wav2vec2-large-xls-r-300m-vietnamese-colab

This model is a fine-tuned version of facebook/wav2vec2-xls-r-300m on the common_voice dataset. It achieves the following results on the evaluation set:

• Loss: 1.6183
• Wer: 0.7769

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: 8
• eval_batch_size: 8
• seed: 42
• gradient_accumulation_steps: 4
• 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: 100
• num_epochs: 50
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
17.3239	3.77	50	8.2850	1.0
4.8101	7.55	100	3.6457	1.0
3.555	11.32	150	3.5135	1.0
3.4525	15.09	200	3.5400	1.0
3.4141	18.87	250	3.4720	1.0
3.3671	22.64	300	3.4015	1.0
3.1574	26.42	350	3.0054	1.0007
2.4479	30.19	400	2.3789	0.9876
1.5488	33.96	450	1.9272	0.9026
0.9813	37.74	500	1.7429	0.8419
0.7087	41.51	550	1.6931	0.8115
0.5322	45.28	600	1.6583	0.7894
0.4502	49.06	650	1.6183	0.7769

Framework versions

• Transformers 4.27.3
• Pytorch 1.10.0+cu113
• Datasets 2.10.1
• Tokenizers 0.13.2