README.md

metadata

license: apache-2.0
tags:
  - generated_from_trainer
metrics:
  - wer
model-index:
  - name: ASR_dear_wav2vec2-thai
    results: []

ASR_dear_wav2vec2-thai

This model is a fine-tuned version of facebook/wav2vec2-large-xlsr-53 on the None dataset. It achieves the following results on the evaluation set:

• Loss: 0.3333
• Wer: 0.3905

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: 32
• eval_batch_size: 16
• 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: 20
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
7.8205	0.75	1000	3.5802	1.0
1.9581	1.5	2000	0.6882	0.7315
0.9012	2.24	3000	0.5229	0.6245
0.7558	2.99	4000	0.4531	0.5812
0.6671	3.74	5000	0.4277	0.5305
0.6083	4.49	6000	0.4067	0.5234
0.5633	5.24	7000	0.3821	0.4831
0.5335	5.98	8000	0.3682	0.4928
0.5021	6.73	9000	0.3578	0.4568
0.4806	7.48	10000	0.3508	0.4609
0.4554	8.23	11000	0.3518	0.4458
0.4361	8.98	12000	0.3375	0.4430
0.411	9.72	13000	0.3363	0.4269
0.3998	10.47	14000	0.3382	0.4221
0.3851	11.22	15000	0.3351	0.4161
0.3713	11.97	16000	0.3353	0.4106
0.3539	12.72	17000	0.3287	0.4084
0.3468	13.46	18000	0.3282	0.4098
0.3369	14.21	19000	0.3278	0.4015
0.3276	14.96	20000	0.3285	0.3968
0.3207	15.71	21000	0.3322	0.3980
0.31	16.45	22000	0.3379	0.3948
0.3043	17.2	23000	0.3264	0.3938
0.2975	17.95	24000	0.3299	0.3933
0.2959	18.7	25000	0.3299	0.3918
0.2898	19.45	26000	0.3333	0.3905

Framework versions

• Transformers 4.27.1
• Pytorch 1.13.1+cu116
• Datasets 2.10.1
• Tokenizers 0.13.2