tomtom5/whisper-small-he

Whisper Small He with punctuation- Tom Apt

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

• Loss: 0.6067
• Wer Ortho: 44.9468
• Wer: 41.3437

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: 16
• seed: 42
• 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: constant_with_warmup
• lr_scheduler_warmup_steps: 50
• training_steps: 500
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer Ortho	Wer
1.0739	0.2242	50	0.9126	51.8174	48.0993
0.7849	0.4484	100	0.7600	49.1209	45.4251
0.6366	0.6726	150	0.6490	47.3700	43.7454
0.5582	0.8969	200	0.6033	46.5499	42.8098
0.3775	1.1211	250	0.6004	46.1584	42.2941
0.3617	1.3453	300	0.5918	46.5352	42.8687
0.3631	1.5695	350	0.5849	45.6339	41.7784
0.3332	1.7937	400	0.5885	44.9690	40.9754
0.3093	2.0179	450	0.5808	44.5109	40.8944
0.1865	2.2422	500	0.6067	44.9468	41.3437

Framework versions

• Transformers 4.48.3
• Pytorch 2.5.1+cu124
• Datasets 3.3.2
• Tokenizers 0.21.0