Upload folder using huggingface_hub

one_mask_per_head_2_latent_vectors/args.json

@@ -1 +1 @@
-{"hellaswag": true, "attention_kind": "selective", "log_dir": "allowing_more_selection_patterns/one_mask_per_head_2_latent_vectors", "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_latent_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": null, "n_latent_masks": 2, "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/one_mask_per_head_2_latent_vectors", "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_latent_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": null, "n_latent_masks": 2, "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}

one_mask_per_head_2_latent_vectors/log2.txt

@@ -1,303 +1,303 @@
 max_steps: 2500
-0 val loss 11.0209
-0 val perplexity 61140.6484
-0 train 11.026832 (lr=8.3916e-07) (hash(x)=23347629)
-10 train 10.224495 (lr=9.2308e-06) (hash(x)=19636655)
-20 train 9.729277 (lr=1.7622e-05) (hash(x)=24490248)
-30 train 9.527868 (lr=2.6014e-05) (hash(x)=21733558)
-40 train 9.272552 (lr=3.4406e-05) (hash(x)=31750616)
-50 train 8.960818 (lr=4.2797e-05) (hash(x)=20041895)
-60 train 8.586679 (lr=5.1189e-05) (hash(x)=21416280)
-70 train 8.241448 (lr=5.9580e-05) (hash(x)=18066780)
-80 train 7.926493 (lr=6.7972e-05) (hash(x)=21809427)
-90 train 7.674655 (lr=7.6364e-05) (hash(x)=22061595)
-100 val loss 7.4587
-100 val perplexity 1734.9619
-100 train 7.408588 (lr=8.4755e-05) (hash(x)=17468914)
-110 train 7.227366 (lr=9.3147e-05) (hash(x)=20421520)
-120 train 7.229856 (lr=1.0154e-04) (hash(x)=23140133)
-130 train 7.097166 (lr=1.0993e-04) (hash(x)=20006835)
-140 train 6.848365 (lr=1.1832e-04) (hash(x)=18628134)
-150 train 6.821269 (lr=1.2671e-04) (hash(x)=21913447)
-160 train 6.769176 (lr=1.3510e-04) (hash(x)=22997297)
-170 train 6.578884 (lr=1.4350e-04) (hash(x)=17904821)
-180 train 6.428210 (lr=1.5189e-04) (hash(x)=18826982)
-190 train 6.516953 (lr=1.6028e-04) (hash(x)=21387716)
-200 val loss 6.3869
-200 val perplexity 594.0179
-200 train 6.309577 (lr=1.6867e-04) (hash(x)=16236395)
-210 train 6.189707 (lr=1.7706e-04) (hash(x)=21428681)
-220 train 6.336513 (lr=1.8545e-04) (hash(x)=20461738)
-230 train 6.159262 (lr=1.9385e-04) (hash(x)=20394865)
-240 train 6.045243 (lr=2.0224e-04) (hash(x)=18028671)
-250 train 5.897091 (lr=2.1063e-04) (hash(x)=18122073)
-260 train 5.813160 (lr=2.1902e-04) (hash(x)=22994143)
-270 train 6.013514 (lr=2.2741e-04) (hash(x)=20646260)
-280 train 5.860374 (lr=2.3580e-04) (hash(x)=23204708)
-290 train 5.776207 (lr=2.4420e-04) (hash(x)=17900252)
-300 val loss 5.8071
-300 val perplexity 332.6622
-300 train 5.650475 (lr=2.5259e-04) (hash(x)=19073819)
-310 train 5.812761 (lr=2.6098e-04) (hash(x)=22378232)
-320 train 5.730812 (lr=2.6937e-04) (hash(x)=19910430)
-330 train 5.659667 (lr=2.7776e-04) (hash(x)=25168753)
-340 train 5.535535 (lr=2.8615e-04) (hash(x)=20665256)
-350 train 5.416691 (lr=2.9455e-04) (hash(x)=19951160)
-360 train 5.629634 (lr=3.0294e-04) (hash(x)=21178202)
-370 train 5.483434 (lr=3.1133e-04) (hash(x)=20435886)
-380 train 5.417364 (lr=3.1972e-04) (hash(x)=21255097)
-390 train 5.422375 (lr=3.2811e-04) (hash(x)=19596330)
-400 val loss 5.3748
-400 val perplexity 215.9036
-400 train 5.398746 (lr=3.3650e-04) (hash(x)=21789426)
-410 train 5.306820 (lr=3.4490e-04) (hash(x)=18366401)
-420 train 5.186624 (lr=3.5329e-04) (hash(x)=18614660)
-430 train 5.335863 (lr=3.6168e-04) (hash(x)=21649063)
-440 train 5.209197 (lr=3.7007e-04) (hash(x)=18505097)
-450 train 5.206759 (lr=3.7846e-04) (hash(x)=23434312)
-460 train 5.083421 (lr=3.8685e-04) (hash(x)=16584569)
-470 train 4.978126 (lr=3.9524e-04) (hash(x)=13447111)
-480 train 5.169206 (lr=4.0364e-04) (hash(x)=22645130)
-490 train 5.041542 (lr=4.1203e-04) (hash(x)=19362777)
-500 val loss 5.0077
-500 val perplexity 149.5551
-500 train 4.943790 (lr=4.2042e-04) (hash(x)=20921874)
-510 train 4.837734 (lr=4.2881e-04) (hash(x)=22123048)
-520 train 4.861088 (lr=4.3720e-04) (hash(x)=21215368)
-530 train 5.019162 (lr=4.4559e-04) (hash(x)=20717936)
-540 train 4.858258 (lr=4.5399e-04) (hash(x)=20118007)
-550 train 4.750133 (lr=4.6238e-04) (hash(x)=21402522)
-560 train 4.671681 (lr=4.7077e-04) (hash(x)=19057180)
-570 train 4.851924 (lr=4.7916e-04) (hash(x)=19360553)
-580 train 4.781807 (lr=4.8755e-04) (hash(x)=18194693)
-590 train 4.677301 (lr=4.9594e-04) (hash(x)=19863826)
-600 val loss 4.6329
-600 val perplexity 102.8118
-600 train 4.585471 (lr=5.0434e-04) (hash(x)=21794551)
-610 train 4.578421 (lr=5.1273e-04) (hash(x)=18685622)
-620 train 4.461172 (lr=5.2112e-04) (hash(x)=23473104)
-630 train 4.555367 (lr=5.2951e-04) (hash(x)=18671672)
-640 train 4.559628 (lr=5.3790e-04) (hash(x)=19234240)
-650 train 4.476555 (lr=5.4629e-04) (hash(x)=23183903)
-660 train 4.321607 (lr=5.5469e-04) (hash(x)=23665543)
-670 train 4.484042 (lr=5.6308e-04) (hash(x)=22055313)
-680 train 4.452967 (lr=5.7147e-04) (hash(x)=22321668)
-690 train 4.545545 (lr=5.7986e-04) (hash(x)=21568926)
-700 val loss 4.4017
-700 val perplexity 81.5878
-700 train 4.404649 (lr=5.8825e-04) (hash(x)=19783143)
-710 train 4.330183 (lr=5.9664e-04) (hash(x)=21565258)
-720 train 4.418242 (lr=5.9999e-04) (hash(x)=16127212)
-730 train 4.336320 (lr=5.9991e-04) (hash(x)=19658174)
-740 train 4.312549 (lr=5.9974e-04) (hash(x)=16454319)
-750 train 4.244321 (lr=5.9949e-04) (hash(x)=17993154)
-760 train 4.170447 (lr=5.9915e-04) (hash(x)=19256989)
-770 train 4.140961 (lr=5.9874e-04) (hash(x)=17351653)
-780 train 4.162821 (lr=5.9824e-04) (hash(x)=21626120)
-790 train 4.300798 (lr=5.9765e-04) (hash(x)=16367412)
-800 val loss 4.2126
-800 val perplexity 67.5346
-800 train 4.165354 (lr=5.9698e-04) (hash(x)=22321504)
-810 train 4.051053 (lr=5.9623e-04) (hash(x)=19869755)
-820 train 4.218919 (lr=5.9540e-04) (hash(x)=24183665)
-830 train 4.191387 (lr=5.9449e-04) (hash(x)=19194113)
-840 train 4.038720 (lr=5.9349e-04) (hash(x)=17710503)
-850 train 4.055954 (lr=5.9241e-04) (hash(x)=17231851)
-860 train 4.191983 (lr=5.9126e-04) (hash(x)=20584272)
-870 train 4.081844 (lr=5.9002e-04) (hash(x)=17468790)
-880 train 4.022425 (lr=5.8869e-04) (hash(x)=22304409)
-890 train 4.110187 (lr=5.8729e-04) (hash(x)=18871567)
-900 val loss 4.0849
-900 val perplexity 59.4347
-900 train 4.060470 (lr=5.8581e-04) (hash(x)=20924720)
-910 train 3.968852 (lr=5.8425e-04) (hash(x)=21163464)
-920 train 3.905523 (lr=5.8262e-04) (hash(x)=20120862)
-930 train 4.075518 (lr=5.8090e-04) (hash(x)=19565198)
-940 train 3.999166 (lr=5.7911e-04) (hash(x)=18517540)
-950 train 3.965435 (lr=5.7723e-04) (hash(x)=18300759)
-960 train 3.995192 (lr=5.7529e-04) (hash(x)=21443001)
-970 train 3.939906 (lr=5.7326e-04) (hash(x)=15210971)
-980 train 4.115149 (lr=5.7116e-04) (hash(x)=20874579)
-990 train 4.031127 (lr=5.6899e-04) (hash(x)=20690802)
-1000 val loss 3.9732
-1000 val perplexity 53.1533
-1000 train 3.894017 (lr=5.6674e-04) (hash(x)=18635877)
-1010 train 4.003831 (lr=5.6442e-04) (hash(x)=21141462)
-1020 train 4.024293 (lr=5.6202e-04) (hash(x)=19483400)
-1030 train 3.927857 (lr=5.5956e-04) (hash(x)=24792825)
-1040 train 3.902182 (lr=5.5702e-04) (hash(x)=16743928)
-1050 train 3.946391 (lr=5.5441e-04) (hash(x)=22540052)
-1060 train 3.970180 (lr=5.5174e-04) (hash(x)=20546637)
-1070 train 3.862578 (lr=5.4899e-04) (hash(x)=13739753)
-1080 train 3.913860 (lr=5.4618e-04) (hash(x)=20071304)
-1090 train 3.873674 (lr=5.4330e-04) (hash(x)=19339656)
-1100 val loss 3.8920
-1100 val perplexity 49.0107
-1100 train 3.858489 (lr=5.4035e-04) (hash(x)=20039628)
-1110 train 3.917438 (lr=5.3734e-04) (hash(x)=19249691)
-1120 train 3.938951 (lr=5.3426e-04) (hash(x)=19048809)
-1130 train 3.863695 (lr=5.3113e-04) (hash(x)=19309000)
-1140 train 3.822195 (lr=5.2792e-04) (hash(x)=21787927)
-1150 train 3.734058 (lr=5.2466e-04) (hash(x)=19998536)
-1160 train 3.882380 (lr=5.2134e-04) (hash(x)=23724694)
-1170 train 3.911989 (lr=5.1796e-04) (hash(x)=17148204)
-1180 train 3.818443 (lr=5.1452e-04) (hash(x)=19511852)
-1190 train 3.791586 (lr=5.1102e-04) (hash(x)=19352862)
-1200 val loss 3.8240
-1200 val perplexity 45.7866
-1200 train 3.737743 (lr=5.0746e-04) (hash(x)=19530144)
-1210 train 3.912196 (lr=5.0386e-04) (hash(x)=19897392)
-1220 train 3.818771 (lr=5.0019e-04) (hash(x)=24402963)
-1230 train 3.735388 (lr=4.9648e-04) (hash(x)=18589754)
-1240 train 3.742621 (lr=4.9271e-04) (hash(x)=21951750)
-1250 train 3.816519 (lr=4.8889e-04) (hash(x)=18611379)
-1260 train 3.832747 (lr=4.8503e-04) (hash(x)=20338521)
-1270 train 3.754608 (lr=4.8111e-04) (hash(x)=18308405)
-1280 train 3.757423 (lr=4.7715e-04) (hash(x)=20920945)
-1290 train 3.719588 (lr=4.7315e-04) (hash(x)=20094152)
-1300 val loss 3.7609
-1300 val perplexity 42.9868
-1300 train 3.835805 (lr=4.6909e-04) (hash(x)=19003947)
-1310 train 3.796214 (lr=4.6500e-04) (hash(x)=19356356)
-1320 train 3.745485 (lr=4.6086e-04) (hash(x)=19390343)
-1330 train 3.685663 (lr=4.5669e-04) (hash(x)=17535982)
-1340 train 3.707920 (lr=4.5247e-04) (hash(x)=21959104)
-1350 train 3.809441 (lr=4.4822e-04) (hash(x)=21982099)
-1360 train 3.764965 (lr=4.4393e-04) (hash(x)=17176243)
-1370 train 3.709131 (lr=4.3960e-04) (hash(x)=21223627)
-1380 train 3.727587 (lr=4.3524e-04) (hash(x)=22462931)
-1390 train 3.752164 (lr=4.3085e-04) (hash(x)=19153464)
-1400 val loss 3.7104
-1400 val perplexity 40.8690
-1400 train 3.641955 (lr=4.2643e-04) (hash(x)=20146496)
-1410 train 3.699762 (lr=4.2197e-04) (hash(x)=21781327)
-1420 train 3.770651 (lr=4.1749e-04) (hash(x)=19234051)
-1430 train 3.817792 (lr=4.1298e-04) (hash(x)=20769898)
-1440 train 3.670071 (lr=4.0845e-04) (hash(x)=22042430)
-1450 train 3.691169 (lr=4.0389e-04) (hash(x)=20527387)
-1460 train 3.677291 (lr=3.9931e-04) (hash(x)=18925766)
-1470 train 3.673635 (lr=3.9470e-04) (hash(x)=18198851)
-1480 train 3.748024 (lr=3.9008e-04) (hash(x)=21847281)
-1490 train 3.716259 (lr=3.8544e-04) (hash(x)=15645274)
-1500 val loss 3.6622
-1500 val perplexity 38.9466
-1500 train 3.583534 (lr=3.8078e-04) (hash(x)=14693370)
-1510 train 3.596366 (lr=3.7610e-04) (hash(x)=21810857)
-1520 train 3.692387 (lr=3.7142e-04) (hash(x)=18980052)
-1530 train 3.720563 (lr=3.6671e-04) (hash(x)=22868842)
-1540 train 3.633126 (lr=3.6200e-04) (hash(x)=17289166)
-1550 train 3.716073 (lr=3.5728e-04) (hash(x)=18109373)
-1560 train 3.750870 (lr=3.5255e-04) (hash(x)=22377619)
-1570 train 3.574374 (lr=3.4781e-04) (hash(x)=20307182)
-1580 train 3.724839 (lr=3.4306e-04) (hash(x)=20391291)
-1590 train 3.589329 (lr=3.3831e-04) (hash(x)=18690939)
-1600 val loss 3.6260
-1600 val perplexity 37.5637
-1600 train 3.602168 (lr=3.3356e-04) (hash(x)=16719886)
-1610 train 3.596222 (lr=3.2881e-04) (hash(x)=16040840)
-1620 train 3.670853 (lr=3.2406e-04) (hash(x)=25795140)
-1630 train 3.673065 (lr=3.1931e-04) (hash(x)=21936964)
-1640 train 3.570467 (lr=3.1456e-04) (hash(x)=11593027)
-1650 train 3.628470 (lr=3.0982e-04) (hash(x)=23256718)
-1660 train 3.589739 (lr=3.0509e-04) (hash(x)=20289140)
-1670 train 3.595504 (lr=3.0036e-04) (hash(x)=18382509)
-1680 train 3.661044 (lr=2.9564e-04) (hash(x)=26604329)
-1690 train 3.628746 (lr=2.9093e-04) (hash(x)=22122534)
-1700 val loss 3.5943
-1700 val perplexity 36.3917
-1700 train 3.642315 (lr=2.8624e-04) (hash(x)=22884812)
-1710 train 3.588021 (lr=2.8156e-04) (hash(x)=18598496)
-1720 train 3.610672 (lr=2.7689e-04) (hash(x)=22147447)
-1730 train 3.614013 (lr=2.7224e-04) (hash(x)=18912135)
-1740 train 3.451508 (lr=2.6761e-04) (hash(x)=21952974)
-1750 train 3.563020 (lr=2.6299e-04) (hash(x)=18390978)
-1760 train 3.575140 (lr=2.5840e-04) (hash(x)=17979288)
-1770 train 3.581527 (lr=2.5383e-04) (hash(x)=18979099)
-1780 train 3.541154 (lr=2.4928e-04) (hash(x)=20511922)
-1790 train 3.599244 (lr=2.4476e-04) (hash(x)=21814060)
-1800 val loss 3.5667
-1800 val perplexity 35.3981
-1800 train 3.586027 (lr=2.4026e-04) (hash(x)=18076243)
-1810 train 3.575942 (lr=2.3580e-04) (hash(x)=19721209)
-1820 train 3.522937 (lr=2.3136e-04) (hash(x)=18140847)
-1830 train 3.585089 (lr=2.2695e-04) (hash(x)=22065413)
-1840 train 3.483308 (lr=2.2257e-04) (hash(x)=21166111)
-1850 train 3.485255 (lr=2.1823e-04) (hash(x)=21904302)
-1860 train 3.554620 (lr=2.1392e-04) (hash(x)=18283828)
-1870 train 3.506842 (lr=2.0965e-04) (hash(x)=20892773)
-1880 train 3.535563 (lr=2.0542e-04) (hash(x)=17781264)
-1890 train 3.503753 (lr=2.0122e-04) (hash(x)=20612962)
-1900 val loss 3.5417
-1900 val perplexity 34.5247
-1900 train 3.597557 (lr=1.9706e-04) (hash(x)=20421058)
-1910 train 3.568954 (lr=1.9295e-04) (hash(x)=18554808)
-1920 train 3.526215 (lr=1.8887e-04) (hash(x)=18189279)
-1930 train 3.543596 (lr=1.8485e-04) (hash(x)=19923279)
-1940 train 3.509617 (lr=1.8086e-04) (hash(x)=21665511)
-1950 train 3.530538 (lr=1.7692e-04) (hash(x)=22038489)
-1960 train 3.503208 (lr=1.7303e-04) (hash(x)=18896956)
-1970 train 3.531210 (lr=1.6919e-04) (hash(x)=20408383)
-1980 train 3.511794 (lr=1.6540e-04) (hash(x)=23039940)
-1990 train 3.644492 (lr=1.6166e-04) (hash(x)=20631172)
-2000 val loss 3.5254
-2000 val perplexity 33.9659
-2000 train 3.461013 (lr=1.5797e-04) (hash(x)=20206106)
-2010 train 3.556440 (lr=1.5433e-04) (hash(x)=21864020)
-2020 train 3.537346 (lr=1.5075e-04) (hash(x)=21391128)
-2030 train 3.503325 (lr=1.4723e-04) (hash(x)=18676636)
-2040 train 3.550397 (lr=1.4376e-04) (hash(x)=15987134)
-2050 train 3.544177 (lr=1.4035e-04) (hash(x)=23940769)
-2060 train 3.522619 (lr=1.3699e-04) (hash(x)=17152840)
-2070 train 3.490402 (lr=1.3370e-04) (hash(x)=22958445)
-2080 train 3.543362 (lr=1.3047e-04) (hash(x)=23616724)
-2090 train 3.512293 (lr=1.2730e-04) (hash(x)=16571283)
-2100 val loss 3.5006
-2100 val perplexity 33.1352
-2100 train 3.510895 (lr=1.2419e-04) (hash(x)=19070327)
-2110 train 3.554395 (lr=1.2115e-04) (hash(x)=20758626)
-2120 train 3.498906 (lr=1.1817e-04) (hash(x)=18397810)
-2130 train 3.488154 (lr=1.1525e-04) (hash(x)=18045234)
-2140 train 3.591183 (lr=1.1241e-04) (hash(x)=18969565)
-2150 train 3.527388 (lr=1.0963e-04) (hash(x)=17149701)
-2160 train 3.446743 (lr=1.0692e-04) (hash(x)=23075443)
-2170 train 3.482064 (lr=1.0427e-04) (hash(x)=17977337)
-2180 train 3.541664 (lr=1.0170e-04) (hash(x)=20657407)
-2190 train 3.530729 (lr=9.9199e-05) (hash(x)=19985933)
-2200 val loss 3.4853
-2200 val perplexity 32.6307
-2200 train 3.483011 (lr=9.6769e-05) (hash(x)=19025387)
-2210 train 3.498638 (lr=9.4412e-05) (hash(x)=23318724)
-2220 train 3.571498 (lr=9.2127e-05) (hash(x)=19308039)
-2230 train 3.414404 (lr=8.9916e-05) (hash(x)=20733682)
-2240 train 3.511224 (lr=8.7779e-05) (hash(x)=18519560)
-2250 train 3.516548 (lr=8.5717e-05) (hash(x)=20299444)
-2260 train 3.474868 (lr=8.3731e-05) (hash(x)=18819615)
-2270 train 3.507273 (lr=8.1821e-05) (hash(x)=18743705)
-2280 train 3.496422 (lr=7.9988e-05) (hash(x)=17190793)
-2290 train 3.527422 (lr=7.8232e-05) (hash(x)=23126906)
-2300 val loss 3.4727
-2300 val perplexity 32.2222
-2300 train 3.461187 (lr=7.6555e-05) (hash(x)=24812267)
-2310 train 3.446771 (lr=7.4956e-05) (hash(x)=20147105)
-2320 train 3.561950 (lr=7.3436e-05) (hash(x)=24619936)
-2330 train 3.426948 (lr=7.1995e-05) (hash(x)=17730658)
-2340 train 3.403539 (lr=7.0635e-05) (hash(x)=19345731)
-2350 train 3.441451 (lr=6.9354e-05) (hash(x)=26415058)
-2360 train 3.478770 (lr=6.8155e-05) (hash(x)=18622999)
-2370 train 3.407625 (lr=6.7036e-05) (hash(x)=18848567)
-2380 train 3.490878 (lr=6.5999e-05) (hash(x)=21903360)
-2390 train 3.504000 (lr=6.5044e-05) (hash(x)=23479216)
-2400 val loss 3.4640
-2400 val perplexity 31.9446
-2400 train 3.501966 (lr=6.4171e-05) (hash(x)=19762013)
-2410 train 3.406570 (lr=6.3380e-05) (hash(x)=20622912)
-2420 train 3.524851 (lr=6.2672e-05) (hash(x)=17669226)
-2430 train 3.452855 (lr=6.2046e-05) (hash(x)=19536719)
-2440 train 3.393947 (lr=6.1504e-05) (hash(x)=19913725)
-2450 train 3.437440 (lr=6.1045e-05) (hash(x)=20782072)
-2460 train 3.405302 (lr=6.0669e-05) (hash(x)=19795002)
-2470 train 3.453721 (lr=6.0376e-05) (hash(x)=17003051)
-2480 train 3.418798 (lr=6.0167e-05) (hash(x)=21414190)
-2490 train 3.468121 (lr=6.0042e-05) (hash(x)=20943620)
-2499 val loss 3.4585
-2499 val perplexity 31.7705
+0 val loss 10.9803
+0 val perplexity 58708.7344
+0 train 10.981787 (lr=8.3916e-07) (hash(x)=17183924)
+10 train 10.119558 (lr=9.2308e-06) (hash(x)=19047484)
+20 train 9.643312 (lr=1.7622e-05) (hash(x)=20001614)
+30 train 9.412062 (lr=2.6014e-05) (hash(x)=21278543)
+40 train 9.157112 (lr=3.4406e-05) (hash(x)=29208111)
+50 train 8.842901 (lr=4.2797e-05) (hash(x)=20801292)
+60 train 8.502390 (lr=5.1189e-05) (hash(x)=27442073)
+70 train 8.139378 (lr=5.9580e-05) (hash(x)=19918406)
+80 train 7.781339 (lr=6.7972e-05) (hash(x)=20569806)
+90 train 7.672777 (lr=7.6364e-05) (hash(x)=26013441)
+100 val loss 7.4532
+100 val perplexity 1725.3317
+100 train 7.426733 (lr=8.4755e-05) (hash(x)=29039793)
+110 train 7.252076 (lr=9.3147e-05) (hash(x)=16550842)
+120 train 7.087349 (lr=1.0154e-04) (hash(x)=17250570)
+130 train 7.114640 (lr=1.0993e-04) (hash(x)=17154654)
+140 train 6.974027 (lr=1.1832e-04) (hash(x)=22437415)
+150 train 6.915896 (lr=1.2671e-04) (hash(x)=18645582)
+160 train 6.661926 (lr=1.3510e-04) (hash(x)=19460261)
+170 train 6.514016 (lr=1.4350e-04) (hash(x)=20953331)
+180 train 6.590486 (lr=1.5189e-04) (hash(x)=20883978)
+190 train 6.469072 (lr=1.6028e-04) (hash(x)=25571428)
+200 val loss 6.4090
+200 val perplexity 607.3052
+200 train 6.332068 (lr=1.6867e-04) (hash(x)=16049634)
+210 train 6.316191 (lr=1.7706e-04) (hash(x)=15886413)
+220 train 6.376137 (lr=1.8545e-04) (hash(x)=21731617)
+230 train 6.155892 (lr=1.9385e-04) (hash(x)=22509449)
+240 train 6.096032 (lr=2.0224e-04) (hash(x)=17396544)
+250 train 6.141424 (lr=2.1063e-04) (hash(x)=17748336)
+260 train 6.024675 (lr=2.1902e-04) (hash(x)=21263532)
+270 train 5.954779 (lr=2.2741e-04) (hash(x)=19623735)
+280 train 5.933280 (lr=2.3580e-04) (hash(x)=18337637)
+290 train 5.886799 (lr=2.4420e-04) (hash(x)=20005903)
+300 val loss 5.8034
+300 val perplexity 331.4403
+300 train 5.822334 (lr=2.5259e-04) (hash(x)=20938811)
+310 train 5.716088 (lr=2.6098e-04) (hash(x)=20688368)
+320 train 5.731483 (lr=2.6937e-04) (hash(x)=19776598)
+330 train 5.648857 (lr=2.7776e-04) (hash(x)=19721837)
+340 train 5.580023 (lr=2.8615e-04) (hash(x)=16578291)
+350 train 5.499309 (lr=2.9455e-04) (hash(x)=19998371)
+360 train 5.606770 (lr=3.0294e-04) (hash(x)=21565819)
+370 train 5.509754 (lr=3.1133e-04) (hash(x)=19484104)
+380 train 5.396148 (lr=3.1972e-04) (hash(x)=20659288)
+390 train 5.321680 (lr=3.2811e-04) (hash(x)=22061179)
+400 val loss 5.3577
+400 val perplexity 212.2278
+400 train 5.426161 (lr=3.3650e-04) (hash(x)=18557926)
+410 train 5.297194 (lr=3.4490e-04) (hash(x)=24034052)
+420 train 5.191974 (lr=3.5329e-04) (hash(x)=19768493)
+430 train 5.275582 (lr=3.6168e-04) (hash(x)=21678884)
+440 train 5.205054 (lr=3.7007e-04) (hash(x)=19589702)
+450 train 5.235847 (lr=3.7846e-04) (hash(x)=19902911)
+460 train 5.006043 (lr=3.8685e-04) (hash(x)=18881029)
+470 train 5.135803 (lr=3.9524e-04) (hash(x)=24329185)
+480 train 5.019996 (lr=4.0364e-04) (hash(x)=18709451)
+490 train 4.990302 (lr=4.1203e-04) (hash(x)=18834705)
+500 val loss 4.9887
+500 val perplexity 146.7455
+500 train 5.025167 (lr=4.2042e-04) (hash(x)=24221098)
+510 train 4.981864 (lr=4.2881e-04) (hash(x)=20010758)
+520 train 4.936800 (lr=4.3720e-04) (hash(x)=18547084)
+530 train 4.757543 (lr=4.4559e-04) (hash(x)=18282458)
+540 train 4.891625 (lr=4.5399e-04) (hash(x)=20671087)
+550 train 4.742090 (lr=4.6238e-04) (hash(x)=18634421)
+560 train 4.680741 (lr=4.7077e-04) (hash(x)=17067920)
+570 train 4.813383 (lr=4.7916e-04) (hash(x)=22276881)
+580 train 4.703204 (lr=4.8755e-04) (hash(x)=22973651)
+590 train 4.692297 (lr=4.9594e-04) (hash(x)=18096900)
+600 val loss 4.6245
+600 val perplexity 101.9488
+600 train 4.571012 (lr=5.0434e-04) (hash(x)=20617414)
+610 train 4.444651 (lr=5.1273e-04) (hash(x)=20722525)
+620 train 4.647902 (lr=5.2112e-04) (hash(x)=20785452)
+630 train 4.590510 (lr=5.2951e-04) (hash(x)=16647858)
+640 train 4.520114 (lr=5.3790e-04) (hash(x)=20798993)
+650 train 4.445772 (lr=5.4629e-04) (hash(x)=20912393)
+660 train 4.281695 (lr=5.5469e-04) (hash(x)=21505604)
+670 train 4.516667 (lr=5.6308e-04) (hash(x)=20263120)
+680 train 4.424880 (lr=5.7147e-04) (hash(x)=20779495)
+690 train 4.345970 (lr=5.7986e-04) (hash(x)=16555678)
+700 val loss 4.3729
+700 val perplexity 79.2770
+700 train 4.194740 (lr=5.8825e-04) (hash(x)=20156261)
+710 train 4.175675 (lr=5.9664e-04) (hash(x)=17064702)
+720 train 4.400195 (lr=5.9999e-04) (hash(x)=19321113)
+730 train 4.339067 (lr=5.9991e-04) (hash(x)=19738613)
+740 train 4.270685 (lr=5.9974e-04) (hash(x)=18585508)
+750 train 4.147637 (lr=5.9949e-04) (hash(x)=18605696)
+760 train 4.313256 (lr=5.9915e-04) (hash(x)=22711193)
+770 train 4.256395 (lr=5.9874e-04) (hash(x)=17163545)
+780 train 4.161443 (lr=5.9824e-04) (hash(x)=22239165)
+790 train 4.263532 (lr=5.9765e-04) (hash(x)=19202751)
+800 val loss 4.2109
+800 val perplexity 67.4170
+800 train 4.271319 (lr=5.9698e-04) (hash(x)=22502753)
+810 train 4.220746 (lr=5.9623e-04) (hash(x)=20197889)
+820 train 4.157265 (lr=5.9540e-04) (hash(x)=19376072)
+830 train 4.014478 (lr=5.9449e-04) (hash(x)=18293711)
+840 train 4.184201 (lr=5.9349e-04) (hash(x)=19061676)
+850 train 4.193909 (lr=5.9241e-04) (hash(x)=20557626)
+860 train 4.115611 (lr=5.9126e-04) (hash(x)=18642266)
+870 train 3.989679 (lr=5.9002e-04) (hash(x)=20957453)
+880 train 4.021313 (lr=5.8869e-04) (hash(x)=21781033)
+890 train 4.133365 (lr=5.8729e-04) (hash(x)=21094646)
+900 val loss 4.0614
+900 val perplexity 58.0554
+900 train 4.118994 (lr=5.8581e-04) (hash(x)=21499150)
+910 train 4.089995 (lr=5.8425e-04) (hash(x)=19423438)
+920 train 3.900768 (lr=5.8262e-04) (hash(x)=20543615)
+930 train 4.088742 (lr=5.8090e-04) (hash(x)=19245236)
+940 train 4.089410 (lr=5.7911e-04) (hash(x)=24035334)
+950 train 4.021673 (lr=5.7723e-04) (hash(x)=16703311)
+960 train 3.992325 (lr=5.7529e-04) (hash(x)=23101805)
+970 train 3.950537 (lr=5.7326e-04) (hash(x)=20320872)
+980 train 3.993726 (lr=5.7116e-04) (hash(x)=20990946)
+990 train 4.003374 (lr=5.6899e-04) (hash(x)=18628956)
+1000 val loss 3.9629
+1000 val perplexity 52.6105
+1000 train 3.938634 (lr=5.6674e-04) (hash(x)=22968748)
+1010 train 3.917498 (lr=5.6442e-04) (hash(x)=20541414)
+1020 train 3.975941 (lr=5.6202e-04) (hash(x)=19274167)
+1030 train 3.934474 (lr=5.5956e-04) (hash(x)=22645317)
+1040 train 3.904500 (lr=5.5702e-04) (hash(x)=17884034)
+1050 train 3.906873 (lr=5.5441e-04) (hash(x)=19299722)
+1060 train 3.918670 (lr=5.5174e-04) (hash(x)=18409931)
+1070 train 3.894806 (lr=5.4899e-04) (hash(x)=22455477)
+1080 train 3.869685 (lr=5.4618e-04) (hash(x)=20085073)
+1090 train 3.896967 (lr=5.4330e-04) (hash(x)=20371024)
+1100 val loss 3.8682
+1100 val perplexity 47.8572
+1100 train 3.847680 (lr=5.4035e-04) (hash(x)=17534387)
+1110 train 3.852636 (lr=5.3734e-04) (hash(x)=18107517)
+1120 train 3.882497 (lr=5.3426e-04) (hash(x)=21862096)
+1130 train 3.877234 (lr=5.3113e-04) (hash(x)=24296067)
+1140 train 3.777543 (lr=5.2792e-04) (hash(x)=25489932)
+1150 train 3.890559 (lr=5.2466e-04) (hash(x)=19256895)
+1160 train 3.802947 (lr=5.2134e-04) (hash(x)=20915643)
+1170 train 3.871340 (lr=5.1796e-04) (hash(x)=22053731)
+1180 train 3.828829 (lr=5.1452e-04) (hash(x)=20012827)
+1190 train 3.814388 (lr=5.1102e-04) (hash(x)=21694459)
+1200 val loss 3.8086
+1200 val perplexity 45.0878
+1200 train 3.850748 (lr=5.0746e-04) (hash(x)=21755992)
+1210 train 3.837916 (lr=5.0386e-04) (hash(x)=21880996)
+1220 train 3.814308 (lr=5.0019e-04) (hash(x)=20520945)
+1230 train 3.783235 (lr=4.9648e-04) (hash(x)=19770784)
+1240 train 3.790468 (lr=4.9271e-04) (hash(x)=18571983)
+1250 train 3.794569 (lr=4.8889e-04) (hash(x)=20735694)
+1260 train 3.746379 (lr=4.8503e-04) (hash(x)=21156163)
+1270 train 3.808167 (lr=4.8111e-04) (hash(x)=21271664)
+1280 train 3.775006 (lr=4.7715e-04) (hash(x)=19315272)
+1290 train 3.743977 (lr=4.7315e-04) (hash(x)=21142191)
+1300 val loss 3.7492
+1300 val perplexity 42.4869
+1300 train 3.781651 (lr=4.6909e-04) (hash(x)=17145293)
+1310 train 3.964586 (lr=4.6500e-04) (hash(x)=17713742)
+1320 train 3.759767 (lr=4.6086e-04) (hash(x)=33348747)
+1330 train 3.714254 (lr=4.5669e-04) (hash(x)=20702111)
+1340 train 3.794268 (lr=4.5247e-04) (hash(x)=17898284)
+1350 train 3.702932 (lr=4.4822e-04) (hash(x)=17904915)
+1360 train 3.716854 (lr=4.4393e-04) (hash(x)=23325893)
+1370 train 3.642303 (lr=4.3960e-04) (hash(x)=20717754)
+1380 train 3.745301 (lr=4.3524e-04) (hash(x)=16859207)
+1390 train 3.706009 (lr=4.3085e-04) (hash(x)=21029500)
+1400 val loss 3.7059
+1400 val perplexity 40.6869
+1400 train 3.646819 (lr=4.2643e-04) (hash(x)=17789655)
+1410 train 3.764728 (lr=4.2197e-04) (hash(x)=19965237)
+1420 train 3.735120 (lr=4.1749e-04) (hash(x)=19262560)
+1430 train 3.640206 (lr=4.1298e-04) (hash(x)=18593169)
+1440 train 3.645439 (lr=4.0845e-04) (hash(x)=20607672)
+1450 train 3.698761 (lr=4.0389e-04) (hash(x)=20682942)
+1460 train 3.649403 (lr=3.9931e-04) (hash(x)=20220347)
+1470 train 3.630159 (lr=3.9470e-04) (hash(x)=15833389)
+1480 train 3.758533 (lr=3.9008e-04) (hash(x)=19495027)
+1490 train 3.668654 (lr=3.8544e-04) (hash(x)=22138106)
+1500 val loss 3.6617
+1500 val perplexity 38.9275
+1500 train 3.621582 (lr=3.8078e-04) (hash(x)=24123427)
+1510 train 3.583689 (lr=3.7610e-04) (hash(x)=19024825)
+1520 train 3.724061 (lr=3.7142e-04) (hash(x)=19780005)
+1530 train 3.714616 (lr=3.6671e-04) (hash(x)=19134410)
+1540 train 3.631273 (lr=3.6200e-04) (hash(x)=19855253)
+1550 train 3.595632 (lr=3.5728e-04) (hash(x)=20259196)
+1560 train 3.597122 (lr=3.5255e-04) (hash(x)=18870222)
+1570 train 3.631028 (lr=3.4781e-04) (hash(x)=16067782)
+1580 train 3.626120 (lr=3.4306e-04) (hash(x)=21604205)
+1590 train 3.600831 (lr=3.3831e-04) (hash(x)=19076298)
+1600 val loss 3.6255
+1600 val perplexity 37.5444
+1600 train 3.623093 (lr=3.3356e-04) (hash(x)=19064055)
+1610 train 3.674186 (lr=3.2881e-04) (hash(x)=19196041)
+1620 train 3.626938 (lr=3.2406e-04) (hash(x)=25625303)
+1630 train 3.576034 (lr=3.1931e-04) (hash(x)=20618762)
+1640 train 3.618860 (lr=3.1456e-04) (hash(x)=21102615)
+1650 train 3.564730 (lr=3.0982e-04) (hash(x)=21088682)
+1660 train 3.549482 (lr=3.0509e-04) (hash(x)=15297157)
+1670 train 3.592760 (lr=3.0036e-04) (hash(x)=19690570)
+1680 train 3.610930 (lr=2.9564e-04) (hash(x)=21445411)
+1690 train 3.594339 (lr=2.9093e-04) (hash(x)=21683876)
+1700 val loss 3.5874
+1700 val perplexity 36.1407
+1700 train 3.602877 (lr=2.8624e-04) (hash(x)=17449466)
+1710 train 3.607208 (lr=2.8156e-04) (hash(x)=16754900)
+1720 train 3.572566 (lr=2.7689e-04) (hash(x)=20736439)
+1730 train 3.607226 (lr=2.7224e-04) (hash(x)=21475325)
+1740 train 3.551026 (lr=2.6761e-04) (hash(x)=31466218)
+1750 train 3.635657 (lr=2.6299e-04) (hash(x)=17740223)
+1760 train 3.570215 (lr=2.5840e-04) (hash(x)=17626006)
+1770 train 3.422887 (lr=2.5383e-04) (hash(x)=22276216)
+1780 train 3.536065 (lr=2.4928e-04) (hash(x)=14277004)
+1790 train 3.561902 (lr=2.4476e-04) (hash(x)=20600488)
+1800 val loss 3.5621
+1800 val perplexity 35.2366
+1800 train 3.574463 (lr=2.4026e-04) (hash(x)=17508978)
+1810 train 3.486950 (lr=2.3580e-04) (hash(x)=22374263)
+1820 train 3.626369 (lr=2.3136e-04) (hash(x)=18959095)
+1830 train 3.537313 (lr=2.2695e-04) (hash(x)=19485836)
+1840 train 3.531831 (lr=2.2257e-04) (hash(x)=24276692)
+1850 train 3.558848 (lr=2.1823e-04) (hash(x)=20554224)
+1860 train 3.547421 (lr=2.1392e-04) (hash(x)=18068030)
+1870 train 3.569391 (lr=2.0965e-04) (hash(x)=18349056)
+1880 train 3.505214 (lr=2.0542e-04) (hash(x)=20640059)
+1890 train 3.511017 (lr=2.0122e-04) (hash(x)=19667347)
+1900 val loss 3.5358
+1900 val perplexity 34.3211
+1900 train 3.568940 (lr=1.9706e-04) (hash(x)=19559109)
+1910 train 3.590243 (lr=1.9295e-04) (hash(x)=22490711)
+1920 train 3.525853 (lr=1.8887e-04) (hash(x)=18268544)
+1930 train 3.506641 (lr=1.8485e-04) (hash(x)=20585037)
+1940 train 3.577510 (lr=1.8086e-04) (hash(x)=30293578)
+1950 train 3.514944 (lr=1.7692e-04) (hash(x)=21836208)
+1960 train 3.449993 (lr=1.7303e-04) (hash(x)=23417084)
+1970 train 3.542269 (lr=1.6919e-04) (hash(x)=19390086)
+1980 train 3.469500 (lr=1.6540e-04) (hash(x)=20129488)
+1990 train 3.513119 (lr=1.6166e-04) (hash(x)=20177413)
+2000 val loss 3.5171
+2000 val perplexity 33.6868
+2000 train 3.443385 (lr=1.5797e-04) (hash(x)=15921967)
+2010 train 3.511441 (lr=1.5433e-04) (hash(x)=20819557)
+2020 train 3.532563 (lr=1.5075e-04) (hash(x)=18729616)
+2030 train 3.469848 (lr=1.4723e-04) (hash(x)=15730659)
+2040 train 3.511942 (lr=1.4376e-04) (hash(x)=18679863)
+2050 train 3.563097 (lr=1.4035e-04) (hash(x)=17110863)
+2060 train 3.472592 (lr=1.3699e-04) (hash(x)=18291952)
+2070 train 3.498121 (lr=1.3370e-04) (hash(x)=18249172)
+2080 train 3.528080 (lr=1.3047e-04) (hash(x)=22686659)
+2090 train 3.483869 (lr=1.2730e-04) (hash(x)=18354325)
+2100 val loss 3.4982
+2100 val perplexity 33.0558
+2100 train 3.473750 (lr=1.2419e-04) (hash(x)=18204536)
+2110 train 3.539428 (lr=1.2115e-04) (hash(x)=22394614)
+2120 train 3.425844 (lr=1.1817e-04) (hash(x)=18121192)
+2130 train 3.468009 (lr=1.1525e-04) (hash(x)=24393898)
+2140 train 3.481289 (lr=1.1241e-04) (hash(x)=27349908)
+2150 train 3.503107 (lr=1.0963e-04) (hash(x)=18262196)
+2160 train 3.434892 (lr=1.0692e-04) (hash(x)=20081235)
+2170 train 3.467141 (lr=1.0427e-04) (hash(x)=16984900)
+2180 train 3.536548 (lr=1.0170e-04) (hash(x)=24294588)
+2190 train 3.494779 (lr=9.9199e-05) (hash(x)=17042450)
+2200 val loss 3.4841
+2200 val perplexity 32.5938
+2200 train 3.383519 (lr=9.6769e-05) (hash(x)=23876593)
+2210 train 3.530521 (lr=9.4412e-05) (hash(x)=19173831)
+2220 train 3.435912 (lr=9.2127e-05) (hash(x)=21211298)
+2230 train 3.437502 (lr=8.9916e-05) (hash(x)=18721700)
+2240 train 3.435278 (lr=8.7779e-05) (hash(x)=19825319)
+2250 train 3.459467 (lr=8.5717e-05) (hash(x)=18378838)
+2260 train 3.463020 (lr=8.3731e-05) (hash(x)=13199654)
+2270 train 3.380137 (lr=8.1821e-05) (hash(x)=18597807)
+2280 train 3.485405 (lr=7.9988e-05) (hash(x)=22012917)
+2290 train 3.404468 (lr=7.8232e-05) (hash(x)=22305542)
+2300 val loss 3.4689
+2300 val perplexity 32.1027
+2300 train 3.500764 (lr=7.6555e-05) (hash(x)=19782726)
+2310 train 3.467016 (lr=7.4956e-05) (hash(x)=22042291)
+2320 train 3.471705 (lr=7.3436e-05) (hash(x)=17370832)
+2330 train 3.506680 (lr=7.1995e-05) (hash(x)=19612132)
+2340 train 3.473424 (lr=7.0635e-05) (hash(x)=18803286)
+2350 train 3.458232 (lr=6.9354e-05) (hash(x)=20196070)
+2360 train 3.427007 (lr=6.8155e-05) (hash(x)=19997454)
+2370 train 3.434407 (lr=6.7036e-05) (hash(x)=16704483)
+2380 train 3.460657 (lr=6.5999e-05) (hash(x)=19131833)
+2390 train 3.406964 (lr=6.5044e-05) (hash(x)=20507035)
+2400 val loss 3.4604
+2400 val perplexity 31.8296
+2400 train 3.519377 (lr=6.4171e-05) (hash(x)=20254090)
+2410 train 3.452719 (lr=6.3380e-05) (hash(x)=17926964)
+2420 train 3.483057 (lr=6.2672e-05) (hash(x)=20607860)
+2430 train 3.416114 (lr=6.2046e-05) (hash(x)=11288168)
+2440 train 3.532155 (lr=6.1504e-05) (hash(x)=21486139)
+2450 train 3.474133 (lr=6.1045e-05) (hash(x)=18994546)
+2460 train 3.407577 (lr=6.0669e-05) (hash(x)=21218220)
+2470 train 3.506916 (lr=6.0376e-05) (hash(x)=20504341)
+2480 train 3.460694 (lr=6.0167e-05) (hash(x)=17937386)
+2490 train 3.374489 (lr=6.0042e-05) (hash(x)=21870434)
+2499 val loss 3.4539
+2499 val perplexity 31.6250

one_mask_per_head_2_latent_vectors/model_02499.pt

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:e6e5b83698a0de24d4f5ad15ec593cbe4fddced132e2c7b15b1afaa74c7ecfd8
+oid sha256:360ed544e74a29df85fed16358bbbe3097bfa42c09a47957d20a37a8832653ed
 size 555249266

one_mask_per_head_2_latent_vectors/optimizer_02499.pt

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:49c48fa9cd63ffdc18e4cbe54f2d3fb3987cb7e18a6b1cdccb6ddc1469074352
+oid sha256:fea366a85ff360d56273b55c5bd1f7553ccc96e0341d8be23722e9f4ff17ab0a
 size 1009849270