Sep26-Mixat-whisper-lg-3-transcript

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: 0.7130
• Wer: 43.1693

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: 16
• eval_batch_size: 8
• seed: 42
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• lr_scheduler_warmup_steps: 500
• num_epochs: 100
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
0.7784	0.4292	100	0.4158	34.8757
0.4942	0.8584	200	0.4306	33.8295
0.4017	1.2876	300	0.4313	38.3124
0.3677	1.7167	400	0.4539	39.1020
0.3498	2.1459	500	0.4611	41.6343
0.2632	2.5751	600	0.4645	37.8113
0.2701	3.0043	700	0.4461	37.3347
0.1499	3.4335	800	0.5147	40.4414
0.1596	3.8627	900	0.5218	41.5292
0.1073	4.2918	1000	0.5668	39.3977
0.0888	4.7210	1100	0.5665	39.4393
0.0738	5.1502	1200	0.6428	39.6104
0.0495	5.5794	1300	0.5914	41.9007
0.0512	6.0086	1400	0.6297	41.4950
0.0315	6.4378	1500	0.6753	44.4477
0.034	6.8670	1600	0.6906	38.4151
0.023	7.2961	1700	0.6998	40.0821
0.0251	7.7253	1800	0.7130	43.1693

Framework versions

• Transformers 4.43.4
• Pytorch 2.4.1
• Datasets 3.0.0
• Tokenizers 0.19.1