andrew-healey commited on
Commit
c2696cf
·
verified ·
1 Parent(s): 71fbb1b

Upload folder using huggingface_hub

Browse files
one_mask_per_head_4_latent_vectors/args.json ADDED
@@ -0,0 +1 @@
 
 
1
+ {"hellaswag": true, "attention_kind": "selective", "log_dir": "one_mask_per_head_4_latent_vectors", "resume_checkpoint": null, "resume_optimizer": false, "add_a_head": false, "add_head_to_start": true, "new_head_init": "normal", "n_heads": 12, "protect_bos_token": true, "prevent_from_masking_myself": true, "max_steps": null, "warmup_steps": null, "group": "shrinking_big_runs", "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": 512, "batch_size": 16, "total_batch_size": 524288, "protection_head_scaling_factor": 1.0, "protection_head_bias": 0.0, "n_sliced_masks": null, "n_latent_masks": 4, "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": 6e-05, "decay_lr": true, "readout_zero_init": false, "query_zero_init": false, "l1_loss": false, "debugpy": false, "key": null, "n_embd": 768}
one_mask_per_head_4_latent_vectors/dataloader_02499.pt ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:55666202dc65099e5a26b66e4e3e7470de2cc1c0921ebdf41ac539630f8a4ecd
3
+ size 964
one_mask_per_head_4_latent_vectors/log2.txt ADDED
@@ -0,0 +1,303 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ max_steps: 2500
2
+ 0 val loss 10.9841
3
+ 0 val perplexity 58927.7227
4
+ 0 train 10.983159 (lr=8.3916e-08) (hash(x)=41882737)
5
+ 10 train 10.976627 (lr=9.2308e-07) (hash(x)=38752233)
6
+ 20 train 10.953777 (lr=1.7622e-06) (hash(x)=41987968)
7
+ 30 train 10.917013 (lr=2.6014e-06) (hash(x)=39173693)
8
+ 40 train 10.866720 (lr=3.4406e-06) (hash(x)=44364470)
9
+ 50 train 10.795599 (lr=4.2797e-06) (hash(x)=37915340)
10
+ 60 train 10.706337 (lr=5.1189e-06) (hash(x)=40075499)
11
+ 70 train 10.607864 (lr=5.9580e-06) (hash(x)=39209274)
12
+ 80 train 10.475047 (lr=6.7972e-06) (hash(x)=35191202)
13
+ 90 train 10.334329 (lr=7.6364e-06) (hash(x)=40296838)
14
+ 100 val loss 10.1578
15
+ 100 val perplexity 25792.2969
16
+ 100 train 10.173817 (lr=8.4755e-06) (hash(x)=38049409)
17
+ 110 train 10.016650 (lr=9.3147e-06) (hash(x)=49174314)
18
+ 120 train 9.920448 (lr=1.0154e-05) (hash(x)=39993942)
19
+ 130 train 9.839408 (lr=1.0993e-05) (hash(x)=34436824)
20
+ 140 train 9.754370 (lr=1.1832e-05) (hash(x)=35630599)
21
+ 150 train 9.717109 (lr=1.2671e-05) (hash(x)=38098135)
22
+ 160 train 9.690331 (lr=1.3510e-05) (hash(x)=46948704)
23
+ 170 train 9.683658 (lr=1.4350e-05) (hash(x)=41276821)
24
+ 180 train 9.674854 (lr=1.5189e-05) (hash(x)=37776315)
25
+ 190 train 9.691936 (lr=1.6028e-05) (hash(x)=39307311)
26
+ 200 val loss 9.6470
27
+ 200 val perplexity 15475.3564
28
+ 200 train 9.658278 (lr=1.6867e-05) (hash(x)=38965293)
29
+ 210 train 9.655929 (lr=1.7706e-05) (hash(x)=43574397)
30
+ 220 train 9.639135 (lr=1.8545e-05) (hash(x)=47662442)
31
+ 230 train 9.637926 (lr=1.9385e-05) (hash(x)=41008662)
32
+ 240 train 9.640918 (lr=2.0224e-05) (hash(x)=37003526)
33
+ 250 train 9.632831 (lr=2.1063e-05) (hash(x)=35863708)
34
+ 260 train 9.607230 (lr=2.1902e-05) (hash(x)=41433145)
35
+ 270 train 9.602919 (lr=2.2741e-05) (hash(x)=39955434)
36
+ 280 train 9.602467 (lr=2.3580e-05) (hash(x)=42053458)
37
+ 290 train 9.623379 (lr=2.4420e-05) (hash(x)=38754802)
38
+ 300 val loss 9.5845
39
+ 300 val perplexity 14536.9893
40
+ 300 train 9.580344 (lr=2.5259e-05) (hash(x)=37473643)
41
+ 310 train 9.555893 (lr=2.6098e-05) (hash(x)=41946823)
42
+ 320 train 9.586332 (lr=2.6937e-05) (hash(x)=37217925)
43
+ 330 train 9.570653 (lr=2.7776e-05) (hash(x)=48653656)
44
+ 340 train 9.535019 (lr=2.8615e-05) (hash(x)=36993091)
45
+ 350 train 9.515607 (lr=2.9455e-05) (hash(x)=38742081)
46
+ 360 train 9.491188 (lr=3.0294e-05) (hash(x)=43576901)
47
+ 370 train 9.488279 (lr=3.1133e-05) (hash(x)=36570538)
48
+ 380 train 9.439842 (lr=3.1972e-05) (hash(x)=43348240)
49
+ 390 train 9.387309 (lr=3.2811e-05) (hash(x)=37875659)
50
+ 400 val loss 9.3864
51
+ 400 val perplexity 11924.7754
52
+ 400 train 9.449704 (lr=3.3650e-05) (hash(x)=41660812)
53
+ 410 train 9.371158 (lr=3.4490e-05) (hash(x)=37968408)
54
+ 420 train 9.374220 (lr=3.5329e-05) (hash(x)=41080176)
55
+ 430 train 9.337018 (lr=3.6168e-05) (hash(x)=34708528)
56
+ 440 train 9.291052 (lr=3.7007e-05) (hash(x)=41041500)
57
+ 450 train 9.333014 (lr=3.7846e-05) (hash(x)=37590081)
58
+ 460 train 9.261171 (lr=3.8685e-05) (hash(x)=42370259)
59
+ 470 train 9.262412 (lr=3.9524e-05) (hash(x)=38174196)
60
+ 480 train 9.258465 (lr=4.0364e-05) (hash(x)=39264243)
61
+ 490 train 9.252340 (lr=4.1203e-05) (hash(x)=39565440)
62
+ 500 val loss 9.2203
63
+ 500 val perplexity 10099.6768
64
+ 500 train 9.213354 (lr=4.2042e-05) (hash(x)=35815196)
65
+ 510 train 9.191642 (lr=4.2881e-05) (hash(x)=39616059)
66
+ 520 train 9.209084 (lr=4.3720e-05) (hash(x)=49078223)
67
+ 530 train 9.207720 (lr=4.4559e-05) (hash(x)=45929312)
68
+ 540 train 9.178532 (lr=4.5399e-05) (hash(x)=39860869)
69
+ 550 train 9.147655 (lr=4.6238e-05) (hash(x)=45524200)
70
+ 560 train 9.100924 (lr=4.7077e-05) (hash(x)=34657559)
71
+ 570 train 9.081181 (lr=4.7916e-05) (hash(x)=34198027)
72
+ 580 train 9.069355 (lr=4.8755e-05) (hash(x)=44856508)
73
+ 590 train 9.087070 (lr=4.9594e-05) (hash(x)=42146659)
74
+ 600 val loss 9.0449
75
+ 600 val perplexity 8475.4053
76
+ 600 train 9.052690 (lr=5.0434e-05) (hash(x)=36673381)
77
+ 610 train 9.028795 (lr=5.1273e-05) (hash(x)=46081639)
78
+ 620 train 9.018815 (lr=5.2112e-05) (hash(x)=41640874)
79
+ 630 train 8.968925 (lr=5.2951e-05) (hash(x)=41224300)
80
+ 640 train 8.966177 (lr=5.3790e-05) (hash(x)=39147021)
81
+ 650 train 8.978804 (lr=5.4629e-05) (hash(x)=36710170)
82
+ 660 train 8.936301 (lr=5.5469e-05) (hash(x)=41488466)
83
+ 670 train 8.938276 (lr=5.6308e-05) (hash(x)=41010960)
84
+ 680 train 8.990581 (lr=5.7147e-05) (hash(x)=39429901)
85
+ 690 train 8.923993 (lr=5.7986e-05) (hash(x)=44800066)
86
+ 700 val loss 8.9057
87
+ 700 val perplexity 7374.2349
88
+ 700 train 8.915691 (lr=5.8825e-05) (hash(x)=46954230)
89
+ 710 train 8.893067 (lr=5.9664e-05) (hash(x)=44729399)
90
+ 720 train 8.870402 (lr=5.9999e-05) (hash(x)=40288354)
91
+ 730 train 8.828152 (lr=5.9991e-05) (hash(x)=44089962)
92
+ 740 train 8.777225 (lr=5.9974e-05) (hash(x)=40026145)
93
+ 750 train 8.734163 (lr=5.9949e-05) (hash(x)=36468229)
94
+ 760 train 8.735678 (lr=5.9915e-05) (hash(x)=39182712)
95
+ 770 train 8.723209 (lr=5.9874e-05) (hash(x)=41005834)
96
+ 780 train 8.713182 (lr=5.9824e-05) (hash(x)=38232985)
97
+ 790 train 8.744566 (lr=5.9765e-05) (hash(x)=38064072)
98
+ 800 val loss 8.7096
99
+ 800 val perplexity 6061.0806
100
+ 800 train 8.710663 (lr=5.9698e-05) (hash(x)=39571816)
101
+ 810 train 8.657696 (lr=5.9623e-05) (hash(x)=41286196)
102
+ 820 train 8.659924 (lr=5.9540e-05) (hash(x)=44007678)
103
+ 830 train 8.713148 (lr=5.9449e-05) (hash(x)=36771154)
104
+ 840 train 8.624725 (lr=5.9349e-05) (hash(x)=44197362)
105
+ 850 train 8.627365 (lr=5.9241e-05) (hash(x)=37351368)
106
+ 860 train 8.719424 (lr=5.9126e-05) (hash(x)=42629178)
107
+ 870 train 8.721700 (lr=5.9002e-05) (hash(x)=36215816)
108
+ 880 train 8.570547 (lr=5.8869e-05) (hash(x)=45418370)
109
+ 890 train 8.583411 (lr=5.8729e-05) (hash(x)=38983291)
110
+ 900 val loss 8.5358
111
+ 900 val perplexity 5093.7202
112
+ 900 train 8.529021 (lr=5.8581e-05) (hash(x)=34016998)
113
+ 910 train 8.491091 (lr=5.8425e-05) (hash(x)=36922232)
114
+ 920 train 8.512975 (lr=5.8262e-05) (hash(x)=43334032)
115
+ 930 train 8.530880 (lr=5.8090e-05) (hash(x)=39139652)
116
+ 940 train 8.476629 (lr=5.7911e-05) (hash(x)=39832832)
117
+ 950 train 8.461159 (lr=5.7723e-05) (hash(x)=38795848)
118
+ 960 train 8.463678 (lr=5.7529e-05) (hash(x)=37215711)
119
+ 970 train 8.508911 (lr=5.7326e-05) (hash(x)=37313575)
120
+ 980 train 9.121012 (lr=5.7116e-05) (hash(x)=46141234)
121
+ 990 train 8.536098 (lr=5.6899e-05) (hash(x)=41240889)
122
+ 1000 val loss 8.4473
123
+ 1000 val perplexity 4662.6045
124
+ 1000 train 8.419992 (lr=5.6674e-05) (hash(x)=40204404)
125
+ 1010 train 8.395643 (lr=5.6442e-05) (hash(x)=34594593)
126
+ 1020 train 8.433323 (lr=5.6202e-05) (hash(x)=38795311)
127
+ 1030 train 8.423841 (lr=5.5956e-05) (hash(x)=43467681)
128
+ 1040 train 9.148621 (lr=5.5702e-05) (hash(x)=38728767)
129
+ 1050 train 9.802886 (lr=5.5441e-05) (hash(x)=42307606)
130
+ 1060 train 9.552670 (lr=5.5174e-05) (hash(x)=43499838)
131
+ 1070 train 8.804494 (lr=5.4899e-05) (hash(x)=42853705)
132
+ 1080 train 8.543422 (lr=5.4618e-05) (hash(x)=38874363)
133
+ 1090 train 8.410303 (lr=5.4330e-05) (hash(x)=44247152)
134
+ 1100 val loss 8.3707
135
+ 1100 val perplexity 4318.6699
136
+ 1100 train 8.350416 (lr=5.4035e-05) (hash(x)=41926226)
137
+ 1110 train 8.324833 (lr=5.3734e-05) (hash(x)=42645790)
138
+ 1120 train 8.302387 (lr=5.3426e-05) (hash(x)=43723936)
139
+ 1130 train 8.270917 (lr=5.3113e-05) (hash(x)=41543766)
140
+ 1140 train 8.235456 (lr=5.2792e-05) (hash(x)=39842285)
141
+ 1150 train 8.263171 (lr=5.2466e-05) (hash(x)=41160965)
142
+ 1160 train 8.265052 (lr=5.2134e-05) (hash(x)=42843998)
143
+ 1170 train 8.272336 (lr=5.1796e-05) (hash(x)=30950803)
144
+ 1180 train 8.270874 (lr=5.1452e-05) (hash(x)=34820484)
145
+ 1190 train 8.244272 (lr=5.1102e-05) (hash(x)=37698786)
146
+ 1200 val loss 8.2746
147
+ 1200 val perplexity 3922.8298
148
+ 1200 train 8.269227 (lr=5.0746e-05) (hash(x)=41861487)
149
+ 1210 train 8.290969 (lr=5.0386e-05) (hash(x)=39863630)
150
+ 1220 train 8.285964 (lr=5.0019e-05) (hash(x)=40242551)
151
+ 1230 train 8.234189 (lr=4.9648e-05) (hash(x)=44311591)
152
+ 1240 train 8.202966 (lr=4.9271e-05) (hash(x)=39781776)
153
+ 1250 train 8.287863 (lr=4.8889e-05) (hash(x)=42390421)
154
+ 1260 train 8.294518 (lr=4.8503e-05) (hash(x)=41152137)
155
+ 1270 train 8.233472 (lr=4.8111e-05) (hash(x)=36946170)
156
+ 1280 train 8.234783 (lr=4.7715e-05) (hash(x)=37347986)
157
+ 1290 train 8.247560 (lr=4.7315e-05) (hash(x)=32033927)
158
+ 1300 val loss 8.2684
159
+ 1300 val perplexity 3898.8071
160
+ 1300 train 8.257142 (lr=4.6909e-05) (hash(x)=44258036)
161
+ 1310 train 8.251493 (lr=4.6500e-05) (hash(x)=46067973)
162
+ 1320 train 8.231394 (lr=4.6086e-05) (hash(x)=39193154)
163
+ 1330 train 8.259173 (lr=4.5669e-05) (hash(x)=45642219)
164
+ 1340 train 8.286476 (lr=4.5247e-05) (hash(x)=40763259)
165
+ 1350 train 8.253614 (lr=4.4822e-05) (hash(x)=38172293)
166
+ 1360 train 8.230739 (lr=4.4393e-05) (hash(x)=37114843)
167
+ 1370 train 8.265316 (lr=4.3960e-05) (hash(x)=40819318)
168
+ 1380 train 8.253040 (lr=4.3524e-05) (hash(x)=42057083)
169
+ 1390 train 8.248776 (lr=4.3085e-05) (hash(x)=38347811)
170
+ 1400 val loss 8.2521
171
+ 1400 val perplexity 3835.5735
172
+ 1400 train 8.246021 (lr=4.2643e-05) (hash(x)=35847481)
173
+ 1410 train 8.260033 (lr=4.2197e-05) (hash(x)=38699246)
174
+ 1420 train 8.295799 (lr=4.1749e-05) (hash(x)=38324524)
175
+ 1430 train 8.247987 (lr=4.1298e-05) (hash(x)=40369167)
176
+ 1440 train 8.269791 (lr=4.0845e-05) (hash(x)=33500630)
177
+ 1450 train 8.239408 (lr=4.0389e-05) (hash(x)=42605844)
178
+ 1460 train 8.244835 (lr=3.9931e-05) (hash(x)=35638705)
179
+ 1470 train 8.228526 (lr=3.9470e-05) (hash(x)=44425926)
180
+ 1480 train 8.260433 (lr=3.9008e-05) (hash(x)=45980198)
181
+ 1490 train 8.248598 (lr=3.8544e-05) (hash(x)=36381167)
182
+ 1500 val loss 8.2578
183
+ 1500 val perplexity 3857.5139
184
+ 1500 train 8.270606 (lr=3.8078e-05) (hash(x)=39218969)
185
+ 1510 train 8.219898 (lr=3.7610e-05) (hash(x)=34963329)
186
+ 1520 train 8.274857 (lr=3.7142e-05) (hash(x)=38493490)
187
+ 1530 train 8.269361 (lr=3.6671e-05) (hash(x)=37779234)
188
+ 1540 train 8.272569 (lr=3.6200e-05) (hash(x)=39600724)
189
+ 1550 train 8.308247 (lr=3.5728e-05) (hash(x)=41652676)
190
+ 1560 train 8.388374 (lr=3.5255e-05) (hash(x)=36794158)
191
+ 1570 train 8.228391 (lr=3.4781e-05) (hash(x)=28528384)
192
+ 1580 train 8.288967 (lr=3.4306e-05) (hash(x)=40202410)
193
+ 1590 train 8.262313 (lr=3.3831e-05) (hash(x)=41923463)
194
+ 1600 val loss 8.2870
195
+ 1600 val perplexity 3971.8389
196
+ 1600 train 8.246229 (lr=3.3356e-05) (hash(x)=39934496)
197
+ 1610 train 8.259838 (lr=3.2881e-05) (hash(x)=40576932)
198
+ 1620 train 8.367721 (lr=3.2406e-05) (hash(x)=41986373)
199
+ 1630 train 8.331136 (lr=3.1931e-05) (hash(x)=38502265)
200
+ 1640 train 8.299128 (lr=3.1456e-05) (hash(x)=38956594)
201
+ 1650 train 8.344444 (lr=3.0982e-05) (hash(x)=37470941)
202
+ 1660 train 8.425151 (lr=3.0509e-05) (hash(x)=35716340)
203
+ 1670 train 8.253879 (lr=3.0036e-05) (hash(x)=39081951)
204
+ 1680 train 8.361001 (lr=2.9564e-05) (hash(x)=44126819)
205
+ 1690 train 8.254582 (lr=2.9093e-05) (hash(x)=39815554)
206
+ 1700 val loss 8.2394
207
+ 1700 val perplexity 3787.2742
208
+ 1700 train 8.254066 (lr=2.8624e-05) (hash(x)=37646590)
209
+ 1710 train 8.199924 (lr=2.8156e-05) (hash(x)=38286621)
210
+ 1720 train 8.252518 (lr=2.7689e-05) (hash(x)=38421820)
211
+ 1730 train 8.222192 (lr=2.7224e-05) (hash(x)=41288435)
212
+ 1740 train 8.198758 (lr=2.6761e-05) (hash(x)=45642859)
213
+ 1750 train 8.162834 (lr=2.6299e-05) (hash(x)=36528352)
214
+ 1760 train 8.202525 (lr=2.5840e-05) (hash(x)=36914578)
215
+ 1770 train 8.202765 (lr=2.5383e-05) (hash(x)=38009157)
216
+ 1780 train 8.163364 (lr=2.4928e-05) (hash(x)=36105857)
217
+ 1790 train 8.222654 (lr=2.4476e-05) (hash(x)=40873205)
218
+ 1800 val loss 8.1983
219
+ 1800 val perplexity 3634.9023
220
+ 1800 train 8.172622 (lr=2.4026e-05) (hash(x)=31431733)
221
+ 1810 train 8.180494 (lr=2.3580e-05) (hash(x)=37889200)
222
+ 1820 train 8.181370 (lr=2.3136e-05) (hash(x)=41974666)
223
+ 1830 train 8.215858 (lr=2.2695e-05) (hash(x)=38425824)
224
+ 1840 train 8.192945 (lr=2.2257e-05) (hash(x)=40983908)
225
+ 1850 train 8.187275 (lr=2.1823e-05) (hash(x)=43685380)
226
+ 1860 train 8.193357 (lr=2.1392e-05) (hash(x)=31777770)
227
+ 1870 train 8.150019 (lr=2.0965e-05) (hash(x)=44109703)
228
+ 1880 train 8.175804 (lr=2.0542e-05) (hash(x)=35678244)
229
+ 1890 train 8.196533 (lr=2.0122e-05) (hash(x)=35649940)
230
+ 1900 val loss 8.1889
231
+ 1900 val perplexity 3600.8040
232
+ 1900 train 8.232388 (lr=1.9706e-05) (hash(x)=43009344)
233
+ 1910 train 8.204277 (lr=1.9295e-05) (hash(x)=41641641)
234
+ 1920 train 8.198946 (lr=1.8887e-05) (hash(x)=41378670)
235
+ 1930 train 8.158615 (lr=1.8485e-05) (hash(x)=42228302)
236
+ 1940 train 8.203094 (lr=1.8086e-05) (hash(x)=35659474)
237
+ 1950 train 8.163998 (lr=1.7692e-05) (hash(x)=43019440)
238
+ 1960 train 8.177827 (lr=1.7303e-05) (hash(x)=36388055)
239
+ 1970 train 8.200537 (lr=1.6919e-05) (hash(x)=38467899)
240
+ 1980 train 8.189949 (lr=1.6540e-05) (hash(x)=37166112)
241
+ 1990 train 8.190762 (lr=1.6166e-05) (hash(x)=39116198)
242
+ 2000 val loss 8.1951
243
+ 2000 val perplexity 3623.0732
244
+ 2000 train 8.126313 (lr=1.5797e-05) (hash(x)=34741552)
245
+ 2010 train 8.202533 (lr=1.5433e-05) (hash(x)=39795810)
246
+ 2020 train 8.169413 (lr=1.5075e-05) (hash(x)=39901562)
247
+ 2030 train 8.176146 (lr=1.4723e-05) (hash(x)=32955586)
248
+ 2040 train 8.219574 (lr=1.4376e-05) (hash(x)=40879614)
249
+ 2050 train 8.198061 (lr=1.4035e-05) (hash(x)=39268533)
250
+ 2060 train 8.175335 (lr=1.3699e-05) (hash(x)=44795608)
251
+ 2070 train 8.141266 (lr=1.3370e-05) (hash(x)=39229893)
252
+ 2080 train 8.176501 (lr=1.3047e-05) (hash(x)=37103514)
253
+ 2090 train 8.144024 (lr=1.2730e-05) (hash(x)=40987512)
254
+ 2100 val loss 8.1827
255
+ 2100 val perplexity 3578.5076
256
+ 2100 train 8.200542 (lr=1.2419e-05) (hash(x)=51452794)
257
+ 2110 train 8.192780 (lr=1.2115e-05) (hash(x)=42262641)
258
+ 2120 train 8.162808 (lr=1.1817e-05) (hash(x)=41984974)
259
+ 2130 train 8.091839 (lr=1.1525e-05) (hash(x)=35439208)
260
+ 2140 train 8.153090 (lr=1.1241e-05) (hash(x)=39817069)
261
+ 2150 train 8.177710 (lr=1.0963e-05) (hash(x)=37989592)
262
+ 2160 train 8.129671 (lr=1.0692e-05) (hash(x)=49885738)
263
+ 2170 train 8.221235 (lr=1.0427e-05) (hash(x)=40654241)
264
+ 2180 train 8.171923 (lr=1.0170e-05) (hash(x)=41950889)
265
+ 2190 train 8.196054 (lr=9.9199e-06) (hash(x)=40294158)
266
+ 2200 val loss 8.1828
267
+ 2200 val perplexity 3578.8455
268
+ 2200 train 8.158991 (lr=9.6769e-06) (hash(x)=40076244)
269
+ 2210 train 8.199405 (lr=9.4412e-06) (hash(x)=36688120)
270
+ 2220 train 8.160173 (lr=9.2127e-06) (hash(x)=37605188)
271
+ 2230 train 8.159690 (lr=8.9916e-06) (hash(x)=35632798)
272
+ 2240 train 8.209606 (lr=8.7779e-06) (hash(x)=42825466)
273
+ 2250 train 8.183066 (lr=8.5717e-06) (hash(x)=35071109)
274
+ 2260 train 8.183180 (lr=8.3731e-06) (hash(x)=47897078)
275
+ 2270 train 8.195320 (lr=8.1821e-06) (hash(x)=46171759)
276
+ 2280 train 8.214214 (lr=7.9988e-06) (hash(x)=41213863)
277
+ 2290 train 8.148228 (lr=7.8232e-06) (hash(x)=40640886)
278
+ 2300 val loss 8.1859
279
+ 2300 val perplexity 3589.9175
280
+ 2300 train 8.158528 (lr=7.6555e-06) (hash(x)=44989210)
281
+ 2310 train 8.207983 (lr=7.4956e-06) (hash(x)=40352035)
282
+ 2320 train 8.199614 (lr=7.3436e-06) (hash(x)=42788091)
283
+ 2330 train 8.166622 (lr=7.1995e-06) (hash(x)=40234639)
284
+ 2340 train 8.157276 (lr=7.0635e-06) (hash(x)=35472103)
285
+ 2350 train 8.230900 (lr=6.9354e-06) (hash(x)=42043124)
286
+ 2360 train 8.152380 (lr=6.8155e-06) (hash(x)=39972275)
287
+ 2370 train 8.164783 (lr=6.7036e-06) (hash(x)=36566627)
288
+ 2380 train 8.217964 (lr=6.5999e-06) (hash(x)=55698840)
289
+ 2390 train 8.176562 (lr=6.5044e-06) (hash(x)=41155464)
290
+ 2400 val loss 8.1862
291
+ 2400 val perplexity 3590.9138
292
+ 2400 train 8.169226 (lr=6.4171e-06) (hash(x)=43485840)
293
+ 2410 train 8.168054 (lr=6.3380e-06) (hash(x)=39685043)
294
+ 2420 train 8.226298 (lr=6.2672e-06) (hash(x)=39496037)
295
+ 2430 train 8.165318 (lr=6.2046e-06) (hash(x)=41962948)
296
+ 2440 train 8.145103 (lr=6.1504e-06) (hash(x)=34986640)
297
+ 2450 train 8.228703 (lr=6.1045e-06) (hash(x)=39123972)
298
+ 2460 train 8.141926 (lr=6.0669e-06) (hash(x)=38112131)
299
+ 2470 train 8.153990 (lr=6.0376e-06) (hash(x)=39029104)
300
+ 2480 train 8.120266 (lr=6.0167e-06) (hash(x)=38048270)
301
+ 2490 train 8.212811 (lr=6.0042e-06) (hash(x)=37369977)
302
+ 2499 val loss 8.1829
303
+ 2499 val perplexity 3579.1697
one_mask_per_head_4_latent_vectors/model_02499.pt ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:c9990944dae65edd7a21d742975ecf03840bc2dd5acaa8363f74178b1068d460
3
+ size 515929202
one_mask_per_head_4_latent_vectors/optimizer_02499.pt ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:3b6a2473fcc2c4db8f645ff0a5d0b940846426b2f7443022e721214dade581a4
3
+ size 1006706614