README.md

metadata

license: cc-by-nc-4.0
tags:
  - generated_from_trainer
metrics:
  - wer
model-index:
  - name: wav2vec2-base-finetune-vi-v6
    results: []

wav2vec2-base-finetune-vi-v6

This model is a fine-tuned version of nguyenvulebinh/wav2vec2-large-vi on the None dataset. It achieves the following results on the evaluation set:

• Loss: 0.1796
• Wer: 0.1328

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: 0.0001
• train_batch_size: 16
• eval_batch_size: 8
• seed: 42
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• lr_scheduler_warmup_steps: 1000
• num_epochs: 22

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
16.006	1.18	500	3.8945	0.9994
3.4476	2.37	1000	3.3364	0.9994
2.1366	3.55	1500	0.4973	0.3117
0.4721	4.74	2000	0.2702	0.1827
0.288	5.92	2500	0.2183	0.1578
0.2313	7.11	3000	0.2134	0.1498
0.2001	8.29	3500	0.1951	0.1448
0.1673	9.48	4000	0.1923	0.1391
0.1575	10.66	4500	0.1835	0.1419
0.1437	11.85	5000	0.1859	0.1382
0.1293	13.03	5500	0.1936	0.1371
0.121	14.22	6000	0.1915	0.1359
0.1159	15.4	6500	0.1814	0.1344
0.1093	16.59	7000	0.1820	0.1342
0.1015	17.77	7500	0.1789	0.1350
0.097	18.96	8000	0.1881	0.1337
0.093	20.14	8500	0.1841	0.1331
0.0928	21.33	9000	0.1796	0.1328

Framework versions

• Transformers 4.30.2
• Pytorch 2.0.0
• Datasets 2.8.0
• Tokenizers 0.13.3