Upload folder using huggingface_hub

lr5.5e-5_total_batch_size61440_baseline_seed1339/args.json

@@ -0,0 +1 @@
+{"hellaswag": true, "attention_kind": "selective", "log_dir": "wider_is_better_10/lr5.5e-5_total_batch_size61440_baseline_seed1339", "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": 1339, "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_1339", "n_embd": 256}

lr5.5e-5_total_batch_size61440_baseline_seed1339/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_seed1339/log2.txt

@@ -0,0 +1,267 @@
+max_steps: 8750
+0 val loss 11.2808
+0 val perplexity 79287.6797
+0 train 11.289314 (lr=1.1000e-07) (hash(x)=150724848)
+100 val loss 9.9941
+100 val perplexity 21895.8184
+100 train 10.007823 (lr=1.1110e-05) (hash(x)=149910534)
+200 val loss 9.1306
+200 val perplexity 9233.5605
+200 train 9.158016 (lr=2.2110e-05) (hash(x)=148123706)
+300 val loss 7.9363
+300 val perplexity 2796.9626
+300 train 7.795696 (lr=3.3110e-05) (hash(x)=146678221)
+400 val loss 7.5281
+400 val perplexity 1859.4856
+400 train 7.413250 (lr=4.4110e-05) (hash(x)=151700982)
+500 val loss 7.3819
+500 val perplexity 1606.6868
+500 train 7.464406 (lr=5.5000e-05) (hash(x)=156182087)
+600 val loss 7.2607
+600 val perplexity 1423.2772
+600 train 7.234007 (lr=5.4982e-05) (hash(x)=149318660)
+700 val loss 7.1635
+700 val perplexity 1291.4326
+700 train 7.201457 (lr=5.4928e-05) (hash(x)=150482428)
+800 val loss 7.0255
+800 val perplexity 1124.9342
+800 train 6.897016 (lr=5.4839e-05) (hash(x)=143268605)
+900 val loss 6.9223
+900 val perplexity 1014.6030
+900 train 6.882633 (lr=5.4713e-05) (hash(x)=152322423)
+1000 val loss 6.8366
+1000 val perplexity 931.3380
+1000 train 6.703583 (lr=5.4553e-05) (hash(x)=147904298)
+1100 val loss 6.7481
+1100 val perplexity 852.4358
+1100 train 6.947027 (lr=5.4357e-05) (hash(x)=154343147)
+1200 val loss 6.6486
+1200 val perplexity 771.6978
+1200 train 6.638902 (lr=5.4126e-05) (hash(x)=141843115)
+1300 val loss 6.5783
+1300 val perplexity 719.2836
+1300 train 6.470258 (lr=5.3860e-05) (hash(x)=145279030)
+1400 val loss 6.5243
+1400 val perplexity 681.4932
+1400 train 6.466841 (lr=5.3561e-05) (hash(x)=152507639)
+1500 val loss 6.4653
+1500 val perplexity 642.4648
+1500 train 6.499565 (lr=5.3227e-05) (hash(x)=148473774)
+1600 val loss 6.4033
+1600 val perplexity 603.8219
+1600 train 6.449642 (lr=5.2860e-05) (hash(x)=151117002)
+1700 val loss 6.3545
+1700 val perplexity 575.0466
+1700 train 6.305005 (lr=5.2461e-05) (hash(x)=138011335)
+1800 val loss 6.3136
+1800 val perplexity 552.0479
+1800 train 6.442955 (lr=5.2029e-05) (hash(x)=171180926)
+1900 val loss 6.2734
+1900 val perplexity 530.2992
+1900 train 6.210119 (lr=5.1565e-05) (hash(x)=141769419)
+2000 val loss 6.2357
+2000 val perplexity 510.6739
+2000 train 6.187378 (lr=5.1071e-05) (hash(x)=151963443)
+2100 val loss 6.2164
+2100 val perplexity 500.8954
+2100 train 6.275733 (lr=5.0547e-05) (hash(x)=162947470)
+2200 val loss 6.1707
+2200 val perplexity 478.4984
+2200 train 6.383195 (lr=4.9993e-05) (hash(x)=154954810)
+2300 val loss 6.1456
+2300 val perplexity 466.6610
+2300 train 6.122678 (lr=4.9410e-05) (hash(x)=151878111)
+2400 val loss 6.1227
+2400 val perplexity 456.1168
+2400 train 6.018513 (lr=4.8800e-05) (hash(x)=158661057)
+2500 val loss 6.0900
+2500 val perplexity 441.4375
+2500 train 6.111250 (lr=4.8162e-05) (hash(x)=150925584)
+2600 val loss 6.0730
+2600 val perplexity 433.9931
+2600 train 5.985307 (lr=4.7499e-05) (hash(x)=144515755)
+2700 val loss 6.0517
+2700 val perplexity 424.8535
+2700 train 6.044211 (lr=4.6811e-05) (hash(x)=153109144)
+2800 val loss 6.0264
+2800 val perplexity 414.2404
+2800 train 5.932133 (lr=4.6099e-05) (hash(x)=151152897)
+2900 val loss 6.0053
+2900 val perplexity 405.5740
+2900 train 5.950101 (lr=4.5364e-05) (hash(x)=145800210)
+3000 val loss 5.9938
+3000 val perplexity 400.9246
+3000 train 5.798833 (lr=4.4606e-05) (hash(x)=141997485)
+3100 val loss 5.9705
+3100 val perplexity 391.7181
+3100 train 5.894557 (lr=4.3828e-05) (hash(x)=154049740)
+3200 val loss 5.9493
+3200 val perplexity 383.4979
+3200 train 5.883973 (lr=4.3031e-05) (hash(x)=150471842)
+3300 val loss 5.9377
+3300 val perplexity 379.0521
+3300 train 5.883282 (lr=4.2215e-05) (hash(x)=149048126)
+3400 val loss 5.9171
+3400 val perplexity 371.3324
+3400 train 6.042283 (lr=4.1381e-05) (hash(x)=161261339)
+3500 val loss 5.8992
+3500 val perplexity 364.7562
+3500 train 5.860901 (lr=4.0532e-05) (hash(x)=157495564)
+3600 val loss 5.8892
+3600 val perplexity 361.1294
+3600 train 5.801177 (lr=3.9667e-05) (hash(x)=144352932)
+3700 val loss 5.8723
+3700 val perplexity 355.0539
+3700 train 5.840976 (lr=3.8789e-05) (hash(x)=149389012)
+3800 val loss 5.8513
+3800 val perplexity 347.6731
+3800 train 5.788841 (lr=3.7898e-05) (hash(x)=146607620)
+3900 val loss 5.8410
+3900 val perplexity 344.1376
+3900 train 5.759561 (lr=3.6996e-05) (hash(x)=143754617)
+4000 val loss 5.8260
+4000 val perplexity 339.0128
+4000 train 5.799609 (lr=3.6085e-05) (hash(x)=156930722)
+4100 val loss 5.8111
+4100 val perplexity 334.0016
+4100 train 5.644769 (lr=3.5165e-05) (hash(x)=147500519)
+4200 val loss 5.8006
+4200 val perplexity 330.4939
+4200 train 5.644540 (lr=3.4238e-05) (hash(x)=143232237)
+4300 val loss 5.7846
+4300 val perplexity 325.2483
+4300 train 5.690441 (lr=3.3305e-05) (hash(x)=146811670)
+4400 val loss 5.7702
+4400 val perplexity 320.5955
+4400 train 5.800915 (lr=3.2368e-05) (hash(x)=158418746)
+4500 val loss 5.7638
+4500 val perplexity 318.5586
+4500 train 5.724743 (lr=3.1428e-05) (hash(x)=156695778)
+4600 val loss 5.7496
+4600 val perplexity 314.0630
+4600 train 5.763687 (lr=3.0486e-05) (hash(x)=147791497)
+4700 val loss 5.7353
+4700 val perplexity 309.6109
+4700 train 5.627360 (lr=2.9543e-05) (hash(x)=155533088)
+4800 val loss 5.7291
+4800 val perplexity 307.7023
+4800 train 5.584979 (lr=2.8602e-05) (hash(x)=138350044)
+4900 val loss 5.7122
+4900 val perplexity 302.5468
+4900 train 5.707434 (lr=2.7663e-05) (hash(x)=143735284)
+5000 val loss 5.7022
+5000 val perplexity 299.5402
+5000 train 5.659809 (lr=2.6728e-05) (hash(x)=154976463)
+5100 val loss 5.6965
+5100 val perplexity 297.8150
+5100 train 5.582112 (lr=2.5798e-05) (hash(x)=149894982)
+5200 val loss 5.6885
+5200 val perplexity 295.4523
+5200 train 5.466450 (lr=2.4874e-05) (hash(x)=159326689)
+5300 val loss 5.6782
+5300 val perplexity 292.4131
+5300 train 5.688088 (lr=2.3958e-05) (hash(x)=159484800)
+5400 val loss 5.6637
+5400 val perplexity 288.2175
+5400 train 5.570019 (lr=2.3051e-05) (hash(x)=140385615)
+5500 val loss 5.6577
+5500 val perplexity 286.4755
+5500 train 5.490870 (lr=2.2155e-05) (hash(x)=148498335)
+5600 val loss 5.6576
+5600 val perplexity 286.4469
+5600 train 5.396508 (lr=2.1271e-05) (hash(x)=151907614)
+5700 val loss 5.6466
+5700 val perplexity 283.3252
+5700 train 5.619106 (lr=2.0399e-05) (hash(x)=155192267)
+5800 val loss 5.6343
+5800 val perplexity 279.8629
+5800 train 5.682648 (lr=1.9542e-05) (hash(x)=153132158)
+5900 val loss 5.6317
+5900 val perplexity 279.1493
+5900 train 5.603333 (lr=1.8700e-05) (hash(x)=161446764)
+6000 val loss 5.6258
+6000 val perplexity 277.5057
+6000 train 5.420910 (lr=1.7875e-05) (hash(x)=151512446)
+6100 val loss 5.6148
+6100 val perplexity 274.4505
+6100 train 5.718103 (lr=1.7068e-05) (hash(x)=188094053)
+6200 val loss 5.6050
+6200 val perplexity 271.7859
+6200 train 5.485103 (lr=1.6280e-05) (hash(x)=149389789)
+6300 val loss 5.6008
+6300 val perplexity 270.6541
+6300 train 5.415108 (lr=1.5512e-05) (hash(x)=138212820)
+6400 val loss 5.5983
+6400 val perplexity 269.9794
+6400 train 5.386089 (lr=1.4766e-05) (hash(x)=146535423)
+6500 val loss 5.5897
+6500 val perplexity 267.6628
+6500 train 5.460159 (lr=1.4042e-05) (hash(x)=145950843)
+6600 val loss 5.5842
+6600 val perplexity 266.1842
+6600 train 5.490510 (lr=1.3342e-05) (hash(x)=141162902)
+6700 val loss 5.5815
+6700 val perplexity 265.4708
+6700 train 5.496548 (lr=1.2666e-05) (hash(x)=153018737)
+6800 val loss 5.5745
+6800 val perplexity 263.6306
+6800 train 5.492538 (lr=1.2016e-05) (hash(x)=155640155)
+6900 val loss 5.5691
+6900 val perplexity 262.2026
+6900 train 5.539137 (lr=1.1392e-05) (hash(x)=153722115)
+7000 val loss 5.5624
+7000 val perplexity 260.4574
+7000 train 5.470424 (lr=1.0795e-05) (hash(x)=146953450)
+7100 val loss 5.5634
+7100 val perplexity 260.7125
+7100 train 5.408412 (lr=1.0227e-05) (hash(x)=137663885)
+7200 val loss 5.5542
+7200 val perplexity 258.3073
+7200 train 5.629156 (lr=9.6875e-06) (hash(x)=146172950)
+7300 val loss 5.5487
+7300 val perplexity 256.9048
+7300 train 5.437962 (lr=9.1780e-06) (hash(x)=150018163)
+7400 val loss 5.5477
+7400 val perplexity 256.6565
+7400 train 5.560954 (lr=8.6990e-06) (hash(x)=145351166)
+7500 val loss 5.5472
+7500 val perplexity 256.5213
+7500 train 5.325307 (lr=8.2513e-06) (hash(x)=145292116)
+7600 val loss 5.5393
+7600 val perplexity 254.4988
+7600 train 5.455224 (lr=7.8355e-06) (hash(x)=150235132)
+7700 val loss 5.5364
+7700 val perplexity 253.7741
+7700 train 5.432760 (lr=7.4522e-06) (hash(x)=154543455)
+7800 val loss 5.5353
+7800 val perplexity 253.4868
+7800 train 5.438547 (lr=7.1019e-06) (hash(x)=142456852)
+7900 val loss 5.5341
+7900 val perplexity 253.1841
+7900 train 5.313053 (lr=6.7852e-06) (hash(x)=147363479)
+8000 val loss 5.5272
+8000 val perplexity 251.4442
+8000 train 5.559174 (lr=6.5025e-06) (hash(x)=156122973)
+8100 val loss 5.5247
+8100 val perplexity 250.8040
+8100 train 5.572883 (lr=6.2543e-06) (hash(x)=156153179)
+8200 val loss 5.5227
+8200 val perplexity 250.3166
+8200 train 5.578476 (lr=6.0408e-06) (hash(x)=146430698)
+8300 val loss 5.5216
+8300 val perplexity 250.0329
+8300 train 5.396613 (lr=5.8625e-06) (hash(x)=143507257)
+8400 val loss 5.5160
+8400 val perplexity 248.6355
+8400 train 5.515770 (lr=5.7195e-06) (hash(x)=166272643)
+8500 val loss 5.5145
+8500 val perplexity 248.2608
+8500 train 5.422902 (lr=5.6121e-06) (hash(x)=143887848)
+8600 val loss 5.5143
+8600 val perplexity 248.2096
+8600 train 5.548209 (lr=5.5404e-06) (hash(x)=156900341)
+8700 val loss 5.5094
+8700 val perplexity 247.0037
+8700 train 5.728339 (lr=5.5045e-06) (hash(x)=146417632)
+8749 val loss 5.5085
+8749 val perplexity 246.7827

lr5.5e-5_total_batch_size61440_baseline_seed1339/model_08749.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:e492987402babee8eedf5f8743cfdd127ab2fe0d387efda9c8a8d4f0cf3263e2
+size 92843394

lr5.5e-5_total_batch_size61440_baseline_seed1339/optimizer_08749.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:ff229e8129232dadc1bbd21707876cb8544243667c1258e2a46cd5e9b19c547d
+size 179406214