Upload folder using huggingface_hub

attention_kindselective_n_heads4_seed1340/args.json

@@ -1 +1 @@
-{"hellaswag": true, "attention_kind": "selective", "log_dir": "wider_is_better_9/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": 8750, "warmup_steps": 500, "group": "wider_is_better_9", "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": 61440, "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_61440_4_1340", "n_embd": 256}
+{"hellaswag": true, "attention_kind": "selective", "log_dir": "wider_is_better_9/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": 8750, "warmup_steps": 500, "group": "wider_is_better_9", "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": 61440, "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": 5.5e-05, "decay_lr": true, "readout_zero_init": false, "query_zero_init": false, "l1_loss": false, "debugpy": false, "key": "5.5e-5_61440_4_1340", "n_embd": 256}

attention_kindselective_n_heads4_seed1340/log2.txt

@@ -1,533 +1,267 @@
 max_steps: 8750
 0 val loss 11.2709
-0 val perplexity 78503.8281
-0 val loss 11.2709
-0 val perplexity 78503.8281
-0 train 11.277061 (lr=7.0000e-08) (hash(x)=164406924)
-0 train 11.277061 (lr=6.0000e-08) (hash(x)=164406924)
-100 val loss 10.1502
-100 val perplexity 25596.3652
-100 train 10.113564 (lr=7.0700e-06) (hash(x)=144903932)
-100 val loss 10.2327
-100 val perplexity 27796.1465
-100 train 10.198875 (lr=6.0600e-06) (hash(x)=144903932)
-200 val loss 9.7291
-200 val perplexity 16799.9590
-200 train 9.783854 (lr=1.4070e-05) (hash(x)=167734596)
-200 val loss 9.8080
-200 val perplexity 18177.9453
-200 train 9.861107 (lr=1.2060e-05) (hash(x)=167734596)
-300 val loss 8.8589
-300 val perplexity 7036.8594
-300 train 8.904337 (lr=2.1070e-05) (hash(x)=149619098)
-300 val loss 8.9841
-300 val perplexity 7975.3730
-300 train 9.031675 (lr=1.8060e-05) (hash(x)=149619098)
-400 val loss 7.9793
-400 val perplexity 2919.9771
-400 train 7.923836 (lr=2.8070e-05) (hash(x)=153710890)
-400 val loss 8.0370
-400 val perplexity 3093.2683
-400 train 7.987201 (lr=2.4060e-05) (hash(x)=153710890)
-500 val loss 7.6324
-500 val perplexity 2063.9663
-500 train 7.716890 (lr=3.5000e-05) (hash(x)=145450636)
-500 val loss 7.6279
-500 val perplexity 2054.7944
-500 train 7.708719 (lr=3.0000e-05) (hash(x)=145450636)
-600 val loss 7.4815
-600 val perplexity 1774.9750
-600 train 7.341413 (lr=3.4989e-05) (hash(x)=145249251)
-600 val loss 7.4696
-600 val perplexity 1753.8359
-600 train 7.332487 (lr=2.9990e-05) (hash(x)=145249251)
-700 val loss 7.3877
-700 val perplexity 1615.9683
-700 train 7.501635 (lr=3.4954e-05) (hash(x)=150475545)
-700 val loss 7.3795
-700 val perplexity 1602.7659
-700 train 7.492246 (lr=2.9961e-05) (hash(x)=150475545)
-800 val loss 7.3168
-800 val perplexity 1505.4194
-800 train 7.283746 (lr=3.4897e-05) (hash(x)=144483776)
-800 val loss 7.3040
-800 val perplexity 1486.2216
-800 train 7.270497 (lr=2.9912e-05) (hash(x)=144483776)
-900 val loss 7.2365
-900 val perplexity 1389.2202
-900 train 7.342116 (lr=3.4818e-05) (hash(x)=157916369)
-900 val loss 7.2214
-900 val perplexity 1368.4169
-900 train 7.322839 (lr=2.9844e-05) (hash(x)=157916369)
-1000 val loss 7.1684
-1000 val perplexity 1297.7793
-1000 train 7.195076 (lr=3.4715e-05) (hash(x)=154856891)
-1000 val loss 7.1551
-1000 val perplexity 1280.6429
-1000 train 7.181716 (lr=2.9756e-05) (hash(x)=154856891)
-1100 val loss 7.1025
-1100 val perplexity 1215.0176
-1100 train 6.832980 (lr=3.4591e-05) (hash(x)=136191502)
-1100 val loss 7.0845
-1100 val perplexity 1193.2688
-1100 train 6.817375 (lr=2.9649e-05) (hash(x)=136191502)
-1200 val loss 7.0249
-1200 val perplexity 1124.2922
-1200 train 6.977524 (lr=3.4444e-05) (hash(x)=148021541)
-1200 val loss 7.0108
-1200 val perplexity 1108.5863
-1200 train 6.962654 (lr=2.9523e-05) (hash(x)=148021541)
-1300 val loss 6.9649
-1300 val perplexity 1058.8282
-1300 train 6.899820 (lr=3.4275e-05) (hash(x)=146005217)
-1300 val loss 6.9486
-1300 val perplexity 1041.7388
-1300 train 6.883662 (lr=2.9378e-05) (hash(x)=146005217)
-1400 val loss 6.9124
-1400 val perplexity 1004.6208
-1400 train 6.817109 (lr=3.4084e-05) (hash(x)=146019502)
-1400 val loss 6.8887
-1400 val perplexity 981.1172
-1400 train 6.794360 (lr=2.9215e-05) (hash(x)=146019502)
-1500 val loss 6.8527
-1500 val perplexity 946.4700
-1500 train 6.795928 (lr=3.3872e-05) (hash(x)=150127281)
-1500 val loss 6.8294
-1500 val perplexity 924.6553
-1500 train 6.769631 (lr=2.9033e-05) (hash(x)=150127281)
-1600 val loss 6.7911
-1600 val perplexity 889.9340
-1600 train 6.759404 (lr=3.3638e-05) (hash(x)=154120875)
-1600 val loss 6.7634
-1600 val perplexity 865.6116
-1600 train 6.731661 (lr=2.8833e-05) (hash(x)=154120875)
-1700 val loss 6.7391
-1700 val perplexity 844.7706
-1700 train 6.959402 (lr=3.3384e-05) (hash(x)=155797680)
-1700 val loss 6.7154
-1700 val perplexity 825.0399
-1700 train 6.934373 (lr=2.8615e-05) (hash(x)=155797680)
-1800 val loss 6.6668
-1800 val perplexity 785.8712
-1800 train 6.665632 (lr=3.3109e-05) (hash(x)=156809396)
-1800 val loss 6.6459
-1800 val perplexity 769.6384
-1800 train 6.640855 (lr=2.8379e-05) (hash(x)=156809396)
-1900 val loss 6.6110
-1900 val perplexity 743.2146
-1900 train 6.502274 (lr=3.2814e-05) (hash(x)=144640294)
-1900 val loss 6.5938
-1900 val perplexity 730.5507
-1900 train 6.482723 (lr=2.8127e-05) (hash(x)=144640294)
-2000 val loss 6.5645
-2000 val perplexity 709.4272
-2000 train 6.437298 (lr=3.2500e-05) (hash(x)=162831106)
-2000 val loss 6.5512
-2000 val perplexity 700.1097
-2000 train 6.425098 (lr=2.7857e-05) (hash(x)=162831106)
-2100 val loss 6.5289
-2100 val perplexity 684.6605
-2100 train 6.518277 (lr=3.2166e-05) (hash(x)=158239484)
-2100 val loss 6.5258
-2100 val perplexity 682.5104
-2100 train 6.522050 (lr=2.7571e-05) (hash(x)=158239484)
-2200 val loss 6.4757
-2200 val perplexity 649.1948
-2200 train 6.351269 (lr=3.1813e-05) (hash(x)=140504180)
-2200 val loss 6.4764
-2200 val perplexity 649.6376
-2200 train 6.350248 (lr=2.7269e-05) (hash(x)=140504180)
-2300 val loss 6.4330
-2300 val perplexity 622.0456
-2300 train 6.415581 (lr=3.1443e-05) (hash(x)=142234024)
-2300 val loss 6.4349
-2300 val perplexity 623.2421
-2300 train 6.418139 (lr=2.6951e-05) (hash(x)=142234024)
-2400 val loss 6.4024
-2400 val perplexity 603.2999
-2400 train 6.397040 (lr=3.1054e-05) (hash(x)=143091562)
-2400 val loss 6.4081
-2400 val perplexity 606.7314
-2400 train 6.401471 (lr=2.6618e-05) (hash(x)=143091562)
-2500 val loss 6.3863
-2500 val perplexity 593.6433
-2500 train 6.290105 (lr=3.0649e-05) (hash(x)=149857456)
-2500 val loss 6.3969
-2500 val perplexity 599.9606
-2500 train 6.296983 (lr=2.6270e-05) (hash(x)=149857456)
-2600 val loss 6.3419
-2600 val perplexity 567.8901
-2600 train 6.379487 (lr=3.0227e-05) (hash(x)=146191551)
-2600 val loss 6.3522
-2600 val perplexity 573.7252
-2600 train 6.395041 (lr=2.5909e-05) (hash(x)=146191551)
-2700 val loss 6.3120
-2700 val perplexity 551.1568
-2700 train 6.235927 (lr=2.9789e-05) (hash(x)=145375752)
-2700 val loss 6.3210
-2700 val perplexity 556.1021
-2700 train 6.239472 (lr=2.5533e-05) (hash(x)=145375752)
-2800 val loss 6.2871
-2800 val perplexity 537.5940
-2800 train 6.166823 (lr=2.9336e-05) (hash(x)=151568014)
-2800 val loss 6.3012
-2800 val perplexity 545.2515
-2800 train 6.188046 (lr=2.5145e-05) (hash(x)=151568014)
-2900 val loss 6.2707
-2900 val perplexity 528.8716
-2900 train 6.104226 (lr=2.8868e-05) (hash(x)=149366597)
-2900 val loss 6.2892
-2900 val perplexity 538.7473
-2900 train 6.116791 (lr=2.4744e-05) (hash(x)=149366597)
-3000 val loss 6.2343
-3000 val perplexity 509.9225
-3000 train 6.170280 (lr=2.8386e-05) (hash(x)=150464442)
-3000 val loss 6.2545
-3000 val perplexity 520.3714
-3000 train 6.192822 (lr=2.4331e-05) (hash(x)=150464442)
-3100 val loss 6.2172
-3100 val perplexity 501.3207
-3100 train 6.346448 (lr=2.7891e-05) (hash(x)=182449036)
-3100 val loss 6.2397
-3100 val perplexity 512.7269
-3100 train 6.366556 (lr=2.3906e-05) (hash(x)=182449036)
-3200 val loss 6.1963
-3200 val perplexity 490.9159
-3200 train 6.149408 (lr=2.7383e-05) (hash(x)=140141286)
-3200 val loss 6.2185
-3200 val perplexity 501.9713
-3200 train 6.174580 (lr=2.3471e-05) (hash(x)=140141286)
-3300 val loss 6.1753
-3300 val perplexity 480.7094
-3300 train 6.136339 (lr=2.6864e-05) (hash(x)=148099414)
-3300 val loss 6.1991
-3300 val perplexity 492.3220
-3300 train 6.160706 (lr=2.3026e-05) (hash(x)=148099414)
-3400 val loss 6.1553
-3400 val perplexity 471.2260
-3400 train 6.015404 (lr=2.6333e-05) (hash(x)=142633951)
-3400 val loss 6.1802
-3400 val perplexity 483.0799
-3400 train 6.039740 (lr=2.2572e-05) (hash(x)=142633951)
-3500 val loss 6.1481
-3500 val perplexity 467.8076
-3500 train 6.004545 (lr=2.5793e-05) (hash(x)=148368965)
-3500 val loss 6.1706
-3500 val perplexity 478.4758
-3500 train 6.029538 (lr=2.2108e-05) (hash(x)=148368965)
-3600 val loss 6.1221
-3600 val perplexity 455.8235
-3600 train 6.150208 (lr=2.5243e-05) (hash(x)=152372067)
-3600 val loss 6.1490
-3600 val perplexity 468.2530
-3600 train 6.173963 (lr=2.1637e-05) (hash(x)=152372067)
-3700 val loss 6.1051
-3700 val perplexity 448.1516
-3700 train 6.197345 (lr=2.4684e-05) (hash(x)=168885609)
-3700 val loss 6.1346
-3700 val perplexity 461.5741
-3700 train 6.226795 (lr=2.1158e-05) (hash(x)=168885609)
-3800 val loss 6.0966
-3800 val perplexity 444.3637
-3800 train 5.907861 (lr=2.4117e-05) (hash(x)=141633734)
-3800 val loss 6.1256
-3800 val perplexity 457.4419
-3800 train 5.937433 (lr=2.0672e-05) (hash(x)=141633734)
-3900 val loss 6.0723
-3900 val perplexity 433.6902
-3900 train 5.966411 (lr=2.3543e-05) (hash(x)=153141007)
-3900 val loss 6.1031
-3900 val perplexity 447.2220
-3900 train 5.999508 (lr=2.0180e-05) (hash(x)=153141007)
-4000 val loss 6.0612
-4000 val perplexity 428.9083
-4000 train 5.994430 (lr=2.2963e-05) (hash(x)=160577202)
-4000 val loss 6.0919
-4000 val perplexity 442.2613
-4000 train 6.026505 (lr=1.9683e-05) (hash(x)=160577202)
-4100 val loss 6.0522
-4100 val perplexity 425.0566
-4100 train 6.113316 (lr=2.2378e-05) (hash(x)=153858169)
-4100 val loss 6.0831
-4100 val perplexity 438.4046
-4100 train 6.139560 (lr=1.9181e-05) (hash(x)=153858169)
-4200 val loss 6.0324
-4200 val perplexity 416.7092
-4200 train 6.035808 (lr=2.1788e-05) (hash(x)=155889149)
-4200 val loss 6.0647
-4200 val perplexity 430.4065
-4200 train 6.064448 (lr=1.8675e-05) (hash(x)=155889149)
-4300 val loss 6.0199
-4300 val perplexity 411.5509
-4300 train 5.834055 (lr=2.1194e-05) (hash(x)=152294662)
-4300 val loss 6.0560
-4300 val perplexity 426.6565
-4300 train 5.866488 (lr=1.8166e-05) (hash(x)=152294662)
-4400 val loss 6.0108
-4400 val perplexity 407.8034
-4400 train 5.899076 (lr=2.0598e-05) (hash(x)=141804386)
-4400 val loss 6.0450
-4400 val perplexity 421.9777
-4400 train 5.932065 (lr=1.7655e-05) (hash(x)=141804386)
-4500 val loss 5.9941
-4500 val perplexity 401.0631
-4500 train 5.893041 (lr=1.9999e-05) (hash(x)=151095242)
-4500 val loss 6.0274
-4500 val perplexity 414.6237
-4500 train 5.928043 (lr=1.7142e-05) (hash(x)=151095242)
-4600 val loss 5.9986
-4600 val perplexity 402.8594
-4600 train 5.935762 (lr=1.9400e-05) (hash(x)=156414699)
-4600 val loss 6.0307
-4600 val perplexity 416.0135
-4600 train 5.966719 (lr=1.6629e-05) (hash(x)=156414699)
-4700 val loss 5.9728
-4700 val perplexity 392.6121
-4700 train 5.997845 (lr=1.8800e-05) (hash(x)=161556686)
-4700 val loss 6.0094
-4700 val perplexity 407.2528
-4700 train 6.033873 (lr=1.6114e-05) (hash(x)=161556686)
-4800 val loss 5.9603
-4800 val perplexity 387.7329
-4800 train 5.870970 (lr=1.8201e-05) (hash(x)=149000293)
-4800 val loss 5.9985
-4800 val perplexity 402.8394
-4800 train 5.906920 (lr=1.5601e-05) (hash(x)=149000293)
-4900 val loss 5.9508
-4900 val perplexity 384.0769
-4900 train 5.752374 (lr=1.7604e-05) (hash(x)=154349989)
-4900 val loss 5.9899
-4900 val perplexity 399.3921
-4900 train 5.792023 (lr=1.5089e-05) (hash(x)=154349989)
-5000 val loss 5.9401
-5000 val perplexity 379.9893
-5000 train 5.769830 (lr=1.7009e-05) (hash(x)=131475967)
-5000 val loss 5.9785
-5000 val perplexity 394.8312
-5000 train 5.818519 (lr=1.4579e-05) (hash(x)=131475967)
-5100 val loss 5.9320
-5100 val perplexity 376.9128
-5100 train 5.830674 (lr=1.6417e-05) (hash(x)=149717902)
-5100 val loss 5.9694
-5100 val perplexity 391.2876
-5100 train 5.866533 (lr=1.4071e-05) (hash(x)=149717902)
-5200 val loss 5.9241
-5200 val perplexity 373.9456
-5200 train 5.780983 (lr=1.5829e-05) (hash(x)=151407999)
-5200 val loss 5.9621
-5200 val perplexity 388.4129
-5200 train 5.815117 (lr=1.3568e-05) (hash(x)=151407999)
-5300 val loss 5.9147
-5300 val perplexity 370.4412
-5300 train 6.021460 (lr=1.5246e-05) (hash(x)=168602728)
-5300 val loss 5.9521
-5300 val perplexity 384.5754
-5300 train 6.062849 (lr=1.3068e-05) (hash(x)=168602728)
-5400 val loss 5.9028
-5400 val perplexity 366.0675
-5400 train 5.860891 (lr=1.4669e-05) (hash(x)=158344511)
-5400 val loss 5.9456
-5400 val perplexity 382.0638
-5400 train 5.902995 (lr=1.2573e-05) (hash(x)=158344511)
-5500 val loss 5.8960
-5500 val perplexity 363.5726
-5500 train 5.857149 (lr=1.4099e-05) (hash(x)=148350057)
-5500 val loss 5.9365
-5500 val perplexity 378.6254
-5500 train 5.899691 (lr=1.2085e-05) (hash(x)=148350057)
-5600 val loss 5.8877
-5600 val perplexity 360.5695
-5600 train 5.956222 (lr=1.3536e-05) (hash(x)=153847323)
-5600 val loss 5.9293
-5600 val perplexity 375.8728
-5600 train 5.996018 (lr=1.1602e-05) (hash(x)=153847323)
-5700 val loss 5.8798
-5700 val perplexity 357.7410
-5700 train 5.829589 (lr=1.2981e-05) (hash(x)=156607405)
-5700 val loss 5.9220
-5700 val perplexity 373.1628
-5700 train 5.870955 (lr=1.1127e-05) (hash(x)=156607405)
-5800 val loss 5.8736
-5800 val perplexity 355.5310
-5800 train 5.641397 (lr=1.2436e-05) (hash(x)=145115031)
-5800 val loss 5.9151
-5800 val perplexity 370.6071
-5800 train 5.676248 (lr=1.0659e-05) (hash(x)=145115031)
-5900 val loss 5.8659
-5900 val perplexity 352.7947
-5900 train 5.737990 (lr=1.1900e-05) (hash(x)=141584622)
-5900 val loss 5.9075
-5900 val perplexity 367.7683
-5900 train 5.776842 (lr=1.0200e-05) (hash(x)=141584622)
-6000 val loss 5.8580
-6000 val perplexity 350.0075
-6000 train 5.787103 (lr=1.1375e-05) (hash(x)=146613857)
-6000 val loss 5.9002
-6000 val perplexity 365.0927
-6000 train 5.830695 (lr=9.7500e-06) (hash(x)=146613857)
-6100 val loss 5.8535
-6100 val perplexity 348.4407
-6100 train 5.786377 (lr=1.0861e-05) (hash(x)=144621768)
-6100 val loss 5.8963
-6100 val perplexity 363.7028
-6100 train 5.814258 (lr=9.3098e-06) (hash(x)=144621768)
-6200 val loss 5.8504
-6200 val perplexity 347.3798
-6200 train 5.775459 (lr=1.0360e-05) (hash(x)=146521760)
-6200 val loss 5.8932
-6200 val perplexity 362.5698
-6200 train 5.820045 (lr=8.8800e-06) (hash(x)=146521760)
-6300 val loss 5.8390
-6300 val perplexity 343.4239
-6300 train 5.671170 (lr=9.8715e-06) (hash(x)=161378136)
-6300 val loss 5.8837
-6300 val perplexity 359.1181
-6300 train 5.718021 (lr=8.4613e-06) (hash(x)=161378136)
-6400 val loss 5.8362
-6400 val perplexity 342.4716
-6400 train 5.602135 (lr=9.3966e-06) (hash(x)=141624235)
-6400 val loss 5.8797
-6400 val perplexity 357.7079
-6400 train 5.647285 (lr=8.0542e-06) (hash(x)=141624235)
-6500 val loss 5.8298
-6500 val perplexity 340.3030
-6500 train 5.743033 (lr=8.9359e-06) (hash(x)=151197095)
-6500 val loss 5.8733
-6500 val perplexity 355.4235
-6500 train 5.777500 (lr=7.6594e-06) (hash(x)=151197095)
-6600 val loss 5.8275
-6600 val perplexity 339.4921
-6600 train 5.840244 (lr=8.4903e-06) (hash(x)=153269571)
-6600 val loss 5.8719
-6600 val perplexity 354.9266
-6600 train 5.881926 (lr=7.2774e-06) (hash(x)=153269571)
-6700 val loss 5.8217
-6700 val perplexity 337.5518
-6700 train 5.783973 (lr=8.0602e-06) (hash(x)=146111181)
-6700 val loss 5.8653
-6700 val perplexity 352.5977
-6700 train 5.830712 (lr=6.9087e-06) (hash(x)=146111181)
-6800 val loss 5.8137
-6800 val perplexity 334.8429
-6800 train 5.840888 (lr=7.6463e-06) (hash(x)=147269760)
-6800 val loss 5.8585
-6800 val perplexity 350.1996
-6800 train 5.885352 (lr=6.5540e-06) (hash(x)=147269760)
-6900 val loss 5.8129
-6900 val perplexity 334.5923
-6900 train 5.878181 (lr=7.2493e-06) (hash(x)=152912762)
-6900 val loss 5.8558
-6900 val perplexity 349.2571
-6900 train 5.918777 (lr=6.2137e-06) (hash(x)=152912762)
-7000 val loss 5.8089
-7000 val perplexity 333.2421
-7000 train 5.886887 (lr=6.8697e-06) (hash(x)=165412343)
-7000 val loss 5.8537
-7000 val perplexity 348.5204
-7000 train 5.931127 (lr=5.8883e-06) (hash(x)=165412343)
-7100 val loss 5.8002
-7100 val perplexity 330.3812
-7100 train 5.863431 (lr=6.5080e-06) (hash(x)=162866028)
-7100 val loss 5.8450
-7100 val perplexity 345.4860
-7100 train 5.909845 (lr=5.5783e-06) (hash(x)=162866028)
-7200 val loss 5.7969
-7200 val perplexity 329.2704
-7200 train 5.771354 (lr=6.1648e-06) (hash(x)=142998115)
-7200 val loss 5.8425
-7200 val perplexity 344.6447
-7200 train 5.816433 (lr=5.2841e-06) (hash(x)=142998115)
-7300 val loss 5.7948
-7300 val perplexity 328.5770
-7300 train 5.564707 (lr=5.8405e-06) (hash(x)=145486999)
-7300 val loss 5.8414
-7300 val perplexity 344.2689
-7300 train 5.614068 (lr=5.0062e-06) (hash(x)=145486999)
-7400 val loss 5.7891
-7400 val perplexity 326.7220
-7400 train 5.758606 (lr=5.5357e-06) (hash(x)=155325873)
-7400 val loss 5.8346
-7400 val perplexity 341.9388
-7400 train 5.804873 (lr=4.7449e-06) (hash(x)=155325873)
-7500 val loss 5.7857
-7500 val perplexity 325.5937
-7500 train 5.746189 (lr=5.2508e-06) (hash(x)=145131256)
-7500 val loss 5.8316
-7500 val perplexity 340.9057
-7500 train 5.791263 (lr=4.5007e-06) (hash(x)=145131256)
-7600 val loss 5.7893
-7600 val perplexity 326.7786
-7600 train 5.710508 (lr=4.9862e-06) (hash(x)=144008365)
-7600 val loss 5.8346
-7600 val perplexity 341.9225
-7600 train 5.753230 (lr=4.2739e-06) (hash(x)=144008365)
-7700 val loss 5.7797
-7700 val perplexity 323.6634
-7700 train 5.858647 (lr=4.7423e-06) (hash(x)=148848532)
-7700 val loss 5.8260
-7700 val perplexity 338.9968
-7700 train 5.898289 (lr=4.0648e-06) (hash(x)=148848532)
-7800 val loss 5.7777
-7800 val perplexity 323.0294
-7800 train 5.758301 (lr=4.5194e-06) (hash(x)=150391642)
-7800 val loss 5.8235
-7800 val perplexity 338.1660
-7800 train 5.806548 (lr=3.8738e-06) (hash(x)=150391642)
-7900 val loss 5.7780
-7900 val perplexity 323.1080
-7900 train 5.534154 (lr=4.3179e-06) (hash(x)=152191414)
-7900 val loss 5.8236
-7900 val perplexity 338.1740
-7900 train 5.581667 (lr=3.7010e-06) (hash(x)=152191414)
-8000 val loss 5.7726
-8000 val perplexity 321.3700
-8000 train 5.851005 (lr=4.1380e-06) (hash(x)=159755587)
-8000 val loss 5.8195
-8000 val perplexity 336.8142
-8000 train 5.901532 (lr=3.5468e-06) (hash(x)=159755587)
-8100 val loss 5.7703
-8100 val perplexity 320.6418
-8100 train 5.791206 (lr=3.9800e-06) (hash(x)=156664468)
-8100 val loss 5.8174
-8100 val perplexity 336.0982
-8100 train 5.838668 (lr=3.4114e-06) (hash(x)=156664468)
-8200 val loss 5.7692
-8200 val perplexity 320.2684
-8200 val loss 5.8163
-8200 val perplexity 335.7272
-8200 train 5.611233 (lr=3.8442e-06) (hash(x)=139457379)
-8200 train 5.657933 (lr=3.2950e-06) (hash(x)=139457379)
-8300 val loss 5.7722
-8300 val perplexity 321.2451
-8300 val loss 5.8194
-8300 val perplexity 336.7803
-8300 train 5.541981 (lr=3.7307e-06) (hash(x)=145478564)
-8300 train 5.593337 (lr=3.1977e-06) (hash(x)=145478564)
-8400 val loss 5.7649
-8400 val perplexity 318.9180
-8400 val loss 5.8112
-8400 val perplexity 334.0220
-8400 train 5.704823 (lr=3.6397e-06) (hash(x)=154982769)
-8400 train 5.757417 (lr=3.1197e-06) (hash(x)=154982769)
-8500 val loss 5.7627
-8500 val perplexity 318.2200
-8500 val loss 5.8100
-8500 val perplexity 333.6224
-8500 train 5.712815 (lr=3.5713e-06) (hash(x)=145798118)
-8500 train 5.755826 (lr=3.0611e-06) (hash(x)=145798118)
-8600 val loss 5.7622
-8600 val perplexity 318.0369
-8600 val loss 5.8088
-8600 val perplexity 333.2224
-8600 train 5.668952 (lr=3.5257e-06) (hash(x)=143231551)
-8600 train 5.717763 (lr=3.0220e-06) (hash(x)=143231551)
-8700 val loss 5.7658
-8700 val perplexity 319.1926
-8700 val loss 5.8134
-8700 val perplexity 334.7692
-8700 train 5.667092 (lr=3.5029e-06) (hash(x)=154780112)
-8700 train 5.715662 (lr=3.0024e-06) (hash(x)=154780112)
-8749 val loss 5.7583
-8749 val perplexity 316.8212
-8749 val loss 5.8060
-8749 val perplexity 332.2806
+0 val perplexity 78502.8516
+0 train 11.277068 (lr=1.1000e-07) (hash(x)=164406924)
+100 val loss 10.0085
+100 val perplexity 22213.9395
+100 train 9.970803 (lr=1.1110e-05) (hash(x)=144903932)
+200 val loss 9.2914
+200 val perplexity 10844.2412
+200 train 9.357403 (lr=2.2110e-05) (hash(x)=167734596)
+300 val loss 8.0630
+300 val perplexity 3174.7661
+300 train 8.075590 (lr=3.3110e-05) (hash(x)=149619098)
+400 val loss 7.5692
+400 val perplexity 1937.4963
+400 train 7.450818 (lr=4.4110e-05) (hash(x)=153710890)
+500 val loss 7.4143
+500 val perplexity 1659.5359
+500 train 7.504872 (lr=5.5000e-05) (hash(x)=145450636)
+600 val loss 7.2988
+600 val perplexity 1478.4788
+600 train 7.166701 (lr=5.4982e-05) (hash(x)=145249251)
+700 val loss 7.1882
+700 val perplexity 1323.7411
+700 train 7.318045 (lr=5.4928e-05) (hash(x)=150475545)
+800 val loss 7.0767
+800 val perplexity 1184.1125
+800 train 7.045276 (lr=5.4839e-05) (hash(x)=144483776)
+900 val loss 6.9781
+900 val perplexity 1072.9233
+900 train 7.080307 (lr=5.4713e-05) (hash(x)=157916369)
+1000 val loss 6.9082
+1000 val perplexity 1000.4556
+1000 train 6.925345 (lr=5.4553e-05) (hash(x)=154856891)
+1100 val loss 6.8256
+1100 val perplexity 921.1058
+1100 train 6.522852 (lr=5.4357e-05) (hash(x)=136191502)
+1200 val loss 6.7440
+1200 val perplexity 848.9789
+1200 train 6.700300 (lr=5.4126e-05) (hash(x)=148021541)
+1300 val loss 6.6680
+1300 val perplexity 786.7943
+1300 train 6.599460 (lr=5.3860e-05) (hash(x)=146005217)
+1400 val loss 6.6030
+1400 val perplexity 737.3328
+1400 train 6.514089 (lr=5.3561e-05) (hash(x)=146019502)
+1500 val loss 6.5557
+1500 val perplexity 703.2271
+1500 train 6.492791 (lr=5.3227e-05) (hash(x)=150127281)
+1600 val loss 6.5065
+1600 val perplexity 669.4506
+1600 train 6.472455 (lr=5.2860e-05) (hash(x)=154120875)
+1700 val loss 6.4733
+1700 val perplexity 647.6284
+1700 train 6.692543 (lr=5.2461e-05) (hash(x)=155797680)
+1800 val loss 6.4172
+1800 val perplexity 612.2867
+1800 train 6.403757 (lr=5.2029e-05) (hash(x)=156809396)
+1900 val loss 6.3877
+1900 val perplexity 594.5070
+1900 train 6.274245 (lr=5.1565e-05) (hash(x)=144640294)
+2000 val loss 6.3473
+2000 val perplexity 570.9570
+2000 train 6.208043 (lr=5.1071e-05) (hash(x)=162831106)
+2100 val loss 6.3462
+2100 val perplexity 570.3412
+2100 train 6.336015 (lr=5.0547e-05) (hash(x)=158239484)
+2200 val loss 6.2871
+2200 val perplexity 537.5853
+2200 train 6.161128 (lr=4.9993e-05) (hash(x)=140504180)
+2300 val loss 6.2516
+2300 val perplexity 518.8192
+2300 train 6.238686 (lr=4.9410e-05) (hash(x)=142234024)
+2400 val loss 6.2270
+2400 val perplexity 506.2221
+2400 train 6.225477 (lr=4.8800e-05) (hash(x)=143091562)
+2500 val loss 6.2183
+2500 val perplexity 501.8651
+2500 train 6.118163 (lr=4.8162e-05) (hash(x)=149857456)
+2600 val loss 6.1767
+2600 val perplexity 481.3983
+2600 train 6.211945 (lr=4.7499e-05) (hash(x)=146191551)
+2700 val loss 6.1543
+2700 val perplexity 470.7182
+2700 train 6.071248 (lr=4.6811e-05) (hash(x)=145375752)
+2800 val loss 6.1304
+2800 val perplexity 459.5979
+2800 train 6.014522 (lr=4.6099e-05) (hash(x)=151568014)
+2900 val loss 6.1218
+2900 val perplexity 455.7011
+2900 train 5.945108 (lr=4.5364e-05) (hash(x)=149366597)
+3000 val loss 6.0883
+3000 val perplexity 440.6820
+3000 train 6.013472 (lr=4.4606e-05) (hash(x)=150464442)
+3100 val loss 6.0724
+3100 val perplexity 433.7043
+3100 train 6.195862 (lr=4.3828e-05) (hash(x)=182449036)
+3200 val loss 6.0531
+3200 val perplexity 425.4502
+3200 train 6.008684 (lr=4.3031e-05) (hash(x)=140141286)
+3300 val loss 6.0351
+3300 val perplexity 417.8260
+3300 train 6.008709 (lr=4.2215e-05) (hash(x)=148099414)
+3400 val loss 6.0145
+3400 val perplexity 409.3233
+3400 train 5.879248 (lr=4.1381e-05) (hash(x)=142633951)
+3500 val loss 6.0038
+3500 val perplexity 404.9803
+3500 train 5.866429 (lr=4.0532e-05) (hash(x)=148368965)
+3600 val loss 5.9817
+3600 val perplexity 396.1001
+3600 train 6.005436 (lr=3.9667e-05) (hash(x)=152372067)
+3700 val loss 5.9637
+3700 val perplexity 389.0613
+3700 train 6.042872 (lr=3.8789e-05) (hash(x)=168885609)
+3800 val loss 5.9535
+3800 val perplexity 385.1143
+3800 train 5.766492 (lr=3.7898e-05) (hash(x)=141633734)
+3900 val loss 5.9283
+3900 val perplexity 375.5027
+3900 train 5.815371 (lr=3.6996e-05) (hash(x)=153141007)
+4000 val loss 5.9161
+4000 val perplexity 370.9533
+4000 train 5.847968 (lr=3.6085e-05) (hash(x)=160577202)
+4100 val loss 5.9108
+4100 val perplexity 368.9956
+4100 train 5.976177 (lr=3.5165e-05) (hash(x)=153858169)
+4200 val loss 5.8859
+4200 val perplexity 359.9236
+4200 train 5.900618 (lr=3.4238e-05) (hash(x)=155889149)
+4300 val loss 5.8756
+4300 val perplexity 356.2298
+4300 train 5.686732 (lr=3.3305e-05) (hash(x)=152294662)
+4400 val loss 5.8614
+4400 val perplexity 351.2089
+4400 train 5.759041 (lr=3.2368e-05) (hash(x)=141804386)
+4500 val loss 5.8419
+4500 val perplexity 344.4301
+4500 train 5.738522 (lr=3.1428e-05) (hash(x)=151095242)
+4600 val loss 5.8479
+4600 val perplexity 346.4965
+4600 train 5.792981 (lr=3.0486e-05) (hash(x)=156414699)
+4700 val loss 5.8213
+4700 val perplexity 337.4196
+4700 train 5.835704 (lr=2.9543e-05) (hash(x)=161556686)
+4800 val loss 5.8061
+4800 val perplexity 332.3093
+4800 train 5.715901 (lr=2.8602e-05) (hash(x)=149000293)
+4900 val loss 5.7968
+4900 val perplexity 329.2426
+4900 train 5.584097 (lr=2.7663e-05) (hash(x)=154349989)
+5000 val loss 5.7843
+5000 val perplexity 325.1438
+5000 train 5.614993 (lr=2.6728e-05) (hash(x)=131475967)
+5100 val loss 5.7728
+5100 val perplexity 321.4481
+5100 train 5.674039 (lr=2.5798e-05) (hash(x)=149717902)
+5200 val loss 5.7695
+5200 val perplexity 320.3890
+5200 train 5.634620 (lr=2.4874e-05) (hash(x)=151407999)
+5300 val loss 5.7545
+5300 val perplexity 315.6197
+5300 train 5.843950 (lr=2.3958e-05) (hash(x)=168602728)
+5400 val loss 5.7444
+5400 val perplexity 312.4250
+5400 train 5.700167 (lr=2.3051e-05) (hash(x)=158344511)
+5500 val loss 5.7369
+5500 val perplexity 310.0940
+5500 train 5.696602 (lr=2.2155e-05) (hash(x)=148350057)
+5600 val loss 5.7255
+5600 val perplexity 306.5844
+5600 train 5.792315 (lr=2.1271e-05) (hash(x)=153847323)
+5700 val loss 5.7168
+5700 val perplexity 303.9281
+5700 train 5.660481 (lr=2.0399e-05) (hash(x)=156607405)
+5800 val loss 5.7113
+5800 val perplexity 302.2715
+5800 train 5.476274 (lr=1.9542e-05) (hash(x)=145115031)
+5900 val loss 5.7032
+5900 val perplexity 299.8224
+5900 train 5.585124 (lr=1.8700e-05) (hash(x)=141584622)
+6000 val loss 5.6926
+6000 val perplexity 296.6544
+6000 train 5.623785 (lr=1.7875e-05) (hash(x)=146613857)
+6100 val loss 5.6872
+6100 val perplexity 295.0699
+6100 train 5.629707 (lr=1.7068e-05) (hash(x)=144621768)
+6200 val loss 5.6809
+6200 val perplexity 293.2248
+6200 train 5.612484 (lr=1.6280e-05) (hash(x)=146521760)
+6300 val loss 5.6744
+6300 val perplexity 291.3021
+6300 train 5.505092 (lr=1.5512e-05) (hash(x)=161378136)
+6400 val loss 5.6678
+6400 val perplexity 289.3883
+6400 train 5.434224 (lr=1.4766e-05) (hash(x)=141624235)
+6500 val loss 5.6614
+6500 val perplexity 287.5368
+6500 train 5.581383 (lr=1.4042e-05) (hash(x)=151197095)
+6600 val loss 5.6591
+6600 val perplexity 286.9044
+6600 train 5.677541 (lr=1.3342e-05) (hash(x)=153269571)
+6700 val loss 5.6526
+6700 val perplexity 285.0374
+6700 train 5.618601 (lr=1.2666e-05) (hash(x)=146111181)
+6800 val loss 5.6434
+6800 val perplexity 282.4169
+6800 train 5.672344 (lr=1.2016e-05) (hash(x)=147269760)
+6900 val loss 5.6390
+6900 val perplexity 281.1924
+6900 train 5.721029 (lr=1.1392e-05) (hash(x)=152912762)
+7000 val loss 5.6363
+7000 val perplexity 280.4270
+7000 train 5.724804 (lr=1.0795e-05) (hash(x)=165412343)
+7100 val loss 5.6272
+7100 val perplexity 277.8830
+7100 train 5.702348 (lr=1.0227e-05) (hash(x)=162866028)
+7200 val loss 5.6242
+7200 val perplexity 277.0432
+7200 train 5.594518 (lr=9.6875e-06) (hash(x)=142998115)
+7300 val loss 5.6216
+7300 val perplexity 276.3317
+7300 train 5.392981 (lr=9.1780e-06) (hash(x)=145486999)
+7400 val loss 5.6144
+7400 val perplexity 274.3558
+7400 train 5.591548 (lr=8.6990e-06) (hash(x)=155325873)
+7500 val loss 5.6108
+7500 val perplexity 273.3524
+7500 train 5.574790 (lr=8.2513e-06) (hash(x)=145131256)
+7600 val loss 5.6154
+7600 val perplexity 274.6144
+7600 train 5.539998 (lr=7.8355e-06) (hash(x)=144008365)
+7700 val loss 5.6057
+7700 val perplexity 271.9604
+7700 train 5.708950 (lr=7.4522e-06) (hash(x)=148848532)
+7800 val loss 5.6015
+7800 val perplexity 270.8214
+7800 train 5.587547 (lr=7.1019e-06) (hash(x)=150391642)
+7900 val loss 5.6012
+7900 val perplexity 270.7429
+7900 train 5.359458 (lr=6.7852e-06) (hash(x)=152191414)
+8000 val loss 5.5965
+8000 val perplexity 269.4868
+8000 train 5.674829 (lr=6.5025e-06) (hash(x)=159755587)
+8100 val loss 5.5929
+8100 val perplexity 268.5056
+8100 train 5.618392 (lr=6.2543e-06) (hash(x)=156664468)
+8200 val loss 5.5916
+8200 val perplexity 268.1588
+8200 train 5.440843 (lr=6.0408e-06) (hash(x)=139457379)
+8300 val loss 5.5967
+8300 val perplexity 269.5344
+8300 train 5.369544 (lr=5.8625e-06) (hash(x)=145478564)
+8400 val loss 5.5865
+8400 val perplexity 266.8061
+8400 train 5.525924 (lr=5.7195e-06) (hash(x)=154982769)
+8500 val loss 5.5846
+8500 val perplexity 266.2822
+8500 train 5.552999 (lr=5.6121e-06) (hash(x)=145798118)
+8600 val loss 5.5844
+8600 val perplexity 266.2398
+8600 train 5.494684 (lr=5.5404e-06) (hash(x)=143231551)
+8700 val loss 5.5872
+8700 val perplexity 266.9780
+8700 train 5.486710 (lr=5.5045e-06) (hash(x)=154780112)
+8749 val loss 5.5793
+8749 val perplexity 264.8789

attention_kindselective_n_heads4_seed1340/model_08749.pt

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:e0deb68cd19fbb766e0c31a28d0c7f0f91e0468f8c0970be55918f414ab5a2e1
+oid sha256:a2b41aec4b5aa02ea394f501290423bcd9f8ce3f46e72eca126939a6364eda72
 size 92843394

attention_kindselective_n_heads4_seed1340/optimizer_08749.pt

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:b680f6fd82e56ab5225beac288979a0f6e12d7c7d4ec8b7753e22389e7775a9d
+oid sha256:77641ea9c6c521b32f7a6a6ed7c39d1f0621a3ef114f834ef2d5b923b8687f0b
 size 179406214