true, "new_head_init": "normal", "n_heads": 2, "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": 3e-05, "decay_lr": true, "readout_zero_init": false, "query_zero_init": false, "l1_loss": false, "debugpy": false, "key": "3e-5_30720_2_1338", "n_embd": 128}

attention_kindselective_n_heads2_seed1338/log2.txt

@@ -1,303 +1,303 @@
 max_steps: 10000
 0 val loss 11.7630
 0 val perplexity 128408.9922
-0 train 11.769258 (lr=2.5000e-07) (hash(x)=150327452)
-100 val loss 10.2404
-100 val perplexity 28012.3633
-100 train 10.227105 (lr=2.5250e-05) (hash(x)=166441190)
-200 val loss 8.6924
-200 val perplexity 5957.4800
-200 train 8.835118 (lr=5.0000e-05) (hash(x)=166780046)
-300 val loss 7.7964
-300 val perplexity 2431.7542
-300 train 7.786602 (lr=4.9988e-05) (hash(x)=159835303)
-400 val loss 7.6744
-400 val perplexity 2152.5623
-400 train 7.489794 (lr=4.9954e-05) (hash(x)=155040610)
-500 val loss 7.6406
-500 val perplexity 2080.9221
-500 train 7.391425 (lr=4.9896e-05) (hash(x)=130190460)
-600 val loss 7.6032
-600 val perplexity 2004.5076
-600 train 7.638520 (lr=4.9815e-05) (hash(x)=155504036)
-700 val loss 7.5646
-700 val perplexity 1928.7394
-700 train 7.406923 (lr=4.9712e-05) (hash(x)=137347213)
-800 val loss 7.5406
-800 val perplexity 1882.9725
-800 train 7.397386 (lr=4.9585e-05) (hash(x)=143823248)
-900 val loss 7.5339
-900 val perplexity 1870.4193
-900 train 7.654196 (lr=4.9436e-05) (hash(x)=156260416)
-1000 val loss 7.5107
-1000 val perplexity 1827.5564
-1000 train 7.477935 (lr=4.9264e-05) (hash(x)=143734685)
-1100 val loss 7.4936
-1100 val perplexity 1796.4606
-1100 train 7.441096 (lr=4.9070e-05) (hash(x)=160013925)
-1200 val loss 7.4976
-1200 val perplexity 1803.6320
-1200 train 7.258204 (lr=4.8854e-05) (hash(x)=150678249)
-1300 val loss 7.4832
-1300 val perplexity 1777.9482
-1300 train 7.282100 (lr=4.8616e-05) (hash(x)=149073315)
-1400 val loss 7.4629
-1400 val perplexity 1742.1178
-1400 train 7.883915 (lr=4.8356e-05) (hash(x)=175802021)
-1500 val loss 7.4301
-1500 val perplexity 1685.9038
-1500 train 7.686608 (lr=4.8074e-05) (hash(x)=171034639)
-1600 val loss 7.4090
-1600 val perplexity 1650.8046
-1600 train 7.636811 (lr=4.7772e-05) (hash(x)=158681215)
-1700 val loss 7.3883
-1700 val perplexity 1616.9302
-1700 train 7.329289 (lr=4.7448e-05) (hash(x)=152116061)
-1800 val loss 7.3710
-1800 val perplexity 1589.1628
-1800 train 7.307854 (lr=4.7105e-05) (hash(x)=146108145)
-1900 val loss 7.3425
-1900 val perplexity 1544.5565
-1900 train 7.182374 (lr=4.6741e-05) (hash(x)=147598108)
-2000 val loss 7.3217
-2000 val perplexity 1512.8422
-2000 train 7.187046 (lr=4.6357e-05) (hash(x)=154996086)
-2100 val loss 7.3005
-2100 val perplexity 1481.1086
-2100 train 7.081825 (lr=4.5954e-05) (hash(x)=153396183)
-2200 val loss 7.2452
-2200 val perplexity 1401.3156
-2200 train 7.233769 (lr=4.5532e-05) (hash(x)=153885445)
-2300 val loss 7.2313
-2300 val perplexity 1382.0316
-2300 train 7.206238 (lr=4.5091e-05) (hash(x)=159666385)
-2400 val loss 7.1577
-2400 val perplexity 1283.9122
-2400 train 7.111083 (lr=4.4633e-05) (hash(x)=142353087)
-2500 val loss 7.1407
-2500 val perplexity 1262.2935
-2500 train 7.140073 (lr=4.4156e-05) (hash(x)=146491718)
-2600 val loss 7.1162
-2600 val perplexity 1231.7444
-2600 train 7.048321 (lr=4.3663e-05) (hash(x)=150750353)
-2700 val loss 7.0994
-2700 val perplexity 1211.1899
-2700 train 6.828227 (lr=4.3153e-05) (hash(x)=129849193)
-2800 val loss 7.0820
-2800 val perplexity 1190.2938
-2800 train 6.952722 (lr=4.2627e-05) (hash(x)=152767913)
-2900 val loss 7.0613
-2900 val perplexity 1165.9741
-2900 train 6.857185 (lr=4.2085e-05) (hash(x)=146531140)
-3000 val loss 7.0340
-3000 val perplexity 1134.5959
-3000 train 7.094491 (lr=4.1529e-05) (hash(x)=151562048)
-3100 val loss 6.9862
-3100 val perplexity 1081.6365
-3100 train 6.972751 (lr=4.0957e-05) (hash(x)=146001424)
-3200 val loss 6.9622
-3200 val perplexity 1055.9473
-3200 train 6.967948 (lr=4.0373e-05) (hash(x)=166486165)
-3300 val loss 6.9229
-3300 val perplexity 1015.2311
-3300 train 6.851268 (lr=3.9775e-05) (hash(x)=150866680)
-3400 val loss 6.8889
-3400 val perplexity 981.2885
-3400 train 6.802827 (lr=3.9164e-05) (hash(x)=143900419)
-3500 val loss 6.8543
-3500 val perplexity 947.9136
-3500 train 6.685315 (lr=3.8541e-05) (hash(x)=148845794)
-3600 val loss 6.8242
-3600 val perplexity 919.8097
-3600 train 6.676590 (lr=3.7907e-05) (hash(x)=145667796)
-3700 val loss 6.7948
-3700 val perplexity 893.1960
-3700 train 6.810184 (lr=3.7262e-05) (hash(x)=163563851)
-3800 val loss 6.7596
-3800 val perplexity 862.3146
-3800 train 6.712791 (lr=3.6608e-05) (hash(x)=147488689)
-3900 val loss 6.7088
-3900 val perplexity 819.6110
-3900 train 6.699515 (lr=3.5944e-05) (hash(x)=148186608)
-4000 val loss 6.6668
-4000 val perplexity 785.9120
-4000 train 6.524370 (lr=3.5271e-05) (hash(x)=142970187)
-4100 val loss 6.6322
-4100 val perplexity 759.1307
-4100 train 6.648249 (lr=3.4590e-05) (hash(x)=141584883)
-4200 val loss 6.6028
-4200 val perplexity 737.1482
-4200 train 6.413905 (lr=3.3902e-05) (hash(x)=145664585)
-4300 val loss 6.5812
-4300 val perplexity 721.4269
-4300 train 6.433772 (lr=3.3207e-05) (hash(x)=143736499)
-4400 val loss 6.5652
-4400 val perplexity 709.9598
-4400 train 6.385487 (lr=3.2507e-05) (hash(x)=151883322)
-4500 val loss 6.5521
-4500 val perplexity 700.7370
-4500 train 6.444251 (lr=3.1801e-05) (hash(x)=153904871)
-4600 val loss 6.5329
-4600 val perplexity 687.4182
-4600 train 6.596064 (lr=3.1091e-05) (hash(x)=154893521)
-4700 val loss 6.4969
-4700 val perplexity 663.1146
-4700 train 6.560022 (lr=3.0377e-05) (hash(x)=152323949)
-4800 val loss 6.4743
-4800 val perplexity 648.2936
-4800 train 6.412653 (lr=2.9661e-05) (hash(x)=154104619)
-4900 val loss 6.4596
-4900 val perplexity 638.8070
-4900 train 6.544138 (lr=2.8942e-05) (hash(x)=146311426)
-5000 val loss 6.4390
-5000 val perplexity 625.7906
-5000 train 6.456110 (lr=2.8221e-05) (hash(x)=156741847)
-5100 val loss 6.4319
-5100 val perplexity 621.3528
-5100 train 6.222180 (lr=2.7500e-05) (hash(x)=142086346)
-5200 val loss 6.4287
-5200 val perplexity 619.3978
-5200 train 6.202271 (lr=2.6779e-05) (hash(x)=150265428)
-5300 val loss 6.4182
-5300 val perplexity 612.8869
-5300 train 6.315850 (lr=2.6058e-05) (hash(x)=151339108)
-5400 val loss 6.4048
-5400 val perplexity 604.7270
-5400 train 6.392514 (lr=2.5339e-05) (hash(x)=154654372)
-5500 val loss 6.3853
-5500 val perplexity 593.0718
-5500 train 6.397058 (lr=2.4623e-05) (hash(x)=150575051)
-5600 val loss 6.3629
-5600 val perplexity 579.9388
-5600 train 6.260410 (lr=2.3909e-05) (hash(x)=140396423)
-5700 val loss 6.3533
-5700 val perplexity 574.3911
-5700 train 6.218089 (lr=2.3199e-05) (hash(x)=144678758)
-5800 val loss 6.3443
-5800 val perplexity 569.2311
-5800 train 6.426583 (lr=2.2493e-05) (hash(x)=151992743)
-5900 val loss 6.3338
-5900 val perplexity 563.2771
-5900 train 6.116439 (lr=2.1793e-05) (hash(x)=144396927)
-6000 val loss 6.3250
-6000 val perplexity 558.3319
-6000 train 6.296118 (lr=2.1098e-05) (hash(x)=165478625)
-6100 val loss 6.3288
-6100 val perplexity 560.4827
-6100 train 6.075109 (lr=2.0410e-05) (hash(x)=147088621)
-6200 val loss 6.3158
-6200 val perplexity 553.2230
-6200 train 6.100374 (lr=1.9729e-05) (hash(x)=140794994)
-6300 val loss 6.2978
-6300 val perplexity 543.3687
-6300 train 6.230536 (lr=1.9056e-05) (hash(x)=134780906)
-6400 val loss 6.2830
-6400 val perplexity 535.3809
-6400 train 6.271345 (lr=1.8392e-05) (hash(x)=149023655)
-6500 val loss 6.2739
-6500 val perplexity 530.5635
-6500 train 6.142905 (lr=1.7738e-05) (hash(x)=147497796)
-6600 val loss 6.2718
-6600 val perplexity 529.4250
-6600 train 6.123541 (lr=1.7093e-05) (hash(x)=152902689)
-6700 val loss 6.2679
-6700 val perplexity 527.3727
-6700 train 6.298330 (lr=1.6459e-05) (hash(x)=153846046)
-6800 val loss 6.2524
-6800 val perplexity 519.2451
-6800 train 6.383592 (lr=1.5836e-05) (hash(x)=158512738)
-6900 val loss 6.2441
-6900 val perplexity 514.9430
-6900 train 6.754239 (lr=1.5225e-05) (hash(x)=156849968)
-7000 val loss 6.2357
-7000 val perplexity 510.6797
-7000 train 6.079944 (lr=1.4627e-05) (hash(x)=142395855)
-7100 val loss 6.2332
-7100 val perplexity 509.3864
-7100 train 6.096469 (lr=1.4043e-05) (hash(x)=147114884)
-7200 val loss 6.2246
-7200 val perplexity 505.0182
-7200 train 6.257214 (lr=1.3471e-05) (hash(x)=156979839)
-7300 val loss 6.2214
-7300 val perplexity 503.4091
-7300 train 5.995510 (lr=1.2915e-05) (hash(x)=145584373)
-7400 val loss 6.2212
-7400 val perplexity 503.3155
-7400 train 5.960420 (lr=1.2373e-05) (hash(x)=141508204)
-7500 val loss 6.2089
-7500 val perplexity 497.1344
-7500 train 6.253492 (lr=1.1847e-05) (hash(x)=148803965)
-7600 val loss 6.1995
-7600 val perplexity 492.4985
-7600 train 6.262319 (lr=1.1337e-05) (hash(x)=151019676)
-7700 val loss 6.1925
-7700 val perplexity 489.0664
-7700 train 6.230704 (lr=1.0844e-05) (hash(x)=143155750)
-7800 val loss 6.1897
-7800 val perplexity 487.7184
-7800 train 6.212282 (lr=1.0367e-05) (hash(x)=152569653)
-7900 val loss 6.1841
-7900 val perplexity 484.9865
-7900 train 6.054117 (lr=9.9088e-06) (hash(x)=143519455)
-8000 val loss 6.1830
-8000 val perplexity 484.4512
-8000 train 6.301287 (lr=9.4682e-06) (hash(x)=161180944)
-8100 val loss 6.1832
-8100 val perplexity 484.5494
-8100 train 6.211595 (lr=9.0461e-06) (hash(x)=154107345)
-8200 val loss 6.1815
-8200 val perplexity 483.7320
-8200 train 6.209146 (lr=8.6430e-06) (hash(x)=152486517)
-8300 val loss 6.1705
-8300 val perplexity 478.4457
-8300 train 6.188221 (lr=8.2593e-06) (hash(x)=156167749)
-8400 val loss 6.1650
-8400 val perplexity 475.7902
-8400 train 6.239934 (lr=7.8953e-06) (hash(x)=149155006)
-8500 val loss 6.1602
-8500 val perplexity 473.5081
-8500 train 6.126613 (lr=7.5515e-06) (hash(x)=147844390)
-8600 val loss 6.1572
-8600 val perplexity 472.1252
-8600 train 6.423381 (lr=7.2282e-06) (hash(x)=165753320)
-8700 val loss 6.1558
-8700 val perplexity 471.4586
-8700 train 5.938447 (lr=6.9257e-06) (hash(x)=146079979)
-8800 val loss 6.1564
-8800 val perplexity 471.7098
-8800 train 6.352519 (lr=6.6444e-06) (hash(x)=172259509)
-8900 val loss 6.1557
-8900 val perplexity 471.4083
-8900 train 5.836442 (lr=6.3845e-06) (hash(x)=145148314)
-9000 val loss 6.1548
-9000 val perplexity 470.9663
-9000 train 6.078220 (lr=6.1462e-06) (hash(x)=144250633)
-9100 val loss 6.1462
-9100 val perplexity 466.9475
-9100 train 6.256772 (lr=5.9300e-06) (hash(x)=157219797)
-9200 val loss 6.1410
-9200 val perplexity 464.5114
-9200 train 6.155202 (lr=5.7359e-06) (hash(x)=142743778)
-9300 val loss 6.1387
-9300 val perplexity 463.4443
-9300 train 6.043135 (lr=5.5641e-06) (hash(x)=139669771)
-9400 val loss 6.1351
-9400 val perplexity 461.7762
-9400 train 6.132237 (lr=5.4149e-06) (hash(x)=145916843)
-9500 val loss 6.1342
-9500 val perplexity 461.3840
-9500 train 6.058453 (lr=5.2884e-06) (hash(x)=150196125)
-9600 val loss 6.1338
-9600 val perplexity 461.1818
-9600 train 6.431315 (lr=5.1847e-06) (hash(x)=160041419)
-9700 val loss 6.1329
-9700 val perplexity 460.7543
-9700 train 5.923302 (lr=5.1040e-06) (hash(x)=139931627)
-9800 val loss 6.1306
-9800 val perplexity 459.7270
-9800 train 6.095959 (lr=5.0462e-06) (hash(x)=150370792)
-9900 val loss 6.1289
-9900 val perplexity 458.9510
-9900 train 5.972460 (lr=5.0116e-06) (hash(x)=153014886)
-9999 val loss 6.1289
-9999 val perplexity 458.9519
+0 train 11.769258 (lr=1.5000e-07) (hash(x)=150327452)
+100 val loss 10.4010
+100 val perplexity 32894.0078
+100 train 10.388559 (lr=1.5150e-05) (hash(x)=166441190)
+200 val loss 9.3220
+200 val perplexity 11180.8262
+200 train 9.420354 (lr=3.0000e-05) (hash(x)=166780046)
+300 val loss 8.1198
+300 val perplexity 3360.3025
+300 train 8.114033 (lr=2.9993e-05) (hash(x)=159835303)
+400 val loss 7.7643
+400 val perplexity 2355.1169
+400 train 7.589100 (lr=2.9972e-05) (hash(x)=155040610)
+500 val loss 7.6660
+500 val perplexity 2134.4812
+500 train 7.423435 (lr=2.9938e-05) (hash(x)=130190460)
+600 val loss 7.6109
+600 val perplexity 2020.0375
+600 train 7.641937 (lr=2.9889e-05) (hash(x)=155504036)
+700 val loss 7.5711
+700 val perplexity 1941.2490
+700 train 7.414425 (lr=2.9827e-05) (hash(x)=137347213)
+800 val loss 7.5448
+800 val perplexity 1890.9220
+800 train 7.399529 (lr=2.9751e-05) (hash(x)=143823248)
+900 val loss 7.5278
+900 val perplexity 1859.0282
+900 train 7.648718 (lr=2.9662e-05) (hash(x)=156260416)
+1000 val loss 7.5065
+1000 val perplexity 1819.8811
+1000 train 7.474764 (lr=2.9558e-05) (hash(x)=143734685)
+1100 val loss 7.4984
+1100 val perplexity 1805.1945
+1100 train 7.444295 (lr=2.9442e-05) (hash(x)=160013925)
+1200 val loss 7.4866
+1200 val perplexity 1784.0110
+1200 train 7.261100 (lr=2.9312e-05) (hash(x)=150678249)
+1300 val loss 7.4759
+1300 val perplexity 1764.9208
+1300 train 7.281197 (lr=2.9169e-05) (hash(x)=149073315)
+1400 val loss 7.4569
+1400 val perplexity 1731.7219
+1400 train 7.886369 (lr=2.9013e-05) (hash(x)=175802021)
+1500 val loss 7.4343
+1500 val perplexity 1693.1479
+1500 train 7.685257 (lr=2.8845e-05) (hash(x)=171034639)
+1600 val loss 7.4127
+1600 val perplexity 1656.9448
+1600 train 7.640902 (lr=2.8663e-05) (hash(x)=158681215)
+1700 val loss 7.3985
+1700 val perplexity 1633.6101
+1700 train 7.341957 (lr=2.8469e-05) (hash(x)=152116061)
+1800 val loss 7.3886
+1800 val perplexity 1617.4152
+1800 train 7.323990 (lr=2.8263e-05) (hash(x)=146108145)
+1900 val loss 7.3732
+1900 val perplexity 1592.7101
+1900 train 7.219169 (lr=2.8044e-05) (hash(x)=147598108)
+2000 val loss 7.3618
+2000 val perplexity 1574.7190
+2000 train 7.232282 (lr=2.7814e-05) (hash(x)=154996086)
+2100 val loss 7.3471
+2100 val perplexity 1551.6771
+2100 train 7.141953 (lr=2.7572e-05) (hash(x)=153396183)
+2200 val loss 7.3012
+2200 val perplexity 1482.1154
+2200 train 7.291854 (lr=2.7319e-05) (hash(x)=153885445)
+2300 val loss 7.2835
+2300 val perplexity 1456.0382
+2300 train 7.268628 (lr=2.7055e-05) (hash(x)=159666385)
+2400 val loss 7.2455
+2400 val perplexity 1401.8269
+2400 train 7.193364 (lr=2.6780e-05) (hash(x)=142353087)
+2500 val loss 7.2123
+2500 val perplexity 1355.9503
+2500 train 7.208140 (lr=2.6494e-05) (hash(x)=146491718)
+2600 val loss 7.1809
+2600 val perplexity 1314.0718
+2600 train 7.118464 (lr=2.6198e-05) (hash(x)=150750353)
+2700 val loss 7.1504
+2700 val perplexity 1274.6262
+2700 train 6.869782 (lr=2.5892e-05) (hash(x)=129849193)
+2800 val loss 7.1357
+2800 val perplexity 1256.0383
+2800 train 6.996274 (lr=2.5576e-05) (hash(x)=152767913)
+2900 val loss 7.1052
+2900 val perplexity 1218.3030
+2900 train 6.898202 (lr=2.5251e-05) (hash(x)=146531140)
+3000 val loss 7.0670
+3000 val perplexity 1172.6589
+3000 train 7.135310 (lr=2.4917e-05) (hash(x)=151562048)
+3100 val loss 7.0301
+3100 val perplexity 1130.1343
+3100 train 7.015220 (lr=2.4574e-05) (hash(x)=146001424)
+3200 val loss 7.0035
+3200 val perplexity 1100.5248
+3200 train 7.012419 (lr=2.4224e-05) (hash(x)=166486165)
+3300 val loss 6.9697
+3300 val perplexity 1063.8522
+3300 train 6.894574 (lr=2.3865e-05) (hash(x)=150866680)
+3400 val loss 6.9519
+3400 val perplexity 1045.1385
+3400 train 6.863424 (lr=2.3498e-05) (hash(x)=143900419)
+3500 val loss 6.9233
+3500 val perplexity 1015.6485
+3500 train 6.757936 (lr=2.3125e-05) (hash(x)=148845794)
+3600 val loss 6.9006
+3600 val perplexity 992.9057
+3600 train 6.751880 (lr=2.2744e-05) (hash(x)=145667796)
+3700 val loss 6.8779
+3700 val perplexity 970.6259
+3700 train 6.897323 (lr=2.2357e-05) (hash(x)=163563851)
+3800 val loss 6.8582
+3800 val perplexity 951.6208
+3800 train 6.815534 (lr=2.1965e-05) (hash(x)=147488689)
+3900 val loss 6.8149
+3900 val perplexity 911.3462
+3900 train 6.802123 (lr=2.1566e-05) (hash(x)=148186608)
+4000 val loss 6.7898
+4000 val perplexity 888.6940
+4000 train 6.638687 (lr=2.1162e-05) (hash(x)=142970187)
+4100 val loss 6.7610
+4100 val perplexity 863.5334
+4100 train 6.777088 (lr=2.0754e-05) (hash(x)=141584883)
+4200 val loss 6.7359
+4200 val perplexity 842.1255
+4200 train 6.537104 (lr=2.0341e-05) (hash(x)=145664585)
+4300 val loss 6.7157
+4300 val perplexity 825.2897
+4300 train 6.550407 (lr=1.9924e-05) (hash(x)=143736499)
+4400 val loss 6.7043
+4400 val perplexity 815.9366
+4400 train 6.524540 (lr=1.9504e-05) (hash(x)=151883322)
+4500 val loss 6.6853
+4500 val perplexity 800.5656
+4500 train 6.591630 (lr=1.9081e-05) (hash(x)=153904871)
+4600 val loss 6.6625
+4600 val perplexity 782.4913
+4600 train 6.722117 (lr=1.8655e-05) (hash(x)=154893521)
+4700 val loss 6.6334
+4700 val perplexity 760.0438
+4700 train 6.709569 (lr=1.8226e-05) (hash(x)=152323949)
+4800 val loss 6.6148
+4800 val perplexity 746.0839
+4800 train 6.544276 (lr=1.7796e-05) (hash(x)=154104619)
+4900 val loss 6.5995
+4900 val perplexity 734.7310
+4900 train 6.678655 (lr=1.7365e-05) (hash(x)=146311426)
+5000 val loss 6.5837
+5000 val perplexity 723.2397
+5000 train 6.608577 (lr=1.6933e-05) (hash(x)=156741847)
+5100 val loss 6.5716
+5100 val perplexity 714.5206
+5100 train 6.365155 (lr=1.6500e-05) (hash(x)=142086346)
+5200 val loss 6.5636
+5200 val perplexity 708.8334
+5200 train 6.348958 (lr=1.6067e-05) (hash(x)=150265428)
+5300 val loss 6.5557
+5300 val perplexity 703.2093
+5300 train 6.452664 (lr=1.5635e-05) (hash(x)=151339108)
+5400 val loss 6.5387
+5400 val perplexity 691.3912
+5400 train 6.534310 (lr=1.5204e-05) (hash(x)=154654372)
+5500 val loss 6.5184
+5500 val perplexity 677.4675
+5500 train 6.529605 (lr=1.4774e-05) (hash(x)=150575051)
+5600 val loss 6.5039
+5600 val perplexity 667.7398
+5600 train 6.391018 (lr=1.4345e-05) (hash(x)=140396423)
+5700 val loss 6.4901
+5700 val perplexity 658.5672
+5700 train 6.374229 (lr=1.3919e-05) (hash(x)=144678758)
+5800 val loss 6.4787
+5800 val perplexity 651.1094
+5800 train 6.583964 (lr=1.3496e-05) (hash(x)=151992743)
+5900 val loss 6.4715
+5900 val perplexity 646.4711
+5900 train 6.251740 (lr=1.3076e-05) (hash(x)=144396927)
+6000 val loss 6.4619
+6000 val perplexity 640.2845
+6000 train 6.434009 (lr=1.2659e-05) (hash(x)=165478625)
+6100 val loss 6.4656
+6100 val perplexity 642.6783
+6100 train 6.216571 (lr=1.2246e-05) (hash(x)=147088621)
+6200 val loss 6.4535
+6200 val perplexity 634.8896
+6200 train 6.240591 (lr=1.1838e-05) (hash(x)=140794994)
+6300 val loss 6.4400
+6300 val perplexity 626.4254
+6300 train 6.364601 (lr=1.1434e-05) (hash(x)=134780906)
+6400 val loss 6.4261
+6400 val perplexity 617.7345
+6400 train 6.413403 (lr=1.1035e-05) (hash(x)=149023655)
+6500 val loss 6.4153
+6500 val perplexity 611.1179
+6500 train 6.281328 (lr=1.0643e-05) (hash(x)=147497796)
+6600 val loss 6.4143
+6600 val perplexity 610.5001
+6600 train 6.261950 (lr=1.0256e-05) (hash(x)=152902689)
+6700 val loss 6.4073
+6700 val perplexity 606.2618
+6700 train 6.460744 (lr=9.8753e-06) (hash(x)=153846046)
+6800 val loss 6.3954
+6800 val perplexity 599.0741
+6800 train 6.533175 (lr=9.5017e-06) (hash(x)=158512738)
+6900 val loss 6.3901
+6900 val perplexity 595.9360
+6900 train 6.872580 (lr=9.1353e-06) (hash(x)=156849968)
+7000 val loss 6.3825
+7000 val perplexity 591.4265
+7000 train 6.211296 (lr=8.7764e-06) (hash(x)=142395855)
+7100 val loss 6.3768
+7100 val perplexity 588.0710
+7100 train 6.247732 (lr=8.4255e-06) (hash(x)=147114884)
+7200 val loss 6.3739
+7200 val perplexity 586.3677
+7200 train 6.394201 (lr=8.0829e-06) (hash(x)=156979839)
+7300 val loss 6.3695
+7300 val perplexity 583.7812
+7300 train 6.134778 (lr=7.7489e-06) (hash(x)=145584373)
+7400 val loss 6.3681
+7400 val perplexity 582.9282
+7400 train 6.108246 (lr=7.4239e-06) (hash(x)=141508204)
+7500 val loss 6.3592
+7500 val perplexity 577.7770
+7500 train 6.410224 (lr=7.1083e-06) (hash(x)=148803965)
+7600 val loss 6.3525
+7600 val perplexity 573.9427
+7600 train 6.412134 (lr=6.8023e-06) (hash(x)=151019676)
+7700 val loss 6.3469
+7700 val perplexity 570.7330
+7700 train 6.363933 (lr=6.5062e-06) (hash(x)=143155750)
+7800 val loss 6.3412
+7800 val perplexity 567.4922
+7800 train 6.365192 (lr=6.2205e-06) (hash(x)=152569653)
+7900 val loss 6.3391
+7900 val perplexity 566.2675
+7900 train 6.198283 (lr=5.9453e-06) (hash(x)=143519455)
+8000 val loss 6.3370
+8000 val perplexity 565.0965
+8000 train 6.521562 (lr=5.6809e-06) (hash(x)=161180944)
+8100 val loss 6.3356
+8100 val perplexity 564.3146
+8100 train 6.367569 (lr=5.4277e-06) (hash(x)=154107345)
+8200 val loss 6.3359
+8200 val perplexity 564.4596
+8200 train 6.350157 (lr=5.1858e-06) (hash(x)=152486517)
+8300 val loss 6.3262
+8300 val perplexity 559.0314
+8300 train 6.355000 (lr=4.9556e-06) (hash(x)=156167749)
+8400 val loss 6.3224
+8400 val perplexity 556.9216
+8400 train 6.399919 (lr=4.7372e-06) (hash(x)=149155006)
+8500 val loss 6.3170
+8500 val perplexity 553.9238
+8500 train 6.291015 (lr=4.5309e-06) (hash(x)=147844390)
+8600 val loss 6.3133
+8600 val perplexity 551.8853
+8600 train 6.573327 (lr=4.3369e-06) (hash(x)=165753320)
+8700 val loss 6.3135
+8700 val perplexity 551.9460
+8700 train 6.098110 (lr=4.1554e-06) (hash(x)=146079979)
+8800 val loss 6.3118
+8800 val perplexity 551.0446
+8800 train 6.534912 (lr=3.9866e-06) (hash(x)=172259509)
+8900 val loss 6.3120
+8900 val perplexity 551.1379
+8900 train 5.987774 (lr=3.8307e-06) (hash(x)=145148314)
+9000 val loss 6.3112
+9000 val perplexity 550.7257
+9000 train 6.234556 (lr=3.6877e-06) (hash(x)=144250633)
+9100 val loss 6.3057
+9100 val perplexity 547.6868
+9100 train 6.414195 (lr=3.5580e-06) (hash(x)=157219797)
+9200 val loss 6.2999
+9200 val perplexity 544.5214
+9200 train 6.288376 (lr=3.4415e-06) (hash(x)=142743778)
+9300 val loss 6.2994
+9300 val perplexity 544.2646
+9300 train 6.183533 (lr=3.3385e-06) (hash(x)=139669771)
+9400 val loss 6.2957
+9400 val perplexity 542.2340
+9400 train 6.265487 (lr=3.2490e-06) (hash(x)=145916843)
+9500 val loss 6.2935
+9500 val perplexity 541.0405
+9500 train 6.209871 (lr=3.1730e-06) (hash(x)=150196125)
+9600 val loss 6.2931
+9600 val perplexity 540.8432
+9600 train 6.577899 (lr=3.1108e-06) (hash(x)=160041419)
+9700 val loss 6.2934
+9700 val perplexity 540.9856
+9700 train 6.079773 (lr=3.0624e-06) (hash(x)=139931627)
+9800 val loss 6.2910
+9800 val perplexity 539.6734
+9800 train 6.253589 (lr=3.0277e-06) (hash(x)=150370792)
+9900 val loss 6.2893
+9900 val perplexity 538.8027
+9900 train 6.149877 (lr=3.0069e-06) (hash(x)=153014886)
+9999 val loss 6.2893
+9999 val perplexity 538.7606

attention_kindselective_n_heads2_seed1338/model_02500.pt

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:ec6bd7e87f52a5c93d8861916e3d82c1ba6393fe528c2119a53e9b72aa371446
+oid sha256:bfe50e723e825bd0786c1a4596ac4cdaf796676df7fdc5da3f87bcf867afc971
 size 38587970

attention_kindselective_n_heads2_seed1338/model_05000.pt

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:9cadf2af730975458af4444f28d26a5cc30c93a8ecab502f773cbe1fb1dc91e8
+oid sha256:e2e8ba950312264226829ac705df43b9cdffaf5bcc036d9b24aef5f3e832494c
 size 38587970

attention_kindselective_n_heads2_seed1338/model_07500.pt

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:da964b640be4bb6a0d647549527e5cb4389be18c0ad3ea8604a371383a3a15b8
+oid sha256:720ce17fda15dca0483b5bee479b09521f324cbda2f55ee89297aeb3815e7127
 size 38587970

attention_kindselective_n_heads2_seed1338/model_09999.pt

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:e0c5cb0d1631c98e783b7f9c6dc0f347e2490fc6240b77dc62fe3a2869067972
+oid sha256:0e1a2367f59732516e841fec0ec047482f7b40395d9a8c4e9bfd7434db667578
 size 38587970

attention_kindselective_n_heads2_seed1338/optimizer_02500.pt

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:7914d38ea51a0b603e2bd0d7b49b0845b68b726646e48c15d0132680add44d7c
+oid sha256:e8266da6cdaef57d288608149918e3d4cc507a0f73a368f4c248cc84d130fa87
 size 70895430

attention_kindselective_n_heads2_seed1338/optimizer_05000.pt

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:f4074661b3b029bf8d3f6f3d765e43fa8d2c27f0ce0850708b58af6f36cbf9dc
+oid sha256:b1cf4773a63f7bd4d9b9b266dfca8bb2631ec02374f0d2e94c402329e5b41680
 size 70895430

attention_kindselective_n_heads2_seed1338/optimizer_07500.pt

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:c222bd337a0e135a7bd57ddee8e40f76f8c7f26e8b95251a5f508a0910654717
+oid sha256:28ffac77284f9c9c07e805be4e544c3bf52c59e2e9c317ae438e92592ac76c33
 size 70895430

attention_kindselective_n_heads2_seed1338/optimizer_09999.pt

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:76c9b27422cd0a25887e6e5a39f1d7a35087a1e53dd1953799b22114616b0758
+oid sha256:d24ecb4199ebaa5047064c533fe7647126557157f3557829d39923f2d8fba6fd
 size 70895430