Upload folder using huggingface_hub

scaling_mha_granularity_2/hd_32_seed_1339/args.json

@@ -0,0 +1 @@
+{"hellaswag": true, "attention_kind": "self", "log_dir": "scaling_mha_granularity_2/hd_32_seed_1339", "resume_checkpoint": null, "resume_optimizer": false, "add_a_head": false, "add_head_to_start": true, "new_head_init": "normal", "n_heads": 8, "n_embd": 256, "head_dim": 32, "protect_bos_token": true, "prevent_from_masking_myself": true, "max_steps": 4375, "warmup_steps": 250, "group": "scaling_mha_granularity_2", "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, "disable_selection_head_linear_combo_bias": false, "assert_latent_matches_no_head": false, "n_latent_masks": null, "init_latent_masks_to_identity": false, "init_latent_masks_to_inverse": false, "latent_mask_scale": null, "latent_mask_runtime_multiplier": null, "latent_mask_sigmoid": false, "S_layernorm": false, "one_head_per_latent_mask": false, "att_conv": false, "att_conv_init": "none", "att_conv_scale": 1.0, "att_conv_weight_decay": false, "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": 128, "total_batch_size": 131072, "protection_head_scaling_factor": 1.0, "protection_head_bias": 0.0, "n_sliced_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, "mup_zero_init": true, "l1_loss": false, "debugpy": false, "key": "hd_32", "latent_mask_precision": "bfloat16", "att_conv_precision": "bfloat16", "attn_precision": "bfloat16", "profile_kind": "none", "dense_attention_kind": "mha", "head_dim_value": 32, "sdpa_iter_size": null, "stabilize_attn_scores": true, "override_use_sdpa": false, "autocast_precision": "bfloat16", "simulate_micro_bs": null, "simulate_micro_bs_2": null, "ckpt_attn": true, "c_proj_scale_init": 1.0}

scaling_mha_granularity_2/hd_32_seed_1339/dataloader_04374.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:1e9422c936c4d62660261ba5451b4b72bd5055f233420bd7b0797ede3ccedd1e
+size 964

scaling_mha_granularity_2/hd_32_seed_1339/log2.txt

@@ -0,0 +1,529 @@
+max_steps: 4375
+0 val loss 11.3317
+0 val perplexity 83421.6484
+0 train 11.334251 (lr=4.1958e-06) (hash(x)=161979327)
+10 train 9.911283 (lr=4.6154e-05) (hash(x)=165941070)
+20 train 9.239605 (lr=8.8112e-05) (hash(x)=152918033)
+30 train 8.527445 (lr=1.3007e-04) (hash(x)=169931607)
+40 train 7.758213 (lr=1.7203e-04) (hash(x)=174342675)
+50 train 7.439511 (lr=2.1399e-04) (hash(x)=163872554)
+60 train 7.420366 (lr=2.5594e-04) (hash(x)=161534908)
+70 train 7.026198 (lr=2.9790e-04) (hash(x)=156776712)
+80 train 7.026145 (lr=3.3986e-04) (hash(x)=164996929)
+90 train 6.863284 (lr=3.8182e-04) (hash(x)=152962169)
+100 val loss 6.7848
+100 val perplexity 884.2640
+100 train 6.640815 (lr=4.2378e-04) (hash(x)=147111701)
+110 train 6.651656 (lr=4.6573e-04) (hash(x)=167595270)
+120 train 6.521910 (lr=5.0769e-04) (hash(x)=152205426)
+130 train 6.408152 (lr=5.4965e-04) (hash(x)=158427335)
+140 train 6.412183 (lr=5.9161e-04) (hash(x)=174588384)
+150 train 6.520604 (lr=6.3357e-04) (hash(x)=157067270)
+160 train 6.472780 (lr=6.7552e-04) (hash(x)=161859595)
+170 train 6.393759 (lr=7.1748e-04) (hash(x)=162739537)
+180 train 6.341454 (lr=7.5944e-04) (hash(x)=170785304)
+190 train 6.265031 (lr=8.0140e-04) (hash(x)=168658088)
+200 val loss 6.2009
+200 val perplexity 493.2097
+200 train 6.234201 (lr=8.4336e-04) (hash(x)=150749086)
+210 train 6.139574 (lr=8.8531e-04) (hash(x)=148789576)
+220 train 6.097281 (lr=9.2727e-04) (hash(x)=153164408)
+230 train 6.222859 (lr=9.6923e-04) (hash(x)=184594366)
+240 train 5.943006 (lr=1.0112e-03) (hash(x)=157817319)
+250 train 6.018112 (lr=1.0531e-03) (hash(x)=149451562)
+260 train 5.894110 (lr=1.0951e-03) (hash(x)=165738161)
+270 train 5.861508 (lr=1.1371e-03) (hash(x)=165724764)
+280 train 5.828928 (lr=1.1790e-03) (hash(x)=155414136)
+290 train 5.741538 (lr=1.2210e-03) (hash(x)=165090903)
+300 val loss 5.8488
+300 val perplexity 346.8324
+300 train 5.616086 (lr=1.2629e-03) (hash(x)=154445545)
+310 train 5.575133 (lr=1.3049e-03) (hash(x)=158949603)
+320 train 5.645050 (lr=1.3469e-03) (hash(x)=161412973)
+330 train 5.738562 (lr=1.3888e-03) (hash(x)=164572249)
+340 train 5.760308 (lr=1.4308e-03) (hash(x)=162202191)
+350 train 5.744545 (lr=1.4727e-03) (hash(x)=158135392)
+360 train 5.683514 (lr=1.5147e-03) (hash(x)=163752391)
+370 train 5.712468 (lr=1.5566e-03) (hash(x)=172758799)
+380 train 5.510674 (lr=1.5986e-03) (hash(x)=144569488)
+390 train 5.481882 (lr=1.6406e-03) (hash(x)=150429229)
+400 val loss 5.4744
+400 val perplexity 238.5159
+400 train 5.459674 (lr=1.6825e-03) (hash(x)=154261530)
+410 train 5.407113 (lr=1.7245e-03) (hash(x)=159172506)
+420 train 5.423469 (lr=1.7664e-03) (hash(x)=154419867)
+430 train 5.447749 (lr=1.8084e-03) (hash(x)=167337099)
+440 train 5.354578 (lr=1.8503e-03) (hash(x)=154535255)
+450 train 5.332942 (lr=1.8923e-03) (hash(x)=153077071)
+460 train 5.235121 (lr=1.9343e-03) (hash(x)=150853070)
+470 train 5.069131 (lr=1.9762e-03) (hash(x)=160701113)
+480 train 5.201829 (lr=2.0182e-03) (hash(x)=154690870)
+490 train 5.008202 (lr=2.0601e-03) (hash(x)=163834286)
+500 val loss 5.2954
+500 val perplexity 199.4213
+500 train 5.083394 (lr=2.1021e-03) (hash(x)=162651717)
+510 train 5.192930 (lr=2.1441e-03) (hash(x)=160448311)
+520 train 5.239585 (lr=2.1860e-03) (hash(x)=165741643)
+530 train 5.376290 (lr=2.2280e-03) (hash(x)=153201264)
+540 train 5.270685 (lr=2.2699e-03) (hash(x)=159027160)
+550 train 5.230485 (lr=2.3119e-03) (hash(x)=170769095)
+560 train 5.262185 (lr=2.3538e-03) (hash(x)=153606190)
+570 train 5.235197 (lr=2.3958e-03) (hash(x)=159708689)
+580 train 5.157146 (lr=2.4378e-03) (hash(x)=179108359)
+590 train 5.207577 (lr=2.4797e-03) (hash(x)=157134241)
+600 val loss 5.1596
+600 val perplexity 174.0882
+600 train 5.178593 (lr=2.5217e-03) (hash(x)=162178356)
+610 train 5.143222 (lr=2.5636e-03) (hash(x)=167646928)
+620 train 5.077677 (lr=2.6056e-03) (hash(x)=160979601)
+630 train 4.972814 (lr=2.6476e-03) (hash(x)=158349468)
+640 train 4.960670 (lr=2.6895e-03) (hash(x)=150370348)
+650 train 4.912189 (lr=2.7315e-03) (hash(x)=152810075)
+660 train 4.880799 (lr=2.7734e-03) (hash(x)=145786262)
+670 train 4.853633 (lr=2.8154e-03) (hash(x)=160779604)
+680 train 4.761813 (lr=2.8573e-03) (hash(x)=152343688)
+690 train 4.776192 (lr=2.8993e-03) (hash(x)=166507600)
+700 val loss 5.0187
+700 val perplexity 151.2199
+700 train 4.757583 (lr=2.9413e-03) (hash(x)=162623700)
+710 train 5.044603 (lr=2.9832e-03) (hash(x)=162293403)
+720 train 5.007554 (lr=3.0000e-03) (hash(x)=159564657)
+730 train 5.015098 (lr=2.9999e-03) (hash(x)=154209107)
+740 train 4.976248 (lr=2.9997e-03) (hash(x)=156485709)
+750 train 4.928605 (lr=2.9994e-03) (hash(x)=160834702)
+760 train 4.941286 (lr=2.9990e-03) (hash(x)=152240275)
+770 train 4.915572 (lr=2.9985e-03) (hash(x)=168903075)
+780 train 5.015688 (lr=2.9979e-03) (hash(x)=166843738)
+790 train 5.027126 (lr=2.9972e-03) (hash(x)=166358104)
+800 val loss 4.8585
+800 val perplexity 128.8319
+800 train 4.823143 (lr=2.9964e-03) (hash(x)=153477747)
+810 train 4.787605 (lr=2.9955e-03) (hash(x)=157954385)
+820 train 4.776272 (lr=2.9945e-03) (hash(x)=174383505)
+830 train 4.774566 (lr=2.9934e-03) (hash(x)=145226181)
+840 train 4.775194 (lr=2.9922e-03) (hash(x)=157665743)
+850 train 4.771248 (lr=2.9909e-03) (hash(x)=158565836)
+860 train 4.773874 (lr=2.9896e-03) (hash(x)=171168105)
+870 train 5.177950 (lr=2.9881e-03) (hash(x)=159017777)
+880 train 4.921308 (lr=2.9865e-03) (hash(x)=168156768)
+890 train 4.829815 (lr=2.9848e-03) (hash(x)=145645245)
+900 val loss 4.7171
+900 val perplexity 111.8424
+900 train 4.666941 (lr=2.9830e-03) (hash(x)=151245521)
+910 train 4.684638 (lr=2.9811e-03) (hash(x)=182614672)
+920 train 4.661321 (lr=2.9792e-03) (hash(x)=167377432)
+930 train 4.598299 (lr=2.9771e-03) (hash(x)=171347228)
+940 train 4.617245 (lr=2.9749e-03) (hash(x)=171588755)
+950 train 4.718965 (lr=2.9726e-03) (hash(x)=159560168)
+960 train 4.613429 (lr=2.9703e-03) (hash(x)=171801357)
+970 train 4.613768 (lr=2.9678e-03) (hash(x)=172513884)
+980 train 4.630521 (lr=2.9652e-03) (hash(x)=170659685)
+990 train 4.666844 (lr=2.9626e-03) (hash(x)=165436358)
+1000 val loss 4.6273
+1000 val perplexity 102.2331
+1000 train 4.728164 (lr=2.9598e-03) (hash(x)=167684690)
+1010 train 4.689438 (lr=2.9570e-03) (hash(x)=127225492)
+1020 train 4.682221 (lr=2.9540e-03) (hash(x)=167575323)
+1030 train 4.611586 (lr=2.9510e-03) (hash(x)=161266302)
+1040 train 4.586234 (lr=2.9478e-03) (hash(x)=155607661)
+1050 train 4.630967 (lr=2.9446e-03) (hash(x)=165685359)
+1060 train 4.493969 (lr=2.9412e-03) (hash(x)=160417824)
+1070 train 4.620078 (lr=2.9378e-03) (hash(x)=173313882)
+1080 train 4.601588 (lr=2.9343e-03) (hash(x)=165485047)
+1090 train 4.456005 (lr=2.9307e-03) (hash(x)=159848305)
+1100 val loss 4.5623
+1100 val perplexity 95.8072
+1100 train 4.620852 (lr=2.9270e-03) (hash(x)=170890601)
+1110 train 4.550446 (lr=2.9231e-03) (hash(x)=172689983)
+1120 train 4.585349 (lr=2.9192e-03) (hash(x)=154762859)
+1130 train 4.504292 (lr=2.9152e-03) (hash(x)=168196871)
+1140 train 4.535262 (lr=2.9112e-03) (hash(x)=156715577)
+1150 train 4.662782 (lr=2.9070e-03) (hash(x)=166381788)
+1160 train 4.582893 (lr=2.9027e-03) (hash(x)=162095130)
+1170 train 4.562312 (lr=2.8983e-03) (hash(x)=166094750)
+1180 train 4.508305 (lr=2.8939e-03) (hash(x)=154965753)
+1190 train 4.481928 (lr=2.8893e-03) (hash(x)=172766447)
+1200 val loss 4.4727
+1200 val perplexity 87.5933
+1200 train 4.397031 (lr=2.8847e-03) (hash(x)=161185334)
+1210 train 4.563327 (lr=2.8800e-03) (hash(x)=167224144)
+1220 train 4.415941 (lr=2.8751e-03) (hash(x)=167252866)
+1230 train 4.429101 (lr=2.8702e-03) (hash(x)=155856924)
+1240 train 4.422760 (lr=2.8652e-03) (hash(x)=155956446)
+1250 train 4.332813 (lr=2.8601e-03) (hash(x)=140794796)
+1260 train 4.442540 (lr=2.8550e-03) (hash(x)=166863869)
+1270 train 4.506233 (lr=2.8497e-03) (hash(x)=166021892)
+1280 train 4.513660 (lr=2.8443e-03) (hash(x)=170503648)
+1290 train 4.427346 (lr=2.8389e-03) (hash(x)=164012719)
+1300 val loss 4.4330
+1300 val perplexity 84.1846
+1300 train 4.487343 (lr=2.8333e-03) (hash(x)=156519834)
+1310 train 4.446633 (lr=2.8277e-03) (hash(x)=165146652)
+1320 train 4.464661 (lr=2.8220e-03) (hash(x)=174391990)
+1330 train 4.459830 (lr=2.8162e-03) (hash(x)=161131201)
+1340 train 4.453235 (lr=2.8103e-03) (hash(x)=162240912)
+1350 train 4.363201 (lr=2.8044e-03) (hash(x)=152462931)
+1360 train 4.374543 (lr=2.7983e-03) (hash(x)=150443278)
+1370 train 4.309171 (lr=2.7922e-03) (hash(x)=153721711)
+1380 train 4.444357 (lr=2.7860e-03) (hash(x)=158070628)
+1390 train 4.474531 (lr=2.7797e-03) (hash(x)=163318827)
+1400 val loss 4.4182
+1400 val perplexity 82.9463
+1400 train 4.387547 (lr=2.7733e-03) (hash(x)=154210336)
+1410 train 4.446886 (lr=2.7668e-03) (hash(x)=162393933)
+1420 train 4.432415 (lr=2.7603e-03) (hash(x)=156659308)
+1430 train 4.393914 (lr=2.7536e-03) (hash(x)=171243439)
+1440 train 4.536616 (lr=2.7469e-03) (hash(x)=178459521)
+1450 train 4.398684 (lr=2.7401e-03) (hash(x)=159166350)
+1460 train 4.308608 (lr=2.7333e-03) (hash(x)=173554569)
+1470 train 4.409107 (lr=2.7263e-03) (hash(x)=158362521)
+1480 train 4.373163 (lr=2.7193e-03) (hash(x)=165899207)
+1490 train 4.401465 (lr=2.7121e-03) (hash(x)=156470939)
+1500 val loss 4.3651
+1500 val perplexity 78.6598
+1500 train 4.371499 (lr=2.7050e-03) (hash(x)=164197780)
+1510 train 4.267107 (lr=2.6977e-03) (hash(x)=146709896)
+1520 train 4.289246 (lr=2.6903e-03) (hash(x)=151304838)
+1530 train 4.320549 (lr=2.6829e-03) (hash(x)=163187333)
+1540 train 4.279408 (lr=2.6754e-03) (hash(x)=153630537)
+1550 train 4.293072 (lr=2.6678e-03) (hash(x)=157299103)
+1560 train 4.204647 (lr=2.6602e-03) (hash(x)=153786477)
+1570 train 4.522542 (lr=2.6525e-03) (hash(x)=155037209)
+1580 train 4.398278 (lr=2.6447e-03) (hash(x)=146157308)
+1590 train 4.447587 (lr=2.6368e-03) (hash(x)=147823317)
+1600 val loss 4.3453
+1600 val perplexity 77.1179
+1600 train 4.439014 (lr=2.6289e-03) (hash(x)=169140159)
+1610 train 4.212669 (lr=2.6208e-03) (hash(x)=157811493)
+1620 train 4.302999 (lr=2.6128e-03) (hash(x)=168940851)
+1630 train 4.301583 (lr=2.6046e-03) (hash(x)=151149213)
+1640 train 4.396276 (lr=2.5964e-03) (hash(x)=161822009)
+1650 train 4.810256 (lr=2.5881e-03) (hash(x)=151056313)
+1660 train 4.314696 (lr=2.5797e-03) (hash(x)=156920306)
+1670 train 4.230519 (lr=2.5713e-03) (hash(x)=159695815)
+1680 train 4.245347 (lr=2.5628e-03) (hash(x)=160232907)
+1690 train 4.237248 (lr=2.5542e-03) (hash(x)=158099357)
+1700 val loss 4.3446
+1700 val perplexity 77.0613
+1700 train 4.296473 (lr=2.5455e-03) (hash(x)=164127381)
+1710 train 4.469089 (lr=2.5368e-03) (hash(x)=156592040)
+1720 train 4.396180 (lr=2.5281e-03) (hash(x)=160220265)
+1730 train 4.331512 (lr=2.5192e-03) (hash(x)=156162204)
+1740 train 4.316258 (lr=2.5103e-03) (hash(x)=152767893)
+1750 train 4.307747 (lr=2.5014e-03) (hash(x)=162707306)
+1760 train 4.330931 (lr=2.4924e-03) (hash(x)=154472004)
+1770 train 4.180199 (lr=2.4833e-03) (hash(x)=160374234)
+1780 train 4.340650 (lr=2.4741e-03) (hash(x)=159709225)
+1790 train 4.378290 (lr=2.4649e-03) (hash(x)=152095944)
+1800 val loss 4.2956
+1800 val perplexity 73.3738
+1800 train 4.354331 (lr=2.4556e-03) (hash(x)=157920255)
+1810 train 4.096696 (lr=2.4463e-03) (hash(x)=167237680)
+1820 train 4.208254 (lr=2.4369e-03) (hash(x)=161255774)
+1830 train 4.250211 (lr=2.4275e-03) (hash(x)=154275481)
+1840 train 4.174768 (lr=2.4180e-03) (hash(x)=152398178)
+1850 train 4.285386 (lr=2.4084e-03) (hash(x)=147646022)
+1860 train 4.308550 (lr=2.3988e-03) (hash(x)=158895559)
+1870 train 4.248223 (lr=2.3891e-03) (hash(x)=162742598)
+1880 train 4.243441 (lr=2.3794e-03) (hash(x)=169911029)
+1890 train 4.282410 (lr=2.3696e-03) (hash(x)=172068270)
+1900 val loss 4.2930
+1900 val perplexity 73.1871
+1900 train 4.376421 (lr=2.3598e-03) (hash(x)=159975457)
+1910 train 4.360821 (lr=2.3499e-03) (hash(x)=178775667)
+1920 train 4.238623 (lr=2.3400e-03) (hash(x)=151180831)
+1930 train 4.254376 (lr=2.3300e-03) (hash(x)=154239412)
+1940 train 4.260098 (lr=2.3200e-03) (hash(x)=139144465)
+1950 train 4.216904 (lr=2.3099e-03) (hash(x)=194482413)
+1960 train 4.230551 (lr=2.2998e-03) (hash(x)=155812344)
+1970 train 4.232997 (lr=2.2896e-03) (hash(x)=157577031)
+1980 train 4.271754 (lr=2.2793e-03) (hash(x)=152045027)
+1990 train 4.180758 (lr=2.2691e-03) (hash(x)=141742221)
+2000 val loss 4.2437
+2000 val perplexity 69.6648
+2000 train 4.298559 (lr=2.2588e-03) (hash(x)=160555812)
+2010 train 4.214628 (lr=2.2484e-03) (hash(x)=160371335)
+2020 train 4.227884 (lr=2.2380e-03) (hash(x)=167890767)
+2030 train 4.301433 (lr=2.2275e-03) (hash(x)=165127715)
+2040 train 4.249403 (lr=2.2170e-03) (hash(x)=151587302)
+2050 train 4.262214 (lr=2.2065e-03) (hash(x)=172347989)
+2060 train 4.339118 (lr=2.1959e-03) (hash(x)=161301919)
+2070 train 4.270619 (lr=2.1853e-03) (hash(x)=161472934)
+2080 train 4.359751 (lr=2.1746e-03) (hash(x)=159777322)
+2090 train 4.271879 (lr=2.1639e-03) (hash(x)=151149770)
+2100 val loss 4.2223
+2100 val perplexity 68.1919
+2100 train 4.099855 (lr=2.1532e-03) (hash(x)=160283520)
+2110 train 4.271060 (lr=2.1424e-03) (hash(x)=154222025)
+2120 train 4.213970 (lr=2.1316e-03) (hash(x)=155638178)
+2130 train 4.129359 (lr=2.1208e-03) (hash(x)=149125384)
+2140 train 4.186398 (lr=2.1099e-03) (hash(x)=159496074)
+2150 train 4.269260 (lr=2.0990e-03) (hash(x)=162019154)
+2160 train 4.428658 (lr=2.0881e-03) (hash(x)=155466928)
+2170 train 4.298334 (lr=2.0771e-03) (hash(x)=173469994)
+2180 train 4.402198 (lr=2.0661e-03) (hash(x)=163448753)
+2190 train 4.242334 (lr=2.0550e-03) (hash(x)=166595123)
+2200 val loss 4.2038
+2200 val perplexity 66.9428
+2200 train 4.123098 (lr=2.0440e-03) (hash(x)=165958875)
+2210 train 4.265956 (lr=2.0329e-03) (hash(x)=165019039)
+2220 train 4.246004 (lr=2.0217e-03) (hash(x)=167325225)
+2230 train 4.074448 (lr=2.0106e-03) (hash(x)=154287439)
+2240 train 4.147588 (lr=1.9994e-03) (hash(x)=158565817)
+2250 train 4.104751 (lr=1.9882e-03) (hash(x)=160592835)
+2260 train 4.150549 (lr=1.9770e-03) (hash(x)=144591648)
+2270 train 4.024544 (lr=1.9657e-03) (hash(x)=150486876)
+2280 train 4.222304 (lr=1.9544e-03) (hash(x)=159194788)
+2290 train 4.295236 (lr=1.9431e-03) (hash(x)=164315345)
+2300 val loss 4.1822
+2300 val perplexity 65.5116
+2300 train 4.226495 (lr=1.9318e-03) (hash(x)=170470665)
+2310 train 4.184696 (lr=1.9205e-03) (hash(x)=176539068)
+2320 train 4.171284 (lr=1.9091e-03) (hash(x)=161008474)
+2330 train 4.245394 (lr=1.8977e-03) (hash(x)=168474380)
+2340 train 4.240894 (lr=1.8863e-03) (hash(x)=156072131)
+2350 train 4.320161 (lr=1.8749e-03) (hash(x)=165127087)
+2360 train 4.323942 (lr=1.8635e-03) (hash(x)=159462801)
+2370 train 4.160083 (lr=1.8520e-03) (hash(x)=161698394)
+2380 train 4.084512 (lr=1.8406e-03) (hash(x)=165588078)
+2390 train 4.089845 (lr=1.8291e-03) (hash(x)=171673672)
+2400 val loss 4.1794
+2400 val perplexity 65.3292
+2400 train 4.157574 (lr=1.8176e-03) (hash(x)=157381516)
+2410 train 4.160076 (lr=1.8061e-03) (hash(x)=163238532)
+2420 train 4.122440 (lr=1.7946e-03) (hash(x)=156178237)
+2430 train 4.001816 (lr=1.7830e-03) (hash(x)=155080617)
+2440 train 4.101940 (lr=1.7715e-03) (hash(x)=167490538)
+2450 train 4.075342 (lr=1.7600e-03) (hash(x)=163757352)
+2460 train 4.034024 (lr=1.7484e-03) (hash(x)=156734811)
+2470 train 4.034389 (lr=1.7368e-03) (hash(x)=161522341)
+2480 train 4.290652 (lr=1.7253e-03) (hash(x)=174032060)
+2490 train 4.246651 (lr=1.7137e-03) (hash(x)=163065818)
+2500 val loss 4.1510
+2500 val perplexity 63.4991
+2500 train 4.302966 (lr=1.7021e-03) (hash(x)=126959780)
+2510 train 4.242518 (lr=1.6906e-03) (hash(x)=153844939)
+2520 train 4.214756 (lr=1.6790e-03) (hash(x)=174018174)
+2530 train 4.193752 (lr=1.6674e-03) (hash(x)=150568427)
+2540 train 4.185742 (lr=1.6558e-03) (hash(x)=161725206)
+2550 train 4.156045 (lr=1.6442e-03) (hash(x)=181346417)
+2560 train 4.144405 (lr=1.6326e-03) (hash(x)=168730011)
+2570 train 4.043176 (lr=1.6210e-03) (hash(x)=155900214)
+2580 train 4.090947 (lr=1.6094e-03) (hash(x)=160014621)
+2590 train 4.066454 (lr=1.5979e-03) (hash(x)=148996689)
+2600 val loss 4.1328
+2600 val perplexity 62.3504
+2600 train 4.140089 (lr=1.5863e-03) (hash(x)=145183968)
+2610 train 3.884699 (lr=1.5747e-03) (hash(x)=159420240)
+2620 train 3.861299 (lr=1.5632e-03) (hash(x)=151169903)
+2630 train 3.876419 (lr=1.5516e-03) (hash(x)=149454811)
+2640 train 3.945082 (lr=1.5400e-03) (hash(x)=155603914)
+2650 train 3.935201 (lr=1.5285e-03) (hash(x)=169545195)
+2660 train 4.290094 (lr=1.5170e-03) (hash(x)=168454079)
+2670 train 4.189615 (lr=1.5054e-03) (hash(x)=163110417)
+2680 train 4.139259 (lr=1.4939e-03) (hash(x)=166075603)
+2690 train 4.165034 (lr=1.4824e-03) (hash(x)=174132258)
+2700 val loss 4.1262
+2700 val perplexity 61.9404
+2700 train 4.173933 (lr=1.4709e-03) (hash(x)=169846940)
+2710 train 4.071883 (lr=1.4594e-03) (hash(x)=151553731)
+2720 train 4.172825 (lr=1.4480e-03) (hash(x)=164024255)
+2730 train 4.192457 (lr=1.4365e-03) (hash(x)=156458097)
+2740 train 4.119302 (lr=1.4251e-03) (hash(x)=154663748)
+2750 train 3.981194 (lr=1.4137e-03) (hash(x)=157548181)
+2760 train 4.095603 (lr=1.4023e-03) (hash(x)=151584783)
+2770 train 4.118092 (lr=1.3909e-03) (hash(x)=152692594)
+2780 train 4.091716 (lr=1.3795e-03) (hash(x)=165952919)
+2790 train 3.984522 (lr=1.3682e-03) (hash(x)=157484183)
+2800 val loss 4.1107
+2800 val perplexity 60.9883
+2800 train 4.059634 (lr=1.3569e-03) (hash(x)=154790034)
+2810 train 3.924847 (lr=1.3456e-03) (hash(x)=157856097)
+2820 train 3.899269 (lr=1.3343e-03) (hash(x)=168903939)
+2830 train 4.023672 (lr=1.3230e-03) (hash(x)=167001475)
+2840 train 3.851925 (lr=1.3118e-03) (hash(x)=158987150)
+2850 train 4.092372 (lr=1.3006e-03) (hash(x)=168695407)
+2860 train 4.219913 (lr=1.2894e-03) (hash(x)=164999034)
+2870 train 4.075723 (lr=1.2783e-03) (hash(x)=191197316)
+2880 train 4.025117 (lr=1.2671e-03) (hash(x)=159625200)
+2890 train 4.187472 (lr=1.2560e-03) (hash(x)=157756126)
+2900 val loss 4.0823
+2900 val perplexity 59.2788
+2900 train 4.037582 (lr=1.2450e-03) (hash(x)=164721377)
+2910 train 4.092578 (lr=1.2339e-03) (hash(x)=155873181)
+2920 train 4.039368 (lr=1.2229e-03) (hash(x)=156477082)
+2930 train 4.044103 (lr=1.2119e-03) (hash(x)=151261633)
+2940 train 3.984043 (lr=1.2010e-03) (hash(x)=160090385)
+2950 train 4.126176 (lr=1.1901e-03) (hash(x)=157300618)
+2960 train 4.041431 (lr=1.1792e-03) (hash(x)=163095555)
+2970 train 4.072598 (lr=1.1684e-03) (hash(x)=149649212)
+2980 train 3.881353 (lr=1.1576e-03) (hash(x)=174517675)
+2990 train 3.853763 (lr=1.1468e-03) (hash(x)=161344491)
+3000 val loss 4.0832
+3000 val perplexity 59.3327
+3000 train 3.876039 (lr=1.1361e-03) (hash(x)=163816905)
+3010 train 3.784514 (lr=1.1254e-03) (hash(x)=157420680)
+3020 train 3.897798 (lr=1.1147e-03) (hash(x)=153843035)
+3030 train 4.091018 (lr=1.1041e-03) (hash(x)=149770373)
+3040 train 4.157246 (lr=1.0935e-03) (hash(x)=165320741)
+3050 train 4.127814 (lr=1.0830e-03) (hash(x)=170181151)
+3060 train 4.049113 (lr=1.0725e-03) (hash(x)=163837230)
+3070 train 4.044464 (lr=1.0620e-03) (hash(x)=156812797)
+3080 train 4.019273 (lr=1.0516e-03) (hash(x)=143663627)
+3090 train 4.045516 (lr=1.0412e-03) (hash(x)=165052478)
+3100 val loss 4.0607
+3100 val perplexity 58.0164
+3100 train 3.976615 (lr=1.0309e-03) (hash(x)=151458705)
+3110 train 4.020579 (lr=1.0207e-03) (hash(x)=158228317)
+3120 train 4.193603 (lr=1.0104e-03) (hash(x)=149856801)
+3130 train 3.864872 (lr=1.0002e-03) (hash(x)=146297847)
+3140 train 3.971745 (lr=9.9011e-04) (hash(x)=172795167)
+3150 train 4.027694 (lr=9.8002e-04) (hash(x)=145936074)
+3160 train 4.109884 (lr=9.6999e-04) (hash(x)=157564303)
+3170 train 4.212304 (lr=9.6000e-04) (hash(x)=157451362)
+3180 train 4.125219 (lr=9.5007e-04) (hash(x)=156366043)
+3190 train 4.084595 (lr=9.4019e-04) (hash(x)=162344697)
+3200 val loss 4.0408
+3200 val perplexity 56.8745
+3200 train 4.105824 (lr=9.3036e-04) (hash(x)=164206298)
+3210 train 4.177977 (lr=9.2058e-04) (hash(x)=149468251)
+3220 train 4.113330 (lr=9.1085e-04) (hash(x)=157808606)
+3230 train 4.244371 (lr=9.0118e-04) (hash(x)=157078690)
+3240 train 4.054189 (lr=8.9157e-04) (hash(x)=165699696)
+3250 train 4.001168 (lr=8.8201e-04) (hash(x)=181352842)
+3260 train 4.096155 (lr=8.7251e-04) (hash(x)=142534383)
+3270 train 4.058337 (lr=8.6307e-04) (hash(x)=165297359)
+3280 train 4.100406 (lr=8.5368e-04) (hash(x)=160326815)
+3290 train 4.027347 (lr=8.4435e-04) (hash(x)=150043063)
+3300 val loss 4.0354
+3300 val perplexity 56.5651
+3300 train 3.978810 (lr=8.3508e-04) (hash(x)=164133091)
+3310 train 3.919091 (lr=8.2588e-04) (hash(x)=145803863)
+3320 train 3.944571 (lr=8.1673e-04) (hash(x)=157282009)
+3330 train 3.880311 (lr=8.0764e-04) (hash(x)=159126415)
+3340 train 4.088424 (lr=7.9862e-04) (hash(x)=164426675)
+3350 train 4.118574 (lr=7.8966e-04) (hash(x)=170424924)
+3360 train 4.062285 (lr=7.8076e-04) (hash(x)=163161621)
+3370 train 4.123874 (lr=7.7192e-04) (hash(x)=160786731)
+3380 train 4.081377 (lr=7.6315e-04) (hash(x)=165954688)
+3390 train 4.131517 (lr=7.5445e-04) (hash(x)=147371107)
+3400 val loss 4.0139
+3400 val perplexity 55.3597
+3400 train 4.066054 (lr=7.4581e-04) (hash(x)=153067604)
+3410 train 4.069908 (lr=7.3724e-04) (hash(x)=161748196)
+3420 train 4.078162 (lr=7.2874e-04) (hash(x)=168297963)
+3430 train 4.074960 (lr=7.2030e-04) (hash(x)=145086892)
+3440 train 3.975844 (lr=7.1193e-04) (hash(x)=173125711)
+3450 train 4.076224 (lr=7.0363e-04) (hash(x)=158080216)
+3460 train 3.899282 (lr=6.9541e-04) (hash(x)=158055750)
+3470 train 4.000287 (lr=6.8725e-04) (hash(x)=151381400)
+3480 train 3.933253 (lr=6.7916e-04) (hash(x)=147262346)
+3490 train 3.910489 (lr=6.7114e-04) (hash(x)=154740086)
+3500 val loss 4.0120
+3500 val perplexity 55.2581
+3500 train 3.838781 (lr=6.6320e-04) (hash(x)=145730968)
+3510 train 3.981884 (lr=6.5533e-04) (hash(x)=151355929)
+3520 train 3.972538 (lr=6.4753e-04) (hash(x)=165136835)
+3530 train 4.049836 (lr=6.3981e-04) (hash(x)=163869056)
+3540 train 3.990377 (lr=6.3216e-04) (hash(x)=161580134)
+3550 train 4.003833 (lr=6.2458e-04) (hash(x)=168595690)
+3560 train 4.040783 (lr=6.1708e-04) (hash(x)=158576466)
+3570 train 4.005232 (lr=6.0966e-04) (hash(x)=157298619)
+3580 train 4.086639 (lr=6.0231e-04) (hash(x)=160892091)
+3590 train 3.958079 (lr=5.9504e-04) (hash(x)=165168008)
+3600 val loss 3.9926
+3600 val perplexity 54.1930
+3600 train 4.059351 (lr=5.8785e-04) (hash(x)=144911638)
+3610 train 4.012804 (lr=5.8074e-04) (hash(x)=150614513)
+3620 train 4.005171 (lr=5.7370e-04) (hash(x)=167695016)
+3630 train 3.921500 (lr=5.6675e-04) (hash(x)=182708233)
+3640 train 4.030973 (lr=5.5987e-04) (hash(x)=154146274)
+3650 train 3.999436 (lr=5.5308e-04) (hash(x)=159129968)
+3660 train 3.908366 (lr=5.4636e-04) (hash(x)=146209497)
+3670 train 3.925957 (lr=5.3973e-04) (hash(x)=145578104)
+3680 train 3.919229 (lr=5.3318e-04) (hash(x)=150243521)
+3690 train 3.875518 (lr=5.2671e-04) (hash(x)=154328313)
+3700 val loss 3.9949
+3700 val perplexity 54.3191
+3700 train 3.846409 (lr=5.2033e-04) (hash(x)=159904582)
+3710 train 4.187193 (lr=5.1402e-04) (hash(x)=175986417)
+3720 train 4.103745 (lr=5.0780e-04) (hash(x)=158509479)
+3730 train 4.122223 (lr=5.0167e-04) (hash(x)=156087760)
+3740 train 4.116159 (lr=4.9562e-04) (hash(x)=177667170)
+3750 train 4.094230 (lr=4.8965e-04) (hash(x)=145256842)
+3760 train 4.051919 (lr=4.8377e-04) (hash(x)=149855635)
+3770 train 4.091427 (lr=4.7798e-04) (hash(x)=168798886)
+3780 train 4.004841 (lr=4.7227e-04) (hash(x)=156277179)
+3790 train 3.970837 (lr=4.6665e-04) (hash(x)=149698945)
+3800 val loss 3.9773
+3800 val perplexity 53.3706
+3800 train 3.929260 (lr=4.6112e-04) (hash(x)=168039654)
+3810 train 4.043870 (lr=4.5567e-04) (hash(x)=154595574)
+3820 train 4.099159 (lr=4.5031e-04) (hash(x)=158303284)
+3830 train 4.108831 (lr=4.4504e-04) (hash(x)=164128644)
+3840 train 4.059931 (lr=4.3986e-04) (hash(x)=166842997)
+3850 train 3.980275 (lr=4.3477e-04) (hash(x)=177681163)
+3860 train 3.951094 (lr=4.2977e-04) (hash(x)=177062575)
+3870 train 3.993991 (lr=4.2486e-04) (hash(x)=158957159)
+3880 train 3.923505 (lr=4.2004e-04) (hash(x)=151567740)
+3890 train 4.006064 (lr=4.1530e-04) (hash(x)=153261675)
+3900 val loss 3.9676
+3900 val perplexity 52.8556
+3900 train 4.277389 (lr=4.1066e-04) (hash(x)=160268683)
+3910 train 3.998404 (lr=4.0611e-04) (hash(x)=153745240)
+3920 train 4.049323 (lr=4.0166e-04) (hash(x)=164537824)
+3930 train 3.996554 (lr=3.9729e-04) (hash(x)=162062116)
+3940 train 3.957098 (lr=3.9302e-04) (hash(x)=168626742)
+3950 train 4.020246 (lr=3.8884e-04) (hash(x)=148753561)
+3960 train 3.949833 (lr=3.8475e-04) (hash(x)=163526242)
+3970 train 4.158437 (lr=3.8076e-04) (hash(x)=159329323)
+3980 train 4.086639 (lr=3.7685e-04) (hash(x)=146502339)
+3990 train 3.783062 (lr=3.7305e-04) (hash(x)=158409708)
+4000 val loss 3.9619
+4000 val perplexity 52.5558
+4000 train 3.963998 (lr=3.6933e-04) (hash(x)=154645137)
+4010 train 3.961229 (lr=3.6572e-04) (hash(x)=160044026)
+4020 train 3.943695 (lr=3.6219e-04) (hash(x)=155544686)
+4030 train 3.930335 (lr=3.5876e-04) (hash(x)=147589421)
+4040 train 4.041538 (lr=3.5543e-04) (hash(x)=158607990)
+4050 train 3.954383 (lr=3.5219e-04) (hash(x)=154108872)
+4060 train 4.031149 (lr=3.4905e-04) (hash(x)=167101465)
+4070 train 3.987827 (lr=3.4600e-04) (hash(x)=160323363)
+4080 train 3.980521 (lr=3.4305e-04) (hash(x)=150423692)
+4090 train 3.957799 (lr=3.4019e-04) (hash(x)=167461885)
+4100 val loss 3.9517
+4100 val perplexity 52.0246
+4100 train 3.993399 (lr=3.3744e-04) (hash(x)=151783377)
+4110 train 3.988193 (lr=3.3477e-04) (hash(x)=165679649)
+4120 train 3.883749 (lr=3.3221e-04) (hash(x)=153643039)
+4130 train 3.934893 (lr=3.2974e-04) (hash(x)=149648749)
+4140 train 3.972370 (lr=3.2737e-04) (hash(x)=162751418)
+4150 train 3.969096 (lr=3.2510e-04) (hash(x)=157047478)
+4160 train 3.903399 (lr=3.2292e-04) (hash(x)=156391812)
+4170 train 3.973576 (lr=3.2085e-04) (hash(x)=180926985)
+4180 train 4.017891 (lr=3.1887e-04) (hash(x)=168699059)
+4190 train 3.942169 (lr=3.1699e-04) (hash(x)=155410372)
+4200 val loss 3.9485
+4200 val perplexity 51.8601
+4200 train 3.952024 (lr=3.1520e-04) (hash(x)=152615252)
+4210 train 3.916412 (lr=3.1352e-04) (hash(x)=155679453)
+4220 train 4.030867 (lr=3.1193e-04) (hash(x)=152954399)
+4230 train 4.039033 (lr=3.1044e-04) (hash(x)=164608084)
+4240 train 4.058977 (lr=3.0905e-04) (hash(x)=162463867)
+4250 train 3.971911 (lr=3.0776e-04) (hash(x)=168832088)
+4260 train 3.945614 (lr=3.0657e-04) (hash(x)=162698882)
+4270 train 3.980678 (lr=3.0548e-04) (hash(x)=180861492)
+4280 train 3.865040 (lr=3.0449e-04) (hash(x)=160824300)
+4290 train 3.930821 (lr=3.0359e-04) (hash(x)=157750938)
+4300 val loss 3.9484
+4300 val perplexity 51.8523
+4300 train 3.880197 (lr=3.0280e-04) (hash(x)=157216833)
+4310 train 3.877450 (lr=3.0210e-04) (hash(x)=176184661)
+4320 train 4.045069 (lr=3.0150e-04) (hash(x)=161537117)
+4330 train 3.992988 (lr=3.0101e-04) (hash(x)=160700216)
+4340 train 3.943339 (lr=3.0061e-04) (hash(x)=175519903)
+4350 train 3.947504 (lr=3.0031e-04) (hash(x)=167988701)
+4360 train 3.902112 (lr=3.0011e-04) (hash(x)=162619084)
+4370 train 4.289888 (lr=3.0001e-04) (hash(x)=166292684)
+4374 val loss 3.9409
+4374 val perplexity 51.4637

scaling_mha_granularity_2/hd_32_seed_1339/model_04374.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:5e6c8c4be1c548887f7dde9b9797f41433882024f8c0faf57b2ef22fd144b743
+size 89692138

scaling_mha_granularity_2/hd_32_seed_1339/optimizer_04374.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:829dd26f38602c1c4540e10ddcef0af0ba4d5e80f7f864cb82ad078c40b4c049
+size 179406342