Whisper-Small-zh-squeezeformer

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

• Loss: 3.0086
• Cer: 103.4976

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: 8
• seed: 42
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• lr_scheduler_warmup_steps: 1500
• training_steps: 33000
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Cer	Validation Loss
2.9928	1.5	1500	95.9754	3.0761
2.6515	3.0	3000	99.3160	2.8994
2.6062	4.5	4500	99.7704	2.8336
2.5246	6.0	6000	100.0351	2.8107
2.7344	7.5	7500	99.8360	2.7851
2.6095	9.0	9000	103.0852	2.7785
2.6911	10.5	10500	104.3667	2.7653
2.7004	12.0	12000	102.5464	2.7534
2.617	13.5	13500	100.7145	2.7050
2.5298	15.0	15000	100.9441	2.6941
2.634	16.5	16500	100.9558	2.6628
2.5629	18.0	18000	100.5552	2.6489
2.8322	19.5	19500	102.8018	2.7431
2.6619	21.0	21000	101.6984	2.7448
2.6874	22.5	22500	104.8258	2.9346
2.5785	24.0	24000	105.6364	2.9642
2.6002	25.5	25500	105.5333	2.9768
2.5124	27.0	27000	104.0621	3.0030
2.5252	28.5	28500	102.7526	2.9799
2.5209	30.0	30000	102.1060	3.0035
2.4584	31.5	31500	2.9897	102.8978
2.4354	33.0	33000	3.0086	103.4976

Framework versions

• Transformers 4.42.0.dev0
• Pytorch 2.1.2
• Datasets 2.21.0
• Tokenizers 0.19.1