Upload folder using huggingface_hub

12_head_baseline_lr_16e-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_16e-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": 0.0016, "decay_lr": true, "readout_zero_init": false, "query_zero_init": false, "l1_loss": false, "debugpy": false, "key": "12_head_baseline_lr_16e-4"}

12_head_baseline_lr_16e-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_16e-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 10.908087 (lr=2.2378e-06) (hash(x)=93356070)
+10 train 10.585678 (lr=2.4615e-05) (hash(x)=91253010)
+20 train 10.376489 (lr=4.6993e-05) (hash(x)=74934453)
+30 train 10.221272 (lr=6.9371e-05) (hash(x)=79278034)
+40 train 9.966092 (lr=9.1748e-05) (hash(x)=80107892)
+50 train 9.673457 (lr=1.1413e-04) (hash(x)=70745428)
+60 train 9.364157 (lr=1.3650e-04) (hash(x)=80077589)
+70 train 8.911022 (lr=1.5888e-04) (hash(x)=76213766)
+80 train 8.597994 (lr=1.8126e-04) (hash(x)=83218328)
+90 train 8.255257 (lr=2.0364e-04) (hash(x)=74855845)
+100 val loss 7.9619
+100 val perplexity 2869.4958
+100 train 7.882020 (lr=2.2601e-04) (hash(x)=82814902)
+110 train 7.654025 (lr=2.4839e-04) (hash(x)=84907741)
+120 train 7.460617 (lr=2.7077e-04) (hash(x)=82613223)
+130 train 7.258423 (lr=2.9315e-04) (hash(x)=83540876)
+140 train 7.208622 (lr=3.1552e-04) (hash(x)=75095216)
+150 train 7.224401 (lr=3.3790e-04) (hash(x)=97190944)
+160 train 7.075915 (lr=3.6028e-04) (hash(x)=82117809)
+170 train 6.947970 (lr=3.8266e-04) (hash(x)=70514724)
+180 train 6.858750 (lr=4.0503e-04) (hash(x)=81029624)
+190 train 6.757197 (lr=4.2741e-04) (hash(x)=78996841)
+200 val loss 6.6985
+200 val perplexity 811.2219
+200 train 6.703852 (lr=4.4979e-04) (hash(x)=79845097)
+210 train 6.577518 (lr=4.7217e-04) (hash(x)=78997683)
+220 train 6.465341 (lr=4.9455e-04) (hash(x)=74895865)
+230 train 6.661087 (lr=5.1692e-04) (hash(x)=80933276)
+240 train 6.347452 (lr=5.3930e-04) (hash(x)=77664606)
+250 train 6.369431 (lr=5.6168e-04) (hash(x)=75903930)
+260 train 6.283101 (lr=5.8406e-04) (hash(x)=82996853)
+270 train 6.215262 (lr=6.0643e-04) (hash(x)=73269568)
+280 train 6.162675 (lr=6.2881e-04) (hash(x)=82906427)
+290 train 6.078569 (lr=6.5119e-04) (hash(x)=71797895)
+300 val loss 6.2113
+300 val perplexity 498.3498
+300 train 5.940141 (lr=6.7357e-04) (hash(x)=85232249)
+310 train 5.895428 (lr=6.9594e-04) (hash(x)=75252489)
+320 train 5.989726 (lr=7.1832e-04) (hash(x)=74394644)
+330 train 6.053756 (lr=7.4070e-04) (hash(x)=82772910)
+340 train 6.087182 (lr=7.6308e-04) (hash(x)=81627464)
+350 train 6.060582 (lr=7.8545e-04) (hash(x)=84678053)
+360 train 6.021919 (lr=8.0783e-04) (hash(x)=81884128)
+370 train 5.993815 (lr=8.3021e-04) (hash(x)=76379242)
+380 train 5.770981 (lr=8.5259e-04) (hash(x)=98182875)
+390 train 5.765016 (lr=8.7497e-04) (hash(x)=79710436)
+400 val loss 5.8146
+400 val perplexity 335.1635
+400 train 5.768349 (lr=8.9734e-04) (hash(x)=79841071)
+410 train 5.680778 (lr=9.1972e-04) (hash(x)=75844151)
+420 train 5.684371 (lr=9.4210e-04) (hash(x)=73125036)
+430 train 5.718620 (lr=9.6448e-04) (hash(x)=84214858)
+440 train 5.610497 (lr=9.8685e-04) (hash(x)=80456994)
+450 train 5.574496 (lr=1.0092e-03) (hash(x)=76735962)
+460 train 5.515234 (lr=1.0316e-03) (hash(x)=81845446)
+470 train 5.331460 (lr=1.0540e-03) (hash(x)=76094689)
+480 train 5.462737 (lr=1.0764e-03) (hash(x)=83806686)
+490 train 5.293181 (lr=1.0987e-03) (hash(x)=84690227)
+500 val loss 5.5773
+500 val perplexity 264.3541
+500 train 5.346482 (lr=1.1211e-03) (hash(x)=71851938)
+510 train 5.434687 (lr=1.1435e-03) (hash(x)=77159346)
+520 train 5.464775 (lr=1.1659e-03) (hash(x)=80755753)
+530 train 5.584145 (lr=1.1883e-03) (hash(x)=75379679)
+540 train 5.473958 (lr=1.2106e-03) (hash(x)=82458619)
+550 train 5.466833 (lr=1.2330e-03) (hash(x)=86366396)
+560 train 5.436893 (lr=1.2554e-03) (hash(x)=85095044)
+570 train 5.428627 (lr=1.2778e-03) (hash(x)=78385159)
+580 train 5.373922 (lr=1.3001e-03) (hash(x)=79342394)
+590 train 5.383455 (lr=1.3225e-03) (hash(x)=70782192)
+600 val loss 5.3651
+600 val perplexity 213.8199
+600 train 5.339742 (lr=1.3449e-03) (hash(x)=76130353)
+610 train 5.240526 (lr=1.3673e-03) (hash(x)=74778440)
+620 train 5.219110 (lr=1.3897e-03) (hash(x)=79129709)
+630 train 5.125548 (lr=1.4120e-03) (hash(x)=76469962)
+640 train 5.127736 (lr=1.4344e-03) (hash(x)=78288049)
+650 train 5.102682 (lr=1.4568e-03) (hash(x)=76641388)
+660 train 5.008476 (lr=1.4792e-03) (hash(x)=75937906)
+670 train 5.015782 (lr=1.5015e-03) (hash(x)=74807157)
+680 train 4.920259 (lr=1.5239e-03) (hash(x)=77490144)
+690 train 4.951772 (lr=1.5463e-03) (hash(x)=72900124)
+700 val loss 5.2004
+700 val perplexity 181.3505
+700 train 4.910363 (lr=1.5687e-03) (hash(x)=73424218)
+710 train 5.168519 (lr=1.5910e-03) (hash(x)=86052345)
+720 train 5.131242 (lr=1.6000e-03) (hash(x)=77462613)
+730 train 5.142790 (lr=1.5999e-03) (hash(x)=79027471)
+740 train 5.113450 (lr=1.5998e-03) (hash(x)=78149992)
+750 train 5.091600 (lr=1.5997e-03) (hash(x)=89147499)
+760 train 5.073165 (lr=1.5995e-03) (hash(x)=74931538)
+770 train 5.035950 (lr=1.5992e-03) (hash(x)=89648204)
+780 train 5.069626 (lr=1.5989e-03) (hash(x)=83708293)
+790 train 5.072124 (lr=1.5985e-03) (hash(x)=84336512)
+800 val loss 5.0365
+800 val perplexity 153.9309
+800 train 4.969589 (lr=1.5981e-03) (hash(x)=75025285)
+810 train 4.874999 (lr=1.5976e-03) (hash(x)=79436536)
+820 train 4.868384 (lr=1.5971e-03) (hash(x)=90467390)
+830 train 4.927909 (lr=1.5965e-03) (hash(x)=77292020)
+840 train 4.835528 (lr=1.5959e-03) (hash(x)=75568927)
+850 train 4.862949 (lr=1.5952e-03) (hash(x)=79671781)
+860 train 4.935530 (lr=1.5944e-03) (hash(x)=84280943)
+870 train 4.918903 (lr=1.5936e-03) (hash(x)=80901294)
+880 train 5.057272 (lr=1.5928e-03) (hash(x)=81437584)
+890 train 4.901940 (lr=1.5919e-03) (hash(x)=73893778)
+900 val loss 4.9172
+900 val perplexity 136.6255
+900 train 4.833457 (lr=1.5909e-03) (hash(x)=74987794)
+910 train 4.840896 (lr=1.5899e-03) (hash(x)=93721374)
+920 train 4.832560 (lr=1.5889e-03) (hash(x)=79149678)
+930 train 4.781965 (lr=1.5878e-03) (hash(x)=83179387)
+940 train 4.785013 (lr=1.5866e-03) (hash(x)=89009978)
+950 train 4.751117 (lr=1.5854e-03) (hash(x)=76483735)
+960 train 4.681861 (lr=1.5841e-03) (hash(x)=78638850)
+970 train 4.707286 (lr=1.5828e-03) (hash(x)=90415206)
+980 train 4.642100 (lr=1.5815e-03) (hash(x)=90819195)
+990 train 4.778181 (lr=1.5800e-03) (hash(x)=85938852)
+1000 val loss 4.7591
+1000 val perplexity 116.6384
+1000 train 4.812365 (lr=1.5786e-03) (hash(x)=82436789)
+1010 train 4.803889 (lr=1.5770e-03) (hash(x)=63022149)
+1020 train 4.754935 (lr=1.5755e-03) (hash(x)=83501199)
+1030 train 4.726191 (lr=1.5738e-03) (hash(x)=77093326)
+1040 train 4.594853 (lr=1.5722e-03) (hash(x)=73437559)
+1050 train 4.689940 (lr=1.5704e-03) (hash(x)=84550388)
+1060 train 4.693380 (lr=1.5687e-03) (hash(x)=86184566)
+1070 train 4.666758 (lr=1.5668e-03) (hash(x)=85644922)
+1080 train 4.602300 (lr=1.5650e-03) (hash(x)=86179801)
+1090 train 4.574203 (lr=1.5630e-03) (hash(x)=82868303)
+1100 val loss 4.6391
+1100 val perplexity 103.4526
+1100 train 4.671437 (lr=1.5610e-03) (hash(x)=96780388)
+1110 train 4.597684 (lr=1.5590e-03) (hash(x)=87223122)
+1120 train 4.568095 (lr=1.5569e-03) (hash(x)=77292786)
+1130 train 4.577935 (lr=1.5548e-03) (hash(x)=88761375)
+1140 train 4.496289 (lr=1.5526e-03) (hash(x)=80956468)
+1150 train 4.658935 (lr=1.5504e-03) (hash(x)=84725462)
+1160 train 4.643101 (lr=1.5481e-03) (hash(x)=76098113)
+1170 train 4.620548 (lr=1.5458e-03) (hash(x)=84387685)
+1180 train 4.562144 (lr=1.5434e-03) (hash(x)=76120321)
+1190 train 4.527261 (lr=1.5410e-03) (hash(x)=85974065)
+1200 val loss 4.5491
+1200 val perplexity 94.5462
+1200 train 4.433625 (lr=1.5385e-03) (hash(x)=79340644)
+1210 train 4.517334 (lr=1.5360e-03) (hash(x)=88345025)
+1220 train 4.429471 (lr=1.5334e-03) (hash(x)=86406230)
+1230 train 4.413530 (lr=1.5308e-03) (hash(x)=80608975)
+1240 train 4.465598 (lr=1.5281e-03) (hash(x)=81635225)
+1250 train 4.304112 (lr=1.5254e-03) (hash(x)=82126376)
+1260 train 4.471310 (lr=1.5226e-03) (hash(x)=83737972)
+1270 train 4.377076 (lr=1.5198e-03) (hash(x)=81132345)
+1280 train 4.466842 (lr=1.5170e-03) (hash(x)=85850404)
+1290 train 4.473235 (lr=1.5141e-03) (hash(x)=83490640)
+1300 val loss 4.4835
+1300 val perplexity 88.5476
+1300 train 4.456028 (lr=1.5111e-03) (hash(x)=77545187)
+1310 train 4.505247 (lr=1.5081e-03) (hash(x)=86412685)
+1320 train 4.396567 (lr=1.5051e-03) (hash(x)=86429640)
+1330 train 4.537487 (lr=1.5020e-03) (hash(x)=79645281)
+1340 train 4.487130 (lr=1.4989e-03) (hash(x)=82668541)
+1350 train 4.348321 (lr=1.4957e-03) (hash(x)=73627248)
+1360 train 4.439166 (lr=1.4924e-03) (hash(x)=74845133)
+1370 train 4.347093 (lr=1.4892e-03) (hash(x)=77850497)
+1380 train 4.367333 (lr=1.4859e-03) (hash(x)=76143954)
+1390 train 4.382529 (lr=1.4825e-03) (hash(x)=80675544)
+1400 val loss 4.4446
+1400 val perplexity 85.1666
+1400 train 4.357868 (lr=1.4791e-03) (hash(x)=76010938)
+1410 train 4.402906 (lr=1.4756e-03) (hash(x)=81260300)
+1420 train 4.514007 (lr=1.4721e-03) (hash(x)=83803629)
+1430 train 4.347075 (lr=1.4686e-03) (hash(x)=83840110)
+1440 train 4.540732 (lr=1.4650e-03) (hash(x)=97711831)
+1450 train 4.406323 (lr=1.4614e-03) (hash(x)=81110139)
+1460 train 4.334305 (lr=1.4577e-03) (hash(x)=90942839)
+1470 train 4.365466 (lr=1.4540e-03) (hash(x)=80721564)
+1480 train 4.466977 (lr=1.4503e-03) (hash(x)=77852059)
+1490 train 4.357713 (lr=1.4465e-03) (hash(x)=80011365)
+1500 val loss 4.3961
+1500 val perplexity 81.1324
+1500 train 4.386594 (lr=1.4426e-03) (hash(x)=84496142)
+1510 train 4.320297 (lr=1.4388e-03) (hash(x)=70186729)
+1520 train 4.342737 (lr=1.4348e-03) (hash(x)=74854227)
+1530 train 4.254687 (lr=1.4309e-03) (hash(x)=77468161)
+1540 train 4.270769 (lr=1.4269e-03) (hash(x)=87853059)
+1550 train 4.199529 (lr=1.4228e-03) (hash(x)=71225436)
+1560 train 4.302473 (lr=1.4188e-03) (hash(x)=84072783)
+1570 train 4.355700 (lr=1.4147e-03) (hash(x)=72723098)
+1580 train 4.367540 (lr=1.4105e-03) (hash(x)=90409866)
+1590 train 4.440462 (lr=1.4063e-03) (hash(x)=87481378)
+1600 val loss 4.3635
+1600 val perplexity 78.5312
+1600 train 4.420328 (lr=1.4021e-03) (hash(x)=77643862)
+1610 train 4.309007 (lr=1.3978e-03) (hash(x)=88862575)
+1620 train 4.331511 (lr=1.3935e-03) (hash(x)=84612581)
+1630 train 4.355415 (lr=1.3891e-03) (hash(x)=87075989)
+1640 train 4.241297 (lr=1.3847e-03) (hash(x)=88277361)
+1650 train 4.359631 (lr=1.3803e-03) (hash(x)=78750236)
+1660 train 4.286689 (lr=1.3758e-03) (hash(x)=82604581)
+1670 train 4.120634 (lr=1.3713e-03) (hash(x)=68482265)
+1680 train 4.243147 (lr=1.3668e-03) (hash(x)=75088835)
+1690 train 4.184025 (lr=1.3622e-03) (hash(x)=66060989)
+1700 val loss 4.3563
+1700 val perplexity 77.9717
+1700 train 4.350269 (lr=1.3576e-03) (hash(x)=79986754)
+1710 train 4.339150 (lr=1.3530e-03) (hash(x)=83657930)
+1720 train 4.278282 (lr=1.3483e-03) (hash(x)=81754135)
+1730 train 4.358900 (lr=1.3436e-03) (hash(x)=78904427)
+1740 train 4.269311 (lr=1.3389e-03) (hash(x)=85920177)
+1750 train 4.320100 (lr=1.3341e-03) (hash(x)=86573211)
+1760 train 4.271739 (lr=1.3293e-03) (hash(x)=81737128)
+1770 train 4.262326 (lr=1.3244e-03) (hash(x)=77400968)
+1780 train 4.277096 (lr=1.3195e-03) (hash(x)=73545497)
+1790 train 4.523623 (lr=1.3146e-03) (hash(x)=71641943)
+1800 val loss 4.3086
+1800 val perplexity 74.3364
+1800 train 4.368937 (lr=1.3097e-03) (hash(x)=87819781)
+1810 train 4.229906 (lr=1.3047e-03) (hash(x)=86870770)
+1820 train 4.238994 (lr=1.2997e-03) (hash(x)=82522211)
+1830 train 4.321395 (lr=1.2947e-03) (hash(x)=61947437)
+1840 train 4.203382 (lr=1.2896e-03) (hash(x)=79865406)
+1850 train 4.171458 (lr=1.2845e-03) (hash(x)=79828721)
+1860 train 4.307742 (lr=1.2794e-03) (hash(x)=80869571)
+1870 train 4.296415 (lr=1.2742e-03) (hash(x)=73780971)
+1880 train 4.255961 (lr=1.2690e-03) (hash(x)=79249549)
+1890 train 4.316472 (lr=1.2638e-03) (hash(x)=81041904)
+1900 val loss 4.2856
+1900 val perplexity 72.6490
+1900 train 4.381341 (lr=1.2586e-03) (hash(x)=82456430)
+1910 train 4.154871 (lr=1.2533e-03) (hash(x)=82222135)
+1920 train 4.186582 (lr=1.2480e-03) (hash(x)=70033249)
+1930 train 4.168818 (lr=1.2427e-03) (hash(x)=72887360)
+1940 train 4.218186 (lr=1.2373e-03) (hash(x)=83251100)
+1950 train 4.256623 (lr=1.2319e-03) (hash(x)=79660266)
+1960 train 4.139914 (lr=1.2265e-03) (hash(x)=78919068)
+1970 train 4.239873 (lr=1.2211e-03) (hash(x)=94690431)
+1980 train 4.079448 (lr=1.2157e-03) (hash(x)=88153756)
+1990 train 4.126514 (lr=1.2102e-03) (hash(x)=77195688)
+2000 val loss 4.2585
+2000 val perplexity 70.7026
+2000 train 4.226669 (lr=1.2047e-03) (hash(x)=81308591)
+2010 train 4.272140 (lr=1.1991e-03) (hash(x)=80205479)
+2020 train 4.186926 (lr=1.1936e-03) (hash(x)=81008704)
+2030 train 4.253856 (lr=1.1880e-03) (hash(x)=82914358)
+2040 train 4.266423 (lr=1.1824e-03) (hash(x)=88064399)
+2050 train 4.266945 (lr=1.1768e-03) (hash(x)=82889390)
+2060 train 4.332013 (lr=1.1712e-03) (hash(x)=80913554)
+2070 train 4.180688 (lr=1.1655e-03) (hash(x)=78121791)
+2080 train 4.310154 (lr=1.1598e-03) (hash(x)=75356657)
+2090 train 4.140240 (lr=1.1541e-03) (hash(x)=74048412)
+2100 val loss 4.2341
+2100 val perplexity 69.0009
+2100 train 4.061413 (lr=1.1484e-03) (hash(x)=68928225)
+2110 train 4.111023 (lr=1.1426e-03) (hash(x)=93734745)
+2120 train 4.108327 (lr=1.1369e-03) (hash(x)=72878164)
+2130 train 4.231514 (lr=1.1311e-03) (hash(x)=71508226)
+2140 train 4.188633 (lr=1.1253e-03) (hash(x)=84168671)
+2150 train 4.130698 (lr=1.1195e-03) (hash(x)=74673639)
+2160 train 4.275507 (lr=1.1136e-03) (hash(x)=75470031)
+2170 train 4.237580 (lr=1.1078e-03) (hash(x)=74307890)
+2180 train 4.350930 (lr=1.1019e-03) (hash(x)=77214245)
+2190 train 4.130011 (lr=1.0960e-03) (hash(x)=88628359)
+2200 val loss 4.2140
+2200 val perplexity 67.6297
+2200 train 4.150084 (lr=1.0901e-03) (hash(x)=74779126)
+2210 train 4.179159 (lr=1.0842e-03) (hash(x)=79817976)
+2220 train 4.136503 (lr=1.0783e-03) (hash(x)=75258996)
+2230 train 4.113448 (lr=1.0723e-03) (hash(x)=88164047)
+2240 train 4.096349 (lr=1.0663e-03) (hash(x)=78580686)
+2250 train 4.034373 (lr=1.0604e-03) (hash(x)=76510617)
+2260 train 4.113445 (lr=1.0544e-03) (hash(x)=76043966)
+2270 train 4.090684 (lr=1.0484e-03) (hash(x)=73062098)
+2280 train 4.314728 (lr=1.0424e-03) (hash(x)=81885909)
+2290 train 4.246823 (lr=1.0363e-03) (hash(x)=81481741)
+2300 val loss 4.1916
+2300 val perplexity 66.1310
+2300 train 4.107804 (lr=1.0303e-03) (hash(x)=82104275)
+2310 train 4.162648 (lr=1.0243e-03) (hash(x)=82137309)
+2320 train 4.138818 (lr=1.0182e-03) (hash(x)=86741167)
+2330 train 4.154326 (lr=1.0121e-03) (hash(x)=84796263)
+2340 train 4.121413 (lr=1.0060e-03) (hash(x)=169966529)
+2350 train 3.998680 (lr=9.9995e-04) (hash(x)=80006061)
+2360 train 4.055999 (lr=9.9385e-04) (hash(x)=71090999)
+2370 train 4.185889 (lr=9.8775e-04) (hash(x)=86641850)
+2380 train 4.082513 (lr=9.8163e-04) (hash(x)=84573253)
+2390 train 4.120616 (lr=9.7551e-04) (hash(x)=74483764)
+2400 val loss 4.1802
+2400 val perplexity 65.3820
+2400 train 4.088716 (lr=9.6938e-04) (hash(x)=78327659)
+2410 train 4.154675 (lr=9.6325e-04) (hash(x)=77273627)
+2420 train 4.019368 (lr=9.5710e-04) (hash(x)=76938049)
+2430 train 4.025048 (lr=9.5096e-04) (hash(x)=98449442)
+2440 train 3.882285 (lr=9.4480e-04) (hash(x)=78157797)
+2450 train 3.928544 (lr=9.3865e-04) (hash(x)=80637582)
+2460 train 3.914698 (lr=9.3248e-04) (hash(x)=80023854)
+2470 train 3.977319 (lr=9.2632e-04) (hash(x)=86678884)
+2480 train 4.200157 (lr=9.2015e-04) (hash(x)=84086469)
+2490 train 4.220301 (lr=9.1398e-04) (hash(x)=83131141)
+2500 val loss 4.1579
+2500 val perplexity 63.9364
+2500 train 4.135242 (lr=9.0780e-04) (hash(x)=82583497)
+2510 train 4.196637 (lr=9.0163e-04) (hash(x)=79164326)
+2520 train 4.106903 (lr=8.9545e-04) (hash(x)=76456503)
+2530 train 4.162382 (lr=8.8927e-04) (hash(x)=82001228)
+2540 train 4.199288 (lr=8.8309e-04) (hash(x)=75615595)
+2550 train 4.142393 (lr=8.7691e-04) (hash(x)=78397869)
+2560 train 4.144168 (lr=8.7073e-04) (hash(x)=75547032)
+2570 train 4.062757 (lr=8.6455e-04) (hash(x)=89201025)
+2580 train 4.045299 (lr=8.5837e-04) (hash(x)=89856704)
+2590 train 4.052350 (lr=8.5220e-04) (hash(x)=82175682)
+2600 val loss 4.1391
+2600 val perplexity 62.7448
+2600 train 4.059687 (lr=8.4602e-04) (hash(x)=78312826)
+2610 train 4.052433 (lr=8.3985e-04) (hash(x)=77066588)
+2620 train 3.889783 (lr=8.3368e-04) (hash(x)=78666061)
+2630 train 3.893958 (lr=8.2752e-04) (hash(x)=93762143)
+2640 train 3.910841 (lr=8.2135e-04) (hash(x)=83191587)
+2650 train 3.834166 (lr=8.1520e-04) (hash(x)=87169585)
+2660 train 4.044852 (lr=8.0904e-04) (hash(x)=86426388)
+2670 train 4.188488 (lr=8.0290e-04) (hash(x)=76692638)
+2680 train 4.100641 (lr=7.9675e-04) (hash(x)=77446063)
+2690 train 4.100773 (lr=7.9062e-04) (hash(x)=79809050)
+2700 val loss 4.1368
+2700 val perplexity 62.6026
+2700 train 4.028737 (lr=7.8449e-04) (hash(x)=83116823)
+2710 train 4.145290 (lr=7.7837e-04) (hash(x)=75622148)
+2720 train 4.066750 (lr=7.7225e-04) (hash(x)=80690512)
+2730 train 4.141237 (lr=7.6615e-04) (hash(x)=78009984)
+2740 train 4.093812 (lr=7.6005e-04) (hash(x)=74205488)
+2750 train 4.098813 (lr=7.5396e-04) (hash(x)=91013332)
+2760 train 4.022225 (lr=7.4788e-04) (hash(x)=76954961)
+2770 train 4.033100 (lr=7.4181e-04) (hash(x)=79390317)
+2780 train 3.899426 (lr=7.3575e-04) (hash(x)=70168783)
+2790 train 4.061967 (lr=7.2970e-04) (hash(x)=76028417)
+2800 val loss 4.1149
+2800 val perplexity 61.2437
+2800 train 3.887700 (lr=7.2366e-04) (hash(x)=77656050)
+2810 train 3.931183 (lr=7.1763e-04) (hash(x)=94248216)
+2820 train 3.998085 (lr=7.1162e-04) (hash(x)=78305078)
+2830 train 3.911000 (lr=7.0562e-04) (hash(x)=79948848)
+2840 train 3.776920 (lr=6.9963e-04) (hash(x)=85341024)
+2850 train 4.082932 (lr=6.9365e-04) (hash(x)=78735170)
+2860 train 4.332425 (lr=6.8769e-04) (hash(x)=71616419)
+2870 train 4.105308 (lr=6.8174e-04) (hash(x)=78656517)
+2880 train 4.120984 (lr=6.7581e-04) (hash(x)=80073987)
+2890 train 4.046520 (lr=6.6989e-04) (hash(x)=76894809)
+2900 val loss 4.0910
+2900 val perplexity 59.7982
+2900 train 4.024076 (lr=6.6398e-04) (hash(x)=80499838)
+2910 train 4.078043 (lr=6.5810e-04) (hash(x)=72673354)
+2920 train 4.060029 (lr=6.5222e-04) (hash(x)=84265768)
+2930 train 4.088078 (lr=6.4637e-04) (hash(x)=79612060)
+2940 train 3.945648 (lr=6.4053e-04) (hash(x)=74970087)
+2950 train 4.095040 (lr=6.3471e-04) (hash(x)=84166818)
+2960 train 4.084397 (lr=6.2891e-04) (hash(x)=89410221)
+2970 train 4.026107 (lr=6.2313e-04) (hash(x)=75672566)
+2980 train 3.882633 (lr=6.1737e-04) (hash(x)=81760314)
+2990 train 3.855840 (lr=6.1162e-04) (hash(x)=80605200)
+3000 val loss 4.0906
+3000 val perplexity 59.7777
+3000 train 3.855405 (lr=6.0590e-04) (hash(x)=83804735)
+3010 train 3.787741 (lr=6.0019e-04) (hash(x)=77015303)
+3020 train 3.848751 (lr=5.9451e-04) (hash(x)=81464523)
+3030 train 4.049000 (lr=5.8884e-04) (hash(x)=80168230)
+3040 train 3.989383 (lr=5.8320e-04) (hash(x)=84817006)
+3050 train 4.096107 (lr=5.7758e-04) (hash(x)=71601811)
+3060 train 4.013893 (lr=5.7199e-04) (hash(x)=85499733)
+3070 train 4.070004 (lr=5.6641e-04) (hash(x)=78670408)
+3080 train 3.962082 (lr=5.6086e-04) (hash(x)=77120468)
+3090 train 4.069730 (lr=5.5533e-04) (hash(x)=77927426)
+3100 val loss 4.0709
+3100 val perplexity 58.6076
+3100 train 4.076867 (lr=5.4983e-04) (hash(x)=83998606)
+3110 train 3.980214 (lr=5.4435e-04) (hash(x)=77990218)
+3120 train 3.895679 (lr=5.3889e-04) (hash(x)=81623970)
+3130 train 3.870850 (lr=5.3346e-04) (hash(x)=75014781)
+3140 train 3.925586 (lr=5.2806e-04) (hash(x)=72591250)
+3150 train 3.941037 (lr=5.2268e-04) (hash(x)=81421847)
+3160 train 3.988408 (lr=5.1733e-04) (hash(x)=75112631)
+3170 train 4.084939 (lr=5.1200e-04) (hash(x)=87518033)
+3180 train 3.935179 (lr=5.0670e-04) (hash(x)=78648348)
+3190 train 4.055359 (lr=5.0143e-04) (hash(x)=78270029)
+3200 val loss 4.0538
+3200 val perplexity 57.6161
+3200 train 4.127747 (lr=4.9619e-04) (hash(x)=83380714)
+3210 train 4.309574 (lr=4.9097e-04) (hash(x)=67804991)
+3220 train 4.039835 (lr=4.8579e-04) (hash(x)=76968804)
+3230 train 3.974635 (lr=4.8063e-04) (hash(x)=75798670)
+3240 train 4.049366 (lr=4.7550e-04) (hash(x)=83192811)
+3250 train 3.985260 (lr=4.7041e-04) (hash(x)=81659789)
+3260 train 4.026311 (lr=4.6534e-04) (hash(x)=80622502)
+3270 train 3.996602 (lr=4.6030e-04) (hash(x)=85436511)
+3280 train 3.938594 (lr=4.5530e-04) (hash(x)=80589180)
+3290 train 4.083184 (lr=4.5032e-04) (hash(x)=85747193)
+3300 val loss 4.0503
+3300 val perplexity 57.4118
+3300 train 3.947705 (lr=4.4538e-04) (hash(x)=74635692)
+3310 train 3.935607 (lr=4.4047e-04) (hash(x)=82045455)
+3320 train 3.805819 (lr=4.3559e-04) (hash(x)=73249173)
+3330 train 3.764654 (lr=4.3074e-04) (hash(x)=71492338)
+3340 train 4.097351 (lr=4.2593e-04) (hash(x)=76374071)
+3350 train 4.063389 (lr=4.2115e-04) (hash(x)=80474064)
+3360 train 3.974170 (lr=4.1640e-04) (hash(x)=84390892)
+3370 train 4.025730 (lr=4.1169e-04) (hash(x)=83399949)
+3380 train 4.074488 (lr=4.0702e-04) (hash(x)=77648059)
+3390 train 4.046433 (lr=4.0237e-04) (hash(x)=80152701)
+3400 val loss 4.0309
+3400 val perplexity 56.3093
+3400 train 4.097097 (lr=3.9777e-04) (hash(x)=81952545)
+3410 train 4.057642 (lr=3.9320e-04) (hash(x)=80908993)
+3420 train 4.097146 (lr=3.8866e-04) (hash(x)=79610037)
+3430 train 3.959424 (lr=3.8416e-04) (hash(x)=87624382)
+3440 train 4.020471 (lr=3.7970e-04) (hash(x)=82336381)
+3450 train 3.993855 (lr=3.7527e-04) (hash(x)=87687835)
+3460 train 3.997487 (lr=3.7088e-04) (hash(x)=76376135)
+3470 train 3.913195 (lr=3.6653e-04) (hash(x)=79331391)
+3480 train 3.820508 (lr=3.6222e-04) (hash(x)=107002681)
+3490 train 3.900601 (lr=3.5794e-04) (hash(x)=77465514)
+3500 val loss 4.0277
+3500 val perplexity 56.1303
+3500 train 3.864505 (lr=3.5371e-04) (hash(x)=88237229)
+3510 train 3.933035 (lr=3.4951e-04) (hash(x)=81011739)
+3520 train 3.978884 (lr=3.4535e-04) (hash(x)=64643427)
+3530 train 4.257866 (lr=3.4123e-04) (hash(x)=78029539)
+3540 train 4.091689 (lr=3.3715e-04) (hash(x)=83188968)
+3550 train 4.111199 (lr=3.3311e-04) (hash(x)=86104185)
+3560 train 4.039367 (lr=3.2911e-04) (hash(x)=81993629)
+3570 train 4.005911 (lr=3.2515e-04) (hash(x)=72719368)
+3580 train 4.020073 (lr=3.2123e-04) (hash(x)=90091487)
+3590 train 3.909998 (lr=3.1736e-04) (hash(x)=77393152)
+3600 val loss 4.0113
+3600 val perplexity 55.2171
+3600 train 4.010381 (lr=3.1352e-04) (hash(x)=73103504)
+3610 train 4.018916 (lr=3.0973e-04) (hash(x)=84429400)
+3620 train 3.989973 (lr=3.0598e-04) (hash(x)=77021795)
+3630 train 3.944215 (lr=3.0227e-04) (hash(x)=92830605)
+3640 train 4.141769 (lr=2.9860e-04) (hash(x)=78313175)
+3650 train 3.938523 (lr=2.9498e-04) (hash(x)=85395549)
+3660 train 3.739498 (lr=2.9139e-04) (hash(x)=78114459)
+3670 train 3.862650 (lr=2.8786e-04) (hash(x)=74968316)
+3680 train 3.785945 (lr=2.8436e-04) (hash(x)=73358737)
+3690 train 3.897454 (lr=2.8091e-04) (hash(x)=76399442)
+3700 val loss 4.0121
+3700 val perplexity 55.2625
+3700 train 3.899099 (lr=2.7751e-04) (hash(x)=74039273)
+3710 train 4.074997 (lr=2.7415e-04) (hash(x)=79299680)
+3720 train 3.988784 (lr=2.7083e-04) (hash(x)=94701498)
+3730 train 4.064685 (lr=2.6756e-04) (hash(x)=75352071)
+3740 train 3.944776 (lr=2.6433e-04) (hash(x)=77909487)
+3750 train 4.048639 (lr=2.6115e-04) (hash(x)=73986730)
+3760 train 4.009073 (lr=2.5801e-04) (hash(x)=79325763)
+3770 train 4.055683 (lr=2.5492e-04) (hash(x)=72457818)
+3780 train 4.194738 (lr=2.5188e-04) (hash(x)=71775590)
+3790 train 4.080719 (lr=2.4888e-04) (hash(x)=82638943)
+3800 val loss 3.9975
+3800 val perplexity 54.4634
+3800 train 4.051365 (lr=2.4593e-04) (hash(x)=79965893)
+3810 train 3.984003 (lr=2.4302e-04) (hash(x)=72598235)
+3820 train 3.971364 (lr=2.4017e-04) (hash(x)=83113889)
+3830 train 4.032384 (lr=2.3736e-04) (hash(x)=74434590)
+3840 train 3.946350 (lr=2.3459e-04) (hash(x)=82860348)
+3850 train 3.877332 (lr=2.3188e-04) (hash(x)=78067565)
+3860 train 3.951835 (lr=2.2921e-04) (hash(x)=82592498)
+3870 train 3.879313 (lr=2.2659e-04) (hash(x)=81820733)
+3880 train 3.898666 (lr=2.2402e-04) (hash(x)=87709040)
+3890 train 3.942389 (lr=2.2150e-04) (hash(x)=70379093)
+3900 val loss 3.9914
+3900 val perplexity 54.1314
+3900 train 4.068113 (lr=2.1902e-04) (hash(x)=76597431)
+3910 train 3.954017 (lr=2.1659e-04) (hash(x)=90490716)
+3920 train 4.053467 (lr=2.1422e-04) (hash(x)=81970659)
+3930 train 4.042645 (lr=2.1189e-04) (hash(x)=81496334)
+3940 train 4.017247 (lr=2.0961e-04) (hash(x)=75717605)
+3950 train 3.962715 (lr=2.0738e-04) (hash(x)=82749357)
+3960 train 3.993573 (lr=2.0520e-04) (hash(x)=89355157)
+3970 train 3.950773 (lr=2.0307e-04) (hash(x)=78980403)
+3980 train 3.922743 (lr=2.0099e-04) (hash(x)=76627217)
+3990 train 3.922173 (lr=1.9896e-04) (hash(x)=72412879)
+4000 val loss 3.9867
+4000 val perplexity 53.8785
+4000 train 3.931885 (lr=1.9698e-04) (hash(x)=83018142)
+4010 train 3.952560 (lr=1.9505e-04) (hash(x)=81272436)
+4020 train 3.878798 (lr=1.9317e-04) (hash(x)=85497482)
+4030 train 3.956520 (lr=1.9134e-04) (hash(x)=85241734)
+4040 train 3.920720 (lr=1.8956e-04) (hash(x)=77925307)
+4050 train 3.915101 (lr=1.8783e-04) (hash(x)=84826179)
+4060 train 3.975927 (lr=1.8616e-04) (hash(x)=83606764)
+4070 train 3.894283 (lr=1.8453e-04) (hash(x)=80567590)
+4080 train 3.937540 (lr=1.8296e-04) (hash(x)=76860998)
+4090 train 3.976709 (lr=1.8144e-04) (hash(x)=74902328)
+4100 val loss 3.9764
+4100 val perplexity 53.3233
+4100 train 4.019157 (lr=1.7997e-04) (hash(x)=82832041)
+4110 train 3.904457 (lr=1.7855e-04) (hash(x)=79143262)
+4120 train 3.923243 (lr=1.7718e-04) (hash(x)=77038149)
+4130 train 4.110155 (lr=1.7586e-04) (hash(x)=86339074)
+4140 train 4.176855 (lr=1.7460e-04) (hash(x)=76686216)
+4150 train 3.951987 (lr=1.7339e-04) (hash(x)=70522682)
+4160 train 3.914981 (lr=1.7223e-04) (hash(x)=90958555)
+4170 train 3.939883 (lr=1.7112e-04) (hash(x)=91463532)
+4180 train 3.928339 (lr=1.7006e-04) (hash(x)=81959329)
+4190 train 3.889983 (lr=1.6906e-04) (hash(x)=83146752)
+4200 val loss 3.9782
+4200 val perplexity 53.4199
+4200 train 3.868438 (lr=1.6811e-04) (hash(x)=78361715)
+4210 train 4.020553 (lr=1.6721e-04) (hash(x)=87364889)
+4220 train 3.872207 (lr=1.6636e-04) (hash(x)=70465156)
+4230 train 4.025944 (lr=1.6557e-04) (hash(x)=84524081)
+4240 train 3.982625 (lr=1.6483e-04) (hash(x)=77824868)
+4250 train 3.992648 (lr=1.6414e-04) (hash(x)=81710711)
+4260 train 3.971971 (lr=1.6350e-04) (hash(x)=76362728)
+4270 train 3.964435 (lr=1.6292e-04) (hash(x)=83115208)
+4280 train 3.878161 (lr=1.6239e-04) (hash(x)=87218314)
+4290 train 4.036193 (lr=1.6192e-04) (hash(x)=74582673)
+4300 val loss 3.9740
+4300 val perplexity 53.1946
+4300 train 3.913922 (lr=1.6149e-04) (hash(x)=77379615)
+4310 train 3.888834 (lr=1.6112e-04) (hash(x)=78669579)
+4320 train 4.008819 (lr=1.6080e-04) (hash(x)=83066608)
+4330 train 3.970701 (lr=1.6054e-04) (hash(x)=83037340)
+4340 train 4.010625 (lr=1.6032e-04) (hash(x)=82849771)
+4350 train 3.946967 (lr=1.6017e-04) (hash(x)=76693985)
+4360 train 4.031130 (lr=1.6006e-04) (hash(x)=77745394)
+4370 train 3.991888 (lr=1.6001e-04) (hash(x)=79954388)
+4374 val loss 3.9683
+4374 val perplexity 52.8972

12_head_baseline_lr_16e-4_head_dim_22/model_04374.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:62a4a342aa370ebdfa889e2ebb3b670295b57068c5c660a8ccdd10e3a9ab8609
+size 96858242

12_head_baseline_lr_16e-4_head_dim_22/optimizer_04374.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:8be828784248db2bb874d69a732aefbe7734a963dc4dbe5adad9693e539bb81f
+size 187435910