Upload folder using huggingface_hub

lr6e-5_total_batch_size61440_baseline_seed1340/args.json

@@ -0,0 +1 @@
+{"hellaswag": true, "attention_kind": "selective", "log_dir": "wider_is_better_10/lr6e-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": 6e-05, "decay_lr": true, "readout_zero_init": false, "query_zero_init": false, "l1_loss": false, "debugpy": false, "key": "6e-5_61440_1340", "n_embd": 256}

lr6e-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

lr6e-5_total_batch_size61440_baseline_seed1340/log2.txt

@@ -0,0 +1,267 @@
+max_steps: 8750
+0 val loss 11.2709
+0 val perplexity 78503.3750
+0 train 11.277069 (lr=1.2000e-07) (hash(x)=164406924)
+100 val loss 9.9849
+100 val perplexity 21697.0371
+100 train 9.946177 (lr=1.2120e-05) (hash(x)=144903932)
+200 val loss 9.1835
+200 val perplexity 9735.4102
+200 train 9.251645 (lr=2.4120e-05) (hash(x)=167734596)
+300 val loss 7.9477
+300 val perplexity 2829.1726
+300 train 7.953303 (lr=3.6120e-05) (hash(x)=149619098)
+400 val loss 7.5520
+400 val perplexity 1904.5880
+400 train 7.431348 (lr=4.8120e-05) (hash(x)=153710890)
+500 val loss 7.4046
+500 val perplexity 1643.5040
+500 train 7.499173 (lr=6.0000e-05) (hash(x)=145450636)
+600 val loss 7.2890
+600 val perplexity 1464.0878
+600 train 7.155668 (lr=5.9980e-05) (hash(x)=145249251)
+700 val loss 7.1905
+700 val perplexity 1326.7148
+700 train 7.319555 (lr=5.9922e-05) (hash(x)=150475545)
+800 val loss 7.0984
+800 val perplexity 1210.0538
+800 train 7.069599 (lr=5.9824e-05) (hash(x)=144483776)
+900 val loss 6.9903
+900 val perplexity 1086.0930
+900 train 7.099010 (lr=5.9687e-05) (hash(x)=157916369)
+1000 val loss 6.9013
+1000 val perplexity 993.6029
+1000 train 6.911705 (lr=5.9512e-05) (hash(x)=154856891)
+1100 val loss 6.8174
+1100 val perplexity 913.6283
+1100 train 6.515013 (lr=5.9298e-05) (hash(x)=136191502)
+1200 val loss 6.7297
+1200 val perplexity 836.8663
+1200 train 6.686831 (lr=5.9046e-05) (hash(x)=148021541)
+1300 val loss 6.6537
+1300 val perplexity 775.6281
+1300 train 6.585151 (lr=5.8757e-05) (hash(x)=146005217)
+1400 val loss 6.5849
+1400 val perplexity 724.0883
+1400 train 6.496242 (lr=5.8430e-05) (hash(x)=146019502)
+1500 val loss 6.5262
+1500 val perplexity 682.8223
+1500 train 6.464975 (lr=5.8066e-05) (hash(x)=150127281)
+1600 val loss 6.4802
+1600 val perplexity 652.1255
+1600 train 6.445440 (lr=5.7666e-05) (hash(x)=154120875)
+1700 val loss 6.4389
+1700 val perplexity 625.7360
+1700 train 6.649782 (lr=5.7230e-05) (hash(x)=155797680)
+1800 val loss 6.3798
+1800 val perplexity 589.7950
+1800 train 6.365313 (lr=5.6759e-05) (hash(x)=156809396)
+1900 val loss 6.3443
+1900 val perplexity 569.2379
+1900 train 6.225636 (lr=5.6253e-05) (hash(x)=144640294)
+2000 val loss 6.3066
+2000 val perplexity 548.1733
+2000 train 6.164998 (lr=5.5714e-05) (hash(x)=162831106)
+2100 val loss 6.2960
+2100 val perplexity 542.4163
+2100 train 6.291697 (lr=5.5142e-05) (hash(x)=158239484)
+2200 val loss 6.2543
+2200 val perplexity 520.2327
+2200 train 6.130803 (lr=5.4537e-05) (hash(x)=140504180)
+2300 val loss 6.2165
+2300 val perplexity 500.9527
+2300 train 6.209535 (lr=5.3902e-05) (hash(x)=142234024)
+2400 val loss 6.1879
+2400 val perplexity 486.8058
+2400 train 6.182105 (lr=5.3236e-05) (hash(x)=143091562)
+2500 val loss 6.1906
+2500 val perplexity 488.1193
+2500 train 6.089002 (lr=5.2541e-05) (hash(x)=149857456)
+2600 val loss 6.1428
+2600 val perplexity 465.3682
+2600 train 6.177594 (lr=5.1817e-05) (hash(x)=146191551)
+2700 val loss 6.1196
+2700 val perplexity 454.6946
+2700 train 6.034596 (lr=5.1067e-05) (hash(x)=145375752)
+2800 val loss 6.0929
+2800 val perplexity 442.6875
+2800 train 5.970033 (lr=5.0290e-05) (hash(x)=151568014)
+2900 val loss 6.0885
+2900 val perplexity 440.7644
+2900 train 5.913503 (lr=4.9487e-05) (hash(x)=149366597)
+3000 val loss 6.0504
+3000 val perplexity 424.2705
+3000 train 5.974408 (lr=4.8662e-05) (hash(x)=150464442)
+3100 val loss 6.0309
+3100 val perplexity 416.1096
+3100 train 6.151582 (lr=4.7813e-05) (hash(x)=182449036)
+3200 val loss 6.0130
+3200 val perplexity 408.6903
+3200 train 5.962421 (lr=4.6943e-05) (hash(x)=140141286)
+3300 val loss 5.9931
+3300 val perplexity 400.6682
+3300 train 5.960182 (lr=4.6052e-05) (hash(x)=148099414)
+3400 val loss 5.9746
+3400 val perplexity 393.3049
+3400 train 5.836246 (lr=4.5143e-05) (hash(x)=142633951)
+3500 val loss 5.9628
+3500 val perplexity 388.7129
+3500 train 5.821012 (lr=4.4216e-05) (hash(x)=148368965)
+3600 val loss 5.9387
+3600 val perplexity 379.4478
+3600 train 5.956468 (lr=4.3273e-05) (hash(x)=152372067)
+3700 val loss 5.9202
+3700 val perplexity 372.5016
+3700 train 5.998601 (lr=4.2315e-05) (hash(x)=168885609)
+3800 val loss 5.9086
+3800 val perplexity 368.1968
+3800 train 5.721739 (lr=4.1343e-05) (hash(x)=141633734)
+3900 val loss 5.8858
+3900 val perplexity 359.8852
+3900 train 5.773929 (lr=4.0360e-05) (hash(x)=153141007)
+4000 val loss 5.8685
+4000 val perplexity 353.7049
+4000 train 5.794589 (lr=3.9365e-05) (hash(x)=160577202)
+4100 val loss 5.8604
+4100 val perplexity 350.8722
+4100 train 5.924521 (lr=3.8362e-05) (hash(x)=153858169)
+4200 val loss 5.8373
+4200 val perplexity 342.8643
+4200 train 5.848408 (lr=3.7351e-05) (hash(x)=155889149)
+4300 val loss 5.8252
+4300 val perplexity 338.7224
+4300 train 5.628758 (lr=3.6333e-05) (hash(x)=152294662)
+4400 val loss 5.8146
+4400 val perplexity 335.1563
+4400 train 5.719494 (lr=3.5311e-05) (hash(x)=141804386)
+4500 val loss 5.7959
+4500 val perplexity 328.9562
+4500 train 5.690154 (lr=3.4285e-05) (hash(x)=151095242)
+4600 val loss 5.8007
+4600 val perplexity 330.5300
+4600 train 5.748248 (lr=3.3257e-05) (hash(x)=156414699)
+4700 val loss 5.7723
+4700 val perplexity 321.2721
+4700 train 5.789610 (lr=3.2229e-05) (hash(x)=161556686)
+4800 val loss 5.7606
+4800 val perplexity 317.5379
+4800 train 5.663333 (lr=3.1202e-05) (hash(x)=149000293)
+4900 val loss 5.7496
+4900 val perplexity 314.0662
+4900 train 5.531809 (lr=3.0178e-05) (hash(x)=154349989)
+5000 val loss 5.7370
+5000 val perplexity 310.1462
+5000 train 5.578991 (lr=2.9157e-05) (hash(x)=131475967)
+5100 val loss 5.7215
+5100 val perplexity 305.3668
+5100 train 5.629333 (lr=2.8143e-05) (hash(x)=149717902)
+5200 val loss 5.7146
+5200 val perplexity 303.2755
+5200 train 5.581986 (lr=2.7135e-05) (hash(x)=151407999)
+5300 val loss 5.7036
+5300 val perplexity 299.9455
+5300 train 5.781487 (lr=2.6136e-05) (hash(x)=168602728)
+5400 val loss 5.6919
+5400 val perplexity 296.4426
+5400 train 5.651140 (lr=2.5147e-05) (hash(x)=158344511)
+5500 val loss 5.6845
+5500 val perplexity 294.2773
+5500 train 5.641092 (lr=2.4169e-05) (hash(x)=148350057)
+5600 val loss 5.6752
+5600 val perplexity 291.5359
+5600 train 5.741850 (lr=2.3204e-05) (hash(x)=153847323)
+5700 val loss 5.6635
+5700 val perplexity 288.1573
+5700 train 5.608263 (lr=2.2253e-05) (hash(x)=156607405)
+5800 val loss 5.6574
+5800 val perplexity 286.4085
+5800 train 5.422741 (lr=2.1318e-05) (hash(x)=145115031)
+5900 val loss 5.6485
+5900 val perplexity 283.8576
+5900 train 5.532207 (lr=2.0400e-05) (hash(x)=141584622)
+6000 val loss 5.6389
+6000 val perplexity 281.1394
+6000 train 5.573113 (lr=1.9500e-05) (hash(x)=146613857)
+6100 val loss 5.6324
+6100 val perplexity 279.3203
+6100 train 5.577152 (lr=1.8620e-05) (hash(x)=144621768)
+6200 val loss 5.6277
+6200 val perplexity 278.0090
+6200 train 5.560519 (lr=1.7760e-05) (hash(x)=146521760)
+6300 val loss 5.6176
+6300 val perplexity 275.2158
+6300 train 5.447705 (lr=1.6923e-05) (hash(x)=161378136)
+6400 val loss 5.6113
+6400 val perplexity 273.5131
+6400 train 5.379691 (lr=1.6108e-05) (hash(x)=141624235)
+6500 val loss 5.6054
+6500 val perplexity 271.8912
+6500 train 5.527293 (lr=1.5319e-05) (hash(x)=151197095)
+6600 val loss 5.6018
+6600 val perplexity 270.9190
+6600 train 5.616758 (lr=1.4555e-05) (hash(x)=153269571)
+6700 val loss 5.5950
+6700 val perplexity 269.0652
+6700 train 5.551866 (lr=1.3817e-05) (hash(x)=146111181)
+6800 val loss 5.5868
+6800 val perplexity 266.8694
+6800 train 5.620859 (lr=1.3108e-05) (hash(x)=147269760)
+6900 val loss 5.5837
+6900 val perplexity 266.0617
+6900 train 5.662534 (lr=1.2427e-05) (hash(x)=152912762)
+7000 val loss 5.5792
+7000 val perplexity 264.8583
+7000 train 5.659291 (lr=1.1777e-05) (hash(x)=165412343)
+7100 val loss 5.5700
+7100 val perplexity 262.4228
+7100 train 5.643213 (lr=1.1157e-05) (hash(x)=162866028)
+7200 val loss 5.5651
+7200 val perplexity 261.1628
+7200 train 5.532481 (lr=1.0568e-05) (hash(x)=142998115)
+7300 val loss 5.5632
+7300 val perplexity 260.6658
+7300 train 5.328614 (lr=1.0012e-05) (hash(x)=145486999)
+7400 val loss 5.5564
+7400 val perplexity 258.8949
+7400 train 5.533556 (lr=9.4899e-06) (hash(x)=155325873)
+7500 val loss 5.5527
+7500 val perplexity 257.9372
+7500 train 5.515679 (lr=9.0014e-06) (hash(x)=145131256)
+7600 val loss 5.5583
+7600 val perplexity 259.3720
+7600 train 5.487406 (lr=8.5478e-06) (hash(x)=144008365)
+7700 val loss 5.5460
+7700 val perplexity 256.2223
+7700 train 5.659881 (lr=8.1297e-06) (hash(x)=148848532)
+7800 val loss 5.5428
+7800 val perplexity 255.3829
+7800 train 5.536234 (lr=7.7476e-06) (hash(x)=150391642)
+7900 val loss 5.5413
+7900 val perplexity 255.0006
+7900 train 5.301567 (lr=7.4021e-06) (hash(x)=152191414)
+8000 val loss 5.5370
+8000 val perplexity 253.9027
+8000 train 5.614108 (lr=7.0937e-06) (hash(x)=159755587)
+8100 val loss 5.5346
+8100 val perplexity 253.2995
+8100 train 5.559530 (lr=6.8229e-06) (hash(x)=156664468)
+8200 val loss 5.5329
+8200 val perplexity 252.8750
+8200 train 5.388103 (lr=6.5900e-06) (hash(x)=139457379)
+8300 val loss 5.5365
+8300 val perplexity 253.7819
+8300 train 5.306564 (lr=6.3954e-06) (hash(x)=145478564)
+8400 val loss 5.5271
+8400 val perplexity 251.4122
+8400 train 5.459111 (lr=6.2395e-06) (hash(x)=154982769)
+8500 val loss 5.5242
+8500 val perplexity 250.6864
+8500 train 5.501402 (lr=6.1223e-06) (hash(x)=145798118)
+8600 val loss 5.5240
+8600 val perplexity 250.6313
+8600 train 5.430560 (lr=6.0440e-06) (hash(x)=143231551)
+8700 val loss 5.5273
+8700 val perplexity 251.4741
+8700 train 5.423115 (lr=6.0049e-06) (hash(x)=154780112)
+8749 val loss 5.5198
+8749 val perplexity 249.5890

lr6e-5_total_batch_size61440_baseline_seed1340/model_08749.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:8b0afb3bec60ca3b5a1ed2252a611c5d2328641481950b52353660aed1f239ba
+size 92843394

lr6e-5_total_batch_size61440_baseline_seed1340/optimizer_08749.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:4abe28ba017c2fd14689b1d38e9805a87ebf99c58a359d3450f460c3256a1222
+size 179406214