andrew-healey commited on
Commit
abc9d47
·
verified ·
1 Parent(s): 12eda2c

Upload folder using huggingface_hub

Browse files
allowing_more_selection_patterns/one_mask_per_head_2_latent_vectors/args.json ADDED
@@ -0,0 +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, "disable_selection_head_linear_combo_bias": false, "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": 2, "init_latent_masks_to_identity": 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_2_latent_vectors"}
allowing_more_selection_patterns/one_mask_per_head_2_latent_vectors/dataloader_02499.pt ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:8f23ea867bba43cdf20aa2c4d5d2e2d1d662882792de37f23ee6c049d369660a
3
+ size 964
allowing_more_selection_patterns/one_mask_per_head_2_latent_vectors/log2.txt ADDED
@@ -0,0 +1,303 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ max_steps: 2500
2
+ 0 val loss 11.0210
3
+ 0 val perplexity 61145.3125
4
+ 0 train 11.026853 (lr=8.3916e-07) (hash(x)=23347629)
5
+ 10 train 10.224984 (lr=9.2308e-06) (hash(x)=19636655)
6
+ 20 train 9.728956 (lr=1.7622e-05) (hash(x)=24490248)
7
+ 30 train 9.528360 (lr=2.6014e-05) (hash(x)=21733558)
8
+ 40 train 9.274856 (lr=3.4406e-05) (hash(x)=31750616)
9
+ 50 train 8.962066 (lr=4.2797e-05) (hash(x)=20041895)
10
+ 60 train 8.584548 (lr=5.1189e-05) (hash(x)=21416280)
11
+ 70 train 8.241666 (lr=5.9580e-05) (hash(x)=18066780)
12
+ 80 train 7.923881 (lr=6.7972e-05) (hash(x)=21809427)
13
+ 90 train 7.676556 (lr=7.6364e-05) (hash(x)=22061595)
14
+ 100 val loss 7.4615
15
+ 100 val perplexity 1739.7021
16
+ 100 train 7.412017 (lr=8.4755e-05) (hash(x)=17468914)
17
+ 110 train 7.220309 (lr=9.3147e-05) (hash(x)=20421520)
18
+ 120 train 7.222252 (lr=1.0154e-04) (hash(x)=23140133)
19
+ 130 train 7.086216 (lr=1.0993e-04) (hash(x)=20006835)
20
+ 140 train 6.857571 (lr=1.1832e-04) (hash(x)=18628134)
21
+ 150 train 6.826988 (lr=1.2671e-04) (hash(x)=21913447)
22
+ 160 train 6.785416 (lr=1.3510e-04) (hash(x)=22997297)
23
+ 170 train 6.582784 (lr=1.4350e-04) (hash(x)=17904821)
24
+ 180 train 6.439573 (lr=1.5189e-04) (hash(x)=18826982)
25
+ 190 train 6.520911 (lr=1.6028e-04) (hash(x)=21387716)
26
+ 200 val loss 6.3877
27
+ 200 val perplexity 594.5059
28
+ 200 train 6.310735 (lr=1.6867e-04) (hash(x)=16236395)
29
+ 210 train 6.192862 (lr=1.7706e-04) (hash(x)=21428681)
30
+ 220 train 6.331400 (lr=1.8545e-04) (hash(x)=20461738)
31
+ 230 train 6.163535 (lr=1.9385e-04) (hash(x)=20394865)
32
+ 240 train 6.053408 (lr=2.0224e-04) (hash(x)=18028671)
33
+ 250 train 5.912908 (lr=2.1063e-04) (hash(x)=18122073)
34
+ 260 train 5.824219 (lr=2.1902e-04) (hash(x)=22994143)
35
+ 270 train 6.008641 (lr=2.2741e-04) (hash(x)=20646260)
36
+ 280 train 5.856314 (lr=2.3580e-04) (hash(x)=23204708)
37
+ 290 train 5.768899 (lr=2.4420e-04) (hash(x)=17900252)
38
+ 300 val loss 5.8088
39
+ 300 val perplexity 333.2105
40
+ 300 train 5.649310 (lr=2.5259e-04) (hash(x)=19073819)
41
+ 310 train 5.817078 (lr=2.6098e-04) (hash(x)=22378232)
42
+ 320 train 5.723641 (lr=2.6937e-04) (hash(x)=19910430)
43
+ 330 train 5.644930 (lr=2.7776e-04) (hash(x)=25168753)
44
+ 340 train 5.543560 (lr=2.8615e-04) (hash(x)=20665256)
45
+ 350 train 5.413407 (lr=2.9455e-04) (hash(x)=19951160)
46
+ 360 train 5.620712 (lr=3.0294e-04) (hash(x)=21178202)
47
+ 370 train 5.493414 (lr=3.1133e-04) (hash(x)=20435886)
48
+ 380 train 5.415110 (lr=3.1972e-04) (hash(x)=21255097)
49
+ 390 train 5.419855 (lr=3.2811e-04) (hash(x)=19596330)
50
+ 400 val loss 5.3653
51
+ 400 val perplexity 213.8510
52
+ 400 train 5.392876 (lr=3.3650e-04) (hash(x)=21789426)
53
+ 410 train 5.303652 (lr=3.4490e-04) (hash(x)=18366401)
54
+ 420 train 5.186001 (lr=3.5329e-04) (hash(x)=18614660)
55
+ 430 train 5.331109 (lr=3.6168e-04) (hash(x)=21649063)
56
+ 440 train 5.207737 (lr=3.7007e-04) (hash(x)=18505097)
57
+ 450 train 5.198619 (lr=3.7846e-04) (hash(x)=23434312)
58
+ 460 train 5.099370 (lr=3.8685e-04) (hash(x)=16584569)
59
+ 470 train 4.970296 (lr=3.9524e-04) (hash(x)=13447111)
60
+ 480 train 5.178278 (lr=4.0364e-04) (hash(x)=22645130)
61
+ 490 train 5.028224 (lr=4.1203e-04) (hash(x)=19362777)
62
+ 500 val loss 5.0263
63
+ 500 val perplexity 152.3755
64
+ 500 train 4.961761 (lr=4.2042e-04) (hash(x)=20921874)
65
+ 510 train 4.832655 (lr=4.2881e-04) (hash(x)=22123048)
66
+ 520 train 4.873633 (lr=4.3720e-04) (hash(x)=21215368)
67
+ 530 train 5.016727 (lr=4.4559e-04) (hash(x)=20717936)
68
+ 540 train 4.863809 (lr=4.5399e-04) (hash(x)=20118007)
69
+ 550 train 4.740807 (lr=4.6238e-04) (hash(x)=21402522)
70
+ 560 train 4.679495 (lr=4.7077e-04) (hash(x)=19057180)
71
+ 570 train 4.865189 (lr=4.7916e-04) (hash(x)=19360553)
72
+ 580 train 4.761452 (lr=4.8755e-04) (hash(x)=18194693)
73
+ 590 train 4.670934 (lr=4.9594e-04) (hash(x)=19863826)
74
+ 600 val loss 4.6322
75
+ 600 val perplexity 102.7406
76
+ 600 train 4.586124 (lr=5.0434e-04) (hash(x)=21794551)
77
+ 610 train 4.578928 (lr=5.1273e-04) (hash(x)=18685622)
78
+ 620 train 4.445580 (lr=5.2112e-04) (hash(x)=23473104)
79
+ 630 train 4.542370 (lr=5.2951e-04) (hash(x)=18671672)
80
+ 640 train 4.556378 (lr=5.3790e-04) (hash(x)=19234240)
81
+ 650 train 4.471832 (lr=5.4629e-04) (hash(x)=23183903)
82
+ 660 train 4.310189 (lr=5.5469e-04) (hash(x)=23665543)
83
+ 670 train 4.470248 (lr=5.6308e-04) (hash(x)=22055313)
84
+ 680 train 4.437671 (lr=5.7147e-04) (hash(x)=22321668)
85
+ 690 train 4.562855 (lr=5.7986e-04) (hash(x)=21568926)
86
+ 700 val loss 4.4191
87
+ 700 val perplexity 83.0237
88
+ 700 train 4.421269 (lr=5.8825e-04) (hash(x)=19783143)
89
+ 710 train 4.315081 (lr=5.9664e-04) (hash(x)=21565258)
90
+ 720 train 4.399453 (lr=5.9999e-04) (hash(x)=16127212)
91
+ 730 train 4.335420 (lr=5.9991e-04) (hash(x)=19658174)
92
+ 740 train 4.297725 (lr=5.9974e-04) (hash(x)=16454319)
93
+ 750 train 4.251865 (lr=5.9949e-04) (hash(x)=17993154)
94
+ 760 train 4.150050 (lr=5.9915e-04) (hash(x)=19256989)
95
+ 770 train 4.126282 (lr=5.9874e-04) (hash(x)=17351653)
96
+ 780 train 4.153135 (lr=5.9824e-04) (hash(x)=21626120)
97
+ 790 train 4.297883 (lr=5.9765e-04) (hash(x)=16367412)
98
+ 800 val loss 4.2014
99
+ 800 val perplexity 66.7825
100
+ 800 train 4.154687 (lr=5.9698e-04) (hash(x)=22321504)
101
+ 810 train 4.044846 (lr=5.9623e-04) (hash(x)=19869755)
102
+ 820 train 4.224068 (lr=5.9540e-04) (hash(x)=24183665)
103
+ 830 train 4.188612 (lr=5.9449e-04) (hash(x)=19194113)
104
+ 840 train 4.031402 (lr=5.9349e-04) (hash(x)=17710503)
105
+ 850 train 4.038771 (lr=5.9241e-04) (hash(x)=17231851)
106
+ 860 train 4.188370 (lr=5.9126e-04) (hash(x)=20584272)
107
+ 870 train 4.070864 (lr=5.9002e-04) (hash(x)=17468790)
108
+ 880 train 4.020459 (lr=5.8869e-04) (hash(x)=22304409)
109
+ 890 train 4.103456 (lr=5.8729e-04) (hash(x)=18871567)
110
+ 900 val loss 4.0760
111
+ 900 val perplexity 58.9118
112
+ 900 train 4.050114 (lr=5.8581e-04) (hash(x)=20924720)
113
+ 910 train 3.960855 (lr=5.8425e-04) (hash(x)=21163464)
114
+ 920 train 3.895747 (lr=5.8262e-04) (hash(x)=20120862)
115
+ 930 train 4.069382 (lr=5.8090e-04) (hash(x)=19565198)
116
+ 940 train 3.996528 (lr=5.7911e-04) (hash(x)=18517540)
117
+ 950 train 3.957483 (lr=5.7723e-04) (hash(x)=18300759)
118
+ 960 train 3.995872 (lr=5.7529e-04) (hash(x)=21443001)
119
+ 970 train 3.933443 (lr=5.7326e-04) (hash(x)=15210971)
120
+ 980 train 4.097782 (lr=5.7116e-04) (hash(x)=20874579)
121
+ 990 train 4.028358 (lr=5.6899e-04) (hash(x)=20690802)
122
+ 1000 val loss 3.9740
123
+ 1000 val perplexity 53.1982
124
+ 1000 train 3.895840 (lr=5.6674e-04) (hash(x)=18635877)
125
+ 1010 train 3.997235 (lr=5.6442e-04) (hash(x)=21141462)
126
+ 1020 train 4.019194 (lr=5.6202e-04) (hash(x)=19483400)
127
+ 1030 train 3.915386 (lr=5.5956e-04) (hash(x)=24792825)
128
+ 1040 train 3.908027 (lr=5.5702e-04) (hash(x)=16743928)
129
+ 1050 train 3.947167 (lr=5.5441e-04) (hash(x)=22540052)
130
+ 1060 train 3.957279 (lr=5.5174e-04) (hash(x)=20546637)
131
+ 1070 train 3.861314 (lr=5.4899e-04) (hash(x)=13739753)
132
+ 1080 train 3.923095 (lr=5.4618e-04) (hash(x)=20071304)
133
+ 1090 train 3.875487 (lr=5.4330e-04) (hash(x)=19339656)
134
+ 1100 val loss 3.8861
135
+ 1100 val perplexity 48.7206
136
+ 1100 train 3.850466 (lr=5.4035e-04) (hash(x)=20039628)
137
+ 1110 train 3.913861 (lr=5.3734e-04) (hash(x)=19249691)
138
+ 1120 train 3.934763 (lr=5.3426e-04) (hash(x)=19048809)
139
+ 1130 train 3.860693 (lr=5.3113e-04) (hash(x)=19309000)
140
+ 1140 train 3.810267 (lr=5.2792e-04) (hash(x)=21787927)
141
+ 1150 train 3.728314 (lr=5.2466e-04) (hash(x)=19998536)
142
+ 1160 train 3.887495 (lr=5.2134e-04) (hash(x)=23724694)
143
+ 1170 train 3.901741 (lr=5.1796e-04) (hash(x)=17148204)
144
+ 1180 train 3.814177 (lr=5.1452e-04) (hash(x)=19511852)
145
+ 1190 train 3.789210 (lr=5.1102e-04) (hash(x)=19352862)
146
+ 1200 val loss 3.8142
147
+ 1200 val perplexity 45.3423
148
+ 1200 train 3.728332 (lr=5.0746e-04) (hash(x)=19530144)
149
+ 1210 train 3.911057 (lr=5.0386e-04) (hash(x)=19897392)
150
+ 1220 train 3.814682 (lr=5.0019e-04) (hash(x)=24402963)
151
+ 1230 train 3.728233 (lr=4.9648e-04) (hash(x)=18589754)
152
+ 1240 train 3.735929 (lr=4.9271e-04) (hash(x)=21951750)
153
+ 1250 train 3.813010 (lr=4.8889e-04) (hash(x)=18611379)
154
+ 1260 train 3.825431 (lr=4.8503e-04) (hash(x)=20338521)
155
+ 1270 train 3.749419 (lr=4.8111e-04) (hash(x)=18308405)
156
+ 1280 train 3.751331 (lr=4.7715e-04) (hash(x)=20920945)
157
+ 1290 train 3.710641 (lr=4.7315e-04) (hash(x)=20094152)
158
+ 1300 val loss 3.7573
159
+ 1300 val perplexity 42.8321
160
+ 1300 train 3.830754 (lr=4.6909e-04) (hash(x)=19003947)
161
+ 1310 train 3.789129 (lr=4.6500e-04) (hash(x)=19356356)
162
+ 1320 train 3.740087 (lr=4.6086e-04) (hash(x)=19390343)
163
+ 1330 train 3.676285 (lr=4.5669e-04) (hash(x)=17535982)
164
+ 1340 train 3.698289 (lr=4.5247e-04) (hash(x)=21959104)
165
+ 1350 train 3.798099 (lr=4.4822e-04) (hash(x)=21982099)
166
+ 1360 train 3.760369 (lr=4.4393e-04) (hash(x)=17176243)
167
+ 1370 train 3.703591 (lr=4.3960e-04) (hash(x)=21223627)
168
+ 1380 train 3.721731 (lr=4.3524e-04) (hash(x)=22462931)
169
+ 1390 train 3.745908 (lr=4.3085e-04) (hash(x)=19153464)
170
+ 1400 val loss 3.7049
171
+ 1400 val perplexity 40.6480
172
+ 1400 train 3.637436 (lr=4.2643e-04) (hash(x)=20146496)
173
+ 1410 train 3.693980 (lr=4.2197e-04) (hash(x)=21781327)
174
+ 1420 train 3.764289 (lr=4.1749e-04) (hash(x)=19234051)
175
+ 1430 train 3.811645 (lr=4.1298e-04) (hash(x)=20769898)
176
+ 1440 train 3.661507 (lr=4.0845e-04) (hash(x)=22042430)
177
+ 1450 train 3.685185 (lr=4.0389e-04) (hash(x)=20527387)
178
+ 1460 train 3.669461 (lr=3.9931e-04) (hash(x)=18925766)
179
+ 1470 train 3.667918 (lr=3.9470e-04) (hash(x)=18198851)
180
+ 1480 train 3.740486 (lr=3.9008e-04) (hash(x)=21847281)
181
+ 1490 train 3.709109 (lr=3.8544e-04) (hash(x)=15645274)
182
+ 1500 val loss 3.6558
183
+ 1500 val perplexity 38.6981
184
+ 1500 train 3.577570 (lr=3.8078e-04) (hash(x)=14693370)
185
+ 1510 train 3.590966 (lr=3.7610e-04) (hash(x)=21810857)
186
+ 1520 train 3.686657 (lr=3.7142e-04) (hash(x)=18980052)
187
+ 1530 train 3.711357 (lr=3.6671e-04) (hash(x)=22868842)
188
+ 1540 train 3.624825 (lr=3.6200e-04) (hash(x)=17289166)
189
+ 1550 train 3.708869 (lr=3.5728e-04) (hash(x)=18109373)
190
+ 1560 train 3.744670 (lr=3.5255e-04) (hash(x)=22377619)
191
+ 1570 train 3.566229 (lr=3.4781e-04) (hash(x)=20307182)
192
+ 1580 train 3.716479 (lr=3.4306e-04) (hash(x)=20391291)
193
+ 1590 train 3.582519 (lr=3.3831e-04) (hash(x)=18690939)
194
+ 1600 val loss 3.6185
195
+ 1600 val perplexity 37.2818
196
+ 1600 train 3.595604 (lr=3.3356e-04) (hash(x)=16719886)
197
+ 1610 train 3.589139 (lr=3.2881e-04) (hash(x)=16040840)
198
+ 1620 train 3.661969 (lr=3.2406e-04) (hash(x)=25795140)
199
+ 1630 train 3.667035 (lr=3.1931e-04) (hash(x)=21936964)
200
+ 1640 train 3.563393 (lr=3.1456e-04) (hash(x)=11593027)
201
+ 1650 train 3.621785 (lr=3.0982e-04) (hash(x)=23256718)
202
+ 1660 train 3.581396 (lr=3.0509e-04) (hash(x)=20289140)
203
+ 1670 train 3.587987 (lr=3.0036e-04) (hash(x)=18382509)
204
+ 1680 train 3.651867 (lr=2.9564e-04) (hash(x)=26604329)
205
+ 1690 train 3.619584 (lr=2.9093e-04) (hash(x)=22122534)
206
+ 1700 val loss 3.5865
207
+ 1700 val perplexity 36.1081
208
+ 1700 train 3.634608 (lr=2.8624e-04) (hash(x)=22884812)
209
+ 1710 train 3.579048 (lr=2.8156e-04) (hash(x)=18598496)
210
+ 1720 train 3.600675 (lr=2.7689e-04) (hash(x)=22147447)
211
+ 1730 train 3.604797 (lr=2.7224e-04) (hash(x)=18912135)
212
+ 1740 train 3.442914 (lr=2.6761e-04) (hash(x)=21952974)
213
+ 1750 train 3.555899 (lr=2.6299e-04) (hash(x)=18390978)
214
+ 1760 train 3.566892 (lr=2.5840e-04) (hash(x)=17979288)
215
+ 1770 train 3.572840 (lr=2.5383e-04) (hash(x)=18979099)
216
+ 1780 train 3.535272 (lr=2.4928e-04) (hash(x)=20511922)
217
+ 1790 train 3.591322 (lr=2.4476e-04) (hash(x)=21814060)
218
+ 1800 val loss 3.5603
219
+ 1800 val perplexity 35.1750
220
+ 1800 train 3.579409 (lr=2.4026e-04) (hash(x)=18076243)
221
+ 1810 train 3.568480 (lr=2.3580e-04) (hash(x)=19721209)
222
+ 1820 train 3.514518 (lr=2.3136e-04) (hash(x)=18140847)
223
+ 1830 train 3.577227 (lr=2.2695e-04) (hash(x)=22065413)
224
+ 1840 train 3.477827 (lr=2.2257e-04) (hash(x)=21166111)
225
+ 1850 train 3.475256 (lr=2.1823e-04) (hash(x)=21904302)
226
+ 1860 train 3.543043 (lr=2.1392e-04) (hash(x)=18283828)
227
+ 1870 train 3.500759 (lr=2.0965e-04) (hash(x)=20892773)
228
+ 1880 train 3.528642 (lr=2.0542e-04) (hash(x)=17781264)
229
+ 1890 train 3.497387 (lr=2.0122e-04) (hash(x)=20612962)
230
+ 1900 val loss 3.5351
231
+ 1900 val perplexity 34.2994
232
+ 1900 train 3.589784 (lr=1.9706e-04) (hash(x)=20421058)
233
+ 1910 train 3.562781 (lr=1.9295e-04) (hash(x)=18554808)
234
+ 1920 train 3.518567 (lr=1.8887e-04) (hash(x)=18189279)
235
+ 1930 train 3.535643 (lr=1.8485e-04) (hash(x)=19923279)
236
+ 1940 train 3.500811 (lr=1.8086e-04) (hash(x)=21665511)
237
+ 1950 train 3.521092 (lr=1.7692e-04) (hash(x)=22038489)
238
+ 1960 train 3.494725 (lr=1.7303e-04) (hash(x)=18896956)
239
+ 1970 train 3.522476 (lr=1.6919e-04) (hash(x)=20408383)
240
+ 1980 train 3.504655 (lr=1.6540e-04) (hash(x)=23039940)
241
+ 1990 train 3.639238 (lr=1.6166e-04) (hash(x)=20631172)
242
+ 2000 val loss 3.5147
243
+ 2000 val perplexity 33.6047
244
+ 2000 train 3.452120 (lr=1.5797e-04) (hash(x)=20206106)
245
+ 2010 train 3.547992 (lr=1.5433e-04) (hash(x)=21864020)
246
+ 2020 train 3.529054 (lr=1.5075e-04) (hash(x)=21391128)
247
+ 2030 train 3.496371 (lr=1.4723e-04) (hash(x)=18676636)
248
+ 2040 train 3.543153 (lr=1.4376e-04) (hash(x)=15987134)
249
+ 2050 train 3.534734 (lr=1.4035e-04) (hash(x)=23940769)
250
+ 2060 train 3.514200 (lr=1.3699e-04) (hash(x)=17152840)
251
+ 2070 train 3.482354 (lr=1.3370e-04) (hash(x)=22958445)
252
+ 2080 train 3.536086 (lr=1.3047e-04) (hash(x)=23616724)
253
+ 2090 train 3.504999 (lr=1.2730e-04) (hash(x)=16571283)
254
+ 2100 val loss 3.4923
255
+ 2100 val perplexity 32.8622
256
+ 2100 train 3.501917 (lr=1.2419e-04) (hash(x)=19070327)
257
+ 2110 train 3.547339 (lr=1.2115e-04) (hash(x)=20758626)
258
+ 2120 train 3.491140 (lr=1.1817e-04) (hash(x)=18397810)
259
+ 2130 train 3.479162 (lr=1.1525e-04) (hash(x)=18045234)
260
+ 2140 train 3.583100 (lr=1.1241e-04) (hash(x)=18969565)
261
+ 2150 train 3.519750 (lr=1.0963e-04) (hash(x)=17149701)
262
+ 2160 train 3.441148 (lr=1.0692e-04) (hash(x)=23075443)
263
+ 2170 train 3.474863 (lr=1.0427e-04) (hash(x)=17977337)
264
+ 2180 train 3.535408 (lr=1.0170e-04) (hash(x)=20657407)
265
+ 2190 train 3.523651 (lr=9.9199e-05) (hash(x)=19985933)
266
+ 2200 val loss 3.4774
267
+ 2200 val perplexity 32.3740
268
+ 2200 train 3.474890 (lr=9.6769e-05) (hash(x)=19025387)
269
+ 2210 train 3.492535 (lr=9.4412e-05) (hash(x)=23318724)
270
+ 2220 train 3.563749 (lr=9.2127e-05) (hash(x)=19308039)
271
+ 2230 train 3.406103 (lr=8.9916e-05) (hash(x)=20733682)
272
+ 2240 train 3.504362 (lr=8.7779e-05) (hash(x)=18519560)
273
+ 2250 train 3.509279 (lr=8.5717e-05) (hash(x)=20299444)
274
+ 2260 train 3.465725 (lr=8.3731e-05) (hash(x)=18819615)
275
+ 2270 train 3.498842 (lr=8.1821e-05) (hash(x)=18743705)
276
+ 2280 train 3.484602 (lr=7.9988e-05) (hash(x)=17190793)
277
+ 2290 train 3.520266 (lr=7.8232e-05) (hash(x)=23126906)
278
+ 2300 val loss 3.4648
279
+ 2300 val perplexity 31.9692
280
+ 2300 train 3.453353 (lr=7.6555e-05) (hash(x)=24812267)
281
+ 2310 train 3.438114 (lr=7.4956e-05) (hash(x)=20147105)
282
+ 2320 train 3.554063 (lr=7.3436e-05) (hash(x)=24619936)
283
+ 2330 train 3.419701 (lr=7.1995e-05) (hash(x)=17730658)
284
+ 2340 train 3.396029 (lr=7.0635e-05) (hash(x)=19345731)
285
+ 2350 train 3.433727 (lr=6.9354e-05) (hash(x)=26415058)
286
+ 2360 train 3.467302 (lr=6.8155e-05) (hash(x)=18622999)
287
+ 2370 train 3.400041 (lr=6.7036e-05) (hash(x)=18848567)
288
+ 2380 train 3.483238 (lr=6.5999e-05) (hash(x)=21903360)
289
+ 2390 train 3.498673 (lr=6.5044e-05) (hash(x)=23479216)
290
+ 2400 val loss 3.4562
291
+ 2400 val perplexity 31.6967
292
+ 2400 train 3.495375 (lr=6.4171e-05) (hash(x)=19762013)
293
+ 2410 train 3.398833 (lr=6.3380e-05) (hash(x)=20622912)
294
+ 2420 train 3.517284 (lr=6.2672e-05) (hash(x)=17669226)
295
+ 2430 train 3.443905 (lr=6.2046e-05) (hash(x)=19536719)
296
+ 2440 train 3.385809 (lr=6.1504e-05) (hash(x)=19913725)
297
+ 2450 train 3.430158 (lr=6.1045e-05) (hash(x)=20782072)
298
+ 2460 train 3.398232 (lr=6.0669e-05) (hash(x)=19795002)
299
+ 2470 train 3.446356 (lr=6.0376e-05) (hash(x)=17003051)
300
+ 2480 train 3.412931 (lr=6.0167e-05) (hash(x)=21414190)
301
+ 2490 train 3.460075 (lr=6.0042e-05) (hash(x)=20943620)
302
+ 2499 val loss 3.4509
303
+ 2499 val perplexity 31.5302
allowing_more_selection_patterns/one_mask_per_head_2_latent_vectors/model_02499.pt ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:74d399e934747a47e67bd9c8047a4e495406eb1676ab43cc544c7a3ce28e64ec
3
+ size 555249458
allowing_more_selection_patterns/one_mask_per_head_2_latent_vectors/optimizer_02499.pt ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:dc6f77658f2840ea7ad86db4aab5f2adb0ac9a4c43dd2c87a1fbff046fe32f19
3
+ size 1009849270