wav2vec2-xlsr-53-Marathi-large

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: 0.3832
• Wer: 0.2102
• Cer: 0.0667

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
6.4763	1.7647	300	3.3057	1.0	1.0
1.3413	3.5294	600	0.6503	0.4479	0.1574
0.4712	5.2941	900	0.4442	0.3447	0.1126
0.2772	7.0588	1200	0.4034	0.2922	0.0970
0.1741	8.8235	1500	0.3750	0.2518	0.0814
0.1213	10.5882	1800	0.3936	0.2435	0.0787
0.0889	12.3529	2100	0.3841	0.2271	0.0736
0.0718	14.1176	2400	0.3675	0.2116	0.0680
0.0546	15.8824	2700	0.3755	0.2134	0.0676
0.047	17.6471	3000	0.3832	0.2102	0.0667

Framework versions

• Transformers 4.41.1
• Pytorch 2.3.0+cu121
• Datasets 1.18.3
• Tokenizers 0.19.1