nijes/real_2e-5

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

• Loss: 0.3138
• Cer: 11.4491

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: 4
• eval_batch_size: 4
• seed: 42
• gradient_accumulation_steps: 4
• total_train_batch_size: 16
• 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: linear
• lr_scheduler_warmup_steps: 500
• num_epochs: 10
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Cer
0.7346	1.0	1114	0.5805	16.9876
0.5369	2.0	2228	0.4435	14.6412
0.3868	3.0	3342	0.3662	13.5838
0.3233	4.0	4456	0.3448	13.1037
0.2241	5.0	5570	0.3309	12.7157
0.2018	6.0	6684	0.3172	11.9463
0.1968	7.0	7798	0.3141	11.5662
0.1874	8.0	8912	0.3162	11.7161
0.1568	9.0	10026	0.3138	11.3294
0.1255	10.0	11140	0.3138	11.4491

Framework versions

• Transformers 4.53.0
• Pytorch 2.5.1+cu124
• Datasets 3.5.1
• Tokenizers 0.21.4