andrew-healey commited on
Commit
f273a3a
·
verified ·
1 Parent(s): 974b811

Upload folder using huggingface_hub

Browse files
scaling_sdpa_granularity_2/hd_2_seed_1340/args.json ADDED
@@ -0,0 +1 @@
 
 
1
+ {"hellaswag": true, "attention_kind": "self", "log_dir": "scaling_sdpa_granularity_2/hd_2_seed_1340", "resume_checkpoint": null, "resume_optimizer": false, "add_a_head": false, "add_head_to_start": true, "new_head_init": "normal", "n_heads": 128, "n_embd": 256, "head_dim": 2, "protect_bos_token": true, "prevent_from_masking_myself": true, "max_steps": 4375, "warmup_steps": 250, "group": "scaling_sdpa_granularity_2", "use_wandb": true, "kill_self_after_run": false, "random_seed": 1340, "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": "hd_2", "latent_mask_precision": "bfloat16", "att_conv_precision": "bfloat16", "attn_precision": "bfloat16", "profile_kind": "none", "dense_attention_kind": "mha", "head_dim_value": 2, "sdpa_iter_size": null, "stabilize_attn_scores": false, "override_use_sdpa": false, "autocast_precision": "bfloat16", "simulate_micro_bs": null, "simulate_micro_bs_2": null, "ckpt_attn": false, "c_proj_scale_init": 1.0}
scaling_sdpa_granularity_2/hd_2_seed_1340/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_granularity_2/hd_2_seed_1340/log2.txt ADDED
@@ -0,0 +1,529 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ max_steps: 4375
2
+ 0 val loss 11.3272
3
+ 0 val perplexity 83052.5312
4
+ 0 train 11.327577 (lr=4.1958e-06) (hash(x)=173692205)
5
+ 10 train 9.911880 (lr=4.6154e-05) (hash(x)=166272335)
6
+ 20 train 9.313235 (lr=8.8112e-05) (hash(x)=152988153)
7
+ 30 train 8.540108 (lr=1.3007e-04) (hash(x)=165375545)
8
+ 40 train 7.747795 (lr=1.7203e-04) (hash(x)=160434435)
9
+ 50 train 7.368715 (lr=2.1399e-04) (hash(x)=155536673)
10
+ 60 train 7.183732 (lr=2.5594e-04) (hash(x)=152324251)
11
+ 70 train 7.128593 (lr=2.9790e-04) (hash(x)=152719350)
12
+ 80 train 7.014344 (lr=3.3986e-04) (hash(x)=164461998)
13
+ 90 train 6.833706 (lr=3.8182e-04) (hash(x)=160609072)
14
+ 100 val loss 6.7698
15
+ 100 val perplexity 871.1029
16
+ 100 train 6.754338 (lr=4.2378e-04) (hash(x)=166364183)
17
+ 110 train 6.703378 (lr=4.6573e-04) (hash(x)=163817364)
18
+ 120 train 6.628330 (lr=5.0769e-04) (hash(x)=165919909)
19
+ 130 train 6.521034 (lr=5.4965e-04) (hash(x)=168185952)
20
+ 140 train 6.425391 (lr=5.9161e-04) (hash(x)=157062392)
21
+ 150 train 6.416339 (lr=6.3357e-04) (hash(x)=162070558)
22
+ 160 train 6.338185 (lr=6.7552e-04) (hash(x)=167448687)
23
+ 170 train 6.401462 (lr=7.1748e-04) (hash(x)=163268047)
24
+ 180 train 6.429371 (lr=7.5944e-04) (hash(x)=162958229)
25
+ 190 train 6.204580 (lr=8.0140e-04) (hash(x)=158951412)
26
+ 200 val loss 6.1822
27
+ 200 val perplexity 484.0721
28
+ 200 train 6.149215 (lr=8.4336e-04) (hash(x)=160898816)
29
+ 210 train 6.093425 (lr=8.8531e-04) (hash(x)=153803852)
30
+ 220 train 6.045036 (lr=9.2727e-04) (hash(x)=150534761)
31
+ 230 train 6.076596 (lr=9.6923e-04) (hash(x)=167269298)
32
+ 240 train 6.116091 (lr=1.0112e-03) (hash(x)=158873716)
33
+ 250 train 6.077937 (lr=1.0531e-03) (hash(x)=163730102)
34
+ 260 train 6.020314 (lr=1.0951e-03) (hash(x)=173556083)
35
+ 270 train 5.908637 (lr=1.1371e-03) (hash(x)=153796179)
36
+ 280 train 5.896715 (lr=1.1790e-03) (hash(x)=168523171)
37
+ 290 train 5.836365 (lr=1.2210e-03) (hash(x)=145409987)
38
+ 300 val loss 5.8192
39
+ 300 val perplexity 336.6917
40
+ 300 train 5.840630 (lr=1.2629e-03) (hash(x)=157831091)
41
+ 310 train 5.683897 (lr=1.3049e-03) (hash(x)=161940559)
42
+ 320 train 5.768365 (lr=1.3469e-03) (hash(x)=151097783)
43
+ 330 train 5.678987 (lr=1.3888e-03) (hash(x)=149560294)
44
+ 340 train 5.665404 (lr=1.4308e-03) (hash(x)=164151118)
45
+ 350 train 5.577121 (lr=1.4727e-03) (hash(x)=154545714)
46
+ 360 train 5.535557 (lr=1.5147e-03) (hash(x)=168490983)
47
+ 370 train 5.550279 (lr=1.5566e-03) (hash(x)=157243384)
48
+ 380 train 5.529683 (lr=1.5986e-03) (hash(x)=161439594)
49
+ 390 train 5.499643 (lr=1.6406e-03) (hash(x)=158011549)
50
+ 400 val loss 5.4721
51
+ 400 val perplexity 237.9690
52
+ 400 train 5.556633 (lr=1.6825e-03) (hash(x)=161977759)
53
+ 410 train 5.502205 (lr=1.7245e-03) (hash(x)=169755458)
54
+ 420 train 5.482793 (lr=1.7664e-03) (hash(x)=165992049)
55
+ 430 train 5.470453 (lr=1.8084e-03) (hash(x)=167325398)
56
+ 440 train 5.428369 (lr=1.8503e-03) (hash(x)=138405929)
57
+ 450 train 5.399467 (lr=1.8923e-03) (hash(x)=148529257)
58
+ 460 train 5.463498 (lr=1.9343e-03) (hash(x)=187877362)
59
+ 470 train 5.355601 (lr=1.9762e-03) (hash(x)=174097227)
60
+ 480 train 5.258920 (lr=2.0182e-03) (hash(x)=163811577)
61
+ 490 train 5.339412 (lr=2.0601e-03) (hash(x)=147406221)
62
+ 500 val loss 5.3036
63
+ 500 val perplexity 201.0689
64
+ 500 train 5.226911 (lr=2.1021e-03) (hash(x)=159337928)
65
+ 510 train 5.242280 (lr=2.1441e-03) (hash(x)=155002527)
66
+ 520 train 5.300801 (lr=2.1860e-03) (hash(x)=163685257)
67
+ 530 train 5.285084 (lr=2.2280e-03) (hash(x)=139648586)
68
+ 540 train 5.312295 (lr=2.2699e-03) (hash(x)=158641796)
69
+ 550 train 5.158271 (lr=2.3119e-03) (hash(x)=163537724)
70
+ 560 train 5.213995 (lr=2.3538e-03) (hash(x)=168609033)
71
+ 570 train 5.262797 (lr=2.3958e-03) (hash(x)=155826061)
72
+ 580 train 5.206078 (lr=2.4378e-03) (hash(x)=166604533)
73
+ 590 train 5.200986 (lr=2.4797e-03) (hash(x)=150461749)
74
+ 600 val loss 5.1420
75
+ 600 val perplexity 171.0608
76
+ 600 train 5.189524 (lr=2.5217e-03) (hash(x)=158810876)
77
+ 610 train 5.538415 (lr=2.5636e-03) (hash(x)=161506635)
78
+ 620 train 5.060720 (lr=2.6056e-03) (hash(x)=153757394)
79
+ 630 train 5.131078 (lr=2.6476e-03) (hash(x)=163276041)
80
+ 640 train 5.085196 (lr=2.6895e-03) (hash(x)=157085458)
81
+ 650 train 5.072533 (lr=2.7315e-03) (hash(x)=156529604)
82
+ 660 train 5.125998 (lr=2.7734e-03) (hash(x)=161460943)
83
+ 670 train 5.108505 (lr=2.8154e-03) (hash(x)=163730633)
84
+ 680 train 5.219978 (lr=2.8573e-03) (hash(x)=172158062)
85
+ 690 train 5.030314 (lr=2.8993e-03) (hash(x)=168091722)
86
+ 700 val loss 5.0244
87
+ 700 val perplexity 152.0800
88
+ 700 train 5.063220 (lr=2.9413e-03) (hash(x)=163165614)
89
+ 710 train 5.215199 (lr=2.9832e-03) (hash(x)=170735476)
90
+ 720 train 4.999315 (lr=3.0000e-03) (hash(x)=166422327)
91
+ 730 train 4.966284 (lr=2.9999e-03) (hash(x)=151626971)
92
+ 740 train 4.947554 (lr=2.9997e-03) (hash(x)=158266349)
93
+ 750 train 4.965805 (lr=2.9994e-03) (hash(x)=165279256)
94
+ 760 train 4.870124 (lr=2.9990e-03) (hash(x)=164452783)
95
+ 770 train 4.891485 (lr=2.9985e-03) (hash(x)=165056209)
96
+ 780 train 4.720287 (lr=2.9979e-03) (hash(x)=153309114)
97
+ 790 train 4.698790 (lr=2.9972e-03) (hash(x)=159931795)
98
+ 800 val loss 4.9581
99
+ 800 val perplexity 142.3197
100
+ 800 train 4.754467 (lr=2.9964e-03) (hash(x)=170348996)
101
+ 810 train 4.986647 (lr=2.9955e-03) (hash(x)=168036306)
102
+ 820 train 4.889466 (lr=2.9945e-03) (hash(x)=160083807)
103
+ 830 train 5.002240 (lr=2.9934e-03) (hash(x)=155107640)
104
+ 840 train 4.849988 (lr=2.9922e-03) (hash(x)=164587669)
105
+ 850 train 4.913556 (lr=2.9909e-03) (hash(x)=154395552)
106
+ 860 train 4.862645 (lr=2.9896e-03) (hash(x)=156970280)
107
+ 870 train 4.823155 (lr=2.9881e-03) (hash(x)=154303917)
108
+ 880 train 4.866235 (lr=2.9865e-03) (hash(x)=156001085)
109
+ 890 train 4.746140 (lr=2.9848e-03) (hash(x)=154724999)
110
+ 900 val loss 4.8083
111
+ 900 val perplexity 122.5237
112
+ 900 train 4.753588 (lr=2.9830e-03) (hash(x)=149318422)
113
+ 910 train 4.805540 (lr=2.9811e-03) (hash(x)=174182960)
114
+ 920 train 4.814501 (lr=2.9792e-03) (hash(x)=160259547)
115
+ 930 train 4.689470 (lr=2.9771e-03) (hash(x)=156727936)
116
+ 940 train 4.736747 (lr=2.9749e-03) (hash(x)=147186760)
117
+ 950 train 4.667428 (lr=2.9726e-03) (hash(x)=156729166)
118
+ 960 train 4.699919 (lr=2.9703e-03) (hash(x)=163690830)
119
+ 970 train 4.706794 (lr=2.9678e-03) (hash(x)=164750190)
120
+ 980 train 4.619019 (lr=2.9652e-03) (hash(x)=165946603)
121
+ 990 train 4.703813 (lr=2.9626e-03) (hash(x)=156682992)
122
+ 1000 val loss 4.7689
123
+ 1000 val perplexity 117.7903
124
+ 1000 train 4.993062 (lr=2.9598e-03) (hash(x)=152147013)
125
+ 1010 train 4.894461 (lr=2.9570e-03) (hash(x)=171157521)
126
+ 1020 train 4.822670 (lr=2.9540e-03) (hash(x)=191295677)
127
+ 1030 train 4.649691 (lr=2.9510e-03) (hash(x)=159879474)
128
+ 1040 train 4.772458 (lr=2.9478e-03) (hash(x)=166566331)
129
+ 1050 train 4.699910 (lr=2.9446e-03) (hash(x)=157178542)
130
+ 1060 train 4.686487 (lr=2.9412e-03) (hash(x)=152806413)
131
+ 1070 train 4.752961 (lr=2.9378e-03) (hash(x)=148403650)
132
+ 1080 train 4.652444 (lr=2.9343e-03) (hash(x)=150234604)
133
+ 1090 train 4.700980 (lr=2.9307e-03) (hash(x)=153101997)
134
+ 1100 val loss 4.6900
135
+ 1100 val perplexity 108.8538
136
+ 1100 train 4.719160 (lr=2.9270e-03) (hash(x)=154889828)
137
+ 1110 train 4.685833 (lr=2.9231e-03) (hash(x)=177021958)
138
+ 1120 train 4.685664 (lr=2.9192e-03) (hash(x)=147360351)
139
+ 1130 train 4.591550 (lr=2.9152e-03) (hash(x)=149540375)
140
+ 1140 train 4.641925 (lr=2.9112e-03) (hash(x)=152068875)
141
+ 1150 train 4.512585 (lr=2.9070e-03) (hash(x)=166896658)
142
+ 1160 train 4.458116 (lr=2.9027e-03) (hash(x)=161056451)
143
+ 1170 train 4.323503 (lr=2.8983e-03) (hash(x)=154114105)
144
+ 1180 train 4.504801 (lr=2.8939e-03) (hash(x)=165060550)
145
+ 1190 train 4.651113 (lr=2.8893e-03) (hash(x)=166969373)
146
+ 1200 val loss 4.6447
147
+ 1200 val perplexity 104.0364
148
+ 1200 train 4.636667 (lr=2.8847e-03) (hash(x)=160075492)
149
+ 1210 train 4.752875 (lr=2.8800e-03) (hash(x)=158231512)
150
+ 1220 train 4.616076 (lr=2.8751e-03) (hash(x)=168549643)
151
+ 1230 train 4.573910 (lr=2.8702e-03) (hash(x)=155189035)
152
+ 1240 train 4.621440 (lr=2.8652e-03) (hash(x)=178512760)
153
+ 1250 train 4.513758 (lr=2.8601e-03) (hash(x)=146425022)
154
+ 1260 train 4.621696 (lr=2.8550e-03) (hash(x)=159136514)
155
+ 1270 train 4.617380 (lr=2.8497e-03) (hash(x)=156001968)
156
+ 1280 train 4.540568 (lr=2.8443e-03) (hash(x)=164259999)
157
+ 1290 train 4.516940 (lr=2.8389e-03) (hash(x)=161354928)
158
+ 1300 val loss 4.5855
159
+ 1300 val perplexity 98.0546
160
+ 1300 train 4.580484 (lr=2.8333e-03) (hash(x)=158995886)
161
+ 1310 train 4.569925 (lr=2.8277e-03) (hash(x)=155383293)
162
+ 1320 train 4.567426 (lr=2.8220e-03) (hash(x)=157798110)
163
+ 1330 train 4.416835 (lr=2.8162e-03) (hash(x)=150111713)
164
+ 1340 train 4.383521 (lr=2.8103e-03) (hash(x)=164386787)
165
+ 1350 train 4.531064 (lr=2.8044e-03) (hash(x)=160484704)
166
+ 1360 train 4.400242 (lr=2.7983e-03) (hash(x)=169124186)
167
+ 1370 train 4.332176 (lr=2.7922e-03) (hash(x)=162460921)
168
+ 1380 train 4.593517 (lr=2.7860e-03) (hash(x)=176469537)
169
+ 1390 train 4.573629 (lr=2.7797e-03) (hash(x)=171186702)
170
+ 1400 val loss 4.5365
171
+ 1400 val perplexity 93.3627
172
+ 1400 train 4.633075 (lr=2.7733e-03) (hash(x)=166459649)
173
+ 1410 train 4.658031 (lr=2.7668e-03) (hash(x)=173618054)
174
+ 1420 train 4.577682 (lr=2.7603e-03) (hash(x)=166595245)
175
+ 1430 train 4.591048 (lr=2.7536e-03) (hash(x)=161021911)
176
+ 1440 train 4.500543 (lr=2.7469e-03) (hash(x)=155839867)
177
+ 1450 train 4.603174 (lr=2.7401e-03) (hash(x)=160040811)
178
+ 1460 train 4.496348 (lr=2.7333e-03) (hash(x)=160394504)
179
+ 1470 train 4.467587 (lr=2.7263e-03) (hash(x)=154986331)
180
+ 1480 train 4.529189 (lr=2.7193e-03) (hash(x)=159568067)
181
+ 1490 train 4.538911 (lr=2.7121e-03) (hash(x)=170775607)
182
+ 1500 val loss 4.5124
183
+ 1500 val perplexity 91.1381
184
+ 1500 train 4.440263 (lr=2.7050e-03) (hash(x)=152612395)
185
+ 1510 train 4.434435 (lr=2.6977e-03) (hash(x)=151245168)
186
+ 1520 train 4.381781 (lr=2.6903e-03) (hash(x)=163463453)
187
+ 1530 train 4.535357 (lr=2.6829e-03) (hash(x)=168631026)
188
+ 1540 train 4.547188 (lr=2.6754e-03) (hash(x)=160083779)
189
+ 1550 train 4.549923 (lr=2.6678e-03) (hash(x)=176913716)
190
+ 1560 train 4.405589 (lr=2.6602e-03) (hash(x)=149889155)
191
+ 1570 train 4.480224 (lr=2.6525e-03) (hash(x)=155207555)
192
+ 1580 train 4.374506 (lr=2.6447e-03) (hash(x)=154467728)
193
+ 1590 train 4.495090 (lr=2.6368e-03) (hash(x)=163338122)
194
+ 1600 val loss 4.4793
195
+ 1600 val perplexity 88.1749
196
+ 1600 train 4.402177 (lr=2.6289e-03) (hash(x)=153840208)
197
+ 1610 train 4.320252 (lr=2.6208e-03) (hash(x)=157648881)
198
+ 1620 train 4.406333 (lr=2.6128e-03) (hash(x)=153418765)
199
+ 1630 train 4.330846 (lr=2.6046e-03) (hash(x)=156530827)
200
+ 1640 train 4.679376 (lr=2.5964e-03) (hash(x)=169919274)
201
+ 1650 train 4.479683 (lr=2.5881e-03) (hash(x)=158235114)
202
+ 1660 train 4.706214 (lr=2.5797e-03) (hash(x)=163545476)
203
+ 1670 train 4.554201 (lr=2.5713e-03) (hash(x)=183148505)
204
+ 1680 train 4.406996 (lr=2.5628e-03) (hash(x)=151198383)
205
+ 1690 train 4.437829 (lr=2.5542e-03) (hash(x)=159798979)
206
+ 1700 val loss 4.4496
207
+ 1700 val perplexity 85.5930
208
+ 1700 train 4.298557 (lr=2.5455e-03) (hash(x)=153937979)
209
+ 1710 train 4.355174 (lr=2.5368e-03) (hash(x)=163663673)
210
+ 1720 train 4.473849 (lr=2.5281e-03) (hash(x)=168209417)
211
+ 1730 train 4.533407 (lr=2.5192e-03) (hash(x)=169627748)
212
+ 1740 train 4.262877 (lr=2.5103e-03) (hash(x)=156391292)
213
+ 1750 train 4.283153 (lr=2.5014e-03) (hash(x)=152353279)
214
+ 1760 train 4.390223 (lr=2.4924e-03) (hash(x)=169569102)
215
+ 1770 train 4.244663 (lr=2.4833e-03) (hash(x)=153444218)
216
+ 1780 train 4.327936 (lr=2.4741e-03) (hash(x)=152591697)
217
+ 1790 train 4.463574 (lr=2.4649e-03) (hash(x)=176923774)
218
+ 1800 val loss 4.4298
219
+ 1800 val perplexity 83.9145
220
+ 1800 train 4.486784 (lr=2.4556e-03) (hash(x)=160268754)
221
+ 1810 train 4.418174 (lr=2.4463e-03) (hash(x)=161145587)
222
+ 1820 train 4.429430 (lr=2.4369e-03) (hash(x)=162293952)
223
+ 1830 train 4.486831 (lr=2.4275e-03) (hash(x)=179707686)
224
+ 1840 train 4.316558 (lr=2.4180e-03) (hash(x)=176247518)
225
+ 1850 train 4.366907 (lr=2.4084e-03) (hash(x)=166603610)
226
+ 1860 train 4.393313 (lr=2.3988e-03) (hash(x)=165957414)
227
+ 1870 train 4.384377 (lr=2.3891e-03) (hash(x)=159922566)
228
+ 1880 train 4.278652 (lr=2.3794e-03) (hash(x)=154743265)
229
+ 1890 train 4.383269 (lr=2.3696e-03) (hash(x)=148872048)
230
+ 1900 val loss 4.4283
231
+ 1900 val perplexity 83.7895
232
+ 1900 train 4.412515 (lr=2.3598e-03) (hash(x)=167836353)
233
+ 1910 train 4.338374 (lr=2.3499e-03) (hash(x)=153969641)
234
+ 1920 train 4.482808 (lr=2.3400e-03) (hash(x)=159154377)
235
+ 1930 train 4.414083 (lr=2.3300e-03) (hash(x)=166351013)
236
+ 1940 train 4.548908 (lr=2.3200e-03) (hash(x)=172131477)
237
+ 1950 train 4.379982 (lr=2.3099e-03) (hash(x)=190945203)
238
+ 1960 train 4.460775 (lr=2.2998e-03) (hash(x)=161296279)
239
+ 1970 train 4.433747 (lr=2.2896e-03) (hash(x)=165088578)
240
+ 1980 train 4.426963 (lr=2.2793e-03) (hash(x)=153534765)
241
+ 1990 train 4.367997 (lr=2.2691e-03) (hash(x)=155548622)
242
+ 2000 val loss 4.3863
243
+ 2000 val perplexity 80.3391
244
+ 2000 train 4.304634 (lr=2.2588e-03) (hash(x)=163314857)
245
+ 2010 train 4.318960 (lr=2.2484e-03) (hash(x)=163661705)
246
+ 2020 train 4.223160 (lr=2.2380e-03) (hash(x)=168141011)
247
+ 2030 train 4.307172 (lr=2.2275e-03) (hash(x)=153626022)
248
+ 2040 train 4.245064 (lr=2.2170e-03) (hash(x)=146318026)
249
+ 2050 train 4.310629 (lr=2.2065e-03) (hash(x)=167739815)
250
+ 2060 train 4.288268 (lr=2.1959e-03) (hash(x)=154999799)
251
+ 2070 train 4.343372 (lr=2.1853e-03) (hash(x)=181501630)
252
+ 2080 train 4.347080 (lr=2.1746e-03) (hash(x)=155805721)
253
+ 2090 train 4.392874 (lr=2.1639e-03) (hash(x)=158244510)
254
+ 2100 val loss 4.3703
255
+ 2100 val perplexity 79.0648
256
+ 2100 train 4.445023 (lr=2.1532e-03) (hash(x)=147877788)
257
+ 2110 train 4.361077 (lr=2.1424e-03) (hash(x)=156688909)
258
+ 2120 train 4.371364 (lr=2.1316e-03) (hash(x)=163897683)
259
+ 2130 train 4.332588 (lr=2.1208e-03) (hash(x)=166670021)
260
+ 2140 train 4.294308 (lr=2.1099e-03) (hash(x)=166034067)
261
+ 2150 train 4.300858 (lr=2.0990e-03) (hash(x)=149358136)
262
+ 2160 train 4.208103 (lr=2.0881e-03) (hash(x)=156501573)
263
+ 2170 train 4.230445 (lr=2.0771e-03) (hash(x)=155673523)
264
+ 2180 train 4.254381 (lr=2.0661e-03) (hash(x)=150360430)
265
+ 2190 train 4.316976 (lr=2.0550e-03) (hash(x)=176159016)
266
+ 2200 val loss 4.3504
267
+ 2200 val perplexity 77.5114
268
+ 2200 train 4.353130 (lr=2.0440e-03) (hash(x)=165442718)
269
+ 2210 train 4.422556 (lr=2.0329e-03) (hash(x)=177529267)
270
+ 2220 train 4.289089 (lr=2.0217e-03) (hash(x)=197393885)
271
+ 2230 train 4.341596 (lr=2.0106e-03) (hash(x)=169437808)
272
+ 2240 train 4.426461 (lr=1.9994e-03) (hash(x)=154335840)
273
+ 2250 train 4.353552 (lr=1.9882e-03) (hash(x)=154514509)
274
+ 2260 train 4.202583 (lr=1.9770e-03) (hash(x)=158439895)
275
+ 2270 train 4.267970 (lr=1.9657e-03) (hash(x)=150326591)
276
+ 2280 train 4.312816 (lr=1.9544e-03) (hash(x)=163533419)
277
+ 2290 train 4.107698 (lr=1.9431e-03) (hash(x)=150576123)
278
+ 2300 val loss 4.3260
279
+ 2300 val perplexity 75.6440
280
+ 2300 train 4.086770 (lr=1.9318e-03) (hash(x)=149109715)
281
+ 2310 train 4.195762 (lr=1.9205e-03) (hash(x)=161192834)
282
+ 2320 train 4.156546 (lr=1.9091e-03) (hash(x)=152309013)
283
+ 2330 train 4.330640 (lr=1.8977e-03) (hash(x)=144666293)
284
+ 2340 train 4.512631 (lr=1.8863e-03) (hash(x)=176885772)
285
+ 2350 train 4.516289 (lr=1.8749e-03) (hash(x)=175195635)
286
+ 2360 train 4.515522 (lr=1.8635e-03) (hash(x)=153998404)
287
+ 2370 train 4.527214 (lr=1.8520e-03) (hash(x)=159393178)
288
+ 2380 train 4.287325 (lr=1.8406e-03) (hash(x)=172339628)
289
+ 2390 train 4.343510 (lr=1.8291e-03) (hash(x)=146795138)
290
+ 2400 val loss 4.3115
291
+ 2400 val perplexity 74.5544
292
+ 2400 train 4.226115 (lr=1.8176e-03) (hash(x)=162512827)
293
+ 2410 train 4.242868 (lr=1.8061e-03) (hash(x)=163561481)
294
+ 2420 train 4.264274 (lr=1.7946e-03) (hash(x)=155454305)
295
+ 2430 train 4.224040 (lr=1.7830e-03) (hash(x)=166346010)
296
+ 2440 train 4.206209 (lr=1.7715e-03) (hash(x)=160028108)
297
+ 2450 train 4.179360 (lr=1.7600e-03) (hash(x)=166851234)
298
+ 2460 train 4.228526 (lr=1.7484e-03) (hash(x)=161312139)
299
+ 2470 train 4.284255 (lr=1.7368e-03) (hash(x)=143069188)
300
+ 2480 train 4.390020 (lr=1.7253e-03) (hash(x)=158251395)
301
+ 2490 train 4.457649 (lr=1.7137e-03) (hash(x)=158119693)
302
+ 2500 val loss 4.2914
303
+ 2500 val perplexity 73.0667
304
+ 2500 train 4.260386 (lr=1.7021e-03) (hash(x)=148144913)
305
+ 2510 train 4.434177 (lr=1.6906e-03) (hash(x)=159474538)
306
+ 2520 train 4.259497 (lr=1.6790e-03) (hash(x)=150836192)
307
+ 2530 train 4.253858 (lr=1.6674e-03) (hash(x)=156823868)
308
+ 2540 train 4.387199 (lr=1.6558e-03) (hash(x)=155702319)
309
+ 2550 train 4.198966 (lr=1.6442e-03) (hash(x)=160471487)
310
+ 2560 train 4.119959 (lr=1.6326e-03) (hash(x)=160276214)
311
+ 2570 train 4.235039 (lr=1.6210e-03) (hash(x)=167637000)
312
+ 2580 train 4.171005 (lr=1.6094e-03) (hash(x)=149819695)
313
+ 2590 train 4.231951 (lr=1.5979e-03) (hash(x)=160749905)
314
+ 2600 val loss 4.2829
315
+ 2600 val perplexity 72.4529
316
+ 2600 train 4.149229 (lr=1.5863e-03) (hash(x)=155617603)
317
+ 2610 train 4.477270 (lr=1.5747e-03) (hash(x)=162181794)
318
+ 2620 train 4.356832 (lr=1.5632e-03) (hash(x)=171300086)
319
+ 2630 train 4.317647 (lr=1.5516e-03) (hash(x)=159707436)
320
+ 2640 train 4.422066 (lr=1.5400e-03) (hash(x)=137869312)
321
+ 2650 train 4.313330 (lr=1.5285e-03) (hash(x)=115611274)
322
+ 2660 train 4.196939 (lr=1.5170e-03) (hash(x)=151140502)
323
+ 2670 train 4.293715 (lr=1.5054e-03) (hash(x)=162821182)
324
+ 2680 train 4.385653 (lr=1.4939e-03) (hash(x)=174840522)
325
+ 2690 train 4.272182 (lr=1.4824e-03) (hash(x)=161323382)
326
+ 2700 val loss 4.2518
327
+ 2700 val perplexity 70.2285
328
+ 2700 train 4.137649 (lr=1.4709e-03) (hash(x)=159630183)
329
+ 2710 train 4.301598 (lr=1.4594e-03) (hash(x)=169557486)
330
+ 2720 train 4.191417 (lr=1.4480e-03) (hash(x)=154059878)
331
+ 2730 train 4.206100 (lr=1.4365e-03) (hash(x)=153014579)
332
+ 2740 train 4.160866 (lr=1.4251e-03) (hash(x)=152040333)
333
+ 2750 train 4.347200 (lr=1.4137e-03) (hash(x)=164487951)
334
+ 2760 train 4.315351 (lr=1.4023e-03) (hash(x)=189559478)
335
+ 2770 train 4.312707 (lr=1.3909e-03) (hash(x)=152017273)
336
+ 2780 train 4.314285 (lr=1.3795e-03) (hash(x)=168478683)
337
+ 2790 train 4.292448 (lr=1.3682e-03) (hash(x)=167047337)
338
+ 2800 val loss 4.2320
339
+ 2800 val perplexity 68.8549
340
+ 2800 train 4.293196 (lr=1.3569e-03) (hash(x)=153213767)
341
+ 2810 train 4.154331 (lr=1.3456e-03) (hash(x)=141332160)
342
+ 2820 train 4.194522 (lr=1.3343e-03) (hash(x)=158788151)
343
+ 2830 train 4.205427 (lr=1.3230e-03) (hash(x)=161375220)
344
+ 2840 train 4.075257 (lr=1.3118e-03) (hash(x)=162935851)
345
+ 2850 train 4.081371 (lr=1.3006e-03) (hash(x)=159464909)
346
+ 2860 train 4.097445 (lr=1.2894e-03) (hash(x)=156126362)
347
+ 2870 train 4.067198 (lr=1.2783e-03) (hash(x)=161380583)
348
+ 2880 train 4.154482 (lr=1.2671e-03) (hash(x)=156964606)
349
+ 2890 train 4.245816 (lr=1.2560e-03) (hash(x)=169458983)
350
+ 2900 val loss 4.2270
351
+ 2900 val perplexity 68.5107
352
+ 2900 train 4.339451 (lr=1.2450e-03) (hash(x)=176573157)
353
+ 2910 train 4.230224 (lr=1.2339e-03) (hash(x)=153552099)
354
+ 2920 train 4.194896 (lr=1.2229e-03) (hash(x)=165236334)
355
+ 2930 train 4.268517 (lr=1.2119e-03) (hash(x)=165262998)
356
+ 2940 train 4.124337 (lr=1.2010e-03) (hash(x)=155086138)
357
+ 2950 train 4.179997 (lr=1.1901e-03) (hash(x)=150891340)
358
+ 2960 train 4.183434 (lr=1.1792e-03) (hash(x)=155191944)
359
+ 2970 train 4.182869 (lr=1.1684e-03) (hash(x)=163561216)
360
+ 2980 train 4.142586 (lr=1.1576e-03) (hash(x)=161288249)
361
+ 2990 train 4.059909 (lr=1.1468e-03) (hash(x)=164593955)
362
+ 3000 val loss 4.2038
363
+ 3000 val perplexity 66.9393
364
+ 3000 train 4.135390 (lr=1.1361e-03) (hash(x)=161887650)
365
+ 3010 train 4.173086 (lr=1.1254e-03) (hash(x)=158376770)
366
+ 3020 train 4.106071 (lr=1.1147e-03) (hash(x)=158653093)
367
+ 3030 train 4.158681 (lr=1.1041e-03) (hash(x)=157601570)
368
+ 3040 train 4.247158 (lr=1.0935e-03) (hash(x)=162074203)
369
+ 3050 train 4.237875 (lr=1.0830e-03) (hash(x)=171875870)
370
+ 3060 train 4.235484 (lr=1.0725e-03) (hash(x)=152843048)
371
+ 3070 train 4.267829 (lr=1.0620e-03) (hash(x)=165808171)
372
+ 3080 train 4.245901 (lr=1.0516e-03) (hash(x)=158272669)
373
+ 3090 train 4.268020 (lr=1.0412e-03) (hash(x)=157688554)
374
+ 3100 val loss 4.1963
375
+ 3100 val perplexity 66.4415
376
+ 3100 train 4.279370 (lr=1.0309e-03) (hash(x)=166005015)
377
+ 3110 train 4.172453 (lr=1.0207e-03) (hash(x)=158763736)
378
+ 3120 train 4.224068 (lr=1.0104e-03) (hash(x)=167439328)
379
+ 3130 train 4.328196 (lr=1.0002e-03) (hash(x)=148048338)
380
+ 3140 train 4.265664 (lr=9.9011e-04) (hash(x)=164522898)
381
+ 3150 train 4.259006 (lr=9.8002e-04) (hash(x)=157242118)
382
+ 3160 train 4.195152 (lr=9.6999e-04) (hash(x)=149556058)
383
+ 3170 train 4.223508 (lr=9.6000e-04) (hash(x)=148563572)
384
+ 3180 train 4.242266 (lr=9.5007e-04) (hash(x)=174824363)
385
+ 3190 train 4.418517 (lr=9.4019e-04) (hash(x)=159427153)
386
+ 3200 val loss 4.1691
387
+ 3200 val perplexity 64.6591
388
+ 3200 train 4.279678 (lr=9.3036e-04) (hash(x)=161394675)
389
+ 3210 train 4.276556 (lr=9.2058e-04) (hash(x)=166134016)
390
+ 3220 train 4.322740 (lr=9.1085e-04) (hash(x)=154731855)
391
+ 3230 train 4.144117 (lr=9.0118e-04) (hash(x)=156158458)
392
+ 3240 train 4.089893 (lr=8.9157e-04) (hash(x)=173128783)
393
+ 3250 train 4.142312 (lr=8.8201e-04) (hash(x)=154070645)
394
+ 3260 train 4.101814 (lr=8.7251e-04) (hash(x)=153348463)
395
+ 3270 train 4.148775 (lr=8.6307e-04) (hash(x)=156976700)
396
+ 3280 train 4.150668 (lr=8.5368e-04) (hash(x)=155420483)
397
+ 3290 train 4.065270 (lr=8.4435e-04) (hash(x)=149021726)
398
+ 3300 val loss 4.1592
399
+ 3300 val perplexity 64.0171
400
+ 3300 train 4.109964 (lr=8.3508e-04) (hash(x)=160700347)
401
+ 3310 train 4.188394 (lr=8.2588e-04) (hash(x)=165468658)
402
+ 3320 train 4.212616 (lr=8.1673e-04) (hash(x)=163266950)
403
+ 3330 train 4.269657 (lr=8.0764e-04) (hash(x)=170100446)
404
+ 3340 train 4.245558 (lr=7.9862e-04) (hash(x)=151366602)
405
+ 3350 train 4.292996 (lr=7.8966e-04) (hash(x)=150644566)
406
+ 3360 train 4.119708 (lr=7.8076e-04) (hash(x)=164233827)
407
+ 3370 train 4.212582 (lr=7.7192e-04) (hash(x)=164188603)
408
+ 3380 train 4.143903 (lr=7.6315e-04) (hash(x)=168215736)
409
+ 3390 train 4.062894 (lr=7.5445e-04) (hash(x)=162692228)
410
+ 3400 val loss 4.1467
411
+ 3400 val perplexity 63.2231
412
+ 3400 train 4.199647 (lr=7.4581e-04) (hash(x)=148500491)
413
+ 3410 train 4.037149 (lr=7.3724e-04) (hash(x)=139154138)
414
+ 3420 train 4.021117 (lr=7.2874e-04) (hash(x)=181460455)
415
+ 3430 train 4.120856 (lr=7.2030e-04) (hash(x)=178896304)
416
+ 3440 train 4.133451 (lr=7.1193e-04) (hash(x)=149701574)
417
+ 3450 train 4.120102 (lr=7.0363e-04) (hash(x)=178339553)
418
+ 3460 train 4.154655 (lr=6.9541e-04) (hash(x)=163112813)
419
+ 3470 train 4.191509 (lr=6.8725e-04) (hash(x)=158012268)
420
+ 3480 train 4.201028 (lr=6.7916e-04) (hash(x)=157067275)
421
+ 3490 train 4.192331 (lr=6.7114e-04) (hash(x)=163370975)
422
+ 3500 val loss 4.1307
423
+ 3500 val perplexity 62.2223
424
+ 3500 train 4.141211 (lr=6.6320e-04) (hash(x)=160383657)
425
+ 3510 train 4.080077 (lr=6.5533e-04) (hash(x)=156523307)
426
+ 3520 train 4.099361 (lr=6.4753e-04) (hash(x)=167614080)
427
+ 3530 train 4.203056 (lr=6.3981e-04) (hash(x)=163104143)
428
+ 3540 train 4.131737 (lr=6.3216e-04) (hash(x)=158510849)
429
+ 3550 train 4.138901 (lr=6.2458e-04) (hash(x)=151149010)
430
+ 3560 train 4.206435 (lr=6.1708e-04) (hash(x)=166504920)
431
+ 3570 train 4.158491 (lr=6.0966e-04) (hash(x)=177141678)
432
+ 3580 train 4.021119 (lr=6.0231e-04) (hash(x)=157572718)
433
+ 3590 train 4.402913 (lr=5.9504e-04) (hash(x)=155138495)
434
+ 3600 val loss 4.1182
435
+ 3600 val perplexity 61.4463
436
+ 3600 train 4.218771 (lr=5.8785e-04) (hash(x)=165538372)
437
+ 3610 train 4.180352 (lr=5.8074e-04) (hash(x)=153930045)
438
+ 3620 train 4.266509 (lr=5.7370e-04) (hash(x)=148944287)
439
+ 3630 train 4.184808 (lr=5.6675e-04) (hash(x)=152951123)
440
+ 3640 train 4.153846 (lr=5.5987e-04) (hash(x)=152974720)
441
+ 3650 train 4.074937 (lr=5.5308e-04) (hash(x)=155442773)
442
+ 3660 train 4.068473 (lr=5.4636e-04) (hash(x)=163439074)
443
+ 3670 train 4.180590 (lr=5.3973e-04) (hash(x)=162484510)
444
+ 3680 train 4.136677 (lr=5.3318e-04) (hash(x)=171702529)
445
+ 3690 train 4.019374 (lr=5.2671e-04) (hash(x)=156564909)
446
+ 3700 val loss 4.1118
447
+ 3700 val perplexity 61.0590
448
+ 3700 train 3.970196 (lr=5.2033e-04) (hash(x)=140880241)
449
+ 3710 train 4.054928 (lr=5.1402e-04) (hash(x)=154078620)
450
+ 3720 train 4.050392 (lr=5.0780e-04) (hash(x)=157029930)
451
+ 3730 train 4.185016 (lr=5.0167e-04) (hash(x)=159656742)
452
+ 3740 train 4.163322 (lr=4.9562e-04) (hash(x)=166856858)
453
+ 3750 train 4.081405 (lr=4.8965e-04) (hash(x)=163228730)
454
+ 3760 train 4.113756 (lr=4.8377e-04) (hash(x)=172270929)
455
+ 3770 train 4.067254 (lr=4.7798e-04) (hash(x)=158075162)
456
+ 3780 train 4.133617 (lr=4.7227e-04) (hash(x)=151894909)
457
+ 3790 train 4.128646 (lr=4.6665e-04) (hash(x)=166661644)
458
+ 3800 val loss 4.0924
459
+ 3800 val perplexity 59.8828
460
+ 3800 train 4.051935 (lr=4.6112e-04) (hash(x)=169003916)
461
+ 3810 train 4.176230 (lr=4.5567e-04) (hash(x)=158320575)
462
+ 3820 train 4.066580 (lr=4.5031e-04) (hash(x)=156723804)
463
+ 3830 train 4.012594 (lr=4.4504e-04) (hash(x)=151045721)
464
+ 3840 train 4.116364 (lr=4.3986e-04) (hash(x)=161384988)
465
+ 3850 train 3.825090 (lr=4.3477e-04) (hash(x)=153163470)
466
+ 3860 train 3.838273 (lr=4.2977e-04) (hash(x)=159294553)
467
+ 3870 train 3.920356 (lr=4.2486e-04) (hash(x)=149254649)
468
+ 3880 train 3.913992 (lr=4.2004e-04) (hash(x)=169983073)
469
+ 3890 train 3.777116 (lr=4.1530e-04) (hash(x)=166230303)
470
+ 3900 val loss 4.0991
471
+ 3900 val perplexity 60.2850
472
+ 3900 train 4.131381 (lr=4.1066e-04) (hash(x)=150193462)
473
+ 3910 train 4.229589 (lr=4.0611e-04) (hash(x)=165783034)
474
+ 3920 train 4.107108 (lr=4.0166e-04) (hash(x)=170963531)
475
+ 3930 train 4.199715 (lr=3.9729e-04) (hash(x)=170794255)
476
+ 3940 train 4.143514 (lr=3.9302e-04) (hash(x)=173233324)
477
+ 3950 train 4.167932 (lr=3.8884e-04) (hash(x)=155491235)
478
+ 3960 train 4.266918 (lr=3.8475e-04) (hash(x)=151130973)
479
+ 3970 train 4.113384 (lr=3.8076e-04) (hash(x)=160788059)
480
+ 3980 train 4.145339 (lr=3.7685e-04) (hash(x)=154935745)
481
+ 3990 train 4.018807 (lr=3.7305e-04) (hash(x)=167636525)
482
+ 4000 val loss 4.0781
483
+ 4000 val perplexity 59.0310
484
+ 4000 train 4.062123 (lr=3.6933e-04) (hash(x)=181909918)
485
+ 4010 train 4.023477 (lr=3.6572e-04) (hash(x)=156869134)
486
+ 4020 train 3.968853 (lr=3.6219e-04) (hash(x)=161833430)
487
+ 4030 train 4.185671 (lr=3.5876e-04) (hash(x)=160284977)
488
+ 4040 train 3.870168 (lr=3.5543e-04) (hash(x)=147725752)
489
+ 4050 train 3.836201 (lr=3.5219e-04) (hash(x)=154595084)
490
+ 4060 train 3.911723 (lr=3.4905e-04) (hash(x)=175075232)
491
+ 4070 train 3.928305 (lr=3.4600e-04) (hash(x)=151811717)
492
+ 4080 train 4.048809 (lr=3.4305e-04) (hash(x)=157540169)
493
+ 4090 train 4.177778 (lr=3.4019e-04) (hash(x)=165996843)
494
+ 4100 val loss 4.0758
495
+ 4100 val perplexity 58.8984
496
+ 4100 train 4.098225 (lr=3.3744e-04) (hash(x)=160351198)
497
+ 4110 train 4.143737 (lr=3.3477e-04) (hash(x)=168096977)
498
+ 4120 train 4.052804 (lr=3.3221e-04) (hash(x)=162609202)
499
+ 4130 train 4.083414 (lr=3.2974e-04) (hash(x)=156667217)
500
+ 4140 train 4.077870 (lr=3.2737e-04) (hash(x)=152750290)
501
+ 4150 train 4.071509 (lr=3.2510e-04) (hash(x)=169619396)
502
+ 4160 train 4.105255 (lr=3.2292e-04) (hash(x)=159107550)
503
+ 4170 train 4.072705 (lr=3.2085e-04) (hash(x)=155951176)
504
+ 4180 train 4.028146 (lr=3.1887e-04) (hash(x)=166284453)
505
+ 4190 train 4.177282 (lr=3.1699e-04) (hash(x)=152671857)
506
+ 4200 val loss 4.0690
507
+ 4200 val perplexity 58.4961
508
+ 4200 train 4.101421 (lr=3.1520e-04) (hash(x)=146023049)
509
+ 4210 train 4.034018 (lr=3.1352e-04) (hash(x)=172064951)
510
+ 4220 train 3.596406 (lr=3.1193e-04) (hash(x)=149621476)
511
+ 4230 train 3.774769 (lr=3.1044e-04) (hash(x)=146485058)
512
+ 4240 train 3.850615 (lr=3.0905e-04) (hash(x)=164185738)
513
+ 4250 train 3.770735 (lr=3.0776e-04) (hash(x)=158666667)
514
+ 4260 train 4.118703 (lr=3.0657e-04) (hash(x)=167422908)
515
+ 4270 train 4.087025 (lr=3.0548e-04) (hash(x)=162494922)
516
+ 4280 train 4.107634 (lr=3.0449e-04) (hash(x)=160033749)
517
+ 4290 train 4.079154 (lr=3.0359e-04) (hash(x)=154077913)
518
+ 4300 val loss 4.0637
519
+ 4300 val perplexity 58.1865
520
+ 4300 train 4.135418 (lr=3.0280e-04) (hash(x)=170440016)
521
+ 4310 train 4.105534 (lr=3.0210e-04) (hash(x)=159947398)
522
+ 4320 train 4.130454 (lr=3.0150e-04) (hash(x)=162316772)
523
+ 4330 train 4.227907 (lr=3.0101e-04) (hash(x)=153263344)
524
+ 4340 train 3.983568 (lr=3.0061e-04) (hash(x)=155575816)
525
+ 4350 train 4.012505 (lr=3.0031e-04) (hash(x)=156835929)
526
+ 4360 train 4.067281 (lr=3.0011e-04) (hash(x)=152507945)
527
+ 4370 train 4.020556 (lr=3.0001e-04) (hash(x)=164438795)
528
+ 4374 val loss 4.0636
529
+ 4374 val perplexity 58.1843
scaling_sdpa_granularity_2/hd_2_seed_1340/model_04374.pt ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:b5d003ad93b06e80cec445c154540241eecfaf05787802b099161b653a2afd6c
3
+ size 89692138
scaling_sdpa_granularity_2/hd_2_seed_1340/optimizer_04374.pt ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:60522522369af70465e3b032c42719cceca602ef9b47552427875f74b3f16057
3
+ size 179406342