Upload folder using huggingface_hub

12_head_baseline_lr_50e-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_50e-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": 0.005, "decay_lr": true, "readout_zero_init": false, "query_zero_init": false, "l1_loss": false, "debugpy": false, "key": "12_head_baseline_lr_50e-4"}

12_head_baseline_lr_50e-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_50e-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=6.9930e-06) (hash(x)=93356070)
+10 train 10.430300 (lr=7.6923e-05) (hash(x)=91253010)
+20 train 10.101633 (lr=1.4685e-04) (hash(x)=74934453)
+30 train 9.611134 (lr=2.1678e-04) (hash(x)=79278034)
+40 train 8.951025 (lr=2.8671e-04) (hash(x)=80107892)
+50 train 8.367085 (lr=3.5664e-04) (hash(x)=70745428)
+60 train 7.951948 (lr=4.2657e-04) (hash(x)=80077589)
+70 train 7.496622 (lr=4.9650e-04) (hash(x)=76213766)
+80 train 7.466031 (lr=5.6643e-04) (hash(x)=83218328)
+90 train 7.272826 (lr=6.3636e-04) (hash(x)=74855845)
+100 val loss 7.1531
+100 val perplexity 1278.0143
+100 train 7.004827 (lr=7.0629e-04) (hash(x)=82814902)
+110 train 6.915068 (lr=7.7622e-04) (hash(x)=84907741)
+120 train 6.776262 (lr=8.4615e-04) (hash(x)=82613223)
+130 train 6.601989 (lr=9.1608e-04) (hash(x)=83540876)
+140 train 6.566031 (lr=9.8601e-04) (hash(x)=75095216)
+150 train 6.635382 (lr=1.0559e-03) (hash(x)=97190944)
+160 train 6.554130 (lr=1.1259e-03) (hash(x)=82117809)
+170 train 6.453229 (lr=1.1958e-03) (hash(x)=70514724)
+180 train 6.373549 (lr=1.2657e-03) (hash(x)=81029624)
+190 train 6.292207 (lr=1.3357e-03) (hash(x)=78996841)
+200 val loss 6.2431
+200 val perplexity 514.4500
+200 train 6.243793 (lr=1.4056e-03) (hash(x)=79845097)
+210 train 6.134615 (lr=1.4755e-03) (hash(x)=78997683)
+220 train 6.044861 (lr=1.5455e-03) (hash(x)=74895865)
+230 train 6.191950 (lr=1.6154e-03) (hash(x)=80933276)
+240 train 5.887405 (lr=1.6853e-03) (hash(x)=77664606)
+250 train 5.918988 (lr=1.7552e-03) (hash(x)=75903930)
+260 train 5.812324 (lr=1.8252e-03) (hash(x)=82996853)
+270 train 5.742345 (lr=1.8951e-03) (hash(x)=73269568)
+280 train 5.675274 (lr=1.9650e-03) (hash(x)=82906427)
+290 train 5.617868 (lr=2.0350e-03) (hash(x)=71797895)
+300 val loss 5.7644
+300 val perplexity 318.7496
+300 train 5.471639 (lr=2.1049e-03) (hash(x)=85232249)
+310 train 5.416192 (lr=2.1748e-03) (hash(x)=75252489)
+320 train 5.496908 (lr=2.2448e-03) (hash(x)=74394644)
+330 train 5.598850 (lr=2.3147e-03) (hash(x)=82772910)
+340 train 5.646424 (lr=2.3846e-03) (hash(x)=81627464)
+350 train 5.628429 (lr=2.4545e-03) (hash(x)=84678053)
+360 train 5.556225 (lr=2.5245e-03) (hash(x)=81884128)
+370 train 5.582003 (lr=2.5944e-03) (hash(x)=76379242)
+380 train 5.384250 (lr=2.6643e-03) (hash(x)=98182875)
+390 train 5.360508 (lr=2.7343e-03) (hash(x)=79710436)
+400 val loss 5.4087
+400 val perplexity 223.3511
+400 train 5.349107 (lr=2.8042e-03) (hash(x)=79841071)
+410 train 5.310011 (lr=2.8741e-03) (hash(x)=75844151)
+420 train 5.309967 (lr=2.9441e-03) (hash(x)=73125036)
+430 train 5.342200 (lr=3.0140e-03) (hash(x)=84214858)
+440 train 5.218863 (lr=3.0839e-03) (hash(x)=80456994)
+450 train 5.220237 (lr=3.1538e-03) (hash(x)=76735962)
+460 train 5.129281 (lr=3.2238e-03) (hash(x)=81845446)
+470 train 4.938518 (lr=3.2937e-03) (hash(x)=76094689)
+480 train 5.068999 (lr=3.3636e-03) (hash(x)=83806686)
+490 train 4.887512 (lr=3.4336e-03) (hash(x)=84690227)
+500 val loss 5.2417
+500 val perplexity 188.9929
+500 train 4.974317 (lr=3.5035e-03) (hash(x)=71851938)
+510 train 5.081150 (lr=3.5734e-03) (hash(x)=77159346)
+520 train 5.132569 (lr=3.6434e-03) (hash(x)=80755753)
+530 train 5.244644 (lr=3.7133e-03) (hash(x)=75379679)
+540 train 5.153445 (lr=3.7832e-03) (hash(x)=82458619)
+550 train 5.119081 (lr=3.8531e-03) (hash(x)=86366396)
+560 train 5.152158 (lr=3.9231e-03) (hash(x)=85095044)
+570 train 5.101776 (lr=3.9930e-03) (hash(x)=78385159)
+580 train 5.036698 (lr=4.0629e-03) (hash(x)=79342394)
+590 train 5.066434 (lr=4.1329e-03) (hash(x)=70782192)
+600 val loss 5.0815
+600 val perplexity 161.0120
+600 train 5.057821 (lr=4.2028e-03) (hash(x)=76130353)
+610 train 4.962499 (lr=4.2727e-03) (hash(x)=74778440)
+620 train 4.927670 (lr=4.3427e-03) (hash(x)=79129709)
+630 train 4.835021 (lr=4.4126e-03) (hash(x)=76469962)
+640 train 4.827167 (lr=4.4825e-03) (hash(x)=78288049)
+650 train 4.770166 (lr=4.5524e-03) (hash(x)=76641388)
+660 train 4.746838 (lr=4.6224e-03) (hash(x)=75937906)
+670 train 4.706330 (lr=4.6923e-03) (hash(x)=74807157)
+680 train 4.589323 (lr=4.7622e-03) (hash(x)=77490144)
+690 train 4.622537 (lr=4.8322e-03) (hash(x)=72900124)
+700 val loss 4.8890
+700 val perplexity 132.8147
+700 train 4.579949 (lr=4.9021e-03) (hash(x)=73424218)
+710 train 4.872239 (lr=4.9720e-03) (hash(x)=86052345)
+720 train 4.863198 (lr=5.0000e-03) (hash(x)=77462613)
+730 train 4.872299 (lr=4.9998e-03) (hash(x)=79027471)
+740 train 4.815014 (lr=4.9995e-03) (hash(x)=78149992)
+750 train 4.746889 (lr=4.9990e-03) (hash(x)=89147499)
+760 train 4.803520 (lr=4.9983e-03) (hash(x)=74931538)
+770 train 4.732812 (lr=4.9975e-03) (hash(x)=89648204)
+780 train 4.789251 (lr=4.9965e-03) (hash(x)=83708293)
+790 train 4.779197 (lr=4.9953e-03) (hash(x)=84336512)
+800 val loss 4.7559
+800 val perplexity 116.2711
+800 train 4.675926 (lr=4.9940e-03) (hash(x)=75025285)
+810 train 4.601695 (lr=4.9925e-03) (hash(x)=79436536)
+820 train 4.598423 (lr=4.9909e-03) (hash(x)=90467390)
+830 train 4.642546 (lr=4.9890e-03) (hash(x)=77292020)
+840 train 4.559494 (lr=4.9871e-03) (hash(x)=75568927)
+850 train 4.589305 (lr=4.9849e-03) (hash(x)=79671781)
+860 train 4.673046 (lr=4.9826e-03) (hash(x)=84280943)
+870 train 4.661456 (lr=4.9801e-03) (hash(x)=80901294)
+880 train 4.779912 (lr=4.9775e-03) (hash(x)=81437584)
+890 train 4.630421 (lr=4.9747e-03) (hash(x)=73893778)
+900 val loss 4.6482
+900 val perplexity 104.3983
+900 train 4.581583 (lr=4.9717e-03) (hash(x)=74987794)
+910 train 4.606688 (lr=4.9686e-03) (hash(x)=93721374)
+920 train 4.583346 (lr=4.9653e-03) (hash(x)=79149678)
+930 train 4.545957 (lr=4.9618e-03) (hash(x)=83179387)
+940 train 4.540257 (lr=4.9582e-03) (hash(x)=89009978)
+950 train 4.534956 (lr=4.9544e-03) (hash(x)=76483735)
+960 train 4.471447 (lr=4.9504e-03) (hash(x)=78638850)
+970 train 4.498500 (lr=4.9463e-03) (hash(x)=90415206)
+980 train 4.445189 (lr=4.9420e-03) (hash(x)=90819195)
+990 train 4.573137 (lr=4.9376e-03) (hash(x)=85938852)
+1000 val loss 4.5595
+1000 val perplexity 95.5402
+1000 train 4.606597 (lr=4.9330e-03) (hash(x)=82436789)
+1010 train 4.669239 (lr=4.9283e-03) (hash(x)=63022149)
+1020 train 4.568455 (lr=4.9233e-03) (hash(x)=83501199)
+1030 train 4.553388 (lr=4.9183e-03) (hash(x)=77093326)
+1040 train 4.435065 (lr=4.9130e-03) (hash(x)=73437559)
+1050 train 4.523121 (lr=4.9076e-03) (hash(x)=84550388)
+1060 train 4.530206 (lr=4.9021e-03) (hash(x)=86184566)
+1070 train 4.514520 (lr=4.8963e-03) (hash(x)=85644922)
+1080 train 4.447129 (lr=4.8905e-03) (hash(x)=86179801)
+1090 train 4.429801 (lr=4.8844e-03) (hash(x)=82868303)
+1100 val loss 4.4971
+1100 val perplexity 89.7549
+1100 train 4.518186 (lr=4.8783e-03) (hash(x)=96780388)
+1110 train 4.460462 (lr=4.8719e-03) (hash(x)=87223122)
+1120 train 4.425699 (lr=4.8654e-03) (hash(x)=77292786)
+1130 train 4.456557 (lr=4.8587e-03) (hash(x)=88761375)
+1140 train 4.367574 (lr=4.8519e-03) (hash(x)=80956468)
+1150 train 4.541785 (lr=4.8450e-03) (hash(x)=84725462)
+1160 train 4.528299 (lr=4.8378e-03) (hash(x)=76098113)
+1170 train 4.499696 (lr=4.8306e-03) (hash(x)=84387685)
+1180 train 4.445394 (lr=4.8231e-03) (hash(x)=76120321)
+1190 train 4.412803 (lr=4.8156e-03) (hash(x)=85974065)
+1200 val loss 4.4306
+1200 val perplexity 83.9796
+1200 train 4.317427 (lr=4.8078e-03) (hash(x)=79340644)
+1210 train 4.405208 (lr=4.7999e-03) (hash(x)=88345025)
+1220 train 4.316737 (lr=4.7919e-03) (hash(x)=86406230)
+1230 train 4.306602 (lr=4.7837e-03) (hash(x)=80608975)
+1240 train 4.357062 (lr=4.7754e-03) (hash(x)=81635225)
+1250 train 4.194213 (lr=4.7669e-03) (hash(x)=82126376)
+1260 train 4.371879 (lr=4.7583e-03) (hash(x)=83737972)
+1270 train 4.287302 (lr=4.7495e-03) (hash(x)=81132345)
+1280 train 4.379058 (lr=4.7405e-03) (hash(x)=85850404)
+1290 train 4.387058 (lr=4.7315e-03) (hash(x)=83490640)
+1300 val loss 4.3955
+1300 val perplexity 81.0832
+1300 train 4.367643 (lr=4.7222e-03) (hash(x)=77545187)
+1310 train 4.416906 (lr=4.7129e-03) (hash(x)=86412685)
+1320 train 4.310229 (lr=4.7034e-03) (hash(x)=86429640)
+1330 train 4.442633 (lr=4.6937e-03) (hash(x)=79645281)
+1340 train 4.397045 (lr=4.6839e-03) (hash(x)=82668541)
+1350 train 4.256176 (lr=4.6740e-03) (hash(x)=73627248)
+1360 train 4.357805 (lr=4.6639e-03) (hash(x)=74845133)
+1370 train 4.271098 (lr=4.6537e-03) (hash(x)=77850497)
+1380 train 4.293922 (lr=4.6433e-03) (hash(x)=76143954)
+1390 train 4.308231 (lr=4.6328e-03) (hash(x)=80675544)
+1400 val loss 4.3755
+1400 val perplexity 79.4791
+1400 train 4.285866 (lr=4.6221e-03) (hash(x)=76010938)
+1410 train 4.335230 (lr=4.6114e-03) (hash(x)=81260300)
+1420 train 4.454363 (lr=4.6004e-03) (hash(x)=83803629)
+1430 train 4.273362 (lr=4.5894e-03) (hash(x)=83840110)
+1440 train 4.463538 (lr=4.5782e-03) (hash(x)=97711831)
+1450 train 4.340655 (lr=4.5669e-03) (hash(x)=81110139)
+1460 train 4.263830 (lr=4.5554e-03) (hash(x)=90942839)
+1470 train 4.304990 (lr=4.5438e-03) (hash(x)=80721564)
+1480 train 4.397264 (lr=4.5321e-03) (hash(x)=77852059)
+1490 train 4.293916 (lr=4.5202e-03) (hash(x)=80011365)
+1500 val loss 4.3331
+1500 val perplexity 76.1835
+1500 train 4.322277 (lr=4.5083e-03) (hash(x)=84496142)
+1510 train 4.259991 (lr=4.4961e-03) (hash(x)=70186729)
+1520 train 4.285318 (lr=4.4839e-03) (hash(x)=74854227)
+1530 train 4.190525 (lr=4.4715e-03) (hash(x)=77468161)
+1540 train 4.207886 (lr=4.4590e-03) (hash(x)=87853059)
+1550 train 4.142836 (lr=4.4464e-03) (hash(x)=71225436)
+1560 train 4.239394 (lr=4.4337e-03) (hash(x)=84072783)
+1570 train 4.294069 (lr=4.4208e-03) (hash(x)=72723098)
+1580 train 4.305942 (lr=4.4078e-03) (hash(x)=90409866)
+1590 train 4.387156 (lr=4.3947e-03) (hash(x)=87481378)
+1600 val loss 4.3110
+1600 val perplexity 74.5172
+1600 train 4.365888 (lr=4.3814e-03) (hash(x)=77643862)
+1610 train 4.259126 (lr=4.3681e-03) (hash(x)=88862575)
+1620 train 4.267086 (lr=4.3546e-03) (hash(x)=84612581)
+1630 train 4.308844 (lr=4.3410e-03) (hash(x)=87075989)
+1640 train 4.194880 (lr=4.3273e-03) (hash(x)=88277361)
+1650 train 4.306574 (lr=4.3134e-03) (hash(x)=78750236)
+1660 train 4.234919 (lr=4.2995e-03) (hash(x)=82604581)
+1670 train 4.076980 (lr=4.2854e-03) (hash(x)=68482265)
+1680 train 4.197222 (lr=4.2713e-03) (hash(x)=75088835)
+1690 train 4.140920 (lr=4.2570e-03) (hash(x)=66060989)
+1700 val loss 4.3141
+1700 val perplexity 74.7461
+1700 train 4.301929 (lr=4.2426e-03) (hash(x)=79986754)
+1710 train 4.292409 (lr=4.2281e-03) (hash(x)=83657930)
+1720 train 4.235638 (lr=4.2135e-03) (hash(x)=81754135)
+1730 train 4.326357 (lr=4.1987e-03) (hash(x)=78904427)
+1740 train 4.217617 (lr=4.1839e-03) (hash(x)=85920177)
+1750 train 4.274431 (lr=4.1690e-03) (hash(x)=86573211)
+1760 train 4.227898 (lr=4.1539e-03) (hash(x)=81737128)
+1770 train 4.222485 (lr=4.1388e-03) (hash(x)=77400968)
+1780 train 4.232876 (lr=4.1235e-03) (hash(x)=73545497)
+1790 train 4.483886 (lr=4.1082e-03) (hash(x)=71641943)
+1800 val loss 4.2705
+1800 val perplexity 71.5581
+1800 train 4.336700 (lr=4.0927e-03) (hash(x)=87819781)
+1810 train 4.183451 (lr=4.0772e-03) (hash(x)=86870770)
+1820 train 4.199247 (lr=4.0616e-03) (hash(x)=82522211)
+1830 train 4.282169 (lr=4.0458e-03) (hash(x)=61947437)
+1840 train 4.163460 (lr=4.0300e-03) (hash(x)=79865406)
+1850 train 4.127339 (lr=4.0140e-03) (hash(x)=79828721)
+1860 train 4.274527 (lr=3.9980e-03) (hash(x)=80869571)
+1870 train 4.251336 (lr=3.9819e-03) (hash(x)=73780971)
+1880 train 4.217798 (lr=3.9657e-03) (hash(x)=79249549)
+1890 train 4.278646 (lr=3.9494e-03) (hash(x)=81041904)
+1900 val loss 4.2519
+1900 val perplexity 70.2374
+1900 train 4.356471 (lr=3.9330e-03) (hash(x)=82456430)
+1910 train 4.110142 (lr=3.9166e-03) (hash(x)=82222135)
+1920 train 4.147755 (lr=3.9000e-03) (hash(x)=70033249)
+1930 train 4.132651 (lr=3.8834e-03) (hash(x)=72887360)
+1940 train 4.181321 (lr=3.8666e-03) (hash(x)=83251100)
+1950 train 4.219366 (lr=3.8498e-03) (hash(x)=79660266)
+1960 train 4.106945 (lr=3.8329e-03) (hash(x)=78919068)
+1970 train 4.202980 (lr=3.8160e-03) (hash(x)=94690431)
+1980 train 4.039069 (lr=3.7989e-03) (hash(x)=88153756)
+1990 train 4.093534 (lr=3.7818e-03) (hash(x)=77195688)
+2000 val loss 4.2265
+2000 val perplexity 68.4801
+2000 train 4.196152 (lr=3.7646e-03) (hash(x)=81308591)
+2010 train 4.241035 (lr=3.7473e-03) (hash(x)=80205479)
+2020 train 4.155095 (lr=3.7300e-03) (hash(x)=81008704)
+2030 train 4.223628 (lr=3.7125e-03) (hash(x)=82914358)
+2040 train 4.233630 (lr=3.6951e-03) (hash(x)=88064399)
+2050 train 4.244870 (lr=3.6775e-03) (hash(x)=82889390)
+2060 train 4.305655 (lr=3.6599e-03) (hash(x)=80913554)
+2070 train 4.147591 (lr=3.6422e-03) (hash(x)=78121791)
+2080 train 4.283460 (lr=3.6244e-03) (hash(x)=75356657)
+2090 train 4.106378 (lr=3.6066e-03) (hash(x)=74048412)
+2100 val loss 4.2043
+2100 val perplexity 66.9739
+2100 train 4.030020 (lr=3.5887e-03) (hash(x)=68928225)
+2110 train 4.080212 (lr=3.5707e-03) (hash(x)=93734745)
+2120 train 4.075750 (lr=3.5527e-03) (hash(x)=72878164)
+2130 train 4.204665 (lr=3.5346e-03) (hash(x)=71508226)
+2140 train 4.165468 (lr=3.5165e-03) (hash(x)=84168671)
+2150 train 4.101372 (lr=3.4983e-03) (hash(x)=74673639)
+2160 train 4.247788 (lr=3.4801e-03) (hash(x)=75470031)
+2170 train 4.206627 (lr=3.4618e-03) (hash(x)=74307890)
+2180 train 4.326437 (lr=3.4435e-03) (hash(x)=77214245)
+2190 train 4.105478 (lr=3.4251e-03) (hash(x)=88628359)
+2200 val loss 4.1806
+2200 val perplexity 65.4071
+2200 train 4.119265 (lr=3.4066e-03) (hash(x)=74779126)
+2210 train 4.158552 (lr=3.3881e-03) (hash(x)=79817976)
+2220 train 4.112025 (lr=3.3696e-03) (hash(x)=75258996)
+2230 train 4.084177 (lr=3.3510e-03) (hash(x)=88164047)
+2240 train 4.070524 (lr=3.3323e-03) (hash(x)=78580686)
+2250 train 4.006957 (lr=3.3137e-03) (hash(x)=76510617)
+2260 train 4.079580 (lr=3.2949e-03) (hash(x)=76043966)
+2270 train 4.057602 (lr=3.2762e-03) (hash(x)=73062098)
+2280 train 4.289168 (lr=3.2574e-03) (hash(x)=81885909)
+2290 train 4.224728 (lr=3.2386e-03) (hash(x)=81481741)
+2300 val loss 4.1613
+2300 val perplexity 64.1528
+2300 train 4.085191 (lr=3.2197e-03) (hash(x)=82104275)
+2310 train 4.130836 (lr=3.2008e-03) (hash(x)=82137309)
+2320 train 4.111669 (lr=3.1818e-03) (hash(x)=86741167)
+2330 train 4.122366 (lr=3.1629e-03) (hash(x)=84796263)
+2340 train 4.093379 (lr=3.1439e-03) (hash(x)=169966529)
+2350 train 3.973513 (lr=3.1248e-03) (hash(x)=80006061)
+2360 train 4.027627 (lr=3.1058e-03) (hash(x)=71090999)
+2370 train 4.158289 (lr=3.0867e-03) (hash(x)=86641850)
+2380 train 4.061723 (lr=3.0676e-03) (hash(x)=84573253)
+2390 train 4.086541 (lr=3.0485e-03) (hash(x)=74483764)
+2400 val loss 4.1580
+2400 val perplexity 63.9419
+2400 train 4.060765 (lr=3.0293e-03) (hash(x)=78327659)
+2410 train 4.128473 (lr=3.0101e-03) (hash(x)=77273627)
+2420 train 3.992177 (lr=2.9910e-03) (hash(x)=76938049)
+2430 train 3.992799 (lr=2.9717e-03) (hash(x)=98449442)
+2440 train 3.857307 (lr=2.9525e-03) (hash(x)=78157797)
+2450 train 3.895611 (lr=2.9333e-03) (hash(x)=80637582)
+2460 train 3.893587 (lr=2.9140e-03) (hash(x)=80023854)
+2470 train 3.948915 (lr=2.8947e-03) (hash(x)=86678884)
+2480 train 4.178443 (lr=2.8755e-03) (hash(x)=84086469)
+2490 train 4.189663 (lr=2.8562e-03) (hash(x)=83131141)
+2500 val loss 4.1307
+2500 val perplexity 62.2204
+2500 train 4.111951 (lr=2.8369e-03) (hash(x)=82583497)
+2510 train 4.171259 (lr=2.8176e-03) (hash(x)=79164326)
+2520 train 4.079795 (lr=2.7983e-03) (hash(x)=76456503)
+2530 train 4.134625 (lr=2.7790e-03) (hash(x)=82001228)
+2540 train 4.166300 (lr=2.7597e-03) (hash(x)=75615595)
+2550 train 4.117345 (lr=2.7403e-03) (hash(x)=78397869)
+2560 train 4.119359 (lr=2.7210e-03) (hash(x)=75547032)
+2570 train 4.028825 (lr=2.7017e-03) (hash(x)=89201025)
+2580 train 4.018310 (lr=2.6824e-03) (hash(x)=89856704)
+2590 train 4.025906 (lr=2.6631e-03) (hash(x)=82175682)
+2600 val loss 4.1127
+2600 val perplexity 61.1128
+2600 train 4.035004 (lr=2.6438e-03) (hash(x)=78312826)
+2610 train 4.021173 (lr=2.6245e-03) (hash(x)=77066588)
+2620 train 3.863435 (lr=2.6053e-03) (hash(x)=78666061)
+2630 train 3.864039 (lr=2.5860e-03) (hash(x)=93762143)
+2640 train 3.875874 (lr=2.5667e-03) (hash(x)=83191587)
+2650 train 3.798448 (lr=2.5475e-03) (hash(x)=87169585)
+2660 train 4.026046 (lr=2.5283e-03) (hash(x)=86426388)
+2670 train 4.160994 (lr=2.5090e-03) (hash(x)=76692638)
+2680 train 4.068765 (lr=2.4899e-03) (hash(x)=77446063)
+2690 train 4.078123 (lr=2.4707e-03) (hash(x)=79809050)
+2700 val loss 4.1070
+2700 val perplexity 60.7635
+2700 train 4.002690 (lr=2.4515e-03) (hash(x)=83116823)
+2710 train 4.112427 (lr=2.4324e-03) (hash(x)=75622148)
+2720 train 4.041700 (lr=2.4133e-03) (hash(x)=80690512)
+2730 train 4.108588 (lr=2.3942e-03) (hash(x)=78009984)
+2740 train 4.065459 (lr=2.3752e-03) (hash(x)=74205488)
+2750 train 4.069176 (lr=2.3561e-03) (hash(x)=91013332)
+2760 train 3.992232 (lr=2.3371e-03) (hash(x)=76954961)
+2770 train 3.997994 (lr=2.3182e-03) (hash(x)=79390317)
+2780 train 3.874724 (lr=2.2992e-03) (hash(x)=70168783)
+2790 train 4.036049 (lr=2.2803e-03) (hash(x)=76028417)
+2800 val loss 4.0879
+2800 val perplexity 59.6133
+2800 train 3.856984 (lr=2.2614e-03) (hash(x)=77656050)
+2810 train 3.900784 (lr=2.2426e-03) (hash(x)=94248216)
+2820 train 3.964212 (lr=2.2238e-03) (hash(x)=78305078)
+2830 train 3.881435 (lr=2.2051e-03) (hash(x)=79948848)
+2840 train 3.742831 (lr=2.1863e-03) (hash(x)=85341024)
+2850 train 4.055018 (lr=2.1677e-03) (hash(x)=78735170)
+2860 train 4.304873 (lr=2.1490e-03) (hash(x)=71616419)
+2870 train 4.074418 (lr=2.1304e-03) (hash(x)=78656517)
+2880 train 4.084019 (lr=2.1119e-03) (hash(x)=80073987)
+2890 train 4.013699 (lr=2.0934e-03) (hash(x)=76894809)
+2900 val loss 4.0597
+2900 val perplexity 57.9597
+2900 train 3.993415 (lr=2.0749e-03) (hash(x)=80499838)
+2910 train 4.047279 (lr=2.0565e-03) (hash(x)=72673354)
+2920 train 4.026742 (lr=2.0382e-03) (hash(x)=84265768)
+2930 train 4.053822 (lr=2.0199e-03) (hash(x)=79612060)
+2940 train 3.906914 (lr=2.0017e-03) (hash(x)=74970087)
+2950 train 4.064762 (lr=1.9835e-03) (hash(x)=84166818)
+2960 train 4.053812 (lr=1.9654e-03) (hash(x)=89410221)
+2970 train 3.989048 (lr=1.9473e-03) (hash(x)=75672566)
+2980 train 3.846205 (lr=1.9293e-03) (hash(x)=81760314)
+2990 train 3.821634 (lr=1.9113e-03) (hash(x)=80605200)
+3000 val loss 4.0582
+3000 val perplexity 57.8694
+3000 train 3.815495 (lr=1.8934e-03) (hash(x)=83804735)
+3010 train 3.742125 (lr=1.8756e-03) (hash(x)=77015303)
+3020 train 3.811002 (lr=1.8578e-03) (hash(x)=81464523)
+3030 train 4.011090 (lr=1.8401e-03) (hash(x)=80168230)
+3040 train 3.949432 (lr=1.8225e-03) (hash(x)=84817006)
+3050 train 4.056327 (lr=1.8049e-03) (hash(x)=71601811)
+3060 train 3.983912 (lr=1.7875e-03) (hash(x)=85499733)
+3070 train 4.039054 (lr=1.7700e-03) (hash(x)=78670408)
+3080 train 3.928184 (lr=1.7527e-03) (hash(x)=77120468)
+3090 train 4.032280 (lr=1.7354e-03) (hash(x)=77927426)
+3100 val loss 4.0360
+3100 val perplexity 56.5974
+3100 train 4.050586 (lr=1.7182e-03) (hash(x)=83998606)
+3110 train 3.949378 (lr=1.7011e-03) (hash(x)=77990218)
+3120 train 3.869966 (lr=1.6840e-03) (hash(x)=81623970)
+3130 train 3.836012 (lr=1.6671e-03) (hash(x)=75014781)
+3140 train 3.889030 (lr=1.6502e-03) (hash(x)=72591250)
+3150 train 3.903394 (lr=1.6334e-03) (hash(x)=81421847)
+3160 train 3.954462 (lr=1.6166e-03) (hash(x)=75112631)
+3170 train 4.047494 (lr=1.6000e-03) (hash(x)=87518033)
+3180 train 3.907936 (lr=1.5834e-03) (hash(x)=78648348)
+3190 train 4.015879 (lr=1.5670e-03) (hash(x)=78270029)
+3200 val loss 4.0147
+3200 val perplexity 55.4043
+3200 train 4.086486 (lr=1.5506e-03) (hash(x)=83380714)
+3210 train 4.277988 (lr=1.5343e-03) (hash(x)=67804991)
+3220 train 4.000352 (lr=1.5181e-03) (hash(x)=76968804)
+3230 train 3.930394 (lr=1.5020e-03) (hash(x)=75798670)
+3240 train 4.009513 (lr=1.4860e-03) (hash(x)=83192811)
+3250 train 3.944531 (lr=1.4700e-03) (hash(x)=81659789)
+3260 train 3.990727 (lr=1.4542e-03) (hash(x)=80622502)
+3270 train 3.956918 (lr=1.4384e-03) (hash(x)=85436511)
+3280 train 3.901253 (lr=1.4228e-03) (hash(x)=80589180)
+3290 train 4.036962 (lr=1.4073e-03) (hash(x)=85747193)
+3300 val loss 4.0101
+3300 val perplexity 55.1509
+3300 train 3.898648 (lr=1.3918e-03) (hash(x)=74635692)
+3310 train 3.888001 (lr=1.3765e-03) (hash(x)=82045455)
+3320 train 3.758686 (lr=1.3612e-03) (hash(x)=73249173)
+3330 train 3.720568 (lr=1.3461e-03) (hash(x)=71492338)
+3340 train 4.057232 (lr=1.3310e-03) (hash(x)=76374071)
+3350 train 4.024077 (lr=1.3161e-03) (hash(x)=80474064)
+3360 train 3.933691 (lr=1.3013e-03) (hash(x)=84390892)
+3370 train 3.978242 (lr=1.2865e-03) (hash(x)=83399949)
+3380 train 4.031515 (lr=1.2719e-03) (hash(x)=77648059)
+3390 train 4.005635 (lr=1.2574e-03) (hash(x)=80152701)
+3400 val loss 3.9865
+3400 val perplexity 53.8656
+3400 train 4.052518 (lr=1.2430e-03) (hash(x)=81952545)
+3410 train 4.013619 (lr=1.2287e-03) (hash(x)=80908993)
+3420 train 4.059756 (lr=1.2146e-03) (hash(x)=79610037)
+3430 train 3.911489 (lr=1.2005e-03) (hash(x)=87624382)
+3440 train 3.973968 (lr=1.1866e-03) (hash(x)=82336381)
+3450 train 3.954776 (lr=1.1727e-03) (hash(x)=87687835)
+3460 train 3.954763 (lr=1.1590e-03) (hash(x)=76376135)
+3470 train 3.869805 (lr=1.1454e-03) (hash(x)=79331391)
+3480 train 3.772729 (lr=1.1319e-03) (hash(x)=107002681)
+3490 train 3.853944 (lr=1.1186e-03) (hash(x)=77465514)
+3500 val loss 3.9819
+3500 val perplexity 53.6197
+3500 train 3.816989 (lr=1.1053e-03) (hash(x)=88237229)
+3510 train 3.883598 (lr=1.0922e-03) (hash(x)=81011739)
+3520 train 3.933651 (lr=1.0792e-03) (hash(x)=64643427)
+3530 train 4.209607 (lr=1.0663e-03) (hash(x)=78029539)
+3540 train 4.045026 (lr=1.0536e-03) (hash(x)=83188968)
+3550 train 4.060884 (lr=1.0410e-03) (hash(x)=86104185)
+3560 train 3.987900 (lr=1.0285e-03) (hash(x)=81993629)
+3570 train 3.959754 (lr=1.0161e-03) (hash(x)=72719368)
+3580 train 3.962993 (lr=1.0039e-03) (hash(x)=90091487)
+3590 train 3.851993 (lr=9.9174e-04) (hash(x)=77393152)
+3600 val loss 3.9614
+3600 val perplexity 52.5282
+3600 train 3.960547 (lr=9.7975e-04) (hash(x)=73103504)
+3610 train 3.971396 (lr=9.6790e-04) (hash(x)=84429400)
+3620 train 3.946180 (lr=9.5617e-04) (hash(x)=77021795)
+3630 train 3.895407 (lr=9.4458e-04) (hash(x)=92830605)
+3640 train 4.092533 (lr=9.3312e-04) (hash(x)=78313175)
+3650 train 3.888989 (lr=9.2180e-04) (hash(x)=85395549)
+3660 train 3.685321 (lr=9.1061e-04) (hash(x)=78114459)
+3670 train 3.811379 (lr=8.9955e-04) (hash(x)=74968316)
+3680 train 3.737785 (lr=8.8863e-04) (hash(x)=73358737)
+3690 train 3.838933 (lr=8.7785e-04) (hash(x)=76399442)
+3700 val loss 3.9618
+3700 val perplexity 52.5492
+3700 train 3.849116 (lr=8.6721e-04) (hash(x)=74039273)
+3710 train 4.026682 (lr=8.5671e-04) (hash(x)=79299680)
+3720 train 3.934015 (lr=8.4634e-04) (hash(x)=94701498)
+3730 train 4.009239 (lr=8.3611e-04) (hash(x)=75352071)
+3740 train 3.894974 (lr=8.2603e-04) (hash(x)=77909487)
+3750 train 3.997064 (lr=8.1609e-04) (hash(x)=73986730)
+3760 train 3.957295 (lr=8.0629e-04) (hash(x)=79325763)
+3770 train 4.005707 (lr=7.9663e-04) (hash(x)=72457818)
+3780 train 4.130510 (lr=7.8712e-04) (hash(x)=71775590)
+3790 train 4.016381 (lr=7.7775e-04) (hash(x)=82638943)
+3800 val loss 3.9431
+3800 val perplexity 51.5779
+3800 train 3.998456 (lr=7.6853e-04) (hash(x)=79965893)
+3810 train 3.935082 (lr=7.5945e-04) (hash(x)=72598235)
+3820 train 3.917046 (lr=7.5052e-04) (hash(x)=83113889)
+3830 train 3.973845 (lr=7.4174e-04) (hash(x)=74434590)
+3840 train 3.892936 (lr=7.3311e-04) (hash(x)=82860348)
+3850 train 3.830428 (lr=7.2462e-04) (hash(x)=78067565)
+3860 train 3.894365 (lr=7.1628e-04) (hash(x)=82592498)
+3870 train 3.820820 (lr=7.0810e-04) (hash(x)=81820733)
+3880 train 3.840778 (lr=7.0006e-04) (hash(x)=87709040)
+3890 train 3.886108 (lr=6.9217e-04) (hash(x)=70379093)
+3900 val loss 3.9349
+3900 val perplexity 51.1557
+3900 train 4.011623 (lr=6.8444e-04) (hash(x)=76597431)
+3910 train 3.894998 (lr=6.7686e-04) (hash(x)=90490716)
+3920 train 3.996272 (lr=6.6943e-04) (hash(x)=81970659)
+3930 train 3.978808 (lr=6.6215e-04) (hash(x)=81496334)
+3940 train 3.960135 (lr=6.5503e-04) (hash(x)=75717605)
+3950 train 3.908638 (lr=6.4806e-04) (hash(x)=82749357)
+3960 train 3.934831 (lr=6.4125e-04) (hash(x)=89355157)
+3970 train 3.893466 (lr=6.3459e-04) (hash(x)=78980403)
+3980 train 3.864629 (lr=6.2809e-04) (hash(x)=76627217)
+3990 train 3.865995 (lr=6.2175e-04) (hash(x)=72412879)
+4000 val loss 3.9282
+4000 val perplexity 50.8141
+4000 train 3.878534 (lr=6.1556e-04) (hash(x)=83018142)
+4010 train 3.890265 (lr=6.0953e-04) (hash(x)=81272436)
+4020 train 3.817843 (lr=6.0365e-04) (hash(x)=85497482)
+4030 train 3.889722 (lr=5.9794e-04) (hash(x)=85241734)
+4040 train 3.856755 (lr=5.9238e-04) (hash(x)=77925307)
+4050 train 3.858809 (lr=5.8698e-04) (hash(x)=84826179)
+4060 train 3.914512 (lr=5.8175e-04) (hash(x)=83606764)
+4070 train 3.830807 (lr=5.7667e-04) (hash(x)=80567590)
+4080 train 3.879648 (lr=5.7175e-04) (hash(x)=76860998)
+4090 train 3.916497 (lr=5.6699e-04) (hash(x)=74902328)
+4100 val loss 3.9151
+4100 val perplexity 50.1523
+4100 train 3.957217 (lr=5.6239e-04) (hash(x)=82832041)
+4110 train 3.835522 (lr=5.5796e-04) (hash(x)=79143262)
+4120 train 3.868289 (lr=5.5368e-04) (hash(x)=77038149)
+4130 train 4.049788 (lr=5.4957e-04) (hash(x)=86339074)
+4140 train 4.115891 (lr=5.4562e-04) (hash(x)=76686216)
+4150 train 3.889693 (lr=5.4183e-04) (hash(x)=70522682)
+4160 train 3.856546 (lr=5.3821e-04) (hash(x)=90958555)
+4170 train 3.879767 (lr=5.3474e-04) (hash(x)=91463532)
+4180 train 3.865762 (lr=5.3144e-04) (hash(x)=81959329)
+4190 train 3.826467 (lr=5.2831e-04) (hash(x)=83146752)
+4200 val loss 3.9196
+4200 val perplexity 50.3806
+4200 train 3.809159 (lr=5.2534e-04) (hash(x)=78361715)
+4210 train 3.956078 (lr=5.2253e-04) (hash(x)=87364889)
+4220 train 3.810987 (lr=5.1988e-04) (hash(x)=70465156)
+4230 train 3.967029 (lr=5.1740e-04) (hash(x)=84524081)
+4240 train 3.917208 (lr=5.1509e-04) (hash(x)=77824868)
+4250 train 3.929652 (lr=5.1294e-04) (hash(x)=81710711)
+4260 train 3.914163 (lr=5.1095e-04) (hash(x)=76362728)
+4270 train 3.906435 (lr=5.0913e-04) (hash(x)=83115208)
+4280 train 3.816528 (lr=5.0748e-04) (hash(x)=87218314)
+4290 train 3.976790 (lr=5.0599e-04) (hash(x)=74582673)
+4300 val loss 3.9113
+4300 val perplexity 49.9646
+4300 train 3.850865 (lr=5.0466e-04) (hash(x)=77379615)
+4310 train 3.826776 (lr=5.0350e-04) (hash(x)=78669579)
+4320 train 3.945709 (lr=5.0251e-04) (hash(x)=83066608)
+4330 train 3.906804 (lr=5.0168e-04) (hash(x)=83037340)
+4340 train 3.939926 (lr=5.0102e-04) (hash(x)=82849771)
+4350 train 3.880484 (lr=5.0052e-04) (hash(x)=76693985)
+4360 train 3.969632 (lr=5.0019e-04) (hash(x)=77745394)
+4370 train 3.930569 (lr=5.0002e-04) (hash(x)=79954388)
+4374 val loss 3.9040
+4374 val perplexity 49.6026

12_head_baseline_lr_50e-4_head_dim_22/model_04374.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:d365ebaf8228f6541ecd53a90868e5935be2de280e557130bc27e59a794e0a47
+size 96858242

12_head_baseline_lr_50e-4_head_dim_22/optimizer_04374.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:f0aa1c2b7986cb66573c8427c24f44f15d5072ff8f5fb9665a5d1d60d97c3ff1
+size 187435910