Upload folder using huggingface_hub

two_masks_shared_1_head/args.json

@@ -1 +1 @@
-{"hellaswag": true, "attention_kind": "selective", "log_dir": "allowing_more_selection_patterns/two_masks_shared_1_head", "resume_checkpoint": null, "resume_optimizer": false, "add_a_head": false, "add_head_to_start": true, "new_head_init": "normal", "n_heads": 12, "n_embd": 768, "head_dim": 64, "protect_bos_token": true, "prevent_from_masking_myself": true, "max_steps": null, "warmup_steps": null, "group": "allowing_more_selection_patterns", "use_wandb": true, "kill_self_after_run": false, "random_seed": 1337, "memory_penalty_epsilon": 0.1, "selection_head_linear_combo": "n_sliced_masks", "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": null, "batch_size": 4, "total_batch_size": null, "protection_head_scaling_factor": 1.0, "protection_head_bias": 0.0, "n_sliced_masks": 2, "n_latent_masks": null, "mask_layernorm": false, "residual_attention_masks": false, "compute_base_shapes": false, "base_shapes_savefile": null, "mup": false, "disable_selection": false, "mup_enable_coord_check_logging": false, "max_lr": null, "decay_lr": true, "readout_zero_init": false, "query_zero_init": false, "l1_loss": false, "debugpy": false, "key": null}
+{"hellaswag": true, "attention_kind": "selective", "log_dir": "allowing_more_selection_patterns/two_masks_shared_1_head", "resume_checkpoint": null, "resume_optimizer": false, "add_a_head": false, "add_head_to_start": true, "new_head_init": "normal", "n_heads": 12, "n_embd": 768, "head_dim": 64, "protect_bos_token": true, "prevent_from_masking_myself": true, "max_steps": null, "warmup_steps": null, "group": "allowing_more_selection_patterns", "use_wandb": true, "kill_self_after_run": false, "random_seed": 1339, "memory_penalty_epsilon": 0.1, "selection_head_linear_combo": "n_sliced_masks", "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": null, "batch_size": 4, "total_batch_size": null, "protection_head_scaling_factor": 1.0, "protection_head_bias": 0.0, "n_sliced_masks": 2, "n_latent_masks": null, "mask_layernorm": false, "residual_attention_masks": false, "compute_base_shapes": false, "base_shapes_savefile": null, "mup": false, "disable_selection": false, "mup_enable_coord_check_logging": false, "max_lr": null, "decay_lr": true, "readout_zero_init": false, "query_zero_init": false, "l1_loss": false, "debugpy": false, "key": null}

two_masks_shared_1_head/log2.txt

@@ -1,303 +1,303 @@
 max_steps: 2500
-0 val loss 11.0522
-0 val perplexity 63080.1992
-0 train 11.053495 (lr=8.3916e-07) (hash(x)=23347629)
-10 train 10.213490 (lr=9.2308e-06) (hash(x)=19636655)
-20 train 9.679847 (lr=1.7622e-05) (hash(x)=24490248)
-30 train 9.489542 (lr=2.6014e-05) (hash(x)=21733558)
-40 train 9.239005 (lr=3.4406e-05) (hash(x)=31750616)
-50 train 8.930004 (lr=4.2797e-05) (hash(x)=20041895)
-60 train 8.572167 (lr=5.1189e-05) (hash(x)=21416280)
-70 train 8.240636 (lr=5.9580e-05) (hash(x)=18066780)
-80 train 7.928938 (lr=6.7972e-05) (hash(x)=21809427)
-90 train 7.686296 (lr=7.6364e-05) (hash(x)=22061595)
-100 val loss 7.4653
-100 val perplexity 1746.4028
-100 train 7.416183 (lr=8.4755e-05) (hash(x)=17468914)
-110 train 7.220499 (lr=9.3147e-05) (hash(x)=20421520)
-120 train 7.225886 (lr=1.0154e-04) (hash(x)=23140133)
-130 train 7.095161 (lr=1.0993e-04) (hash(x)=20006835)
-140 train 6.849404 (lr=1.1832e-04) (hash(x)=18628134)
-150 train 6.836167 (lr=1.2671e-04) (hash(x)=21913447)
-160 train 6.757919 (lr=1.3510e-04) (hash(x)=22997297)
-170 train 6.564948 (lr=1.4350e-04) (hash(x)=17904821)
-180 train 6.430589 (lr=1.5189e-04) (hash(x)=18826982)
-190 train 6.505245 (lr=1.6028e-04) (hash(x)=21387716)
-200 val loss 6.3835
-200 val perplexity 592.0068
-200 train 6.308176 (lr=1.6867e-04) (hash(x)=16236395)
-210 train 6.176318 (lr=1.7706e-04) (hash(x)=21428681)
-220 train 6.334815 (lr=1.8545e-04) (hash(x)=20461738)
-230 train 6.144704 (lr=1.9385e-04) (hash(x)=20394865)
-240 train 6.047451 (lr=2.0224e-04) (hash(x)=18028671)
-250 train 5.904708 (lr=2.1063e-04) (hash(x)=18122073)
-260 train 5.835814 (lr=2.1902e-04) (hash(x)=22994143)
-270 train 6.015463 (lr=2.2741e-04) (hash(x)=20646260)
-280 train 5.862163 (lr=2.3580e-04) (hash(x)=23204708)
-290 train 5.774846 (lr=2.4420e-04) (hash(x)=17900252)
-300 val loss 5.8051
-300 val perplexity 331.9947
-300 train 5.650853 (lr=2.5259e-04) (hash(x)=19073819)
-310 train 5.827680 (lr=2.6098e-04) (hash(x)=22378232)
-320 train 5.732667 (lr=2.6937e-04) (hash(x)=19910430)
-330 train 5.677101 (lr=2.7776e-04) (hash(x)=25168753)
-340 train 5.544260 (lr=2.8615e-04) (hash(x)=20665256)
-350 train 5.432979 (lr=2.9455e-04) (hash(x)=19951160)
-360 train 5.615287 (lr=3.0294e-04) (hash(x)=21178202)
-370 train 5.482204 (lr=3.1133e-04) (hash(x)=20435886)
-380 train 5.408401 (lr=3.1972e-04) (hash(x)=21255097)
-390 train 5.417608 (lr=3.2811e-04) (hash(x)=19596330)
-400 val loss 5.3747
-400 val perplexity 215.8802
-400 train 5.400533 (lr=3.3650e-04) (hash(x)=21789426)
-410 train 5.304148 (lr=3.4490e-04) (hash(x)=18366401)
-420 train 5.191407 (lr=3.5329e-04) (hash(x)=18614660)
-430 train 5.322891 (lr=3.6168e-04) (hash(x)=21649063)
-440 train 5.202750 (lr=3.7007e-04) (hash(x)=18505097)
-450 train 5.195105 (lr=3.7846e-04) (hash(x)=23434312)
-460 train 5.106994 (lr=3.8685e-04) (hash(x)=16584569)
-470 train 4.979374 (lr=3.9524e-04) (hash(x)=13447111)
-480 train 5.198777 (lr=4.0364e-04) (hash(x)=22645130)
-490 train 5.037940 (lr=4.1203e-04) (hash(x)=19362777)
-500 val loss 5.0225
-500 val perplexity 151.7935
-500 train 4.956991 (lr=4.2042e-04) (hash(x)=20921874)
-510 train 4.836232 (lr=4.2881e-04) (hash(x)=22123048)
-520 train 4.888699 (lr=4.3720e-04) (hash(x)=21215368)
-530 train 5.016878 (lr=4.4559e-04) (hash(x)=20717936)
-540 train 4.862735 (lr=4.5399e-04) (hash(x)=20118007)
-550 train 4.751873 (lr=4.6238e-04) (hash(x)=21402522)
-560 train 4.704627 (lr=4.7077e-04) (hash(x)=19057180)
-570 train 4.852227 (lr=4.7916e-04) (hash(x)=19360553)
-580 train 4.787922 (lr=4.8755e-04) (hash(x)=18194693)
-590 train 4.659945 (lr=4.9594e-04) (hash(x)=19863826)
-600 val loss 4.6399
-600 val perplexity 103.5357
-600 train 4.590632 (lr=5.0434e-04) (hash(x)=21794551)
-610 train 4.552748 (lr=5.1273e-04) (hash(x)=18685622)
-620 train 4.433325 (lr=5.2112e-04) (hash(x)=23473104)
-630 train 4.537632 (lr=5.2951e-04) (hash(x)=18671672)
-640 train 4.550228 (lr=5.3790e-04) (hash(x)=19234240)
-650 train 4.443274 (lr=5.4629e-04) (hash(x)=23183903)
-660 train 4.294300 (lr=5.5469e-04) (hash(x)=23665543)
-670 train 4.447758 (lr=5.6308e-04) (hash(x)=22055313)
-680 train 4.416853 (lr=5.7147e-04) (hash(x)=22321668)
-690 train 4.514027 (lr=5.7986e-04) (hash(x)=21568926)
-700 val loss 4.3671
-700 val perplexity 78.8165
-700 train 4.372666 (lr=5.8825e-04) (hash(x)=19783143)
-710 train 4.289976 (lr=5.9664e-04) (hash(x)=21565258)
-720 train 4.387317 (lr=5.9999e-04) (hash(x)=16127212)
-730 train 4.297870 (lr=5.9991e-04) (hash(x)=19658174)
-740 train 4.264292 (lr=5.9974e-04) (hash(x)=16454319)
-750 train 4.217838 (lr=5.9949e-04) (hash(x)=17993154)
-760 train 4.113863 (lr=5.9915e-04) (hash(x)=19256989)
-770 train 4.092350 (lr=5.9874e-04) (hash(x)=17351653)
-780 train 4.125974 (lr=5.9824e-04) (hash(x)=21626120)
-790 train 4.253780 (lr=5.9765e-04) (hash(x)=16367412)
-800 val loss 4.1793
-800 val perplexity 65.3211
-800 train 4.132425 (lr=5.9698e-04) (hash(x)=22321504)
-810 train 4.005775 (lr=5.9623e-04) (hash(x)=19869755)
-820 train 4.195062 (lr=5.9540e-04) (hash(x)=24183665)
-830 train 4.152798 (lr=5.9449e-04) (hash(x)=19194113)
-840 train 4.012914 (lr=5.9349e-04) (hash(x)=17710503)
-850 train 4.012138 (lr=5.9241e-04) (hash(x)=17231851)
-860 train 4.158671 (lr=5.9126e-04) (hash(x)=20584272)
-870 train 4.041473 (lr=5.9002e-04) (hash(x)=17468790)
-880 train 3.975993 (lr=5.8869e-04) (hash(x)=22304409)
-890 train 4.075308 (lr=5.8729e-04) (hash(x)=18871567)
-900 val loss 4.0451
-900 val perplexity 57.1184
-900 train 4.022658 (lr=5.8581e-04) (hash(x)=20924720)
-910 train 3.935108 (lr=5.8425e-04) (hash(x)=21163464)
-920 train 3.861070 (lr=5.8262e-04) (hash(x)=20120862)
-930 train 4.050533 (lr=5.8090e-04) (hash(x)=19565198)
-940 train 3.963954 (lr=5.7911e-04) (hash(x)=18517540)
-950 train 3.925745 (lr=5.7723e-04) (hash(x)=18300759)
-960 train 3.961215 (lr=5.7529e-04) (hash(x)=21443001)
-970 train 3.910380 (lr=5.7326e-04) (hash(x)=15210971)
-980 train 4.085726 (lr=5.7116e-04) (hash(x)=20874579)
-990 train 3.993262 (lr=5.6899e-04) (hash(x)=20690802)
-1000 val loss 3.9449
-1000 val perplexity 51.6737
-1000 train 3.866000 (lr=5.6674e-04) (hash(x)=18635877)
-1010 train 3.973155 (lr=5.6442e-04) (hash(x)=21141462)
-1020 train 3.988773 (lr=5.6202e-04) (hash(x)=19483400)
-1030 train 3.893528 (lr=5.5956e-04) (hash(x)=24792825)
-1040 train 3.875763 (lr=5.5702e-04) (hash(x)=16743928)
-1050 train 3.915400 (lr=5.5441e-04) (hash(x)=22540052)
-1060 train 3.929815 (lr=5.5174e-04) (hash(x)=20546637)
-1070 train 3.835329 (lr=5.4899e-04) (hash(x)=13739753)
-1080 train 3.884538 (lr=5.4618e-04) (hash(x)=20071304)
-1090 train 3.840072 (lr=5.4330e-04) (hash(x)=19339656)
-1100 val loss 3.8676
-1100 val perplexity 47.8297
-1100 train 3.832666 (lr=5.4035e-04) (hash(x)=20039628)
-1110 train 3.890791 (lr=5.3734e-04) (hash(x)=19249691)
-1120 train 3.909864 (lr=5.3426e-04) (hash(x)=19048809)
-1130 train 3.836288 (lr=5.3113e-04) (hash(x)=19309000)
-1140 train 3.793789 (lr=5.2792e-04) (hash(x)=21787927)
-1150 train 3.704907 (lr=5.2466e-04) (hash(x)=19998536)
-1160 train 3.852681 (lr=5.2134e-04) (hash(x)=23724694)
-1170 train 3.878023 (lr=5.1796e-04) (hash(x)=17148204)
-1180 train 3.788407 (lr=5.1452e-04) (hash(x)=19511852)
-1190 train 3.765954 (lr=5.1102e-04) (hash(x)=19352862)
-1200 val loss 3.7932
-1200 val perplexity 44.3961
-1200 train 3.706720 (lr=5.0746e-04) (hash(x)=19530144)
-1210 train 3.881899 (lr=5.0386e-04) (hash(x)=19897392)
-1220 train 3.791504 (lr=5.0019e-04) (hash(x)=24402963)
-1230 train 3.707231 (lr=4.9648e-04) (hash(x)=18589754)
-1240 train 3.717966 (lr=4.9271e-04) (hash(x)=21951750)
-1250 train 3.792468 (lr=4.8889e-04) (hash(x)=18611379)
-1260 train 3.808415 (lr=4.8503e-04) (hash(x)=20338521)
-1270 train 3.728000 (lr=4.8111e-04) (hash(x)=18308405)
-1280 train 3.730972 (lr=4.7715e-04) (hash(x)=20920945)
-1290 train 3.692427 (lr=4.7315e-04) (hash(x)=20094152)
-1300 val loss 3.7383
-1300 val perplexity 42.0261
-1300 train 3.811040 (lr=4.6909e-04) (hash(x)=19003947)
-1310 train 3.774107 (lr=4.6500e-04) (hash(x)=19356356)
-1320 train 3.723505 (lr=4.6086e-04) (hash(x)=19390343)
-1330 train 3.659417 (lr=4.5669e-04) (hash(x)=17535982)
-1340 train 3.682254 (lr=4.5247e-04) (hash(x)=21959104)
-1350 train 3.782424 (lr=4.4822e-04) (hash(x)=21982099)
-1360 train 3.741298 (lr=4.4393e-04) (hash(x)=17176243)
-1370 train 3.684681 (lr=4.3960e-04) (hash(x)=21223627)
-1380 train 3.705378 (lr=4.3524e-04) (hash(x)=22462931)
-1390 train 3.728876 (lr=4.3085e-04) (hash(x)=19153464)
-1400 val loss 3.6897
-1400 val perplexity 40.0328
-1400 train 3.619301 (lr=4.2643e-04) (hash(x)=20146496)
-1410 train 3.676383 (lr=4.2197e-04) (hash(x)=21781327)
-1420 train 3.746097 (lr=4.1749e-04) (hash(x)=19234051)
-1430 train 3.793772 (lr=4.1298e-04) (hash(x)=20769898)
-1440 train 3.646108 (lr=4.0845e-04) (hash(x)=22042430)
-1450 train 3.668874 (lr=4.0389e-04) (hash(x)=20527387)
-1460 train 3.652778 (lr=3.9931e-04) (hash(x)=18925766)
-1470 train 3.650412 (lr=3.9470e-04) (hash(x)=18198851)
-1480 train 3.724410 (lr=3.9008e-04) (hash(x)=21847281)
-1490 train 3.692933 (lr=3.8544e-04) (hash(x)=15645274)
-1500 val loss 3.6399
-1500 val perplexity 38.0881
-1500 train 3.563986 (lr=3.8078e-04) (hash(x)=14693370)
-1510 train 3.577874 (lr=3.7610e-04) (hash(x)=21810857)
-1520 train 3.667368 (lr=3.7142e-04) (hash(x)=18980052)
-1530 train 3.698538 (lr=3.6671e-04) (hash(x)=22868842)
-1540 train 3.613892 (lr=3.6200e-04) (hash(x)=17289166)
-1550 train 3.693062 (lr=3.5728e-04) (hash(x)=18109373)
-1560 train 3.726691 (lr=3.5255e-04) (hash(x)=22377619)
-1570 train 3.554958 (lr=3.4781e-04) (hash(x)=20307182)
-1580 train 3.700760 (lr=3.4306e-04) (hash(x)=20391291)
-1590 train 3.569774 (lr=3.3831e-04) (hash(x)=18690939)
-1600 val loss 3.6051
-1600 val perplexity 36.7853
-1600 train 3.581419 (lr=3.3356e-04) (hash(x)=16719886)
-1610 train 3.573832 (lr=3.2881e-04) (hash(x)=16040840)
-1620 train 3.649099 (lr=3.2406e-04) (hash(x)=25795140)
-1630 train 3.653820 (lr=3.1931e-04) (hash(x)=21936964)
-1640 train 3.551316 (lr=3.1456e-04) (hash(x)=11593027)
-1650 train 3.608443 (lr=3.0982e-04) (hash(x)=23256718)
-1660 train 3.570829 (lr=3.0509e-04) (hash(x)=20289140)
-1670 train 3.573119 (lr=3.0036e-04) (hash(x)=18382509)
-1680 train 3.638904 (lr=2.9564e-04) (hash(x)=26604329)
-1690 train 3.607218 (lr=2.9093e-04) (hash(x)=22122534)
-1700 val loss 3.5739
-1700 val perplexity 35.6544
-1700 train 3.623316 (lr=2.8624e-04) (hash(x)=22884812)
-1710 train 3.567406 (lr=2.8156e-04) (hash(x)=18598496)
-1720 train 3.586696 (lr=2.7689e-04) (hash(x)=22147447)
-1730 train 3.591927 (lr=2.7224e-04) (hash(x)=18912135)
-1740 train 3.431834 (lr=2.6761e-04) (hash(x)=21952974)
-1750 train 3.543264 (lr=2.6299e-04) (hash(x)=18390978)
-1760 train 3.554070 (lr=2.5840e-04) (hash(x)=17979288)
-1770 train 3.561543 (lr=2.5383e-04) (hash(x)=18979099)
-1780 train 3.524708 (lr=2.4928e-04) (hash(x)=20511922)
-1790 train 3.578063 (lr=2.4476e-04) (hash(x)=21814060)
-1800 val loss 3.5471
-1800 val perplexity 34.7117
-1800 train 3.566007 (lr=2.4026e-04) (hash(x)=18076243)
-1810 train 3.556169 (lr=2.3580e-04) (hash(x)=19721209)
-1820 train 3.502210 (lr=2.3136e-04) (hash(x)=18140847)
-1830 train 3.564292 (lr=2.2695e-04) (hash(x)=22065413)
-1840 train 3.464599 (lr=2.2257e-04) (hash(x)=21166111)
-1850 train 3.465952 (lr=2.1823e-04) (hash(x)=21904302)
-1860 train 3.532963 (lr=2.1392e-04) (hash(x)=18283828)
-1870 train 3.487225 (lr=2.0965e-04) (hash(x)=20892773)
-1880 train 3.517797 (lr=2.0542e-04) (hash(x)=17781264)
-1890 train 3.485482 (lr=2.0122e-04) (hash(x)=20612962)
-1900 val loss 3.5233
-1900 val perplexity 33.8949
-1900 train 3.577407 (lr=1.9706e-04) (hash(x)=20421058)
-1910 train 3.551716 (lr=1.9295e-04) (hash(x)=18554808)
-1920 train 3.508257 (lr=1.8887e-04) (hash(x)=18189279)
-1930 train 3.524584 (lr=1.8485e-04) (hash(x)=19923279)
-1940 train 3.488195 (lr=1.8086e-04) (hash(x)=21665511)
-1950 train 3.513012 (lr=1.7692e-04) (hash(x)=22038489)
-1960 train 3.484293 (lr=1.7303e-04) (hash(x)=18896956)
-1970 train 3.512303 (lr=1.6919e-04) (hash(x)=20408383)
-1980 train 3.495337 (lr=1.6540e-04) (hash(x)=23039940)
-1990 train 3.628411 (lr=1.6166e-04) (hash(x)=20631172)
-2000 val loss 3.5039
-2000 val perplexity 33.2435
-2000 train 3.442245 (lr=1.5797e-04) (hash(x)=20206106)
-2010 train 3.536399 (lr=1.5433e-04) (hash(x)=21864020)
-2020 train 3.519025 (lr=1.5075e-04) (hash(x)=21391128)
-2030 train 3.486156 (lr=1.4723e-04) (hash(x)=18676636)
-2040 train 3.532384 (lr=1.4376e-04) (hash(x)=15987134)
-2050 train 3.525572 (lr=1.4035e-04) (hash(x)=23940769)
-2060 train 3.502497 (lr=1.3699e-04) (hash(x)=17152840)
-2070 train 3.471947 (lr=1.3370e-04) (hash(x)=22958445)
-2080 train 3.523313 (lr=1.3047e-04) (hash(x)=23616724)
-2090 train 3.494230 (lr=1.2730e-04) (hash(x)=16571283)
+0 val loss 10.9488
+0 val perplexity 56887.5898
+0 train 10.951686 (lr=8.3916e-07) (hash(x)=17183924)
+10 train 10.135524 (lr=9.2308e-06) (hash(x)=19047484)
+20 train 9.673862 (lr=1.7622e-05) (hash(x)=20001614)
+30 train 9.445292 (lr=2.6014e-05) (hash(x)=21278543)
+40 train 9.199447 (lr=3.4406e-05) (hash(x)=29208111)
+50 train 8.886049 (lr=4.2797e-05) (hash(x)=20801292)
+60 train 8.546229 (lr=5.1189e-05) (hash(x)=27442073)
+70 train 8.178991 (lr=5.9580e-05) (hash(x)=19918406)
+80 train 7.815042 (lr=6.7972e-05) (hash(x)=20569806)
+90 train 7.704312 (lr=7.6364e-05) (hash(x)=26013441)
+100 val loss 7.4804
+100 val perplexity 1772.9517
+100 train 7.454827 (lr=8.4755e-05) (hash(x)=29039793)
+110 train 7.284087 (lr=9.3147e-05) (hash(x)=16550842)
+120 train 7.125364 (lr=1.0154e-04) (hash(x)=17250570)
+130 train 7.142025 (lr=1.0993e-04) (hash(x)=17154654)
+140 train 7.005312 (lr=1.1832e-04) (hash(x)=22437415)
+150 train 6.937147 (lr=1.2671e-04) (hash(x)=18645582)
+160 train 6.682060 (lr=1.3510e-04) (hash(x)=19460261)
+170 train 6.518916 (lr=1.4350e-04) (hash(x)=20953331)
+180 train 6.617326 (lr=1.5189e-04) (hash(x)=20883978)
+190 train 6.475650 (lr=1.6028e-04) (hash(x)=25571428)
+200 val loss 6.4173
+200 val perplexity 612.3538
+200 train 6.341202 (lr=1.6867e-04) (hash(x)=16049634)
+210 train 6.323199 (lr=1.7706e-04) (hash(x)=15886413)
+220 train 6.372123 (lr=1.8545e-04) (hash(x)=21731617)
+230 train 6.157201 (lr=1.9385e-04) (hash(x)=22509449)
+240 train 6.092800 (lr=2.0224e-04) (hash(x)=17396544)
+250 train 6.146547 (lr=2.1063e-04) (hash(x)=17748336)
+260 train 6.021659 (lr=2.1902e-04) (hash(x)=21263532)
+270 train 5.948525 (lr=2.2741e-04) (hash(x)=19623735)
+280 train 5.953936 (lr=2.3580e-04) (hash(x)=18337637)
+290 train 5.895530 (lr=2.4420e-04) (hash(x)=20005903)
+300 val loss 5.8078
+300 val perplexity 332.8762
+300 train 5.825626 (lr=2.5259e-04) (hash(x)=20938811)
+310 train 5.713806 (lr=2.6098e-04) (hash(x)=20688368)
+320 train 5.733479 (lr=2.6937e-04) (hash(x)=19776598)
+330 train 5.649756 (lr=2.7776e-04) (hash(x)=19721837)
+340 train 5.585549 (lr=2.8615e-04) (hash(x)=16578291)
+350 train 5.494850 (lr=2.9455e-04) (hash(x)=19998371)
+360 train 5.607419 (lr=3.0294e-04) (hash(x)=21565819)
+370 train 5.499692 (lr=3.1133e-04) (hash(x)=19484104)
+380 train 5.373439 (lr=3.1972e-04) (hash(x)=20659288)
+390 train 5.311124 (lr=3.2811e-04) (hash(x)=22061179)
+400 val loss 5.3585
+400 val perplexity 212.4125
+400 train 5.427289 (lr=3.3650e-04) (hash(x)=18557926)
+410 train 5.299348 (lr=3.4490e-04) (hash(x)=24034052)
+420 train 5.164496 (lr=3.5329e-04) (hash(x)=19768493)
+430 train 5.270503 (lr=3.6168e-04) (hash(x)=21678884)
+440 train 5.195265 (lr=3.7007e-04) (hash(x)=19589702)
+450 train 5.209245 (lr=3.7846e-04) (hash(x)=19902911)
+460 train 5.006237 (lr=3.8685e-04) (hash(x)=18881029)
+470 train 5.112347 (lr=3.9524e-04) (hash(x)=24329185)
+480 train 5.001166 (lr=4.0364e-04) (hash(x)=18709451)
+490 train 4.981668 (lr=4.1203e-04) (hash(x)=18834705)
+500 val loss 4.9851
+500 val perplexity 146.2175
+500 train 5.022027 (lr=4.2042e-04) (hash(x)=24221098)
+510 train 4.966081 (lr=4.2881e-04) (hash(x)=20010758)
+520 train 4.950898 (lr=4.3720e-04) (hash(x)=18547084)
+530 train 4.755854 (lr=4.4559e-04) (hash(x)=18282458)
+540 train 4.898161 (lr=4.5399e-04) (hash(x)=20671087)
+550 train 4.718338 (lr=4.6238e-04) (hash(x)=18634421)
+560 train 4.665814 (lr=4.7077e-04) (hash(x)=17067920)
+570 train 4.803200 (lr=4.7916e-04) (hash(x)=22276881)
+580 train 4.702220 (lr=4.8755e-04) (hash(x)=22973651)
+590 train 4.670744 (lr=4.9594e-04) (hash(x)=18096900)
+600 val loss 4.5845
+600 val perplexity 97.9530
+600 train 4.527422 (lr=5.0434e-04) (hash(x)=20617414)
+610 train 4.382463 (lr=5.1273e-04) (hash(x)=20722525)
+620 train 4.606997 (lr=5.2112e-04) (hash(x)=20785452)
+630 train 4.548081 (lr=5.2951e-04) (hash(x)=16647858)
+640 train 4.474239 (lr=5.3790e-04) (hash(x)=20798993)
+650 train 4.408963 (lr=5.4629e-04) (hash(x)=20912393)
+660 train 4.225210 (lr=5.5469e-04) (hash(x)=21505604)
+670 train 4.492852 (lr=5.6308e-04) (hash(x)=20263120)
+680 train 4.389547 (lr=5.7147e-04) (hash(x)=20779495)
+690 train 4.297709 (lr=5.7986e-04) (hash(x)=16555678)
+700 val loss 4.3320
+700 val perplexity 76.0999
+700 train 4.155384 (lr=5.8825e-04) (hash(x)=20156261)
+710 train 4.126278 (lr=5.9664e-04) (hash(x)=17064702)
+720 train 4.341045 (lr=5.9999e-04) (hash(x)=19321113)
+730 train 4.295744 (lr=5.9991e-04) (hash(x)=19738613)
+740 train 4.215635 (lr=5.9974e-04) (hash(x)=18585508)
+750 train 4.109127 (lr=5.9949e-04) (hash(x)=18605696)
+760 train 4.259497 (lr=5.9915e-04) (hash(x)=22711193)
+770 train 4.218299 (lr=5.9874e-04) (hash(x)=17163545)
+780 train 4.106490 (lr=5.9824e-04) (hash(x)=22239165)
+790 train 4.215418 (lr=5.9765e-04) (hash(x)=19202751)
+800 val loss 4.1635
+800 val perplexity 64.2947
+800 train 4.222058 (lr=5.9698e-04) (hash(x)=22502753)
+810 train 4.190634 (lr=5.9623e-04) (hash(x)=20197889)
+820 train 4.121151 (lr=5.9540e-04) (hash(x)=19376072)
+830 train 3.969591 (lr=5.9449e-04) (hash(x)=18293711)
+840 train 4.132769 (lr=5.9349e-04) (hash(x)=19061676)
+850 train 4.141328 (lr=5.9241e-04) (hash(x)=20557626)
+860 train 4.064749 (lr=5.9126e-04) (hash(x)=18642266)
+870 train 3.946215 (lr=5.9002e-04) (hash(x)=20957453)
+880 train 3.982539 (lr=5.8869e-04) (hash(x)=21781033)
+890 train 4.088262 (lr=5.8729e-04) (hash(x)=21094646)
+900 val loss 4.0193
+900 val perplexity 55.6642
+900 train 4.075872 (lr=5.8581e-04) (hash(x)=21499150)
+910 train 4.048689 (lr=5.8425e-04) (hash(x)=19423438)
+920 train 3.857939 (lr=5.8262e-04) (hash(x)=20543615)
+930 train 4.030903 (lr=5.8090e-04) (hash(x)=19245236)
+940 train 4.040757 (lr=5.7911e-04) (hash(x)=24035334)
+950 train 3.977630 (lr=5.7723e-04) (hash(x)=16703311)
+960 train 3.958583 (lr=5.7529e-04) (hash(x)=23101805)
+970 train 3.912422 (lr=5.7326e-04) (hash(x)=20320872)
+980 train 3.960725 (lr=5.7116e-04) (hash(x)=20990946)
+990 train 3.963059 (lr=5.6899e-04) (hash(x)=18628956)
+1000 val loss 3.9232
+1000 val perplexity 50.5619
+1000 train 3.900304 (lr=5.6674e-04) (hash(x)=22968748)
+1010 train 3.885020 (lr=5.6442e-04) (hash(x)=20541414)
+1020 train 3.939539 (lr=5.6202e-04) (hash(x)=19274167)
+1030 train 3.892586 (lr=5.5956e-04) (hash(x)=22645317)
+1040 train 3.873803 (lr=5.5702e-04) (hash(x)=17884034)
+1050 train 3.875090 (lr=5.5441e-04) (hash(x)=19299722)
+1060 train 3.879480 (lr=5.5174e-04) (hash(x)=18409931)
+1070 train 3.862430 (lr=5.4899e-04) (hash(x)=22455477)
+1080 train 3.831441 (lr=5.4618e-04) (hash(x)=20085073)
+1090 train 3.863496 (lr=5.4330e-04) (hash(x)=20371024)
+1100 val loss 3.8371
+1100 val perplexity 46.3896
+1100 train 3.813580 (lr=5.4035e-04) (hash(x)=17534387)
+1110 train 3.813799 (lr=5.3734e-04) (hash(x)=18107517)
+1120 train 3.846587 (lr=5.3426e-04) (hash(x)=21862096)
+1130 train 3.848328 (lr=5.3113e-04) (hash(x)=24296067)
+1140 train 3.752687 (lr=5.2792e-04) (hash(x)=25489932)
+1150 train 3.855810 (lr=5.2466e-04) (hash(x)=19256895)
+1160 train 3.764308 (lr=5.2134e-04) (hash(x)=20915643)
+1170 train 3.841542 (lr=5.1796e-04) (hash(x)=22053731)
+1180 train 3.789248 (lr=5.1452e-04) (hash(x)=20012827)
+1190 train 3.786664 (lr=5.1102e-04) (hash(x)=21694459)
+1200 val loss 3.7737
+1200 val perplexity 43.5401
+1200 train 3.812212 (lr=5.0746e-04) (hash(x)=21755992)
+1210 train 3.810848 (lr=5.0386e-04) (hash(x)=21880996)
+1220 train 3.786426 (lr=5.0019e-04) (hash(x)=20520945)
+1230 train 3.753773 (lr=4.9648e-04) (hash(x)=19770784)
+1240 train 3.759912 (lr=4.9271e-04) (hash(x)=18571983)
+1250 train 3.768283 (lr=4.8889e-04) (hash(x)=20735694)
+1260 train 3.719514 (lr=4.8503e-04) (hash(x)=21156163)
+1270 train 3.778144 (lr=4.8111e-04) (hash(x)=21271664)
+1280 train 3.745692 (lr=4.7715e-04) (hash(x)=19315272)
+1290 train 3.715318 (lr=4.7315e-04) (hash(x)=21142191)
+1300 val loss 3.7219
+1300 val perplexity 41.3416
+1300 train 3.753943 (lr=4.6909e-04) (hash(x)=17145293)
+1310 train 3.937163 (lr=4.6500e-04) (hash(x)=17713742)
+1320 train 3.733192 (lr=4.6086e-04) (hash(x)=33348747)
+1330 train 3.692661 (lr=4.5669e-04) (hash(x)=20702111)
+1340 train 3.764495 (lr=4.5247e-04) (hash(x)=17898284)
+1350 train 3.678736 (lr=4.4822e-04) (hash(x)=17904915)
+1360 train 3.688644 (lr=4.4393e-04) (hash(x)=23325893)
+1370 train 3.617370 (lr=4.3960e-04) (hash(x)=20717754)
+1380 train 3.718314 (lr=4.3524e-04) (hash(x)=16859207)
+1390 train 3.682452 (lr=4.3085e-04) (hash(x)=21029500)
+1400 val loss 3.6796
+1400 val perplexity 39.6308
+1400 train 3.622112 (lr=4.2643e-04) (hash(x)=17789655)
+1410 train 3.738389 (lr=4.2197e-04) (hash(x)=19965237)
+1420 train 3.711900 (lr=4.1749e-04) (hash(x)=19262560)
+1430 train 3.613701 (lr=4.1298e-04) (hash(x)=18593169)
+1440 train 3.620387 (lr=4.0845e-04) (hash(x)=20607672)
+1450 train 3.673327 (lr=4.0389e-04) (hash(x)=20682942)
+1460 train 3.624961 (lr=3.9931e-04) (hash(x)=20220347)
+1470 train 3.605336 (lr=3.9470e-04) (hash(x)=15833389)
+1480 train 3.733973 (lr=3.9008e-04) (hash(x)=19495027)
+1490 train 3.639123 (lr=3.8544e-04) (hash(x)=22138106)
+1500 val loss 3.6380
+1500 val perplexity 38.0140
+1500 train 3.597368 (lr=3.8078e-04) (hash(x)=24123427)
+1510 train 3.562565 (lr=3.7610e-04) (hash(x)=19024825)
+1520 train 3.700214 (lr=3.7142e-04) (hash(x)=19780005)
+1530 train 3.689910 (lr=3.6671e-04) (hash(x)=19134410)
+1540 train 3.610766 (lr=3.6200e-04) (hash(x)=19855253)
+1550 train 3.574905 (lr=3.5728e-04) (hash(x)=20259196)
+1560 train 3.576008 (lr=3.5255e-04) (hash(x)=18870222)
+1570 train 3.609393 (lr=3.4781e-04) (hash(x)=16067782)
+1580 train 3.603217 (lr=3.4306e-04) (hash(x)=21604205)
+1590 train 3.577602 (lr=3.3831e-04) (hash(x)=19076298)
+1600 val loss 3.6016
+1600 val perplexity 36.6555
+1600 train 3.598683 (lr=3.3356e-04) (hash(x)=19064055)
+1610 train 3.654874 (lr=3.2881e-04) (hash(x)=19196041)
+1620 train 3.605009 (lr=3.2406e-04) (hash(x)=25625303)
+1630 train 3.558200 (lr=3.1931e-04) (hash(x)=20618762)
+1640 train 3.597315 (lr=3.1456e-04) (hash(x)=21102615)
+1650 train 3.545415 (lr=3.0982e-04) (hash(x)=21088682)
+1660 train 3.532047 (lr=3.0509e-04) (hash(x)=15297157)
+1670 train 3.572251 (lr=3.0036e-04) (hash(x)=19690570)
+1680 train 3.591368 (lr=2.9564e-04) (hash(x)=21445411)
+1690 train 3.572603 (lr=2.9093e-04) (hash(x)=21683876)
+1700 val loss 3.5670
+1700 val perplexity 35.4106
+1700 train 3.582284 (lr=2.8624e-04) (hash(x)=17449466)
+1710 train 3.586621 (lr=2.8156e-04) (hash(x)=16754900)
+1720 train 3.553272 (lr=2.7689e-04) (hash(x)=20736439)
+1730 train 3.587787 (lr=2.7224e-04) (hash(x)=21475325)
+1740 train 3.529853 (lr=2.6761e-04) (hash(x)=31466218)
+1750 train 3.612159 (lr=2.6299e-04) (hash(x)=17740223)
+1760 train 3.549626 (lr=2.5840e-04) (hash(x)=17626006)
+1770 train 3.402314 (lr=2.5383e-04) (hash(x)=22276216)
+1780 train 3.513784 (lr=2.4928e-04) (hash(x)=14277004)
+1790 train 3.544051 (lr=2.4476e-04) (hash(x)=20600488)
+1800 val loss 3.5419
+1800 val perplexity 34.5320
+1800 train 3.554114 (lr=2.4026e-04) (hash(x)=17508978)
+1810 train 3.469589 (lr=2.3580e-04) (hash(x)=22374263)
+1820 train 3.606226 (lr=2.3136e-04) (hash(x)=18959095)
+1830 train 3.517645 (lr=2.2695e-04) (hash(x)=19485836)
+1840 train 3.511595 (lr=2.2257e-04) (hash(x)=24276692)
+1850 train 3.540552 (lr=2.1823e-04) (hash(x)=20554224)
+1860 train 3.528264 (lr=2.1392e-04) (hash(x)=18068030)
+1870 train 3.551932 (lr=2.0965e-04) (hash(x)=18349056)
+1880 train 3.486073 (lr=2.0542e-04) (hash(x)=20640059)
+1890 train 3.491437 (lr=2.0122e-04) (hash(x)=19667347)
+1900 val loss 3.5176
+1900 val perplexity 33.7042
+1900 train 3.550257 (lr=1.9706e-04) (hash(x)=19559109)
+1910 train 3.572239 (lr=1.9295e-04) (hash(x)=22490711)
+1920 train 3.507478 (lr=1.8887e-04) (hash(x)=18268544)
+1930 train 3.487029 (lr=1.8485e-04) (hash(x)=20585037)
+1940 train 3.558938 (lr=1.8086e-04) (hash(x)=30293578)
+1950 train 3.495139 (lr=1.7692e-04) (hash(x)=21836208)
+1960 train 3.432023 (lr=1.7303e-04) (hash(x)=23417084)
+1970 train 3.523047 (lr=1.6919e-04) (hash(x)=19390086)
+1980 train 3.451552 (lr=1.6540e-04) (hash(x)=20129488)
+1990 train 3.495123 (lr=1.6166e-04) (hash(x)=20177413)
+2000 val loss 3.4995
+2000 val perplexity 33.0993
+2000 train 3.423361 (lr=1.5797e-04) (hash(x)=15921967)
+2010 train 3.493405 (lr=1.5433e-04) (hash(x)=20819557)
+2020 train 3.516616 (lr=1.5075e-04) (hash(x)=18729616)
+2030 train 3.453966 (lr=1.4723e-04) (hash(x)=15730659)
+2040 train 3.492836 (lr=1.4376e-04) (hash(x)=18679863)
+2050 train 3.545012 (lr=1.4035e-04) (hash(x)=17110863)
+2060 train 3.454457 (lr=1.3699e-04) (hash(x)=18291952)
+2070 train 3.482400 (lr=1.3370e-04) (hash(x)=18249172)
+2080 train 3.508815 (lr=1.3047e-04) (hash(x)=22686659)
+2090 train 3.466354 (lr=1.2730e-04) (hash(x)=18354325)
 2100 val loss 3.4820
-2100 val perplexity 32.5238
-2100 train 3.491362 (lr=1.2419e-04) (hash(x)=19070327)
-2110 train 3.535779 (lr=1.2115e-04) (hash(x)=20758626)
-2120 train 3.479458 (lr=1.1817e-04) (hash(x)=18397810)
-2130 train 3.471737 (lr=1.1525e-04) (hash(x)=18045234)
-2140 train 3.574210 (lr=1.1241e-04) (hash(x)=18969565)
-2150 train 3.509843 (lr=1.0963e-04) (hash(x)=17149701)
-2160 train 3.429842 (lr=1.0692e-04) (hash(x)=23075443)
-2170 train 3.465482 (lr=1.0427e-04) (hash(x)=17977337)
-2180 train 3.525461 (lr=1.0170e-04) (hash(x)=20657407)
-2190 train 3.511199 (lr=9.9199e-05) (hash(x)=19985933)
-2200 val loss 3.4672
-2200 val perplexity 32.0459
-2200 train 3.464933 (lr=9.6769e-05) (hash(x)=19025387)
-2210 train 3.482505 (lr=9.4412e-05) (hash(x)=23318724)
-2220 train 3.553313 (lr=9.2127e-05) (hash(x)=19308039)
-2230 train 3.397718 (lr=8.9916e-05) (hash(x)=20733682)
-2240 train 3.494338 (lr=8.7779e-05) (hash(x)=18519560)
-2250 train 3.498209 (lr=8.5717e-05) (hash(x)=20299444)
-2260 train 3.455821 (lr=8.3731e-05) (hash(x)=18819615)
-2270 train 3.490780 (lr=8.1821e-05) (hash(x)=18743705)
-2280 train 3.474558 (lr=7.9988e-05) (hash(x)=17190793)
-2290 train 3.510101 (lr=7.8232e-05) (hash(x)=23126906)
-2300 val loss 3.4553
-2300 val perplexity 31.6677
-2300 train 3.444954 (lr=7.6555e-05) (hash(x)=24812267)
-2310 train 3.427062 (lr=7.4956e-05) (hash(x)=20147105)
-2320 train 3.543566 (lr=7.3436e-05) (hash(x)=24619936)
-2330 train 3.408771 (lr=7.1995e-05) (hash(x)=17730658)
-2340 train 3.384672 (lr=7.0635e-05) (hash(x)=19345731)
-2350 train 3.424534 (lr=6.9354e-05) (hash(x)=26415058)
-2360 train 3.455724 (lr=6.8155e-05) (hash(x)=18622999)
-2370 train 3.392233 (lr=6.7036e-05) (hash(x)=18848567)
-2380 train 3.472850 (lr=6.5999e-05) (hash(x)=21903360)
-2390 train 3.487927 (lr=6.5044e-05) (hash(x)=23479216)
-2400 val loss 3.4462
-2400 val perplexity 31.3810
-2400 train 3.483463 (lr=6.4171e-05) (hash(x)=19762013)
-2410 train 3.389079 (lr=6.3380e-05) (hash(x)=20622912)
-2420 train 3.504714 (lr=6.2672e-05) (hash(x)=17669226)
-2430 train 3.433876 (lr=6.2046e-05) (hash(x)=19536719)
-2440 train 3.376764 (lr=6.1504e-05) (hash(x)=19913725)
-2450 train 3.419426 (lr=6.1045e-05) (hash(x)=20782072)
-2460 train 3.390014 (lr=6.0669e-05) (hash(x)=19795002)
-2470 train 3.435155 (lr=6.0376e-05) (hash(x)=17003051)
-2480 train 3.404618 (lr=6.0167e-05) (hash(x)=21414190)
-2490 train 3.448616 (lr=6.0042e-05) (hash(x)=20943620)
-2499 val loss 3.4412
-2499 val perplexity 31.2248
+2100 val perplexity 32.5245
+2100 train 3.457326 (lr=1.2419e-04) (hash(x)=18204536)
+2110 train 3.521772 (lr=1.2115e-04) (hash(x)=22394614)
+2120 train 3.408096 (lr=1.1817e-04) (hash(x)=18121192)
+2130 train 3.453637 (lr=1.1525e-04) (hash(x)=24393898)
+2140 train 3.465610 (lr=1.1241e-04) (hash(x)=27349908)
+2150 train 3.482891 (lr=1.0963e-04) (hash(x)=18262196)
+2160 train 3.419465 (lr=1.0692e-04) (hash(x)=20081235)
+2170 train 3.449390 (lr=1.0427e-04) (hash(x)=16984900)
+2180 train 3.519099 (lr=1.0170e-04) (hash(x)=24294588)
+2190 train 3.479918 (lr=9.9199e-05) (hash(x)=17042450)
+2200 val loss 3.4680
+2200 val perplexity 32.0718
+2200 train 3.367420 (lr=9.6769e-05) (hash(x)=23876593)
+2210 train 3.512331 (lr=9.4412e-05) (hash(x)=19173831)
+2220 train 3.416461 (lr=9.2127e-05) (hash(x)=21211298)
+2230 train 3.420935 (lr=8.9916e-05) (hash(x)=18721700)
+2240 train 3.419350 (lr=8.7779e-05) (hash(x)=19825319)
+2250 train 3.443780 (lr=8.5717e-05) (hash(x)=18378838)
+2260 train 3.446432 (lr=8.3731e-05) (hash(x)=13199654)
+2270 train 3.363626 (lr=8.1821e-05) (hash(x)=18597807)
+2280 train 3.468322 (lr=7.9988e-05) (hash(x)=22012917)
+2290 train 3.387262 (lr=7.8232e-05) (hash(x)=22305542)
+2300 val loss 3.4534
+2300 val perplexity 31.6067
+2300 train 3.484706 (lr=7.6555e-05) (hash(x)=19782726)
+2310 train 3.450092 (lr=7.4956e-05) (hash(x)=22042291)
+2320 train 3.455256 (lr=7.3436e-05) (hash(x)=17370832)
+2330 train 3.489407 (lr=7.1995e-05) (hash(x)=19612132)
+2340 train 3.457868 (lr=7.0635e-05) (hash(x)=18803286)
+2350 train 3.441510 (lr=6.9354e-05) (hash(x)=20196070)
+2360 train 3.408381 (lr=6.8155e-05) (hash(x)=19997454)
+2370 train 3.416731 (lr=6.7036e-05) (hash(x)=16704483)
+2380 train 3.445230 (lr=6.5999e-05) (hash(x)=19131833)
+2390 train 3.390046 (lr=6.5044e-05) (hash(x)=20507035)
+2400 val loss 3.4443
+2400 val perplexity 31.3220
+2400 train 3.499394 (lr=6.4171e-05) (hash(x)=20254090)
+2410 train 3.435609 (lr=6.3380e-05) (hash(x)=17926964)
+2420 train 3.467933 (lr=6.2672e-05) (hash(x)=20607860)
+2430 train 3.399081 (lr=6.2046e-05) (hash(x)=11288168)
+2440 train 3.517715 (lr=6.1504e-05) (hash(x)=21486139)
+2450 train 3.457528 (lr=6.1045e-05) (hash(x)=18994546)
+2460 train 3.392504 (lr=6.0669e-05) (hash(x)=21218220)
+2470 train 3.491202 (lr=6.0376e-05) (hash(x)=20504341)
+2480 train 3.445621 (lr=6.0167e-05) (hash(x)=17937386)
+2490 train 3.357094 (lr=6.0042e-05) (hash(x)=21870434)
+2499 val loss 3.4380
+2499 val perplexity 31.1246

two_masks_shared_1_head/model_02499.pt

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:0ad668c93057ed378077caeab4d124428a33738419a3d62b74a5c7a4b40152e6
+oid sha256:7bc16c44d15fcd3d1ac25457f4c99d76c000e5a58bfb5b1c2f844bd0dac9b7a1
 size 555239810

two_masks_shared_1_head/optimizer_02499.pt

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:9fd1a03b60591e489e20e75a99eecd10f6e66dd0bcb419ad4cd1eebdebdddebf
+oid sha256:e7a7cc6e18a223907b419bbd51073c538c92d7ba3779c33fb748e505fb74a53a
 size 1009827078