Upload folder using huggingface_hub

lr5.5e-5_total_batch_size61440_baseline_seed1338/args.json

@@ -0,0 +1 @@
+{"hellaswag": true, "attention_kind": "selective", "log_dir": "wider_is_better_10/lr5.5e-5_total_batch_size61440_baseline_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": 8750, "warmup_steps": 500, "group": "wider_is_better_10", "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": 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_1338", "n_embd": 256}

lr5.5e-5_total_batch_size61440_baseline_seed1338/dataloader_08749.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:953385078aa3787b69fc6857dfd48b0a2cd2f4d27c6f8892e01211aca53d07f5
+size 964

lr5.5e-5_total_batch_size61440_baseline_seed1338/log2.txt

@@ -0,0 +1,267 @@
+max_steps: 8750
+0 val loss 11.2646
+0 val perplexity 78014.3047
+0 train 11.267750 (lr=1.1000e-07) (hash(x)=150327452)
+100 val loss 9.9450
+100 val perplexity 20847.2383
+100 train 9.996918 (lr=1.1110e-05) (hash(x)=166780046)
+200 val loss 9.1141
+200 val perplexity 9082.8682
+200 train 9.053102 (lr=2.2110e-05) (hash(x)=155040610)
+300 val loss 8.0183
+300 val perplexity 3035.9395
+300 train 8.014933 (lr=3.3110e-05) (hash(x)=155504036)
+400 val loss 7.6026
+400 val perplexity 2003.3218
+400 train 7.466523 (lr=4.4110e-05) (hash(x)=143823248)
+500 val loss 7.4847
+500 val perplexity 1780.6073
+500 train 7.454373 (lr=5.5000e-05) (hash(x)=143734685)
+600 val loss 7.4313
+600 val perplexity 1687.9413
+600 train 7.214120 (lr=5.4982e-05) (hash(x)=150678249)
+700 val loss 7.3474
+700 val perplexity 1552.1353
+700 train 7.486725 (lr=5.4928e-05) (hash(x)=175802021)
+800 val loss 7.2545
+800 val perplexity 1414.5203
+800 train 7.265283 (lr=5.4839e-05) (hash(x)=158681215)
+900 val loss 7.1814
+900 val perplexity 1314.7568
+900 train 7.077361 (lr=5.4713e-05) (hash(x)=146108145)
+1000 val loss 7.1197
+1000 val perplexity 1236.0907
+1000 train 6.961311 (lr=5.4553e-05) (hash(x)=154996086)
+1100 val loss 7.0403
+1100 val perplexity 1141.7784
+1100 train 7.024014 (lr=5.4357e-05) (hash(x)=153885445)
+1200 val loss 6.9773
+1200 val perplexity 1072.0734
+1200 train 6.931175 (lr=5.4126e-05) (hash(x)=142353087)
+1300 val loss 6.9208
+1300 val perplexity 1013.1246
+1300 train 6.828637 (lr=5.3860e-05) (hash(x)=150750353)
+1400 val loss 6.8724
+1400 val perplexity 965.2719
+1400 train 6.695466 (lr=5.3561e-05) (hash(x)=152767913)
+1500 val loss 6.7946
+1500 val perplexity 893.0278
+1500 train 6.830484 (lr=5.3227e-05) (hash(x)=151562048)
+1600 val loss 6.7141
+1600 val perplexity 823.9701
+1600 train 6.712261 (lr=5.2860e-05) (hash(x)=166486165)
+1700 val loss 6.6574
+1700 val perplexity 778.4987
+1700 train 6.303072 (lr=5.2461e-05) (hash(x)=130835396)
+1800 val loss 6.6013
+1800 val perplexity 736.0822
+1800 train 6.648942 (lr=5.2029e-05) (hash(x)=158851816)
+1900 val loss 6.5568
+1900 val perplexity 703.9802
+1900 train 6.589365 (lr=5.1565e-05) (hash(x)=153313879)
+2000 val loss 6.4923
+2000 val perplexity 660.0706
+2000 train 6.424527 (lr=5.1071e-05) (hash(x)=158245023)
+2100 val loss 6.4371
+2100 val perplexity 624.5764
+2100 train 6.263038 (lr=5.0547e-05) (hash(x)=157204896)
+2200 val loss 6.4097
+2200 val perplexity 607.7307
+2200 train 6.247912 (lr=4.9993e-05) (hash(x)=137541932)
+2300 val loss 6.3787
+2300 val perplexity 589.1797
+2300 train 6.332857 (lr=4.9410e-05) (hash(x)=150149692)
+2400 val loss 6.3293
+2400 val perplexity 560.7457
+2400 train 6.315440 (lr=4.8800e-05) (hash(x)=151730720)
+2500 val loss 6.3038
+2500 val perplexity 546.6617
+2500 train 6.080341 (lr=4.8162e-05) (hash(x)=143406752)
+2600 val loss 6.2849
+2600 val perplexity 536.4228
+2600 train 6.115074 (lr=4.7499e-05) (hash(x)=157272496)
+2700 val loss 6.2555
+2700 val perplexity 520.8868
+2700 train 6.251719 (lr=4.6811e-05) (hash(x)=155342327)
+2800 val loss 6.2251
+2800 val perplexity 505.2550
+2800 train 6.103718 (lr=4.6099e-05) (hash(x)=140626679)
+2900 val loss 6.2003
+2900 val perplexity 492.8994
+2900 train 6.058414 (lr=4.5364e-05) (hash(x)=144953350)
+3000 val loss 6.1796
+3000 val perplexity 482.8010
+3000 train 6.076258 (lr=4.4606e-05) (hash(x)=172449837)
+3100 val loss 6.1717
+3100 val perplexity 478.9760
+3100 train 5.958810 (lr=4.3828e-05) (hash(x)=141710086)
+3200 val loss 6.1354
+3200 val perplexity 461.9237
+3200 train 6.065051 (lr=4.3031e-05) (hash(x)=151299772)
+3300 val loss 6.1224
+3300 val perplexity 455.9704
+3300 train 5.989041 (lr=4.2215e-05) (hash(x)=146473110)
+3400 val loss 6.1019
+3400 val perplexity 446.7232
+3400 train 6.188746 (lr=4.1381e-05) (hash(x)=153954157)
+3500 val loss 6.0829
+3500 val perplexity 438.3020
+3500 train 6.019620 (lr=4.0532e-05) (hash(x)=153717336)
+3600 val loss 6.0678
+3600 val perplexity 431.7148
+3600 train 5.823302 (lr=3.9667e-05) (hash(x)=144965161)
+3700 val loss 6.0552
+3700 val perplexity 426.3289
+3700 train 5.833506 (lr=3.8789e-05) (hash(x)=125969741)
+3800 val loss 6.0324
+3800 val perplexity 416.7287
+3800 train 5.894809 (lr=3.7898e-05) (hash(x)=155070487)
+3900 val loss 6.0151
+3900 val perplexity 409.5789
+3900 train 5.890672 (lr=3.6996e-05) (hash(x)=149444644)
+4000 val loss 6.0067
+4000 val perplexity 406.1317
+4000 train 5.864836 (lr=3.6085e-05) (hash(x)=151663033)
+4100 val loss 5.9990
+4100 val perplexity 403.0375
+4100 train 5.953324 (lr=3.5165e-05) (hash(x)=143688282)
+4200 val loss 5.9740
+4200 val perplexity 393.0714
+4200 train 5.973290 (lr=3.4238e-05) (hash(x)=163361651)
+4300 val loss 5.9575
+4300 val perplexity 386.6399
+4300 train 5.988450 (lr=3.3305e-05) (hash(x)=153619361)
+4400 val loss 5.9552
+4400 val perplexity 385.7703
+4400 train 6.269998 (lr=3.2368e-05) (hash(x)=168527064)
+4500 val loss 5.9402
+4500 val perplexity 380.0159
+4500 train 5.754234 (lr=3.1428e-05) (hash(x)=125588037)
+4600 val loss 5.9200
+4600 val perplexity 372.4245
+4600 train 5.880640 (lr=3.0486e-05) (hash(x)=143710941)
+4700 val loss 5.9086
+4700 val perplexity 368.2047
+4700 train 5.740867 (lr=2.9543e-05) (hash(x)=150952742)
+4800 val loss 5.8953
+4800 val perplexity 363.3102
+4800 train 5.765694 (lr=2.8602e-05) (hash(x)=145323659)
+4900 val loss 5.8917
+4900 val perplexity 362.0049
+4900 train 5.903539 (lr=2.7663e-05) (hash(x)=153151397)
+5000 val loss 5.8786
+5000 val perplexity 357.2975
+5000 train 5.775647 (lr=2.6728e-05) (hash(x)=143182059)
+5100 val loss 5.8666
+5100 val perplexity 353.0298
+5100 train 5.900901 (lr=2.5798e-05) (hash(x)=170083586)
+5200 val loss 5.8507
+5200 val perplexity 347.4714
+5200 train 5.741638 (lr=2.4874e-05) (hash(x)=149363919)
+5300 val loss 5.8443
+5300 val perplexity 345.2589
+5300 train 5.728531 (lr=2.3958e-05) (hash(x)=152033784)
+5400 val loss 5.8293
+5400 val perplexity 340.1364
+5400 train 5.937089 (lr=2.3051e-05) (hash(x)=154614289)
+5500 val loss 5.8217
+5500 val perplexity 337.5583
+5500 train 5.912975 (lr=2.2155e-05) (hash(x)=157745174)
+5600 val loss 5.8152
+5600 val perplexity 335.3488
+5600 train 5.713684 (lr=2.1271e-05) (hash(x)=147693222)
+5700 val loss 5.8051
+5700 val perplexity 331.9754
+5700 train 5.742182 (lr=2.0399e-05) (hash(x)=149784627)
+5800 val loss 5.8038
+5800 val perplexity 331.5405
+5800 train 5.724194 (lr=1.9542e-05) (hash(x)=158620729)
+5900 val loss 5.7913
+5900 val perplexity 327.4529
+5900 train 5.737638 (lr=1.8700e-05) (hash(x)=159763910)
+6000 val loss 5.7816
+6000 val perplexity 324.2631
+6000 train 5.717546 (lr=1.7875e-05) (hash(x)=147640561)
+6100 val loss 5.7737
+6100 val perplexity 321.7176
+6100 train 5.750793 (lr=1.7068e-05) (hash(x)=156613394)
+6200 val loss 5.7690
+6200 val perplexity 320.2155
+6200 train 5.712326 (lr=1.6280e-05) (hash(x)=186221290)
+6300 val loss 5.7566
+6300 val perplexity 316.2713
+6300 train 5.636555 (lr=1.5512e-05) (hash(x)=152081419)
+6400 val loss 5.7545
+6400 val perplexity 315.6057
+6400 train 5.707115 (lr=1.4766e-05) (hash(x)=154808349)
+6500 val loss 5.7490
+6500 val perplexity 313.8748
+6500 train 5.720559 (lr=1.4042e-05) (hash(x)=159437208)
+6600 val loss 5.7469
+6600 val perplexity 313.2175
+6600 train 5.559500 (lr=1.3342e-05) (hash(x)=157933074)
+6700 val loss 5.7381
+6700 val perplexity 310.4663
+6700 train 5.716550 (lr=1.2666e-05) (hash(x)=161560240)
+6800 val loss 5.7332
+6800 val perplexity 308.9702
+6800 train 5.711833 (lr=1.2016e-05) (hash(x)=155424292)
+6900 val loss 5.7303
+6900 val perplexity 308.0646
+6900 train 5.724535 (lr=1.1392e-05) (hash(x)=148561470)
+7000 val loss 5.7225
+7000 val perplexity 305.6771
+7000 train 5.620068 (lr=1.0795e-05) (hash(x)=141527450)
+7100 val loss 5.7189
+7100 val perplexity 304.5638
+7100 train 5.676641 (lr=1.0227e-05) (hash(x)=151066339)
+7200 val loss 5.7204
+7200 val perplexity 305.0164
+7200 train 5.519127 (lr=9.6875e-06) (hash(x)=155231264)
+7300 val loss 5.7100
+7300 val perplexity 301.8799
+7300 train 5.728126 (lr=9.1780e-06) (hash(x)=150281149)
+7400 val loss 5.7074
+7400 val perplexity 301.0882
+7400 train 5.593415 (lr=8.6990e-06) (hash(x)=148421717)
+7500 val loss 5.7063
+7500 val perplexity 300.7696
+7500 train 5.468148 (lr=8.2513e-06) (hash(x)=146921118)
+7600 val loss 5.6994
+7600 val perplexity 298.6937
+7600 train 5.747622 (lr=7.8355e-06) (hash(x)=150660048)
+7700 val loss 5.6970
+7700 val perplexity 297.9716
+7700 train 5.572972 (lr=7.4522e-06) (hash(x)=148059852)
+7800 val loss 5.6964
+7800 val perplexity 297.7930
+7800 train 5.387359 (lr=7.1019e-06) (hash(x)=148331002)
+7900 val loss 5.6909
+7900 val perplexity 296.1748
+7900 train 5.637100 (lr=6.7852e-06) (hash(x)=164923883)
+8000 val loss 5.6881
+8000 val perplexity 295.3373
+8000 train 5.572701 (lr=6.5025e-06) (hash(x)=143545384)
+8100 val loss 5.6874
+8100 val perplexity 295.1346
+8100 train 5.503888 (lr=6.2543e-06) (hash(x)=160686959)
+8200 val loss 5.6850
+8200 val perplexity 294.4196
+8200 train 5.642776 (lr=6.0408e-06) (hash(x)=156501889)
+8300 val loss 5.6837
+8300 val perplexity 294.0495
+8300 train 5.522572 (lr=5.8625e-06) (hash(x)=142716875)
+8400 val loss 5.6778
+8400 val perplexity 292.2994
+8400 train 5.594144 (lr=5.7195e-06) (hash(x)=154436684)
+8500 val loss 5.6765
+8500 val perplexity 291.9289
+8500 train 5.754575 (lr=5.6121e-06) (hash(x)=147965839)
+8600 val loss 5.6745
+8600 val perplexity 291.3358
+8600 train 5.474033 (lr=5.5404e-06) (hash(x)=145228097)
+8700 val loss 5.6704
+8700 val perplexity 290.1638
+8700 train 5.862656 (lr=5.5045e-06) (hash(x)=152910357)
+8749 val loss 5.6702
+8749 val perplexity 290.0913

lr5.5e-5_total_batch_size61440_baseline_seed1338/model_08749.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:8d0fd315b76ba2b1d4b77dbf9499ec9f86e5a48ef6e0a27d5c83e9a09b268919
+size 92843394

lr5.5e-5_total_batch_size61440_baseline_seed1338/optimizer_08749.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:a5447e122fb4faf2794ff71e941bac8a5d3e7dffe0af880b62dfa07758237f4a
+size 179406214