wav2vec2-Odia-large-xlsr53

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: 0.2083
• Wer: 0.1897
• Cer: 0.0476

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: 6
• eval_batch_size: 8
• seed: 42
• gradient_accumulation_steps: 4
• total_train_batch_size: 24
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• lr_scheduler_warmup_steps: 500
• training_steps: 3000
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer	Cer
6.9058	2.3622	300	3.1227	1.0	0.8690
1.1432	4.7244	600	0.4002	0.4333	0.1134
0.2628	7.0866	900	0.3145	0.3314	0.0850
0.1368	9.4488	1200	0.2585	0.2716	0.0686
0.0865	11.8110	1500	0.2332	0.2524	0.0619
0.0596	14.1732	1800	0.2253	0.2196	0.0538
0.0445	16.5354	2100	0.2202	0.2100	0.0527
0.0324	18.8976	2400	0.2126	0.2001	0.0511
0.0264	21.2598	2700	0.2089	0.1966	0.0498
0.0211	23.6220	3000	0.2083	0.1897	0.0476

Framework versions

• Transformers 4.41.2
• Pytorch 2.3.0+cu121
• Datasets 1.18.3
• Tokenizers 0.19.1