Upload folder using huggingface_hub

lr5.5e-5_total_batch_size61440_baseline_seed1340/args.json

@@ -0,0 +1 @@
+{"hellaswag": true, "attention_kind": "selective", "log_dir": "wider_is_better_10/lr5.5e-5_total_batch_size61440_baseline_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_10", "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_1340", "n_embd": 256}

lr5.5e-5_total_batch_size61440_baseline_seed1340/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_seed1340/log2.txt

@@ -0,0 +1,267 @@
+max_steps: 8750
+0 val loss 11.2709
+0 val perplexity 78502.8516
+0 train 11.277068 (lr=1.1000e-07) (hash(x)=164406924)
+100 val loss 10.0006
+100 val perplexity 22040.1660
+100 train 9.961754 (lr=1.1110e-05) (hash(x)=144903932)
+200 val loss 9.2243
+200 val perplexity 10140.2500
+200 train 9.291431 (lr=2.2110e-05) (hash(x)=167734596)
+300 val loss 7.9598
+300 val perplexity 2863.6021
+300 train 7.966219 (lr=3.3110e-05) (hash(x)=149619098)
+400 val loss 7.5558
+400 val perplexity 1911.7671
+400 train 7.434297 (lr=4.4110e-05) (hash(x)=153710890)
+500 val loss 7.4021
+500 val perplexity 1639.4824
+500 train 7.499963 (lr=5.5000e-05) (hash(x)=145450636)
+600 val loss 7.2682
+600 val perplexity 1433.9047
+600 train 7.137627 (lr=5.4982e-05) (hash(x)=145249251)
+700 val loss 7.1397
+700 val perplexity 1261.0992
+700 train 7.269415 (lr=5.4928e-05) (hash(x)=150475545)
+800 val loss 7.0391
+800 val perplexity 1140.3866
+800 train 7.009733 (lr=5.4839e-05) (hash(x)=144483776)
+900 val loss 6.9339
+900 val perplexity 1026.4821
+900 train 7.044923 (lr=5.4713e-05) (hash(x)=157916369)
+1000 val loss 6.8542
+1000 val perplexity 947.8123
+1000 train 6.866351 (lr=5.4553e-05) (hash(x)=154856891)
+1100 val loss 6.7677
+1100 val perplexity 869.3527
+1100 train 6.461109 (lr=5.4357e-05) (hash(x)=136191502)
+1200 val loss 6.6756
+1200 val perplexity 792.8586
+1200 train 6.634073 (lr=5.4126e-05) (hash(x)=148021541)
+1300 val loss 6.6044
+1300 val perplexity 738.3351
+1300 train 6.532261 (lr=5.3860e-05) (hash(x)=146005217)
+1400 val loss 6.5427
+1400 val perplexity 694.1486
+1400 train 6.450285 (lr=5.3561e-05) (hash(x)=146019502)
+1500 val loss 6.4922
+1500 val perplexity 659.9768
+1500 train 6.425323 (lr=5.3227e-05) (hash(x)=150127281)
+1600 val loss 6.4346
+1600 val perplexity 623.0635
+1600 train 6.401301 (lr=5.2860e-05) (hash(x)=154120875)
+1700 val loss 6.4050
+1700 val perplexity 604.8397
+1700 train 6.618564 (lr=5.2461e-05) (hash(x)=155797680)
+1800 val loss 6.3490
+1800 val perplexity 571.9486
+1800 train 6.333819 (lr=5.2029e-05) (hash(x)=156809396)
+1900 val loss 6.3121
+1900 val perplexity 551.1878
+1900 train 6.189785 (lr=5.1565e-05) (hash(x)=144640294)
+2000 val loss 6.2812
+2000 val perplexity 534.4522
+2000 train 6.130698 (lr=5.1071e-05) (hash(x)=162831106)
+2100 val loss 6.2677
+2100 val perplexity 527.2629
+2100 train 6.257947 (lr=5.0547e-05) (hash(x)=158239484)
+2200 val loss 6.2255
+2200 val perplexity 505.4960
+2200 train 6.106677 (lr=4.9993e-05) (hash(x)=140504180)
+2300 val loss 6.1917
+2300 val perplexity 488.6992
+2300 train 6.181066 (lr=4.9410e-05) (hash(x)=142234024)
+2400 val loss 6.1692
+2400 val perplexity 477.8210
+2400 train 6.162827 (lr=4.8800e-05) (hash(x)=143091562)
+2500 val loss 6.1724
+2500 val perplexity 479.3504
+2500 train 6.071291 (lr=4.8162e-05) (hash(x)=149857456)
+2600 val loss 6.1294
+2600 val perplexity 459.1714
+2600 train 6.167506 (lr=4.7499e-05) (hash(x)=146191551)
+2700 val loss 6.1052
+2700 val perplexity 448.1638
+2700 train 6.025500 (lr=4.6811e-05) (hash(x)=145375752)
+2800 val loss 6.0891
+2800 val perplexity 441.0058
+2800 train 5.967486 (lr=4.6099e-05) (hash(x)=151568014)
+2900 val loss 6.0740
+2900 val perplexity 434.4293
+2900 train 5.897992 (lr=4.5364e-05) (hash(x)=149366597)
+3000 val loss 6.0426
+3000 val perplexity 420.9740
+3000 train 5.969383 (lr=4.4606e-05) (hash(x)=150464442)
+3100 val loss 6.0239
+3100 val perplexity 413.1702
+3100 train 6.142439 (lr=4.3828e-05) (hash(x)=182449036)
+3200 val loss 6.0103
+3200 val perplexity 407.6222
+3200 train 5.959648 (lr=4.3031e-05) (hash(x)=140141286)
+3300 val loss 5.9877
+3300 val perplexity 398.4962
+3300 train 5.955009 (lr=4.2215e-05) (hash(x)=148099414)
+3400 val loss 5.9704
+3400 val perplexity 391.6514
+3400 train 5.831068 (lr=4.1381e-05) (hash(x)=142633951)
+3500 val loss 5.9661
+3500 val perplexity 389.9897
+3500 train 5.822116 (lr=4.0532e-05) (hash(x)=148368965)
+3600 val loss 5.9460
+3600 val perplexity 382.2267
+3600 train 5.964525 (lr=3.9667e-05) (hash(x)=152372067)
+3700 val loss 5.9264
+3700 val perplexity 374.7864
+3700 train 6.004961 (lr=3.8789e-05) (hash(x)=168885609)
+3800 val loss 5.9162
+3800 val perplexity 370.9991
+3800 train 5.729640 (lr=3.7898e-05) (hash(x)=141633734)
+3900 val loss 5.8952
+3900 val perplexity 363.2991
+3900 train 5.785316 (lr=3.6996e-05) (hash(x)=153141007)
+4000 val loss 5.8782
+4000 val perplexity 357.1673
+4000 train 5.808302 (lr=3.6085e-05) (hash(x)=160577202)
+4100 val loss 5.8725
+4100 val perplexity 355.1257
+4100 train 5.940940 (lr=3.5165e-05) (hash(x)=153858169)
+4200 val loss 5.8506
+4200 val perplexity 347.4344
+4200 train 5.859368 (lr=3.4238e-05) (hash(x)=155889149)
+4300 val loss 5.8405
+4300 val perplexity 343.9608
+4300 train 5.649345 (lr=3.3305e-05) (hash(x)=152294662)
+4400 val loss 5.8269
+4400 val perplexity 339.2892
+4400 train 5.728466 (lr=3.2368e-05) (hash(x)=141804386)
+4500 val loss 5.8074
+4500 val perplexity 332.7563
+4500 train 5.704861 (lr=3.1428e-05) (hash(x)=151095242)
+4600 val loss 5.8144
+4600 val perplexity 335.0947
+4600 train 5.760637 (lr=3.0486e-05) (hash(x)=156414699)
+4700 val loss 5.7900
+4700 val perplexity 327.0099
+4700 train 5.808534 (lr=2.9543e-05) (hash(x)=161556686)
+4800 val loss 5.7760
+4800 val perplexity 322.4820
+4800 train 5.682988 (lr=2.8602e-05) (hash(x)=149000293)
+4900 val loss 5.7693
+4900 val perplexity 320.3153
+4900 train 5.552223 (lr=2.7663e-05) (hash(x)=154349989)
+5000 val loss 5.7542
+5000 val perplexity 315.5229
+5000 train 5.591222 (lr=2.6728e-05) (hash(x)=131475967)
+5100 val loss 5.7435
+5100 val perplexity 312.1526
+5100 train 5.646534 (lr=2.5798e-05) (hash(x)=149717902)
+5200 val loss 5.7352
+5200 val perplexity 309.5698
+5200 train 5.606053 (lr=2.4874e-05) (hash(x)=151407999)
+5300 val loss 5.7244
+5300 val perplexity 306.2366
+5300 train 5.810199 (lr=2.3958e-05) (hash(x)=168602728)
+5400 val loss 5.7136
+5400 val perplexity 302.9520
+5400 train 5.673536 (lr=2.3051e-05) (hash(x)=158344511)
+5500 val loss 5.7070
+5500 val perplexity 300.9659
+5500 train 5.661861 (lr=2.2155e-05) (hash(x)=148350057)
+5600 val loss 5.6953
+5600 val perplexity 297.4600
+5600 train 5.764478 (lr=2.1271e-05) (hash(x)=153847323)
+5700 val loss 5.6870
+5700 val perplexity 295.0194
+5700 train 5.626769 (lr=2.0399e-05) (hash(x)=156607405)
+5800 val loss 5.6821
+5800 val perplexity 293.5533
+5800 train 5.448245 (lr=1.9542e-05) (hash(x)=145115031)
+5900 val loss 5.6727
+5900 val perplexity 290.8273
+5900 train 5.555720 (lr=1.8700e-05) (hash(x)=141584622)
+6000 val loss 5.6635
+6000 val perplexity 288.1493
+6000 train 5.595728 (lr=1.7875e-05) (hash(x)=146613857)
+6100 val loss 5.6598
+6100 val perplexity 287.1042
+6100 train 5.582310 (lr=1.7068e-05) (hash(x)=144621768)
+6200 val loss 5.6544
+6200 val perplexity 285.5421
+6200 train 5.585410 (lr=1.6280e-05) (hash(x)=146521760)
+6300 val loss 5.6451
+6300 val perplexity 282.9159
+6300 train 5.472491 (lr=1.5512e-05) (hash(x)=161378136)
+6400 val loss 5.6410
+6400 val perplexity 281.7580
+6400 train 5.406197 (lr=1.4766e-05) (hash(x)=141624235)
+6500 val loss 5.6335
+6500 val perplexity 279.6512
+6500 train 5.553033 (lr=1.4042e-05) (hash(x)=151197095)
+6600 val loss 5.6336
+6600 val perplexity 279.6686
+6600 train 5.649274 (lr=1.3342e-05) (hash(x)=153269571)
+6700 val loss 5.6260
+6700 val perplexity 277.5412
+6700 train 5.582949 (lr=1.2666e-05) (hash(x)=146111181)
+6800 val loss 5.6161
+6800 val perplexity 274.8114
+6800 train 5.649564 (lr=1.2016e-05) (hash(x)=147269760)
+6900 val loss 5.6140
+6900 val perplexity 274.2294
+6900 train 5.692091 (lr=1.1392e-05) (hash(x)=152912762)
+7000 val loss 5.6105
+7000 val perplexity 273.2679
+7000 train 5.682277 (lr=1.0795e-05) (hash(x)=165412343)
+7100 val loss 5.6023
+7100 val perplexity 271.0528
+7100 train 5.675092 (lr=1.0227e-05) (hash(x)=162866028)
+7200 val loss 5.5975
+7200 val perplexity 269.7498
+7200 train 5.563646 (lr=9.6875e-06) (hash(x)=142998115)
+7300 val loss 5.5947
+7300 val perplexity 269.0092
+7300 train 5.360698 (lr=9.1780e-06) (hash(x)=145486999)
+7400 val loss 5.5895
+7400 val perplexity 267.5955
+7400 train 5.557508 (lr=8.6990e-06) (hash(x)=155325873)
+7500 val loss 5.5860
+7500 val perplexity 266.6570
+7500 train 5.549754 (lr=8.2513e-06) (hash(x)=145131256)
+7600 val loss 5.5895
+7600 val perplexity 267.6042
+7600 train 5.515909 (lr=7.8355e-06) (hash(x)=144008365)
+7700 val loss 5.5784
+7700 val perplexity 264.6376
+7700 train 5.683969 (lr=7.4522e-06) (hash(x)=148848532)
+7800 val loss 5.5767
+7800 val perplexity 264.1926
+7800 train 5.565325 (lr=7.1019e-06) (hash(x)=150391642)
+7900 val loss 5.5764
+7900 val perplexity 264.1082
+7900 train 5.332288 (lr=6.7852e-06) (hash(x)=152191414)
+8000 val loss 5.5714
+8000 val perplexity 262.7974
+8000 train 5.654792 (lr=6.5025e-06) (hash(x)=159755587)
+8100 val loss 5.5688
+8100 val perplexity 262.1242
+8100 train 5.594342 (lr=6.2543e-06) (hash(x)=156664468)
+8200 val loss 5.5677
+8200 val perplexity 261.8188
+8200 train 5.412155 (lr=6.0408e-06) (hash(x)=139457379)
+8300 val loss 5.5713
+8300 val perplexity 262.7819
+8300 train 5.336612 (lr=5.8625e-06) (hash(x)=145478564)
+8400 val loss 5.5625
+8400 val perplexity 260.4675
+8400 train 5.495065 (lr=5.7195e-06) (hash(x)=154982769)
+8500 val loss 5.5605
+8500 val perplexity 259.9456
+8500 train 5.530403 (lr=5.6121e-06) (hash(x)=145798118)
+8600 val loss 5.5599
+8600 val perplexity 259.7968
+8600 train 5.470276 (lr=5.5404e-06) (hash(x)=143231551)
+8700 val loss 5.5630
+8700 val perplexity 260.6076
+8700 train 5.452856 (lr=5.5045e-06) (hash(x)=154780112)
+8749 val loss 5.5559
+8749 val perplexity 258.7468

lr5.5e-5_total_batch_size61440_baseline_seed1340/model_08749.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:ec0656437d30ca642fc81b89fb1236e73f7ccd5c904e5786f0ca59a55eee3ba3
+size 92843394

lr5.5e-5_total_batch_size61440_baseline_seed1340/optimizer_08749.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:4583a6ebc690b53435d66b1b2d8db39c670af0104e07e19f488f019ac68c1e6d
+size 179406214