wav2vec2-commonvoice-tamil

This model is a fine-tuned version of Harveenchadha/vakyansh-wav2vec2-tamil-tam-250 on the common_voice dataset. It achieves the following results on the evaluation set:

• Loss: 3.3415
• Wer: 1.0

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: 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_steps: 400
• num_epochs: 30
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
5.384	1.69	200	3.3400	1.0
3.3085	3.39	400	3.3609	1.0
3.3008	5.08	600	3.3331	1.0
3.2852	6.78	800	3.3492	1.0
3.2908	8.47	1000	3.3318	1.0
3.2865	10.17	1200	3.3501	1.0
3.2826	11.86	1400	3.3403	1.0
3.2875	13.56	1600	3.3335	1.0
3.2899	15.25	1800	3.3311	1.0
3.2755	16.95	2000	3.3617	1.0
3.2877	18.64	2200	3.3317	1.0
3.2854	20.34	2400	3.3560	1.0
3.2878	22.03	2600	3.3332	1.0
3.2766	23.73	2800	3.3317	1.0
3.2943	25.42	3000	3.3737	1.0
3.2845	27.12	3200	3.3347	1.0
3.2765	28.81	3400	3.3415	1.0

Framework versions

• Transformers 4.11.3
• Pytorch 1.10.0+cu111
• Datasets 1.18.3
• Tokenizers 0.10.3