andrew-healey commited on
Commit
963b183
·
verified ·
1 Parent(s): 3e97235

Upload folder using huggingface_hub

Browse files
allowing_more_selection_patterns/one_mask_per_head_1_latent_vector_degenerate_seed_1338/args.json ADDED
@@ -0,0 +1 @@
 
 
1
+ {"hellaswag": true, "attention_kind": "selective", "log_dir": "allowing_more_selection_patterns/one_mask_per_head_1_latent_vector_degenerate_seed_1338", "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": 1338, "memory_penalty_epsilon": 0.1, "selection_head_linear_combo": "n_latent_masks", "selection_head_linear_combo_scale": 0.0, "disable_selection_head_linear_combo_bias": true, "assert_latent_matches_no_head": 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": 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": 1, "init_latent_masks_to_identity": true, "init_latent_masks_to_inverse": false, "latent_mask_scale": null, "latent_mask_sigmoid": false, "S_layernorm": false, "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": "one_mask_per_head_1_latent_vector_degenerate", "latent_mask_precision": "bfloat16"}
allowing_more_selection_patterns/one_mask_per_head_1_latent_vector_degenerate_seed_1338/dataloader_02499.pt ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:86bf2d12aa2d7d065a4e8fd1660928dd629c2417a5062dd8aecee75fb283886f
3
+ size 964
allowing_more_selection_patterns/one_mask_per_head_1_latent_vector_degenerate_seed_1338/log2.txt ADDED
@@ -0,0 +1,303 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ max_steps: 2500
2
+ 0 val loss 10.9653
3
+ 0 val perplexity 57829.3633
4
+ 0 train 10.964844 (lr=8.3916e-07) (hash(x)=20472174)
5
+ 10 train 10.149137 (lr=9.2308e-06) (hash(x)=18772821)
6
+ 20 train 9.635406 (lr=1.7622e-05) (hash(x)=18169834)
7
+ 30 train 9.403813 (lr=2.6014e-05) (hash(x)=21772648)
8
+ 40 train 9.198076 (lr=3.4406e-05) (hash(x)=19737653)
9
+ 50 train 8.879588 (lr=4.2797e-05) (hash(x)=18397495)
10
+ 60 train 8.527857 (lr=5.1189e-05) (hash(x)=19313033)
11
+ 70 train 8.126386 (lr=5.9580e-05) (hash(x)=19235256)
12
+ 80 train 7.880077 (lr=6.7972e-05) (hash(x)=15263116)
13
+ 90 train 7.608851 (lr=7.6364e-05) (hash(x)=20660013)
14
+ 100 val loss 7.4652
15
+ 100 val perplexity 1746.1788
16
+ 100 train 7.425018 (lr=8.4755e-05) (hash(x)=19046599)
17
+ 110 train 7.326130 (lr=9.3147e-05) (hash(x)=24079550)
18
+ 120 train 7.092932 (lr=1.0154e-04) (hash(x)=26797340)
19
+ 130 train 7.108392 (lr=1.0993e-04) (hash(x)=20619718)
20
+ 140 train 6.925122 (lr=1.1832e-04) (hash(x)=18313305)
21
+ 150 train 6.869478 (lr=1.2671e-04) (hash(x)=20825391)
22
+ 160 train 6.646358 (lr=1.3510e-04) (hash(x)=20505694)
23
+ 170 train 6.525844 (lr=1.4350e-04) (hash(x)=24020301)
24
+ 180 train 6.584141 (lr=1.5189e-04) (hash(x)=23124134)
25
+ 190 train 6.422416 (lr=1.6028e-04) (hash(x)=17851881)
26
+ 200 val loss 6.3819
27
+ 200 val perplexity 591.0219
28
+ 200 train 6.362373 (lr=1.6867e-04) (hash(x)=21138724)
29
+ 210 train 6.190283 (lr=1.7706e-04) (hash(x)=16978198)
30
+ 220 train 6.121952 (lr=1.8545e-04) (hash(x)=20812935)
31
+ 230 train 6.197385 (lr=1.9385e-04) (hash(x)=20619471)
32
+ 240 train 6.132792 (lr=2.0224e-04) (hash(x)=17706413)
33
+ 250 train 5.943398 (lr=2.1063e-04) (hash(x)=21557598)
34
+ 260 train 5.892084 (lr=2.1902e-04) (hash(x)=18077648)
35
+ 270 train 6.004688 (lr=2.2741e-04) (hash(x)=17053957)
36
+ 280 train 5.930441 (lr=2.3580e-04) (hash(x)=20526478)
37
+ 290 train 5.760366 (lr=2.4420e-04) (hash(x)=22226244)
38
+ 300 val loss 5.7749
39
+ 300 val perplexity 322.1271
40
+ 300 train 5.687343 (lr=2.5259e-04) (hash(x)=19583531)
41
+ 310 train 5.543598 (lr=2.6098e-04) (hash(x)=18986313)
42
+ 320 train 5.761025 (lr=2.6937e-04) (hash(x)=23629157)
43
+ 330 train 5.663108 (lr=2.7776e-04) (hash(x)=21312406)
44
+ 340 train 5.482827 (lr=2.8615e-04) (hash(x)=19278398)
45
+ 350 train 5.504622 (lr=2.9455e-04) (hash(x)=19370729)
46
+ 360 train 5.337851 (lr=3.0294e-04) (hash(x)=20663223)
47
+ 370 train 5.504222 (lr=3.1133e-04) (hash(x)=22952038)
48
+ 380 train 5.452213 (lr=3.1972e-04) (hash(x)=21902455)
49
+ 390 train 5.441855 (lr=3.2811e-04) (hash(x)=21953350)
50
+ 400 val loss 5.3378
51
+ 400 val perplexity 208.0621
52
+ 400 train 5.401706 (lr=3.3650e-04) (hash(x)=22915018)
53
+ 410 train 5.319546 (lr=3.4490e-04) (hash(x)=18301453)
54
+ 420 train 5.226571 (lr=3.5329e-04) (hash(x)=17422968)
55
+ 430 train 5.097789 (lr=3.6168e-04) (hash(x)=20616951)
56
+ 440 train 5.279465 (lr=3.7007e-04) (hash(x)=19912399)
57
+ 450 train 5.125679 (lr=3.7846e-04) (hash(x)=21154839)
58
+ 460 train 5.072629 (lr=3.8685e-04) (hash(x)=18919228)
59
+ 470 train 5.048383 (lr=3.9524e-04) (hash(x)=20916162)
60
+ 480 train 4.943963 (lr=4.0364e-04) (hash(x)=19136877)
61
+ 490 train 5.049657 (lr=4.1203e-04) (hash(x)=22631452)
62
+ 500 val loss 4.9411
63
+ 500 val perplexity 139.9249
64
+ 500 train 4.969503 (lr=4.2042e-04) (hash(x)=21454149)
65
+ 510 train 4.809808 (lr=4.2881e-04) (hash(x)=18263355)
66
+ 520 train 4.776715 (lr=4.3720e-04) (hash(x)=21540112)
67
+ 530 train 4.775882 (lr=4.4559e-04) (hash(x)=20378460)
68
+ 540 train 4.856249 (lr=4.5399e-04) (hash(x)=25242015)
69
+ 550 train 4.734632 (lr=4.6238e-04) (hash(x)=18067565)
70
+ 560 train 4.629473 (lr=4.7077e-04) (hash(x)=18532926)
71
+ 570 train 4.525438 (lr=4.7916e-04) (hash(x)=17014430)
72
+ 580 train 4.551474 (lr=4.8755e-04) (hash(x)=21587296)
73
+ 590 train 4.646731 (lr=4.9594e-04) (hash(x)=19956044)
74
+ 600 val loss 4.5973
75
+ 600 val perplexity 99.2118
76
+ 600 train 4.645379 (lr=5.0434e-04) (hash(x)=20658639)
77
+ 610 train 4.542713 (lr=5.1273e-04) (hash(x)=18301842)
78
+ 620 train 4.500952 (lr=5.2112e-04) (hash(x)=18675952)
79
+ 630 train 4.600066 (lr=5.2951e-04) (hash(x)=17602704)
80
+ 640 train 4.523263 (lr=5.3790e-04) (hash(x)=21491802)
81
+ 650 train 4.522380 (lr=5.4629e-04) (hash(x)=20557423)
82
+ 660 train 4.388939 (lr=5.5469e-04) (hash(x)=21188117)
83
+ 670 train 4.503300 (lr=5.6308e-04) (hash(x)=20683583)
84
+ 680 train 4.433919 (lr=5.7147e-04) (hash(x)=24879052)
85
+ 690 train 4.322607 (lr=5.7986e-04) (hash(x)=19538285)
86
+ 700 val loss 4.3354
87
+ 700 val perplexity 76.3560
88
+ 700 train 4.351713 (lr=5.8825e-04) (hash(x)=18338040)
89
+ 710 train 4.347413 (lr=5.9664e-04) (hash(x)=21087427)
90
+ 720 train 4.283252 (lr=5.9999e-04) (hash(x)=20546196)
91
+ 730 train 4.307380 (lr=5.9991e-04) (hash(x)=19605436)
92
+ 740 train 4.272770 (lr=5.9974e-04) (hash(x)=20510264)
93
+ 750 train 4.327110 (lr=5.9949e-04) (hash(x)=19782550)
94
+ 760 train 4.216037 (lr=5.9915e-04) (hash(x)=20997594)
95
+ 770 train 4.088340 (lr=5.9874e-04) (hash(x)=20733511)
96
+ 780 train 4.186656 (lr=5.9824e-04) (hash(x)=17584884)
97
+ 790 train 4.122427 (lr=5.9765e-04) (hash(x)=20952808)
98
+ 800 val loss 4.1762
99
+ 800 val perplexity 65.1149
100
+ 800 train 4.070551 (lr=5.9698e-04) (hash(x)=16092551)
101
+ 810 train 4.112890 (lr=5.9623e-04) (hash(x)=16116155)
102
+ 820 train 4.177746 (lr=5.9540e-04) (hash(x)=22196607)
103
+ 830 train 4.065981 (lr=5.9449e-04) (hash(x)=21490054)
104
+ 840 train 3.967163 (lr=5.9349e-04) (hash(x)=19471853)
105
+ 850 train 4.118798 (lr=5.9241e-04) (hash(x)=40176264)
106
+ 860 train 4.113213 (lr=5.9126e-04) (hash(x)=19241841)
107
+ 870 train 3.951173 (lr=5.9002e-04) (hash(x)=19010654)
108
+ 880 train 4.056929 (lr=5.8869e-04) (hash(x)=21812805)
109
+ 890 train 4.155994 (lr=5.8729e-04) (hash(x)=23266756)
110
+ 900 val loss 4.0417
111
+ 900 val perplexity 56.9229
112
+ 900 train 4.016778 (lr=5.8581e-04) (hash(x)=17198172)
113
+ 910 train 3.915779 (lr=5.8425e-04) (hash(x)=17654181)
114
+ 920 train 4.011151 (lr=5.8262e-04) (hash(x)=18651087)
115
+ 930 train 3.970577 (lr=5.8090e-04) (hash(x)=18812778)
116
+ 940 train 3.877449 (lr=5.7911e-04) (hash(x)=19229059)
117
+ 950 train 3.995198 (lr=5.7723e-04) (hash(x)=16244689)
118
+ 960 train 3.949887 (lr=5.7529e-04) (hash(x)=14252464)
119
+ 970 train 3.863349 (lr=5.7326e-04) (hash(x)=20812699)
120
+ 980 train 4.040447 (lr=5.7116e-04) (hash(x)=19274156)
121
+ 990 train 3.902785 (lr=5.6899e-04) (hash(x)=23893376)
122
+ 1000 val loss 3.9364
123
+ 1000 val perplexity 51.2362
124
+ 1000 train 3.793566 (lr=5.6674e-04) (hash(x)=20084739)
125
+ 1010 train 3.875041 (lr=5.6442e-04) (hash(x)=20537558)
126
+ 1020 train 3.964690 (lr=5.6202e-04) (hash(x)=20432294)
127
+ 1030 train 3.875535 (lr=5.5956e-04) (hash(x)=22466028)
128
+ 1040 train 3.844952 (lr=5.5702e-04) (hash(x)=17894926)
129
+ 1050 train 3.941573 (lr=5.5441e-04) (hash(x)=21738864)
130
+ 1060 train 3.860307 (lr=5.5174e-04) (hash(x)=18724801)
131
+ 1070 train 3.818421 (lr=5.4899e-04) (hash(x)=21845067)
132
+ 1080 train 3.775688 (lr=5.4618e-04) (hash(x)=17269002)
133
+ 1090 train 3.900982 (lr=5.4330e-04) (hash(x)=20777124)
134
+ 1100 val loss 3.8591
135
+ 1100 val perplexity 47.4222
136
+ 1100 train 3.829301 (lr=5.4035e-04) (hash(x)=21658646)
137
+ 1110 train 3.802896 (lr=5.3734e-04) (hash(x)=25217492)
138
+ 1120 train 3.866534 (lr=5.3426e-04) (hash(x)=17597309)
139
+ 1130 train 3.845275 (lr=5.3113e-04) (hash(x)=15251388)
140
+ 1140 train 3.831608 (lr=5.2792e-04) (hash(x)=16526754)
141
+ 1150 train 3.791356 (lr=5.2466e-04) (hash(x)=16839845)
142
+ 1160 train 3.805662 (lr=5.2134e-04) (hash(x)=19271941)
143
+ 1170 train 3.794544 (lr=5.1796e-04) (hash(x)=20159915)
144
+ 1180 train 3.760350 (lr=5.1452e-04) (hash(x)=18768657)
145
+ 1190 train 3.805496 (lr=5.1102e-04) (hash(x)=19691298)
146
+ 1200 val loss 3.7893
147
+ 1200 val perplexity 44.2264
148
+ 1200 train 3.802570 (lr=5.0746e-04) (hash(x)=20619530)
149
+ 1210 train 3.739477 (lr=5.0386e-04) (hash(x)=20327868)
150
+ 1220 train 3.728197 (lr=5.0019e-04) (hash(x)=17297005)
151
+ 1230 train 3.749378 (lr=4.9648e-04) (hash(x)=22743384)
152
+ 1240 train 3.752196 (lr=4.9271e-04) (hash(x)=19764130)
153
+ 1250 train 3.657744 (lr=4.8889e-04) (hash(x)=19685029)
154
+ 1260 train 3.719430 (lr=4.8503e-04) (hash(x)=21621842)
155
+ 1270 train 3.723538 (lr=4.8111e-04) (hash(x)=21024799)
156
+ 1280 train 3.727242 (lr=4.7715e-04) (hash(x)=20459305)
157
+ 1290 train 3.703695 (lr=4.7315e-04) (hash(x)=21945208)
158
+ 1300 val loss 3.7467
159
+ 1300 val perplexity 42.3791
160
+ 1300 train 3.756041 (lr=4.6909e-04) (hash(x)=19421909)
161
+ 1310 train 3.678718 (lr=4.6500e-04) (hash(x)=22065431)
162
+ 1320 train 3.651360 (lr=4.6086e-04) (hash(x)=18310987)
163
+ 1330 train 3.748827 (lr=4.5669e-04) (hash(x)=18046899)
164
+ 1340 train 3.665390 (lr=4.5247e-04) (hash(x)=19877417)
165
+ 1350 train 3.804711 (lr=4.4822e-04) (hash(x)=21157293)
166
+ 1360 train 3.681175 (lr=4.4393e-04) (hash(x)=20260598)
167
+ 1370 train 3.683757 (lr=4.3960e-04) (hash(x)=20189167)
168
+ 1380 train 3.673514 (lr=4.3524e-04) (hash(x)=22193383)
169
+ 1390 train 3.670562 (lr=4.3085e-04) (hash(x)=16407046)
170
+ 1400 val loss 3.6948
171
+ 1400 val perplexity 40.2357
172
+ 1400 train 3.683640 (lr=4.2643e-04) (hash(x)=22770666)
173
+ 1410 train 3.701669 (lr=4.2197e-04) (hash(x)=22418421)
174
+ 1420 train 3.766599 (lr=4.1749e-04) (hash(x)=17883057)
175
+ 1430 train 3.603860 (lr=4.1298e-04) (hash(x)=18408864)
176
+ 1440 train 3.593240 (lr=4.0845e-04) (hash(x)=19699467)
177
+ 1450 train 3.656230 (lr=4.0389e-04) (hash(x)=21354581)
178
+ 1460 train 3.682720 (lr=3.9931e-04) (hash(x)=20422335)
179
+ 1470 train 3.585140 (lr=3.9470e-04) (hash(x)=18953370)
180
+ 1480 train 3.695054 (lr=3.9008e-04) (hash(x)=23442747)
181
+ 1490 train 3.741323 (lr=3.8544e-04) (hash(x)=20126187)
182
+ 1500 val loss 3.6550
183
+ 1500 val perplexity 38.6672
184
+ 1500 train 3.682213 (lr=3.8078e-04) (hash(x)=20342754)
185
+ 1510 train 3.641150 (lr=3.7610e-04) (hash(x)=19036313)
186
+ 1520 train 3.687203 (lr=3.7142e-04) (hash(x)=22164007)
187
+ 1530 train 3.654878 (lr=3.6671e-04) (hash(x)=24857054)
188
+ 1540 train 3.623704 (lr=3.6200e-04) (hash(x)=21281762)
189
+ 1550 train 3.582209 (lr=3.5728e-04) (hash(x)=19650272)
190
+ 1560 train 3.684850 (lr=3.5255e-04) (hash(x)=20137905)
191
+ 1570 train 3.521481 (lr=3.4781e-04) (hash(x)=14416059)
192
+ 1580 train 3.653732 (lr=3.4306e-04) (hash(x)=23684747)
193
+ 1590 train 3.496739 (lr=3.3831e-04) (hash(x)=20092448)
194
+ 1600 val loss 3.6176
195
+ 1600 val perplexity 37.2491
196
+ 1600 train 3.579444 (lr=3.3356e-04) (hash(x)=18725986)
197
+ 1610 train 3.565784 (lr=3.2881e-04) (hash(x)=27357245)
198
+ 1620 train 3.479879 (lr=3.2406e-04) (hash(x)=20033464)
199
+ 1630 train 3.708945 (lr=3.1931e-04) (hash(x)=20930109)
200
+ 1640 train 3.647739 (lr=3.1456e-04) (hash(x)=19398603)
201
+ 1650 train 3.683493 (lr=3.0982e-04) (hash(x)=20959448)
202
+ 1660 train 3.402952 (lr=3.0509e-04) (hash(x)=19692923)
203
+ 1670 train 3.660202 (lr=3.0036e-04) (hash(x)=23448284)
204
+ 1680 train 3.650345 (lr=2.9564e-04) (hash(x)=17968371)
205
+ 1690 train 3.579404 (lr=2.9093e-04) (hash(x)=15669180)
206
+ 1700 val loss 3.5855
207
+ 1700 val perplexity 36.0696
208
+ 1700 train 3.393665 (lr=2.8624e-04) (hash(x)=17816072)
209
+ 1710 train 3.593935 (lr=2.8156e-04) (hash(x)=19307466)
210
+ 1720 train 3.637766 (lr=2.7689e-04) (hash(x)=22978699)
211
+ 1730 train 3.606547 (lr=2.7224e-04) (hash(x)=23143048)
212
+ 1740 train 3.479164 (lr=2.6761e-04) (hash(x)=18722687)
213
+ 1750 train 3.409755 (lr=2.6299e-04) (hash(x)=18477212)
214
+ 1760 train 3.590554 (lr=2.5840e-04) (hash(x)=19884708)
215
+ 1770 train 3.603837 (lr=2.5383e-04) (hash(x)=20256662)
216
+ 1780 train 3.572342 (lr=2.4928e-04) (hash(x)=18622247)
217
+ 1790 train 3.535561 (lr=2.4476e-04) (hash(x)=21400648)
218
+ 1800 val loss 3.5610
219
+ 1800 val perplexity 35.2001
220
+ 1800 train 3.362972 (lr=2.4026e-04) (hash(x)=23156122)
221
+ 1810 train 3.604962 (lr=2.3580e-04) (hash(x)=13466715)
222
+ 1820 train 3.534439 (lr=2.3136e-04) (hash(x)=18810586)
223
+ 1830 train 3.533193 (lr=2.2695e-04) (hash(x)=20834847)
224
+ 1840 train 3.355483 (lr=2.2257e-04) (hash(x)=14702111)
225
+ 1850 train 3.590870 (lr=2.1823e-04) (hash(x)=18470456)
226
+ 1860 train 3.582299 (lr=2.1392e-04) (hash(x)=20701741)
227
+ 1870 train 3.515582 (lr=2.0965e-04) (hash(x)=19669016)
228
+ 1880 train 3.496598 (lr=2.0542e-04) (hash(x)=20381930)
229
+ 1890 train 3.273951 (lr=2.0122e-04) (hash(x)=21303953)
230
+ 1900 val loss 3.5348
231
+ 1900 val perplexity 34.2885
232
+ 1900 train 3.557092 (lr=1.9706e-04) (hash(x)=22119135)
233
+ 1910 train 3.610139 (lr=1.9295e-04) (hash(x)=19343834)
234
+ 1920 train 3.508832 (lr=1.8887e-04) (hash(x)=20131130)
235
+ 1930 train 3.357190 (lr=1.8485e-04) (hash(x)=21254128)
236
+ 1940 train 3.516062 (lr=1.8086e-04) (hash(x)=20816939)
237
+ 1950 train 3.559673 (lr=1.7692e-04) (hash(x)=19820308)
238
+ 1960 train 3.556221 (lr=1.7303e-04) (hash(x)=28009259)
239
+ 1970 train 3.459218 (lr=1.6919e-04) (hash(x)=19370262)
240
+ 1980 train 3.281868 (lr=1.6540e-04) (hash(x)=19401165)
241
+ 1990 train 3.559883 (lr=1.6166e-04) (hash(x)=18628676)
242
+ 2000 val loss 3.5102
243
+ 2000 val perplexity 33.4536
244
+ 2000 train 3.556651 (lr=1.5797e-04) (hash(x)=20151188)
245
+ 2010 train 3.560676 (lr=1.5433e-04) (hash(x)=18053856)
246
+ 2020 train 3.555864 (lr=1.5075e-04) (hash(x)=20481771)
247
+ 2030 train 3.581336 (lr=1.4723e-04) (hash(x)=18607449)
248
+ 2040 train 3.465315 (lr=1.4376e-04) (hash(x)=21897936)
249
+ 2050 train 3.493999 (lr=1.4035e-04) (hash(x)=19648528)
250
+ 2060 train 3.518195 (lr=1.3699e-04) (hash(x)=21541784)
251
+ 2070 train 3.538033 (lr=1.3370e-04) (hash(x)=19466104)
252
+ 2080 train 3.507005 (lr=1.3047e-04) (hash(x)=19540614)
253
+ 2090 train 3.487651 (lr=1.2730e-04) (hash(x)=18783653)
254
+ 2100 val loss 3.4960
255
+ 2100 val perplexity 32.9817
256
+ 2100 train 3.423492 (lr=1.2419e-04) (hash(x)=30861019)
257
+ 2110 train 3.449903 (lr=1.2115e-04) (hash(x)=20187147)
258
+ 2120 train 3.550241 (lr=1.1817e-04) (hash(x)=17350218)
259
+ 2130 train 3.589391 (lr=1.1525e-04) (hash(x)=21668918)
260
+ 2140 train 3.452075 (lr=1.1241e-04) (hash(x)=16953068)
261
+ 2150 train 3.563969 (lr=1.0963e-04) (hash(x)=22435234)
262
+ 2160 train 3.457865 (lr=1.0692e-04) (hash(x)=14277862)
263
+ 2170 train 3.532546 (lr=1.0427e-04) (hash(x)=21051545)
264
+ 2180 train 3.505602 (lr=1.0170e-04) (hash(x)=16728289)
265
+ 2190 train 3.509662 (lr=9.9199e-05) (hash(x)=18519101)
266
+ 2200 val loss 3.4758
267
+ 2200 val perplexity 32.3226
268
+ 2200 train 3.510487 (lr=9.6769e-05) (hash(x)=19727249)
269
+ 2210 train 3.360144 (lr=9.4412e-05) (hash(x)=18890853)
270
+ 2220 train 3.518687 (lr=9.2127e-05) (hash(x)=20660564)
271
+ 2230 train 3.533267 (lr=8.9916e-05) (hash(x)=22774110)
272
+ 2240 train 3.499394 (lr=8.7779e-05) (hash(x)=19389926)
273
+ 2250 train 3.532896 (lr=8.5717e-05) (hash(x)=17041691)
274
+ 2260 train 3.413473 (lr=8.3731e-05) (hash(x)=20807441)
275
+ 2270 train 3.540822 (lr=8.1821e-05) (hash(x)=22538437)
276
+ 2280 train 3.563267 (lr=7.9988e-05) (hash(x)=21551546)
277
+ 2290 train 3.437593 (lr=7.8232e-05) (hash(x)=24167699)
278
+ 2300 val loss 3.4634
279
+ 2300 val perplexity 31.9268
280
+ 2300 train 3.348315 (lr=7.6555e-05) (hash(x)=16362157)
281
+ 2310 train 3.321218 (lr=7.4956e-05) (hash(x)=23722796)
282
+ 2320 train 3.548417 (lr=7.3436e-05) (hash(x)=18346470)
283
+ 2330 train 3.499416 (lr=7.1995e-05) (hash(x)=17096354)
284
+ 2340 train 3.500489 (lr=7.0635e-05) (hash(x)=23151918)
285
+ 2350 train 3.315239 (lr=6.9354e-05) (hash(x)=17604704)
286
+ 2360 train 3.455883 (lr=6.8155e-05) (hash(x)=20390213)
287
+ 2370 train 3.502772 (lr=6.7036e-05) (hash(x)=24875449)
288
+ 2380 train 3.524223 (lr=6.5999e-05) (hash(x)=21185640)
289
+ 2390 train 3.458211 (lr=6.5044e-05) (hash(x)=14839939)
290
+ 2400 val loss 3.4516
291
+ 2400 val perplexity 31.5520
292
+ 2400 train 3.525647 (lr=6.4171e-05) (hash(x)=17498688)
293
+ 2410 train 3.393266 (lr=6.3380e-05) (hash(x)=20544025)
294
+ 2420 train 3.402240 (lr=6.2672e-05) (hash(x)=21093275)
295
+ 2430 train 3.387079 (lr=6.2046e-05) (hash(x)=19270991)
296
+ 2440 train 3.512996 (lr=6.1504e-05) (hash(x)=17536544)
297
+ 2450 train 3.443870 (lr=6.1045e-05) (hash(x)=16978919)
298
+ 2460 train 3.446483 (lr=6.0669e-05) (hash(x)=18893211)
299
+ 2470 train 3.424699 (lr=6.0376e-05) (hash(x)=17072230)
300
+ 2480 train 3.343539 (lr=6.0167e-05) (hash(x)=21601992)
301
+ 2490 train 3.478189 (lr=6.0042e-05) (hash(x)=23385916)
302
+ 2499 val loss 3.4453
303
+ 2499 val perplexity 31.3542
allowing_more_selection_patterns/one_mask_per_head_1_latent_vector_degenerate_seed_1338/model_02499.pt ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:a56a8848993993ef75b1d1baceafcdd3a8bb0ff4a5dbb75ffb43f54bf93b450d
3
+ size 555244826
allowing_more_selection_patterns/one_mask_per_head_1_latent_vector_degenerate_seed_1338/optimizer_02499.pt ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:301236b9c726bc3b495839bca87caee7ae5a5a2d1d6925f60cc8b0c83ea8d731
3
+ size 1009837470