Upload folder using huggingface_hub

lr6e-5_total_batch_size61440_baseline_seed1341/args.json

@@ -0,0 +1 @@
+{"hellaswag": true, "attention_kind": "selective", "log_dir": "wider_is_better_10/lr6e-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": 6e-05, "decay_lr": true, "readout_zero_init": false, "query_zero_init": false, "l1_loss": false, "debugpy": false, "key": "6e-5_61440_1341", "n_embd": 256}

lr6e-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

lr6e-5_total_batch_size61440_baseline_seed1341/log2.txt

@@ -0,0 +1,267 @@
+max_steps: 8750
+0 val loss 11.3095
+0 val perplexity 81594.5625
+0 train 11.313038 (lr=1.2000e-07) (hash(x)=145079536)
+100 val loss 10.0135
+100 val perplexity 22325.9512
+100 train 10.007975 (lr=1.2120e-05) (hash(x)=155800595)
+200 val loss 9.1023
+200 val perplexity 8976.3135
+200 train 9.047848 (lr=2.4120e-05) (hash(x)=145606733)
+300 val loss 7.8128
+300 val perplexity 2472.1331
+300 train 7.767262 (lr=3.6120e-05) (hash(x)=150367139)
+400 val loss 7.5184
+400 val perplexity 1841.6793
+400 train 7.828914 (lr=4.8120e-05) (hash(x)=155747374)
+500 val loss 7.3954
+500 val perplexity 1628.4949
+500 train 7.283290 (lr=6.0000e-05) (hash(x)=140604760)
+600 val loss 7.3038
+600 val perplexity 1485.8616
+600 train 7.240885 (lr=5.9980e-05) (hash(x)=148404734)
+700 val loss 7.2217
+700 val perplexity 1368.7478
+700 train 7.174208 (lr=5.9922e-05) (hash(x)=148115934)
+800 val loss 7.1430
+800 val perplexity 1265.2288
+800 train 7.026456 (lr=5.9824e-05) (hash(x)=137464699)
+900 val loss 7.0783
+900 val perplexity 1185.9524
+900 train 6.891728 (lr=5.9687e-05) (hash(x)=143886042)
+1000 val loss 7.0191
+1000 val perplexity 1117.8007
+1000 train 7.024662 (lr=5.9512e-05) (hash(x)=163799796)
+1100 val loss 6.9561
+1100 val perplexity 1049.5629
+1100 train 6.912311 (lr=5.9298e-05) (hash(x)=144592844)
+1200 val loss 6.8763
+1200 val perplexity 969.0346
+1200 train 7.189436 (lr=5.9046e-05) (hash(x)=204706354)
+1300 val loss 6.7876
+1300 val perplexity 886.7493
+1300 train 6.720086 (lr=5.8757e-05) (hash(x)=150862210)
+1400 val loss 6.6863
+1400 val perplexity 801.3672
+1400 train 6.681302 (lr=5.8430e-05) (hash(x)=147766811)
+1500 val loss 6.6032
+1500 val perplexity 737.4558
+1500 train 6.421759 (lr=5.8066e-05) (hash(x)=135925327)
+1600 val loss 6.5544
+1600 val perplexity 702.3112
+1600 train 6.521947 (lr=5.7666e-05) (hash(x)=160440642)
+1700 val loss 6.4916
+1700 val perplexity 659.5621
+1700 train 6.604711 (lr=5.7230e-05) (hash(x)=151184106)
+1800 val loss 6.4506
+1800 val perplexity 633.1057
+1800 train 6.345433 (lr=5.6759e-05) (hash(x)=148100580)
+1900 val loss 6.4345
+1900 val perplexity 622.9780
+1900 train 6.264120 (lr=5.6253e-05) (hash(x)=149434659)
+2000 val loss 6.3658
+2000 val perplexity 581.6387
+2000 train 6.263128 (lr=5.5714e-05) (hash(x)=152285486)
+2100 val loss 6.3185
+2100 val perplexity 554.7516
+2100 train 6.063070 (lr=5.5142e-05) (hash(x)=144294295)
+2200 val loss 6.2887
+2200 val perplexity 538.4678
+2200 train 6.300180 (lr=5.4537e-05) (hash(x)=175030215)
+2300 val loss 6.2594
+2300 val perplexity 522.9048
+2300 train 6.070185 (lr=5.3902e-05) (hash(x)=150831428)
+2400 val loss 6.2105
+2400 val perplexity 497.9686
+2400 train 6.392653 (lr=5.3236e-05) (hash(x)=140808297)
+2500 val loss 6.1788
+2500 val perplexity 482.4130
+2500 train 6.160472 (lr=5.2541e-05) (hash(x)=153160275)
+2600 val loss 6.1571
+2600 val perplexity 472.0579
+2600 train 5.952056 (lr=5.1817e-05) (hash(x)=133990623)
+2700 val loss 6.1319
+2700 val perplexity 460.3261
+2700 train 6.103174 (lr=5.1067e-05) (hash(x)=142860944)
+2800 val loss 6.1009
+2800 val perplexity 446.2404
+2800 train 6.118627 (lr=5.0290e-05) (hash(x)=137959511)
+2900 val loss 6.0720
+2900 val perplexity 433.5335
+2900 train 6.024552 (lr=4.9487e-05) (hash(x)=147009873)
+3000 val loss 6.0508
+3000 val perplexity 424.4506
+3000 train 5.904181 (lr=4.8662e-05) (hash(x)=158264841)
+3100 val loss 6.0405
+3100 val perplexity 420.1201
+3100 train 5.874842 (lr=4.7813e-05) (hash(x)=139232251)
+3200 val loss 6.0076
+3200 val perplexity 406.5200
+3200 train 6.047667 (lr=4.6943e-05) (hash(x)=153436104)
+3300 val loss 5.9899
+3300 val perplexity 399.3808
+3300 train 5.815275 (lr=4.6052e-05) (hash(x)=149681831)
+3400 val loss 5.9729
+3400 val perplexity 392.6516
+3400 train 5.846929 (lr=4.5143e-05) (hash(x)=168425516)
+3500 val loss 5.9473
+3500 val perplexity 382.7263
+3500 train 5.973871 (lr=4.4216e-05) (hash(x)=163104338)
+3600 val loss 5.9283
+3600 val perplexity 375.5199
+3600 train 5.898438 (lr=4.3273e-05) (hash(x)=165109772)
+3700 val loss 5.9255
+3700 val perplexity 374.4577
+3700 train 5.662436 (lr=4.2315e-05) (hash(x)=153420306)
+3800 val loss 5.8938
+3800 val perplexity 362.7888
+3800 train 5.902781 (lr=4.1343e-05) (hash(x)=160168863)
+3900 val loss 5.8821
+3900 val perplexity 358.5515
+3900 train 5.799942 (lr=4.0360e-05) (hash(x)=153906073)
+4000 val loss 5.8668
+4000 val perplexity 353.1042
+4000 train 5.816743 (lr=3.9365e-05) (hash(x)=151055067)
+4100 val loss 5.8531
+4100 val perplexity 348.3010
+4100 train 5.831659 (lr=3.8362e-05) (hash(x)=149629830)
+4200 val loss 5.8380
+4200 val perplexity 343.0844
+4200 train 5.723072 (lr=3.7351e-05) (hash(x)=143101381)
+4300 val loss 5.8252
+4300 val perplexity 338.7227
+4300 train 5.726717 (lr=3.6333e-05) (hash(x)=149712044)
+4400 val loss 5.8073
+4400 val perplexity 332.7252
+4400 train 5.734195 (lr=3.5311e-05) (hash(x)=153446449)
+4500 val loss 5.7931
+4500 val perplexity 328.0166
+4500 train 5.742892 (lr=3.4285e-05) (hash(x)=146086947)
+4600 val loss 5.7831
+4600 val perplexity 324.7725
+4600 train 5.625107 (lr=3.3257e-05) (hash(x)=153800173)
+4700 val loss 5.7672
+4700 val perplexity 319.6543
+4700 train 5.838467 (lr=3.2229e-05) (hash(x)=155962726)
+4800 val loss 5.7620
+4800 val perplexity 317.9879
+4800 train 6.006111 (lr=3.1202e-05) (hash(x)=142045616)
+4900 val loss 5.7472
+4900 val perplexity 313.3211
+4900 train 5.717638 (lr=3.0178e-05) (hash(x)=143418248)
+5000 val loss 5.7375
+5000 val perplexity 310.2754
+5000 train 5.661459 (lr=2.9157e-05) (hash(x)=145789790)
+5100 val loss 5.7320
+5100 val perplexity 308.5795
+5100 train 5.716419 (lr=2.8143e-05) (hash(x)=137795633)
+5200 val loss 5.7207
+5200 val perplexity 305.1143
+5200 train 5.870852 (lr=2.7135e-05) (hash(x)=148907132)
+5300 val loss 5.7103
+5300 val perplexity 301.9531
+5300 train 5.586517 (lr=2.6136e-05) (hash(x)=152343580)
+5400 val loss 5.7010
+5400 val perplexity 299.1613
+5400 train 5.554786 (lr=2.5147e-05) (hash(x)=148578264)
+5500 val loss 5.6897
+5500 val perplexity 295.7905
+5500 train 5.848770 (lr=2.4169e-05) (hash(x)=145635833)
+5600 val loss 5.6819
+5600 val perplexity 293.5197
+5600 train 5.671008 (lr=2.3204e-05) (hash(x)=156337844)
+5700 val loss 5.6771
+5700 val perplexity 292.1105
+5700 train 5.547239 (lr=2.2253e-05) (hash(x)=147168506)
+5800 val loss 5.6649
+5800 val perplexity 288.5723
+5800 train 5.709722 (lr=2.1318e-05) (hash(x)=159566920)
+5900 val loss 5.6571
+5900 val perplexity 286.3130
+5900 train 5.712388 (lr=2.0400e-05) (hash(x)=158273929)
+6000 val loss 5.6559
+6000 val perplexity 285.9781
+6000 train 5.626070 (lr=1.9500e-05) (hash(x)=156649749)
+6100 val loss 5.6450
+6100 val perplexity 282.8661
+6100 train 5.562746 (lr=1.8620e-05) (hash(x)=146812388)
+6200 val loss 5.6399
+6200 val perplexity 281.4344
+6200 train 5.474884 (lr=1.7760e-05) (hash(x)=143522146)
+6300 val loss 5.6348
+6300 val perplexity 280.0072
+6300 train 5.450123 (lr=1.6923e-05) (hash(x)=150124474)
+6400 val loss 5.6263
+6400 val perplexity 277.6254
+6400 train 5.660075 (lr=1.6108e-05) (hash(x)=141242117)
+6500 val loss 5.6180
+6500 val perplexity 275.3304
+6500 train 5.491308 (lr=1.5319e-05) (hash(x)=143529762)
+6600 val loss 5.6132
+6600 val perplexity 274.0089
+6600 train 5.398042 (lr=1.4555e-05) (hash(x)=136948374)
+6700 val loss 5.6126
+6700 val perplexity 273.8648
+6700 train 5.404612 (lr=1.3817e-05) (hash(x)=146268592)
+6800 val loss 5.6039
+6800 val perplexity 271.4702
+6800 train 5.554190 (lr=1.3108e-05) (hash(x)=152676836)
+6900 val loss 5.5955
+6900 val perplexity 269.2032
+6900 train 5.514298 (lr=1.2427e-05) (hash(x)=134657776)
+7000 val loss 5.5926
+7000 val perplexity 268.4449
+7000 train 5.538960 (lr=1.1777e-05) (hash(x)=166721861)
+7100 val loss 5.5907
+7100 val perplexity 267.9311
+7100 train 5.383312 (lr=1.1157e-05) (hash(x)=135496702)
+7200 val loss 5.5871
+7200 val perplexity 266.9684
+7200 train 5.696351 (lr=1.0568e-05) (hash(x)=155567461)
+7300 val loss 5.5781
+7300 val perplexity 264.5772
+7300 train 5.382541 (lr=1.0012e-05) (hash(x)=142803829)
+7400 val loss 5.5778
+7400 val perplexity 264.4870
+7400 train 5.401532 (lr=9.4899e-06) (hash(x)=145294178)
+7500 val loss 5.5745
+7500 val perplexity 263.6072
+7500 train 5.330779 (lr=9.0014e-06) (hash(x)=150573713)
+7600 val loss 5.5718
+7600 val perplexity 262.9147
+7600 train 5.581925 (lr=8.5478e-06) (hash(x)=142771511)
+7700 val loss 5.5662
+7700 val perplexity 261.4331
+7700 train 5.472789 (lr=8.1297e-06) (hash(x)=143602175)
+7800 val loss 5.5628
+7800 val perplexity 260.5639
+7800 train 5.623899 (lr=7.7476e-06) (hash(x)=152379862)
+7900 val loss 5.5615
+7900 val perplexity 260.2248
+7900 train 5.372259 (lr=7.4021e-06) (hash(x)=146655921)
+8000 val loss 5.5627
+8000 val perplexity 260.5373
+8000 train 5.703586 (lr=7.0937e-06) (hash(x)=148262482)
+8100 val loss 5.5532
+8100 val perplexity 258.0580
+8100 train 5.498971 (lr=6.8229e-06) (hash(x)=147683655)
+8200 val loss 5.5533
+8200 val perplexity 258.0817
+8200 train 5.670504 (lr=6.5900e-06) (hash(x)=157312987)
+8300 val loss 5.5501
+8300 val perplexity 257.2672
+8300 train 5.504878 (lr=6.3954e-06) (hash(x)=141107543)
+8400 val loss 5.5457
+8400 val perplexity 256.1457
+8400 train 5.547677 (lr=6.2395e-06) (hash(x)=141323024)
+8500 val loss 5.5439
+8500 val perplexity 255.6770
+8500 train 5.511901 (lr=6.1223e-06) (hash(x)=150696521)
+8600 val loss 5.5438
+8600 val perplexity 255.6536
+8600 train 5.472732 (lr=6.0440e-06) (hash(x)=162288191)
+8700 val loss 5.5389
+8700 val perplexity 254.3990
+8700 train 5.406146 (lr=6.0049e-06) (hash(x)=152860941)
+8749 val loss 5.5381
+8749 val perplexity 254.2068

lr6e-5_total_batch_size61440_baseline_seed1341/model_08749.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:4de5af6bc2e9ce295e10570d20596c0d92b1ec5a8f7163e74f2afadfab1ccba7
+size 92843394

lr6e-5_total_batch_size61440_baseline_seed1341/optimizer_08749.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:6195253d78111832c707bcda8bc41a95e099b7b0f7812f63fb1c2aec29c1c336
+size 179406214