Upload folder using huggingface_hub

12_head_baseline_lr_45e-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_45e-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": 0.0045, "decay_lr": true, "readout_zero_init": false, "query_zero_init": false, "l1_loss": false, "debugpy": false, "key": "12_head_baseline_lr_45e-4"}

12_head_baseline_lr_45e-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_45e-4_head_dim_22/log2.txt

@@ -0,0 +1,529 @@
+max_steps: 4375
+0 val loss 10.9076
+0 val perplexity 54591.7773
+0 train 10.908084 (lr=6.2937e-06) (hash(x)=93356070)
+10 train 10.443457 (lr=6.9231e-05) (hash(x)=91253010)
+20 train 10.142874 (lr=1.3217e-04) (hash(x)=74934453)
+30 train 9.697568 (lr=1.9510e-04) (hash(x)=79278034)
+40 train 9.081161 (lr=2.5804e-04) (hash(x)=80107892)
+50 train 8.511380 (lr=3.2098e-04) (hash(x)=70745428)
+60 train 8.070782 (lr=3.8392e-04) (hash(x)=80077589)
+70 train 7.568213 (lr=4.4685e-04) (hash(x)=76213766)
+80 train 7.513824 (lr=5.0979e-04) (hash(x)=83218328)
+90 train 7.329232 (lr=5.7273e-04) (hash(x)=74855845)
+100 val loss 7.2050
+100 val perplexity 1346.0970
+100 train 7.062106 (lr=6.3566e-04) (hash(x)=82814902)
+110 train 6.990048 (lr=6.9860e-04) (hash(x)=84907741)
+120 train 6.825825 (lr=7.6154e-04) (hash(x)=82613223)
+130 train 6.650343 (lr=8.2448e-04) (hash(x)=83540876)
+140 train 6.620094 (lr=8.8741e-04) (hash(x)=75095216)
+150 train 6.659968 (lr=9.5035e-04) (hash(x)=97190944)
+160 train 6.605835 (lr=1.0133e-03) (hash(x)=82117809)
+170 train 6.489573 (lr=1.0762e-03) (hash(x)=70514724)
+180 train 6.414136 (lr=1.1392e-03) (hash(x)=81029624)
+190 train 6.338988 (lr=1.2021e-03) (hash(x)=78996841)
+200 val loss 6.2787
+200 val perplexity 533.1180
+200 train 6.279589 (lr=1.2650e-03) (hash(x)=79845097)
+210 train 6.174238 (lr=1.3280e-03) (hash(x)=78997683)
+220 train 6.087176 (lr=1.3909e-03) (hash(x)=74895865)
+230 train 6.230381 (lr=1.4538e-03) (hash(x)=80933276)
+240 train 5.939016 (lr=1.5168e-03) (hash(x)=77664606)
+250 train 5.969964 (lr=1.5797e-03) (hash(x)=75903930)
+260 train 5.862676 (lr=1.6427e-03) (hash(x)=82996853)
+270 train 5.791356 (lr=1.7056e-03) (hash(x)=73269568)
+280 train 5.717883 (lr=1.7685e-03) (hash(x)=82906427)
+290 train 5.655074 (lr=1.8315e-03) (hash(x)=71797895)
+300 val loss 5.8015
+300 val perplexity 330.7952
+300 train 5.506413 (lr=1.8944e-03) (hash(x)=85232249)
+310 train 5.458512 (lr=1.9573e-03) (hash(x)=75252489)
+320 train 5.539351 (lr=2.0203e-03) (hash(x)=74394644)
+330 train 5.633944 (lr=2.0832e-03) (hash(x)=82772910)
+340 train 5.683032 (lr=2.1462e-03) (hash(x)=81627464)
+350 train 5.663598 (lr=2.2091e-03) (hash(x)=84678053)
+360 train 5.603345 (lr=2.2720e-03) (hash(x)=81884128)
+370 train 5.607181 (lr=2.3350e-03) (hash(x)=76379242)
+380 train 5.418147 (lr=2.3979e-03) (hash(x)=98182875)
+390 train 5.386377 (lr=2.4608e-03) (hash(x)=79710436)
+400 val loss 5.4384
+400 val perplexity 230.0843
+400 train 5.381654 (lr=2.5238e-03) (hash(x)=79841071)
+410 train 5.346811 (lr=2.5867e-03) (hash(x)=75844151)
+420 train 5.346627 (lr=2.6497e-03) (hash(x)=73125036)
+430 train 5.369853 (lr=2.7126e-03) (hash(x)=84214858)
+440 train 5.257258 (lr=2.7755e-03) (hash(x)=80456994)
+450 train 5.249259 (lr=2.8385e-03) (hash(x)=76735962)
+460 train 5.160303 (lr=2.9014e-03) (hash(x)=81845446)
+470 train 4.971720 (lr=2.9643e-03) (hash(x)=76094689)
+480 train 5.105975 (lr=3.0273e-03) (hash(x)=83806686)
+490 train 4.914362 (lr=3.0902e-03) (hash(x)=84690227)
+500 val loss 5.2602
+500 val perplexity 192.5164
+500 train 4.994784 (lr=3.1531e-03) (hash(x)=71851938)
+510 train 5.113018 (lr=3.2161e-03) (hash(x)=77159346)
+520 train 5.160901 (lr=3.2790e-03) (hash(x)=80755753)
+530 train 5.291695 (lr=3.3420e-03) (hash(x)=75379679)
+540 train 5.182175 (lr=3.4049e-03) (hash(x)=82458619)
+550 train 5.140905 (lr=3.4678e-03) (hash(x)=86366396)
+560 train 5.175700 (lr=3.5308e-03) (hash(x)=85095044)
+570 train 5.135133 (lr=3.5937e-03) (hash(x)=78385159)
+580 train 5.064587 (lr=3.6566e-03) (hash(x)=79342394)
+590 train 5.105096 (lr=3.7196e-03) (hash(x)=70782192)
+600 val loss 5.1162
+600 val perplexity 166.6962
+600 train 5.092052 (lr=3.7825e-03) (hash(x)=76130353)
+610 train 5.003191 (lr=3.8455e-03) (hash(x)=74778440)
+620 train 4.959255 (lr=3.9084e-03) (hash(x)=79129709)
+630 train 4.857925 (lr=3.9713e-03) (hash(x)=76469962)
+640 train 4.858615 (lr=4.0343e-03) (hash(x)=78288049)
+650 train 4.818135 (lr=4.0972e-03) (hash(x)=76641388)
+660 train 4.738489 (lr=4.1601e-03) (hash(x)=75937906)
+670 train 4.723015 (lr=4.2231e-03) (hash(x)=74807157)
+680 train 4.624310 (lr=4.2860e-03) (hash(x)=77490144)
+690 train 4.658890 (lr=4.3490e-03) (hash(x)=72900124)
+700 val loss 4.9349
+700 val perplexity 139.0642
+700 train 4.624784 (lr=4.4119e-03) (hash(x)=73424218)
+710 train 4.898812 (lr=4.4748e-03) (hash(x)=86052345)
+720 train 4.864446 (lr=4.5000e-03) (hash(x)=77462613)
+730 train 4.880635 (lr=4.4998e-03) (hash(x)=79027471)
+740 train 4.834323 (lr=4.4995e-03) (hash(x)=78149992)
+750 train 4.768178 (lr=4.4991e-03) (hash(x)=89147499)
+760 train 4.816968 (lr=4.4985e-03) (hash(x)=74931538)
+770 train 4.755541 (lr=4.4977e-03) (hash(x)=89648204)
+780 train 4.798131 (lr=4.4968e-03) (hash(x)=83708293)
+790 train 4.803608 (lr=4.4958e-03) (hash(x)=84336512)
+800 val loss 4.7703
+800 val perplexity 117.9501
+800 train 4.695293 (lr=4.4946e-03) (hash(x)=75025285)
+810 train 4.641448 (lr=4.4933e-03) (hash(x)=79436536)
+820 train 4.617040 (lr=4.4918e-03) (hash(x)=90467390)
+830 train 4.657061 (lr=4.4901e-03) (hash(x)=77292020)
+840 train 4.578790 (lr=4.4884e-03) (hash(x)=75568927)
+850 train 4.599410 (lr=4.4864e-03) (hash(x)=79671781)
+860 train 4.679811 (lr=4.4843e-03) (hash(x)=84280943)
+870 train 4.676815 (lr=4.4821e-03) (hash(x)=80901294)
+880 train 4.796723 (lr=4.4797e-03) (hash(x)=81437584)
+890 train 4.646313 (lr=4.4772e-03) (hash(x)=73893778)
+900 val loss 4.6685
+900 val perplexity 106.5372
+900 train 4.599154 (lr=4.4745e-03) (hash(x)=74987794)
+910 train 4.607971 (lr=4.4717e-03) (hash(x)=93721374)
+920 train 4.595990 (lr=4.4687e-03) (hash(x)=79149678)
+930 train 4.552395 (lr=4.4656e-03) (hash(x)=83179387)
+940 train 4.552142 (lr=4.4624e-03) (hash(x)=89009978)
+950 train 4.545175 (lr=4.4589e-03) (hash(x)=76483735)
+960 train 4.481332 (lr=4.4554e-03) (hash(x)=78638850)
+970 train 4.508463 (lr=4.4517e-03) (hash(x)=90415206)
+980 train 4.452473 (lr=4.4478e-03) (hash(x)=90819195)
+990 train 4.584301 (lr=4.4438e-03) (hash(x)=85938852)
+1000 val loss 4.5637
+1000 val perplexity 95.9401
+1000 train 4.611748 (lr=4.4397e-03) (hash(x)=82436789)
+1010 train 4.675111 (lr=4.4354e-03) (hash(x)=63022149)
+1020 train 4.580532 (lr=4.4310e-03) (hash(x)=83501199)
+1030 train 4.559902 (lr=4.4264e-03) (hash(x)=77093326)
+1040 train 4.442497 (lr=4.4217e-03) (hash(x)=73437559)
+1050 train 4.536575 (lr=4.4169e-03) (hash(x)=84550388)
+1060 train 4.531255 (lr=4.4119e-03) (hash(x)=86184566)
+1070 train 4.523239 (lr=4.4067e-03) (hash(x)=85644922)
+1080 train 4.454116 (lr=4.4014e-03) (hash(x)=86179801)
+1090 train 4.433668 (lr=4.3960e-03) (hash(x)=82868303)
+1100 val loss 4.5036
+1100 val perplexity 90.3414
+1100 train 4.528584 (lr=4.3904e-03) (hash(x)=96780388)
+1110 train 4.469618 (lr=4.3847e-03) (hash(x)=87223122)
+1120 train 4.442015 (lr=4.3789e-03) (hash(x)=77292786)
+1130 train 4.460906 (lr=4.3729e-03) (hash(x)=88761375)
+1140 train 4.365977 (lr=4.3667e-03) (hash(x)=80956468)
+1150 train 4.549721 (lr=4.3605e-03) (hash(x)=84725462)
+1160 train 4.530141 (lr=4.3541e-03) (hash(x)=76098113)
+1170 train 4.506246 (lr=4.3475e-03) (hash(x)=84387685)
+1180 train 4.452580 (lr=4.3408e-03) (hash(x)=76120321)
+1190 train 4.411777 (lr=4.3340e-03) (hash(x)=85974065)
+1200 val loss 4.4343
+1200 val perplexity 84.2964
+1200 train 4.318137 (lr=4.3270e-03) (hash(x)=79340644)
+1210 train 4.411776 (lr=4.3199e-03) (hash(x)=88345025)
+1220 train 4.317564 (lr=4.3127e-03) (hash(x)=86406230)
+1230 train 4.311738 (lr=4.3053e-03) (hash(x)=80608975)
+1240 train 4.358839 (lr=4.2978e-03) (hash(x)=81635225)
+1250 train 4.201019 (lr=4.2902e-03) (hash(x)=82126376)
+1260 train 4.382804 (lr=4.2824e-03) (hash(x)=83737972)
+1270 train 4.291987 (lr=4.2745e-03) (hash(x)=81132345)
+1280 train 4.381249 (lr=4.2665e-03) (hash(x)=85850404)
+1290 train 4.393670 (lr=4.2583e-03) (hash(x)=83490640)
+1300 val loss 4.3980
+1300 val perplexity 81.2899
+1300 train 4.371847 (lr=4.2500e-03) (hash(x)=77545187)
+1310 train 4.411253 (lr=4.2416e-03) (hash(x)=86412685)
+1320 train 4.309211 (lr=4.2330e-03) (hash(x)=86429640)
+1330 train 4.446244 (lr=4.2243e-03) (hash(x)=79645281)
+1340 train 4.398903 (lr=4.2155e-03) (hash(x)=82668541)
+1350 train 4.268805 (lr=4.2066e-03) (hash(x)=73627248)
+1360 train 4.364138 (lr=4.1975e-03) (hash(x)=74845133)
+1370 train 4.269295 (lr=4.1883e-03) (hash(x)=77850497)
+1380 train 4.297641 (lr=4.1790e-03) (hash(x)=76143954)
+1390 train 4.313132 (lr=4.1695e-03) (hash(x)=80675544)
+1400 val loss 4.3720
+1400 val perplexity 79.1992
+1400 train 4.282452 (lr=4.1599e-03) (hash(x)=76010938)
+1410 train 4.334379 (lr=4.1502e-03) (hash(x)=81260300)
+1420 train 4.448101 (lr=4.1404e-03) (hash(x)=83803629)
+1430 train 4.272176 (lr=4.1305e-03) (hash(x)=83840110)
+1440 train 4.466006 (lr=4.1204e-03) (hash(x)=97711831)
+1450 train 4.339498 (lr=4.1102e-03) (hash(x)=81110139)
+1460 train 4.269872 (lr=4.0999e-03) (hash(x)=90942839)
+1470 train 4.301402 (lr=4.0894e-03) (hash(x)=80721564)
+1480 train 4.399260 (lr=4.0789e-03) (hash(x)=77852059)
+1490 train 4.295176 (lr=4.0682e-03) (hash(x)=80011365)
+1500 val loss 4.3353
+1500 val perplexity 76.3459
+1500 train 4.321362 (lr=4.0574e-03) (hash(x)=84496142)
+1510 train 4.260795 (lr=4.0465e-03) (hash(x)=70186729)
+1520 train 4.287123 (lr=4.0355e-03) (hash(x)=74854227)
+1530 train 4.189298 (lr=4.0244e-03) (hash(x)=77468161)
+1540 train 4.206603 (lr=4.0131e-03) (hash(x)=87853059)
+1550 train 4.142653 (lr=4.0018e-03) (hash(x)=71225436)
+1560 train 4.241271 (lr=3.9903e-03) (hash(x)=84072783)
+1570 train 4.296734 (lr=3.9787e-03) (hash(x)=72723098)
+1580 train 4.313349 (lr=3.9670e-03) (hash(x)=90409866)
+1590 train 4.386318 (lr=3.9552e-03) (hash(x)=87481378)
+1600 val loss 4.3083
+1600 val perplexity 74.3125
+1600 train 4.370247 (lr=3.9433e-03) (hash(x)=77643862)
+1610 train 4.255949 (lr=3.9313e-03) (hash(x)=88862575)
+1620 train 4.271427 (lr=3.9191e-03) (hash(x)=84612581)
+1630 train 4.306268 (lr=3.9069e-03) (hash(x)=87075989)
+1640 train 4.192508 (lr=3.8945e-03) (hash(x)=88277361)
+1650 train 4.308311 (lr=3.8821e-03) (hash(x)=78750236)
+1660 train 4.241352 (lr=3.8695e-03) (hash(x)=82604581)
+1670 train 4.071190 (lr=3.8569e-03) (hash(x)=68482265)
+1680 train 4.195350 (lr=3.8441e-03) (hash(x)=75088835)
+1690 train 4.149308 (lr=3.8313e-03) (hash(x)=66060989)
+1700 val loss 4.3185
+1700 val perplexity 75.0763
+1700 train 4.305271 (lr=3.8183e-03) (hash(x)=79986754)
+1710 train 4.299456 (lr=3.8053e-03) (hash(x)=83657930)
+1720 train 4.237231 (lr=3.7921e-03) (hash(x)=81754135)
+1730 train 4.324254 (lr=3.7789e-03) (hash(x)=78904427)
+1740 train 4.223693 (lr=3.7655e-03) (hash(x)=85920177)
+1750 train 4.275943 (lr=3.7521e-03) (hash(x)=86573211)
+1760 train 4.229624 (lr=3.7385e-03) (hash(x)=81737128)
+1770 train 4.220276 (lr=3.7249e-03) (hash(x)=77400968)
+1780 train 4.234477 (lr=3.7112e-03) (hash(x)=73545497)
+1790 train 4.480901 (lr=3.6974e-03) (hash(x)=71641943)
+1800 val loss 4.2672
+1800 val perplexity 71.3212
+1800 train 4.330043 (lr=3.6835e-03) (hash(x)=87819781)
+1810 train 4.186752 (lr=3.6695e-03) (hash(x)=86870770)
+1820 train 4.195448 (lr=3.6554e-03) (hash(x)=82522211)
+1830 train 4.275971 (lr=3.6412e-03) (hash(x)=61947437)
+1840 train 4.162815 (lr=3.6270e-03) (hash(x)=79865406)
+1850 train 4.134509 (lr=3.6126e-03) (hash(x)=79828721)
+1860 train 4.274059 (lr=3.5982e-03) (hash(x)=80869571)
+1870 train 4.257416 (lr=3.5837e-03) (hash(x)=73780971)
+1880 train 4.216194 (lr=3.5691e-03) (hash(x)=79249549)
+1890 train 4.276030 (lr=3.5545e-03) (hash(x)=81041904)
+1900 val loss 4.2464
+1900 val perplexity 69.8503
+1900 train 4.337872 (lr=3.5397e-03) (hash(x)=82456430)
+1910 train 4.113013 (lr=3.5249e-03) (hash(x)=82222135)
+1920 train 4.148304 (lr=3.5100e-03) (hash(x)=70033249)
+1930 train 4.133564 (lr=3.4950e-03) (hash(x)=72887360)
+1940 train 4.179838 (lr=3.4800e-03) (hash(x)=83251100)
+1950 train 4.221179 (lr=3.4648e-03) (hash(x)=79660266)
+1960 train 4.100946 (lr=3.4496e-03) (hash(x)=78919068)
+1970 train 4.204841 (lr=3.4344e-03) (hash(x)=94690431)
+1980 train 4.035365 (lr=3.4190e-03) (hash(x)=88153756)
+1990 train 4.091574 (lr=3.4036e-03) (hash(x)=77195688)
+2000 val loss 4.2241
+2000 val perplexity 68.3133
+2000 train 4.192498 (lr=3.3881e-03) (hash(x)=81308591)
+2010 train 4.237671 (lr=3.3726e-03) (hash(x)=80205479)
+2020 train 4.155099 (lr=3.3570e-03) (hash(x)=81008704)
+2030 train 4.220112 (lr=3.3413e-03) (hash(x)=82914358)
+2040 train 4.228340 (lr=3.3255e-03) (hash(x)=88064399)
+2050 train 4.234824 (lr=3.3097e-03) (hash(x)=82889390)
+2060 train 4.304838 (lr=3.2939e-03) (hash(x)=80913554)
+2070 train 4.146251 (lr=3.2779e-03) (hash(x)=78121791)
+2080 train 4.277553 (lr=3.2620e-03) (hash(x)=75356657)
+2090 train 4.113638 (lr=3.2459e-03) (hash(x)=74048412)
+2100 val loss 4.2002
+2100 val perplexity 66.7007
+2100 train 4.026247 (lr=3.2298e-03) (hash(x)=68928225)
+2110 train 4.075083 (lr=3.2137e-03) (hash(x)=93734745)
+2120 train 4.074932 (lr=3.1974e-03) (hash(x)=72878164)
+2130 train 4.203812 (lr=3.1812e-03) (hash(x)=71508226)
+2140 train 4.167001 (lr=3.1649e-03) (hash(x)=84168671)
+2150 train 4.096613 (lr=3.1485e-03) (hash(x)=74673639)
+2160 train 4.246060 (lr=3.1321e-03) (hash(x)=75470031)
+2170 train 4.205618 (lr=3.1156e-03) (hash(x)=74307890)
+2180 train 4.320757 (lr=3.0991e-03) (hash(x)=77214245)
+2190 train 4.107280 (lr=3.0825e-03) (hash(x)=88628359)
+2200 val loss 4.1782
+2200 val perplexity 65.2485
+2200 train 4.116131 (lr=3.0659e-03) (hash(x)=74779126)
+2210 train 4.153569 (lr=3.0493e-03) (hash(x)=79817976)
+2220 train 4.108807 (lr=3.0326e-03) (hash(x)=75258996)
+2230 train 4.085124 (lr=3.0159e-03) (hash(x)=88164047)
+2240 train 4.068874 (lr=2.9991e-03) (hash(x)=78580686)
+2250 train 4.000635 (lr=2.9823e-03) (hash(x)=76510617)
+2260 train 4.081006 (lr=2.9655e-03) (hash(x)=76043966)
+2270 train 4.060118 (lr=2.9486e-03) (hash(x)=73062098)
+2280 train 4.289019 (lr=2.9317e-03) (hash(x)=81885909)
+2290 train 4.226622 (lr=2.9147e-03) (hash(x)=81481741)
+2300 val loss 4.1595
+2300 val perplexity 64.0401
+2300 train 4.080135 (lr=2.8977e-03) (hash(x)=82104275)
+2310 train 4.127464 (lr=2.8807e-03) (hash(x)=82137309)
+2320 train 4.109167 (lr=2.8637e-03) (hash(x)=86741167)
+2330 train 4.123013 (lr=2.8466e-03) (hash(x)=84796263)
+2340 train 4.097818 (lr=2.8295e-03) (hash(x)=169966529)
+2350 train 3.969773 (lr=2.8124e-03) (hash(x)=80006061)
+2360 train 4.027162 (lr=2.7952e-03) (hash(x)=71090999)
+2370 train 4.161279 (lr=2.7780e-03) (hash(x)=86641850)
+2380 train 4.057930 (lr=2.7608e-03) (hash(x)=84573253)
+2390 train 4.088683 (lr=2.7436e-03) (hash(x)=74483764)
+2400 val loss 4.1556
+2400 val perplexity 63.7924
+2400 train 4.058002 (lr=2.7264e-03) (hash(x)=78327659)
+2410 train 4.128336 (lr=2.7091e-03) (hash(x)=77273627)
+2420 train 3.987896 (lr=2.6919e-03) (hash(x)=76938049)
+2430 train 3.989423 (lr=2.6746e-03) (hash(x)=98449442)
+2440 train 3.851939 (lr=2.6573e-03) (hash(x)=78157797)
+2450 train 3.892995 (lr=2.6399e-03) (hash(x)=80637582)
+2460 train 3.890267 (lr=2.6226e-03) (hash(x)=80023854)
+2470 train 3.950493 (lr=2.6053e-03) (hash(x)=86678884)
+2480 train 4.176673 (lr=2.5879e-03) (hash(x)=84086469)
+2490 train 4.191201 (lr=2.5706e-03) (hash(x)=83131141)
+2500 val loss 4.1286
+2500 val perplexity 62.0892
+2500 train 4.106947 (lr=2.5532e-03) (hash(x)=82583497)
+2510 train 4.164584 (lr=2.5358e-03) (hash(x)=79164326)
+2520 train 4.075799 (lr=2.5185e-03) (hash(x)=76456503)
+2530 train 4.139409 (lr=2.5011e-03) (hash(x)=82001228)
+2540 train 4.164934 (lr=2.4837e-03) (hash(x)=75615595)
+2550 train 4.111584 (lr=2.4663e-03) (hash(x)=78397869)
+2560 train 4.114524 (lr=2.4489e-03) (hash(x)=75547032)
+2570 train 4.025485 (lr=2.4315e-03) (hash(x)=89201025)
+2580 train 4.013738 (lr=2.4142e-03) (hash(x)=89856704)
+2590 train 4.025880 (lr=2.3968e-03) (hash(x)=82175682)
+2600 val loss 4.1125
+2600 val perplexity 61.0998
+2600 train 4.033395 (lr=2.3794e-03) (hash(x)=78312826)
+2610 train 4.021948 (lr=2.3621e-03) (hash(x)=77066588)
+2620 train 3.863116 (lr=2.3447e-03) (hash(x)=78666061)
+2630 train 3.870616 (lr=2.3274e-03) (hash(x)=93762143)
+2640 train 3.876052 (lr=2.3101e-03) (hash(x)=83191587)
+2650 train 3.796335 (lr=2.2927e-03) (hash(x)=87169585)
+2660 train 4.019372 (lr=2.2754e-03) (hash(x)=86426388)
+2670 train 4.157969 (lr=2.2581e-03) (hash(x)=76692638)
+2680 train 4.071080 (lr=2.2409e-03) (hash(x)=77446063)
+2690 train 4.072331 (lr=2.2236e-03) (hash(x)=79809050)
+2700 val loss 4.1050
+2700 val perplexity 60.6413
+2700 train 4.000932 (lr=2.2064e-03) (hash(x)=83116823)
+2710 train 4.110765 (lr=2.1892e-03) (hash(x)=75622148)
+2720 train 4.038945 (lr=2.1720e-03) (hash(x)=80690512)
+2730 train 4.107310 (lr=2.1548e-03) (hash(x)=78009984)
+2740 train 4.060920 (lr=2.1376e-03) (hash(x)=74205488)
+2750 train 4.066218 (lr=2.1205e-03) (hash(x)=91013332)
+2760 train 3.990166 (lr=2.1034e-03) (hash(x)=76954961)
+2770 train 3.998373 (lr=2.0863e-03) (hash(x)=79390317)
+2780 train 3.870346 (lr=2.0693e-03) (hash(x)=70168783)
+2790 train 4.035718 (lr=2.0523e-03) (hash(x)=76028417)
+2800 val loss 4.0851
+2800 val perplexity 59.4501
+2800 train 3.849011 (lr=2.0353e-03) (hash(x)=77656050)
+2810 train 3.898592 (lr=2.0183e-03) (hash(x)=94248216)
+2820 train 3.963553 (lr=2.0014e-03) (hash(x)=78305078)
+2830 train 3.878233 (lr=1.9845e-03) (hash(x)=79948848)
+2840 train 3.739441 (lr=1.9677e-03) (hash(x)=85341024)
+2850 train 4.051101 (lr=1.9509e-03) (hash(x)=78735170)
+2860 train 4.297779 (lr=1.9341e-03) (hash(x)=71616419)
+2870 train 4.074628 (lr=1.9174e-03) (hash(x)=78656517)
+2880 train 4.082485 (lr=1.9007e-03) (hash(x)=80073987)
+2890 train 4.018866 (lr=1.8841e-03) (hash(x)=76894809)
+2900 val loss 4.0575
+2900 val perplexity 57.8300
+2900 train 3.993490 (lr=1.8675e-03) (hash(x)=80499838)
+2910 train 4.041917 (lr=1.8509e-03) (hash(x)=72673354)
+2920 train 4.026372 (lr=1.8344e-03) (hash(x)=84265768)
+2930 train 4.051604 (lr=1.8179e-03) (hash(x)=79612060)
+2940 train 3.907071 (lr=1.8015e-03) (hash(x)=74970087)
+2950 train 4.063517 (lr=1.7851e-03) (hash(x)=84166818)
+2960 train 4.051497 (lr=1.7688e-03) (hash(x)=89410221)
+2970 train 3.989285 (lr=1.7526e-03) (hash(x)=75672566)
+2980 train 3.844017 (lr=1.7363e-03) (hash(x)=81760314)
+2990 train 3.816067 (lr=1.7202e-03) (hash(x)=80605200)
+3000 val loss 4.0572
+3000 val perplexity 57.8105
+3000 train 3.813120 (lr=1.7041e-03) (hash(x)=83804735)
+3010 train 3.740401 (lr=1.6880e-03) (hash(x)=77015303)
+3020 train 3.808123 (lr=1.6721e-03) (hash(x)=81464523)
+3030 train 4.010704 (lr=1.6561e-03) (hash(x)=80168230)
+3040 train 3.948796 (lr=1.6403e-03) (hash(x)=84817006)
+3050 train 4.053991 (lr=1.6245e-03) (hash(x)=71601811)
+3060 train 3.983120 (lr=1.6087e-03) (hash(x)=85499733)
+3070 train 4.036233 (lr=1.5930e-03) (hash(x)=78670408)
+3080 train 3.929919 (lr=1.5774e-03) (hash(x)=77120468)
+3090 train 4.034234 (lr=1.5619e-03) (hash(x)=77927426)
+3100 val loss 4.0355
+3100 val perplexity 56.5739
+3100 train 4.049586 (lr=1.5464e-03) (hash(x)=83998606)
+3110 train 3.948550 (lr=1.5310e-03) (hash(x)=77990218)
+3120 train 3.865429 (lr=1.5156e-03) (hash(x)=81623970)
+3130 train 3.837754 (lr=1.5004e-03) (hash(x)=75014781)
+3140 train 3.888442 (lr=1.4852e-03) (hash(x)=72591250)
+3150 train 3.903487 (lr=1.4700e-03) (hash(x)=81421847)
+3160 train 3.953110 (lr=1.4550e-03) (hash(x)=75112631)
+3170 train 4.053273 (lr=1.4400e-03) (hash(x)=87518033)
+3180 train 3.901290 (lr=1.4251e-03) (hash(x)=78648348)
+3190 train 4.015341 (lr=1.4103e-03) (hash(x)=78270029)
+3200 val loss 4.0139
+3200 val perplexity 55.3612
+3200 train 4.084314 (lr=1.3955e-03) (hash(x)=83380714)
+3210 train 4.274916 (lr=1.3809e-03) (hash(x)=67804991)
+3220 train 3.995890 (lr=1.3663e-03) (hash(x)=76968804)
+3230 train 3.929096 (lr=1.3518e-03) (hash(x)=75798670)
+3240 train 4.010128 (lr=1.3374e-03) (hash(x)=83192811)
+3250 train 3.945243 (lr=1.3230e-03) (hash(x)=81659789)
+3260 train 3.988765 (lr=1.3088e-03) (hash(x)=80622502)
+3270 train 3.957424 (lr=1.2946e-03) (hash(x)=85436511)
+3280 train 3.899053 (lr=1.2805e-03) (hash(x)=80589180)
+3290 train 4.038465 (lr=1.2665e-03) (hash(x)=85747193)
+3300 val loss 4.0083
+3300 val perplexity 55.0524
+3300 train 3.902628 (lr=1.2526e-03) (hash(x)=74635692)
+3310 train 3.892558 (lr=1.2388e-03) (hash(x)=82045455)
+3320 train 3.759993 (lr=1.2251e-03) (hash(x)=73249173)
+3330 train 3.721480 (lr=1.2115e-03) (hash(x)=71492338)
+3340 train 4.054098 (lr=1.1979e-03) (hash(x)=76374071)
+3350 train 4.025442 (lr=1.1845e-03) (hash(x)=80474064)
+3360 train 3.933820 (lr=1.1711e-03) (hash(x)=84390892)
+3370 train 3.981359 (lr=1.1579e-03) (hash(x)=83399949)
+3380 train 4.029132 (lr=1.1447e-03) (hash(x)=77648059)
+3390 train 4.004344 (lr=1.1317e-03) (hash(x)=80152701)
+3400 val loss 3.9864
+3400 val perplexity 53.8591
+3400 train 4.048268 (lr=1.1187e-03) (hash(x)=81952545)
+3410 train 4.011857 (lr=1.1059e-03) (hash(x)=80908993)
+3420 train 4.056953 (lr=1.0931e-03) (hash(x)=79610037)
+3430 train 3.914837 (lr=1.0805e-03) (hash(x)=87624382)
+3440 train 3.977443 (lr=1.0679e-03) (hash(x)=82336381)
+3450 train 3.954830 (lr=1.0555e-03) (hash(x)=87687835)
+3460 train 3.955319 (lr=1.0431e-03) (hash(x)=76376135)
+3470 train 3.868557 (lr=1.0309e-03) (hash(x)=79331391)
+3480 train 3.772404 (lr=1.0187e-03) (hash(x)=107002681)
+3490 train 3.852919 (lr=1.0067e-03) (hash(x)=77465514)
+3500 val loss 3.9817
+3500 val perplexity 53.6057
+3500 train 3.817164 (lr=9.9480e-04) (hash(x)=88237229)
+3510 train 3.885977 (lr=9.8299e-04) (hash(x)=81011739)
+3520 train 3.937080 (lr=9.7130e-04) (hash(x)=64643427)
+3530 train 4.215271 (lr=9.5971e-04) (hash(x)=78029539)
+3540 train 4.041749 (lr=9.4823e-04) (hash(x)=83188968)
+3550 train 4.058108 (lr=9.3687e-04) (hash(x)=86104185)
+3560 train 3.987685 (lr=9.2562e-04) (hash(x)=81993629)
+3570 train 3.958878 (lr=9.1449e-04) (hash(x)=72719368)
+3580 train 3.969758 (lr=9.0347e-04) (hash(x)=90091487)
+3590 train 3.851887 (lr=8.9257e-04) (hash(x)=77393152)
+3600 val loss 3.9614
+3600 val perplexity 52.5317
+3600 train 3.962326 (lr=8.8178e-04) (hash(x)=73103504)
+3610 train 3.968733 (lr=8.7111e-04) (hash(x)=84429400)
+3620 train 3.945470 (lr=8.6056e-04) (hash(x)=77021795)
+3630 train 3.895486 (lr=8.5012e-04) (hash(x)=92830605)
+3640 train 4.092023 (lr=8.3981e-04) (hash(x)=78313175)
+3650 train 3.886850 (lr=8.2962e-04) (hash(x)=85395549)
+3660 train 3.683842 (lr=8.1955e-04) (hash(x)=78114459)
+3670 train 3.810094 (lr=8.0960e-04) (hash(x)=74968316)
+3680 train 3.738171 (lr=7.9977e-04) (hash(x)=73358737)
+3690 train 3.839586 (lr=7.9007e-04) (hash(x)=76399442)
+3700 val loss 3.9627
+3700 val perplexity 52.5992
+3700 train 3.845065 (lr=7.8049e-04) (hash(x)=74039273)
+3710 train 4.022370 (lr=7.7103e-04) (hash(x)=79299680)
+3720 train 3.934250 (lr=7.6171e-04) (hash(x)=94701498)
+3730 train 4.011451 (lr=7.5250e-04) (hash(x)=75352071)
+3740 train 3.893168 (lr=7.4343e-04) (hash(x)=77909487)
+3750 train 4.002237 (lr=7.3448e-04) (hash(x)=73986730)
+3760 train 3.957610 (lr=7.2566e-04) (hash(x)=79325763)
+3770 train 4.005772 (lr=7.1697e-04) (hash(x)=72457818)
+3780 train 4.127856 (lr=7.0841e-04) (hash(x)=71775590)
+3790 train 4.019744 (lr=6.9998e-04) (hash(x)=82638943)
+3800 val loss 3.9437
+3800 val perplexity 51.6109
+3800 train 3.997448 (lr=6.9168e-04) (hash(x)=79965893)
+3810 train 3.932855 (lr=6.8351e-04) (hash(x)=72598235)
+3820 train 3.913646 (lr=6.7547e-04) (hash(x)=83113889)
+3830 train 3.975082 (lr=6.6757e-04) (hash(x)=74434590)
+3840 train 3.892490 (lr=6.5979e-04) (hash(x)=82860348)
+3850 train 3.830576 (lr=6.5216e-04) (hash(x)=78067565)
+3860 train 3.891256 (lr=6.4465e-04) (hash(x)=82592498)
+3870 train 3.823926 (lr=6.3729e-04) (hash(x)=81820733)
+3880 train 3.844683 (lr=6.3005e-04) (hash(x)=87709040)
+3890 train 3.890274 (lr=6.2296e-04) (hash(x)=70379093)
+3900 val loss 3.9364
+3900 val perplexity 51.2330
+3900 train 4.010964 (lr=6.1600e-04) (hash(x)=76597431)
+3910 train 3.894915 (lr=6.0917e-04) (hash(x)=90490716)
+3920 train 3.998918 (lr=6.0249e-04) (hash(x)=81970659)
+3930 train 3.987105 (lr=5.9594e-04) (hash(x)=81496334)
+3940 train 3.960097 (lr=5.8953e-04) (hash(x)=75717605)
+3950 train 3.910129 (lr=5.8326e-04) (hash(x)=82749357)
+3960 train 3.937377 (lr=5.7713e-04) (hash(x)=89355157)
+3970 train 3.892755 (lr=5.7113e-04) (hash(x)=78980403)
+3980 train 3.867150 (lr=5.6528e-04) (hash(x)=76627217)
+3990 train 3.864689 (lr=5.5957e-04) (hash(x)=72412879)
+4000 val loss 3.9293
+4000 val perplexity 50.8732
+4000 train 3.879589 (lr=5.5400e-04) (hash(x)=83018142)
+4010 train 3.890895 (lr=5.4857e-04) (hash(x)=81272436)
+4020 train 3.820300 (lr=5.4329e-04) (hash(x)=85497482)
+4030 train 3.891135 (lr=5.3814e-04) (hash(x)=85241734)
+4040 train 3.858341 (lr=5.3314e-04) (hash(x)=77925307)
+4050 train 3.858371 (lr=5.2829e-04) (hash(x)=84826179)
+4060 train 3.919471 (lr=5.2357e-04) (hash(x)=83606764)
+4070 train 3.832870 (lr=5.1900e-04) (hash(x)=80567590)
+4080 train 3.881743 (lr=5.1457e-04) (hash(x)=76860998)
+4090 train 3.920039 (lr=5.1029e-04) (hash(x)=74902328)
+4100 val loss 3.9170
+4100 val perplexity 50.2482
+4100 train 3.962889 (lr=5.0615e-04) (hash(x)=82832041)
+4110 train 3.838349 (lr=5.0216e-04) (hash(x)=79143262)
+4120 train 3.868998 (lr=4.9831e-04) (hash(x)=77038149)
+4130 train 4.053898 (lr=4.9461e-04) (hash(x)=86339074)
+4140 train 4.119166 (lr=4.9106e-04) (hash(x)=76686216)
+4150 train 3.892898 (lr=4.8765e-04) (hash(x)=70522682)
+4160 train 3.857199 (lr=4.8439e-04) (hash(x)=90958555)
+4170 train 3.882116 (lr=4.8127e-04) (hash(x)=91463532)
+4180 train 3.865572 (lr=4.7830e-04) (hash(x)=81959329)
+4190 train 3.826999 (lr=4.7548e-04) (hash(x)=83146752)
+4200 val loss 3.9227
+4200 val perplexity 50.5357
+4200 train 3.814270 (lr=4.7280e-04) (hash(x)=78361715)
+4210 train 3.961164 (lr=4.7028e-04) (hash(x)=87364889)
+4220 train 3.812204 (lr=4.6790e-04) (hash(x)=70465156)
+4230 train 3.964916 (lr=4.6566e-04) (hash(x)=84524081)
+4240 train 3.918658 (lr=4.6358e-04) (hash(x)=77824868)
+4250 train 3.934568 (lr=4.6164e-04) (hash(x)=81710711)
+4260 train 3.915440 (lr=4.5986e-04) (hash(x)=76362728)
+4270 train 3.906548 (lr=4.5822e-04) (hash(x)=83115208)
+4280 train 3.815990 (lr=4.5673e-04) (hash(x)=87218314)
+4290 train 3.976292 (lr=4.5539e-04) (hash(x)=74582673)
+4300 val loss 3.9136
+4300 val perplexity 50.0772
+4300 train 3.852392 (lr=4.5419e-04) (hash(x)=77379615)
+4310 train 3.829798 (lr=4.5315e-04) (hash(x)=78669579)
+4320 train 3.948654 (lr=4.5226e-04) (hash(x)=83066608)
+4330 train 3.912447 (lr=4.5151e-04) (hash(x)=83037340)
+4340 train 3.941407 (lr=4.5091e-04) (hash(x)=82849771)
+4350 train 3.881882 (lr=4.5047e-04) (hash(x)=76693985)
+4360 train 3.970262 (lr=4.5017e-04) (hash(x)=77745394)
+4370 train 3.933635 (lr=4.5002e-04) (hash(x)=79954388)
+4374 val loss 3.9066
+4374 val perplexity 49.7292

12_head_baseline_lr_45e-4_head_dim_22/model_04374.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:8409ef4dfee8bfb6966369cf2037056343152a0b78c9e17579342c0772c76243
+size 96858242

12_head_baseline_lr_45e-4_head_dim_22/optimizer_04374.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:d5b146f67e53e670ef3dfa1f2ceeabe8a7ccd1ab3bac68931e3264e2ad61e8ce
+size 187435910