jun-han/Whisper-squeezeformer-v6

Whisper-squeezeformer-N6SQU-full-per-norn

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

• Loss: 0.1775
• Wer: 8.1539

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: 20
• eval_batch_size: 8
• seed: 42
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• lr_scheduler_warmup_steps: 2500
• training_steps: 30000
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
4.7944	1.0	2500	3.7862	136.5585
3.8012	2.0	5000	2.9923	107.0298
0.7205	3.0	7500	0.2924	17.2474
0.1473	4.0	10000	0.2399	13.8276
0.0847	5.0	12500	0.2228	11.7677
0.0505	6.0	15000	0.2200	12.2071
0.1802	7.0	17500	0.1782	9.7554
0.107	8.0	20000	0.1681	9.1962
0.0693	9.0	22500	0.1669	8.2205
0.0462	10.0	25000	0.1700	8.9604
0.032	11.0	27500	0.1745	8.5952
0.024	12.0	30000	0.1775	8.1539

Framework versions

• Transformers 4.45.2
• Pytorch 2.4.0
• Datasets 3.1.0
• Tokenizers 0.20.0