whisper-grain-lg_cv_only

This model is a fine-tuned version of openai/whisper-small on the common_voice_17_0 dataset. It achieves the following results on the evaluation set:

• Loss: 0.7301
• Wer: 0.3085
• Cer: 0.1022

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: 5e-05
• train_batch_size: 8
• eval_batch_size: 4
• seed: 42
• gradient_accumulation_steps: 2
• total_train_batch_size: 16
• optimizer: Use adamw_hf with betas=(0.9,0.999) and epsilon=1e-08 and optimizer_args=No additional optimizer arguments
• lr_scheduler_type: linear
• lr_scheduler_warmup_ratio: 0.1
• num_epochs: 100
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer	Cer
2.047	1.0	4442	0.5846	0.7958	0.2884
0.6167	2.0	8884	0.4595	0.3765	0.1123
0.4114	3.0	13326	0.4360	0.3274	0.0961
0.3021	4.0	17768	0.4581	0.3233	0.1041
0.2397	5.0	22210	0.4812	0.4780	0.2962
0.2033	6.0	26652	0.5091	0.3106	0.0936
0.1828	7.0	31094	0.5121	0.3105	0.1066
0.1722	8.0	35536	0.5459	0.3140	0.0999
0.1662	9.0	39978	0.5600	0.3358	0.1088
0.1628	10.0	44420	0.5772	0.3287	0.1064
0.1501	11.0	48862	0.5920	0.3217	0.1069
0.1247	12.0	53304	0.5935	0.3236	0.1368
0.1052	13.0	57746	0.6017	0.2989	0.0963
0.0906	14.0	62188	0.6221	0.3205	0.1056
0.0804	15.0	66630	0.6380	0.3011	0.0977
0.0715	16.0	71072	0.6551	0.3064	0.1012
0.0648	17.0	75514	0.6800	0.3128	0.1049
0.0589	18.0	79956	0.6752	0.2997	0.0979
0.0533	19.0	84398	0.6970	0.3079	0.0985
0.0498	20.0	88840	0.7228	0.3043	0.0981
0.0451	21.0	93282	0.7327	0.3137	0.1042
0.0424	22.0	97724	0.7152	0.3002	0.1002
0.04	23.0	102166	0.7301	0.3085	0.1022

Framework versions

• Transformers 4.46.1
• Pytorch 2.1.0+cu118
• Datasets 3.1.0
• Tokenizers 0.20.1