fine-tune-Wav2Vec2-XLS-R-300M-Indonesia

This model is a fine-tuned version of facebook/wav2vec2-xls-r-300m on the common_voice_10_0 dataset. It achieves the following results on the evaluation set:

• Loss: 0.7924
• Wer: 0.4307

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: 8
• seed: 42
• gradient_accumulation_steps: 2
• total_train_batch_size: 64
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• lr_scheduler_warmup_steps: 500
• num_epochs: 100
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
2.8421	6.33	500	0.7758	0.7386
0.1875	12.66	1000	0.6039	0.5712
0.1138	18.99	1500	0.6463	0.5170
0.0774	25.32	2000	0.6814	0.4914
0.0655	31.65	2500	0.7010	0.4778
0.0536	37.97	3000	0.7267	0.4761
0.0471	44.3	3500	0.7552	0.4693
0.0402	50.63	4000	0.7474	0.4612
0.0366	56.96	4500	0.7705	0.4560
0.0324	63.29	5000	0.7253	0.4472
0.0293	69.62	5500	0.7660	0.4441
0.0254	75.95	6000	0.7816	0.4398
0.0226	82.28	6500	0.7988	0.4378
0.0202	88.61	7000	0.8009	0.4305
0.0185	94.94	7500	0.7924	0.4307

Framework versions

• Transformers 4.22.2
• Pytorch 1.10.0+cu113
• Datasets 2.5.2
• Tokenizers 0.12.1