Upload folder using huggingface_hub

attention_kindselective_n_heads4_seed1342/args.json

@@ -1 +1 @@
-{"hellaswag": true, "attention_kind": "selective", "log_dir": "wider_is_better_9/attention_kindselective_n_heads4_seed1342", "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": 1342, "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": 3.5e-05, "decay_lr": true, "readout_zero_init": false, "query_zero_init": false, "l1_loss": false, "debugpy": false, "key": "3.5e-5_61440_4_1342", "n_embd": 256}
+{"hellaswag": true, "attention_kind": "selective", "log_dir": "wider_is_better_9/attention_kindselective_n_heads4_seed1342", "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": 1342, "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": 4.5e-05, "decay_lr": true, "readout_zero_init": false, "query_zero_init": false, "l1_loss": false, "debugpy": false, "key": "4.5e-5_61440_4_1342", "n_embd": 256}

attention_kindselective_n_heads4_seed1342/log2.txt

@@ -1,534 +1,267 @@
 max_steps: 8750
-max_steps: 8750
-0 val loss 11.2898
-0 val perplexity 80004.5469
 0 val loss 11.2898
 0 val perplexity 80004.5469
-0 train 11.289051 (lr=7.0000e-08) (hash(x)=150332693)
-0 train 11.289051 (lr=6.0000e-08) (hash(x)=150332693)
-100 val loss 10.1794
-100 val perplexity 26354.9375
-100 val loss 10.1043
-100 val perplexity 24447.8750
-100 train 10.142110 (lr=6.0600e-06) (hash(x)=149277319)
-100 train 10.061918 (lr=7.0700e-06) (hash(x)=149277319)
-200 val loss 9.6815
-200 val perplexity 16018.4980
-200 train 9.752102 (lr=1.2060e-05) (hash(x)=155934075)
-200 val loss 9.6158
-200 val perplexity 15000.1738
-200 train 9.686650 (lr=1.4070e-05) (hash(x)=155934075)
-300 val loss 8.7807
-300 val perplexity 6507.2988
-300 train 8.821930 (lr=1.8060e-05) (hash(x)=165402628)
-300 val loss 8.6249
-300 val perplexity 5568.3394
-300 train 8.665680 (lr=2.1070e-05) (hash(x)=165402628)
-400 val loss 7.9230
-400 val perplexity 2759.9607
-400 train 7.778969 (lr=2.4060e-05) (hash(x)=139168939)
-400 val loss 7.8122
-400 val perplexity 2470.4434
-400 train 7.653944 (lr=2.8070e-05) (hash(x)=139168939)
-500 val loss 7.5502
-500 val perplexity 1901.0621
-500 train 7.481020 (lr=3.0000e-05) (hash(x)=147134338)
-500 val loss 7.5233
-500 val perplexity 1850.6179
-500 train 7.451856 (lr=3.5000e-05) (hash(x)=147134338)
-600 val loss 7.4285
-600 val perplexity 1683.2417
-600 train 7.348928 (lr=2.9990e-05) (hash(x)=159086459)
-600 val loss 7.4080
-600 val perplexity 1649.0697
-600 train 7.328346 (lr=3.4989e-05) (hash(x)=159086459)
-700 val loss 7.3568
-700 val perplexity 1566.8014
-700 train 7.395999 (lr=2.9961e-05) (hash(x)=152399099)
-700 val loss 7.3374
-700 val perplexity 1536.7510
-700 train 7.378353 (lr=3.4954e-05) (hash(x)=152399099)
-800 val loss 7.2636
-800 val perplexity 1427.3462
-800 train 7.262696 (lr=2.9912e-05) (hash(x)=156548843)
-800 val loss 7.2412
-800 val perplexity 1395.8164
-800 train 7.243083 (lr=3.4897e-05) (hash(x)=156548843)
-900 val loss 7.1888
-900 val perplexity 1324.5342
-900 train 7.158834 (lr=2.9844e-05) (hash(x)=157889911)
-900 val loss 7.1467
-900 val perplexity 1269.9143
-900 train 7.119726 (lr=3.4818e-05) (hash(x)=157889911)
-1000 val loss 7.1303
-1000 val perplexity 1249.2928
-1000 train 6.941356 (lr=2.9756e-05) (hash(x)=142943707)
-1000 val loss 7.0711
-1000 val perplexity 1177.5012
-1000 train 6.876932 (lr=3.4715e-05) (hash(x)=142943707)
-1100 val loss 7.0742
-1100 val perplexity 1181.1373
-1100 train 7.080656 (lr=2.9649e-05) (hash(x)=148938689)
-1100 val loss 7.0131
-1100 val perplexity 1111.0625
-1100 train 7.017052 (lr=3.4591e-05) (hash(x)=148938689)
-1200 val loss 7.0067
-1200 val perplexity 1104.0490
-1200 train 7.136171 (lr=2.9523e-05) (hash(x)=161962903)
-1200 val loss 6.9357
-1200 val perplexity 1028.3834
-1200 train 7.066735 (lr=3.4444e-05) (hash(x)=161962903)
-1300 val loss 6.9482
-1300 val perplexity 1041.2402
-1300 train 6.917204 (lr=2.9378e-05) (hash(x)=145979235)
-1300 val loss 6.8651
-1300 val perplexity 958.2105
-1300 train 6.840671 (lr=3.4275e-05) (hash(x)=145979235)
-1400 val loss 6.9017
-1400 val perplexity 993.9407
-1400 train 6.849684 (lr=2.9215e-05) (hash(x)=159157507)
-1400 val loss 6.8026
-1400 val perplexity 900.1846
-1400 train 6.751725 (lr=3.4084e-05) (hash(x)=159157507)
-1500 val loss 6.8547
-1500 val perplexity 948.3015
-1500 train 6.894870 (lr=2.9033e-05) (hash(x)=145943130)
-1500 val loss 6.7512
-1500 val perplexity 855.1268
-1500 train 6.790061 (lr=3.3872e-05) (hash(x)=145943130)
-1600 val loss 6.7962
-1600 val perplexity 894.4285
-1600 train 6.587233 (lr=2.8833e-05) (hash(x)=143249173)
-1600 val loss 6.6718
-1600 val perplexity 789.8278
-1600 train 6.468535 (lr=3.3638e-05) (hash(x)=143249173)
-1700 val loss 6.7377
-1700 val perplexity 843.5904
-1700 train 6.812438 (lr=2.8615e-05) (hash(x)=173907143)
-1700 val loss 6.6123
-1700 val perplexity 744.1997
-1700 train 6.683813 (lr=3.3384e-05) (hash(x)=173907143)
-1800 val loss 6.6922
-1800 val perplexity 806.0727
-1800 train 6.619445 (lr=2.8379e-05) (hash(x)=138363755)
-1800 val loss 6.5679
-1800 val perplexity 711.8691
-1800 train 6.490389 (lr=3.3109e-05) (hash(x)=138363755)
-1900 val loss 6.6449
-1900 val perplexity 768.8578
-1900 train 6.460667 (lr=2.8127e-05) (hash(x)=146341390)
-1900 val loss 6.5231
-1900 val perplexity 680.7050
-1900 train 6.339457 (lr=3.2814e-05) (hash(x)=146341390)
-2000 val loss 6.5923
-2000 val perplexity 729.4580
-2000 train 6.609406 (lr=2.7857e-05) (hash(x)=154559671)
-2000 val loss 6.4736
-2000 val perplexity 647.8088
-2000 train 6.496451 (lr=3.2500e-05) (hash(x)=154559671)
-2100 val loss 6.5569
-2100 val perplexity 704.1078
-2100 train 6.621999 (lr=2.7571e-05) (hash(x)=155607137)
-2100 val loss 6.4424
-2100 val perplexity 627.8850
-2100 train 6.498925 (lr=3.2166e-05) (hash(x)=155607137)
-2200 val loss 6.5175
-2200 val perplexity 676.9085
-2200 train 6.371143 (lr=2.7269e-05) (hash(x)=153010221)
-2200 val loss 6.4053
-2200 val perplexity 605.0538
-2200 train 6.259152 (lr=3.1813e-05) (hash(x)=153010221)
-2300 val loss 6.4806
-2300 val perplexity 652.3843
-2300 train 6.504387 (lr=2.6951e-05) (hash(x)=160097777)
-2300 val loss 6.3697
-2300 val perplexity 583.9057
-2300 train 6.387885 (lr=3.1443e-05) (hash(x)=160097777)
-2400 val loss 6.4437
-2400 val perplexity 628.7560
-2400 train 6.401982 (lr=2.6618e-05) (hash(x)=135590094)
-2400 val loss 6.3429
-2400 val perplexity 568.4287
-2400 train 6.308095 (lr=3.1054e-05) (hash(x)=135590094)
-2500 val loss 6.4175
-2500 val perplexity 612.4648
-2500 train 6.538583 (lr=2.6270e-05) (hash(x)=147042929)
-2500 val loss 6.3173
-2500 val perplexity 554.0828
-2500 train 6.432256 (lr=3.0649e-05) (hash(x)=147042929)
-2600 val loss 6.3833
-2600 val perplexity 591.8883
-2600 train 6.348204 (lr=2.5909e-05) (hash(x)=163793881)
-2600 val loss 6.2817
-2600 val perplexity 534.7117
-2600 train 6.245691 (lr=3.0227e-05) (hash(x)=163793881)
-2700 val loss 6.3584
-2700 val perplexity 577.3007
-2700 train 6.389218 (lr=2.5533e-05) (hash(x)=153598698)
-2700 val loss 6.2576
-2700 val perplexity 521.9579
-2700 train 6.286983 (lr=2.9789e-05) (hash(x)=153598698)
-2800 val loss 6.3386
-2800 val perplexity 565.9789
-2800 train 6.140002 (lr=2.5145e-05) (hash(x)=133439803)
-2800 val loss 6.2427
-2800 val perplexity 514.2557
-2800 train 6.040616 (lr=2.9336e-05) (hash(x)=133439803)
-2900 val loss 6.3106
-2900 val perplexity 550.3931
-2900 train 6.254204 (lr=2.4744e-05) (hash(x)=148239158)
-2900 val loss 6.2123
-2900 val perplexity 498.8553
-2900 train 6.154753 (lr=2.8868e-05) (hash(x)=148239158)
-3000 val loss 6.2870
-3000 val perplexity 537.5233
-3000 train 6.281301 (lr=2.4331e-05) (hash(x)=150991971)
-3000 val loss 6.1935
-3000 val perplexity 489.5722
-3000 train 6.193192 (lr=2.8386e-05) (hash(x)=150991971)
-3100 val loss 6.2690
-3100 val perplexity 527.9424
-3100 train 6.275957 (lr=2.3906e-05) (hash(x)=157791832)
-3100 val loss 6.1763
-3100 val perplexity 481.2045
-3100 train 6.180146 (lr=2.7891e-05) (hash(x)=157791832)
-3200 val loss 6.2427
-3200 val perplexity 514.2594
-3200 train 6.317775 (lr=2.3471e-05) (hash(x)=163746305)
-3200 val loss 6.1504
-3200 val perplexity 468.9072
-3200 train 6.224569 (lr=2.7383e-05) (hash(x)=163746305)
-3300 val loss 6.2315
-3300 val perplexity 508.5287
-3300 train 6.160985 (lr=2.3026e-05) (hash(x)=155120269)
-3300 val loss 6.1413
-3300 val perplexity 464.6666
-3300 train 6.069816 (lr=2.6864e-05) (hash(x)=155120269)
-3400 val loss 6.2016
-3400 val perplexity 493.5572
-3400 train 6.141113 (lr=2.2572e-05) (hash(x)=152748451)
-3400 val loss 6.1122
-3400 val perplexity 451.3326
-3400 train 6.053649 (lr=2.6333e-05) (hash(x)=152748451)
-3500 val loss 6.1898
-3500 val perplexity 487.7594
-3500 train 6.065310 (lr=2.2108e-05) (hash(x)=147395004)
-3500 val loss 6.0981
-3500 val perplexity 445.0312
-3500 train 5.977302 (lr=2.5793e-05) (hash(x)=147395004)
-3600 val loss 6.1764
-3600 val perplexity 481.2518
-3600 train 6.201883 (lr=2.1637e-05) (hash(x)=157554250)
-3600 val loss 6.0854
-3600 val perplexity 439.3748
-3600 train 6.107175 (lr=2.5243e-05) (hash(x)=157554250)
-3700 val loss 6.1583
-3700 val perplexity 472.6462
-3700 train 6.011199 (lr=2.1158e-05) (hash(x)=147953321)
-3700 val loss 6.0681
-3700 val perplexity 431.8563
-3700 train 5.921242 (lr=2.4684e-05) (hash(x)=147953321)
-3800 val loss 6.1382
-3800 val perplexity 463.2388
-3800 train 5.997552 (lr=2.0672e-05) (hash(x)=146898609)
-3800 val loss 6.0475
-3800 val perplexity 423.0692
-3800 train 5.903200 (lr=2.4117e-05) (hash(x)=146898609)
-3900 val loss 6.1327
-3900 val perplexity 460.6937
-3900 train 6.182793 (lr=2.0180e-05) (hash(x)=150115282)
-3900 val loss 6.0465
-3900 val perplexity 422.6220
-3900 train 6.085420 (lr=2.3543e-05) (hash(x)=150115282)
-4000 val loss 6.1066
-4000 val perplexity 448.7970
-4000 train 5.963685 (lr=1.9683e-05) (hash(x)=142448619)
-4000 val loss 6.0158
-4000 val perplexity 409.8366
-4000 train 5.875909 (lr=2.2963e-05) (hash(x)=142448619)
-4100 val loss 6.1067
-4100 val perplexity 448.8636
-4100 train 5.829423 (lr=1.9181e-05) (hash(x)=141244123)
-4100 val loss 6.0114
-4100 val perplexity 408.0352
-4100 train 5.743440 (lr=2.2378e-05) (hash(x)=141244123)
-4200 val loss 6.0938
-4200 val perplexity 443.1073
-4200 train 6.123459 (lr=1.8675e-05) (hash(x)=165913661)
-4200 val loss 6.0059
-4200 val perplexity 405.8220
-4200 train 6.037457 (lr=2.1788e-05) (hash(x)=165913661)
-4300 val loss 6.0734
-4300 val perplexity 434.1442
-4300 train 6.109362 (lr=1.8166e-05) (hash(x)=155744823)
-4300 val loss 5.9803
-4300 val perplexity 395.5565
-4300 train 6.013528 (lr=2.1194e-05) (hash(x)=155744823)
-4400 val loss 6.0636
-4400 val perplexity 429.9022
-4400 train 6.056406 (lr=1.7655e-05) (hash(x)=160461243)
-4400 val loss 5.9719
-4400 val perplexity 392.2354
-4400 train 5.965618 (lr=2.0598e-05) (hash(x)=160461243)
-4500 val loss 6.0538
-4500 val perplexity 425.7418
-4500 train 5.980674 (lr=1.7142e-05) (hash(x)=156073129)
-4500 val loss 5.9604
-4500 val perplexity 387.7795
-4500 train 5.879493 (lr=1.9999e-05) (hash(x)=156073129)
-4600 val loss 6.0368
-4600 val perplexity 418.5724
-4600 train 6.086256 (lr=1.6629e-05) (hash(x)=152294132)
-4600 val loss 5.9469
-4600 val perplexity 382.5828
-4600 train 5.993891 (lr=1.9400e-05) (hash(x)=152294132)
-4700 val loss 6.0269
-4700 val perplexity 414.4126
-4700 train 5.966726 (lr=1.6114e-05) (hash(x)=140904569)
-4700 val loss 5.9346
-4700 val perplexity 377.8774
-4700 train 5.868339 (lr=1.8800e-05) (hash(x)=140904569)
-4800 val loss 6.0212
-4800 val perplexity 412.0571
-4800 train 5.943006 (lr=1.5601e-05) (hash(x)=156191635)
-4800 val loss 5.9285
-4800 val perplexity 375.5807
-4800 train 5.838706 (lr=1.8201e-05) (hash(x)=156191635)
-4900 val loss 6.0084
-4900 val perplexity 406.8464
-4900 train 6.088559 (lr=1.5089e-05) (hash(x)=145063976)
-4900 val loss 5.9164
-4900 val perplexity 371.0731
-4900 train 6.010023 (lr=1.7604e-05) (hash(x)=145063976)
-5000 val loss 6.0026
-5000 val perplexity 404.4739
-5000 train 5.947785 (lr=1.4579e-05) (hash(x)=160110619)
-5000 val loss 5.9095
-5000 val perplexity 368.5036
-5000 train 5.861709 (lr=1.7009e-05) (hash(x)=160110619)
-5100 val loss 5.9873
-5100 val perplexity 398.3270
-5100 train 6.010914 (lr=1.4071e-05) (hash(x)=156270070)
-5100 val loss 5.8941
-5100 val perplexity 362.8853
-5100 train 5.915792 (lr=1.6417e-05) (hash(x)=156270070)
-5200 val loss 5.9787
-5200 val perplexity 394.9133
-5200 train 5.849009 (lr=1.3568e-05) (hash(x)=138954242)
-5200 val loss 5.8882
-5200 val perplexity 360.7615
-5200 train 5.765492 (lr=1.5829e-05) (hash(x)=138954242)
-5300 val loss 5.9750
-5300 val perplexity 393.4499
-5300 train 5.825052 (lr=1.3068e-05) (hash(x)=146472367)
-5300 val loss 5.8781
-5300 val perplexity 357.1307
-5300 train 5.724410 (lr=1.5246e-05) (hash(x)=146472367)
-5400 val loss 5.9621
-5400 val perplexity 388.4326
-5400 train 5.902071 (lr=1.2573e-05) (hash(x)=146753405)
-5400 val loss 5.8667
-5400 val perplexity 353.0806
-5400 train 5.797910 (lr=1.4669e-05) (hash(x)=146753405)
-5500 val loss 5.9555
-5500 val perplexity 385.8689
-5500 train 5.900256 (lr=1.2085e-05) (hash(x)=147757398)
-5500 val loss 5.8596
-5500 val perplexity 350.5704
-5500 train 5.799724 (lr=1.4099e-05) (hash(x)=147757398)
-5600 val loss 5.9502
-5600 val perplexity 383.8358
-5600 train 5.831348 (lr=1.1602e-05) (hash(x)=151291757)
-5600 val loss 5.8539
-5600 val perplexity 348.5839
-5600 train 5.729513 (lr=1.3536e-05) (hash(x)=151291757)
-5700 val loss 5.9380
-5700 val perplexity 379.1620
-5700 train 5.980654 (lr=1.1127e-05) (hash(x)=156274392)
-5700 val loss 5.8421
-5700 val perplexity 344.4909
-5700 train 5.884508 (lr=1.2981e-05) (hash(x)=156274392)
-5800 val loss 5.9322
-5800 val perplexity 376.9763
-5800 train 5.895813 (lr=1.0659e-05) (hash(x)=152612266)
-5800 val loss 5.8349
-5800 val perplexity 342.0416
-5800 train 5.799439 (lr=1.2436e-05) (hash(x)=152612266)
-5900 val loss 5.9274
-5900 val perplexity 375.1920
-5900 train 6.081093 (lr=1.0200e-05) (hash(x)=148973201)
-5900 val loss 5.8301
-5900 val perplexity 340.3840
-5900 train 5.991538 (lr=1.1900e-05) (hash(x)=148973201)
-6000 val loss 5.9201
-6000 val perplexity 372.4622
-6000 train 5.883375 (lr=9.7500e-06) (hash(x)=158311490)
-6000 val loss 5.8212
-6000 val perplexity 337.3903
-6000 train 5.787454 (lr=1.1375e-05) (hash(x)=158311490)
-6100 val loss 5.9126
-6100 val perplexity 369.6491
-6100 train 5.831893 (lr=9.3098e-06) (hash(x)=147541319)
-6100 val loss 5.8151
-6100 val perplexity 335.3258
-6100 train 5.741688 (lr=1.0861e-05) (hash(x)=147541319)
-6200 val loss 5.9105
-6200 val perplexity 368.8990
-6200 train 5.971362 (lr=8.8800e-06) (hash(x)=154753980)
-6200 val loss 5.8099
-6200 val perplexity 333.5760
-6200 train 5.873833 (lr=1.0360e-05) (hash(x)=154753980)
-6300 val loss 5.9034
-6300 val perplexity 366.2755
-6300 train 5.801030 (lr=8.4613e-06) (hash(x)=140798998)
-6300 val loss 5.8038
-6300 val perplexity 331.5592
-6300 train 5.703178 (lr=9.8715e-06) (hash(x)=140798998)
-6400 val loss 5.8987
-6400 val perplexity 364.5718
-6400 train 5.783006 (lr=8.0542e-06) (hash(x)=149299516)
-6400 val loss 5.8005
-6400 val perplexity 330.4565
-6400 train 5.687269 (lr=9.3966e-06) (hash(x)=149299516)
-6500 val loss 5.8946
-6500 val perplexity 363.0667
-6500 train 5.893285 (lr=7.6594e-06) (hash(x)=148495806)
-6500 val loss 5.7949
-6500 val perplexity 328.6337
-6500 train 5.787768 (lr=8.9359e-06) (hash(x)=148495806)
-6600 val loss 5.8868
-6600 val perplexity 360.2643
-6600 train 5.860414 (lr=7.2774e-06) (hash(x)=153664387)
-6600 val loss 5.7871
-6600 val perplexity 326.0732
-6600 train 5.761164 (lr=8.4903e-06) (hash(x)=153664387)
-6700 val loss 5.8848
-6700 val perplexity 359.5459
-6700 train 5.866759 (lr=6.9087e-06) (hash(x)=153333491)
-6700 val loss 5.7843
-6700 val perplexity 325.1652
-6700 train 5.766811 (lr=8.0602e-06) (hash(x)=153333491)
-6800 val loss 5.8856
-6800 val perplexity 359.8354
-6800 train 5.644751 (lr=6.5540e-06) (hash(x)=150149382)
-6800 val loss 5.7835
-6800 val perplexity 324.8955
-6800 train 5.536487 (lr=7.6463e-06) (hash(x)=150149382)
-6900 val loss 5.8772
-6900 val perplexity 356.8184
-6900 train 5.949944 (lr=6.2137e-06) (hash(x)=152379767)
-6900 val loss 5.7778
-6900 val perplexity 323.0596
-6900 train 5.856633 (lr=7.2493e-06) (hash(x)=152379767)
-7000 val loss 5.8711
-7000 val perplexity 354.6424
-7000 train 5.916762 (lr=5.8883e-06) (hash(x)=150432537)
-7000 val loss 5.7705
-7000 val perplexity 320.6822
-7000 train 5.818938 (lr=6.8697e-06) (hash(x)=150432537)
-7100 val loss 5.8700
-7100 val perplexity 354.2435
-7100 train 5.747322 (lr=5.5783e-06) (hash(x)=148277836)
-7100 val loss 5.7707
-7100 val perplexity 320.7480
-7100 train 5.649139 (lr=6.5080e-06) (hash(x)=148277836)
-7200 val loss 5.8696
-7200 val perplexity 354.0941
-7200 train 5.876884 (lr=5.2841e-06) (hash(x)=168591659)
-7200 val loss 5.7699
-7200 val perplexity 320.5116
-7200 train 5.778949 (lr=6.1648e-06) (hash(x)=168591659)
-7300 val loss 5.8627
-7300 val perplexity 351.6757
-7300 train 5.839412 (lr=5.0062e-06) (hash(x)=152764851)
-7300 val loss 5.7618
-7300 val perplexity 317.9327
-7300 train 5.739857 (lr=5.8405e-06) (hash(x)=152764851)
-7400 val loss 5.8589
-7400 val perplexity 350.3225
-7400 train 5.809220 (lr=4.7449e-06) (hash(x)=142920131)
-7400 val loss 5.7583
-7400 val perplexity 316.8126
-7400 train 5.714902 (lr=5.5357e-06) (hash(x)=142920131)
-7500 val loss 5.8567
-7500 val perplexity 349.5636
-7500 train 5.858554 (lr=4.5007e-06) (hash(x)=146954649)
-7500 val loss 5.7558
-7500 val perplexity 316.0316
-7500 train 5.757757 (lr=5.2508e-06) (hash(x)=146954649)
-7600 val loss 5.8574
-7600 val perplexity 349.8061
-7600 train 5.648992 (lr=4.2739e-06) (hash(x)=144529566)
-7600 val loss 5.7568
-7600 val perplexity 316.3331
-7600 train 5.546754 (lr=4.9862e-06) (hash(x)=144529566)
-7700 val loss 5.8515
-7700 val perplexity 347.7727
-7700 train 5.806117 (lr=4.0648e-06) (hash(x)=142669945)
-7700 val loss 5.7503
-7700 val perplexity 314.2869
-7700 train 5.707147 (lr=4.7423e-06) (hash(x)=142669945)
-7800 val loss 5.8481
-7800 val perplexity 346.5742
-7800 train 5.919096 (lr=3.8738e-06) (hash(x)=161563764)
-7800 val loss 5.7467
-7800 val perplexity 313.1556
-7800 train 5.816212 (lr=4.5194e-06) (hash(x)=161563764)
-7900 val loss 5.8457
-7900 val perplexity 345.7390
-7900 train 5.643712 (lr=3.7010e-06) (hash(x)=141516532)
-7900 val loss 5.7451
-7900 val perplexity 312.6412
-7900 train 5.545619 (lr=4.3179e-06) (hash(x)=141516532)
-8000 val loss 5.8473
-8000 val perplexity 346.2927
-8000 train 5.716923 (lr=3.5468e-06) (hash(x)=152418457)
-8000 val loss 5.7454
-8000 val perplexity 312.7387
-8000 train 5.618537 (lr=4.1380e-06) (hash(x)=152418457)
-8100 val loss 5.8420
-8100 val perplexity 344.4742
-8100 train 5.831784 (lr=3.4114e-06) (hash(x)=161109470)
-8100 val loss 5.7407
-8100 val perplexity 311.2804
-8100 train 5.731179 (lr=3.9800e-06) (hash(x)=161109470)
-8200 val loss 5.8385
-8200 val perplexity 343.2647
-8200 train 5.958980 (lr=3.2950e-06) (hash(x)=154408822)
-8200 val loss 5.7371
-8200 val perplexity 310.1499
-8200 train 5.859767 (lr=3.8442e-06) (hash(x)=154408822)
-8300 val loss 5.8363
-8300 val perplexity 342.5162
-8300 train 5.717995 (lr=3.1977e-06) (hash(x)=146431691)
-8300 val loss 5.7353
-8300 val perplexity 309.5942
-8300 train 5.619819 (lr=3.7307e-06) (hash(x)=146431691)
-8400 val loss 5.8365
-8400 val perplexity 342.5848
-8400 train 5.609816 (lr=3.1197e-06) (hash(x)=150182527)
-8400 val loss 5.7352
-8400 val perplexity 309.5880
-8400 train 5.512958 (lr=3.6397e-06) (hash(x)=150182527)
-8500 val loss 5.8311
-8500 val perplexity 340.7312
-8500 train 5.901333 (lr=3.0611e-06) (hash(x)=154828709)
-8500 val loss 5.7297
-8500 val perplexity 307.8887
-8500 train 5.799944 (lr=3.5713e-06) (hash(x)=154828709)
-8600 val loss 5.8306
-8600 val perplexity 340.5702
-8600 train 5.737008 (lr=3.0220e-06) (hash(x)=154995428)
-8600 val loss 5.7294
-8600 val perplexity 307.7847
-8600 train 5.641396 (lr=3.5257e-06) (hash(x)=154995428)
-8700 val loss 5.8302
-8700 val perplexity 340.4246
-8700 train 5.724449 (lr=3.0024e-06) (hash(x)=142111043)
-8700 val loss 5.7292
-8700 val perplexity 307.7150
-8700 train 5.626750 (lr=3.5029e-06) (hash(x)=142111043)
-8749 val loss 5.8284
-8749 val perplexity 339.8080
-8749 val loss 5.7264
-8749 val perplexity 306.8476
+0 train 11.289043 (lr=9.0000e-08) (hash(x)=150332693)
+100 val loss 10.0243
+100 val perplexity 22568.6875
+100 train 9.978326 (lr=9.0900e-06) (hash(x)=149277319)
+200 val loss 9.3785
+200 val perplexity 11831.2090
+200 train 9.445846 (lr=1.8090e-05) (hash(x)=155934075)
+300 val loss 8.2194
+300 val perplexity 3712.2971
+300 train 8.254610 (lr=2.7090e-05) (hash(x)=165402628)
+400 val loss 7.6163
+400 val perplexity 2031.0946
+400 train 7.421521 (lr=3.6090e-05) (hash(x)=139168939)
+500 val loss 7.4349
+500 val perplexity 1694.0548
+500 train 7.357471 (lr=4.5000e-05) (hash(x)=147134338)
+600 val loss 7.3305
+600 val perplexity 1526.1565
+600 train 7.243377 (lr=4.4985e-05) (hash(x)=159086459)
+700 val loss 7.2503
+700 val perplexity 1408.4863
+700 train 7.288620 (lr=4.4941e-05) (hash(x)=152399099)
+800 val loss 7.1363
+800 val perplexity 1256.7866
+800 train 7.140655 (lr=4.4868e-05) (hash(x)=156548843)
+900 val loss 7.0484
+900 val perplexity 1151.0243
+900 train 7.021039 (lr=4.4766e-05) (hash(x)=157889911)
+1000 val loss 6.9766
+1000 val perplexity 1071.3145
+1000 train 6.774483 (lr=4.4634e-05) (hash(x)=142943707)
+1100 val loss 6.9135
+1100 val perplexity 1005.7903
+1100 train 6.912020 (lr=4.4474e-05) (hash(x)=148938689)
+1200 val loss 6.8143
+1200 val perplexity 910.7801
+1200 train 6.954645 (lr=4.4285e-05) (hash(x)=161962903)
+1300 val loss 6.7411
+1300 val perplexity 846.4569
+1300 train 6.723616 (lr=4.4068e-05) (hash(x)=145979235)
+1400 val loss 6.6754
+1400 val perplexity 792.6756
+1400 train 6.625307 (lr=4.3822e-05) (hash(x)=159157507)
+1500 val loss 6.6153
+1500 val perplexity 746.3971
+1500 train 6.649820 (lr=4.3549e-05) (hash(x)=145943130)
+1600 val loss 6.5401
+1600 val perplexity 692.3466
+1600 train 6.341069 (lr=4.3249e-05) (hash(x)=143249173)
+1700 val loss 6.4872
+1700 val perplexity 656.6964
+1700 train 6.551693 (lr=4.2922e-05) (hash(x)=173907143)
+1800 val loss 6.4411
+1800 val perplexity 627.1124
+1800 train 6.368075 (lr=4.2569e-05) (hash(x)=138363755)
+1900 val loss 6.4086
+1900 val perplexity 607.0637
+1900 train 6.216325 (lr=4.2190e-05) (hash(x)=146341390)
+2000 val loss 6.3554
+2000 val perplexity 575.6113
+2000 train 6.384245 (lr=4.1785e-05) (hash(x)=154559671)
+2100 val loss 6.3281
+2100 val perplexity 560.0664
+2100 train 6.378413 (lr=4.1356e-05) (hash(x)=155607137)
+2200 val loss 6.2939
+2200 val perplexity 541.2359
+2200 train 6.150013 (lr=4.0903e-05) (hash(x)=153010221)
+2300 val loss 6.2609
+2300 val perplexity 523.7111
+2300 train 6.278090 (lr=4.0426e-05) (hash(x)=160097777)
+2400 val loss 6.2372
+2400 val perplexity 511.4395
+2400 train 6.214860 (lr=3.9927e-05) (hash(x)=135590094)
+2500 val loss 6.2120
+2500 val perplexity 498.6850
+2500 train 6.329730 (lr=3.9406e-05) (hash(x)=147042929)
+2600 val loss 6.1775
+2600 val perplexity 481.7953
+2600 train 6.138986 (lr=3.8863e-05) (hash(x)=163793881)
+2700 val loss 6.1626
+2700 val perplexity 474.6808
+2700 train 6.195170 (lr=3.8300e-05) (hash(x)=153598698)
+2800 val loss 6.1463
+2800 val perplexity 466.9887
+2800 train 5.944228 (lr=3.7717e-05) (hash(x)=133439803)
+2900 val loss 6.1207
+2900 val perplexity 455.1814
+2900 train 6.059956 (lr=3.7116e-05) (hash(x)=148239158)
+3000 val loss 6.0949
+3000 val perplexity 443.6075
+3000 train 6.099676 (lr=3.6496e-05) (hash(x)=150991971)
+3100 val loss 6.0827
+3100 val perplexity 438.2101
+3100 train 6.087112 (lr=3.5860e-05) (hash(x)=157791832)
+3200 val loss 6.0558
+3200 val perplexity 426.5774
+3200 train 6.124968 (lr=3.5207e-05) (hash(x)=163746305)
+3300 val loss 6.0481
+3300 val perplexity 423.3175
+3300 train 5.978697 (lr=3.4539e-05) (hash(x)=155120269)
+3400 val loss 6.0166
+3400 val perplexity 410.1862
+3400 train 5.958898 (lr=3.3857e-05) (hash(x)=152748451)
+3500 val loss 6.0043
+3500 val perplexity 405.1582
+3500 train 5.885794 (lr=3.3162e-05) (hash(x)=147395004)
+3600 val loss 5.9914
+3600 val perplexity 399.9552
+3600 train 6.009531 (lr=3.2455e-05) (hash(x)=157554250)
+3700 val loss 5.9740
+3700 val perplexity 393.0799
+3700 train 5.826887 (lr=3.1736e-05) (hash(x)=147953321)
+3800 val loss 5.9551
+3800 val perplexity 385.7162
+3800 train 5.805262 (lr=3.1008e-05) (hash(x)=146898609)
+3900 val loss 5.9564
+3900 val perplexity 386.2080
+3900 train 5.996837 (lr=3.0270e-05) (hash(x)=150115282)
+4000 val loss 5.9218
+4000 val perplexity 373.0663
+4000 train 5.790571 (lr=2.9524e-05) (hash(x)=142448619)
+4100 val loss 5.9223
+4100 val perplexity 373.2632
+4100 train 5.654961 (lr=2.8771e-05) (hash(x)=141244123)
+4200 val loss 5.9140
+4200 val perplexity 370.1866
+4200 train 5.927835 (lr=2.8013e-05) (hash(x)=165913661)
+4300 val loss 5.8861
+4300 val perplexity 360.0132
+4300 train 5.917614 (lr=2.7250e-05) (hash(x)=155744823)
+4400 val loss 5.8769
+4400 val perplexity 356.6874
+4400 train 5.872905 (lr=2.6483e-05) (hash(x)=160461243)
+4500 val loss 5.8657
+4500 val perplexity 352.7454
+4500 train 5.776689 (lr=2.5714e-05) (hash(x)=156073129)
+4600 val loss 5.8505
+4600 val perplexity 347.4088
+4600 train 5.897415 (lr=2.4943e-05) (hash(x)=152294132)
+4700 val loss 5.8393
+4700 val perplexity 343.5292
+4700 train 5.771152 (lr=2.4172e-05) (hash(x)=140904569)
+4800 val loss 5.8338
+4800 val perplexity 341.6528
+4800 train 5.739367 (lr=2.3402e-05) (hash(x)=156191635)
+4900 val loss 5.8181
+4900 val perplexity 336.3423
+4900 train 5.924228 (lr=2.2633e-05) (hash(x)=145063976)
+5000 val loss 5.8120
+5000 val perplexity 334.2776
+5000 train 5.754147 (lr=2.1868e-05) (hash(x)=160110619)
+5100 val loss 5.7954
+5100 val perplexity 328.7745
+5100 train 5.821685 (lr=2.1107e-05) (hash(x)=156270070)
+5200 val loss 5.7867
+5200 val perplexity 325.9519
+5200 train 5.674892 (lr=2.0351e-05) (hash(x)=138954242)
+5300 val loss 5.7796
+5300 val perplexity 323.6199
+5300 train 5.633004 (lr=1.9602e-05) (hash(x)=146472367)
+5400 val loss 5.7699
+5400 val perplexity 320.5030
+5400 train 5.702404 (lr=1.8860e-05) (hash(x)=146753405)
+5500 val loss 5.7575
+5500 val perplexity 316.5528
+5500 train 5.706319 (lr=1.8127e-05) (hash(x)=147757398)
+5600 val loss 5.7557
+5600 val perplexity 315.9973
+5600 train 5.624521 (lr=1.7403e-05) (hash(x)=151291757)
+5700 val loss 5.7424
+5700 val perplexity 311.8269
+5700 train 5.793109 (lr=1.6690e-05) (hash(x)=156274392)
+5800 val loss 5.7336
+5800 val perplexity 309.0707
+5800 train 5.702478 (lr=1.5989e-05) (hash(x)=152612266)
+5900 val loss 5.7301
+5900 val perplexity 307.9881
+5900 train 5.893963 (lr=1.5300e-05) (hash(x)=148973201)
+6000 val loss 5.7217
+6000 val perplexity 305.4335
+6000 train 5.687572 (lr=1.4625e-05) (hash(x)=158311490)
+6100 val loss 5.7135
+6100 val perplexity 302.9321
+6100 train 5.645096 (lr=1.3965e-05) (hash(x)=147541319)
+6200 val loss 5.7129
+6200 val perplexity 302.7402
+6200 train 5.777384 (lr=1.3320e-05) (hash(x)=154753980)
+6300 val loss 5.7037
+6300 val perplexity 299.9899
+6300 train 5.601553 (lr=1.2692e-05) (hash(x)=140798998)
+6400 val loss 5.6993
+6400 val perplexity 298.6710
+6400 train 5.588508 (lr=1.2081e-05) (hash(x)=149299516)
+6500 val loss 5.6953
+6500 val perplexity 297.4683
+6500 train 5.682362 (lr=1.1489e-05) (hash(x)=148495806)
+6600 val loss 5.6878
+6600 val perplexity 295.2325
+6600 train 5.665105 (lr=1.0916e-05) (hash(x)=153664387)
+6700 val loss 5.6849
+6700 val perplexity 294.3740
+6700 train 5.661620 (lr=1.0363e-05) (hash(x)=153333491)
+6800 val loss 5.6854
+6800 val perplexity 294.5452
+6800 train 5.438633 (lr=9.8310e-06) (hash(x)=150149382)
+6900 val loss 5.6780
+6900 val perplexity 292.3573
+6900 train 5.763326 (lr=9.3205e-06) (hash(x)=152379767)
+7000 val loss 5.6708
+7000 val perplexity 290.2598
+7000 train 5.719643 (lr=8.8324e-06) (hash(x)=150432537)
+7100 val loss 5.6688
+7100 val perplexity 289.6923
+7100 train 5.545722 (lr=8.3674e-06) (hash(x)=148277836)
+7200 val loss 5.6698
+7200 val perplexity 289.9790
+7200 train 5.664272 (lr=7.9261e-06) (hash(x)=168591659)
+7300 val loss 5.6615
+7300 val perplexity 287.5750
+7300 train 5.642782 (lr=7.5093e-06) (hash(x)=152764851)
+7400 val loss 5.6584
+7400 val perplexity 286.6996
+7400 train 5.621808 (lr=7.1174e-06) (hash(x)=142920131)
+7500 val loss 5.6559
+7500 val perplexity 285.9665
+7500 train 5.656323 (lr=6.7511e-06) (hash(x)=146954649)
+7600 val loss 5.6564
+7600 val perplexity 286.1283
+7600 train 5.434806 (lr=6.4109e-06) (hash(x)=144529566)
+7700 val loss 5.6503
+7700 val perplexity 284.3646
+7700 train 5.609683 (lr=6.0972e-06) (hash(x)=142669945)
+7800 val loss 5.6460
+7800 val perplexity 283.1537
+7800 train 5.704182 (lr=5.8107e-06) (hash(x)=161563764)
+7900 val loss 5.6443
+7900 val perplexity 282.6702
+7900 train 5.439746 (lr=5.5515e-06) (hash(x)=141516532)
+8000 val loss 5.6455
+8000 val perplexity 283.0064
+8000 train 5.515144 (lr=5.3203e-06) (hash(x)=152418457)
+8100 val loss 5.6398
+8100 val perplexity 281.4167
+8100 train 5.624706 (lr=5.1172e-06) (hash(x)=161109470)
+8200 val loss 5.6360
+8200 val perplexity 280.3446
+8200 train 5.758985 (lr=4.9425e-06) (hash(x)=154408822)
+8300 val loss 5.6342
+8300 val perplexity 279.8434
+8300 train 5.520042 (lr=4.7966e-06) (hash(x)=146431691)
+8400 val loss 5.6345
+8400 val perplexity 279.9092
+8400 train 5.412016 (lr=4.6796e-06) (hash(x)=150182527)
+8500 val loss 5.6286
+8500 val perplexity 278.2667
+8500 train 5.693805 (lr=4.5917e-06) (hash(x)=154828709)
+8600 val loss 5.6284
+8600 val perplexity 278.2239
+8600 train 5.539424 (lr=4.5330e-06) (hash(x)=154995428)
+8700 val loss 5.6285
+8700 val perplexity 278.2420
+8700 train 5.523919 (lr=4.5037e-06) (hash(x)=142111043)
+8749 val loss 5.6262
+8749 val perplexity 277.6001

attention_kindselective_n_heads4_seed1342/model_08749.pt

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:50ff8744afab0bf33a1f184229c55f6f938e2f46b13c2d1aae2f9e9fc3be54cd
+oid sha256:3b784ac0421732561a9feaba30a9e9fef056d1e13bb74eb961fa766d7412a975
 size 92843394

attention_kindselective_n_heads4_seed1342/optimizer_08749.pt

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:51e2ecb9dfd6d5a8f8b29373903243ccda842f45d2f42561a1f91995ae236f24
+oid sha256:a42419c28aa5e439c9bce28733a36c3eb5bf07c6395ea15a1c7f729048403a90
 size 179406214