xlsr_hindi_LMless_300m_finetuned

This model is a fine-tuned version of facebook/wav2vec2-xls-r-300m on the common_voice_16_1 dataset. It achieves the following results on the evaluation set:

• Loss: 1.0218
• Wer : 0.5493

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: 32
• eval_batch_size: 8
• seed: 42
• gradient_accumulation_steps: 3
• total_train_batch_size: 96
• 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
No log	2.8846	50	12.4919	1.0
15.3302	5.7692	100	7.5919	1.0
15.3302	8.6538	150	4.6126	1.0
5.1316	11.5385	200	3.6522	1.0
5.1316	14.4231	250	3.5474	1.0
3.3791	17.3077	300	3.5372	1.0
3.3791	20.1923	350	3.1750	0.9995
2.6935	23.0769	400	1.6399	0.8194
2.6935	25.9615	450	1.1040	0.6393
0.7418	28.8462	500	1.0218	0.5493

Framework versions

• Transformers 4.40.2
• Pytorch 2.2.1+cu121
• Datasets 2.19.1
• Tokenizers 0.19.1