whisper_large_v3_turbo_finetuned_en_id_v1

This model is a fine-tuned version of openai/whisper-large-v3-turbo on the None dataset. It achieves the following results on the evaluation set:

• eval_loss: 0.1362
• eval_runtime: 263.9538
• eval_samples_per_second: 5.906
• eval_steps_per_second: 5.906
• epoch: 3 (checkpoint model)
• step: 687 (checkpoint model)

Training and Evaluation Data

Step	Epoch	Training Loss	Eval Loss	Grad Norm	Learning Rate	Eval Samples/s	Eval Steps/s	Eval Runtime (s)
229	1.00	0.3841	0.1563	2.216	9.77e-06	5.91	5.91	263.95
458	2.00	0.1163	0.1449	1.855	9.36e-06	5.91	5.91	263.95
687	3.00	0.0693	0.1363	1.538	8.96e-06	5.91	5.91	263.95
917	4.00	0.0408	0.1376	1.367	8.55e-06	5.89	5.89	264.59
1146	5.00	0.0229	0.1398	0.982	8.14e-06	5.90	5.90	264.43
1375	6.00	0.0126	0.1450	0.936	7.73e-06	5.91	5.91	263.95
1604	7.00	0.0082	0.1476	0.362	7.33e-06	5.91	5.91	263.95

Training hyperparameters

The following hyperparameters were used during training:

• learning_rate: 1e-05
• train_batch_size: 8
• eval_batch_size: 1
• seed: 42
• gradient_accumulation_steps: 4
• 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: 100
• num_epochs: 25

Framework versions

• Transformers 4.41.2
• Pytorch 2.2.2+cu121
• Datasets 2.18.0
• Tokenizers 0.19.1