smutuvi/ndizi-whisper-small-optimized_1

ndizi-whisper-small-optimized

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: 0.7279
• Wer: 0.3118
• Cer: 0.1242

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: 2
• total_train_batch_size: 8
• optimizer: Use OptimizerNames.ADAMW_TORCH_FUSED 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: 200
• num_epochs: 15
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer	Cer
0.8495	1.0	268	0.6232	0.3815	0.1517
0.4653	2.0	536	0.5281	0.3612	0.1496
0.3145	3.0	804	0.5185	0.3169	0.1267
0.1865	4.0	1072	0.5390	0.3227	0.1308
0.1128	5.0	1340	0.5657	0.3236	0.1284
0.0654	6.0	1608	0.6066	0.3150	0.1245
0.0276	7.0	1876	0.6278	0.3210	0.1302
0.0194	8.0	2144	0.6498	0.3197	0.1253
0.0078	9.0	2412	0.6749	0.3240	0.1292
0.0052	10.0	2680	0.6909	0.3154	0.1271
0.0024	11.0	2948	0.7041	0.3107	0.1232
0.0028	12.0	3216	0.7126	0.3165	0.1258
0.0016	13.0	3484	0.7215	0.3142	0.1255
0.0012	14.0	3752	0.7256	0.3120	0.1241
0.0011	15.0	4020	0.7279	0.3118	0.1242

Framework versions

• Transformers 4.57.1
• Pytorch 2.8.0+cu128
• Datasets 3.6.0
• Tokenizers 0.22.1