jun-han/Whisper-squeezeformer-v7

Whisper-squeezeformer-N6SQU-

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: 0.1579
• Wer: 5.4340

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: 45000
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
4.7861	1.0	2500	3.8746	131.6000
2.7936	2.0	5000	0.2788	14.1395
0.1896	3.0	7500	0.2055	10.4534
0.1024	4.0	10000	0.1973	8.6903
0.0602	5.0	12500	0.1949	8.9470
0.1756	6.0	15000	0.1584	7.5034
0.1005	7.0	17500	0.1525	6.7046
0.0619	8.0	20000	0.1549	6.7712
0.2214	9.0	22500	0.1455	6.3185
0.1398	10.0	25000	0.1445	6.1625
0.1967	11.0	27500	0.1302	5.5177
0.1329	12.0	30000	0.1298	5.5482
0.1778	13.0	32500	0.1227	5.3237
0.1281	14.0	35000	0.1235	5.1792
0.3553	15.0	37500	0.1238	5.2362
0.2678	16.0	40000	0.1211	5.0670
0.8916	17.0	42500	0.1431	5.3618
0.8058	18.0	45000	0.1579	5.4340

Framework versions

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