Edit model card

w2v2_ablation_with_ling_head-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: 0.4160
  • Wer: 0.0829

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
119.4146 0.94 100 91.5098 18.6362
74.7874 1.89 200 12.2829 0.9952
6.9045 2.83 300 5.2350 1.0
5.1208 3.77 400 5.0365 1.0
4.7305 4.72 500 4.9151 1.0
4.4972 5.66 600 4.9305 1.0
4.3922 6.6 700 4.7878 1.0
4.3448 7.55 800 4.6496 1.0
4.2246 8.49 900 4.6107 1.0
3.9903 9.43 1000 3.7007 0.8971
2.8741 10.38 1100 1.8526 0.3769
1.5912 11.32 1200 1.0932 0.2104
1.1157 12.26 1300 0.8455 0.1694
0.8761 13.21 1400 0.7110 0.1416
0.7465 14.15 1500 0.6333 0.1364
0.6291 15.09 1600 0.5729 0.1346
0.5745 16.04 1700 0.5506 0.1200
0.5359 16.98 1800 0.5359 0.1182
0.4804 17.92 1900 0.4879 0.1104
0.4833 18.87 2000 0.5131 0.1068
0.4384 19.81 2100 0.4859 0.1108
0.4273 20.75 2200 0.4690 0.1078
0.4111 21.7 2300 0.4541 0.1019
0.3879 22.64 2400 0.4585 0.1074
0.3926 23.58 2500 0.4487 0.1033
0.3499 24.53 2600 0.4447 0.0981
0.3537 25.47 2700 0.4481 0.0994
0.3207 26.42 2800 0.4383 0.1086
0.3149 27.36 2900 0.4449 0.1065
0.3042 28.3 3000 0.4315 0.0999
0.3181 29.25 3100 0.4423 0.1073
0.3099 30.19 3200 0.4329 0.0984
0.3079 31.13 3300 0.4308 0.1022
0.2524 32.08 3400 0.4050 0.0919
0.274 33.02 3500 0.4013 0.0981
0.2687 33.96 3600 0.4088 0.0948
0.2496 34.91 3700 0.4083 0.0883
0.2555 35.85 3800 0.4105 0.1002
0.2466 36.79 3900 0.3890 0.0915
0.2519 37.74 4000 0.3962 0.0959
0.2541 38.68 4100 0.4004 0.0971
0.2311 39.62 4200 0.4018 0.0978
0.2306 40.57 4300 0.4000 0.0931
0.2414 41.51 4400 0.4023 0.0972
0.2484 42.45 4500 0.4031 0.0928
0.2167 43.4 4600 0.3899 0.0858
0.1952 44.34 4700 0.3920 0.0845
0.2115 45.28 4800 0.3941 0.0832
0.209 46.23 4900 0.4096 0.0890
0.1999 47.17 5000 0.4065 0.0853
0.1844 48.11 5100 0.4080 0.0886
0.1938 49.06 5200 0.4058 0.0875
0.1775 50.0 5300 0.4028 0.0861
0.1757 50.94 5400 0.4024 0.0832
0.1861 51.89 5500 0.4095 0.0844
0.1967 52.83 5600 0.4128 0.0843
0.1887 53.77 5700 0.4062 0.0864
0.1684 54.72 5800 0.4170 0.0922
0.1697 55.66 5900 0.4141 0.0904
0.1711 56.6 6000 0.4080 0.0798
0.1731 57.55 6100 0.4066 0.0817
0.1614 58.49 6200 0.4160 0.0883
0.1668 59.43 6300 0.4112 0.0881
0.1535 60.38 6400 0.4117 0.0834
0.1723 61.32 6500 0.4185 0.0836
0.1667 62.26 6600 0.4210 0.0817
0.1667 63.21 6700 0.4202 0.0903
0.1618 64.15 6800 0.4165 0.0868
0.1583 65.09 6900 0.4082 0.0851
0.163 66.04 7000 0.4131 0.0869
0.1514 66.98 7100 0.4114 0.0847
0.1504 67.92 7200 0.4122 0.0842
0.1713 68.87 7300 0.4175 0.0887
0.1562 69.81 7400 0.4174 0.0872
0.1421 70.75 7500 0.4186 0.0848
0.141 71.7 7600 0.4238 0.0896
0.1345 72.64 7700 0.4201 0.0868
0.1381 73.58 7800 0.4179 0.0849
0.1293 74.53 7900 0.4140 0.0832
0.1391 75.47 8000 0.4162 0.0826
0.1467 76.42 8100 0.4159 0.0838
0.1454 77.36 8200 0.4169 0.0861
0.1348 78.3 8300 0.4193 0.0884
0.1634 79.25 8400 0.4196 0.0866
0.1435 80.19 8500 0.4206 0.0874
0.1445 81.13 8600 0.4166 0.0843
0.1429 82.08 8700 0.4148 0.0815
0.1238 83.02 8800 0.4177 0.0826
0.129 83.96 8900 0.4146 0.0829
0.1335 84.91 9000 0.4177 0.0836
0.1409 85.85 9100 0.4187 0.0843
0.1333 86.79 9200 0.4186 0.0852
0.1372 87.74 9300 0.4171 0.0828
0.1364 88.68 9400 0.4180 0.0860
0.137 89.62 9500 0.4179 0.0838
0.1346 90.57 9600 0.4181 0.0828
0.137 91.51 9700 0.4173 0.0825
0.1501 92.45 9800 0.4184 0.0841
0.136 93.4 9900 0.4162 0.0822
0.1278 94.34 10000 0.4161 0.0828
0.1381 95.28 10100 0.4167 0.0829
0.138 96.23 10200 0.4166 0.0829
0.1484 97.17 10300 0.4161 0.0826
0.14 98.11 10400 0.4160 0.0831
0.1254 99.06 10500 0.4160 0.0831
0.145 100.0 10600 0.4160 0.0829

Framework versions

  • Transformers 4.35.2
  • Pytorch 1.13.1+cu117
  • Datasets 2.12.0
  • Tokenizers 0.14.1
Downloads last month
2
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-best_on_tp0.025_tl10_fp0.001_fl16

Finetuned
(56)
this model