wav2vec2-mms-1b-malayalam-colab-CV17.0-v2

This model is a fine-tuned version of facebook/mms-1b-all on the common_voice_17_0 dataset. It achieves the following results on the evaluation set:

• Loss: 0.2965
• Wer: 0.5284
• Cer: 0.0934
• Bleu: 0.1997

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: 0.001
• 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_ratio: 0.15
• training_steps: 2000
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer	Cer	Bleu
5.5563	3.1496	200	0.3157	0.5580	0.1055	0.1800
0.3888	6.2992	400	0.2983	0.5471	0.1003	0.1906
0.3328	9.4488	600	0.3008	0.5542	0.1002	0.1634
0.3006	12.5984	800	0.2821	0.5368	0.0984	0.1888
0.2743	15.7480	1000	0.2913	0.5329	0.0968	0.1813
0.2461	18.8976	1200	0.2822	0.5319	0.0957	0.1937
0.2346	22.0472	1400	0.2933	0.5335	0.0942	0.1848
0.2112	25.1969	1600	0.2885	0.5300	0.0947	0.1900
0.2006	28.3465	1800	0.2944	0.5329	0.0939	0.1870
0.1879	31.4961	2000	0.2965	0.5284	0.0934	0.1997

Framework versions

• Transformers 4.42.4
• Pytorch 2.4.0+cu121
• Datasets 2.21.0
• Tokenizers 0.19.1