andrew-healey commited on
Commit
a436473
·
verified ·
1 Parent(s): 8e53b09

Upload folder using huggingface_hub

Browse files
scaling_sdpa_nh/sdpa_nh_32_seed_1339/args.json ADDED
@@ -0,0 +1 @@
 
 
1
+ {"hellaswag": true, "attention_kind": "self", "log_dir": "scaling_sdpa_nh/sdpa_nh_32_seed_1339", "resume_checkpoint": null, "resume_optimizer": false, "add_a_head": false, "add_head_to_start": true, "new_head_init": "normal", "n_heads": 32, "n_embd": 256, "head_dim": 32, "protect_bos_token": true, "prevent_from_masking_myself": true, "max_steps": 4375, "warmup_steps": 250, "group": "scaling_sdpa_nh", "use_wandb": true, "kill_self_after_run": false, "random_seed": 1339, "memory_penalty_epsilon": 0.1, "selection_head_linear_combo": "none", "selection_head_linear_combo_scale": 1.0, "disable_selection_head_linear_combo_bias": false, "assert_latent_matches_no_head": false, "n_latent_masks": null, "init_latent_masks_to_identity": false, "init_latent_masks_to_inverse": false, "latent_mask_scale": null, "latent_mask_runtime_multiplier": null, "latent_mask_sigmoid": false, "S_layernorm": false, "one_head_per_latent_mask": false, "att_conv": false, "att_conv_init": "none", "att_conv_scale": 1.0, "att_conv_weight_decay": 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": 256, "batch_size": 128, "total_batch_size": 131072, "protection_head_scaling_factor": 1.0, "protection_head_bias": 0.0, "n_sliced_masks": null, "mask_layernorm": false, "residual_attention_masks": false, "compute_base_shapes": false, "base_shapes_savefile": null, "mup": true, "disable_selection": false, "mup_enable_coord_check_logging": false, "max_lr": 0.003, "decay_lr": true, "readout_zero_init": false, "query_zero_init": false, "mup_zero_init": true, "l1_loss": false, "debugpy": false, "key": "sdpa_nh_32", "latent_mask_precision": "bfloat16", "att_conv_precision": "bfloat16", "attn_precision": "bfloat16", "profile_kind": "none", "dense_attention_kind": "mha", "head_dim_value": 32, "sdpa_iter_size": null, "stabilize_attn_scores": false, "override_use_sdpa": false, "autocast_precision": "bfloat16", "simulate_micro_bs": null, "simulate_micro_bs_2": null}
scaling_sdpa_nh/sdpa_nh_32_seed_1339/dataloader_04374.pt ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:1e9422c936c4d62660261ba5451b4b72bd5055f233420bd7b0797ede3ccedd1e
3
+ size 964
scaling_sdpa_nh/sdpa_nh_32_seed_1339/log2.txt ADDED
@@ -0,0 +1,529 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ max_steps: 4375
2
+ 0 val loss 11.1894
3
+ 0 val perplexity 72359.6094
4
+ 0 train 11.199350 (lr=4.1958e-06) (hash(x)=161979327)
5
+ 10 train 9.842110 (lr=4.6154e-05) (hash(x)=165941070)
6
+ 20 train 9.173056 (lr=8.8112e-05) (hash(x)=152918033)
7
+ 30 train 8.442722 (lr=1.3007e-04) (hash(x)=169931607)
8
+ 40 train 7.690411 (lr=1.7203e-04) (hash(x)=174342675)
9
+ 50 train 7.406435 (lr=2.1399e-04) (hash(x)=163872554)
10
+ 60 train 7.399722 (lr=2.5594e-04) (hash(x)=161534908)
11
+ 70 train 7.000638 (lr=2.9790e-04) (hash(x)=156776712)
12
+ 80 train 7.005642 (lr=3.3986e-04) (hash(x)=164996929)
13
+ 90 train 6.828156 (lr=3.8182e-04) (hash(x)=152962169)
14
+ 100 val loss 6.7505
15
+ 100 val perplexity 854.4653
16
+ 100 train 6.605455 (lr=4.2378e-04) (hash(x)=147111701)
17
+ 110 train 6.621009 (lr=4.6573e-04) (hash(x)=167595270)
18
+ 120 train 6.505689 (lr=5.0769e-04) (hash(x)=152205426)
19
+ 130 train 6.401568 (lr=5.4965e-04) (hash(x)=158427335)
20
+ 140 train 6.411443 (lr=5.9161e-04) (hash(x)=174588384)
21
+ 150 train 6.485283 (lr=6.3357e-04) (hash(x)=157067270)
22
+ 160 train 6.437401 (lr=6.7552e-04) (hash(x)=161859595)
23
+ 170 train 6.363762 (lr=7.1748e-04) (hash(x)=162739537)
24
+ 180 train 6.322475 (lr=7.5944e-04) (hash(x)=170785304)
25
+ 190 train 6.236233 (lr=8.0140e-04) (hash(x)=168658088)
26
+ 200 val loss 6.1655
27
+ 200 val perplexity 476.0240
28
+ 200 train 6.200137 (lr=8.4336e-04) (hash(x)=150749086)
29
+ 210 train 6.105766 (lr=8.8531e-04) (hash(x)=148789576)
30
+ 220 train 6.045851 (lr=9.2727e-04) (hash(x)=153164408)
31
+ 230 train 6.189363 (lr=9.6923e-04) (hash(x)=184594366)
32
+ 240 train 5.898108 (lr=1.0112e-03) (hash(x)=157817319)
33
+ 250 train 5.964208 (lr=1.0531e-03) (hash(x)=149451562)
34
+ 260 train 5.865019 (lr=1.0951e-03) (hash(x)=165738161)
35
+ 270 train 5.819511 (lr=1.1371e-03) (hash(x)=165724764)
36
+ 280 train 5.743745 (lr=1.1790e-03) (hash(x)=155414136)
37
+ 290 train 5.668654 (lr=1.2210e-03) (hash(x)=165090903)
38
+ 300 val loss 5.7754
39
+ 300 val perplexity 322.2777
40
+ 300 train 5.542324 (lr=1.2629e-03) (hash(x)=154445545)
41
+ 310 train 5.474517 (lr=1.3049e-03) (hash(x)=158949603)
42
+ 320 train 5.547161 (lr=1.3469e-03) (hash(x)=161412973)
43
+ 330 train 5.649278 (lr=1.3888e-03) (hash(x)=164572249)
44
+ 340 train 5.686551 (lr=1.4308e-03) (hash(x)=162202191)
45
+ 350 train 5.653337 (lr=1.4727e-03) (hash(x)=158135392)
46
+ 360 train 5.611431 (lr=1.5147e-03) (hash(x)=163752391)
47
+ 370 train 5.620819 (lr=1.5566e-03) (hash(x)=172758799)
48
+ 380 train 5.416392 (lr=1.5986e-03) (hash(x)=144569488)
49
+ 390 train 5.385399 (lr=1.6406e-03) (hash(x)=150429229)
50
+ 400 val loss 5.3980
51
+ 400 val perplexity 220.9603
52
+ 400 train 5.385469 (lr=1.6825e-03) (hash(x)=154261530)
53
+ 410 train 5.334976 (lr=1.7245e-03) (hash(x)=159172506)
54
+ 420 train 5.331872 (lr=1.7664e-03) (hash(x)=154419867)
55
+ 430 train 5.368871 (lr=1.8084e-03) (hash(x)=167337099)
56
+ 440 train 5.258461 (lr=1.8503e-03) (hash(x)=154535255)
57
+ 450 train 5.244998 (lr=1.8923e-03) (hash(x)=153077071)
58
+ 460 train 5.155660 (lr=1.9343e-03) (hash(x)=150853070)
59
+ 470 train 4.973045 (lr=1.9762e-03) (hash(x)=160701113)
60
+ 480 train 5.106442 (lr=2.0182e-03) (hash(x)=154690870)
61
+ 490 train 4.912395 (lr=2.0601e-03) (hash(x)=163834286)
62
+ 500 val loss 5.2082
63
+ 500 val perplexity 182.7736
64
+ 500 train 4.999027 (lr=2.1021e-03) (hash(x)=162651717)
65
+ 510 train 5.131568 (lr=2.1441e-03) (hash(x)=160448311)
66
+ 520 train 5.179554 (lr=2.1860e-03) (hash(x)=165741643)
67
+ 530 train 5.288843 (lr=2.2280e-03) (hash(x)=153201264)
68
+ 540 train 5.191005 (lr=2.2699e-03) (hash(x)=159027160)
69
+ 550 train 5.138835 (lr=2.3119e-03) (hash(x)=170769095)
70
+ 560 train 5.170374 (lr=2.3538e-03) (hash(x)=153606190)
71
+ 570 train 5.140438 (lr=2.3958e-03) (hash(x)=159708689)
72
+ 580 train 5.070537 (lr=2.4378e-03) (hash(x)=179108359)
73
+ 590 train 5.079934 (lr=2.4797e-03) (hash(x)=157134241)
74
+ 600 val loss 5.0565
75
+ 600 val perplexity 157.0335
76
+ 600 train 5.084392 (lr=2.5217e-03) (hash(x)=162178356)
77
+ 610 train 4.998819 (lr=2.5636e-03) (hash(x)=167646928)
78
+ 620 train 4.960254 (lr=2.6056e-03) (hash(x)=160979601)
79
+ 630 train 4.865294 (lr=2.6476e-03) (hash(x)=158349468)
80
+ 640 train 4.860706 (lr=2.6895e-03) (hash(x)=150370348)
81
+ 650 train 4.818883 (lr=2.7315e-03) (hash(x)=152810075)
82
+ 660 train 4.757813 (lr=2.7734e-03) (hash(x)=145786262)
83
+ 670 train 4.730083 (lr=2.8154e-03) (hash(x)=160779604)
84
+ 680 train 4.629196 (lr=2.8573e-03) (hash(x)=152343688)
85
+ 690 train 4.665431 (lr=2.8993e-03) (hash(x)=166507600)
86
+ 700 val loss 4.9133
87
+ 700 val perplexity 136.0893
88
+ 700 train 4.649576 (lr=2.9413e-03) (hash(x)=162623700)
89
+ 710 train 4.932231 (lr=2.9832e-03) (hash(x)=162293403)
90
+ 720 train 4.888051 (lr=3.0000e-03) (hash(x)=159564657)
91
+ 730 train 4.897137 (lr=2.9999e-03) (hash(x)=154209107)
92
+ 740 train 4.846026 (lr=2.9997e-03) (hash(x)=156485709)
93
+ 750 train 4.804501 (lr=2.9994e-03) (hash(x)=160834702)
94
+ 760 train 4.834312 (lr=2.9990e-03) (hash(x)=152240275)
95
+ 770 train 4.809090 (lr=2.9985e-03) (hash(x)=168903075)
96
+ 780 train 4.862814 (lr=2.9979e-03) (hash(x)=166843738)
97
+ 790 train 4.892451 (lr=2.9972e-03) (hash(x)=166358104)
98
+ 800 val loss 4.7398
99
+ 800 val perplexity 114.4074
100
+ 800 train 4.710567 (lr=2.9964e-03) (hash(x)=153477747)
101
+ 810 train 4.664695 (lr=2.9955e-03) (hash(x)=157954385)
102
+ 820 train 4.665728 (lr=2.9945e-03) (hash(x)=174383505)
103
+ 830 train 4.665391 (lr=2.9934e-03) (hash(x)=145226181)
104
+ 840 train 4.679838 (lr=2.9922e-03) (hash(x)=157665743)
105
+ 850 train 4.670352 (lr=2.9909e-03) (hash(x)=158565836)
106
+ 860 train 4.656913 (lr=2.9896e-03) (hash(x)=171168105)
107
+ 870 train 5.076761 (lr=2.9881e-03) (hash(x)=159017777)
108
+ 880 train 4.811701 (lr=2.9865e-03) (hash(x)=168156768)
109
+ 890 train 4.751247 (lr=2.9848e-03) (hash(x)=145645245)
110
+ 900 val loss 4.6350
111
+ 900 val perplexity 103.0315
112
+ 900 train 4.581655 (lr=2.9830e-03) (hash(x)=151245521)
113
+ 910 train 4.601138 (lr=2.9811e-03) (hash(x)=182614672)
114
+ 920 train 4.572731 (lr=2.9792e-03) (hash(x)=167377432)
115
+ 930 train 4.523597 (lr=2.9771e-03) (hash(x)=171347228)
116
+ 940 train 4.528168 (lr=2.9749e-03) (hash(x)=171588755)
117
+ 950 train 4.635194 (lr=2.9726e-03) (hash(x)=159560168)
118
+ 960 train 4.516045 (lr=2.9703e-03) (hash(x)=171801357)
119
+ 970 train 4.519415 (lr=2.9678e-03) (hash(x)=172513884)
120
+ 980 train 4.547640 (lr=2.9652e-03) (hash(x)=170659685)
121
+ 990 train 4.577935 (lr=2.9626e-03) (hash(x)=165436358)
122
+ 1000 val loss 4.5389
123
+ 1000 val perplexity 93.5844
124
+ 1000 train 4.635589 (lr=2.9598e-03) (hash(x)=167684690)
125
+ 1010 train 4.609678 (lr=2.9570e-03) (hash(x)=127225492)
126
+ 1020 train 4.604395 (lr=2.9540e-03) (hash(x)=167575323)
127
+ 1030 train 4.539838 (lr=2.9510e-03) (hash(x)=161266302)
128
+ 1040 train 4.516424 (lr=2.9478e-03) (hash(x)=155607661)
129
+ 1050 train 4.546844 (lr=2.9446e-03) (hash(x)=165685359)
130
+ 1060 train 4.405491 (lr=2.9412e-03) (hash(x)=160417824)
131
+ 1070 train 4.542985 (lr=2.9378e-03) (hash(x)=173313882)
132
+ 1080 train 4.523608 (lr=2.9343e-03) (hash(x)=165485047)
133
+ 1090 train 4.387110 (lr=2.9307e-03) (hash(x)=159848305)
134
+ 1100 val loss 4.4833
135
+ 1100 val perplexity 88.5236
136
+ 1100 train 4.539196 (lr=2.9270e-03) (hash(x)=170890601)
137
+ 1110 train 4.470822 (lr=2.9231e-03) (hash(x)=172689983)
138
+ 1120 train 4.503944 (lr=2.9192e-03) (hash(x)=154762859)
139
+ 1130 train 4.422890 (lr=2.9152e-03) (hash(x)=168196871)
140
+ 1140 train 4.466641 (lr=2.9112e-03) (hash(x)=156715577)
141
+ 1150 train 4.576858 (lr=2.9070e-03) (hash(x)=166381788)
142
+ 1160 train 4.503989 (lr=2.9027e-03) (hash(x)=162095130)
143
+ 1170 train 4.486695 (lr=2.8983e-03) (hash(x)=166094750)
144
+ 1180 train 4.428436 (lr=2.8939e-03) (hash(x)=154965753)
145
+ 1190 train 4.405046 (lr=2.8893e-03) (hash(x)=172766447)
146
+ 1200 val loss 4.3988
147
+ 1200 val perplexity 81.3524
148
+ 1200 train 4.323671 (lr=2.8847e-03) (hash(x)=161185334)
149
+ 1210 train 4.490150 (lr=2.8800e-03) (hash(x)=167224144)
150
+ 1220 train 4.333463 (lr=2.8751e-03) (hash(x)=167252866)
151
+ 1230 train 4.353264 (lr=2.8702e-03) (hash(x)=155856924)
152
+ 1240 train 4.340919 (lr=2.8652e-03) (hash(x)=155956446)
153
+ 1250 train 4.256556 (lr=2.8601e-03) (hash(x)=140794796)
154
+ 1260 train 4.373973 (lr=2.8550e-03) (hash(x)=166863869)
155
+ 1270 train 4.433874 (lr=2.8497e-03) (hash(x)=166021892)
156
+ 1280 train 4.444230 (lr=2.8443e-03) (hash(x)=170503648)
157
+ 1290 train 4.350812 (lr=2.8389e-03) (hash(x)=164012719)
158
+ 1300 val loss 4.3637
159
+ 1300 val perplexity 78.5504
160
+ 1300 train 4.424841 (lr=2.8333e-03) (hash(x)=156519834)
161
+ 1310 train 4.368719 (lr=2.8277e-03) (hash(x)=165146652)
162
+ 1320 train 4.401155 (lr=2.8220e-03) (hash(x)=174391990)
163
+ 1330 train 4.388462 (lr=2.8162e-03) (hash(x)=161131201)
164
+ 1340 train 4.380292 (lr=2.8103e-03) (hash(x)=162240912)
165
+ 1350 train 4.298312 (lr=2.8044e-03) (hash(x)=152462931)
166
+ 1360 train 4.307581 (lr=2.7983e-03) (hash(x)=150443278)
167
+ 1370 train 4.251284 (lr=2.7922e-03) (hash(x)=153721711)
168
+ 1380 train 4.387814 (lr=2.7860e-03) (hash(x)=158070628)
169
+ 1390 train 4.410990 (lr=2.7797e-03) (hash(x)=163318827)
170
+ 1400 val loss 4.3516
171
+ 1400 val perplexity 77.6006
172
+ 1400 train 4.314656 (lr=2.7733e-03) (hash(x)=154210336)
173
+ 1410 train 4.371486 (lr=2.7668e-03) (hash(x)=162393933)
174
+ 1420 train 4.366269 (lr=2.7603e-03) (hash(x)=156659308)
175
+ 1430 train 4.331207 (lr=2.7536e-03) (hash(x)=171243439)
176
+ 1440 train 4.474599 (lr=2.7469e-03) (hash(x)=178459521)
177
+ 1450 train 4.321321 (lr=2.7401e-03) (hash(x)=159166350)
178
+ 1460 train 4.245605 (lr=2.7333e-03) (hash(x)=173554569)
179
+ 1470 train 4.348864 (lr=2.7263e-03) (hash(x)=158362521)
180
+ 1480 train 4.304917 (lr=2.7193e-03) (hash(x)=165899207)
181
+ 1490 train 4.340619 (lr=2.7121e-03) (hash(x)=156470939)
182
+ 1500 val loss 4.2992
183
+ 1500 val perplexity 73.6380
184
+ 1500 train 4.310606 (lr=2.7050e-03) (hash(x)=164197780)
185
+ 1510 train 4.209189 (lr=2.6977e-03) (hash(x)=146709896)
186
+ 1520 train 4.227100 (lr=2.6903e-03) (hash(x)=151304838)
187
+ 1530 train 4.256722 (lr=2.6829e-03) (hash(x)=163187333)
188
+ 1540 train 4.210342 (lr=2.6754e-03) (hash(x)=153630537)
189
+ 1550 train 4.225683 (lr=2.6678e-03) (hash(x)=157299103)
190
+ 1560 train 4.139333 (lr=2.6602e-03) (hash(x)=153786477)
191
+ 1570 train 4.457505 (lr=2.6525e-03) (hash(x)=155037209)
192
+ 1580 train 4.335843 (lr=2.6447e-03) (hash(x)=146157308)
193
+ 1590 train 4.381704 (lr=2.6368e-03) (hash(x)=147823317)
194
+ 1600 val loss 4.2792
195
+ 1600 val perplexity 72.1797
196
+ 1600 train 4.369462 (lr=2.6289e-03) (hash(x)=169140159)
197
+ 1610 train 4.142849 (lr=2.6208e-03) (hash(x)=157811493)
198
+ 1620 train 4.245614 (lr=2.6128e-03) (hash(x)=168940851)
199
+ 1630 train 4.231154 (lr=2.6046e-03) (hash(x)=151149213)
200
+ 1640 train 4.333345 (lr=2.5964e-03) (hash(x)=161822009)
201
+ 1650 train 4.750437 (lr=2.5881e-03) (hash(x)=151056313)
202
+ 1660 train 4.250002 (lr=2.5797e-03) (hash(x)=156920306)
203
+ 1670 train 4.164579 (lr=2.5713e-03) (hash(x)=159695815)
204
+ 1680 train 4.179887 (lr=2.5628e-03) (hash(x)=160232907)
205
+ 1690 train 4.167913 (lr=2.5542e-03) (hash(x)=158099357)
206
+ 1700 val loss 4.2783
207
+ 1700 val perplexity 72.1149
208
+ 1700 train 4.232100 (lr=2.5455e-03) (hash(x)=164127381)
209
+ 1710 train 4.392782 (lr=2.5368e-03) (hash(x)=156592040)
210
+ 1720 train 4.317292 (lr=2.5281e-03) (hash(x)=160220265)
211
+ 1730 train 4.253538 (lr=2.5192e-03) (hash(x)=156162204)
212
+ 1740 train 4.251254 (lr=2.5103e-03) (hash(x)=152767893)
213
+ 1750 train 4.232057 (lr=2.5014e-03) (hash(x)=162707306)
214
+ 1760 train 4.261207 (lr=2.4924e-03) (hash(x)=154472004)
215
+ 1770 train 4.118498 (lr=2.4833e-03) (hash(x)=160374234)
216
+ 1780 train 4.275365 (lr=2.4741e-03) (hash(x)=159709225)
217
+ 1790 train 4.306963 (lr=2.4649e-03) (hash(x)=152095944)
218
+ 1800 val loss 4.2274
219
+ 1800 val perplexity 68.5402
220
+ 1800 train 4.286317 (lr=2.4556e-03) (hash(x)=157920255)
221
+ 1810 train 4.033162 (lr=2.4463e-03) (hash(x)=167237680)
222
+ 1820 train 4.143323 (lr=2.4369e-03) (hash(x)=161255774)
223
+ 1830 train 4.172377 (lr=2.4275e-03) (hash(x)=154275481)
224
+ 1840 train 4.116464 (lr=2.4180e-03) (hash(x)=152398178)
225
+ 1850 train 4.222289 (lr=2.4084e-03) (hash(x)=147646022)
226
+ 1860 train 4.252814 (lr=2.3988e-03) (hash(x)=158895559)
227
+ 1870 train 4.170678 (lr=2.3891e-03) (hash(x)=162742598)
228
+ 1880 train 4.179760 (lr=2.3794e-03) (hash(x)=169911029)
229
+ 1890 train 4.210052 (lr=2.3696e-03) (hash(x)=172068270)
230
+ 1900 val loss 4.2225
231
+ 1900 val perplexity 68.2027
232
+ 1900 train 4.300045 (lr=2.3598e-03) (hash(x)=159975457)
233
+ 1910 train 4.296904 (lr=2.3499e-03) (hash(x)=178775667)
234
+ 1920 train 4.168103 (lr=2.3400e-03) (hash(x)=151180831)
235
+ 1930 train 4.182503 (lr=2.3300e-03) (hash(x)=154239412)
236
+ 1940 train 4.188622 (lr=2.3200e-03) (hash(x)=139144465)
237
+ 1950 train 4.147552 (lr=2.3099e-03) (hash(x)=194482413)
238
+ 1960 train 4.164912 (lr=2.2998e-03) (hash(x)=155812344)
239
+ 1970 train 4.171468 (lr=2.2896e-03) (hash(x)=157577031)
240
+ 1980 train 4.205054 (lr=2.2793e-03) (hash(x)=152045027)
241
+ 1990 train 4.112964 (lr=2.2691e-03) (hash(x)=141742221)
242
+ 2000 val loss 4.1775
243
+ 2000 val perplexity 65.2017
244
+ 2000 train 4.228210 (lr=2.2588e-03) (hash(x)=160555812)
245
+ 2010 train 4.154474 (lr=2.2484e-03) (hash(x)=160371335)
246
+ 2020 train 4.155124 (lr=2.2380e-03) (hash(x)=167890767)
247
+ 2030 train 4.229740 (lr=2.2275e-03) (hash(x)=165127715)
248
+ 2040 train 4.173424 (lr=2.2170e-03) (hash(x)=151587302)
249
+ 2050 train 4.187147 (lr=2.2065e-03) (hash(x)=172347989)
250
+ 2060 train 4.270131 (lr=2.1959e-03) (hash(x)=161301919)
251
+ 2070 train 4.205544 (lr=2.1853e-03) (hash(x)=161472934)
252
+ 2080 train 4.292228 (lr=2.1746e-03) (hash(x)=159777322)
253
+ 2090 train 4.203932 (lr=2.1639e-03) (hash(x)=151149770)
254
+ 2100 val loss 4.1540
255
+ 2100 val perplexity 63.6911
256
+ 2100 train 4.031109 (lr=2.1532e-03) (hash(x)=160283520)
257
+ 2110 train 4.203350 (lr=2.1424e-03) (hash(x)=154222025)
258
+ 2120 train 4.141645 (lr=2.1316e-03) (hash(x)=155638178)
259
+ 2130 train 4.061255 (lr=2.1208e-03) (hash(x)=149125384)
260
+ 2140 train 4.114100 (lr=2.1099e-03) (hash(x)=159496074)
261
+ 2150 train 4.203867 (lr=2.0990e-03) (hash(x)=162019154)
262
+ 2160 train 4.358819 (lr=2.0881e-03) (hash(x)=155466928)
263
+ 2170 train 4.232115 (lr=2.0771e-03) (hash(x)=173469994)
264
+ 2180 train 4.336219 (lr=2.0661e-03) (hash(x)=163448753)
265
+ 2190 train 4.169088 (lr=2.0550e-03) (hash(x)=166595123)
266
+ 2200 val loss 4.1331
267
+ 2200 val perplexity 62.3686
268
+ 2200 train 4.052068 (lr=2.0440e-03) (hash(x)=165958875)
269
+ 2210 train 4.196562 (lr=2.0329e-03) (hash(x)=165019039)
270
+ 2220 train 4.173222 (lr=2.0217e-03) (hash(x)=167325225)
271
+ 2230 train 4.006720 (lr=2.0106e-03) (hash(x)=154287439)
272
+ 2240 train 4.066934 (lr=1.9994e-03) (hash(x)=158565817)
273
+ 2250 train 4.037912 (lr=1.9882e-03) (hash(x)=160592835)
274
+ 2260 train 4.083261 (lr=1.9770e-03) (hash(x)=144591648)
275
+ 2270 train 3.956563 (lr=1.9657e-03) (hash(x)=150486876)
276
+ 2280 train 4.153441 (lr=1.9544e-03) (hash(x)=159194788)
277
+ 2290 train 4.226479 (lr=1.9431e-03) (hash(x)=164315345)
278
+ 2300 val loss 4.1103
279
+ 2300 val perplexity 60.9640
280
+ 2300 train 4.148618 (lr=1.9318e-03) (hash(x)=170470665)
281
+ 2310 train 4.108979 (lr=1.9205e-03) (hash(x)=176539068)
282
+ 2320 train 4.089654 (lr=1.9091e-03) (hash(x)=161008474)
283
+ 2330 train 4.169914 (lr=1.8977e-03) (hash(x)=168474380)
284
+ 2340 train 4.167517 (lr=1.8863e-03) (hash(x)=156072131)
285
+ 2350 train 4.252087 (lr=1.8749e-03) (hash(x)=165127087)
286
+ 2360 train 4.246985 (lr=1.8635e-03) (hash(x)=159462801)
287
+ 2370 train 4.087796 (lr=1.8520e-03) (hash(x)=161698394)
288
+ 2380 train 4.017075 (lr=1.8406e-03) (hash(x)=165588078)
289
+ 2390 train 4.011868 (lr=1.8291e-03) (hash(x)=171673672)
290
+ 2400 val loss 4.1042
291
+ 2400 val perplexity 60.5956
292
+ 2400 train 4.085863 (lr=1.8176e-03) (hash(x)=157381516)
293
+ 2410 train 4.093005 (lr=1.8061e-03) (hash(x)=163238532)
294
+ 2420 train 4.043815 (lr=1.7946e-03) (hash(x)=156178237)
295
+ 2430 train 3.928093 (lr=1.7830e-03) (hash(x)=155080617)
296
+ 2440 train 4.031720 (lr=1.7715e-03) (hash(x)=167490538)
297
+ 2450 train 4.000449 (lr=1.7600e-03) (hash(x)=163757352)
298
+ 2460 train 3.959746 (lr=1.7484e-03) (hash(x)=156734811)
299
+ 2470 train 3.956488 (lr=1.7368e-03) (hash(x)=161522341)
300
+ 2480 train 4.218634 (lr=1.7253e-03) (hash(x)=174032060)
301
+ 2490 train 4.184583 (lr=1.7137e-03) (hash(x)=163065818)
302
+ 2500 val loss 4.0779
303
+ 2500 val perplexity 59.0216
304
+ 2500 train 4.236787 (lr=1.7021e-03) (hash(x)=126959780)
305
+ 2510 train 4.162998 (lr=1.6906e-03) (hash(x)=153844939)
306
+ 2520 train 4.146119 (lr=1.6790e-03) (hash(x)=174018174)
307
+ 2530 train 4.114050 (lr=1.6674e-03) (hash(x)=150568427)
308
+ 2540 train 4.121467 (lr=1.6558e-03) (hash(x)=161725206)
309
+ 2550 train 4.081499 (lr=1.6442e-03) (hash(x)=181346417)
310
+ 2560 train 4.073201 (lr=1.6326e-03) (hash(x)=168730011)
311
+ 2570 train 3.960889 (lr=1.6210e-03) (hash(x)=155900214)
312
+ 2580 train 4.014501 (lr=1.6094e-03) (hash(x)=160014621)
313
+ 2590 train 3.992140 (lr=1.5979e-03) (hash(x)=148996689)
314
+ 2600 val loss 4.0586
315
+ 2600 val perplexity 57.8915
316
+ 2600 train 4.066885 (lr=1.5863e-03) (hash(x)=145183968)
317
+ 2610 train 3.807775 (lr=1.5747e-03) (hash(x)=159420240)
318
+ 2620 train 3.786449 (lr=1.5632e-03) (hash(x)=151169903)
319
+ 2630 train 3.799556 (lr=1.5516e-03) (hash(x)=149454811)
320
+ 2640 train 3.868265 (lr=1.5400e-03) (hash(x)=155603914)
321
+ 2650 train 3.857804 (lr=1.5285e-03) (hash(x)=169545195)
322
+ 2660 train 4.215327 (lr=1.5170e-03) (hash(x)=168454079)
323
+ 2670 train 4.109705 (lr=1.5054e-03) (hash(x)=163110417)
324
+ 2680 train 4.053902 (lr=1.4939e-03) (hash(x)=166075603)
325
+ 2690 train 4.088150 (lr=1.4824e-03) (hash(x)=174132258)
326
+ 2700 val loss 4.0493
327
+ 2700 val perplexity 57.3545
328
+ 2700 train 4.095256 (lr=1.4709e-03) (hash(x)=169846940)
329
+ 2710 train 3.993697 (lr=1.4594e-03) (hash(x)=151553731)
330
+ 2720 train 4.095570 (lr=1.4480e-03) (hash(x)=164024255)
331
+ 2730 train 4.111537 (lr=1.4365e-03) (hash(x)=156458097)
332
+ 2740 train 4.047830 (lr=1.4251e-03) (hash(x)=154663748)
333
+ 2750 train 3.906610 (lr=1.4137e-03) (hash(x)=157548181)
334
+ 2760 train 4.018479 (lr=1.4023e-03) (hash(x)=151584783)
335
+ 2770 train 4.044423 (lr=1.3909e-03) (hash(x)=152692594)
336
+ 2780 train 4.011673 (lr=1.3795e-03) (hash(x)=165952919)
337
+ 2790 train 3.910402 (lr=1.3682e-03) (hash(x)=157484183)
338
+ 2800 val loss 4.0318
339
+ 2800 val perplexity 56.3600
340
+ 2800 train 3.986262 (lr=1.3569e-03) (hash(x)=154790034)
341
+ 2810 train 3.840214 (lr=1.3456e-03) (hash(x)=157856097)
342
+ 2820 train 3.825213 (lr=1.3343e-03) (hash(x)=168903939)
343
+ 2830 train 3.944788 (lr=1.3230e-03) (hash(x)=167001475)
344
+ 2840 train 3.773647 (lr=1.3118e-03) (hash(x)=158987150)
345
+ 2850 train 4.009000 (lr=1.3006e-03) (hash(x)=168695407)
346
+ 2860 train 4.143414 (lr=1.2894e-03) (hash(x)=164999034)
347
+ 2870 train 3.989305 (lr=1.2783e-03) (hash(x)=191197316)
348
+ 2880 train 3.947282 (lr=1.2671e-03) (hash(x)=159625200)
349
+ 2890 train 4.110631 (lr=1.2560e-03) (hash(x)=157756126)
350
+ 2900 val loss 4.0043
351
+ 2900 val perplexity 54.8339
352
+ 2900 train 3.962940 (lr=1.2450e-03) (hash(x)=164721377)
353
+ 2910 train 4.015709 (lr=1.2339e-03) (hash(x)=155873181)
354
+ 2920 train 3.964198 (lr=1.2229e-03) (hash(x)=156477082)
355
+ 2930 train 3.962638 (lr=1.2119e-03) (hash(x)=151261633)
356
+ 2940 train 3.909863 (lr=1.2010e-03) (hash(x)=160090385)
357
+ 2950 train 4.053516 (lr=1.1901e-03) (hash(x)=157300618)
358
+ 2960 train 3.966956 (lr=1.1792e-03) (hash(x)=163095555)
359
+ 2970 train 4.001521 (lr=1.1684e-03) (hash(x)=149649212)
360
+ 2980 train 3.799201 (lr=1.1576e-03) (hash(x)=174517675)
361
+ 2990 train 3.769811 (lr=1.1468e-03) (hash(x)=161344491)
362
+ 3000 val loss 4.0074
363
+ 3000 val perplexity 55.0046
364
+ 3000 train 3.801960 (lr=1.1361e-03) (hash(x)=163816905)
365
+ 3010 train 3.711115 (lr=1.1254e-03) (hash(x)=157420680)
366
+ 3020 train 3.820984 (lr=1.1147e-03) (hash(x)=153843035)
367
+ 3030 train 4.010468 (lr=1.1041e-03) (hash(x)=149770373)
368
+ 3040 train 4.074637 (lr=1.0935e-03) (hash(x)=165320741)
369
+ 3050 train 4.051455 (lr=1.0830e-03) (hash(x)=170181151)
370
+ 3060 train 3.963952 (lr=1.0725e-03) (hash(x)=163837230)
371
+ 3070 train 3.966324 (lr=1.0620e-03) (hash(x)=156812797)
372
+ 3080 train 3.942455 (lr=1.0516e-03) (hash(x)=143663627)
373
+ 3090 train 3.965731 (lr=1.0412e-03) (hash(x)=165052478)
374
+ 3100 val loss 3.9821
375
+ 3100 val perplexity 53.6308
376
+ 3100 train 3.901545 (lr=1.0309e-03) (hash(x)=151458705)
377
+ 3110 train 3.947926 (lr=1.0207e-03) (hash(x)=158228317)
378
+ 3120 train 4.127236 (lr=1.0104e-03) (hash(x)=149856801)
379
+ 3130 train 3.787742 (lr=1.0002e-03) (hash(x)=146297847)
380
+ 3140 train 3.892088 (lr=9.9011e-04) (hash(x)=172795167)
381
+ 3150 train 3.944777 (lr=9.8002e-04) (hash(x)=145936074)
382
+ 3160 train 4.032974 (lr=9.6999e-04) (hash(x)=157564303)
383
+ 3170 train 4.134198 (lr=9.6000e-04) (hash(x)=157451362)
384
+ 3180 train 4.043130 (lr=9.5007e-04) (hash(x)=156366043)
385
+ 3190 train 4.006887 (lr=9.4019e-04) (hash(x)=162344697)
386
+ 3200 val loss 3.9607
387
+ 3200 val perplexity 52.4958
388
+ 3200 train 4.024298 (lr=9.3036e-04) (hash(x)=164206298)
389
+ 3210 train 4.104543 (lr=9.2058e-04) (hash(x)=149468251)
390
+ 3220 train 4.033402 (lr=9.1085e-04) (hash(x)=157808606)
391
+ 3230 train 4.167614 (lr=9.0118e-04) (hash(x)=157078690)
392
+ 3240 train 3.973851 (lr=8.9157e-04) (hash(x)=165699696)
393
+ 3250 train 3.923892 (lr=8.8201e-04) (hash(x)=181352842)
394
+ 3260 train 4.021019 (lr=8.7251e-04) (hash(x)=142534383)
395
+ 3270 train 3.979468 (lr=8.6307e-04) (hash(x)=165297359)
396
+ 3280 train 4.024129 (lr=8.5368e-04) (hash(x)=160326815)
397
+ 3290 train 3.943507 (lr=8.4435e-04) (hash(x)=150043063)
398
+ 3300 val loss 3.9550
399
+ 3300 val perplexity 52.1941
400
+ 3300 train 3.898534 (lr=8.3508e-04) (hash(x)=164133091)
401
+ 3310 train 3.840547 (lr=8.2588e-04) (hash(x)=145803863)
402
+ 3320 train 3.866417 (lr=8.1673e-04) (hash(x)=157282009)
403
+ 3330 train 3.799544 (lr=8.0764e-04) (hash(x)=159126415)
404
+ 3340 train 4.004915 (lr=7.9862e-04) (hash(x)=164426675)
405
+ 3350 train 4.035877 (lr=7.8966e-04) (hash(x)=170424924)
406
+ 3360 train 3.976892 (lr=7.8076e-04) (hash(x)=163161621)
407
+ 3370 train 4.040255 (lr=7.7192e-04) (hash(x)=160786731)
408
+ 3380 train 3.996120 (lr=7.6315e-04) (hash(x)=165954688)
409
+ 3390 train 4.054119 (lr=7.5445e-04) (hash(x)=147371107)
410
+ 3400 val loss 3.9328
411
+ 3400 val perplexity 51.0476
412
+ 3400 train 3.995572 (lr=7.4581e-04) (hash(x)=153067604)
413
+ 3410 train 3.987434 (lr=7.3724e-04) (hash(x)=161748196)
414
+ 3420 train 3.998423 (lr=7.2874e-04) (hash(x)=168297963)
415
+ 3430 train 3.991735 (lr=7.2030e-04) (hash(x)=145086892)
416
+ 3440 train 3.895310 (lr=7.1193e-04) (hash(x)=173125711)
417
+ 3450 train 4.003996 (lr=7.0363e-04) (hash(x)=158080216)
418
+ 3460 train 3.821947 (lr=6.9541e-04) (hash(x)=158055750)
419
+ 3470 train 3.913529 (lr=6.8725e-04) (hash(x)=151381400)
420
+ 3480 train 3.857453 (lr=6.7916e-04) (hash(x)=147262346)
421
+ 3490 train 3.833146 (lr=6.7114e-04) (hash(x)=154740086)
422
+ 3500 val loss 3.9300
423
+ 3500 val perplexity 50.9086
424
+ 3500 train 3.752352 (lr=6.6320e-04) (hash(x)=145730968)
425
+ 3510 train 3.900675 (lr=6.5533e-04) (hash(x)=151355929)
426
+ 3520 train 3.888025 (lr=6.4753e-04) (hash(x)=165136835)
427
+ 3530 train 3.964739 (lr=6.3981e-04) (hash(x)=163869056)
428
+ 3540 train 3.906909 (lr=6.3216e-04) (hash(x)=161580134)
429
+ 3550 train 3.920521 (lr=6.2458e-04) (hash(x)=168595690)
430
+ 3560 train 3.954974 (lr=6.1708e-04) (hash(x)=158576466)
431
+ 3570 train 3.922025 (lr=6.0966e-04) (hash(x)=157298619)
432
+ 3580 train 4.007148 (lr=6.0231e-04) (hash(x)=160892091)
433
+ 3590 train 3.873895 (lr=5.9504e-04) (hash(x)=165168008)
434
+ 3600 val loss 3.9107
435
+ 3600 val perplexity 49.9334
436
+ 3600 train 3.978968 (lr=5.8785e-04) (hash(x)=144911638)
437
+ 3610 train 3.930417 (lr=5.8074e-04) (hash(x)=150614513)
438
+ 3620 train 3.924148 (lr=5.7370e-04) (hash(x)=167695016)
439
+ 3630 train 3.844071 (lr=5.6675e-04) (hash(x)=182708233)
440
+ 3640 train 3.949391 (lr=5.5987e-04) (hash(x)=154146274)
441
+ 3650 train 3.915379 (lr=5.5308e-04) (hash(x)=159129968)
442
+ 3660 train 3.827741 (lr=5.4636e-04) (hash(x)=146209497)
443
+ 3670 train 3.848766 (lr=5.3973e-04) (hash(x)=145578104)
444
+ 3680 train 3.837877 (lr=5.3318e-04) (hash(x)=150243521)
445
+ 3690 train 3.796276 (lr=5.2671e-04) (hash(x)=154328313)
446
+ 3700 val loss 3.9099
447
+ 3700 val perplexity 49.8960
448
+ 3700 train 3.769648 (lr=5.2033e-04) (hash(x)=159904582)
449
+ 3710 train 4.100586 (lr=5.1402e-04) (hash(x)=175986417)
450
+ 3720 train 4.022589 (lr=5.0780e-04) (hash(x)=158509479)
451
+ 3730 train 4.031912 (lr=5.0167e-04) (hash(x)=156087760)
452
+ 3740 train 4.030684 (lr=4.9562e-04) (hash(x)=177667170)
453
+ 3750 train 4.013712 (lr=4.8965e-04) (hash(x)=145256842)
454
+ 3760 train 3.970976 (lr=4.8377e-04) (hash(x)=149855635)
455
+ 3770 train 4.004295 (lr=4.7798e-04) (hash(x)=168798886)
456
+ 3780 train 3.920371 (lr=4.7227e-04) (hash(x)=156277179)
457
+ 3790 train 3.887872 (lr=4.6665e-04) (hash(x)=149698945)
458
+ 3800 val loss 3.8932
459
+ 3800 val perplexity 49.0683
460
+ 3800 train 3.844043 (lr=4.6112e-04) (hash(x)=168039654)
461
+ 3810 train 3.962731 (lr=4.5567e-04) (hash(x)=154595574)
462
+ 3820 train 4.016512 (lr=4.5031e-04) (hash(x)=158303284)
463
+ 3830 train 4.030608 (lr=4.4504e-04) (hash(x)=164128644)
464
+ 3840 train 3.977919 (lr=4.3986e-04) (hash(x)=166842997)
465
+ 3850 train 3.895818 (lr=4.3477e-04) (hash(x)=177681163)
466
+ 3860 train 3.866384 (lr=4.2977e-04) (hash(x)=177062575)
467
+ 3870 train 3.911023 (lr=4.2486e-04) (hash(x)=158957159)
468
+ 3880 train 3.842605 (lr=4.2004e-04) (hash(x)=151567740)
469
+ 3890 train 3.918954 (lr=4.1530e-04) (hash(x)=153261675)
470
+ 3900 val loss 3.8814
471
+ 3900 val perplexity 48.4939
472
+ 3900 train 4.193359 (lr=4.1066e-04) (hash(x)=160268683)
473
+ 3910 train 3.909543 (lr=4.0611e-04) (hash(x)=153745240)
474
+ 3920 train 3.961729 (lr=4.0166e-04) (hash(x)=164537824)
475
+ 3930 train 3.914365 (lr=3.9729e-04) (hash(x)=162062116)
476
+ 3940 train 3.870928 (lr=3.9302e-04) (hash(x)=168626742)
477
+ 3950 train 3.933808 (lr=3.8884e-04) (hash(x)=148753561)
478
+ 3960 train 3.864079 (lr=3.8475e-04) (hash(x)=163526242)
479
+ 3970 train 4.082196 (lr=3.8076e-04) (hash(x)=159329323)
480
+ 3980 train 4.005297 (lr=3.7685e-04) (hash(x)=146502339)
481
+ 3990 train 3.701346 (lr=3.7305e-04) (hash(x)=158409708)
482
+ 4000 val loss 3.8770
483
+ 4000 val perplexity 48.2815
484
+ 4000 train 3.878196 (lr=3.6933e-04) (hash(x)=154645137)
485
+ 4010 train 3.883752 (lr=3.6572e-04) (hash(x)=160044026)
486
+ 4020 train 3.863085 (lr=3.6219e-04) (hash(x)=155544686)
487
+ 4030 train 3.844070 (lr=3.5876e-04) (hash(x)=147589421)
488
+ 4040 train 3.954780 (lr=3.5543e-04) (hash(x)=158607990)
489
+ 4050 train 3.865170 (lr=3.5219e-04) (hash(x)=154108872)
490
+ 4060 train 3.940969 (lr=3.4905e-04) (hash(x)=167101465)
491
+ 4070 train 3.902925 (lr=3.4600e-04) (hash(x)=160323363)
492
+ 4080 train 3.895875 (lr=3.4305e-04) (hash(x)=150423692)
493
+ 4090 train 3.873333 (lr=3.4019e-04) (hash(x)=167461885)
494
+ 4100 val loss 3.8650
495
+ 4100 val perplexity 47.7030
496
+ 4100 train 3.905963 (lr=3.3744e-04) (hash(x)=151783377)
497
+ 4110 train 3.902719 (lr=3.3477e-04) (hash(x)=165679649)
498
+ 4120 train 3.803249 (lr=3.3221e-04) (hash(x)=153643039)
499
+ 4130 train 3.856675 (lr=3.2974e-04) (hash(x)=149648749)
500
+ 4140 train 3.889102 (lr=3.2737e-04) (hash(x)=162751418)
501
+ 4150 train 3.884369 (lr=3.2510e-04) (hash(x)=157047478)
502
+ 4160 train 3.818530 (lr=3.2292e-04) (hash(x)=156391812)
503
+ 4170 train 3.890290 (lr=3.2085e-04) (hash(x)=180926985)
504
+ 4180 train 3.929876 (lr=3.1887e-04) (hash(x)=168699059)
505
+ 4190 train 3.856908 (lr=3.1699e-04) (hash(x)=155410372)
506
+ 4200 val loss 3.8643
507
+ 4200 val perplexity 47.6677
508
+ 4200 train 3.866704 (lr=3.1520e-04) (hash(x)=152615252)
509
+ 4210 train 3.830707 (lr=3.1352e-04) (hash(x)=155679453)
510
+ 4220 train 3.950214 (lr=3.1193e-04) (hash(x)=152954399)
511
+ 4230 train 3.946784 (lr=3.1044e-04) (hash(x)=164608084)
512
+ 4240 train 3.977905 (lr=3.0905e-04) (hash(x)=162463867)
513
+ 4250 train 3.887748 (lr=3.0776e-04) (hash(x)=168832088)
514
+ 4260 train 3.859124 (lr=3.0657e-04) (hash(x)=162698882)
515
+ 4270 train 3.895552 (lr=3.0548e-04) (hash(x)=180861492)
516
+ 4280 train 3.784168 (lr=3.0449e-04) (hash(x)=160824300)
517
+ 4290 train 3.847311 (lr=3.0359e-04) (hash(x)=157750938)
518
+ 4300 val loss 3.8606
519
+ 4300 val perplexity 47.4958
520
+ 4300 train 3.794927 (lr=3.0280e-04) (hash(x)=157216833)
521
+ 4310 train 3.786776 (lr=3.0210e-04) (hash(x)=176184661)
522
+ 4320 train 3.959943 (lr=3.0150e-04) (hash(x)=161537117)
523
+ 4330 train 3.905605 (lr=3.0101e-04) (hash(x)=160700216)
524
+ 4340 train 3.852973 (lr=3.0061e-04) (hash(x)=175519903)
525
+ 4350 train 3.861788 (lr=3.0031e-04) (hash(x)=167988701)
526
+ 4360 train 3.819558 (lr=3.0011e-04) (hash(x)=162619084)
527
+ 4370 train 4.208399 (lr=3.0001e-04) (hash(x)=166292684)
528
+ 4374 val loss 3.8544
529
+ 4374 val perplexity 47.1983
scaling_sdpa_nh/sdpa_nh_32_seed_1339/model_04374.pt ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:cc1d4620a26f8f8f3277ad6fa38780f19e8a0594e2719d9814b84a66da1e9fa6
3
+ size 127551402
scaling_sdpa_nh/sdpa_nh_32_seed_1339/optimizer_04374.pt ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:3f5a208be45e7cc96b147abaceaddd02747256d293148a22d57ed02e74ee5d6b
3
+ size 255125126