wav2vec2-large-xlsr-hindi

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: 1.0220
• Wer: 0.5697

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
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
5.6122	1.81	400	3.3749	1.0
1.6592	3.61	800	1.0003	0.7554
0.7745	5.42	1200	0.9482	0.6972
0.6286	7.22	1600	1.0754	0.6750
0.5413	9.03	2000	0.9040	0.6405
0.4833	10.84	2400	0.9086	0.6116
0.4331	12.64	2800	0.9273	0.6283
0.4047	14.45	3200	1.0076	0.6138
0.3739	16.25	3600	0.9818	0.6018
0.3445	18.06	4000	0.9948	0.5952
0.3305	19.86	4400	0.9897	0.5834
0.3107	21.67	4800	1.0022	0.5751
0.2879	23.48	5200	1.0235	0.5744
0.2836	25.28	5600	1.0238	0.5765
0.2706	27.09	6000	1.0276	0.5694
0.2656	28.89	6400	1.0220	0.5697

Framework versions

• Transformers 4.36.2
• Pytorch 2.1.2+cu121
• Datasets 2.16.1
• Tokenizers 0.15.0