Skip to content

Commit

Permalink
Experiments and modeling updates
Browse files Browse the repository at this point in the history
  • Loading branch information
@hivaze
hivaze committed Apr 6, 2023
1 parent c061d05 commit 15da477
Show file tree
Hide file tree
Showing 29 changed files with 3,093 additions and 3,606 deletions.
1,289 changes: 1,289 additions & 0 deletions ctc_training.ipynb
View file

Large diffs are not rendered by default.

214 changes: 214 additions & 0 deletions experiments/cnn_v2_128_64seq_alstm_1h_2l_100e/experiment_info.json
View file
Original file line number Diff line number Diff line change
@@ -0,0 +1,214 @@
{
"best_epoch": {
"number": 99,
"train_loss": 0.20460229263275484,
"eval_loss": 0.22667
},
"history": {
"train": [
7.474794321422335,
4.601671568955047,
4.482716554327856,
4.407654206964034,
4.3926886968974825,
4.383409385439716,
4.3643438242658785,
4.351827054084102,
4.327183475977258,
4.288188101370124,
4.253056737441051,
4.237245583836036,
4.21497398086741,
4.192919181872018,
4.151153878320621,
4.060844231255447,
3.969439503512805,
3.40532013434398,
2.736350201353242,
2.22719536853742,
1.8379024131388604,
1.5868311954450003,
1.3268549925164332,
1.1804814663114427,
1.0782012494304511,
0.96314545749109,
0.8742181915271131,
0.8160303212419341,
0.7613078876386715,
0.6958911320831203,
0.6655475225629686,
0.6185498237609863,
0.5898696904695486,
0.5567171335974826,
0.5388461951213547,
0.5185936478119862,
0.494251296112809,
0.48876763408697105,
0.4858396004272413,
0.4552773640880102,
0.4359423936922339,
0.4229008147988138,
0.4196196322954154,
0.41002785480475124,
0.388508718602265,
0.3894110869003248,
0.3760479224633567,
0.3761454537699494,
0.35906109421313565,
0.3515298596665829,
0.3504030772402317,
0.348206295620037,
0.33598872528800483,
0.33573182398759865,
0.3274257752714278,
0.31507090243357644,
0.3082205647531944,
0.3046629768383654,
0.3046883503092995,
0.30675201268890234,
0.2912692327665377,
0.2862265879594827,
0.2999667408345621,
0.2881554548121706,
0.28957333689249015,
0.2706757129370412,
0.28019605331783054,
0.28175595025473005,
0.28264813370342495,
0.26669829423669017,
0.26676003144511695,
0.25708295310599893,
0.2691444132146956,
0.25465964101537875,
0.25381804719755924,
0.2537935504430457,
0.25597068384478366,
0.2472354068031794,
0.2406197392110583,
0.24760646495638014,
0.2428064272750782,
0.23173385933984683,
0.2284139649400228,
0.22771800394299663,
0.24517416803142691,
0.23691140134123306,
0.25361957712264,
0.23134631597543065,
0.2205837278426448,
0.22987172818636592,
0.2239124786250199,
0.21986827141121973,
0.22484416750412953,
0.2204116718678535,
0.21964902006372622,
0.21922560705791547,
0.2114865249848064,
0.2108864010889319,
0.20367810262154928,
0.20460229263275484
],
"eval": [
4.62078,
4.54395,
4.42116,
4.40495,
4.40167,
4.37782,
4.35313,
4.3496,
4.30549,
4.27316,
4.28197,
4.23966,
4.21443,
4.17726,
4.11107,
4.01558,
3.73457,
3.23988,
2.57004,
1.99632,
1.76847,
1.46295,
1.21081,
1.1731,
0.97638,
0.88813,
0.87675,
0.76712,
0.72723,
0.67653,
0.63932,
0.66086,
0.59021,
0.53745,
0.5251,
0.52956,
0.51291,
0.49835,
0.47941,
0.44904,
0.44594,
0.42173,
0.42981,
0.41637,
0.41631,
0.39542,
0.4144,
0.40706,
0.37007,
0.36168,
0.38227,
0.34657,
0.36647,
0.36619,
0.34927,
0.33928,
0.33536,
0.33481,
0.31706,
0.3353,
0.31337,
0.30883,
0.31053,
0.32962,
0.29774,
0.29235,
0.2872,
0.2878,
0.30269,
0.27265,
0.30007,
0.28004,
0.27551,
0.2679,
0.25541,
0.30045,
0.27299,
0.27866,
0.27718,
0.27015,
0.26678,
0.24985,
0.25283,
0.28619,
0.26582,
0.27605,
0.25317,
0.2408,
0.26005,
0.25337,
0.26107,
0.25116,
0.23573,
0.25207,
0.24146,
0.24163,
0.22955,
0.23934,
0.22944,
0.22667
]
},
"architecture": "OCR_CARNN(\n (encoder): CNNImageEncoderV2(\n (layers): Sequential(\n (0): ConvBlock(\n (bn): BatchNorm2d(3, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n (conv): Conv2d(3, 32, kernel_size=(9, 9), stride=(1, 1), padding=(4, 4), bias=False)\n (pooling): MaxPool2d(kernel_size=2, stride=2, padding=0, dilation=1, ceil_mode=False)\n (activation): Hardswish()\n )\n (1): ConvBlock(\n (bn): BatchNorm2d(32, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n (conv): Conv2d(32, 64, kernel_size=(7, 7), stride=(1, 1), padding=(3, 3), bias=False)\n (pooling): MaxPool2d(kernel_size=2, stride=2, padding=0, dilation=1, ceil_mode=False)\n (activation): Hardswish()\n )\n (2): ConvBlock(\n (bn): BatchNorm2d(64, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n (conv): Conv2d(64, 128, kernel_size=(5, 5), stride=(1, 1), padding=(2, 2), bias=False)\n (pooling): MaxPool2d(kernel_size=(2, 1), stride=(2, 1), padding=0, dilation=1, ceil_mode=False)\n (activation): Hardswish()\n )\n (3): ConvBlock(\n (bn): BatchNorm2d(128, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n (conv): Conv2d(128, 128, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)\n (pooling): MaxPool2d(kernel_size=(2, 1), stride=(2, 1), padding=0, dilation=1, ceil_mode=False)\n (activation): Hardswish()\n )\n (4): ConvBlock(\n (bn): BatchNorm2d(128, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n (conv): Conv2d(128, 128, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)\n (pooling): MaxPool2d(kernel_size=(2, 1), stride=(2, 1), padding=0, dilation=1, ceil_mode=False)\n (activation): Hardswish()\n )\n (5): ConvBlock(\n (bn): BatchNorm2d(128, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n (conv): Conv2d(128, 128, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)\n (pooling): MaxPool2d(kernel_size=(2, 1), stride=(2, 1), padding=0, dilation=1, ceil_mode=False)\n (activation): Hardswish()\n )\n )\n (dropout): Dropout(p=0.1, inplace=False)\n (out_net): Sequential(\n (0): LayerNorm((128,), eps=1e-05, elementwise_affine=True)\n (1): Linear(in_features=128, out_features=128, bias=True)\n )\n )\n (decoder): SelfAttenBiLSTMImageDecoder(\n (norm): BatchNorm1d(64, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n (self_atten): MultiheadAttention(\n (out_proj): NonDynamicallyQuantizableLinear(in_features=128, out_features=128, bias=True)\n )\n (rnn): LSTM(128, 128, num_layers=2, dropout=0.1, bidirectional=True)\n (out_proj): Linear(in_features=256, out_features=65, bias=True)\n )\n (softmax): LogSoftmax(dim=-1)\n)"
}
0 ...8_64seq_alstm_2h_2l_100e/tokenizer.pickle → ...8_64seq_alstm_1h_2l_100e/tokenizer.pickle
View file
File renamed without changes.
180 changes: 0 additions & 180 deletions experiments/cnn_v2_128_64seq_alstm_2h_2l_100e/experiment_info.json
View file

This file was deleted.

Loading

0 comments on commit 15da477

Please sign in to comment.