v3-turbo-cv17-telephonic-lora

This model is a fine-tuned version of openai/whisper-large-v3-turbo on the CommonVoice-17_tr_bandpass_filter_telephonic dataset. It achieves the following results on the evaluation set:

• Loss: 0.1411
• Wer: 14.2090

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: 1e-05
• train_batch_size: 4
• eval_batch_size: 8
• seed: 42
• gradient_accumulation_steps: 2
• total_train_batch_size: 8
• optimizer: Use OptimizerNames.ADAMW_TORCH with betas=(0.9,0.999) and epsilon=1e-08 and optimizer_args=No additional optimizer arguments
• lr_scheduler_type: cosine
• training_steps: 5000
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
0.454	0.1138	500	0.1633	15.3845
0.1463	0.2276	1000	0.1525	14.9965
0.1393	0.3414	1500	0.1482	14.7002
0.1344	0.4552	2000	0.1466	14.4383
0.1305	0.5690	2500	0.1442	14.3084
0.1235	0.6828	3000	0.1427	14.2510
0.129	0.7966	3500	0.1418	14.2434
0.1259	0.9104	4000	0.1416	14.1765
0.1169	1.0241	4500	0.1412	14.2185
0.1103	1.1379	5000	0.1411	14.2090

Framework versions

• PEFT 0.14.0
• Transformers 4.48.3
• Pytorch 2.4.1+cu121
• Datasets 3.1.0
• Tokenizers 0.21.0