whisper-v3-raw-segments

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

• Loss: 1.0487
• Wer: 38.5643

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: 2e-05
• train_batch_size: 8
• eval_batch_size: 8
• seed: 42
• gradient_accumulation_steps: 4
• total_train_batch_size: 32
• optimizer: Use adamw_torch with betas=(0.9,0.999) and epsilon=1e-08 and optimizer_args=No additional optimizer arguments
• lr_scheduler_type: linear
• num_epochs: 10
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
0.6455	1.0	77	0.5327	38.8982
0.2879	2.0	154	0.5643	38.3139
0.1475	3.0	231	0.6085	39.4825
0.0782	4.0	308	0.6817	39.4825
0.0474	5.0	385	0.7397	39.3990
0.0278	6.0	462	0.8067	38.3139
0.018	7.0	539	0.8780	41.0684
0.0089	8.0	616	0.9437	38.9816
0.0039	9.0	693	1.0166	38.0634
0.0014	9.8758	760	1.0487	38.5643

Framework versions

• Transformers 4.47.0
• Pytorch 2.5.1+cu124
• Datasets 3.2.0
• Tokenizers 0.21.0