andrew-healey commited on
Commit
52d2247
·
verified ·
1 Parent(s): b854b75

Upload folder using huggingface_hub

Browse files
allowing_more_selection_patterns/one_mask_per_head_1_latent_vector_degenerate_no_compile_seed_1337/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_no_compile_seed_1337", "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": 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": false, "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, "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_no_compile"}
allowing_more_selection_patterns/one_mask_per_head_1_latent_vector_degenerate_no_compile_seed_1337/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_1_latent_vector_degenerate_no_compile_seed_1337/log2.txt ADDED
@@ -0,0 +1,314 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ max_steps: 2500
2
+ 0 val loss 10.9623
3
+ 0 val perplexity 57661.6719
4
+ 0 hella 0.2511
5
+ 0 train 10.960283 (lr=8.3916e-07) (hash(x)=23347629)
6
+ 10 train 10.120608 (lr=9.2308e-06) (hash(x)=19636655)
7
+ 20 train 9.648245 (lr=1.7622e-05) (hash(x)=24490248)
8
+ 30 train 9.470486 (lr=2.6014e-05) (hash(x)=21733558)
9
+ 40 train 9.215852 (lr=3.4406e-05) (hash(x)=31750616)
10
+ 50 train 8.905017 (lr=4.2797e-05) (hash(x)=20041895)
11
+ 60 train 8.542735 (lr=5.1189e-05) (hash(x)=21416280)
12
+ 70 train 8.224788 (lr=5.9580e-05) (hash(x)=18066780)
13
+ 80 train 7.907933 (lr=6.7972e-05) (hash(x)=21809427)
14
+ 90 train 7.681328 (lr=7.6364e-05) (hash(x)=22061595)
15
+ 100 val loss 7.4752
16
+ 100 val perplexity 1763.7076
17
+ 100 train 7.425412 (lr=8.4755e-05) (hash(x)=17468914)
18
+ 110 train 7.234653 (lr=9.3147e-05) (hash(x)=20421520)
19
+ 120 train 7.243821 (lr=1.0154e-04) (hash(x)=23140133)
20
+ 130 train 7.114510 (lr=1.0993e-04) (hash(x)=20006835)
21
+ 140 train 6.870553 (lr=1.1832e-04) (hash(x)=18628134)
22
+ 150 train 6.846054 (lr=1.2671e-04) (hash(x)=21913447)
23
+ 160 train 6.771260 (lr=1.3510e-04) (hash(x)=22997297)
24
+ 170 train 6.582151 (lr=1.4350e-04) (hash(x)=17904821)
25
+ 180 train 6.434290 (lr=1.5189e-04) (hash(x)=18826982)
26
+ 190 train 6.532411 (lr=1.6028e-04) (hash(x)=21387716)
27
+ 200 val loss 6.4037
28
+ 200 val perplexity 604.0753
29
+ 200 train 6.322908 (lr=1.6867e-04) (hash(x)=16236395)
30
+ 210 train 6.193309 (lr=1.7706e-04) (hash(x)=21428681)
31
+ 220 train 6.356368 (lr=1.8545e-04) (hash(x)=20461738)
32
+ 230 train 6.153878 (lr=1.9385e-04) (hash(x)=20394865)
33
+ 240 train 6.057524 (lr=2.0224e-04) (hash(x)=18028671)
34
+ 250 hella 0.2421
35
+ 250 train 5.924955 (lr=2.1063e-04) (hash(x)=18122073)
36
+ 260 train 5.832840 (lr=2.1902e-04) (hash(x)=22994143)
37
+ 270 train 6.027573 (lr=2.2741e-04) (hash(x)=20646260)
38
+ 280 train 5.859165 (lr=2.3580e-04) (hash(x)=23204708)
39
+ 290 train 5.773631 (lr=2.4420e-04) (hash(x)=17900252)
40
+ 300 val loss 5.8129
41
+ 300 val perplexity 334.5804
42
+ 300 train 5.651948 (lr=2.5259e-04) (hash(x)=19073819)
43
+ 310 train 5.837531 (lr=2.6098e-04) (hash(x)=22378232)
44
+ 320 train 5.728408 (lr=2.6937e-04) (hash(x)=19910430)
45
+ 330 train 5.664508 (lr=2.7776e-04) (hash(x)=25168753)
46
+ 340 train 5.529865 (lr=2.8615e-04) (hash(x)=20665256)
47
+ 350 train 5.409571 (lr=2.9455e-04) (hash(x)=19951160)
48
+ 360 train 5.625602 (lr=3.0294e-04) (hash(x)=21178202)
49
+ 370 train 5.469856 (lr=3.1133e-04) (hash(x)=20435886)
50
+ 380 train 5.410357 (lr=3.1972e-04) (hash(x)=21255097)
51
+ 390 train 5.417028 (lr=3.2811e-04) (hash(x)=19596330)
52
+ 400 val loss 5.3557
53
+ 400 val perplexity 211.8215
54
+ 400 train 5.385563 (lr=3.3650e-04) (hash(x)=21789426)
55
+ 410 train 5.285660 (lr=3.4490e-04) (hash(x)=18366401)
56
+ 420 train 5.171840 (lr=3.5329e-04) (hash(x)=18614660)
57
+ 430 train 5.323102 (lr=3.6168e-04) (hash(x)=21649063)
58
+ 440 train 5.197217 (lr=3.7007e-04) (hash(x)=18505097)
59
+ 450 train 5.180646 (lr=3.7846e-04) (hash(x)=23434312)
60
+ 460 train 5.075920 (lr=3.8685e-04) (hash(x)=16584569)
61
+ 470 train 4.954670 (lr=3.9524e-04) (hash(x)=13447111)
62
+ 480 train 5.164924 (lr=4.0364e-04) (hash(x)=22645130)
63
+ 490 train 5.034174 (lr=4.1203e-04) (hash(x)=19362777)
64
+ 500 val loss 4.9969
65
+ 500 val perplexity 147.9479
66
+ 500 hella 0.2390
67
+ 500 train 4.929436 (lr=4.2042e-04) (hash(x)=20921874)
68
+ 510 train 4.829317 (lr=4.2881e-04) (hash(x)=22123048)
69
+ 520 train 4.867614 (lr=4.3720e-04) (hash(x)=21215368)
70
+ 530 train 5.037164 (lr=4.4559e-04) (hash(x)=20717936)
71
+ 540 train 4.858409 (lr=4.5399e-04) (hash(x)=20118007)
72
+ 550 train 4.748128 (lr=4.6238e-04) (hash(x)=21402522)
73
+ 560 train 4.653287 (lr=4.7077e-04) (hash(x)=19057180)
74
+ 570 train 4.849389 (lr=4.7916e-04) (hash(x)=19360553)
75
+ 580 train 4.751447 (lr=4.8755e-04) (hash(x)=18194693)
76
+ 590 train 4.638285 (lr=4.9594e-04) (hash(x)=19863826)
77
+ 600 val loss 4.5958
78
+ 600 val perplexity 99.0636
79
+ 600 train 4.551629 (lr=5.0434e-04) (hash(x)=21794551)
80
+ 610 train 4.516992 (lr=5.1273e-04) (hash(x)=18685622)
81
+ 620 train 4.424405 (lr=5.2112e-04) (hash(x)=23473104)
82
+ 630 train 4.503919 (lr=5.2951e-04) (hash(x)=18671672)
83
+ 640 train 4.530106 (lr=5.3790e-04) (hash(x)=19234240)
84
+ 650 train 4.417812 (lr=5.4629e-04) (hash(x)=23183903)
85
+ 660 train 4.260354 (lr=5.5469e-04) (hash(x)=23665543)
86
+ 670 train 4.417737 (lr=5.6308e-04) (hash(x)=22055313)
87
+ 680 train 4.418112 (lr=5.7147e-04) (hash(x)=22321668)
88
+ 690 train 4.512354 (lr=5.7986e-04) (hash(x)=21568926)
89
+ 700 val loss 4.3490
90
+ 700 val perplexity 77.4039
91
+ 700 train 4.355264 (lr=5.8825e-04) (hash(x)=19783143)
92
+ 710 train 4.269092 (lr=5.9664e-04) (hash(x)=21565258)
93
+ 720 train 4.359616 (lr=5.9999e-04) (hash(x)=16127212)
94
+ 730 train 4.295231 (lr=5.9991e-04) (hash(x)=19658174)
95
+ 740 train 4.247321 (lr=5.9974e-04) (hash(x)=16454319)
96
+ 750 hella 0.2513
97
+ 750 train 4.199901 (lr=5.9949e-04) (hash(x)=17993154)
98
+ 760 train 4.093071 (lr=5.9915e-04) (hash(x)=19256989)
99
+ 770 train 4.095788 (lr=5.9874e-04) (hash(x)=17351653)
100
+ 780 train 4.105547 (lr=5.9824e-04) (hash(x)=21626120)
101
+ 790 train 4.246477 (lr=5.9765e-04) (hash(x)=16367412)
102
+ 800 val loss 4.1701
103
+ 800 val perplexity 64.7216
104
+ 800 train 4.121781 (lr=5.9698e-04) (hash(x)=22321504)
105
+ 810 train 4.009194 (lr=5.9623e-04) (hash(x)=19869755)
106
+ 820 train 4.164668 (lr=5.9540e-04) (hash(x)=24183665)
107
+ 830 train 4.150872 (lr=5.9449e-04) (hash(x)=19194113)
108
+ 840 train 3.993633 (lr=5.9349e-04) (hash(x)=17710503)
109
+ 850 train 4.009343 (lr=5.9241e-04) (hash(x)=17231851)
110
+ 860 train 4.145665 (lr=5.9126e-04) (hash(x)=20584272)
111
+ 870 train 4.041808 (lr=5.9002e-04) (hash(x)=17468790)
112
+ 880 train 3.977543 (lr=5.8869e-04) (hash(x)=22304409)
113
+ 890 train 4.065569 (lr=5.8729e-04) (hash(x)=18871567)
114
+ 900 val loss 4.0323
115
+ 900 val perplexity 56.3898
116
+ 900 train 4.011298 (lr=5.8581e-04) (hash(x)=20924720)
117
+ 910 train 3.928891 (lr=5.8425e-04) (hash(x)=21163464)
118
+ 920 train 3.860510 (lr=5.8262e-04) (hash(x)=20120862)
119
+ 930 train 4.038430 (lr=5.8090e-04) (hash(x)=19565198)
120
+ 940 train 3.955804 (lr=5.7911e-04) (hash(x)=18517540)
121
+ 950 train 3.922330 (lr=5.7723e-04) (hash(x)=18300759)
122
+ 960 train 3.960672 (lr=5.7529e-04) (hash(x)=21443001)
123
+ 970 train 3.904534 (lr=5.7326e-04) (hash(x)=15210971)
124
+ 980 train 4.069089 (lr=5.7116e-04) (hash(x)=20874579)
125
+ 990 train 3.989326 (lr=5.6899e-04) (hash(x)=20690802)
126
+ 1000 val loss 3.9394
127
+ 1000 val perplexity 51.3874
128
+ 1000 hella 0.2544
129
+ 1000 train 3.860886 (lr=5.6674e-04) (hash(x)=18635877)
130
+ 1010 train 3.966814 (lr=5.6442e-04) (hash(x)=21141462)
131
+ 1020 train 3.985730 (lr=5.6202e-04) (hash(x)=19483400)
132
+ 1030 train 3.892252 (lr=5.5956e-04) (hash(x)=24792825)
133
+ 1040 train 3.875563 (lr=5.5702e-04) (hash(x)=16743928)
134
+ 1050 train 3.912506 (lr=5.5441e-04) (hash(x)=22540052)
135
+ 1060 train 3.930166 (lr=5.5174e-04) (hash(x)=20546637)
136
+ 1070 train 3.833379 (lr=5.4899e-04) (hash(x)=13739753)
137
+ 1080 train 3.884339 (lr=5.4618e-04) (hash(x)=20071304)
138
+ 1090 train 3.842955 (lr=5.4330e-04) (hash(x)=19339656)
139
+ 1100 val loss 3.8580
140
+ 1100 val perplexity 47.3719
141
+ 1100 train 3.824395 (lr=5.4035e-04) (hash(x)=20039628)
142
+ 1110 train 3.893573 (lr=5.3734e-04) (hash(x)=19249691)
143
+ 1120 train 3.908659 (lr=5.3426e-04) (hash(x)=19048809)
144
+ 1130 train 3.833630 (lr=5.3113e-04) (hash(x)=19309000)
145
+ 1140 train 3.796860 (lr=5.2792e-04) (hash(x)=21787927)
146
+ 1150 train 3.701379 (lr=5.2466e-04) (hash(x)=19998536)
147
+ 1160 train 3.863658 (lr=5.2134e-04) (hash(x)=23724694)
148
+ 1170 train 3.877430 (lr=5.1796e-04) (hash(x)=17148204)
149
+ 1180 train 3.790870 (lr=5.1452e-04) (hash(x)=19511852)
150
+ 1190 train 3.764167 (lr=5.1102e-04) (hash(x)=19352862)
151
+ 1200 val loss 3.7960
152
+ 1200 val perplexity 44.5226
153
+ 1200 train 3.709783 (lr=5.0746e-04) (hash(x)=19530144)
154
+ 1210 train 3.885121 (lr=5.0386e-04) (hash(x)=19897392)
155
+ 1220 train 3.794478 (lr=5.0019e-04) (hash(x)=24402963)
156
+ 1230 train 3.711987 (lr=4.9648e-04) (hash(x)=18589754)
157
+ 1240 train 3.719692 (lr=4.9271e-04) (hash(x)=21951750)
158
+ 1250 hella 0.2553
159
+ 1250 train 3.792992 (lr=4.8889e-04) (hash(x)=18611379)
160
+ 1260 train 3.807710 (lr=4.8503e-04) (hash(x)=20338521)
161
+ 1270 train 3.729153 (lr=4.8111e-04) (hash(x)=18308405)
162
+ 1280 train 3.735471 (lr=4.7715e-04) (hash(x)=20920945)
163
+ 1290 train 3.698793 (lr=4.7315e-04) (hash(x)=20094152)
164
+ 1300 val loss 3.7386
165
+ 1300 val perplexity 42.0394
166
+ 1300 train 3.811048 (lr=4.6909e-04) (hash(x)=19003947)
167
+ 1310 train 3.770312 (lr=4.6500e-04) (hash(x)=19356356)
168
+ 1320 train 3.729848 (lr=4.6086e-04) (hash(x)=19390343)
169
+ 1330 train 3.660179 (lr=4.5669e-04) (hash(x)=17535982)
170
+ 1340 train 3.685437 (lr=4.5247e-04) (hash(x)=21959104)
171
+ 1350 train 3.785408 (lr=4.4822e-04) (hash(x)=21982099)
172
+ 1360 train 3.744889 (lr=4.4393e-04) (hash(x)=17176243)
173
+ 1370 train 3.691411 (lr=4.3960e-04) (hash(x)=21223627)
174
+ 1380 train 3.706606 (lr=4.3524e-04) (hash(x)=22462931)
175
+ 1390 train 3.730671 (lr=4.3085e-04) (hash(x)=19153464)
176
+ 1400 val loss 3.6867
177
+ 1400 val perplexity 39.9138
178
+ 1400 train 3.616882 (lr=4.2643e-04) (hash(x)=20146496)
179
+ 1410 train 3.682813 (lr=4.2197e-04) (hash(x)=21781327)
180
+ 1420 train 3.751625 (lr=4.1749e-04) (hash(x)=19234051)
181
+ 1430 train 3.799494 (lr=4.1298e-04) (hash(x)=20769898)
182
+ 1440 train 3.647286 (lr=4.0845e-04) (hash(x)=22042430)
183
+ 1450 train 3.672754 (lr=4.0389e-04) (hash(x)=20527387)
184
+ 1460 train 3.659091 (lr=3.9931e-04) (hash(x)=18925766)
185
+ 1470 train 3.655443 (lr=3.9470e-04) (hash(x)=18198851)
186
+ 1480 train 3.730088 (lr=3.9008e-04) (hash(x)=21847281)
187
+ 1490 train 3.696964 (lr=3.8544e-04) (hash(x)=15645274)
188
+ 1500 val loss 3.6421
189
+ 1500 val perplexity 38.1734
190
+ 1500 hella 0.2610
191
+ 1500 train 3.568343 (lr=3.8078e-04) (hash(x)=14693370)
192
+ 1510 train 3.579729 (lr=3.7610e-04) (hash(x)=21810857)
193
+ 1520 train 3.674245 (lr=3.7142e-04) (hash(x)=18980052)
194
+ 1530 train 3.698456 (lr=3.6671e-04) (hash(x)=22868842)
195
+ 1540 train 3.615157 (lr=3.6200e-04) (hash(x)=17289166)
196
+ 1550 train 3.699608 (lr=3.5728e-04) (hash(x)=18109373)
197
+ 1560 train 3.731543 (lr=3.5255e-04) (hash(x)=22377619)
198
+ 1570 train 3.559052 (lr=3.4781e-04) (hash(x)=20307182)
199
+ 1580 train 3.706031 (lr=3.4306e-04) (hash(x)=20391291)
200
+ 1590 train 3.572526 (lr=3.3831e-04) (hash(x)=18690939)
201
+ 1600 val loss 3.6096
202
+ 1600 val perplexity 36.9512
203
+ 1600 train 3.587132 (lr=3.3356e-04) (hash(x)=16719886)
204
+ 1610 train 3.578436 (lr=3.2881e-04) (hash(x)=16040840)
205
+ 1620 train 3.651986 (lr=3.2406e-04) (hash(x)=25795140)
206
+ 1630 train 3.658085 (lr=3.1931e-04) (hash(x)=21936964)
207
+ 1640 train 3.556605 (lr=3.1456e-04) (hash(x)=11593027)
208
+ 1650 train 3.610035 (lr=3.0982e-04) (hash(x)=23256718)
209
+ 1660 train 3.572417 (lr=3.0509e-04) (hash(x)=20289140)
210
+ 1670 train 3.578953 (lr=3.0036e-04) (hash(x)=18382509)
211
+ 1680 train 3.644212 (lr=2.9564e-04) (hash(x)=26604329)
212
+ 1690 train 3.610823 (lr=2.9093e-04) (hash(x)=22122534)
213
+ 1700 val loss 3.5775
214
+ 1700 val perplexity 35.7834
215
+ 1700 train 3.626693 (lr=2.8624e-04) (hash(x)=22884812)
216
+ 1710 train 3.574154 (lr=2.8156e-04) (hash(x)=18598496)
217
+ 1720 train 3.593128 (lr=2.7689e-04) (hash(x)=22147447)
218
+ 1730 train 3.598692 (lr=2.7224e-04) (hash(x)=18912135)
219
+ 1740 train 3.436403 (lr=2.6761e-04) (hash(x)=21952974)
220
+ 1750 hella 0.2667
221
+ 1750 train 3.548981 (lr=2.6299e-04) (hash(x)=18390978)
222
+ 1760 train 3.557056 (lr=2.5840e-04) (hash(x)=17979288)
223
+ 1770 train 3.567672 (lr=2.5383e-04) (hash(x)=18979099)
224
+ 1780 train 3.531941 (lr=2.4928e-04) (hash(x)=20511922)
225
+ 1790 train 3.584124 (lr=2.4476e-04) (hash(x)=21814060)
226
+ 1800 val loss 3.5512
227
+ 1800 val perplexity 34.8563
228
+ 1800 train 3.570031 (lr=2.4026e-04) (hash(x)=18076243)
229
+ 1810 train 3.559240 (lr=2.3580e-04) (hash(x)=19721209)
230
+ 1820 train 3.506674 (lr=2.3136e-04) (hash(x)=18140847)
231
+ 1830 train 3.570811 (lr=2.2695e-04) (hash(x)=22065413)
232
+ 1840 train 3.472082 (lr=2.2257e-04) (hash(x)=21166111)
233
+ 1850 train 3.470281 (lr=2.1823e-04) (hash(x)=21904302)
234
+ 1860 train 3.536438 (lr=2.1392e-04) (hash(x)=18283828)
235
+ 1870 train 3.493150 (lr=2.0965e-04) (hash(x)=20892773)
236
+ 1880 train 3.524343 (lr=2.0542e-04) (hash(x)=17781264)
237
+ 1890 train 3.492336 (lr=2.0122e-04) (hash(x)=20612962)
238
+ 1900 val loss 3.5279
239
+ 1900 val perplexity 34.0516
240
+ 1900 train 3.583488 (lr=1.9706e-04) (hash(x)=20421058)
241
+ 1910 train 3.556489 (lr=1.9295e-04) (hash(x)=18554808)
242
+ 1920 train 3.513366 (lr=1.8887e-04) (hash(x)=18189279)
243
+ 1930 train 3.529909 (lr=1.8485e-04) (hash(x)=19923279)
244
+ 1940 train 3.495233 (lr=1.8086e-04) (hash(x)=21665511)
245
+ 1950 train 3.516702 (lr=1.7692e-04) (hash(x)=22038489)
246
+ 1960 train 3.492492 (lr=1.7303e-04) (hash(x)=18896956)
247
+ 1970 train 3.517446 (lr=1.6919e-04) (hash(x)=20408383)
248
+ 1980 train 3.498108 (lr=1.6540e-04) (hash(x)=23039940)
249
+ 1990 train 3.633665 (lr=1.6166e-04) (hash(x)=20631172)
250
+ 2000 val loss 3.5076
251
+ 2000 val perplexity 33.3688
252
+ 2000 hella 0.2668
253
+ 2000 train 3.447230 (lr=1.5797e-04) (hash(x)=20206106)
254
+ 2010 train 3.543045 (lr=1.5433e-04) (hash(x)=21864020)
255
+ 2020 train 3.525219 (lr=1.5075e-04) (hash(x)=21391128)
256
+ 2030 train 3.490790 (lr=1.4723e-04) (hash(x)=18676636)
257
+ 2040 train 3.538269 (lr=1.4376e-04) (hash(x)=15987134)
258
+ 2050 train 3.531995 (lr=1.4035e-04) (hash(x)=23940769)
259
+ 2060 train 3.508464 (lr=1.3699e-04) (hash(x)=17152840)
260
+ 2070 train 3.480815 (lr=1.3370e-04) (hash(x)=22958445)
261
+ 2080 train 3.529466 (lr=1.3047e-04) (hash(x)=23616724)
262
+ 2090 train 3.501508 (lr=1.2730e-04) (hash(x)=16571283)
263
+ 2100 val loss 3.4872
264
+ 2100 val perplexity 32.6934
265
+ 2100 train 3.496928 (lr=1.2419e-04) (hash(x)=19070327)
266
+ 2110 train 3.542429 (lr=1.2115e-04) (hash(x)=20758626)
267
+ 2120 train 3.486337 (lr=1.1817e-04) (hash(x)=18397810)
268
+ 2130 train 3.477817 (lr=1.1525e-04) (hash(x)=18045234)
269
+ 2140 train 3.578724 (lr=1.1241e-04) (hash(x)=18969565)
270
+ 2150 train 3.515861 (lr=1.0963e-04) (hash(x)=17149701)
271
+ 2160 train 3.438516 (lr=1.0692e-04) (hash(x)=23075443)
272
+ 2170 train 3.470445 (lr=1.0427e-04) (hash(x)=17977337)
273
+ 2180 train 3.530679 (lr=1.0170e-04) (hash(x)=20657407)
274
+ 2190 train 3.517409 (lr=9.9199e-05) (hash(x)=19985933)
275
+ 2200 val loss 3.4722
276
+ 2200 val perplexity 32.2079
277
+ 2200 train 3.470954 (lr=9.6769e-05) (hash(x)=19025387)
278
+ 2210 train 3.489388 (lr=9.4412e-05) (hash(x)=23318724)
279
+ 2220 train 3.560081 (lr=9.2127e-05) (hash(x)=19308039)
280
+ 2230 train 3.404163 (lr=8.9916e-05) (hash(x)=20733682)
281
+ 2240 train 3.498767 (lr=8.7779e-05) (hash(x)=18519560)
282
+ 2250 hella 0.2650
283
+ 2250 train 3.504050 (lr=8.5717e-05) (hash(x)=20299444)
284
+ 2260 train 3.461646 (lr=8.3731e-05) (hash(x)=18819615)
285
+ 2270 train 3.495849 (lr=8.1821e-05) (hash(x)=18743705)
286
+ 2280 train 3.491463 (lr=7.9988e-05) (hash(x)=17190793)
287
+ 2290 train 3.518027 (lr=7.8232e-05) (hash(x)=23126906)
288
+ 2300 val loss 3.4614
289
+ 2300 val perplexity 31.8603
290
+ 2300 train 3.451773 (lr=7.6555e-05) (hash(x)=24812267)
291
+ 2310 train 3.435097 (lr=7.4956e-05) (hash(x)=20147105)
292
+ 2320 train 3.548457 (lr=7.3436e-05) (hash(x)=24619936)
293
+ 2330 train 3.417677 (lr=7.1995e-05) (hash(x)=17730658)
294
+ 2340 train 3.393165 (lr=7.0635e-05) (hash(x)=19345731)
295
+ 2350 train 3.429430 (lr=6.9354e-05) (hash(x)=26415058)
296
+ 2360 train 3.467775 (lr=6.8155e-05) (hash(x)=18622999)
297
+ 2370 train 3.397594 (lr=6.7036e-05) (hash(x)=18848567)
298
+ 2380 train 3.479826 (lr=6.5999e-05) (hash(x)=21903360)
299
+ 2390 train 3.495298 (lr=6.5044e-05) (hash(x)=23479216)
300
+ 2400 val loss 3.4526
301
+ 2400 val perplexity 31.5814
302
+ 2400 train 3.492115 (lr=6.4171e-05) (hash(x)=19762013)
303
+ 2410 train 3.395568 (lr=6.3380e-05) (hash(x)=20622912)
304
+ 2420 train 3.511413 (lr=6.2672e-05) (hash(x)=17669226)
305
+ 2430 train 3.441263 (lr=6.2046e-05) (hash(x)=19536719)
306
+ 2440 train 3.384212 (lr=6.1504e-05) (hash(x)=19913725)
307
+ 2450 train 3.426318 (lr=6.1045e-05) (hash(x)=20782072)
308
+ 2460 train 3.396527 (lr=6.0669e-05) (hash(x)=19795002)
309
+ 2470 train 3.442552 (lr=6.0376e-05) (hash(x)=17003051)
310
+ 2480 train 3.410668 (lr=6.0167e-05) (hash(x)=21414190)
311
+ 2490 train 3.456452 (lr=6.0042e-05) (hash(x)=20943620)
312
+ 2499 val loss 3.4467
313
+ 2499 val perplexity 31.3974
314
+ 2499 hella 0.2686
allowing_more_selection_patterns/one_mask_per_head_1_latent_vector_degenerate_no_compile_seed_1337/model_02499.pt ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:f2ab0fab5642e1c3982bb0304eb2ecc4c8b84e2fd738001dde8250ee867fbf90
3
+ size 555244762
allowing_more_selection_patterns/one_mask_per_head_1_latent_vector_degenerate_no_compile_seed_1337/optimizer_02499.pt ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:02ad9f44eb96f7151229cb7b2c9d9762c7b482f41c1c0be4b7cdebf915ddf153
3
+ size 1009837470