Upload folder using huggingface_hub

lr6e-5_total_batch_size61440_baseline_seed1339/args.json

@@ -0,0 +1 @@
+{"hellaswag": true, "attention_kind": "selective", "log_dir": "wider_is_better_10/lr6e-5_total_batch_size61440_baseline_seed1339", "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": 1339, "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_1339", "n_embd": 256}

lr6e-5_total_batch_size61440_baseline_seed1339/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_seed1339/log2.txt

@@ -0,0 +1,267 @@
+max_steps: 8750
+0 val loss 11.2808
+0 val perplexity 79287.8281
+0 train 11.289343 (lr=1.2000e-07) (hash(x)=150724848)
+100 val loss 9.9755
+100 val perplexity 21492.5566
+100 train 9.989188 (lr=1.2120e-05) (hash(x)=149910534)
+200 val loss 9.0484
+200 val perplexity 8504.8945
+200 train 9.075457 (lr=2.4120e-05) (hash(x)=148123706)
+300 val loss 7.8743
+300 val perplexity 2628.8279
+300 train 7.727704 (lr=3.6120e-05) (hash(x)=146678221)
+400 val loss 7.5006
+400 val perplexity 1809.1740
+400 train 7.385983 (lr=4.8120e-05) (hash(x)=151700982)
+500 val loss 7.3579
+500 val perplexity 1568.5027
+500 train 7.441451 (lr=6.0000e-05) (hash(x)=156182087)
+600 val loss 7.2430
+600 val perplexity 1398.2352
+600 train 7.216476 (lr=5.9980e-05) (hash(x)=149318660)
+700 val loss 7.1223
+700 val perplexity 1239.3071
+700 train 7.160754 (lr=5.9922e-05) (hash(x)=150482428)
+800 val loss 6.9518
+800 val perplexity 1045.0728
+800 train 6.828316 (lr=5.9824e-05) (hash(x)=143268605)
+900 val loss 6.8351
+900 val perplexity 929.9632
+900 train 6.803361 (lr=5.9687e-05) (hash(x)=152322423)
+1000 val loss 6.7339
+1000 val perplexity 840.4306
+1000 train 6.601456 (lr=5.9512e-05) (hash(x)=147904298)
+1100 val loss 6.6412
+1100 val perplexity 766.0239
+1100 train 6.844751 (lr=5.9298e-05) (hash(x)=154343147)
+1200 val loss 6.5514
+1200 val perplexity 700.1945
+1200 train 6.538544 (lr=5.9046e-05) (hash(x)=141843115)
+1300 val loss 6.4751
+1300 val perplexity 648.7869
+1300 train 6.367088 (lr=5.8757e-05) (hash(x)=145279030)
+1400 val loss 6.4372
+1400 val perplexity 624.6503
+1400 train 6.378782 (lr=5.8430e-05) (hash(x)=152507639)
+1500 val loss 6.3972
+1500 val perplexity 600.1918
+1500 train 6.431919 (lr=5.8066e-05) (hash(x)=148473774)
+1600 val loss 6.3373
+1600 val perplexity 565.2644
+1600 train 6.383267 (lr=5.7666e-05) (hash(x)=151117002)
+1700 val loss 6.2935
+1700 val perplexity 541.0338
+1700 train 6.245098 (lr=5.7230e-05) (hash(x)=138011335)
+1800 val loss 6.2530
+1800 val perplexity 519.5478
+1800 train 6.376237 (lr=5.6759e-05) (hash(x)=171180926)
+1900 val loss 6.2118
+1900 val perplexity 498.6025
+1900 train 6.152692 (lr=5.6253e-05) (hash(x)=141769419)
+2000 val loss 6.1771
+2000 val perplexity 481.5746
+2000 train 6.130800 (lr=5.5714e-05) (hash(x)=151963443)
+2100 val loss 6.1687
+2100 val perplexity 477.5611
+2100 train 6.221868 (lr=5.5142e-05) (hash(x)=162947470)
+2200 val loss 6.1192
+2200 val perplexity 454.4800
+2200 train 6.321251 (lr=5.4537e-05) (hash(x)=154954810)
+2300 val loss 6.0971
+2300 val perplexity 444.5767
+2300 train 6.072069 (lr=5.3902e-05) (hash(x)=151878111)
+2400 val loss 6.0779
+2400 val perplexity 436.0949
+2400 train 5.980672 (lr=5.3236e-05) (hash(x)=158661057)
+2500 val loss 6.0451
+2500 val perplexity 422.0493
+2500 train 6.070344 (lr=5.2541e-05) (hash(x)=150925584)
+2600 val loss 6.0258
+2600 val perplexity 413.9886
+2600 train 5.940624 (lr=5.1817e-05) (hash(x)=144515755)
+2700 val loss 6.0047
+2700 val perplexity 405.3316
+2700 train 5.989003 (lr=5.1067e-05) (hash(x)=153109144)
+2800 val loss 5.9778
+2800 val perplexity 394.5779
+2800 train 5.888316 (lr=5.0290e-05) (hash(x)=151152897)
+2900 val loss 5.9601
+2900 val perplexity 387.6386
+2900 train 5.905348 (lr=4.9487e-05) (hash(x)=145800210)
+3000 val loss 5.9444
+3000 val perplexity 381.6084
+3000 train 5.756807 (lr=4.8662e-05) (hash(x)=141997485)
+3100 val loss 5.9257
+3100 val perplexity 374.5420
+3100 train 5.849598 (lr=4.7813e-05) (hash(x)=154049740)
+3200 val loss 5.9004
+3200 val perplexity 365.1790
+3200 train 5.827687 (lr=4.6943e-05) (hash(x)=150471842)
+3300 val loss 5.8913
+3300 val perplexity 361.8876
+3300 train 5.839287 (lr=4.6052e-05) (hash(x)=149048126)
+3400 val loss 5.8647
+3400 val perplexity 352.3678
+3400 train 5.989864 (lr=4.5143e-05) (hash(x)=161261339)
+3500 val loss 5.8504
+3500 val perplexity 347.3798
+3500 train 5.810623 (lr=4.4216e-05) (hash(x)=157495564)
+3600 val loss 5.8381
+3600 val perplexity 343.1161
+3600 train 5.751511 (lr=4.3273e-05) (hash(x)=144352932)
+3700 val loss 5.8167
+3700 val perplexity 335.8481
+3700 train 5.782929 (lr=4.2315e-05) (hash(x)=149389012)
+3800 val loss 5.7984
+3800 val perplexity 329.7632
+3800 train 5.741716 (lr=4.1343e-05) (hash(x)=146607620)
+3900 val loss 5.7860
+3900 val perplexity 325.7057
+3900 train 5.700026 (lr=4.0360e-05) (hash(x)=143754617)
+4000 val loss 5.7704
+4000 val perplexity 320.6554
+4000 train 5.739624 (lr=3.9365e-05) (hash(x)=156930722)
+4100 val loss 5.7582
+4100 val perplexity 316.7666
+4100 train 5.593564 (lr=3.8362e-05) (hash(x)=147500519)
+4200 val loss 5.7446
+4200 val perplexity 312.4859
+4200 train 5.583457 (lr=3.7351e-05) (hash(x)=143232237)
+4300 val loss 5.7327
+4300 val perplexity 308.8044
+4300 train 5.638025 (lr=3.6333e-05) (hash(x)=146811670)
+4400 val loss 5.7176
+4400 val perplexity 304.1682
+4400 train 5.749782 (lr=3.5311e-05) (hash(x)=158418746)
+4500 val loss 5.7106
+4500 val perplexity 302.0473
+4500 train 5.676314 (lr=3.4285e-05) (hash(x)=156695778)
+4600 val loss 5.6956
+4600 val perplexity 297.5486
+4600 train 5.706344 (lr=3.3257e-05) (hash(x)=147791497)
+4700 val loss 5.6819
+4700 val perplexity 293.5062
+4700 train 5.575708 (lr=3.2229e-05) (hash(x)=155533088)
+4800 val loss 5.6756
+4800 val perplexity 291.6633
+4800 train 5.531428 (lr=3.1202e-05) (hash(x)=138350044)
+4900 val loss 5.6626
+4900 val perplexity 287.9030
+4900 train 5.649870 (lr=3.0178e-05) (hash(x)=143735284)
+5000 val loss 5.6509
+5000 val perplexity 284.5386
+5000 train 5.603555 (lr=2.9157e-05) (hash(x)=154976463)
+5100 val loss 5.6497
+5100 val perplexity 284.1965
+5100 train 5.533969 (lr=2.8143e-05) (hash(x)=149894982)
+5200 val loss 5.6381
+5200 val perplexity 280.9320
+5200 train 5.412635 (lr=2.7135e-05) (hash(x)=159326689)
+5300 val loss 5.6300
+5300 val perplexity 278.6674
+5300 train 5.643821 (lr=2.6136e-05) (hash(x)=159484800)
+5400 val loss 5.6156
+5400 val perplexity 274.6644
+5400 train 5.522407 (lr=2.5147e-05) (hash(x)=140385615)
+5500 val loss 5.6077
+5500 val perplexity 272.5098
+5500 train 5.436176 (lr=2.4169e-05) (hash(x)=148498335)
+5600 val loss 5.6077
+5600 val perplexity 272.5115
+5600 train 5.349473 (lr=2.3204e-05) (hash(x)=151907614)
+5700 val loss 5.5970
+5700 val perplexity 269.6220
+5700 train 5.567425 (lr=2.2253e-05) (hash(x)=155192267)
+5800 val loss 5.5849
+5800 val perplexity 266.3655
+5800 train 5.635519 (lr=2.1318e-05) (hash(x)=153132158)
+5900 val loss 5.5835
+5900 val perplexity 265.9950
+5900 train 5.555301 (lr=2.0400e-05) (hash(x)=161446764)
+6000 val loss 5.5777
+6000 val perplexity 264.4673
+6000 train 5.361888 (lr=1.9500e-05) (hash(x)=151512446)
+6100 val loss 5.5663
+6100 val perplexity 261.4657
+6100 train 5.658627 (lr=1.8620e-05) (hash(x)=188094053)
+6200 val loss 5.5557
+6200 val perplexity 258.7140
+6200 train 5.433425 (lr=1.7760e-05) (hash(x)=149389789)
+6300 val loss 5.5506
+6300 val perplexity 257.3858
+6300 train 5.367990 (lr=1.6923e-05) (hash(x)=138212820)
+6400 val loss 5.5488
+6400 val perplexity 256.9208
+6400 train 5.334530 (lr=1.6108e-05) (hash(x)=146535423)
+6500 val loss 5.5397
+6500 val perplexity 254.6079
+6500 train 5.409243 (lr=1.5319e-05) (hash(x)=145950843)
+6600 val loss 5.5365
+6600 val perplexity 253.7826
+6600 train 5.443093 (lr=1.4555e-05) (hash(x)=141162902)
+6700 val loss 5.5333
+6700 val perplexity 252.9709
+6700 train 5.449463 (lr=1.3817e-05) (hash(x)=153018737)
+6800 val loss 5.5259
+6800 val perplexity 251.1181
+6800 train 5.435163 (lr=1.3108e-05) (hash(x)=155640155)
+6900 val loss 5.5232
+6900 val perplexity 250.4237
+6900 train 5.494747 (lr=1.2427e-05) (hash(x)=153722115)
+7000 val loss 5.5155
+7000 val perplexity 248.5069
+7000 train 5.431949 (lr=1.1777e-05) (hash(x)=146953450)
+7100 val loss 5.5169
+7100 val perplexity 248.8593
+7100 train 5.365694 (lr=1.1157e-05) (hash(x)=137663885)
+7200 val loss 5.5062
+7200 val perplexity 246.2229
+7200 train 5.588296 (lr=1.0568e-05) (hash(x)=146172950)
+7300 val loss 5.5022
+7300 val perplexity 245.2201
+7300 train 5.388580 (lr=1.0012e-05) (hash(x)=150018163)
+7400 val loss 5.4998
+7400 val perplexity 244.6358
+7400 train 5.517736 (lr=9.4899e-06) (hash(x)=145351166)
+7500 val loss 5.5001
+7500 val perplexity 244.7241
+7500 train 5.276795 (lr=9.0014e-06) (hash(x)=145292116)
+7600 val loss 5.4925
+7600 val perplexity 242.8609
+7600 train 5.412445 (lr=8.5478e-06) (hash(x)=150235132)
+7700 val loss 5.4889
+7700 val perplexity 241.9937
+7700 train 5.387060 (lr=8.1297e-06) (hash(x)=154543455)
+7800 val loss 5.4877
+7800 val perplexity 241.7115
+7800 train 5.389268 (lr=7.7476e-06) (hash(x)=142456852)
+7900 val loss 5.4883
+7900 val perplexity 241.8375
+7900 train 5.262912 (lr=7.4021e-06) (hash(x)=147363479)
+8000 val loss 5.4807
+8000 val perplexity 240.0065
+8000 train 5.512944 (lr=7.0937e-06) (hash(x)=156122973)
+8100 val loss 5.4789
+8100 val perplexity 239.5800
+8100 train 5.522630 (lr=6.8229e-06) (hash(x)=156153179)
+8200 val loss 5.4765
+8200 val perplexity 238.9997
+8200 train 5.532199 (lr=6.5900e-06) (hash(x)=146430698)
+8300 val loss 5.4750
+8300 val perplexity 238.6505
+8300 train 5.345642 (lr=6.3954e-06) (hash(x)=143507257)
+8400 val loss 5.4696
+8400 val perplexity 237.3676
+8400 train 5.448027 (lr=6.2395e-06) (hash(x)=166272643)
+8500 val loss 5.4683
+8500 val perplexity 237.0685
+8500 train 5.377180 (lr=6.1223e-06) (hash(x)=143887848)
+8600 val loss 5.4685
+8600 val perplexity 237.1143
+8600 train 5.497317 (lr=6.0440e-06) (hash(x)=156900341)
+8700 val loss 5.4631
+8700 val perplexity 235.8200
+8700 train 5.699231 (lr=6.0049e-06) (hash(x)=146417632)
+8749 val loss 5.4630
+8749 val perplexity 235.8106

lr6e-5_total_batch_size61440_baseline_seed1339/model_08749.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:f188a713ea40200ac4ca85dc1d82bf92556e91399b18dcacb2cc5204603dede9
+size 92843394

lr6e-5_total_batch_size61440_baseline_seed1339/optimizer_08749.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:9eda9d71b255206a809f36baa6d1b5ca98af40cf9797504040890c078386df5b
+size 179406214