Upload folder using huggingface_hub

12_head_baseline_lr_20e-4_head_dim_22/args.json

@@ -0,0 +1 @@
+{"hellaswag": true, "attention_kind": "selective", "log_dir": "logs/repro_selective_pattern_rankings/12_head_baseline_lr_20e-4_head_dim_22", "resume_checkpoint": null, "resume_optimizer": false, "add_a_head": false, "add_head_to_start": true, "new_head_init": "normal", "n_heads": 12, "n_embd": 264, "head_dim": 22, "protect_bos_token": true, "prevent_from_masking_myself": true, "max_steps": 4375, "warmup_steps": 250, "group": "repro_selective_pattern_rankings", "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": 64, "total_batch_size": 131072, "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": 0.002, "decay_lr": true, "readout_zero_init": false, "query_zero_init": false, "l1_loss": false, "debugpy": false, "key": "12_head_baseline_lr_20e-4"}

12_head_baseline_lr_20e-4_head_dim_22/dataloader_04374.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:8ec6682852ed60a9d42cb4047300f65a04f87328c0ad7a4516be84d11b28f216
+size 964

12_head_baseline_lr_20e-4_head_dim_22/log2.txt

@@ -0,0 +1,529 @@
+max_steps: 4375
+0 val loss 10.9076
+0 val perplexity 54591.8281
+0 train 10.908085 (lr=2.7972e-06) (hash(x)=93356070)
+10 train 10.552961 (lr=3.0769e-05) (hash(x)=91253010)
+20 train 10.344660 (lr=5.8741e-05) (hash(x)=74934453)
+30 train 10.147306 (lr=8.6713e-05) (hash(x)=79278034)
+40 train 9.833697 (lr=1.1469e-04) (hash(x)=80107892)
+50 train 9.479548 (lr=1.4266e-04) (hash(x)=70745428)
+60 train 9.117208 (lr=1.7063e-04) (hash(x)=80077589)
+70 train 8.613481 (lr=1.9860e-04) (hash(x)=76213766)
+80 train 8.296818 (lr=2.2657e-04) (hash(x)=83218328)
+90 train 7.948500 (lr=2.5455e-04) (hash(x)=74855845)
+100 val loss 7.6924
+100 val perplexity 2191.6199
+100 train 7.598456 (lr=2.8252e-04) (hash(x)=82814902)
+110 train 7.435596 (lr=3.1049e-04) (hash(x)=84907741)
+120 train 7.293886 (lr=3.3846e-04) (hash(x)=82613223)
+130 train 7.108490 (lr=3.6643e-04) (hash(x)=83540876)
+140 train 7.044240 (lr=3.9441e-04) (hash(x)=75095216)
+150 train 7.063421 (lr=4.2238e-04) (hash(x)=97190944)
+160 train 6.933561 (lr=4.5035e-04) (hash(x)=82117809)
+170 train 6.812375 (lr=4.7832e-04) (hash(x)=70514724)
+180 train 6.728166 (lr=5.0629e-04) (hash(x)=81029624)
+190 train 6.641076 (lr=5.3427e-04) (hash(x)=78996841)
+200 val loss 6.5840
+200 val perplexity 723.4366
+200 train 6.586956 (lr=5.6224e-04) (hash(x)=79845097)
+210 train 6.473219 (lr=5.9021e-04) (hash(x)=78997683)
+220 train 6.363735 (lr=6.1818e-04) (hash(x)=74895865)
+230 train 6.538910 (lr=6.4615e-04) (hash(x)=80933276)
+240 train 6.253383 (lr=6.7413e-04) (hash(x)=77664606)
+250 train 6.267797 (lr=7.0210e-04) (hash(x)=75903930)
+260 train 6.183492 (lr=7.3007e-04) (hash(x)=82996853)
+270 train 6.115453 (lr=7.5804e-04) (hash(x)=73269568)
+280 train 6.062278 (lr=7.8601e-04) (hash(x)=82906427)
+290 train 5.983430 (lr=8.1399e-04) (hash(x)=71797895)
+300 val loss 6.1122
+300 val perplexity 451.3156
+300 train 5.840255 (lr=8.4196e-04) (hash(x)=85232249)
+310 train 5.794100 (lr=8.6993e-04) (hash(x)=75252489)
+320 train 5.873136 (lr=8.9790e-04) (hash(x)=74394644)
+330 train 5.946835 (lr=9.2587e-04) (hash(x)=82772910)
+340 train 5.983040 (lr=9.5385e-04) (hash(x)=81627464)
+350 train 5.963180 (lr=9.8182e-04) (hash(x)=84678053)
+360 train 5.917547 (lr=1.0098e-03) (hash(x)=81884128)
+370 train 5.900195 (lr=1.0378e-03) (hash(x)=76379242)
+380 train 5.673262 (lr=1.0657e-03) (hash(x)=98182875)
+390 train 5.670824 (lr=1.0937e-03) (hash(x)=79710436)
+400 val loss 5.7162
+400 val perplexity 303.7545
+400 train 5.669680 (lr=1.1217e-03) (hash(x)=79841071)
+410 train 5.592827 (lr=1.1497e-03) (hash(x)=75844151)
+420 train 5.591699 (lr=1.1776e-03) (hash(x)=73125036)
+430 train 5.628320 (lr=1.2056e-03) (hash(x)=84214858)
+440 train 5.518014 (lr=1.2336e-03) (hash(x)=80456994)
+450 train 5.482885 (lr=1.2615e-03) (hash(x)=76735962)
+460 train 5.422392 (lr=1.2895e-03) (hash(x)=81845446)
+470 train 5.234500 (lr=1.3175e-03) (hash(x)=76094689)
+480 train 5.356008 (lr=1.3455e-03) (hash(x)=83806686)
+490 train 5.188223 (lr=1.3734e-03) (hash(x)=84690227)
+500 val loss 5.4936
+500 val perplexity 243.1230
+500 train 5.256655 (lr=1.4014e-03) (hash(x)=71851938)
+510 train 5.342268 (lr=1.4294e-03) (hash(x)=77159346)
+520 train 5.381430 (lr=1.4573e-03) (hash(x)=80755753)
+530 train 5.488706 (lr=1.4853e-03) (hash(x)=75379679)
+540 train 5.385969 (lr=1.5133e-03) (hash(x)=82458619)
+550 train 5.379599 (lr=1.5413e-03) (hash(x)=86366396)
+560 train 5.350755 (lr=1.5692e-03) (hash(x)=85095044)
+570 train 5.336123 (lr=1.5972e-03) (hash(x)=78385159)
+580 train 5.273055 (lr=1.6252e-03) (hash(x)=79342394)
+590 train 5.297306 (lr=1.6531e-03) (hash(x)=70782192)
+600 val loss 5.2783
+600 val perplexity 196.0441
+600 train 5.259060 (lr=1.6811e-03) (hash(x)=76130353)
+610 train 5.160896 (lr=1.7091e-03) (hash(x)=74778440)
+620 train 5.143505 (lr=1.7371e-03) (hash(x)=79129709)
+630 train 5.043120 (lr=1.7650e-03) (hash(x)=76469962)
+640 train 5.053054 (lr=1.7930e-03) (hash(x)=78288049)
+650 train 5.012689 (lr=1.8210e-03) (hash(x)=76641388)
+660 train 4.910568 (lr=1.8490e-03) (hash(x)=75937906)
+670 train 4.922427 (lr=1.8769e-03) (hash(x)=74807157)
+680 train 4.830317 (lr=1.9049e-03) (hash(x)=77490144)
+690 train 4.861750 (lr=1.9329e-03) (hash(x)=72900124)
+700 val loss 5.1221
+700 val perplexity 167.6951
+700 train 4.820829 (lr=1.9608e-03) (hash(x)=73424218)
+710 train 5.090122 (lr=1.9888e-03) (hash(x)=86052345)
+720 train 5.051780 (lr=2.0000e-03) (hash(x)=77462613)
+730 train 5.061331 (lr=1.9999e-03) (hash(x)=79027471)
+740 train 5.037912 (lr=1.9998e-03) (hash(x)=78149992)
+750 train 5.017835 (lr=1.9996e-03) (hash(x)=89147499)
+760 train 5.001096 (lr=1.9993e-03) (hash(x)=74931538)
+770 train 4.955994 (lr=1.9990e-03) (hash(x)=89648204)
+780 train 4.990937 (lr=1.9986e-03) (hash(x)=83708293)
+790 train 4.983702 (lr=1.9981e-03) (hash(x)=84336512)
+800 val loss 4.9564
+800 val perplexity 142.0809
+800 train 4.885051 (lr=1.9976e-03) (hash(x)=75025285)
+810 train 4.794215 (lr=1.9970e-03) (hash(x)=79436536)
+820 train 4.788100 (lr=1.9963e-03) (hash(x)=90467390)
+830 train 4.839355 (lr=1.9956e-03) (hash(x)=77292020)
+840 train 4.753167 (lr=1.9948e-03) (hash(x)=75568927)
+850 train 4.770946 (lr=1.9940e-03) (hash(x)=79671781)
+860 train 4.841572 (lr=1.9930e-03) (hash(x)=84280943)
+870 train 4.832933 (lr=1.9920e-03) (hash(x)=80901294)
+880 train 4.960783 (lr=1.9910e-03) (hash(x)=81437584)
+890 train 4.802401 (lr=1.9899e-03) (hash(x)=73893778)
+900 val loss 4.8102
+900 val perplexity 122.7548
+900 train 4.737334 (lr=1.9887e-03) (hash(x)=74987794)
+910 train 4.747663 (lr=1.9874e-03) (hash(x)=93721374)
+920 train 4.731197 (lr=1.9861e-03) (hash(x)=79149678)
+930 train 4.684422 (lr=1.9847e-03) (hash(x)=83179387)
+940 train 4.686351 (lr=1.9833e-03) (hash(x)=89009978)
+950 train 4.654570 (lr=1.9818e-03) (hash(x)=76483735)
+960 train 4.586953 (lr=1.9802e-03) (hash(x)=78638850)
+970 train 4.614932 (lr=1.9785e-03) (hash(x)=90415206)
+980 train 4.548272 (lr=1.9768e-03) (hash(x)=90819195)
+990 train 4.686238 (lr=1.9750e-03) (hash(x)=85938852)
+1000 val loss 4.6626
+1000 val perplexity 105.9062
+1000 train 4.711349 (lr=1.9732e-03) (hash(x)=82436789)
+1010 train 4.747443 (lr=1.9713e-03) (hash(x)=63022149)
+1020 train 4.669138 (lr=1.9693e-03) (hash(x)=83501199)
+1030 train 4.650716 (lr=1.9673e-03) (hash(x)=77093326)
+1040 train 4.525937 (lr=1.9652e-03) (hash(x)=73437559)
+1050 train 4.610815 (lr=1.9630e-03) (hash(x)=84550388)
+1060 train 4.617441 (lr=1.9608e-03) (hash(x)=86184566)
+1070 train 4.598150 (lr=1.9585e-03) (hash(x)=85644922)
+1080 train 4.531865 (lr=1.9562e-03) (hash(x)=86179801)
+1090 train 4.505293 (lr=1.9538e-03) (hash(x)=82868303)
+1100 val loss 4.5696
+1100 val perplexity 96.5059
+1100 train 4.596138 (lr=1.9513e-03) (hash(x)=96780388)
+1110 train 4.528399 (lr=1.9488e-03) (hash(x)=87223122)
+1120 train 4.500387 (lr=1.9462e-03) (hash(x)=77292786)
+1130 train 4.525281 (lr=1.9435e-03) (hash(x)=88761375)
+1140 train 4.437323 (lr=1.9408e-03) (hash(x)=80956468)
+1150 train 4.607018 (lr=1.9380e-03) (hash(x)=84725462)
+1160 train 4.587046 (lr=1.9351e-03) (hash(x)=76098113)
+1170 train 4.567883 (lr=1.9322e-03) (hash(x)=84387685)
+1180 train 4.509290 (lr=1.9293e-03) (hash(x)=76120321)
+1190 train 4.470008 (lr=1.9262e-03) (hash(x)=85974065)
+1200 val loss 4.4936
+1200 val perplexity 89.4414
+1200 train 4.378562 (lr=1.9231e-03) (hash(x)=79340644)
+1210 train 4.466261 (lr=1.9200e-03) (hash(x)=88345025)
+1220 train 4.386754 (lr=1.9168e-03) (hash(x)=86406230)
+1230 train 4.368605 (lr=1.9135e-03) (hash(x)=80608975)
+1240 train 4.415116 (lr=1.9102e-03) (hash(x)=81635225)
+1250 train 4.258708 (lr=1.9068e-03) (hash(x)=82126376)
+1260 train 4.426104 (lr=1.9033e-03) (hash(x)=83737972)
+1270 train 4.333998 (lr=1.8998e-03) (hash(x)=81132345)
+1280 train 4.427336 (lr=1.8962e-03) (hash(x)=85850404)
+1290 train 4.433538 (lr=1.8926e-03) (hash(x)=83490640)
+1300 val loss 4.4413
+1300 val perplexity 84.8834
+1300 train 4.414694 (lr=1.8889e-03) (hash(x)=77545187)
+1310 train 4.460745 (lr=1.8852e-03) (hash(x)=86412685)
+1320 train 4.352343 (lr=1.8813e-03) (hash(x)=86429640)
+1330 train 4.498969 (lr=1.8775e-03) (hash(x)=79645281)
+1340 train 4.446202 (lr=1.8736e-03) (hash(x)=82668541)
+1350 train 4.308775 (lr=1.8696e-03) (hash(x)=73627248)
+1360 train 4.403217 (lr=1.8656e-03) (hash(x)=74845133)
+1370 train 4.309646 (lr=1.8615e-03) (hash(x)=77850497)
+1380 train 4.333172 (lr=1.8573e-03) (hash(x)=76143954)
+1390 train 4.347454 (lr=1.8531e-03) (hash(x)=80675544)
+1400 val loss 4.4110
+1400 val perplexity 82.3496
+1400 train 4.325371 (lr=1.8489e-03) (hash(x)=76010938)
+1410 train 4.371292 (lr=1.8445e-03) (hash(x)=81260300)
+1420 train 4.486557 (lr=1.8402e-03) (hash(x)=83803629)
+1430 train 4.311521 (lr=1.8358e-03) (hash(x)=83840110)
+1440 train 4.504490 (lr=1.8313e-03) (hash(x)=97711831)
+1450 train 4.381348 (lr=1.8268e-03) (hash(x)=81110139)
+1460 train 4.306368 (lr=1.8222e-03) (hash(x)=90942839)
+1470 train 4.338898 (lr=1.8175e-03) (hash(x)=80721564)
+1480 train 4.436263 (lr=1.8128e-03) (hash(x)=77852059)
+1490 train 4.326926 (lr=1.8081e-03) (hash(x)=80011365)
+1500 val loss 4.3650
+1500 val perplexity 78.6458
+1500 train 4.358942 (lr=1.8033e-03) (hash(x)=84496142)
+1510 train 4.288649 (lr=1.7985e-03) (hash(x)=70186729)
+1520 train 4.313070 (lr=1.7936e-03) (hash(x)=74854227)
+1530 train 4.224393 (lr=1.7886e-03) (hash(x)=77468161)
+1540 train 4.234926 (lr=1.7836e-03) (hash(x)=87853059)
+1550 train 4.169269 (lr=1.7786e-03) (hash(x)=71225436)
+1560 train 4.275188 (lr=1.7735e-03) (hash(x)=84072783)
+1570 train 4.320570 (lr=1.7683e-03) (hash(x)=72723098)
+1580 train 4.335917 (lr=1.7631e-03) (hash(x)=90409866)
+1590 train 4.417427 (lr=1.7579e-03) (hash(x)=87481378)
+1600 val loss 4.3356
+1600 val perplexity 76.3741
+1600 train 4.395532 (lr=1.7526e-03) (hash(x)=77643862)
+1610 train 4.283550 (lr=1.7472e-03) (hash(x)=88862575)
+1620 train 4.299943 (lr=1.7418e-03) (hash(x)=84612581)
+1630 train 4.331640 (lr=1.7364e-03) (hash(x)=87075989)
+1640 train 4.219054 (lr=1.7309e-03) (hash(x)=88277361)
+1650 train 4.333176 (lr=1.7254e-03) (hash(x)=78750236)
+1660 train 4.263692 (lr=1.7198e-03) (hash(x)=82604581)
+1670 train 4.094774 (lr=1.7142e-03) (hash(x)=68482265)
+1680 train 4.219302 (lr=1.7085e-03) (hash(x)=75088835)
+1690 train 4.161540 (lr=1.7028e-03) (hash(x)=66060989)
+1700 val loss 4.3351
+1700 val perplexity 76.3307
+1700 train 4.327582 (lr=1.6970e-03) (hash(x)=79986754)
+1710 train 4.313794 (lr=1.6912e-03) (hash(x)=83657930)
+1720 train 4.252811 (lr=1.6854e-03) (hash(x)=81754135)
+1730 train 4.339839 (lr=1.6795e-03) (hash(x)=78904427)
+1740 train 4.242384 (lr=1.6736e-03) (hash(x)=85920177)
+1750 train 4.293923 (lr=1.6676e-03) (hash(x)=86573211)
+1760 train 4.249685 (lr=1.6616e-03) (hash(x)=81737128)
+1770 train 4.240215 (lr=1.6555e-03) (hash(x)=77400968)
+1780 train 4.255862 (lr=1.6494e-03) (hash(x)=73545497)
+1790 train 4.501863 (lr=1.6433e-03) (hash(x)=71641943)
+1800 val loss 4.2889
+1800 val perplexity 72.8868
+1800 train 4.352567 (lr=1.6371e-03) (hash(x)=87819781)
+1810 train 4.204506 (lr=1.6309e-03) (hash(x)=86870770)
+1820 train 4.216088 (lr=1.6246e-03) (hash(x)=82522211)
+1830 train 4.299448 (lr=1.6183e-03) (hash(x)=61947437)
+1840 train 4.182921 (lr=1.6120e-03) (hash(x)=79865406)
+1850 train 4.147946 (lr=1.6056e-03) (hash(x)=79828721)
+1860 train 4.287210 (lr=1.5992e-03) (hash(x)=80869571)
+1870 train 4.274639 (lr=1.5928e-03) (hash(x)=73780971)
+1880 train 4.236541 (lr=1.5863e-03) (hash(x)=79249549)
+1890 train 4.300196 (lr=1.5798e-03) (hash(x)=81041904)
+1900 val loss 4.2645
+1900 val perplexity 71.1314
+1900 train 4.364532 (lr=1.5732e-03) (hash(x)=82456430)
+1910 train 4.127520 (lr=1.5666e-03) (hash(x)=82222135)
+1920 train 4.167559 (lr=1.5600e-03) (hash(x)=70033249)
+1930 train 4.153447 (lr=1.5533e-03) (hash(x)=72887360)
+1940 train 4.201384 (lr=1.5467e-03) (hash(x)=83251100)
+1950 train 4.237510 (lr=1.5399e-03) (hash(x)=79660266)
+1960 train 4.119651 (lr=1.5332e-03) (hash(x)=78919068)
+1970 train 4.216429 (lr=1.5264e-03) (hash(x)=94690431)
+1980 train 4.052589 (lr=1.5196e-03) (hash(x)=88153756)
+1990 train 4.112547 (lr=1.5127e-03) (hash(x)=77195688)
+2000 val loss 4.2406
+2000 val perplexity 69.4492
+2000 train 4.205284 (lr=1.5058e-03) (hash(x)=81308591)
+2010 train 4.257005 (lr=1.4989e-03) (hash(x)=80205479)
+2020 train 4.169465 (lr=1.4920e-03) (hash(x)=81008704)
+2030 train 4.237891 (lr=1.4850e-03) (hash(x)=82914358)
+2040 train 4.245559 (lr=1.4780e-03) (hash(x)=88064399)
+2050 train 4.240659 (lr=1.4710e-03) (hash(x)=82889390)
+2060 train 4.315722 (lr=1.4639e-03) (hash(x)=80913554)
+2070 train 4.161486 (lr=1.4569e-03) (hash(x)=78121791)
+2080 train 4.293556 (lr=1.4498e-03) (hash(x)=75356657)
+2090 train 4.120970 (lr=1.4426e-03) (hash(x)=74048412)
+2100 val loss 4.2186
+2100 val perplexity 67.9363
+2100 train 4.042845 (lr=1.4355e-03) (hash(x)=68928225)
+2110 train 4.099514 (lr=1.4283e-03) (hash(x)=93734745)
+2120 train 4.090335 (lr=1.4211e-03) (hash(x)=72878164)
+2130 train 4.216095 (lr=1.4139e-03) (hash(x)=71508226)
+2140 train 4.178213 (lr=1.4066e-03) (hash(x)=84168671)
+2150 train 4.116589 (lr=1.3993e-03) (hash(x)=74673639)
+2160 train 4.256698 (lr=1.3920e-03) (hash(x)=75470031)
+2170 train 4.224009 (lr=1.3847e-03) (hash(x)=74307890)
+2180 train 4.333496 (lr=1.3774e-03) (hash(x)=77214245)
+2190 train 4.115132 (lr=1.3700e-03) (hash(x)=88628359)
+2200 val loss 4.1990
+2200 val perplexity 66.6176
+2200 train 4.135565 (lr=1.3626e-03) (hash(x)=74779126)
+2210 train 4.168253 (lr=1.3552e-03) (hash(x)=79817976)
+2220 train 4.123131 (lr=1.3478e-03) (hash(x)=75258996)
+2230 train 4.095799 (lr=1.3404e-03) (hash(x)=88164047)
+2240 train 4.078424 (lr=1.3329e-03) (hash(x)=78580686)
+2250 train 4.017454 (lr=1.3255e-03) (hash(x)=76510617)
+2260 train 4.095739 (lr=1.3180e-03) (hash(x)=76043966)
+2270 train 4.074729 (lr=1.3105e-03) (hash(x)=73062098)
+2280 train 4.297091 (lr=1.3030e-03) (hash(x)=81885909)
+2290 train 4.230886 (lr=1.2954e-03) (hash(x)=81481741)
+2300 val loss 4.1751
+2300 val perplexity 65.0454
+2300 train 4.092732 (lr=1.2879e-03) (hash(x)=82104275)
+2310 train 4.145487 (lr=1.2803e-03) (hash(x)=82137309)
+2320 train 4.124151 (lr=1.2727e-03) (hash(x)=86741167)
+2330 train 4.142951 (lr=1.2652e-03) (hash(x)=84796263)
+2340 train 4.103671 (lr=1.2576e-03) (hash(x)=169966529)
+2350 train 3.981705 (lr=1.2499e-03) (hash(x)=80006061)
+2360 train 4.042752 (lr=1.2423e-03) (hash(x)=71090999)
+2370 train 4.170042 (lr=1.2347e-03) (hash(x)=86641850)
+2380 train 4.070052 (lr=1.2270e-03) (hash(x)=84573253)
+2390 train 4.098938 (lr=1.2194e-03) (hash(x)=74483764)
+2400 val loss 4.1648
+2400 val perplexity 64.3780
+2400 train 4.072681 (lr=1.2117e-03) (hash(x)=78327659)
+2410 train 4.138642 (lr=1.2041e-03) (hash(x)=77273627)
+2420 train 4.004621 (lr=1.1964e-03) (hash(x)=76938049)
+2430 train 4.011081 (lr=1.1887e-03) (hash(x)=98449442)
+2440 train 3.869775 (lr=1.1810e-03) (hash(x)=78157797)
+2450 train 3.908055 (lr=1.1733e-03) (hash(x)=80637582)
+2460 train 3.900242 (lr=1.1656e-03) (hash(x)=80023854)
+2470 train 3.966435 (lr=1.1579e-03) (hash(x)=86678884)
+2480 train 4.191215 (lr=1.1502e-03) (hash(x)=84086469)
+2490 train 4.208256 (lr=1.1425e-03) (hash(x)=83131141)
+2500 val loss 4.1437
+2500 val perplexity 63.0367
+2500 train 4.122083 (lr=1.1348e-03) (hash(x)=82583497)
+2510 train 4.182195 (lr=1.1270e-03) (hash(x)=79164326)
+2520 train 4.089431 (lr=1.1193e-03) (hash(x)=76456503)
+2530 train 4.147231 (lr=1.1116e-03) (hash(x)=82001228)
+2540 train 4.180750 (lr=1.1039e-03) (hash(x)=75615595)
+2550 train 4.129670 (lr=1.0961e-03) (hash(x)=78397869)
+2560 train 4.130214 (lr=1.0884e-03) (hash(x)=75547032)
+2570 train 4.045819 (lr=1.0807e-03) (hash(x)=89201025)
+2580 train 4.029406 (lr=1.0730e-03) (hash(x)=89856704)
+2590 train 4.040164 (lr=1.0652e-03) (hash(x)=82175682)
+2600 val loss 4.1264
+2600 val perplexity 61.9522
+2600 train 4.047859 (lr=1.0575e-03) (hash(x)=78312826)
+2610 train 4.034362 (lr=1.0498e-03) (hash(x)=77066588)
+2620 train 3.876209 (lr=1.0421e-03) (hash(x)=78666061)
+2630 train 3.877981 (lr=1.0344e-03) (hash(x)=93762143)
+2640 train 3.889488 (lr=1.0267e-03) (hash(x)=83191587)
+2650 train 3.815217 (lr=1.0190e-03) (hash(x)=87169585)
+2660 train 4.033377 (lr=1.0113e-03) (hash(x)=86426388)
+2670 train 4.172971 (lr=1.0036e-03) (hash(x)=76692638)
+2680 train 4.085076 (lr=9.9594e-04) (hash(x)=77446063)
+2690 train 4.086094 (lr=9.8827e-04) (hash(x)=79809050)
+2700 val loss 4.1247
+2700 val perplexity 61.8476
+2700 train 4.015488 (lr=9.8061e-04) (hash(x)=83116823)
+2710 train 4.128306 (lr=9.7296e-04) (hash(x)=75622148)
+2720 train 4.057120 (lr=9.6532e-04) (hash(x)=80690512)
+2730 train 4.126090 (lr=9.5768e-04) (hash(x)=78009984)
+2740 train 4.075282 (lr=9.5006e-04) (hash(x)=74205488)
+2750 train 4.084890 (lr=9.4245e-04) (hash(x)=91013332)
+2760 train 4.005525 (lr=9.3485e-04) (hash(x)=76954961)
+2770 train 4.016811 (lr=9.2726e-04) (hash(x)=79390317)
+2780 train 3.886596 (lr=9.1969e-04) (hash(x)=70168783)
+2790 train 4.047701 (lr=9.1212e-04) (hash(x)=76028417)
+2800 val loss 4.1018
+2800 val perplexity 60.4476
+2800 train 3.873746 (lr=9.0458e-04) (hash(x)=77656050)
+2810 train 3.917989 (lr=8.9704e-04) (hash(x)=94248216)
+2820 train 3.984176 (lr=8.8952e-04) (hash(x)=78305078)
+2830 train 3.893653 (lr=8.8202e-04) (hash(x)=79948848)
+2840 train 3.760536 (lr=8.7453e-04) (hash(x)=85341024)
+2850 train 4.068945 (lr=8.6706e-04) (hash(x)=78735170)
+2860 train 4.313912 (lr=8.5961e-04) (hash(x)=71616419)
+2870 train 4.094830 (lr=8.5217e-04) (hash(x)=78656517)
+2880 train 4.102984 (lr=8.4476e-04) (hash(x)=80073987)
+2890 train 4.029589 (lr=8.3736e-04) (hash(x)=76894809)
+2900 val loss 4.0770
+2900 val perplexity 58.9687
+2900 train 4.010001 (lr=8.2998e-04) (hash(x)=80499838)
+2910 train 4.062317 (lr=8.2262e-04) (hash(x)=72673354)
+2920 train 4.046411 (lr=8.1528e-04) (hash(x)=84265768)
+2930 train 4.069248 (lr=8.0796e-04) (hash(x)=79612060)
+2940 train 3.927702 (lr=8.0067e-04) (hash(x)=74970087)
+2950 train 4.082783 (lr=7.9339e-04) (hash(x)=84166818)
+2960 train 4.066880 (lr=7.8614e-04) (hash(x)=89410221)
+2970 train 4.010108 (lr=7.7891e-04) (hash(x)=75672566)
+2980 train 3.865549 (lr=7.7171e-04) (hash(x)=81760314)
+2990 train 3.840216 (lr=7.6453e-04) (hash(x)=80605200)
+3000 val loss 4.0757
+3000 val perplexity 58.8896
+3000 train 3.833838 (lr=7.5737e-04) (hash(x)=83804735)
+3010 train 3.762861 (lr=7.5024e-04) (hash(x)=77015303)
+3020 train 3.829176 (lr=7.4313e-04) (hash(x)=81464523)
+3030 train 4.027744 (lr=7.3606e-04) (hash(x)=80168230)
+3040 train 3.973993 (lr=7.2900e-04) (hash(x)=84817006)
+3050 train 4.080138 (lr=7.2198e-04) (hash(x)=71601811)
+3060 train 4.001876 (lr=7.1498e-04) (hash(x)=85499733)
+3070 train 4.054987 (lr=7.0801e-04) (hash(x)=78670408)
+3080 train 3.948816 (lr=7.0107e-04) (hash(x)=77120468)
+3090 train 4.054901 (lr=6.9416e-04) (hash(x)=77927426)
+3100 val loss 4.0558
+3100 val perplexity 57.7295
+3100 train 4.068103 (lr=6.8728e-04) (hash(x)=83998606)
+3110 train 3.963448 (lr=6.8043e-04) (hash(x)=77990218)
+3120 train 3.884058 (lr=6.7362e-04) (hash(x)=81623970)
+3130 train 3.853804 (lr=6.6683e-04) (hash(x)=75014781)
+3140 train 3.913841 (lr=6.6007e-04) (hash(x)=72591250)
+3150 train 3.921466 (lr=6.5335e-04) (hash(x)=81421847)
+3160 train 3.973674 (lr=6.4666e-04) (hash(x)=75112631)
+3170 train 4.071921 (lr=6.4000e-04) (hash(x)=87518033)
+3180 train 3.918107 (lr=6.3338e-04) (hash(x)=78648348)
+3190 train 4.041168 (lr=6.2679e-04) (hash(x)=78270029)
+3200 val loss 4.0374
+3200 val perplexity 56.6781
+3200 train 4.112720 (lr=6.2024e-04) (hash(x)=83380714)
+3210 train 4.297076 (lr=6.1372e-04) (hash(x)=67804991)
+3220 train 4.024940 (lr=6.0724e-04) (hash(x)=76968804)
+3230 train 3.959510 (lr=6.0079e-04) (hash(x)=75798670)
+3240 train 4.033435 (lr=5.9438e-04) (hash(x)=83192811)
+3250 train 3.971035 (lr=5.8801e-04) (hash(x)=81659789)
+3260 train 4.011578 (lr=5.8167e-04) (hash(x)=80622502)
+3270 train 3.983234 (lr=5.7538e-04) (hash(x)=85436511)
+3280 train 3.919018 (lr=5.6912e-04) (hash(x)=80589180)
+3290 train 4.064373 (lr=5.6290e-04) (hash(x)=85747193)
+3300 val loss 4.0338
+3300 val perplexity 56.4738
+3300 train 3.928417 (lr=5.5672e-04) (hash(x)=74635692)
+3310 train 3.916089 (lr=5.5058e-04) (hash(x)=82045455)
+3320 train 3.791219 (lr=5.4449e-04) (hash(x)=73249173)
+3330 train 3.746655 (lr=5.3843e-04) (hash(x)=71492338)
+3340 train 4.078082 (lr=5.3241e-04) (hash(x)=76374071)
+3350 train 4.046371 (lr=5.2644e-04) (hash(x)=80474064)
+3360 train 3.961680 (lr=5.2051e-04) (hash(x)=84390892)
+3370 train 4.011613 (lr=5.1462e-04) (hash(x)=83399949)
+3380 train 4.056945 (lr=5.0877e-04) (hash(x)=77648059)
+3390 train 4.028938 (lr=5.0297e-04) (hash(x)=80152701)
+3400 val loss 4.0148
+3400 val perplexity 55.4130
+3400 train 4.075192 (lr=4.9721e-04) (hash(x)=81952545)
+3410 train 4.041352 (lr=4.9149e-04) (hash(x)=80908993)
+3420 train 4.080109 (lr=4.8582e-04) (hash(x)=79610037)
+3430 train 3.951145 (lr=4.8020e-04) (hash(x)=87624382)
+3440 train 4.006751 (lr=4.7462e-04) (hash(x)=82336381)
+3450 train 3.981426 (lr=4.6909e-04) (hash(x)=87687835)
+3460 train 3.984341 (lr=4.6360e-04) (hash(x)=76376135)
+3470 train 3.896631 (lr=4.5816e-04) (hash(x)=79331391)
+3480 train 3.803572 (lr=4.5277e-04) (hash(x)=107002681)
+3490 train 3.881458 (lr=4.4743e-04) (hash(x)=77465514)
+3500 val loss 4.0119
+3500 val perplexity 55.2491
+3500 train 3.850668 (lr=4.4213e-04) (hash(x)=88237229)
+3510 train 3.916492 (lr=4.3689e-04) (hash(x)=81011739)
+3520 train 3.961740 (lr=4.3169e-04) (hash(x)=64643427)
+3530 train 4.241804 (lr=4.2654e-04) (hash(x)=78029539)
+3540 train 4.067226 (lr=4.2144e-04) (hash(x)=83188968)
+3550 train 4.093814 (lr=4.1639e-04) (hash(x)=86104185)
+3560 train 4.017192 (lr=4.1139e-04) (hash(x)=81993629)
+3570 train 3.989879 (lr=4.0644e-04) (hash(x)=72719368)
+3580 train 4.001433 (lr=4.0154e-04) (hash(x)=90091487)
+3590 train 3.885998 (lr=3.9670e-04) (hash(x)=77393152)
+3600 val loss 3.9942
+3600 val perplexity 54.2806
+3600 train 3.994248 (lr=3.9190e-04) (hash(x)=73103504)
+3610 train 4.000402 (lr=3.8716e-04) (hash(x)=84429400)
+3620 train 3.977069 (lr=3.8247e-04) (hash(x)=77021795)
+3630 train 3.925595 (lr=3.7783e-04) (hash(x)=92830605)
+3640 train 4.124040 (lr=3.7325e-04) (hash(x)=78313175)
+3650 train 3.924363 (lr=3.6872e-04) (hash(x)=85395549)
+3660 train 3.717732 (lr=3.6424e-04) (hash(x)=78114459)
+3670 train 3.846315 (lr=3.5982e-04) (hash(x)=74968316)
+3680 train 3.768297 (lr=3.5545e-04) (hash(x)=73358737)
+3690 train 3.877346 (lr=3.5114e-04) (hash(x)=76399442)
+3700 val loss 3.9953
+3700 val perplexity 54.3400
+3700 train 3.878586 (lr=3.4688e-04) (hash(x)=74039273)
+3710 train 4.052510 (lr=3.4268e-04) (hash(x)=79299680)
+3720 train 3.968278 (lr=3.3854e-04) (hash(x)=94701498)
+3730 train 4.049542 (lr=3.3445e-04) (hash(x)=75352071)
+3740 train 3.932052 (lr=3.3041e-04) (hash(x)=77909487)
+3750 train 4.036614 (lr=3.2644e-04) (hash(x)=73986730)
+3760 train 3.994642 (lr=3.2252e-04) (hash(x)=79325763)
+3770 train 4.039020 (lr=3.1865e-04) (hash(x)=72457818)
+3780 train 4.167808 (lr=3.1485e-04) (hash(x)=71775590)
+3790 train 4.063184 (lr=3.1110e-04) (hash(x)=82638943)
+3800 val loss 3.9797
+3800 val perplexity 53.5036
+3800 train 4.034631 (lr=3.0741e-04) (hash(x)=79965893)
+3810 train 3.967500 (lr=3.0378e-04) (hash(x)=72598235)
+3820 train 3.950491 (lr=3.0021e-04) (hash(x)=83113889)
+3830 train 4.011137 (lr=2.9670e-04) (hash(x)=74434590)
+3840 train 3.929249 (lr=2.9324e-04) (hash(x)=82860348)
+3850 train 3.860030 (lr=2.8985e-04) (hash(x)=78067565)
+3860 train 3.932343 (lr=2.8651e-04) (hash(x)=82592498)
+3870 train 3.866020 (lr=2.8324e-04) (hash(x)=81820733)
+3880 train 3.882616 (lr=2.8002e-04) (hash(x)=87709040)
+3890 train 3.927842 (lr=2.7687e-04) (hash(x)=70379093)
+3900 val loss 3.9745
+3900 val perplexity 53.2225
+3900 train 4.049213 (lr=2.7378e-04) (hash(x)=76597431)
+3910 train 3.936760 (lr=2.7074e-04) (hash(x)=90490716)
+3920 train 4.036233 (lr=2.6777e-04) (hash(x)=81970659)
+3930 train 4.024949 (lr=2.6486e-04) (hash(x)=81496334)
+3940 train 3.998729 (lr=2.6201e-04) (hash(x)=75717605)
+3950 train 3.946081 (lr=2.5923e-04) (hash(x)=82749357)
+3960 train 3.972913 (lr=2.5650e-04) (hash(x)=89355157)
+3970 train 3.931072 (lr=2.5384e-04) (hash(x)=78980403)
+3980 train 3.900640 (lr=2.5124e-04) (hash(x)=76627217)
+3990 train 3.904887 (lr=2.4870e-04) (hash(x)=72412879)
+4000 val loss 3.9684
+4000 val perplexity 52.9000
+4000 train 3.916356 (lr=2.4622e-04) (hash(x)=83018142)
+4010 train 3.928426 (lr=2.4381e-04) (hash(x)=81272436)
+4020 train 3.856344 (lr=2.4146e-04) (hash(x)=85497482)
+4030 train 3.931446 (lr=2.3918e-04) (hash(x)=85241734)
+4040 train 3.900393 (lr=2.3695e-04) (hash(x)=77925307)
+4050 train 3.897156 (lr=2.3479e-04) (hash(x)=84826179)
+4060 train 3.958513 (lr=2.3270e-04) (hash(x)=83606764)
+4070 train 3.872245 (lr=2.3067e-04) (hash(x)=80567590)
+4080 train 3.920675 (lr=2.2870e-04) (hash(x)=76860998)
+4090 train 3.958889 (lr=2.2680e-04) (hash(x)=74902328)
+4100 val loss 3.9585
+4100 val perplexity 52.3789
+4100 train 3.996765 (lr=2.2496e-04) (hash(x)=82832041)
+4110 train 3.882298 (lr=2.2318e-04) (hash(x)=79143262)
+4120 train 3.909158 (lr=2.2147e-04) (hash(x)=77038149)
+4130 train 4.091127 (lr=2.1983e-04) (hash(x)=86339074)
+4140 train 4.158182 (lr=2.1825e-04) (hash(x)=76686216)
+4150 train 3.932431 (lr=2.1673e-04) (hash(x)=70522682)
+4160 train 3.893674 (lr=2.1528e-04) (hash(x)=90958555)
+4170 train 3.918903 (lr=2.1390e-04) (hash(x)=91463532)
+4180 train 3.908016 (lr=2.1258e-04) (hash(x)=81959329)
+4190 train 3.867520 (lr=2.1132e-04) (hash(x)=83146752)
+4200 val loss 3.9611
+4200 val perplexity 52.5151
+4200 train 3.850982 (lr=2.1013e-04) (hash(x)=78361715)
+4210 train 4.002067 (lr=2.0901e-04) (hash(x)=87364889)
+4220 train 3.850836 (lr=2.0795e-04) (hash(x)=70465156)
+4230 train 4.006497 (lr=2.0696e-04) (hash(x)=84524081)
+4240 train 3.961092 (lr=2.0604e-04) (hash(x)=77824868)
+4250 train 3.975957 (lr=2.0518e-04) (hash(x)=81710711)
+4260 train 3.955402 (lr=2.0438e-04) (hash(x)=76362728)
+4270 train 3.941320 (lr=2.0365e-04) (hash(x)=83115208)
+4280 train 3.859817 (lr=2.0299e-04) (hash(x)=87218314)
+4290 train 4.017165 (lr=2.0239e-04) (hash(x)=74582673)
+4300 val loss 3.9554
+4300 val perplexity 52.2144
+4300 train 3.893432 (lr=2.0186e-04) (hash(x)=77379615)
+4310 train 3.867355 (lr=2.0140e-04) (hash(x)=78669579)
+4320 train 3.986726 (lr=2.0100e-04) (hash(x)=83066608)
+4330 train 3.953562 (lr=2.0067e-04) (hash(x)=83037340)
+4340 train 3.987670 (lr=2.0041e-04) (hash(x)=82849771)
+4350 train 3.929417 (lr=2.0021e-04) (hash(x)=76693985)
+4360 train 4.015263 (lr=2.0007e-04) (hash(x)=77745394)
+4370 train 3.973243 (lr=2.0001e-04) (hash(x)=79954388)
+4374 val loss 3.9497
+4374 val perplexity 51.9194

12_head_baseline_lr_20e-4_head_dim_22/model_04374.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:5e48315cff54ba6eee9d03e2a69887c619274a28bbfeff7289f89dd7eb4ad1d6
+size 96858242

12_head_baseline_lr_20e-4_head_dim_22/optimizer_04374.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:3aab757dd0ebfd884305e80b148673fa57aa08bac03b16059df008e248fa1f08
+size 187435910