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

This model is a fine-tuned version of facebook/wav2vec2-xls-r-300m on the common_voice_16_0 dataset. It achieves the following results on the evaluation set:

• Loss: 0.6983
• Wer: 0.4948

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: 30

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
5.491	1.8059	400	1.3703	0.9679
0.6981	3.6117	800	0.7041	0.6607
0.3758	5.4176	1200	0.6709	0.6185
0.2736	7.2235	1600	0.7170	0.5925
0.2089	9.0293	2000	0.6445	0.5722
0.1686	10.8352	2400	0.7004	0.5770
0.1408	12.6411	2800	0.7097	0.5735
0.1227	14.4470	3200	0.6763	0.5533
0.1056	16.2528	3600	0.7245	0.5484
0.0923	18.0587	4000	0.7198	0.5480
0.083	19.8646	4400	0.6568	0.5251
0.0742	21.6704	4800	0.7183	0.5252
0.0647	23.4763	5200	0.7306	0.5141
0.0574	25.2822	5600	0.7236	0.5063
0.052	27.0880	6000	0.7234	0.4969
0.0478	28.8939	6400	0.6983	0.4948

Framework versions

• Transformers 4.40.1
• Pytorch 2.2.1+cu121
• Datasets 2.19.0
• Tokenizers 0.19.1