Upload folder using huggingface_hub

lr6e-5_total_batch_size61440_baseline_seed1338/args.json

@@ -0,0 +1 @@
+{"hellaswag": true, "attention_kind": "selective", "log_dir": "wider_is_better_10/lr6e-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": 6e-05, "decay_lr": true, "readout_zero_init": false, "query_zero_init": false, "l1_loss": false, "debugpy": false, "key": "6e-5_61440_1338", "n_embd": 256}

lr6e-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

lr6e-5_total_batch_size61440_baseline_seed1338/log2.txt

@@ -0,0 +1,267 @@
+max_steps: 8750
+0 val loss 11.2647
+0 val perplexity 78014.9766
+0 train 11.267744 (lr=1.2000e-07) (hash(x)=150327452)
+100 val loss 9.9336
+100 val perplexity 20610.7285
+100 train 9.985141 (lr=1.2120e-05) (hash(x)=166780046)
+200 val loss 9.0294
+200 val perplexity 8344.8027
+200 train 8.960388 (lr=2.4120e-05) (hash(x)=155040610)
+300 val loss 7.8849
+300 val perplexity 2656.8672
+300 train 7.875908 (lr=3.6120e-05) (hash(x)=155504036)
+400 val loss 7.5169
+400 val perplexity 1838.8458
+400 train 7.375343 (lr=4.8120e-05) (hash(x)=143823248)
+500 val loss 7.3942
+500 val perplexity 1626.4989
+500 train 7.352885 (lr=6.0000e-05) (hash(x)=143734685)
+600 val loss 7.3002
+600 val perplexity 1480.6045
+600 train 7.071487 (lr=5.9980e-05) (hash(x)=150678249)
+700 val loss 7.1863
+700 val perplexity 1321.1469
+700 train 7.308282 (lr=5.9922e-05) (hash(x)=175802021)
+800 val loss 7.0785
+800 val perplexity 1186.1412
+800 train 7.097569 (lr=5.9824e-05) (hash(x)=158681215)
+900 val loss 6.9944
+900 val perplexity 1090.4926
+900 train 6.885390 (lr=5.9687e-05) (hash(x)=146108145)
+1000 val loss 6.9300
+1000 val perplexity 1022.4719
+1000 train 6.768590 (lr=5.9512e-05) (hash(x)=154996086)
+1100 val loss 6.8340
+1100 val perplexity 928.9164
+1100 train 6.815024 (lr=5.9298e-05) (hash(x)=153885445)
+1200 val loss 6.7258
+1200 val perplexity 833.6080
+1200 train 6.690398 (lr=5.9046e-05) (hash(x)=142353087)
+1300 val loss 6.6548
+1300 val perplexity 776.5233
+1300 train 6.562291 (lr=5.8757e-05) (hash(x)=150750353)
+1400 val loss 6.5954
+1400 val perplexity 731.7207
+1400 train 6.402224 (lr=5.8430e-05) (hash(x)=152767913)
+1500 val loss 6.5287
+1500 val perplexity 684.5015
+1500 train 6.578063 (lr=5.8066e-05) (hash(x)=151562048)
+1600 val loss 6.4561
+1600 val perplexity 636.5933
+1600 train 6.462641 (lr=5.7666e-05) (hash(x)=166486165)
+1700 val loss 6.4188
+1700 val perplexity 613.2521
+1700 train 6.066719 (lr=5.7230e-05) (hash(x)=130835396)
+1800 val loss 6.3702
+1800 val perplexity 584.1725
+1800 train 6.418849 (lr=5.6759e-05) (hash(x)=158851816)
+1900 val loss 6.3417
+1900 val perplexity 567.7599
+1900 train 6.367229 (lr=5.6253e-05) (hash(x)=153313879)
+2000 val loss 6.2922
+2000 val perplexity 540.3171
+2000 train 6.221293 (lr=5.5714e-05) (hash(x)=158245023)
+2100 val loss 6.2494
+2100 val perplexity 517.6829
+2100 train 6.069878 (lr=5.5142e-05) (hash(x)=157204896)
+2200 val loss 6.2325
+2200 val perplexity 509.0488
+2200 train 6.072013 (lr=5.4537e-05) (hash(x)=137541932)
+2300 val loss 6.2104
+2300 val perplexity 497.8807
+2300 train 6.167417 (lr=5.3902e-05) (hash(x)=150149692)
+2400 val loss 6.1706
+2400 val perplexity 478.4685
+2400 train 6.160289 (lr=5.3236e-05) (hash(x)=151730720)
+2500 val loss 6.1445
+2500 val perplexity 466.1251
+2500 train 5.940173 (lr=5.2541e-05) (hash(x)=143406752)
+2600 val loss 6.1269
+2600 val perplexity 457.9996
+2600 train 5.958959 (lr=5.1817e-05) (hash(x)=157272496)
+2700 val loss 6.1058
+2700 val perplexity 448.4577
+2700 train 6.074640 (lr=5.1067e-05) (hash(x)=155342327)
+2800 val loss 6.0700
+2800 val perplexity 432.6698
+2800 train 5.952314 (lr=5.0290e-05) (hash(x)=140626679)
+2900 val loss 6.0527
+2900 val perplexity 425.2800
+2900 train 5.913211 (lr=4.9487e-05) (hash(x)=144953350)
+3000 val loss 6.0327
+3000 val perplexity 416.8304
+3000 train 5.921506 (lr=4.8662e-05) (hash(x)=172449837)
+3100 val loss 6.0246
+3100 val perplexity 413.4940
+3100 train 5.808658 (lr=4.7813e-05) (hash(x)=141710086)
+3200 val loss 5.9924
+3200 val perplexity 400.3599
+3200 train 5.923226 (lr=4.6943e-05) (hash(x)=151299772)
+3300 val loss 5.9797
+3300 val perplexity 395.3212
+3300 train 5.854528 (lr=4.6052e-05) (hash(x)=146473110)
+3400 val loss 5.9551
+3400 val perplexity 385.7184
+3400 train 6.046208 (lr=4.5143e-05) (hash(x)=153954157)
+3500 val loss 5.9403
+3500 val perplexity 380.0634
+3500 train 5.882200 (lr=4.4216e-05) (hash(x)=153717336)
+3600 val loss 5.9216
+3600 val perplexity 373.0199
+3600 train 5.677235 (lr=4.3273e-05) (hash(x)=144965161)
+3700 val loss 5.9148
+3700 val perplexity 370.4974
+3700 train 5.697598 (lr=4.2315e-05) (hash(x)=125969741)
+3800 val loss 5.8978
+3800 val perplexity 364.2434
+3800 train 5.762879 (lr=4.1343e-05) (hash(x)=155070487)
+3900 val loss 5.8740
+3900 val perplexity 355.6602
+3900 train 5.745014 (lr=4.0360e-05) (hash(x)=149444644)
+4000 val loss 5.8619
+4000 val perplexity 351.4026
+4000 train 5.713954 (lr=3.9365e-05) (hash(x)=151663033)
+4100 val loss 5.8587
+4100 val perplexity 350.2661
+4100 train 5.818701 (lr=3.8362e-05) (hash(x)=143688282)
+4200 val loss 5.8332
+4200 val perplexity 341.4621
+4200 train 5.833591 (lr=3.7351e-05) (hash(x)=163361651)
+4300 val loss 5.8158
+4300 val perplexity 335.5430
+4300 train 5.848325 (lr=3.6333e-05) (hash(x)=153619361)
+4400 val loss 5.8089
+4400 val perplexity 333.2647
+4400 train 6.146502 (lr=3.5311e-05) (hash(x)=168527064)
+4500 val loss 5.7962
+4500 val perplexity 329.0342
+4500 train 5.626137 (lr=3.4285e-05) (hash(x)=125588037)
+4600 val loss 5.7776
+4600 val perplexity 322.9897
+4600 train 5.734975 (lr=3.3257e-05) (hash(x)=143710941)
+4700 val loss 5.7627
+4700 val perplexity 318.2065
+4700 train 5.605650 (lr=3.2229e-05) (hash(x)=150952742)
+4800 val loss 5.7550
+4800 val perplexity 315.7542
+4800 train 5.635776 (lr=3.1202e-05) (hash(x)=145323659)
+4900 val loss 5.7465
+4900 val perplexity 313.1005
+4900 train 5.764939 (lr=3.0178e-05) (hash(x)=153151397)
+5000 val loss 5.7356
+5000 val perplexity 309.7007
+5000 train 5.642522 (lr=2.9157e-05) (hash(x)=143182059)
+5100 val loss 5.7215
+5100 val perplexity 305.3588
+5100 train 5.749349 (lr=2.8143e-05) (hash(x)=170083586)
+5200 val loss 5.7102
+5200 val perplexity 301.9301
+5200 train 5.585383 (lr=2.7135e-05) (hash(x)=149363919)
+5300 val loss 5.7075
+5300 val perplexity 301.1300
+5300 train 5.586699 (lr=2.6136e-05) (hash(x)=152033784)
+5400 val loss 5.6932
+5400 val perplexity 296.8299
+5400 train 5.808794 (lr=2.5147e-05) (hash(x)=154614289)
+5500 val loss 5.6842
+5500 val perplexity 294.1708
+5500 train 5.785327 (lr=2.4169e-05) (hash(x)=157745174)
+5600 val loss 5.6782
+5600 val perplexity 292.4264
+5600 train 5.576692 (lr=2.3204e-05) (hash(x)=147693222)
+5700 val loss 5.6670
+5700 val perplexity 289.1793
+5700 train 5.597318 (lr=2.2253e-05) (hash(x)=149784627)
+5800 val loss 5.6628
+5800 val perplexity 287.9644
+5800 train 5.590798 (lr=2.1318e-05) (hash(x)=158620729)
+5900 val loss 5.6525
+5900 val perplexity 284.9917
+5900 train 5.605389 (lr=2.0400e-05) (hash(x)=159763910)
+6000 val loss 5.6399
+6000 val perplexity 281.4313
+6000 train 5.578279 (lr=1.9500e-05) (hash(x)=147640561)
+6100 val loss 5.6363
+6100 val perplexity 280.4095
+6100 train 5.620928 (lr=1.8620e-05) (hash(x)=156613394)
+6200 val loss 5.6307
+6200 val perplexity 278.8485
+6200 train 5.551859 (lr=1.7760e-05) (hash(x)=186221290)
+6300 val loss 5.6201
+6300 val perplexity 275.9275
+6300 train 5.495809 (lr=1.6923e-05) (hash(x)=152081419)
+6400 val loss 5.6201
+6400 val perplexity 275.9156
+6400 train 5.572321 (lr=1.6108e-05) (hash(x)=154808349)
+6500 val loss 5.6129
+6500 val perplexity 273.9359
+6500 train 5.576208 (lr=1.5319e-05) (hash(x)=159437208)
+6600 val loss 5.6097
+6600 val perplexity 273.0613
+6600 train 5.423267 (lr=1.4555e-05) (hash(x)=157933074)
+6700 val loss 5.6014
+6700 val perplexity 270.8051
+6700 train 5.575732 (lr=1.3817e-05) (hash(x)=161560240)
+6800 val loss 5.5984
+6800 val perplexity 270.0009
+6800 train 5.584086 (lr=1.3108e-05) (hash(x)=155424292)
+6900 val loss 5.5966
+6900 val perplexity 269.5205
+6900 train 5.592770 (lr=1.2427e-05) (hash(x)=148561470)
+7000 val loss 5.5890
+7000 val perplexity 267.4669
+7000 train 5.489427 (lr=1.1777e-05) (hash(x)=141527450)
+7100 val loss 5.5831
+7100 val perplexity 265.9005
+7100 train 5.542429 (lr=1.1157e-05) (hash(x)=151066339)
+7200 val loss 5.5841
+7200 val perplexity 266.1620
+7200 train 5.375564 (lr=1.0568e-05) (hash(x)=155231264)
+7300 val loss 5.5749
+7300 val perplexity 263.7362
+7300 train 5.603450 (lr=1.0012e-05) (hash(x)=150281149)
+7400 val loss 5.5720
+7400 val perplexity 262.9702
+7400 train 5.461077 (lr=9.4899e-06) (hash(x)=148421717)
+7500 val loss 5.5728
+7500 val perplexity 263.1751
+7500 train 5.334098 (lr=9.0014e-06) (hash(x)=146921118)
+7600 val loss 5.5641
+7600 val perplexity 260.8827
+7600 train 5.609095 (lr=8.5478e-06) (hash(x)=150660048)
+7700 val loss 5.5636
+7700 val perplexity 260.7608
+7700 train 5.437018 (lr=8.1297e-06) (hash(x)=148059852)
+7800 val loss 5.5628
+7800 val perplexity 260.5605
+7800 train 5.247862 (lr=7.7476e-06) (hash(x)=148331002)
+7900 val loss 5.5566
+7900 val perplexity 258.9312
+7900 train 5.493481 (lr=7.4021e-06) (hash(x)=164923883)
+8000 val loss 5.5553
+8000 val perplexity 258.6062
+8000 train 5.435534 (lr=7.0937e-06) (hash(x)=143545384)
+8100 val loss 5.5540
+8100 val perplexity 258.2666
+8100 train 5.367274 (lr=6.8229e-06) (hash(x)=160686959)
+8200 val loss 5.5510
+8200 val perplexity 257.4983
+8200 train 5.499079 (lr=6.5900e-06) (hash(x)=156501889)
+8300 val loss 5.5496
+8300 val perplexity 257.1336
+8300 train 5.395368 (lr=6.3954e-06) (hash(x)=142716875)
+8400 val loss 5.5444
+8400 val perplexity 255.8118
+8400 train 5.468668 (lr=6.2395e-06) (hash(x)=154436684)
+8500 val loss 5.5432
+8500 val perplexity 255.4954
+8500 train 5.627841 (lr=6.1223e-06) (hash(x)=147965839)
+8600 val loss 5.5435
+8600 val perplexity 255.5600
+8600 train 5.331046 (lr=6.0440e-06) (hash(x)=145228097)
+8700 val loss 5.5380
+8700 val perplexity 254.1711
+8700 train 5.729925 (lr=6.0049e-06) (hash(x)=152910357)
+8749 val loss 5.5380
+8749 val perplexity 254.1640

lr6e-5_total_batch_size61440_baseline_seed1338/model_08749.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:a22cb36b4300267e1f0888b6265676c4446ae2849f6ea80468b3d61f02907b44
+size 92843394

lr6e-5_total_batch_size61440_baseline_seed1338/optimizer_08749.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:2b140ca4c4682d48238d6e31c4f162ad18a5b51918cac8998ddefae669a39995
+size 179406214