Upload folder using huggingface_hub

scaling_mha_nh_3/mha_nh_32_seed_1339/args.json

@@ -0,0 +1 @@
+{"hellaswag": true, "attention_kind": "dense", "log_dir": "scaling_mha_nh_3/mha_nh_32_seed_1339", "resume_checkpoint": null, "resume_optimizer": false, "add_a_head": false, "add_head_to_start": true, "new_head_init": "normal", "n_heads": 32, "n_embd": 256, "head_dim": 32, "protect_bos_token": true, "prevent_from_masking_myself": true, "max_steps": 4375, "warmup_steps": 250, "group": "scaling_mha_nh_3", "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": 32, "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": "mha_nh_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": false, "override_use_sdpa": false, "autocast_precision": "bfloat16", "simulate_micro_bs": null, "simulate_micro_bs_2": null, "ckpt_attn": false, "c_proj_scale_init": 1.0}

scaling_mha_nh_3/mha_nh_32_seed_1339/dataloader_04374.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:6031fd3e2855a036f7a5531cc24555aabd1115f9dd6618b8b2ca6f55279ef0b2
+size 964

scaling_mha_nh_3/mha_nh_32_seed_1339/log2.txt

@@ -0,0 +1,529 @@
+max_steps: 4375
+0 val loss 11.2849
+0 val perplexity 79608.3281
+0 train 11.284409 (lr=4.1958e-06) (hash(x)=44992657)
+10 train 9.831003 (lr=4.6154e-05) (hash(x)=33468021)
+20 train 9.167021 (lr=8.8112e-05) (hash(x)=40941803)
+30 train 8.426884 (lr=1.3007e-04) (hash(x)=36715902)
+40 train 7.686311 (lr=1.7203e-04) (hash(x)=32710993)
+50 train 7.409697 (lr=2.1399e-04) (hash(x)=43839896)
+60 train 7.373839 (lr=2.5594e-04) (hash(x)=40675468)
+70 train 6.972479 (lr=2.9790e-04) (hash(x)=34592631)
+80 train 6.985819 (lr=3.3986e-04) (hash(x)=44444845)
+90 train 6.830372 (lr=3.8182e-04) (hash(x)=41965258)
+100 val loss 6.7575
+100 val perplexity 860.5209
+100 train 6.615165 (lr=4.2378e-04) (hash(x)=41284750)
+110 train 6.619215 (lr=4.6573e-04) (hash(x)=41118734)
+120 train 6.503584 (lr=5.0769e-04) (hash(x)=37537547)
+130 train 6.395005 (lr=5.4965e-04) (hash(x)=43625179)
+140 train 6.405569 (lr=5.9161e-04) (hash(x)=41940760)
+150 train 6.491913 (lr=6.3357e-04) (hash(x)=39210431)
+160 train 6.452270 (lr=6.7552e-04) (hash(x)=41128294)
+170 train 6.369477 (lr=7.1748e-04) (hash(x)=41590227)
+180 train 6.320257 (lr=7.5944e-04) (hash(x)=38084871)
+190 train 6.255620 (lr=8.0140e-04) (hash(x)=34534333)
+200 val loss 6.1875
+200 val perplexity 486.6088
+200 train 6.217568 (lr=8.4336e-04) (hash(x)=41614299)
+210 train 6.124395 (lr=8.8531e-04) (hash(x)=35551597)
+220 train 6.082658 (lr=9.2727e-04) (hash(x)=44615432)
+230 train 6.211925 (lr=9.6923e-04) (hash(x)=37078768)
+240 train 5.943058 (lr=1.0112e-03) (hash(x)=40644078)
+250 train 6.002269 (lr=1.0531e-03) (hash(x)=44636066)
+260 train 5.900775 (lr=1.0951e-03) (hash(x)=43247952)
+270 train 5.853110 (lr=1.1371e-03) (hash(x)=37989370)
+280 train 5.799560 (lr=1.1790e-03) (hash(x)=36229730)
+290 train 5.713433 (lr=1.2210e-03) (hash(x)=35032996)
+300 val loss 5.8299
+300 val perplexity 340.3291
+300 train 5.593707 (lr=1.2629e-03) (hash(x)=33868636)
+310 train 5.537184 (lr=1.3049e-03) (hash(x)=39782705)
+320 train 5.609524 (lr=1.3469e-03) (hash(x)=43525765)
+330 train 5.694166 (lr=1.3888e-03) (hash(x)=38074926)
+340 train 5.716688 (lr=1.4308e-03) (hash(x)=50125702)
+350 train 5.696824 (lr=1.4727e-03) (hash(x)=35062222)
+360 train 5.649176 (lr=1.5147e-03) (hash(x)=39671001)
+370 train 5.653901 (lr=1.5566e-03) (hash(x)=35748364)
+380 train 5.466178 (lr=1.5986e-03) (hash(x)=39962666)
+390 train 5.444090 (lr=1.6406e-03) (hash(x)=39213360)
+400 val loss 5.4408
+400 val perplexity 230.6329
+400 train 5.425432 (lr=1.6825e-03) (hash(x)=37919584)
+410 train 5.361440 (lr=1.7245e-03) (hash(x)=42001527)
+420 train 5.358880 (lr=1.7664e-03) (hash(x)=33940005)
+430 train 5.396614 (lr=1.8084e-03) (hash(x)=41749217)
+440 train 5.287040 (lr=1.8503e-03) (hash(x)=45219739)
+450 train 5.276079 (lr=1.8923e-03) (hash(x)=38515090)
+460 train 5.189280 (lr=1.9343e-03) (hash(x)=37518607)
+470 train 5.021706 (lr=1.9762e-03) (hash(x)=46830716)
+480 train 5.136236 (lr=2.0182e-03) (hash(x)=49262598)
+490 train 4.945209 (lr=2.0601e-03) (hash(x)=41370559)
+500 val loss 5.2413
+500 val perplexity 188.9094
+500 train 5.027570 (lr=2.1021e-03) (hash(x)=35237459)
+510 train 5.133389 (lr=2.1441e-03) (hash(x)=35070579)
+520 train 5.178512 (lr=2.1860e-03) (hash(x)=37355110)
+530 train 5.281877 (lr=2.2280e-03) (hash(x)=39398044)
+540 train 5.207469 (lr=2.2699e-03) (hash(x)=39156676)
+550 train 5.170529 (lr=2.3119e-03) (hash(x)=38553909)
+560 train 5.205350 (lr=2.3538e-03) (hash(x)=36555371)
+570 train 5.167586 (lr=2.3958e-03) (hash(x)=42048991)
+580 train 5.096127 (lr=2.4378e-03) (hash(x)=37359399)
+590 train 5.102664 (lr=2.4797e-03) (hash(x)=39254663)
+600 val loss 5.0726
+600 val perplexity 159.5900
+600 train 5.093618 (lr=2.5217e-03) (hash(x)=38982539)
+610 train 5.020190 (lr=2.5636e-03) (hash(x)=35859145)
+620 train 5.000105 (lr=2.6056e-03) (hash(x)=35593542)
+630 train 4.892186 (lr=2.6476e-03) (hash(x)=39124418)
+640 train 4.903229 (lr=2.6895e-03) (hash(x)=39107374)
+650 train 4.846175 (lr=2.7315e-03) (hash(x)=34529151)
+660 train 4.772969 (lr=2.7734e-03) (hash(x)=39483931)
+670 train 4.773403 (lr=2.8154e-03) (hash(x)=43008010)
+680 train 4.683468 (lr=2.8573e-03) (hash(x)=33469328)
+690 train 4.700742 (lr=2.8993e-03) (hash(x)=35025253)
+700 val loss 4.9400
+700 val perplexity 139.7772
+700 train 4.669359 (lr=2.9413e-03) (hash(x)=43280188)
+710 train 4.965689 (lr=2.9832e-03) (hash(x)=40999263)
+720 train 4.938477 (lr=3.0000e-03) (hash(x)=38304742)
+730 train 4.958840 (lr=2.9999e-03) (hash(x)=34705188)
+740 train 4.903385 (lr=2.9997e-03) (hash(x)=40028964)
+750 train 4.862506 (lr=2.9994e-03) (hash(x)=39051591)
+760 train 4.878346 (lr=2.9990e-03) (hash(x)=41286198)
+770 train 4.817084 (lr=2.9985e-03) (hash(x)=44512967)
+780 train 4.847969 (lr=2.9979e-03) (hash(x)=45171627)
+790 train 4.820498 (lr=2.9972e-03) (hash(x)=46345673)
+800 val loss 4.7923
+800 val perplexity 120.5835
+800 train 4.750757 (lr=2.9964e-03) (hash(x)=35529545)
+810 train 4.680483 (lr=2.9955e-03) (hash(x)=42321793)
+820 train 4.693810 (lr=2.9945e-03) (hash(x)=49327569)
+830 train 4.762928 (lr=2.9934e-03) (hash(x)=41325057)
+840 train 4.687345 (lr=2.9922e-03) (hash(x)=37305614)
+850 train 4.635846 (lr=2.9909e-03) (hash(x)=38000800)
+860 train 4.756247 (lr=2.9896e-03) (hash(x)=44047697)
+870 train 4.755713 (lr=2.9881e-03) (hash(x)=41576206)
+880 train 4.722601 (lr=2.9865e-03) (hash(x)=44685458)
+890 train 4.700705 (lr=2.9848e-03) (hash(x)=40398597)
+900 val loss 4.6682
+900 val perplexity 106.5071
+900 train 4.632981 (lr=2.9830e-03) (hash(x)=39312673)
+910 train 4.640726 (lr=2.9811e-03) (hash(x)=38881875)
+920 train 4.679230 (lr=2.9792e-03) (hash(x)=39140166)
+930 train 4.567174 (lr=2.9771e-03) (hash(x)=39075781)
+940 train 4.585098 (lr=2.9749e-03) (hash(x)=46887071)
+950 train 4.565897 (lr=2.9726e-03) (hash(x)=39740603)
+960 train 4.633887 (lr=2.9703e-03) (hash(x)=39590268)
+970 train 4.574183 (lr=2.9678e-03) (hash(x)=43580052)
+980 train 4.524231 (lr=2.9652e-03) (hash(x)=40226735)
+990 train 4.627963 (lr=2.9626e-03) (hash(x)=40723896)
+1000 val loss 4.5654
+1000 val perplexity 96.0977
+1000 train 4.572906 (lr=2.9598e-03) (hash(x)=39408050)
+1010 train 4.745452 (lr=2.9570e-03) (hash(x)=32247950)
+1020 train 4.642047 (lr=2.9540e-03) (hash(x)=39049938)
+1030 train 4.606352 (lr=2.9510e-03) (hash(x)=37453736)
+1040 train 4.516580 (lr=2.9478e-03) (hash(x)=36649666)
+1050 train 4.585294 (lr=2.9446e-03) (hash(x)=41404097)
+1060 train 4.572114 (lr=2.9412e-03) (hash(x)=40701599)
+1070 train 4.585525 (lr=2.9378e-03) (hash(x)=41787393)
+1080 train 4.477023 (lr=2.9343e-03) (hash(x)=42325465)
+1090 train 4.508667 (lr=2.9307e-03) (hash(x)=45018925)
+1100 val loss 4.4948
+1100 val perplexity 89.5547
+1100 train 4.538201 (lr=2.9270e-03) (hash(x)=53751982)
+1110 train 4.465182 (lr=2.9231e-03) (hash(x)=43118376)
+1120 train 4.489386 (lr=2.9192e-03) (hash(x)=41489942)
+1130 train 4.508797 (lr=2.9152e-03) (hash(x)=48020487)
+1140 train 4.369557 (lr=2.9112e-03) (hash(x)=43266072)
+1150 train 4.596982 (lr=2.9070e-03) (hash(x)=41790132)
+1160 train 4.533031 (lr=2.9027e-03) (hash(x)=33739090)
+1170 train 4.501829 (lr=2.8983e-03) (hash(x)=43885030)
+1180 train 4.451200 (lr=2.8939e-03) (hash(x)=39046714)
+1190 train 4.484913 (lr=2.8893e-03) (hash(x)=44387023)
+1200 val loss 4.4181
+1200 val perplexity 82.9398
+1200 train 4.342537 (lr=2.8847e-03) (hash(x)=36667141)
+1210 train 4.419269 (lr=2.8800e-03) (hash(x)=45487179)
+1220 train 4.396080 (lr=2.8751e-03) (hash(x)=43456413)
+1230 train 4.329468 (lr=2.8702e-03) (hash(x)=42473632)
+1240 train 4.400002 (lr=2.8652e-03) (hash(x)=42966315)
+1250 train 4.245781 (lr=2.8601e-03) (hash(x)=42930524)
+1260 train 4.399171 (lr=2.8550e-03) (hash(x)=43467676)
+1270 train 4.320784 (lr=2.8497e-03) (hash(x)=40872580)
+1280 train 4.391618 (lr=2.8443e-03) (hash(x)=40619157)
+1290 train 4.409763 (lr=2.8389e-03) (hash(x)=44406165)
+1300 val loss 4.3727
+1300 val perplexity 79.2569
+1300 train 4.367556 (lr=2.8333e-03) (hash(x)=40808029)
+1310 train 4.551178 (lr=2.8277e-03) (hash(x)=46125736)
+1320 train 4.301901 (lr=2.8220e-03) (hash(x)=41698487)
+1330 train 4.341690 (lr=2.8162e-03) (hash(x)=39302878)
+1340 train 4.457324 (lr=2.8103e-03) (hash(x)=41826369)
+1350 train 4.270795 (lr=2.8044e-03) (hash(x)=37061413)
+1360 train 4.387803 (lr=2.7983e-03) (hash(x)=40770974)
+1370 train 4.311044 (lr=2.7922e-03) (hash(x)=38738718)
+1380 train 4.299088 (lr=2.7860e-03) (hash(x)=34961558)
+1390 train 4.349629 (lr=2.7797e-03) (hash(x)=38559342)
+1400 val loss 4.3539
+1400 val perplexity 77.7826
+1400 train 4.305126 (lr=2.7733e-03) (hash(x)=38840020)
+1410 train 4.465403 (lr=2.7668e-03) (hash(x)=44448100)
+1420 train 4.517131 (lr=2.7603e-03) (hash(x)=39199838)
+1430 train 4.314208 (lr=2.7536e-03) (hash(x)=41669873)
+1440 train 4.448372 (lr=2.7469e-03) (hash(x)=53253836)
+1450 train 4.369516 (lr=2.7401e-03) (hash(x)=41320631)
+1460 train 4.283396 (lr=2.7333e-03) (hash(x)=48462803)
+1470 train 4.325759 (lr=2.7263e-03) (hash(x)=42356864)
+1480 train 4.366220 (lr=2.7193e-03) (hash(x)=41118068)
+1490 train 4.277613 (lr=2.7121e-03) (hash(x)=37242854)
+1500 val loss 4.3002
+1500 val perplexity 73.7170
+1500 train 4.328384 (lr=2.7050e-03) (hash(x)=40571031)
+1510 train 4.283280 (lr=2.6977e-03) (hash(x)=35384608)
+1520 train 4.307294 (lr=2.6903e-03) (hash(x)=35341056)
+1530 train 4.199541 (lr=2.6829e-03) (hash(x)=36471779)
+1540 train 4.235432 (lr=2.6754e-03) (hash(x)=40356153)
+1550 train 4.273212 (lr=2.6678e-03) (hash(x)=40133377)
+1560 train 4.204952 (lr=2.6602e-03) (hash(x)=33722160)
+1570 train 4.362757 (lr=2.6525e-03) (hash(x)=39555295)
+1580 train 4.255520 (lr=2.6447e-03) (hash(x)=35950666)
+1590 train 4.406691 (lr=2.6368e-03) (hash(x)=36528535)
+1600 val loss 4.2883
+1600 val perplexity 72.8438
+1600 train 4.304619 (lr=2.6289e-03) (hash(x)=42613017)
+1610 train 4.445381 (lr=2.6208e-03) (hash(x)=42845965)
+1620 train 4.323762 (lr=2.6128e-03) (hash(x)=36137074)
+1630 train 4.292446 (lr=2.6046e-03) (hash(x)=37560569)
+1640 train 4.381917 (lr=2.5964e-03) (hash(x)=43680614)
+1650 train 4.239828 (lr=2.5881e-03) (hash(x)=37281218)
+1660 train 4.280923 (lr=2.5797e-03) (hash(x)=36165222)
+1670 train 4.224460 (lr=2.5713e-03) (hash(x)=50222513)
+1680 train 4.120292 (lr=2.5628e-03) (hash(x)=35239173)
+1690 train 4.166812 (lr=2.5542e-03) (hash(x)=40287349)
+1700 val loss 4.2866
+1700 val perplexity 72.7201
+1700 train 4.354477 (lr=2.5455e-03) (hash(x)=37444780)
+1710 train 4.344234 (lr=2.5368e-03) (hash(x)=38962723)
+1720 train 4.336098 (lr=2.5281e-03) (hash(x)=33882305)
+1730 train 4.320095 (lr=2.5192e-03) (hash(x)=36501540)
+1740 train 4.315228 (lr=2.5103e-03) (hash(x)=38723266)
+1750 train 4.245676 (lr=2.5014e-03) (hash(x)=38772789)
+1760 train 4.323306 (lr=2.4924e-03) (hash(x)=37707898)
+1770 train 4.188629 (lr=2.4833e-03) (hash(x)=30529327)
+1780 train 4.270228 (lr=2.4741e-03) (hash(x)=37885464)
+1790 train 4.282052 (lr=2.4649e-03) (hash(x)=39013967)
+1800 val loss 4.2391
+1800 val perplexity 69.3466
+1800 train 4.320072 (lr=2.4556e-03) (hash(x)=34906955)
+1810 train 4.196305 (lr=2.4463e-03) (hash(x)=40790682)
+1820 train 4.062162 (lr=2.4369e-03) (hash(x)=39912223)
+1830 train 4.192920 (lr=2.4275e-03) (hash(x)=39304486)
+1840 train 4.163272 (lr=2.4180e-03) (hash(x)=37247975)
+1850 train 4.153632 (lr=2.4084e-03) (hash(x)=41340614)
+1860 train 4.314875 (lr=2.3988e-03) (hash(x)=39924627)
+1870 train 4.292680 (lr=2.3891e-03) (hash(x)=40235993)
+1880 train 4.184875 (lr=2.3794e-03) (hash(x)=37693476)
+1890 train 4.294076 (lr=2.3696e-03) (hash(x)=40449864)
+1900 val loss 4.2114
+1900 val perplexity 67.4502
+1900 train 4.197103 (lr=2.3598e-03) (hash(x)=37907749)
+1910 train 4.171101 (lr=2.3499e-03) (hash(x)=34547901)
+1920 train 4.203171 (lr=2.3400e-03) (hash(x)=37771788)
+1930 train 4.074045 (lr=2.3300e-03) (hash(x)=36049454)
+1940 train 4.171709 (lr=2.3200e-03) (hash(x)=29564336)
+1950 train 4.164992 (lr=2.3099e-03) (hash(x)=41689281)
+1960 train 4.074987 (lr=2.2998e-03) (hash(x)=41976699)
+1970 train 4.280608 (lr=2.2896e-03) (hash(x)=41538621)
+1980 train 4.160500 (lr=2.2793e-03) (hash(x)=36490902)
+1990 train 4.192510 (lr=2.2691e-03) (hash(x)=42739315)
+2000 val loss 4.1902
+2000 val perplexity 66.0373
+2000 train 4.201960 (lr=2.2588e-03) (hash(x)=38600074)
+2010 train 4.217104 (lr=2.2484e-03) (hash(x)=34555488)
+2020 train 4.238808 (lr=2.2380e-03) (hash(x)=37227345)
+2030 train 4.186332 (lr=2.2275e-03) (hash(x)=38172386)
+2040 train 4.301264 (lr=2.2170e-03) (hash(x)=39340786)
+2050 train 4.147235 (lr=2.2065e-03) (hash(x)=37237298)
+2060 train 4.382494 (lr=2.1959e-03) (hash(x)=35894782)
+2070 train 4.204191 (lr=2.1853e-03) (hash(x)=44161861)
+2080 train 4.218318 (lr=2.1746e-03) (hash(x)=38062558)
+2090 train 4.066265 (lr=2.1639e-03) (hash(x)=38608301)
+2100 val loss 4.1659
+2100 val perplexity 64.4518
+2100 train 4.097606 (lr=2.1532e-03) (hash(x)=47611537)
+2110 train 4.093205 (lr=2.1424e-03) (hash(x)=38076651)
+2120 train 4.110522 (lr=2.1316e-03) (hash(x)=39870893)
+2130 train 4.249832 (lr=2.1208e-03) (hash(x)=48123321)
+2140 train 4.279020 (lr=2.1099e-03) (hash(x)=44124977)
+2150 train 4.180139 (lr=2.0990e-03) (hash(x)=42181374)
+2160 train 4.216885 (lr=2.0881e-03) (hash(x)=41837878)
+2170 train 4.226863 (lr=2.0771e-03) (hash(x)=42161375)
+2180 train 4.200527 (lr=2.0661e-03) (hash(x)=41792986)
+2190 train 4.164966 (lr=2.0550e-03) (hash(x)=41872416)
+2200 val loss 4.1439
+2200 val perplexity 63.0464
+2200 train 4.139199 (lr=2.0440e-03) (hash(x)=37395985)
+2210 train 4.192711 (lr=2.0329e-03) (hash(x)=40616096)
+2220 train 4.094735 (lr=2.0217e-03) (hash(x)=40118423)
+2230 train 4.005547 (lr=2.0106e-03) (hash(x)=39641586)
+2240 train 4.093243 (lr=1.9994e-03) (hash(x)=34720116)
+2250 train 4.038753 (lr=1.9882e-03) (hash(x)=43641508)
+2260 train 4.055352 (lr=1.9770e-03) (hash(x)=31179786)
+2270 train 4.031299 (lr=1.9657e-03) (hash(x)=42721932)
+2280 train 4.354287 (lr=1.9544e-03) (hash(x)=38474505)
+2290 train 4.181806 (lr=1.9431e-03) (hash(x)=41398545)
+2300 val loss 4.1240
+2300 val perplexity 61.8057
+2300 train 4.113652 (lr=1.9318e-03) (hash(x)=44131094)
+2310 train 4.191688 (lr=1.9205e-03) (hash(x)=41667948)
+2320 train 4.271818 (lr=1.9091e-03) (hash(x)=36339232)
+2330 train 4.146030 (lr=1.8977e-03) (hash(x)=41630644)
+2340 train 4.235251 (lr=1.8863e-03) (hash(x)=41278955)
+2350 train 4.110110 (lr=1.8749e-03) (hash(x)=39491976)
+2360 train 4.176768 (lr=1.8635e-03) (hash(x)=41771238)
+2370 train 4.264560 (lr=1.8520e-03) (hash(x)=45374570)
+2380 train 4.142990 (lr=1.8406e-03) (hash(x)=41419267)
+2390 train 4.181340 (lr=1.8291e-03) (hash(x)=38869169)
+2400 val loss 4.1123
+2400 val perplexity 61.0857
+2400 train 4.078676 (lr=1.8176e-03) (hash(x)=38619293)
+2410 train 3.900402 (lr=1.8061e-03) (hash(x)=41600240)
+2420 train 4.027859 (lr=1.7946e-03) (hash(x)=40891045)
+2430 train 4.028064 (lr=1.7830e-03) (hash(x)=41023249)
+2440 train 3.996095 (lr=1.7715e-03) (hash(x)=33813452)
+2450 train 3.941894 (lr=1.7600e-03) (hash(x)=38464119)
+2460 train 3.836749 (lr=1.7484e-03) (hash(x)=40699982)
+2470 train 3.848376 (lr=1.7368e-03) (hash(x)=38254854)
+2480 train 4.139463 (lr=1.7253e-03) (hash(x)=42736069)
+2490 train 4.147765 (lr=1.7137e-03) (hash(x)=41928525)
+2500 val loss 4.0912
+2500 val perplexity 59.8138
+2500 train 4.107133 (lr=1.7021e-03) (hash(x)=45864011)
+2510 train 4.118594 (lr=1.6906e-03) (hash(x)=41380960)
+2520 train 4.108972 (lr=1.6790e-03) (hash(x)=38366545)
+2530 train 4.111085 (lr=1.6674e-03) (hash(x)=41045176)
+2540 train 4.141745 (lr=1.6558e-03) (hash(x)=43435705)
+2550 train 4.096506 (lr=1.6442e-03) (hash(x)=40693090)
+2560 train 4.015873 (lr=1.6326e-03) (hash(x)=40502478)
+2570 train 4.044238 (lr=1.6210e-03) (hash(x)=38610920)
+2580 train 3.969744 (lr=1.6094e-03) (hash(x)=38333499)
+2590 train 4.043532 (lr=1.5979e-03) (hash(x)=41982736)
+2600 val loss 4.0733
+2600 val perplexity 58.7498
+2600 train 4.021665 (lr=1.5863e-03) (hash(x)=37724702)
+2610 train 3.962540 (lr=1.5747e-03) (hash(x)=39564630)
+2620 train 3.908698 (lr=1.5632e-03) (hash(x)=38465803)
+2630 train 3.826732 (lr=1.5516e-03) (hash(x)=36955007)
+2640 train 3.864953 (lr=1.5400e-03) (hash(x)=32994568)
+2650 train 4.049616 (lr=1.5285e-03) (hash(x)=41165765)
+2660 train 4.064397 (lr=1.5170e-03) (hash(x)=42934000)
+2670 train 4.260743 (lr=1.5054e-03) (hash(x)=42726316)
+2680 train 4.149772 (lr=1.4939e-03) (hash(x)=35880318)
+2690 train 4.029663 (lr=1.4824e-03) (hash(x)=41747329)
+2700 val loss 4.0628
+2700 val perplexity 58.1368
+2700 train 4.156364 (lr=1.4709e-03) (hash(x)=40259630)
+2710 train 4.272419 (lr=1.4594e-03) (hash(x)=39895798)
+2720 train 4.002565 (lr=1.4480e-03) (hash(x)=36146683)
+2730 train 3.984866 (lr=1.4365e-03) (hash(x)=36181984)
+2740 train 3.963217 (lr=1.4251e-03) (hash(x)=43700349)
+2750 train 4.028590 (lr=1.4137e-03) (hash(x)=41610597)
+2760 train 4.020451 (lr=1.4023e-03) (hash(x)=31183639)
+2770 train 4.011908 (lr=1.3909e-03) (hash(x)=37722489)
+2780 train 3.957323 (lr=1.3795e-03) (hash(x)=47290688)
+2790 train 3.995935 (lr=1.3682e-03) (hash(x)=41205574)
+2800 val loss 4.0468
+2800 val perplexity 57.2118
+2800 train 3.849253 (lr=1.3569e-03) (hash(x)=40257962)
+2810 train 3.819733 (lr=1.3456e-03) (hash(x)=39529014)
+2820 train 3.842391 (lr=1.3343e-03) (hash(x)=42244749)
+2830 train 3.894398 (lr=1.3230e-03) (hash(x)=32820090)
+2840 train 3.849078 (lr=1.3118e-03) (hash(x)=40315769)
+2850 train 4.093531 (lr=1.3006e-03) (hash(x)=41524462)
+2860 train 4.084390 (lr=1.2894e-03) (hash(x)=38365734)
+2870 train 4.221619 (lr=1.2783e-03) (hash(x)=37682602)
+2880 train 4.069720 (lr=1.2671e-03) (hash(x)=39162991)
+2890 train 3.924066 (lr=1.2560e-03) (hash(x)=33316384)
+2900 val loss 4.0230
+2900 val perplexity 55.8688
+2900 train 3.979747 (lr=1.2450e-03) (hash(x)=37271132)
+2910 train 4.163024 (lr=1.2339e-03) (hash(x)=35586242)
+2920 train 4.051622 (lr=1.2229e-03) (hash(x)=33320586)
+2930 train 4.063995 (lr=1.2119e-03) (hash(x)=43531361)
+2940 train 3.954144 (lr=1.2010e-03) (hash(x)=37368286)
+2950 train 3.991661 (lr=1.1901e-03) (hash(x)=40363394)
+2960 train 3.957000 (lr=1.1792e-03) (hash(x)=42253792)
+2970 train 3.902286 (lr=1.1684e-03) (hash(x)=38072598)
+2980 train 3.970243 (lr=1.1576e-03) (hash(x)=41470557)
+2990 train 3.898298 (lr=1.1468e-03) (hash(x)=42600033)
+3000 val loss 4.0212
+3000 val perplexity 55.7673
+3000 train 3.847022 (lr=1.1361e-03) (hash(x)=46890983)
+3010 train 3.721937 (lr=1.1254e-03) (hash(x)=37986759)
+3020 train 3.846706 (lr=1.1147e-03) (hash(x)=36270703)
+3030 train 4.112988 (lr=1.1041e-03) (hash(x)=38228599)
+3040 train 3.962567 (lr=1.0935e-03) (hash(x)=44344296)
+3050 train 4.049918 (lr=1.0830e-03) (hash(x)=39962297)
+3060 train 4.037074 (lr=1.0725e-03) (hash(x)=39817394)
+3070 train 4.023964 (lr=1.0620e-03) (hash(x)=41763868)
+3080 train 4.106663 (lr=1.0516e-03) (hash(x)=42343051)
+3090 train 4.106384 (lr=1.0412e-03) (hash(x)=39204893)
+3100 val loss 3.9974
+3100 val perplexity 54.4559
+3100 train 4.006607 (lr=1.0309e-03) (hash(x)=39353599)
+3110 train 4.074608 (lr=1.0207e-03) (hash(x)=36118073)
+3120 train 3.767110 (lr=1.0104e-03) (hash(x)=44070271)
+3130 train 3.802597 (lr=1.0002e-03) (hash(x)=43368151)
+3140 train 3.878448 (lr=9.9011e-04) (hash(x)=40612203)
+3150 train 4.015577 (lr=9.8002e-04) (hash(x)=39598447)
+3160 train 4.071764 (lr=9.6999e-04) (hash(x)=43897682)
+3170 train 4.051023 (lr=9.6000e-04) (hash(x)=39029243)
+3180 train 4.060163 (lr=9.5007e-04) (hash(x)=38586684)
+3190 train 4.012737 (lr=9.4019e-04) (hash(x)=39299115)
+3200 val loss 3.9783
+3200 val perplexity 53.4270
+3200 train 4.042639 (lr=9.3036e-04) (hash(x)=41064949)
+3210 train 4.026763 (lr=9.2058e-04) (hash(x)=41752822)
+3220 train 4.034617 (lr=9.1085e-04) (hash(x)=37484805)
+3230 train 3.996578 (lr=9.0118e-04) (hash(x)=40514919)
+3240 train 4.053064 (lr=8.9157e-04) (hash(x)=42440343)
+3250 train 3.949322 (lr=8.8201e-04) (hash(x)=32464136)
+3260 train 3.966358 (lr=8.7251e-04) (hash(x)=47270558)
+3270 train 3.907974 (lr=8.6307e-04) (hash(x)=39307546)
+3280 train 3.942009 (lr=8.5368e-04) (hash(x)=40410299)
+3290 train 3.937995 (lr=8.4435e-04) (hash(x)=40541497)
+3300 val loss 3.9715
+3300 val perplexity 53.0619
+3300 train 3.841106 (lr=8.3508e-04) (hash(x)=38089139)
+3310 train 3.872728 (lr=8.2588e-04) (hash(x)=42006576)
+3320 train 3.963868 (lr=8.1673e-04) (hash(x)=40422556)
+3330 train 3.756885 (lr=8.0764e-04) (hash(x)=39021643)
+3340 train 4.019528 (lr=7.9862e-04) (hash(x)=40434305)
+3350 train 4.049447 (lr=7.8966e-04) (hash(x)=40833559)
+3360 train 4.047432 (lr=7.8076e-04) (hash(x)=40045231)
+3370 train 4.009585 (lr=7.7192e-04) (hash(x)=34668317)
+3380 train 4.006077 (lr=7.6315e-04) (hash(x)=42738568)
+3390 train 4.101512 (lr=7.5445e-04) (hash(x)=38980166)
+3400 val loss 3.9511
+3400 val perplexity 51.9939
+3400 train 4.009703 (lr=7.4581e-04) (hash(x)=43878176)
+3410 train 4.051007 (lr=7.3724e-04) (hash(x)=38201991)
+3420 train 4.029405 (lr=7.2874e-04) (hash(x)=34905889)
+3430 train 4.029083 (lr=7.2030e-04) (hash(x)=41680405)
+3440 train 3.987942 (lr=7.1193e-04) (hash(x)=39948184)
+3450 train 4.036534 (lr=7.0363e-04) (hash(x)=33623728)
+3460 train 3.910428 (lr=6.9541e-04) (hash(x)=41027647)
+3470 train 3.944430 (lr=6.8725e-04) (hash(x)=42779482)
+3480 train 3.956617 (lr=6.7916e-04) (hash(x)=41579288)
+3490 train 3.816310 (lr=6.7114e-04) (hash(x)=37174846)
+3500 val loss 3.9469
+3500 val perplexity 51.7727
+3500 train 3.690612 (lr=6.6320e-04) (hash(x)=37082090)
+3510 train 3.885297 (lr=6.5533e-04) (hash(x)=43131340)
+3520 train 3.804730 (lr=6.4753e-04) (hash(x)=42457250)
+3530 train 4.040538 (lr=6.3981e-04) (hash(x)=39504492)
+3540 train 3.982667 (lr=6.3216e-04) (hash(x)=36225997)
+3550 train 4.110770 (lr=6.2458e-04) (hash(x)=40912087)
+3560 train 3.950840 (lr=6.1708e-04) (hash(x)=41545925)
+3570 train 4.040921 (lr=6.0966e-04) (hash(x)=32553193)
+3580 train 3.881988 (lr=6.0231e-04) (hash(x)=38799856)
+3590 train 3.928826 (lr=5.9504e-04) (hash(x)=40772075)
+3600 val loss 3.9267
+3600 val perplexity 50.7413
+3600 train 3.969097 (lr=5.8785e-04) (hash(x)=39299903)
+3610 train 4.017958 (lr=5.8074e-04) (hash(x)=39503247)
+3620 train 3.905115 (lr=5.7370e-04) (hash(x)=38730668)
+3630 train 3.928154 (lr=5.6675e-04) (hash(x)=40503276)
+3640 train 3.936673 (lr=5.5987e-04) (hash(x)=36504573)
+3650 train 3.955630 (lr=5.5308e-04) (hash(x)=44903075)
+3660 train 3.867089 (lr=5.4636e-04) (hash(x)=33800148)
+3670 train 3.778673 (lr=5.3973e-04) (hash(x)=39404541)
+3680 train 3.720395 (lr=5.3318e-04) (hash(x)=44777432)
+3690 train 3.818221 (lr=5.2671e-04) (hash(x)=45649913)
+3700 val loss 3.9291
+3700 val perplexity 50.8611
+3700 train 3.948917 (lr=5.2033e-04) (hash(x)=33812537)
+3710 train 4.091668 (lr=5.1402e-04) (hash(x)=36161222)
+3720 train 3.969955 (lr=5.0780e-04) (hash(x)=43690616)
+3730 train 3.976103 (lr=5.0167e-04) (hash(x)=35462313)
+3740 train 4.267167 (lr=4.9562e-04) (hash(x)=44674254)
+3750 train 3.958643 (lr=4.8965e-04) (hash(x)=44276297)
+3760 train 3.996600 (lr=4.8377e-04) (hash(x)=38492178)
+3770 train 4.002790 (lr=4.7798e-04) (hash(x)=44347531)
+3780 train 3.919357 (lr=4.7227e-04) (hash(x)=33792059)
+3790 train 4.008738 (lr=4.6665e-04) (hash(x)=38549318)
+3800 val loss 3.9102
+3800 val perplexity 49.9068
+3800 train 3.893719 (lr=4.6112e-04) (hash(x)=41437448)
+3810 train 3.794036 (lr=4.5567e-04) (hash(x)=36656768)
+3820 train 3.952337 (lr=4.5031e-04) (hash(x)=46387893)
+3830 train 3.956923 (lr=4.4504e-04) (hash(x)=44312129)
+3840 train 3.876998 (lr=4.3986e-04) (hash(x)=40341896)
+3850 train 3.925651 (lr=4.3477e-04) (hash(x)=43589160)
+3860 train 3.904959 (lr=4.2977e-04) (hash(x)=40388945)
+3870 train 3.807533 (lr=4.2486e-04) (hash(x)=33336679)
+3880 train 3.902813 (lr=4.2004e-04) (hash(x)=37643311)
+3890 train 3.927500 (lr=4.1530e-04) (hash(x)=38272927)
+3900 val loss 3.9024
+3900 val perplexity 49.5198
+3900 train 3.871929 (lr=4.1066e-04) (hash(x)=41925748)
+3910 train 3.964934 (lr=4.0611e-04) (hash(x)=38377253)
+3920 train 3.967288 (lr=4.0166e-04) (hash(x)=41616611)
+3930 train 3.890297 (lr=3.9729e-04) (hash(x)=32301827)
+3940 train 3.944790 (lr=3.9302e-04) (hash(x)=47697363)
+3950 train 3.896632 (lr=3.8884e-04) (hash(x)=37867767)
+3960 train 3.893064 (lr=3.8475e-04) (hash(x)=37187295)
+3970 train 3.948962 (lr=3.8076e-04) (hash(x)=41952752)
+3980 train 3.919329 (lr=3.7685e-04) (hash(x)=38358660)
+3990 train 3.787942 (lr=3.7305e-04) (hash(x)=40207878)
+4000 val loss 3.8958
+4000 val perplexity 49.1976
+4000 train 3.876563 (lr=3.6933e-04) (hash(x)=39134015)
+4010 train 3.945871 (lr=3.6572e-04) (hash(x)=38313135)
+4020 train 3.876674 (lr=3.6219e-04) (hash(x)=40710513)
+4030 train 4.005750 (lr=3.5876e-04) (hash(x)=35530165)
+4040 train 3.929139 (lr=3.5543e-04) (hash(x)=39494066)
+4050 train 3.918313 (lr=3.5219e-04) (hash(x)=40861777)
+4060 train 3.895581 (lr=3.4905e-04) (hash(x)=36151939)
+4070 train 3.862994 (lr=3.4600e-04) (hash(x)=39008241)
+4080 train 3.950357 (lr=3.4305e-04) (hash(x)=40395746)
+4090 train 3.984824 (lr=3.4019e-04) (hash(x)=38855480)
+4100 val loss 3.8839
+4100 val perplexity 48.6128
+4100 train 3.923942 (lr=3.3744e-04) (hash(x)=45013254)
+4110 train 3.926099 (lr=3.3477e-04) (hash(x)=39934818)
+4120 train 3.929821 (lr=3.3221e-04) (hash(x)=33417705)
+4130 train 3.862675 (lr=3.2974e-04) (hash(x)=41267978)
+4140 train 3.860227 (lr=3.2737e-04) (hash(x)=38961609)
+4150 train 3.864541 (lr=3.2510e-04) (hash(x)=42173878)
+4160 train 3.826050 (lr=3.2292e-04) (hash(x)=36867405)
+4170 train 3.805605 (lr=3.2085e-04) (hash(x)=38991954)
+4180 train 3.872111 (lr=3.1887e-04) (hash(x)=39032592)
+4190 train 3.848115 (lr=3.1699e-04) (hash(x)=61329402)
+4200 val loss 3.8850
+4200 val perplexity 48.6676
+4200 train 3.862508 (lr=3.1520e-04) (hash(x)=40455200)
+4210 train 3.892125 (lr=3.1352e-04) (hash(x)=40913417)
+4220 train 3.916988 (lr=3.1193e-04) (hash(x)=38809704)
+4230 train 3.905368 (lr=3.1044e-04) (hash(x)=39560204)
+4240 train 3.906725 (lr=3.0905e-04) (hash(x)=41496549)
+4250 train 3.941051 (lr=3.0776e-04) (hash(x)=33994768)
+4260 train 3.925737 (lr=3.0657e-04) (hash(x)=39658687)
+4270 train 3.809272 (lr=3.0548e-04) (hash(x)=42583064)
+4280 train 3.873438 (lr=3.0449e-04) (hash(x)=40635811)
+4290 train 3.773278 (lr=3.0359e-04) (hash(x)=35313841)
+4300 val loss 3.8802
+4300 val perplexity 48.4321
+4300 train 3.952422 (lr=3.0280e-04) (hash(x)=39332245)
+4310 train 3.848435 (lr=3.0210e-04) (hash(x)=43732999)
+4320 train 3.860212 (lr=3.0150e-04) (hash(x)=40831835)
+4330 train 3.993501 (lr=3.0101e-04) (hash(x)=41813368)
+4340 train 3.896585 (lr=3.0061e-04) (hash(x)=36952586)
+4350 train 3.916270 (lr=3.0031e-04) (hash(x)=42470745)
+4360 train 3.846249 (lr=3.0011e-04) (hash(x)=39581495)
+4370 train 4.033044 (lr=3.0001e-04) (hash(x)=35914538)
+4374 val loss 3.8734
+4374 val perplexity 48.1062

scaling_mha_nh_3/mha_nh_32_seed_1339/model_04374.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:e8770deb2ab238185dbe849cbbcc380a80e26714d73713580a0e68537faa71ed
+size 130706626

scaling_mha_nh_3/mha_nh_32_seed_1339/optimizer_04374.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:5e8c7f1ae87ae21c791bcf3e46b0eb2d7b1401e217a1568a3e7b5e0b1650d2bd
+size 255125126