JacobLinCool/wft-test-model-merged

wft-test-model

This model is a fine-tuned version of openai/whisper-tiny on the hf-internal-testing/librispeech_asr_dummy dataset. It achieves the following results on the evaluation set:

• Loss: 0.1248
• Wer: 4.7244
• Cer: 92.6847
• Decode Time: 0.5481
• Wer Time: 0.0069
• Cer Time: 0.0040

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: 0.0005
• train_batch_size: 4
• 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: 50
• training_steps: 100

Training results

Training Loss	Epoch	Step	Validation Loss	Wer	Cer	Decode Time	Wer Time	Cer Time
2.4107	0.1	10	1.9892	303.5433	117.1875	0.5449	0.0307	0.0039
1.2109	1.01	20	1.1659	155.1181	91.2642	0.5278	0.0062	0.0036
0.8855	1.11	30	0.8104	30.7087	56.8182	0.4832	0.0069	0.0041
0.4367	2.02	40	0.6315	25.1969	74.5739	0.5295	0.0058	0.0034
0.4398	2.12	50	0.4566	17.3228	91.9744	0.6078	0.0055	0.0030
0.2291	3.03	60	0.3006	9.0551	100.7102	0.5659	0.0058	0.0031
0.2281	3.13	70	0.2144	7.4803	90.4830	0.5507	0.0046	0.0030
0.111	4.04	80	0.1736	5.9055	89.3466	0.6595	0.0063	0.0032
0.0695	4.14	90	0.1345	4.7244	87.9261	0.6369	0.0402	0.0182
0.0761	5.05	100	0.1248	4.7244	92.6847	0.5481	0.0069	0.0040

Framework versions

• PEFT 0.13.2
• Transformers 4.45.2
• Pytorch 2.5.0
• Datasets 3.0.2
• Tokenizers 0.20.1