Edit model card

w2v2_ablation_with_ling_head-drop0.1-not-load-best-wer-best_on_tp0.025_tl10_fp0.001_fl16

This model is a fine-tuned version of nguyenvulebinh/wav2vec2-base-vietnamese-250h on an unknown dataset. It achieves the following results on the evaluation set:

  • Loss: 0.4141
  • Wer: 0.0914

Model description

More information needed

Intended uses & limitations

More information needed

Training and evaluation data

More information needed

Training procedure

Training hyperparameters

The following hyperparameters were used during training:

  • learning_rate: 2e-05
  • train_batch_size: 8
  • eval_batch_size: 16
  • seed: 42
  • distributed_type: multi-GPU
  • num_devices: 4
  • total_train_batch_size: 32
  • total_eval_batch_size: 64
  • optimizer: Adam with betas=(0.9,0.98) and epsilon=1e-08
  • lr_scheduler_type: cosine
  • lr_scheduler_warmup_ratio: 0.1
  • num_epochs: 100
  • mixed_precision_training: Native AMP

Training results

Training Loss Epoch Step Validation Loss Wer
119.415 0.94 100 91.5112 18.6364
74.7916 1.89 200 12.2928 0.9951
6.9068 2.83 300 5.2345 1.0
5.1207 3.77 400 5.0365 1.0
4.7306 4.72 500 4.9152 1.0
4.4974 5.66 600 4.9315 1.0
4.3923 6.6 700 4.7918 1.0
4.3447 7.55 800 4.6447 1.0
4.225 8.49 900 4.6061 1.0
3.9805 9.43 1000 3.6422 0.8733
2.8303 10.38 1100 1.7824 0.3489
1.5807 11.32 1200 1.0908 0.2162
1.1284 12.26 1300 0.8473 0.1640
0.8703 13.21 1400 0.7322 0.1423
0.7576 14.15 1500 0.6551 0.1325
0.6256 15.09 1600 0.6027 0.1387
0.594 16.04 1700 0.5550 0.1300
0.5492 16.98 1800 0.5200 0.1159
0.476 17.92 1900 0.5012 0.1091
0.4822 18.87 2000 0.5112 0.1074
0.4351 19.81 2100 0.4985 0.1179
0.4169 20.75 2200 0.4712 0.1061
0.3957 21.7 2300 0.4613 0.0988
0.3885 22.64 2400 0.4610 0.1025
0.3827 23.58 2500 0.4509 0.0978
0.3468 24.53 2600 0.4549 0.0951
0.3451 25.47 2700 0.4556 0.1019
0.3234 26.42 2800 0.4554 0.1104
0.31 27.36 2900 0.4568 0.0988
0.3026 28.3 3000 0.4211 0.0965
0.2905 29.25 3100 0.4305 0.0911
0.2964 30.19 3200 0.4379 0.0990
0.302 31.13 3300 0.4379 0.0943
0.2576 32.08 3400 0.4293 0.0933
0.2771 33.02 3500 0.4239 0.0928
0.268 33.96 3600 0.4228 0.0894
0.2458 34.91 3700 0.4288 0.0899
0.2553 35.85 3800 0.4312 0.0966
0.2424 36.79 3900 0.4162 0.0917
0.2501 37.74 4000 0.4088 0.0840
0.2498 38.68 4100 0.4144 0.0921
0.2273 39.62 4200 0.4154 0.0863
0.23 40.57 4300 0.4157 0.0868
0.2409 41.51 4400 0.4033 0.0826
0.248 42.45 4500 0.4122 0.0847
0.218 43.4 4600 0.4052 0.0848
0.1979 44.34 4700 0.4063 0.0887
0.2091 45.28 4800 0.4078 0.0823
0.2097 46.23 4900 0.4177 0.0893
0.2017 47.17 5000 0.4295 0.0887
0.1899 48.11 5100 0.4177 0.0919
0.195 49.06 5200 0.4109 0.0880
0.179 50.0 5300 0.4089 0.0879
0.1773 50.94 5400 0.4071 0.0843
0.1889 51.89 5500 0.4072 0.0885
0.1987 52.83 5600 0.4033 0.0873
0.1979 53.77 5700 0.4033 0.0928
0.1777 54.72 5800 0.4077 0.0898
0.1742 55.66 5900 0.3969 0.0838
0.1678 56.6 6000 0.3997 0.0806
0.1726 57.55 6100 0.3978 0.0885
0.1602 58.49 6200 0.3967 0.0860
0.1681 59.43 6300 0.4039 0.0901
0.1594 60.38 6400 0.3992 0.0856
0.171 61.32 6500 0.4058 0.0890
0.1691 62.26 6600 0.4078 0.0842
0.1724 63.21 6700 0.4161 0.0903
0.172 64.15 6800 0.4121 0.0899
0.1717 65.09 6900 0.4111 0.0878
0.1775 66.04 7000 0.4109 0.0926
0.1607 66.98 7100 0.4080 0.0908
0.1606 67.92 7200 0.4070 0.0930
0.1801 68.87 7300 0.4096 0.0908
0.16 69.81 7400 0.4030 0.0933
0.1433 70.75 7500 0.4059 0.0920
0.1473 71.7 7600 0.4120 0.0979
0.1396 72.64 7700 0.4062 0.0922
0.1429 73.58 7800 0.4079 0.0899
0.1332 74.53 7900 0.4055 0.0851
0.1429 75.47 8000 0.4081 0.0922
0.1528 76.42 8100 0.4083 0.0853
0.1547 77.36 8200 0.4139 0.0945
0.1384 78.3 8300 0.4111 0.0933
0.1696 79.25 8400 0.4132 0.0943
0.1483 80.19 8500 0.4139 0.0906
0.1547 81.13 8600 0.4156 0.0959
0.149 82.08 8700 0.4119 0.0905
0.1294 83.02 8800 0.4145 0.0945
0.1383 83.96 8900 0.4151 0.0917
0.1356 84.91 9000 0.4165 0.0952
0.1491 85.85 9100 0.4188 0.0950
0.1395 86.79 9200 0.4174 0.0950
0.1439 87.74 9300 0.4151 0.0919
0.1421 88.68 9400 0.4152 0.0931
0.1443 89.62 9500 0.4160 0.0944
0.1429 90.57 9600 0.4138 0.0928
0.1397 91.51 9700 0.4149 0.0918
0.155 92.45 9800 0.4144 0.0915
0.1406 93.4 9900 0.4139 0.0921
0.1328 94.34 10000 0.4140 0.0929
0.1461 95.28 10100 0.4142 0.0914
0.1455 96.23 10200 0.4142 0.0913
0.155 97.17 10300 0.4139 0.0914
0.147 98.11 10400 0.4140 0.0918
0.1298 99.06 10500 0.4140 0.0917
0.1508 100.0 10600 0.4141 0.0914

Framework versions

  • Transformers 4.35.2
  • Pytorch 1.13.1+cu117
  • Datasets 2.12.0
  • Tokenizers 0.14.1
Downloads last month
1
Safetensors
Model size
98.8M params
Tensor type
FP16
·
Inference API
Unable to determine this model’s pipeline type. Check the docs .

Model tree for tuanio/w2v2_ablation_with_ling_head-drop0.1-not-load-best-wer-best_on_tp0.025_tl10_fp0.001_fl16

Finetuned
(56)
this model