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: 0.1185
• Wer: 5.9055
• Cer: 83.2386
• Decode Time: 0.5299
• Wer Time: 0.0047
• Cer Time: 0.0030

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
2.4079	0.1	10	1.9885	312.2047	119.2472	0.5334	0.0169	0.0041
1.2303	1.01	20	1.1646	258.2677	100.0	0.5213	1.4057	0.0046
0.8667	1.11	30	0.8100	37.7953	52.3438	0.5008	0.0396	0.0045
0.4517	2.02	40	0.6337	40.9449	73.7926	0.5137	0.0217	0.0030
0.4352	2.12	50	0.4493	16.5354	88.1392	0.5203	0.0054	0.0032
0.2341	3.03	60	0.2922	8.2677	97.5852	0.5434	0.0060	0.0032
0.2233	3.13	70	0.2026	9.0551	83.5227	0.5359	0.0063	0.0033
0.1098	4.04	80	0.1665	5.9055	83.9489	0.5316	0.0056	0.0029
0.0678	4.14	90	0.1279	7.0866	81.1080	0.5388	0.0079	0.0038
0.078	5.05	100	0.1185	5.9055	83.2386	0.5299	0.0047	0.0030

Framework versions

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