nijes/mixed_1e-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: 0.2613
• Cer: 10.9862

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: 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
0.698	1.0	1114	0.4865	14.7030
0.5294	2.0	2228	0.3672	13.2286
0.3749	3.0	3342	0.3083	12.6513
0.3479	4.0	4456	0.2905	12.6933
0.2457	5.0	5570	0.2796	11.9239
0.2168	6.0	6684	0.2719	11.9318
0.2455	7.0	7798	0.2724	11.4557
0.2279	8.0	8912	0.2631	11.1006
0.2242	9.0	10026	0.2637	10.9585
0.1664	10.0	11140	0.2613	10.9862

Framework versions

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