Upload folder using huggingface_hub

12_head_baseline_lr_30e-4_head_dim_22_fixed/args.json

@@ -0,0 +1 @@
+{"hellaswag": true, "attention_kind": "selective", "log_dir": "logs/repro_selective_pattern_rankings/12_head_baseline_lr_30e-4_head_dim_22_fixed", "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": 0.003, "decay_lr": true, "readout_zero_init": false, "query_zero_init": false, "l1_loss": false, "debugpy": false, "key": "12_head_baseline_lr_30e-4_fixed"}

12_head_baseline_lr_30e-4_head_dim_22_fixed/dataloader_04374.pt

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

12_head_baseline_lr_30e-4_head_dim_22_fixed/log2.txt

@@ -0,0 +1,529 @@
+max_steps: 4375
+0 val loss 11.4043
+0 val perplexity 89706.9531
+0 train 11.388920 (lr=4.1958e-06) (hash(x)=93356070)
+10 train 10.043298 (lr=4.6154e-05) (hash(x)=91253010)
+20 train 9.453281 (lr=8.8112e-05) (hash(x)=74934453)
+30 train 8.737671 (lr=1.3007e-04) (hash(x)=79278034)
+40 train 7.968776 (lr=1.7203e-04) (hash(x)=80107892)
+50 train 7.709503 (lr=2.1399e-04) (hash(x)=70745428)
+60 train 7.745905 (lr=2.5594e-04) (hash(x)=80077589)
+70 train 7.417418 (lr=2.9790e-04) (hash(x)=76213766)
+80 train 7.403365 (lr=3.3986e-04) (hash(x)=83218328)
+90 train 7.185566 (lr=3.8182e-04) (hash(x)=74855845)
+100 val loss 7.0940
+100 val perplexity 1204.6813
+100 train 6.932849 (lr=4.2378e-04) (hash(x)=82814902)
+110 train 6.916564 (lr=4.6573e-04) (hash(x)=84907741)
+120 train 6.732979 (lr=5.0769e-04) (hash(x)=82613223)
+130 train 6.583001 (lr=5.4965e-04) (hash(x)=83540876)
+140 train 6.556883 (lr=5.9161e-04) (hash(x)=75095216)
+150 train 6.624824 (lr=6.3357e-04) (hash(x)=97190944)
+160 train 6.543280 (lr=6.7552e-04) (hash(x)=82117809)
+170 train 6.432925 (lr=7.1748e-04) (hash(x)=70514724)
+180 train 6.356050 (lr=7.5944e-04) (hash(x)=81029624)
+190 train 6.287621 (lr=8.0140e-04) (hash(x)=78996841)
+200 val loss 6.2176
+200 val perplexity 501.5091
+200 train 6.226252 (lr=8.4336e-04) (hash(x)=79845097)
+210 train 6.118777 (lr=8.8531e-04) (hash(x)=78997683)
+220 train 6.037190 (lr=9.2727e-04) (hash(x)=74895865)
+230 train 6.162255 (lr=9.6923e-04) (hash(x)=80933276)
+240 train 5.857391 (lr=1.0112e-03) (hash(x)=77664606)
+250 train 5.915548 (lr=1.0531e-03) (hash(x)=75903930)
+260 train 5.802463 (lr=1.0951e-03) (hash(x)=82996853)
+270 train 5.732442 (lr=1.1371e-03) (hash(x)=73269568)
+280 train 5.678641 (lr=1.1790e-03) (hash(x)=82906427)
+290 train 5.583721 (lr=1.2210e-03) (hash(x)=71797895)
+300 val loss 5.7490
+300 val perplexity 313.8758
+300 train 5.451455 (lr=1.2629e-03) (hash(x)=85232249)
+310 train 5.394286 (lr=1.3049e-03) (hash(x)=75252489)
+320 train 5.486943 (lr=1.3469e-03) (hash(x)=74394644)
+330 train 5.591196 (lr=1.3888e-03) (hash(x)=82772910)
+340 train 5.650424 (lr=1.4308e-03) (hash(x)=81627464)
+350 train 5.606482 (lr=1.4727e-03) (hash(x)=84678053)
+360 train 5.556487 (lr=1.5147e-03) (hash(x)=81884128)
+370 train 5.570803 (lr=1.5566e-03) (hash(x)=76379242)
+380 train 5.388386 (lr=1.5986e-03) (hash(x)=98182875)
+390 train 5.355709 (lr=1.6406e-03) (hash(x)=79710436)
+400 val loss 5.4114
+400 val perplexity 223.9486
+400 train 5.356648 (lr=1.6825e-03) (hash(x)=79841071)
+410 train 5.310948 (lr=1.7245e-03) (hash(x)=75844151)
+420 train 5.313613 (lr=1.7664e-03) (hash(x)=73125036)
+430 train 5.345565 (lr=1.8084e-03) (hash(x)=84214858)
+440 train 5.231837 (lr=1.8503e-03) (hash(x)=80456994)
+450 train 5.219643 (lr=1.8923e-03) (hash(x)=76735962)
+460 train 5.132965 (lr=1.9343e-03) (hash(x)=81845446)
+470 train 4.943850 (lr=1.9762e-03) (hash(x)=76094689)
+480 train 5.093268 (lr=2.0182e-03) (hash(x)=83806686)
+490 train 4.888348 (lr=2.0601e-03) (hash(x)=84690227)
+500 val loss 5.2281
+500 val perplexity 186.4464
+500 train 4.959856 (lr=2.1021e-03) (hash(x)=71851938)
+510 train 5.084060 (lr=2.1441e-03) (hash(x)=77159346)
+520 train 5.129706 (lr=2.1860e-03) (hash(x)=80755753)
+530 train 5.231335 (lr=2.2280e-03) (hash(x)=75379679)
+540 train 5.137349 (lr=2.2699e-03) (hash(x)=82458619)
+550 train 5.077714 (lr=2.3119e-03) (hash(x)=86366396)
+560 train 5.145063 (lr=2.3538e-03) (hash(x)=85095044)
+570 train 5.063125 (lr=2.3958e-03) (hash(x)=78385159)
+580 train 5.002679 (lr=2.4378e-03) (hash(x)=79342394)
+590 train 5.030479 (lr=2.4797e-03) (hash(x)=70782192)
+600 val loss 5.0526
+600 val perplexity 156.4267
+600 train 5.018241 (lr=2.5217e-03) (hash(x)=76130353)
+610 train 4.924848 (lr=2.5636e-03) (hash(x)=74778440)
+620 train 4.912838 (lr=2.6056e-03) (hash(x)=79129709)
+630 train 4.805475 (lr=2.6476e-03) (hash(x)=76469962)
+640 train 4.797305 (lr=2.6895e-03) (hash(x)=78288049)
+650 train 4.723791 (lr=2.7315e-03) (hash(x)=76641388)
+660 train 4.653028 (lr=2.7734e-03) (hash(x)=75937906)
+670 train 4.664919 (lr=2.8154e-03) (hash(x)=74807157)
+680 train 4.554918 (lr=2.8573e-03) (hash(x)=77490144)
+690 train 4.594469 (lr=2.8993e-03) (hash(x)=72900124)
+700 val loss 4.8745
+700 val perplexity 130.9103
+700 train 4.574590 (lr=2.9413e-03) (hash(x)=73424218)
+710 train 4.843815 (lr=2.9832e-03) (hash(x)=86052345)
+720 train 4.824827 (lr=3.0000e-03) (hash(x)=77462613)
+730 train 4.836173 (lr=2.9999e-03) (hash(x)=79027471)
+740 train 4.794456 (lr=2.9997e-03) (hash(x)=78149992)
+750 train 4.738410 (lr=2.9994e-03) (hash(x)=89147499)
+760 train 4.795321 (lr=2.9990e-03) (hash(x)=74931538)
+770 train 4.732946 (lr=2.9985e-03) (hash(x)=89648204)
+780 train 4.791467 (lr=2.9979e-03) (hash(x)=83708293)
+790 train 4.788226 (lr=2.9972e-03) (hash(x)=84336512)
+800 val loss 4.7541
+800 val perplexity 116.0588
+800 train 4.673043 (lr=2.9964e-03) (hash(x)=75025285)
+810 train 4.591257 (lr=2.9955e-03) (hash(x)=79436536)
+820 train 4.598249 (lr=2.9945e-03) (hash(x)=90467390)
+830 train 4.634099 (lr=2.9934e-03) (hash(x)=77292020)
+840 train 4.557719 (lr=2.9922e-03) (hash(x)=75568927)
+850 train 4.589141 (lr=2.9909e-03) (hash(x)=79671781)
+860 train 4.676084 (lr=2.9896e-03) (hash(x)=84280943)
+870 train 4.662921 (lr=2.9881e-03) (hash(x)=80901294)
+880 train 4.779596 (lr=2.9865e-03) (hash(x)=81437584)
+890 train 4.638093 (lr=2.9848e-03) (hash(x)=73893778)
+900 val loss 4.6626
+900 val perplexity 105.9162
+900 train 4.596845 (lr=2.9830e-03) (hash(x)=74987794)
+910 train 4.612024 (lr=2.9811e-03) (hash(x)=93721374)
+920 train 4.591785 (lr=2.9792e-03) (hash(x)=79149678)
+930 train 4.549619 (lr=2.9771e-03) (hash(x)=83179387)
+940 train 4.550191 (lr=2.9749e-03) (hash(x)=89009978)
+950 train 4.537589 (lr=2.9726e-03) (hash(x)=76483735)
+960 train 4.469532 (lr=2.9703e-03) (hash(x)=78638850)
+970 train 4.508464 (lr=2.9678e-03) (hash(x)=90415206)
+980 train 4.453831 (lr=2.9652e-03) (hash(x)=90819195)
+990 train 4.578433 (lr=2.9626e-03) (hash(x)=85938852)
+1000 val loss 4.5610
+1000 val perplexity 95.6804
+1000 train 4.603090 (lr=2.9598e-03) (hash(x)=82436789)
+1010 train 4.671502 (lr=2.9570e-03) (hash(x)=63022149)
+1020 train 4.575079 (lr=2.9540e-03) (hash(x)=83501199)
+1030 train 4.554188 (lr=2.9510e-03) (hash(x)=77093326)
+1040 train 4.438592 (lr=2.9478e-03) (hash(x)=73437559)
+1050 train 4.522750 (lr=2.9446e-03) (hash(x)=84550388)
+1060 train 4.522533 (lr=2.9412e-03) (hash(x)=86184566)
+1070 train 4.514493 (lr=2.9378e-03) (hash(x)=85644922)
+1080 train 4.457847 (lr=2.9343e-03) (hash(x)=86179801)
+1090 train 4.429684 (lr=2.9307e-03) (hash(x)=82868303)
+1100 val loss 4.4985
+1100 val perplexity 89.8779
+1100 train 4.521542 (lr=2.9270e-03) (hash(x)=96780388)
+1110 train 4.465613 (lr=2.9231e-03) (hash(x)=87223122)
+1120 train 4.433199 (lr=2.9192e-03) (hash(x)=77292786)
+1130 train 4.460567 (lr=2.9152e-03) (hash(x)=88761375)
+1140 train 4.366926 (lr=2.9112e-03) (hash(x)=80956468)
+1150 train 4.536956 (lr=2.9070e-03) (hash(x)=84725462)
+1160 train 4.522674 (lr=2.9027e-03) (hash(x)=76098113)
+1170 train 4.503212 (lr=2.8983e-03) (hash(x)=84387685)
+1180 train 4.448418 (lr=2.8939e-03) (hash(x)=76120321)
+1190 train 4.407722 (lr=2.8893e-03) (hash(x)=85974065)
+1200 val loss 4.4300
+1200 val perplexity 83.9291
+1200 train 4.317886 (lr=2.8847e-03) (hash(x)=79340644)
+1210 train 4.409569 (lr=2.8800e-03) (hash(x)=88345025)
+1220 train 4.316388 (lr=2.8751e-03) (hash(x)=86406230)
+1230 train 4.306210 (lr=2.8702e-03) (hash(x)=80608975)
+1240 train 4.366783 (lr=2.8652e-03) (hash(x)=81635225)
+1250 train 4.204933 (lr=2.8601e-03) (hash(x)=82126376)
+1260 train 4.378597 (lr=2.8550e-03) (hash(x)=83737972)
+1270 train 4.290668 (lr=2.8497e-03) (hash(x)=81132345)
+1280 train 4.376769 (lr=2.8443e-03) (hash(x)=85850404)
+1290 train 4.386447 (lr=2.8389e-03) (hash(x)=83490640)
+1300 val loss 4.3921
+1300 val perplexity 80.8139
+1300 train 4.361177 (lr=2.8333e-03) (hash(x)=77545187)
+1310 train 4.413182 (lr=2.8277e-03) (hash(x)=86412685)
+1320 train 4.310979 (lr=2.8220e-03) (hash(x)=86429640)
+1330 train 4.441265 (lr=2.8162e-03) (hash(x)=79645281)
+1340 train 4.399456 (lr=2.8103e-03) (hash(x)=82668541)
+1350 train 4.262652 (lr=2.8044e-03) (hash(x)=73627248)
+1360 train 4.364335 (lr=2.7983e-03) (hash(x)=74845133)
+1370 train 4.270746 (lr=2.7922e-03) (hash(x)=77850497)
+1380 train 4.296601 (lr=2.7860e-03) (hash(x)=76143954)
+1390 train 4.302392 (lr=2.7797e-03) (hash(x)=80675544)
+1400 val loss 4.3688
+1400 val perplexity 78.9512
+1400 train 4.278130 (lr=2.7733e-03) (hash(x)=76010938)
+1410 train 4.333201 (lr=2.7668e-03) (hash(x)=81260300)
+1420 train 4.449539 (lr=2.7603e-03) (hash(x)=83803629)
+1430 train 4.271910 (lr=2.7536e-03) (hash(x)=83840110)
+1440 train 4.463119 (lr=2.7469e-03) (hash(x)=97711831)
+1450 train 4.337279 (lr=2.7401e-03) (hash(x)=81110139)
+1460 train 4.263162 (lr=2.7333e-03) (hash(x)=90942839)
+1470 train 4.302293 (lr=2.7263e-03) (hash(x)=80721564)
+1480 train 4.401114 (lr=2.7193e-03) (hash(x)=77852059)
+1490 train 4.292821 (lr=2.7121e-03) (hash(x)=80011365)
+1500 val loss 4.3303
+1500 val perplexity 75.9686
+1500 train 4.313121 (lr=2.7050e-03) (hash(x)=84496142)
+1510 train 4.262228 (lr=2.6977e-03) (hash(x)=70186729)
+1520 train 4.280502 (lr=2.6903e-03) (hash(x)=74854227)
+1530 train 4.186149 (lr=2.6829e-03) (hash(x)=77468161)
+1540 train 4.206680 (lr=2.6754e-03) (hash(x)=87853059)
+1550 train 4.141838 (lr=2.6678e-03) (hash(x)=71225436)
+1560 train 4.242230 (lr=2.6602e-03) (hash(x)=84072783)
+1570 train 4.295146 (lr=2.6525e-03) (hash(x)=72723098)
+1580 train 4.308866 (lr=2.6447e-03) (hash(x)=90409866)
+1590 train 4.376564 (lr=2.6368e-03) (hash(x)=87481378)
+1600 val loss 4.3073
+1600 val perplexity 74.2368
+1600 train 4.367268 (lr=2.6289e-03) (hash(x)=77643862)
+1610 train 4.254496 (lr=2.6208e-03) (hash(x)=88862575)
+1620 train 4.286589 (lr=2.6128e-03) (hash(x)=84612581)
+1630 train 4.305496 (lr=2.6046e-03) (hash(x)=87075989)
+1640 train 4.190815 (lr=2.5964e-03) (hash(x)=88277361)
+1650 train 4.310169 (lr=2.5881e-03) (hash(x)=78750236)
+1660 train 4.223742 (lr=2.5797e-03) (hash(x)=82604581)
+1670 train 4.071836 (lr=2.5713e-03) (hash(x)=68482265)
+1680 train 4.191428 (lr=2.5628e-03) (hash(x)=75088835)
+1690 train 4.136374 (lr=2.5542e-03) (hash(x)=66060989)
+1700 val loss 4.3078
+1700 val perplexity 74.2753
+1700 train 4.293228 (lr=2.5455e-03) (hash(x)=79986754)
+1710 train 4.287001 (lr=2.5368e-03) (hash(x)=83657930)
+1720 train 4.225898 (lr=2.5281e-03) (hash(x)=81754135)
+1730 train 4.317577 (lr=2.5192e-03) (hash(x)=78904427)
+1740 train 4.218430 (lr=2.5103e-03) (hash(x)=85920177)
+1750 train 4.266397 (lr=2.5014e-03) (hash(x)=86573211)
+1760 train 4.227085 (lr=2.4924e-03) (hash(x)=81737128)
+1770 train 4.215204 (lr=2.4833e-03) (hash(x)=77400968)
+1780 train 4.236235 (lr=2.4741e-03) (hash(x)=73545497)
+1790 train 4.477455 (lr=2.4649e-03) (hash(x)=71641943)
+1800 val loss 4.2629
+1800 val perplexity 71.0181
+1800 train 4.327610 (lr=2.4556e-03) (hash(x)=87819781)
+1810 train 4.186983 (lr=2.4463e-03) (hash(x)=86870770)
+1820 train 4.192216 (lr=2.4369e-03) (hash(x)=82522211)
+1830 train 4.280381 (lr=2.4275e-03) (hash(x)=61947437)
+1840 train 4.156124 (lr=2.4180e-03) (hash(x)=79865406)
+1850 train 4.125253 (lr=2.4084e-03) (hash(x)=79828721)
+1860 train 4.263417 (lr=2.3988e-03) (hash(x)=80869571)
+1870 train 4.251936 (lr=2.3891e-03) (hash(x)=73780971)
+1880 train 4.212151 (lr=2.3794e-03) (hash(x)=79249549)
+1890 train 4.271846 (lr=2.3696e-03) (hash(x)=81041904)
+1900 val loss 4.2443
+1900 val perplexity 69.7046
+1900 train 4.342704 (lr=2.3598e-03) (hash(x)=82456430)
+1910 train 4.109939 (lr=2.3499e-03) (hash(x)=82222135)
+1920 train 4.145533 (lr=2.3400e-03) (hash(x)=70033249)
+1930 train 4.130339 (lr=2.3300e-03) (hash(x)=72887360)
+1940 train 4.174621 (lr=2.3200e-03) (hash(x)=83251100)
+1950 train 4.220342 (lr=2.3099e-03) (hash(x)=79660266)
+1960 train 4.099016 (lr=2.2998e-03) (hash(x)=78919068)
+1970 train 4.197801 (lr=2.2896e-03) (hash(x)=94690431)
+1980 train 4.034377 (lr=2.2793e-03) (hash(x)=88153756)
+1990 train 4.088666 (lr=2.2691e-03) (hash(x)=77195688)
+2000 val loss 4.2243
+2000 val perplexity 68.3237
+2000 train 4.188674 (lr=2.2588e-03) (hash(x)=81308591)
+2010 train 4.239821 (lr=2.2484e-03) (hash(x)=80205479)
+2020 train 4.144857 (lr=2.2380e-03) (hash(x)=81008704)
+2030 train 4.216527 (lr=2.2275e-03) (hash(x)=82914358)
+2040 train 4.219554 (lr=2.2170e-03) (hash(x)=88064399)
+2050 train 4.230384 (lr=2.2065e-03) (hash(x)=82889390)
+2060 train 4.298309 (lr=2.1959e-03) (hash(x)=80913554)
+2070 train 4.141866 (lr=2.1853e-03) (hash(x)=78121791)
+2080 train 4.275607 (lr=2.1746e-03) (hash(x)=75356657)
+2090 train 4.097907 (lr=2.1639e-03) (hash(x)=74048412)
+2100 val loss 4.1990
+2100 val perplexity 66.6177
+2100 train 4.021699 (lr=2.1532e-03) (hash(x)=68928225)
+2110 train 4.074705 (lr=2.1424e-03) (hash(x)=93734745)
+2120 train 4.071416 (lr=2.1316e-03) (hash(x)=72878164)
+2130 train 4.203925 (lr=2.1208e-03) (hash(x)=71508226)
+2140 train 4.159962 (lr=2.1099e-03) (hash(x)=84168671)
+2150 train 4.096533 (lr=2.0990e-03) (hash(x)=74673639)
+2160 train 4.246063 (lr=2.0881e-03) (hash(x)=75470031)
+2170 train 4.199344 (lr=2.0771e-03) (hash(x)=74307890)
+2180 train 4.318949 (lr=2.0661e-03) (hash(x)=77214245)
+2190 train 4.100140 (lr=2.0550e-03) (hash(x)=88628359)
+2200 val loss 4.1773
+2200 val perplexity 65.1866
+2200 train 4.113553 (lr=2.0440e-03) (hash(x)=74779126)
+2210 train 4.153378 (lr=2.0329e-03) (hash(x)=79817976)
+2220 train 4.104673 (lr=2.0217e-03) (hash(x)=75258996)
+2230 train 4.077724 (lr=2.0106e-03) (hash(x)=88164047)
+2240 train 4.059766 (lr=1.9994e-03) (hash(x)=78580686)
+2250 train 4.005801 (lr=1.9882e-03) (hash(x)=76510617)
+2260 train 4.075488 (lr=1.9770e-03) (hash(x)=76043966)
+2270 train 4.053638 (lr=1.9657e-03) (hash(x)=73062098)
+2280 train 4.284164 (lr=1.9544e-03) (hash(x)=81885909)
+2290 train 4.217714 (lr=1.9431e-03) (hash(x)=81481741)
+2300 val loss 4.1581
+2300 val perplexity 63.9487
+2300 train 4.074558 (lr=1.9318e-03) (hash(x)=82104275)
+2310 train 4.128089 (lr=1.9205e-03) (hash(x)=82137309)
+2320 train 4.103353 (lr=1.9091e-03) (hash(x)=86741167)
+2330 train 4.119576 (lr=1.8977e-03) (hash(x)=84796263)
+2340 train 4.100761 (lr=1.8863e-03) (hash(x)=169966529)
+2350 train 3.966264 (lr=1.8749e-03) (hash(x)=80006061)
+2360 train 4.021591 (lr=1.8635e-03) (hash(x)=71090999)
+2370 train 4.152897 (lr=1.8520e-03) (hash(x)=86641850)
+2380 train 4.049273 (lr=1.8406e-03) (hash(x)=84573253)
+2390 train 4.086118 (lr=1.8291e-03) (hash(x)=74483764)
+2400 val loss 4.1483
+2400 val perplexity 63.3294
+2400 train 4.053940 (lr=1.8176e-03) (hash(x)=78327659)
+2410 train 4.123806 (lr=1.8061e-03) (hash(x)=77273627)
+2420 train 3.984845 (lr=1.7946e-03) (hash(x)=76938049)
+2430 train 3.990595 (lr=1.7830e-03) (hash(x)=98449442)
+2440 train 3.849964 (lr=1.7715e-03) (hash(x)=78157797)
+2450 train 3.894181 (lr=1.7600e-03) (hash(x)=80637582)
+2460 train 3.879771 (lr=1.7484e-03) (hash(x)=80023854)
+2470 train 3.942410 (lr=1.7368e-03) (hash(x)=86678884)
+2480 train 4.170700 (lr=1.7253e-03) (hash(x)=84086469)
+2490 train 4.187392 (lr=1.7137e-03) (hash(x)=83131141)
+2500 val loss 4.1241
+2500 val perplexity 61.8104
+2500 train 4.101756 (lr=1.7021e-03) (hash(x)=82583497)
+2510 train 4.166438 (lr=1.6906e-03) (hash(x)=79164326)
+2520 train 4.071899 (lr=1.6790e-03) (hash(x)=76456503)
+2530 train 4.130459 (lr=1.6674e-03) (hash(x)=82001228)
+2540 train 4.162688 (lr=1.6558e-03) (hash(x)=75615595)
+2550 train 4.109509 (lr=1.6442e-03) (hash(x)=78397869)
+2560 train 4.115657 (lr=1.6326e-03) (hash(x)=75547032)
+2570 train 4.024044 (lr=1.6210e-03) (hash(x)=89201025)
+2580 train 4.008725 (lr=1.6094e-03) (hash(x)=89856704)
+2590 train 4.022644 (lr=1.5979e-03) (hash(x)=82175682)
+2600 val loss 4.1071
+2600 val perplexity 60.7685
+2600 train 4.029530 (lr=1.5863e-03) (hash(x)=78312826)
+2610 train 4.020201 (lr=1.5747e-03) (hash(x)=77066588)
+2620 train 3.863490 (lr=1.5632e-03) (hash(x)=78666061)
+2630 train 3.859653 (lr=1.5516e-03) (hash(x)=93762143)
+2640 train 3.873468 (lr=1.5400e-03) (hash(x)=83191587)
+2650 train 3.798436 (lr=1.5285e-03) (hash(x)=87169585)
+2660 train 4.019514 (lr=1.5170e-03) (hash(x)=86426388)
+2670 train 4.155437 (lr=1.5054e-03) (hash(x)=76692638)
+2680 train 4.063534 (lr=1.4939e-03) (hash(x)=77446063)
+2690 train 4.067085 (lr=1.4824e-03) (hash(x)=79809050)
+2700 val loss 4.1035
+2700 val perplexity 60.5493
+2700 train 3.994519 (lr=1.4709e-03) (hash(x)=83116823)
+2710 train 4.104850 (lr=1.4594e-03) (hash(x)=75622148)
+2720 train 4.035907 (lr=1.4480e-03) (hash(x)=80690512)
+2730 train 4.109308 (lr=1.4365e-03) (hash(x)=78009984)
+2740 train 4.061550 (lr=1.4251e-03) (hash(x)=74205488)
+2750 train 4.065698 (lr=1.4137e-03) (hash(x)=91013332)
+2760 train 3.985155 (lr=1.4023e-03) (hash(x)=76954961)
+2770 train 4.002871 (lr=1.3909e-03) (hash(x)=79390317)
+2780 train 3.868957 (lr=1.3795e-03) (hash(x)=70168783)
+2790 train 4.024764 (lr=1.3682e-03) (hash(x)=76028417)
+2800 val loss 4.0799
+2800 val perplexity 59.1413
+2800 train 3.851883 (lr=1.3569e-03) (hash(x)=77656050)
+2810 train 3.896259 (lr=1.3456e-03) (hash(x)=94248216)
+2820 train 3.959035 (lr=1.3343e-03) (hash(x)=78305078)
+2830 train 3.870515 (lr=1.3230e-03) (hash(x)=79948848)
+2840 train 3.740088 (lr=1.3118e-03) (hash(x)=85341024)
+2850 train 4.051746 (lr=1.3006e-03) (hash(x)=78735170)
+2860 train 4.296257 (lr=1.2894e-03) (hash(x)=71616419)
+2870 train 4.072713 (lr=1.2783e-03) (hash(x)=78656517)
+2880 train 4.080023 (lr=1.2671e-03) (hash(x)=80073987)
+2890 train 4.015083 (lr=1.2560e-03) (hash(x)=76894809)
+2900 val loss 4.0560
+2900 val perplexity 57.7418
+2900 train 3.996126 (lr=1.2450e-03) (hash(x)=80499838)
+2910 train 4.044046 (lr=1.2339e-03) (hash(x)=72673354)
+2920 train 4.026986 (lr=1.2229e-03) (hash(x)=84265768)
+2930 train 4.051304 (lr=1.2119e-03) (hash(x)=79612060)
+2940 train 3.906831 (lr=1.2010e-03) (hash(x)=74970087)
+2950 train 4.065031 (lr=1.1901e-03) (hash(x)=84166818)
+2960 train 4.048571 (lr=1.1792e-03) (hash(x)=89410221)
+2970 train 3.989050 (lr=1.1684e-03) (hash(x)=75672566)
+2980 train 3.848064 (lr=1.1576e-03) (hash(x)=81760314)
+2990 train 3.818724 (lr=1.1468e-03) (hash(x)=80605200)
+3000 val loss 4.0554
+3000 val perplexity 57.7099
+3000 train 3.814094 (lr=1.1361e-03) (hash(x)=83804735)
+3010 train 3.746550 (lr=1.1254e-03) (hash(x)=77015303)
+3020 train 3.810715 (lr=1.1147e-03) (hash(x)=81464523)
+3030 train 4.008684 (lr=1.1041e-03) (hash(x)=80168230)
+3040 train 3.954852 (lr=1.0935e-03) (hash(x)=84817006)
+3050 train 4.065544 (lr=1.0830e-03) (hash(x)=71601811)
+3060 train 3.979119 (lr=1.0725e-03) (hash(x)=85499733)
+3070 train 4.035703 (lr=1.0620e-03) (hash(x)=78670408)
+3080 train 3.928230 (lr=1.0516e-03) (hash(x)=77120468)
+3090 train 4.036397 (lr=1.0412e-03) (hash(x)=77927426)
+3100 val loss 4.0364
+3100 val perplexity 56.6220
+3100 train 4.057170 (lr=1.0309e-03) (hash(x)=83998606)
+3110 train 3.947271 (lr=1.0207e-03) (hash(x)=77990218)
+3120 train 3.863707 (lr=1.0104e-03) (hash(x)=81623970)
+3130 train 3.838758 (lr=1.0002e-03) (hash(x)=75014781)
+3140 train 3.892305 (lr=9.9011e-04) (hash(x)=72591250)
+3150 train 3.904748 (lr=9.8002e-04) (hash(x)=81421847)
+3160 train 3.953761 (lr=9.6999e-04) (hash(x)=75112631)
+3170 train 4.048624 (lr=9.6000e-04) (hash(x)=87518033)
+3180 train 3.899073 (lr=9.5007e-04) (hash(x)=78648348)
+3190 train 4.015284 (lr=9.4019e-04) (hash(x)=78270029)
+3200 val loss 4.0143
+3200 val perplexity 55.3825
+3200 train 4.089229 (lr=9.3036e-04) (hash(x)=83380714)
+3210 train 4.273848 (lr=9.2058e-04) (hash(x)=67804991)
+3220 train 3.999769 (lr=9.1085e-04) (hash(x)=76968804)
+3230 train 3.929533 (lr=9.0118e-04) (hash(x)=75798670)
+3240 train 4.009161 (lr=8.9157e-04) (hash(x)=83192811)
+3250 train 3.943414 (lr=8.8201e-04) (hash(x)=81659789)
+3260 train 3.991196 (lr=8.7251e-04) (hash(x)=80622502)
+3270 train 3.957937 (lr=8.6307e-04) (hash(x)=85436511)
+3280 train 3.899525 (lr=8.5368e-04) (hash(x)=80589180)
+3290 train 4.035955 (lr=8.4435e-04) (hash(x)=85747193)
+3300 val loss 4.0099
+3300 val perplexity 55.1402
+3300 train 3.905746 (lr=8.3508e-04) (hash(x)=74635692)
+3310 train 3.894986 (lr=8.2588e-04) (hash(x)=82045455)
+3320 train 3.764027 (lr=8.1673e-04) (hash(x)=73249173)
+3330 train 3.721456 (lr=8.0764e-04) (hash(x)=71492338)
+3340 train 4.056547 (lr=7.9862e-04) (hash(x)=76374071)
+3350 train 4.028903 (lr=7.8966e-04) (hash(x)=80474064)
+3360 train 3.936519 (lr=7.8076e-04) (hash(x)=84390892)
+3370 train 3.980539 (lr=7.7192e-04) (hash(x)=83399949)
+3380 train 4.033925 (lr=7.6315e-04) (hash(x)=77648059)
+3390 train 4.005970 (lr=7.5445e-04) (hash(x)=80152701)
+3400 val loss 3.9879
+3400 val perplexity 53.9391
+3400 train 4.059071 (lr=7.4581e-04) (hash(x)=81952545)
+3410 train 4.010522 (lr=7.3724e-04) (hash(x)=80908993)
+3420 train 4.058775 (lr=7.2874e-04) (hash(x)=79610037)
+3430 train 3.921782 (lr=7.2030e-04) (hash(x)=87624382)
+3440 train 3.981541 (lr=7.1193e-04) (hash(x)=82336381)
+3450 train 3.956544 (lr=7.0363e-04) (hash(x)=87687835)
+3460 train 3.959896 (lr=6.9541e-04) (hash(x)=76376135)
+3470 train 3.873620 (lr=6.8725e-04) (hash(x)=79331391)
+3480 train 3.779906 (lr=6.7916e-04) (hash(x)=107002681)
+3490 train 3.859548 (lr=6.7114e-04) (hash(x)=77465514)
+3500 val loss 3.9855
+3500 val perplexity 53.8124
+3500 train 3.821056 (lr=6.6320e-04) (hash(x)=88237229)
+3510 train 3.889826 (lr=6.5533e-04) (hash(x)=81011739)
+3520 train 3.940324 (lr=6.4753e-04) (hash(x)=64643427)
+3530 train 4.217909 (lr=6.3981e-04) (hash(x)=78029539)
+3540 train 4.046010 (lr=6.3216e-04) (hash(x)=83188968)
+3550 train 4.066934 (lr=6.2458e-04) (hash(x)=86104185)
+3560 train 3.991896 (lr=6.1708e-04) (hash(x)=81993629)
+3570 train 3.964536 (lr=6.0966e-04) (hash(x)=72719368)
+3580 train 3.971838 (lr=6.0231e-04) (hash(x)=90091487)
+3590 train 3.857211 (lr=5.9504e-04) (hash(x)=77393152)
+3600 val loss 3.9650
+3600 val perplexity 52.7180
+3600 train 3.966793 (lr=5.8785e-04) (hash(x)=73103504)
+3610 train 3.975624 (lr=5.8074e-04) (hash(x)=84429400)
+3620 train 3.946476 (lr=5.7370e-04) (hash(x)=77021795)
+3630 train 3.901858 (lr=5.6675e-04) (hash(x)=92830605)
+3640 train 4.100672 (lr=5.5987e-04) (hash(x)=78313175)
+3650 train 3.900242 (lr=5.5308e-04) (hash(x)=85395549)
+3660 train 3.687737 (lr=5.4636e-04) (hash(x)=78114459)
+3670 train 3.819327 (lr=5.3973e-04) (hash(x)=74968316)
+3680 train 3.745363 (lr=5.3318e-04) (hash(x)=73358737)
+3690 train 3.852147 (lr=5.2671e-04) (hash(x)=76399442)
+3700 val loss 3.9663
+3700 val perplexity 52.7865
+3700 train 3.847090 (lr=5.2033e-04) (hash(x)=74039273)
+3710 train 4.027522 (lr=5.1402e-04) (hash(x)=79299680)
+3720 train 3.940223 (lr=5.0780e-04) (hash(x)=94701498)
+3730 train 4.020823 (lr=5.0167e-04) (hash(x)=75352071)
+3740 train 3.900819 (lr=4.9562e-04) (hash(x)=77909487)
+3750 train 4.007624 (lr=4.8965e-04) (hash(x)=73986730)
+3760 train 3.964523 (lr=4.8377e-04) (hash(x)=79325763)
+3770 train 4.012294 (lr=4.7798e-04) (hash(x)=72457818)
+3780 train 4.143191 (lr=4.7227e-04) (hash(x)=71775590)
+3790 train 4.020677 (lr=4.6665e-04) (hash(x)=82638943)
+3800 val loss 3.9487
+3800 val perplexity 51.8669
+3800 train 4.005657 (lr=4.6112e-04) (hash(x)=79965893)
+3810 train 3.938763 (lr=4.5567e-04) (hash(x)=72598235)
+3820 train 3.923743 (lr=4.5031e-04) (hash(x)=83113889)
+3830 train 3.979820 (lr=4.4504e-04) (hash(x)=74434590)
+3840 train 3.897368 (lr=4.3986e-04) (hash(x)=82860348)
+3850 train 3.837399 (lr=4.3477e-04) (hash(x)=78067565)
+3860 train 3.901173 (lr=4.2977e-04) (hash(x)=82592498)
+3870 train 3.832375 (lr=4.2486e-04) (hash(x)=81820733)
+3880 train 3.849381 (lr=4.2004e-04) (hash(x)=87709040)
+3890 train 3.892334 (lr=4.1530e-04) (hash(x)=70379093)
+3900 val loss 3.9420
+3900 val perplexity 51.5216
+3900 train 4.021118 (lr=4.1066e-04) (hash(x)=76597431)
+3910 train 3.899689 (lr=4.0611e-04) (hash(x)=90490716)
+3920 train 4.005448 (lr=4.0166e-04) (hash(x)=81970659)
+3930 train 3.991372 (lr=3.9729e-04) (hash(x)=81496334)
+3940 train 3.965497 (lr=3.9302e-04) (hash(x)=75717605)
+3950 train 3.914184 (lr=3.8884e-04) (hash(x)=82749357)
+3960 train 3.946499 (lr=3.8475e-04) (hash(x)=89355157)
+3970 train 3.899633 (lr=3.8076e-04) (hash(x)=78980403)
+3980 train 3.872687 (lr=3.7685e-04) (hash(x)=76627217)
+3990 train 3.875857 (lr=3.7305e-04) (hash(x)=72412879)
+4000 val loss 3.9360
+4000 val perplexity 51.2142
+4000 train 3.885509 (lr=3.6933e-04) (hash(x)=83018142)
+4010 train 3.900593 (lr=3.6572e-04) (hash(x)=81272436)
+4020 train 3.831450 (lr=3.6219e-04) (hash(x)=85497482)
+4030 train 3.899762 (lr=3.5876e-04) (hash(x)=85241734)
+4040 train 3.868155 (lr=3.5543e-04) (hash(x)=77925307)
+4050 train 3.867801 (lr=3.5219e-04) (hash(x)=84826179)
+4060 train 3.928291 (lr=3.4905e-04) (hash(x)=83606764)
+4070 train 3.836564 (lr=3.4600e-04) (hash(x)=80567590)
+4080 train 3.889646 (lr=3.4305e-04) (hash(x)=76860998)
+4090 train 3.929055 (lr=3.4019e-04) (hash(x)=74902328)
+4100 val loss 3.9238
+4100 val perplexity 50.5898
+4100 train 3.972485 (lr=3.3744e-04) (hash(x)=82832041)
+4110 train 3.853017 (lr=3.3477e-04) (hash(x)=79143262)
+4120 train 3.874489 (lr=3.3221e-04) (hash(x)=77038149)
+4130 train 4.061419 (lr=3.2974e-04) (hash(x)=86339074)
+4140 train 4.127346 (lr=3.2737e-04) (hash(x)=76686216)
+4150 train 3.898166 (lr=3.2510e-04) (hash(x)=70522682)
+4160 train 3.863660 (lr=3.2292e-04) (hash(x)=90958555)
+4170 train 3.892589 (lr=3.2085e-04) (hash(x)=91463532)
+4180 train 3.877108 (lr=3.1887e-04) (hash(x)=81959329)
+4190 train 3.832473 (lr=3.1699e-04) (hash(x)=83146752)
+4200 val loss 3.9265
+4200 val perplexity 50.7272
+4200 train 3.822560 (lr=3.1520e-04) (hash(x)=78361715)
+4210 train 3.965407 (lr=3.1352e-04) (hash(x)=87364889)
+4220 train 3.820172 (lr=3.1193e-04) (hash(x)=70465156)
+4230 train 3.974707 (lr=3.1044e-04) (hash(x)=84524081)
+4240 train 3.930824 (lr=3.0905e-04) (hash(x)=77824868)
+4250 train 3.941558 (lr=3.0776e-04) (hash(x)=81710711)
+4260 train 3.926866 (lr=3.0657e-04) (hash(x)=76362728)
+4270 train 3.915299 (lr=3.0548e-04) (hash(x)=83115208)
+4280 train 3.829508 (lr=3.0449e-04) (hash(x)=87218314)
+4290 train 3.987516 (lr=3.0359e-04) (hash(x)=74582673)
+4300 val loss 3.9204
+4300 val perplexity 50.4226
+4300 train 3.864719 (lr=3.0280e-04) (hash(x)=77379615)
+4310 train 3.835655 (lr=3.0210e-04) (hash(x)=78669579)
+4320 train 3.953920 (lr=3.0150e-04) (hash(x)=83066608)
+4330 train 3.918398 (lr=3.0101e-04) (hash(x)=83037340)
+4340 train 3.954510 (lr=3.0061e-04) (hash(x)=82849771)
+4350 train 3.891350 (lr=3.0031e-04) (hash(x)=76693985)
+4360 train 3.983877 (lr=3.0011e-04) (hash(x)=77745394)
+4370 train 3.944788 (lr=3.0001e-04) (hash(x)=79954388)
+4374 val loss 3.9142
+4374 val perplexity 50.1103

12_head_baseline_lr_30e-4_head_dim_22_fixed/model_04374.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:68b8b61105eae85648f4622a3ca5416e8306af374ccba2764f0bfb70c5b3ee9f
+size 96858242

12_head_baseline_lr_30e-4_head_dim_22_fixed/optimizer_04374.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:9194fc582e4ee3578e1fe4ab88ec8b99f82d8a38bd8ddb6de835ee9b5c90c13a
+size 187435910