wav2vec2-large-xls-r-300m-j-kana-colab

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: 0.7188
• Wer: 0.1285

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: 2
• eval_batch_size: 8
• seed: 42
• gradient_accumulation_steps: 8
• total_train_batch_size: 16
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• lr_scheduler_warmup_steps: 500
• num_epochs: 15
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
No log	1.0	397	3.8381	0.9571
No log	2.0	794	0.8909	0.2265
4.0962	3.0	1191	0.8076	0.2054
4.0962	4.0	1588	0.7300	0.1809
4.0962	5.0	1985	0.7322	0.1761
0.6325	6.0	2382	0.6478	0.1524
0.6325	7.0	2779	0.6559	0.1472
0.408	8.0	3176	0.6925	0.1500
0.408	9.0	3573	0.7567	0.1582
0.408	10.0	3970	0.6687	0.1358
0.29	11.0	4367	0.7223	0.1418
0.29	12.0	4764	0.7082	0.1328
0.2152	13.0	5161	0.7114	0.1340
0.2152	14.0	5558	0.7082	0.1280
0.2152	15.0	5955	0.7188	0.1285

Framework versions

• Transformers 4.21.2
• Pytorch 1.10.0+cu113
• Datasets 2.4.0
• Tokenizers 0.12.1