wav2vec2-base-Odia-large

This model is a fine-tuned version of facebook/wav2vec2-base-960h on the None dataset. It achieves the following results on the evaluation set:

• Loss: 0.3174
• Wer: 0.2448
• Cer: 0.0666

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
6.2874	2.3622	300	3.3817	1.0	1.0
2.5772	4.7244	600	1.1573	0.7875	0.2934
0.8599	7.0866	900	0.6319	0.5302	0.1579
0.532	9.4488	1200	0.5208	0.4332	0.1278
0.374	11.8110	1500	0.4485	0.3917	0.1110
0.272	14.1732	1800	0.3939	0.3383	0.0928
0.2015	16.5354	2100	0.3646	0.3040	0.0824
0.152	18.8976	2400	0.3415	0.2700	0.0741
0.1146	21.2598	2700	0.3278	0.2584	0.0691
0.0939	23.6220	3000	0.3174	0.2448	0.0666

Framework versions

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