miosipof/asr2_aug_v3

Whisper Medium IT

This model is a fine-tuned version of openai/whisper-medium on the b-brave-balanced-augmented dataset. It achieves the following results on the evaluation set:

• Loss: 0.6232
• Wer: 38.6005
• Cer: 25.6421
• Lr: 0.0000

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: 0.0003
• train_batch_size: 16
• eval_batch_size: 16
• seed: 42
• gradient_accumulation_steps: 4
• total_train_batch_size: 64
• 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
• lr_scheduler_warmup_ratio: 0.3
• num_epochs: 8
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer	Cer	Lr
2.1513	1.0	128	1.3441	68.8488	39.3538	0.0001
0.7324	2.0	256	0.8601	52.5959	33.6785	0.0002
0.4633	3.0	384	0.6653	48.7585	33.4714	0.0003
0.301	4.0	512	0.6633	39.5034	27.8790	0.0002
0.1653	5.0	640	0.6030	42.2122	27.8376	0.0002
0.1096	6.0	768	0.6118	38.1490	25.9321	0.0001
0.0641	7.0	896	0.6399	39.5034	25.9321	0.0001
0.042	7.9430	1016	0.6232	38.6005	25.6421	0.0000

Framework versions

• PEFT 0.14.0
• Transformers 4.49.0
• Pytorch 2.2.0
• Datasets 3.3.2
• Tokenizers 0.21.1