w2v2_ablation_with_4-layer-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: 0.4122
  • Wer: 0.0792

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
115.505 0.94 100 81.4452 18.6578
51.5639 1.89 200 5.2045 1.0
5.0643 2.83 300 5.2778 1.0
4.769 3.77 400 4.8450 1.0
4.5284 4.72 500 4.6743 1.0
4.424 5.66 600 4.6898 1.0
4.368 6.6 700 4.5734 1.0
4.2534 7.55 800 4.3091 1.0
4.0562 8.49 900 3.9265 0.9752
3.5949 9.43 1000 2.8375 0.6954
2.3633 10.38 1100 1.4117 0.2572
1.331 11.32 1200 0.9633 0.1877
0.9938 12.26 1300 0.7780 0.1505
0.8026 13.21 1400 0.6777 0.1499
0.7257 14.15 1500 0.5969 0.1356
0.6029 15.09 1600 0.5835 0.1203
0.5681 16.04 1700 0.5685 0.1325
0.5349 16.98 1800 0.5383 0.1270
0.4683 17.92 1900 0.5036 0.1126
0.4761 18.87 2000 0.5119 0.1173
0.4224 19.81 2100 0.5188 0.1277
0.4313 20.75 2200 0.5012 0.1180
0.3931 21.7 2300 0.4858 0.1200
0.3891 22.64 2400 0.4647 0.1120
0.3914 23.58 2500 0.4528 0.1144
0.3504 24.53 2600 0.4701 0.1044
0.3616 25.47 2700 0.4736 0.1048
0.3184 26.42 2800 0.4503 0.1084
0.3244 27.36 2900 0.4476 0.1024
0.3153 28.3 3000 0.4354 0.1101
0.3065 29.25 3100 0.4419 0.0965
0.3146 30.19 3200 0.4464 0.0997
0.3153 31.13 3300 0.4289 0.0859
0.269 32.08 3400 0.4388 0.1040
0.2974 33.02 3500 0.4407 0.1024
0.2943 33.96 3600 0.4631 0.0962
0.2728 34.91 3700 0.4320 0.0962
0.2816 35.85 3800 0.4459 0.0984
0.2647 36.79 3900 0.4249 0.0966
0.2788 37.74 4000 0.4118 0.0954
0.2786 38.68 4100 0.4420 0.1003
0.2669 39.62 4200 0.4459 0.1147
0.2569 40.57 4300 0.4257 0.0885
0.2616 41.51 4400 0.4151 0.0928
0.2724 42.45 4500 0.4109 0.0948
0.2332 43.4 4600 0.4189 0.0909
0.2225 44.34 4700 0.4212 0.0952
0.2431 45.28 4800 0.4214 0.0941
0.2369 46.23 4900 0.4097 0.0930
0.2326 47.17 5000 0.4093 0.0944
0.2019 48.11 5100 0.4003 0.0950
0.2074 49.06 5200 0.4333 0.0938
0.1947 50.0 5300 0.4136 0.0952
0.1965 50.94 5400 0.4271 0.0851
0.2035 51.89 5500 0.4170 0.0861
0.2072 52.83 5600 0.4090 0.0831
0.2022 53.77 5700 0.4258 0.0858
0.1807 54.72 5800 0.4159 0.0833
0.1912 55.66 5900 0.4286 0.0846
0.1805 56.6 6000 0.4354 0.0805
0.1906 57.55 6100 0.4126 0.0829
0.1799 58.49 6200 0.4310 0.0823
0.1861 59.43 6300 0.4363 0.0873
0.1823 60.38 6400 0.4247 0.0814
0.1951 61.32 6500 0.4414 0.0866
0.1852 62.26 6600 0.4278 0.0798
0.1877 63.21 6700 0.4242 0.0836
0.1818 64.15 6800 0.4220 0.0826
0.177 65.09 6900 0.4151 0.0861
0.1773 66.04 7000 0.4220 0.0856
0.1682 66.98 7100 0.4215 0.0813
0.1659 67.92 7200 0.4229 0.0807
0.1879 68.87 7300 0.4152 0.0819
0.1704 69.81 7400 0.4113 0.0839
0.155 70.75 7500 0.4139 0.0843
0.1563 71.7 7600 0.4159 0.0834
0.1489 72.64 7700 0.4119 0.0827
0.1501 73.58 7800 0.4113 0.0819
0.1363 74.53 7900 0.4110 0.0820
0.154 75.47 8000 0.4130 0.0788
0.1595 76.42 8100 0.4154 0.0789
0.1574 77.36 8200 0.4149 0.0808
0.1407 78.3 8300 0.4152 0.0827
0.1759 79.25 8400 0.4149 0.0829
0.1512 80.19 8500 0.4198 0.0831
0.1544 81.13 8600 0.4087 0.0786
0.1523 82.08 8700 0.4139 0.0797
0.1346 83.02 8800 0.4116 0.0782
0.1429 83.96 8900 0.4113 0.0782
0.135 84.91 9000 0.4119 0.0802
0.1461 85.85 9100 0.4162 0.0800
0.1389 86.79 9200 0.4166 0.0802
0.1458 87.74 9300 0.4182 0.0804
0.144 88.68 9400 0.4169 0.0792
0.1415 89.62 9500 0.4156 0.0792
0.1472 90.57 9600 0.4147 0.0792
0.1428 91.51 9700 0.4152 0.0796
0.1529 92.45 9800 0.4129 0.0790
0.1447 93.4 9900 0.4128 0.0795
0.1353 94.34 10000 0.4129 0.0797
0.1488 95.28 10100 0.4130 0.0796
0.1454 96.23 10200 0.4124 0.0793
0.16 97.17 10300 0.4122 0.0794
0.1533 98.11 10400 0.4123 0.0792
0.1313 99.06 10500 0.4120 0.0794
0.155 100.0 10600 0.4122 0.0792

Framework versions

  • Transformers 4.35.2
  • Pytorch 1.13.1+cu117
  • Datasets 2.12.0
  • Tokenizers 0.14.1
Downloads last month
29
Safetensors
Model size
118M params
Tensor type
FP16
·
Inference Providers NEW
This model is not currently available via any of the supported third-party Inference Providers, and HF Inference API was unable to determine this model’s pipeline type.

Model tree for tuanio/w2v2_ablation_with_4-layer-ling_head-best_on_tp0.025_tl10_fp0.001_fl16

Finetuned
(56)
this model