Markusaksli's workspace
Runs
17
Name
17 visualized
State
Notes
User
Tags
Created
Runtime
Sweep
batch_size
clip_norm
d_embed
d_model
dff
dropout_rate
epochs
learning_rate
lr_steps
mem_len
min_learning_rate
num_heads
num_layers
sequence_length
transposition_range
warmup_steps
accuracy
batch_accuracy
batch_fetch_time
batch_loss
batch_train_time
gen_time
learning_rate
loss
seq_train_time
val_acc
val_accuracy
val_loss
epoch_time
Failed
-
markusaksli
39m 9s
-
16
-
-
128
512
0.1
1000
-
-
-
-
8
8
1024
8
-
0.78004
0.74226
-
0.83085
-
2.27933
-
0.67885
0.02763
-
0.57317
1.66433
140.51869
Killed
-
markusaksli
1h 8m 54s
-
16
-
-
128
512
0.5
1000
-
-
-
-
8
4
1024
8
-
0.68506
0.68617
-
1.00987
-
1.19306
-
1.02006
0.0073347
-
0.59693
1.51797
77.01747
Killed
-
markusaksli
2h 59m 56s
-
16
-
-
256
1024
0.5
1000
-
-
-
-
8
6
1024
6
-
0.94404
0.94226
-
0.18024
-
1.68456
-
0.17425
0.011813
-
0.5657
2.63834
109.993
Killed
Final run for generation
markusaksli
1h 10m 58s
-
12
-
-
256
512
0.4
1000
-
-
-
-
8
6
1024
6
-
0.89061
0.89388
-
0.32585
-
1.68506
-
0.33632
0.012505
-
0.55702
2.01242
-
Killed
Huge data augmentation (seq_len // 2 iterations) with agressive dropout leading to interesting generalization
markusaksli
1h 18m 39s
-
8
-
-
256
1024
0.3
1000
-
-
-
-
8
8
1024
6
-
0.94817
0.95028
-
0.15008
-
2.16655
-
0.15622
0.016375
-
0.56916
2.35881
-
Killed
entire expanded DS as input, still managed to get to over 0.9
markusaksli
32m 59s
-
8
-
-
256
1024
0.1
1000
-
-
-
-
8
8
1024
4
-
0.94151
0.9468
-
0.15655
-
2.2552
-
0.17198
0.017875
-
0.54407
2.10467
-
Killed
-
markusaksli
2h 40m 26s
-
32
-
-
256
1024
0.5
1000
-
-
-
-
8
8
512
12
-
0.98929
0.98941
-
0.034
-
2.163
-
0.034774
0.0048125
-
0.5566
5.00352
-
Killed
Better generalization with bigger dropout
markusaksli
39m 18s
-
32
-
-
256
1024
0.3
1000
-
-
-
-
8
8
512
12
-
0.96943
0.97112
-
0.089994
-
2.16
-
0.096292
0.0047188
-
0.5596
2.5446
-
Killed
much faster than the frankenstein enc-dec model but now manages to overfit all 32 songs
markusaksli
37m 4s
-
32
-
-
256
2048
0.1
1000
-
-
-
-
8
6
512
6
-
0.99094
0.99146
-
0.026557
-
1.691
-
0.027711
0.0040625
-
0.54402
3.6636
-
Killed
-
markusaksli
38m 37s
-
10
0.25
300
300
1500
0.1
-
0.001
200000
500
0.004
10
16
500
4
0
-
-
-
3.22623
-
5.365
0.00099966
3.22623
0.0692
-
-
2.82249
-
Killed
-
markusaksli
30m 33s
-
32
0.25
512
512
1048
0.1
-
0.0001
5000
128
0.000001
8
8
128
4
20
-
-
-
0.8492
-
2.696
1.0000e-10
0.87297
0.0023438
-
-
2.02555
-
Killed
-
markusaksli
19m 14s
-
16
-
-
512
2048
0.1
-
-
-
-
-
16
8
512
6
-
-
-
-
2.07518
-
-
0.000098036
2.13168
0.023125
-
-
2.3179
-
Failed
-
markusaksli
33m 30s
-
4
-
-
256
1024
0.1
1000
-
-
-
-
4
4
2048
4
-
0.99042
0.99042
-
0.032035
-
2.903
-
0.032035
0.044725
-
0.51293
2.51749
-
Killed
-
markusaksli
39m 1s
-
4
-
-
256
1024
0.1
1000
-
-
-
-
4
4
2048
4
-
0.99894
0.99901
-
0.0037318
-
-
-
0.0038545
0.04425
-
0.56503
2.16507
-
Killed
-
markusaksli
1h 52m 55s
-
4
-
-
256
1024
0.2
1000
-
-
-
-
8
4
2048
4
-
0.99375
0.99403
-
0.017749
-
-
-
0.0215
0.06775
-
-
-
-
Killed
even 6 songs transposed +- 6 tones (4 transpositions) fitted to over 0.9 (needs more masking or teacher forcing for generalization?)
markusaksli
23m 19s
-
1
-
-
512
1024
0.1
1000
-
-
-
-
8
4
2048
6
-
0.93409
0.94107
-
0.1898
-
-
-
0.20795
0.14579
-
-
-
-
Killed
-
markusaksli
19m 47s
-
1
-
-
512
1024
0.1
1000
-
-
-
-
8
4
2048
12
-
0.9732
0.9732
-
0.10749
-
-
-
0.10749
0.14544
-
-
-
-
1-17
of 17