Upload folder using huggingface_hub

lr6.5e-5_total_batch_size61440_baseline_seed1341/args.json

@@ -0,0 +1 @@
+{"hellaswag": true, "attention_kind": "selective", "log_dir": "wider_is_better_10/lr6.5e-5_total_batch_size61440_baseline_seed1341", "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": 1341, "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": 6.5e-05, "decay_lr": true, "readout_zero_init": false, "query_zero_init": false, "l1_loss": false, "debugpy": false, "key": "6.5e-5_61440_1341", "n_embd": 256}

lr6.5e-5_total_batch_size61440_baseline_seed1341/dataloader_08749.pt

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

lr6.5e-5_total_batch_size61440_baseline_seed1341/log2.txt

@@ -0,0 +1,267 @@
+max_steps: 8750
+0 val loss 11.3095
+0 val perplexity 81594.3281
+0 train 11.313001 (lr=1.3000e-07) (hash(x)=145079536)
+100 val loss 10.0001
+100 val perplexity 22027.9355
+100 train 9.995198 (lr=1.3130e-05) (hash(x)=155800595)
+200 val loss 8.9902
+200 val perplexity 8023.8325
+200 train 8.933511 (lr=2.6130e-05) (hash(x)=145606733)
+300 val loss 7.7076
+300 val perplexity 2225.1543
+300 train 7.657198 (lr=3.9130e-05) (hash(x)=150367139)
+400 val loss 7.4630
+400 val perplexity 1742.3130
+400 train 7.785594 (lr=5.2130e-05) (hash(x)=155747374)
+500 val loss 7.3237
+500 val perplexity 1515.8309
+500 train 7.205816 (lr=6.5000e-05) (hash(x)=140604760)
+600 val loss 7.1997
+600 val perplexity 1338.9630
+600 train 7.123516 (lr=6.4979e-05) (hash(x)=148404734)
+700 val loss 7.0857
+700 val perplexity 1194.7924
+700 train 7.033662 (lr=6.4915e-05) (hash(x)=148115934)
+800 val loss 6.9645
+800 val perplexity 1058.4285
+800 train 6.828290 (lr=6.4809e-05) (hash(x)=137464699)
+900 val loss 6.8784
+900 val perplexity 971.0778
+900 train 6.691501 (lr=6.4661e-05) (hash(x)=143886042)
+1000 val loss 6.8082
+1000 val perplexity 905.2700
+1000 train 6.816955 (lr=6.4471e-05) (hash(x)=163799796)
+1100 val loss 6.7196
+1100 val perplexity 828.5250
+1100 train 6.690909 (lr=6.4240e-05) (hash(x)=144592844)
+1200 val loss 6.6617
+1200 val perplexity 781.8486
+1200 train 6.994630 (lr=6.3967e-05) (hash(x)=204706354)
+1300 val loss 6.6133
+1300 val perplexity 744.9322
+1300 train 6.556497 (lr=6.3653e-05) (hash(x)=150862210)
+1400 val loss 6.5609
+1400 val perplexity 706.9273
+1400 train 6.562226 (lr=6.3299e-05) (hash(x)=147766811)
+1500 val loss 6.5088
+1500 val perplexity 670.9980
+1500 train 6.338181 (lr=6.2905e-05) (hash(x)=135925327)
+1600 val loss 6.4923
+1600 val perplexity 660.0234
+1600 train 6.468217 (lr=6.2471e-05) (hash(x)=160440642)
+1700 val loss 6.4376
+1700 val perplexity 624.8875
+1700 train 6.557316 (lr=6.1999e-05) (hash(x)=151184106)
+1800 val loss 6.4049
+1800 val perplexity 604.8204
+1800 train 6.300439 (lr=6.1489e-05) (hash(x)=148100580)
+1900 val loss 6.3855
+1900 val perplexity 593.2050
+1900 train 6.220203 (lr=6.0941e-05) (hash(x)=149434659)
+2000 val loss 6.3328
+2000 val perplexity 562.7365
+2000 train 6.239842 (lr=6.0357e-05) (hash(x)=152285486)
+2100 val loss 6.3034
+2100 val perplexity 546.4256
+2100 train 6.049372 (lr=5.9737e-05) (hash(x)=144294295)
+2200 val loss 6.2816
+2200 val perplexity 534.6292
+2200 train 6.292001 (lr=5.9082e-05) (hash(x)=175030215)
+2300 val loss 6.2663
+2300 val perplexity 526.5107
+2300 train 6.082758 (lr=5.8394e-05) (hash(x)=150831428)
+2400 val loss 6.2304
+2400 val perplexity 507.9553
+2400 train 6.401173 (lr=5.7672e-05) (hash(x)=140808297)
+2500 val loss 6.2126
+2500 val perplexity 498.9756
+2500 train 6.188481 (lr=5.6919e-05) (hash(x)=153160275)
+2600 val loss 6.1917
+2600 val perplexity 488.6997
+2600 train 5.981144 (lr=5.6135e-05) (hash(x)=133990623)
+2700 val loss 6.1773
+2700 val perplexity 481.7082
+2700 train 6.149596 (lr=5.5322e-05) (hash(x)=142860944)
+2800 val loss 6.1537
+2800 val perplexity 470.4557
+2800 train 6.161194 (lr=5.4480e-05) (hash(x)=137959511)
+2900 val loss 6.1184
+2900 val perplexity 454.1553
+2900 train 6.069050 (lr=5.3611e-05) (hash(x)=147009873)
+3000 val loss 6.1078
+3000 val perplexity 449.3561
+3000 train 5.956011 (lr=5.2717e-05) (hash(x)=158264841)
+3100 val loss 6.1045
+3100 val perplexity 447.8760
+3100 train 5.935299 (lr=5.1797e-05) (hash(x)=139232251)
+3200 val loss 6.0727
+3200 val perplexity 433.8486
+3200 train 6.115242 (lr=5.0855e-05) (hash(x)=153436104)
+3300 val loss 6.0525
+3300 val perplexity 425.1632
+3300 train 5.884038 (lr=4.9890e-05) (hash(x)=149681831)
+3400 val loss 6.0517
+3400 val perplexity 424.8363
+3400 train 5.925019 (lr=4.8905e-05) (hash(x)=168425516)
+3500 val loss 6.0297
+3500 val perplexity 415.5702
+3500 train 6.052419 (lr=4.7901e-05) (hash(x)=163104338)
+3600 val loss 6.0140
+3600 val perplexity 409.1073
+3600 train 5.982936 (lr=4.6879e-05) (hash(x)=165109772)
+3700 val loss 6.0035
+3700 val perplexity 404.8543
+3700 train 5.743422 (lr=4.5841e-05) (hash(x)=153420306)
+3800 val loss 5.9841
+3800 val perplexity 397.0655
+3800 train 6.001788 (lr=4.4789e-05) (hash(x)=160168863)
+3900 val loss 5.9675
+3900 val perplexity 390.5187
+3900 train 5.878839 (lr=4.3723e-05) (hash(x)=153906073)
+4000 val loss 5.9534
+4000 val perplexity 385.0477
+4000 train 5.898636 (lr=4.2646e-05) (hash(x)=151055067)
+4100 val loss 5.9383
+4100 val perplexity 379.2809
+4100 train 5.920074 (lr=4.1559e-05) (hash(x)=149629830)
+4200 val loss 5.9215
+4200 val perplexity 372.9710
+4200 train 5.803367 (lr=4.0463e-05) (hash(x)=143101381)
+4300 val loss 5.9091
+4300 val perplexity 368.3901
+4300 train 5.817513 (lr=3.9361e-05) (hash(x)=149712044)
+4400 val loss 5.8921
+4400 val perplexity 362.1600
+4400 train 5.819164 (lr=3.8253e-05) (hash(x)=153446449)
+4500 val loss 5.8803
+4500 val perplexity 357.9161
+4500 train 5.826147 (lr=3.7142e-05) (hash(x)=146086947)
+4600 val loss 5.8704
+4600 val perplexity 354.3936
+4600 train 5.715634 (lr=3.6028e-05) (hash(x)=153800173)
+4700 val loss 5.8539
+4700 val perplexity 348.5879
+4700 train 5.917669 (lr=3.4915e-05) (hash(x)=155962726)
+4800 val loss 5.8462
+4800 val perplexity 345.9281
+4800 train 6.081252 (lr=3.3802e-05) (hash(x)=142045616)
+4900 val loss 5.8335
+4900 val perplexity 341.5412
+4900 train 5.802567 (lr=3.2693e-05) (hash(x)=143418248)
+5000 val loss 5.8227
+5000 val perplexity 337.8939
+5000 train 5.742158 (lr=3.1587e-05) (hash(x)=145789790)
+5100 val loss 5.8153
+5100 val perplexity 335.3846
+5100 train 5.797321 (lr=3.0488e-05) (hash(x)=137795633)
+5200 val loss 5.8010
+5200 val perplexity 330.6296
+5200 train 5.963263 (lr=2.9396e-05) (hash(x)=148907132)
+5300 val loss 5.7919
+5300 val perplexity 327.6216
+5300 train 5.666360 (lr=2.8314e-05) (hash(x)=152343580)
+5400 val loss 5.7834
+5400 val perplexity 324.8591
+5400 train 5.640688 (lr=2.7243e-05) (hash(x)=148578264)
+5500 val loss 5.7673
+5500 val perplexity 319.6873
+5500 train 5.925671 (lr=2.6183e-05) (hash(x)=145635833)
+5600 val loss 5.7618
+5600 val perplexity 317.9293
+5600 train 5.748021 (lr=2.5138e-05) (hash(x)=156337844)
+5700 val loss 5.7589
+5700 val perplexity 317.0084
+5700 train 5.632628 (lr=2.4108e-05) (hash(x)=147168506)
+5800 val loss 5.7470
+5800 val perplexity 313.2369
+5800 train 5.801088 (lr=2.3095e-05) (hash(x)=159566920)
+5900 val loss 5.7376
+5900 val perplexity 310.3252
+5900 train 5.795163 (lr=2.2100e-05) (hash(x)=158273929)
+6000 val loss 5.7379
+6000 val perplexity 310.4150
+6000 train 5.709027 (lr=2.1125e-05) (hash(x)=156649749)
+6100 val loss 5.7249
+6100 val perplexity 306.3973
+6100 train 5.643465 (lr=2.0171e-05) (hash(x)=146812388)
+6200 val loss 5.7202
+6200 val perplexity 304.9746
+6200 train 5.548894 (lr=1.9240e-05) (hash(x)=143522146)
+6300 val loss 5.7153
+6300 val perplexity 303.4803
+6300 train 5.537858 (lr=1.8333e-05) (hash(x)=150124474)
+6400 val loss 5.7068
+6400 val perplexity 300.8941
+6400 train 5.730670 (lr=1.7451e-05) (hash(x)=141242117)
+6500 val loss 5.6993
+6500 val perplexity 298.6642
+6500 train 5.578855 (lr=1.6595e-05) (hash(x)=143529762)
+6600 val loss 5.6954
+6600 val perplexity 297.5073
+6600 train 5.486333 (lr=1.5768e-05) (hash(x)=136948374)
+6700 val loss 5.6937
+6700 val perplexity 296.9866
+6700 train 5.499346 (lr=1.4969e-05) (hash(x)=146268592)
+6800 val loss 5.6864
+6800 val perplexity 294.8358
+6800 train 5.636919 (lr=1.4200e-05) (hash(x)=152676836)
+6900 val loss 5.6763
+6900 val perplexity 291.8597
+6900 train 5.594341 (lr=1.3463e-05) (hash(x)=134657776)
+7000 val loss 5.6748
+7000 val perplexity 291.4367
+7000 train 5.625916 (lr=1.2758e-05) (hash(x)=166721861)
+7100 val loss 5.6733
+7100 val perplexity 291.0062
+7100 train 5.475239 (lr=1.2086e-05) (hash(x)=135496702)
+7200 val loss 5.6678
+7200 val perplexity 289.3969
+7200 train 5.775498 (lr=1.1449e-05) (hash(x)=155567461)
+7300 val loss 5.6599
+7300 val perplexity 287.1293
+7300 train 5.460237 (lr=1.0847e-05) (hash(x)=142803829)
+7400 val loss 5.6586
+7400 val perplexity 286.7600
+7400 train 5.483219 (lr=1.0281e-05) (hash(x)=145294178)
+7500 val loss 5.6567
+7500 val perplexity 286.2040
+7500 train 5.416491 (lr=9.7516e-06) (hash(x)=150573713)
+7600 val loss 5.6536
+7600 val perplexity 285.3220
+7600 train 5.668944 (lr=9.2601e-06) (hash(x)=142771511)
+7700 val loss 5.6465
+7700 val perplexity 283.3120
+7700 train 5.548744 (lr=8.8071e-06) (hash(x)=143602175)
+7800 val loss 5.6434
+7800 val perplexity 282.4251
+7800 train 5.693456 (lr=8.3932e-06) (hash(x)=152379862)
+7900 val loss 5.6425
+7900 val perplexity 282.1734
+7900 train 5.465147 (lr=8.0189e-06) (hash(x)=146655921)
+8000 val loss 5.6439
+8000 val perplexity 282.5518
+8000 train 5.776010 (lr=7.6848e-06) (hash(x)=148262482)
+8100 val loss 5.6344
+8100 val perplexity 279.8912
+8100 train 5.576598 (lr=7.3914e-06) (hash(x)=147683655)
+8200 val loss 5.6331
+8200 val perplexity 279.5405
+8200 train 5.751260 (lr=7.1392e-06) (hash(x)=157312987)
+8300 val loss 5.6306
+8300 val perplexity 278.8158
+8300 train 5.590483 (lr=6.9284e-06) (hash(x)=141107543)
+8400 val loss 5.6260
+8400 val perplexity 277.5457
+8400 train 5.621188 (lr=6.7594e-06) (hash(x)=141323024)
+8500 val loss 5.6245
+8500 val perplexity 277.1367
+8500 train 5.586369 (lr=6.6324e-06) (hash(x)=150696521)
+8600 val loss 5.6241
+8600 val perplexity 277.0347
+8600 train 5.567379 (lr=6.5477e-06) (hash(x)=162288191)
+8700 val loss 5.6196
+8700 val perplexity 275.7833
+8700 train 5.487623 (lr=6.5053e-06) (hash(x)=152860941)
+8749 val loss 5.6180
+8749 val perplexity 275.3439

lr6.5e-5_total_batch_size61440_baseline_seed1341/model_08749.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:68f77d8af9b0f725382024b28844e2e0f30f66880bb1b0c1378fb9700f21124a
+size 92843394

lr6.5e-5_total_batch_size61440_baseline_seed1341/optimizer_08749.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:8720c2cc66b479863bf89a0d3b02b65f5b709da8bef4330a1d852da382d8bacf
+size 179406214