wav2vec2-base-Malyalam-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: 0.4124
• Wer: 0.4294
• Cer: 0.0876

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
5.755	2.1622	300	3.4394	1.0	1.0
1.5526	4.3243	600	0.7064	0.7488	0.1865
0.6682	6.4865	900	0.5412	0.6252	0.1383
0.4698	8.6486	1200	0.4891	0.5767	0.1271
0.3632	10.8108	1500	0.3958	0.5024	0.1004
0.2918	12.9730	1800	0.4565	0.5020	0.1055
0.2335	15.1351	2100	0.4298	0.4723	0.0981
0.1956	17.2973	2400	0.3805	0.4466	0.0888
0.1665	19.4595	2700	0.4039	0.4311	0.0889
0.1447	21.6216	3000	0.4124	0.4294	0.0876

Framework versions

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