wav2vec2-base-thai-5-google-colab

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: 0.5367
• Wer: 0.5438

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: 0.0003
• train_batch_size: 16
• eval_batch_size: 8
• seed: 42
• gradient_accumulation_steps: 2
• total_train_batch_size: 32
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• lr_scheduler_warmup_steps: 500
• num_epochs: 20
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
5.8705	0.88	400	3.6557	1.0
3.2446	1.77	800	1.8761	1.0173
1.6206	2.65	1200	0.9131	0.8479
1.2166	3.54	1600	0.7688	0.7733
1.0517	4.42	2000	0.6680	0.7191
0.9463	5.31	2400	0.6290	0.6903
0.8679	6.19	2800	0.5944	0.6736
0.8053	7.08	3200	0.5609	0.6405
0.7408	7.96	3600	0.5476	0.6294
0.6992	8.85	4000	0.5386	0.6046
0.6593	9.73	4400	0.5338	0.5962
0.6276	10.62	4800	0.5410	0.6087
0.5988	11.5	5200	0.5257	0.5850
0.5625	12.39	5600	0.4970	0.5780
0.5382	13.27	6000	0.5132	0.5638
0.5208	14.16	6400	0.5323	0.5696
0.5041	15.04	6800	0.5287	0.5644
0.4742	15.93	7200	0.5302	0.5744
0.4679	16.81	7600	0.5347	0.5501
0.453	17.7	8000	0.5413	0.5498
0.4424	18.58	8400	0.5418	0.5429
0.4283	19.47	8800	0.5367	0.5438

Framework versions

• Transformers 4.35.2
• Pytorch 2.1.0+cu121
• Datasets 2.16.1
• Tokenizers 0.15.1