jun-han/Whisper-squeezeformer-v2

Whisper-squeezeformer-V9-mutliconv

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: 0.1799
• Wer: 11.3645

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
4.7863	1.0	2500	3.8844	119.2635
3.9844	2.0	5000	3.7326	132.0431
3.459	3.0	7500	1.2427	72.6510
0.4598	4.0	10000	0.4232	23.4841
0.1931	5.0	12500	0.3479	20.3629
0.1096	6.0	15000	0.3255	17.5574
0.0631	7.0	17500	0.3251	16.8252
0.2473	8.0	20000	0.2452	14.4781
0.1433	9.0	22500	0.2323	13.0782
0.3067	10.0	25000	0.2224	14.5675
0.2769	11.0	27500	0.2015	13.2323
0.1853	12.0	30000	0.2004	14.1015
0.1308	13.0	32500	0.2011	13.2494
0.2435	14.0	35000	0.1830	11.9066
0.1786	15.0	37500	0.1804	12.7948
0.4194	16.0	40000	0.1833	10.3469
0.3283	17.0	42500	0.1818	10.2785
0.2936	18.0	45000	0.1799	11.3645

Framework versions

• Transformers 4.45.2
• Pytorch 2.4.0
• Datasets 3.2.0
• Tokenizers 0.20.3