Modelpublic

chengyili2005/whisper-medium-english-spanish-lora

sourceHugging Faceapache-2.0updated 7mo agoView on Hugging Face

0likes7downloads

Model Card

Whisper Medium English-Spanish - Chengyi Li

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

—Loss: 0.9114
—Wer: 54.8305

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: 5e-05
—trainbatchsize: 4
—evalbatchsize: 4
—seed: 42
—gradientaccumulationsteps: 4
—totaltrainbatch_size: 16
—optimizer: Use OptimizerNames.ADAMWTORCH with betas=(0.9,0.999) and epsilon=1e-08 and optimizerargs=No additional optimizer arguments
—lrschedulertype: linear
—lrschedulerwarmup_steps: 500
—training_steps: 20000

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
1.3085	1.7688	500	1.2604	68.9315
1.016	3.5350	1000	1.0430	80.9213
0.9305	5.3012	1500	0.9117	73.3717
0.8359	7.0673	2000	0.8675	68.3941
0.7932	8.8361	2500	0.8538	60.2687
0.7181	10.6023	3000	0.8497	59.5138
0.7049	12.3685	3500	0.8520	59.6929
0.638	14.1346	4000	0.8520	60.5502
0.6844	15.9035	4500	0.8531	55.3167
0.6867	17.6696	5000	0.8622	55.7901
0.6535	19.4358	5500	0.8628	59.1043
0.6441	21.2019	6000	0.8786	54.3442
0.6291	22.9708	6500	0.8818	51.7210
0.5844	24.7369	7000	0.8897	53.7044
0.5489	26.5031	7500	0.9018	55.3807
0.525	28.2693	8000	0.9114	54.8305

Framework versions

—PEFT 0.18.1
—Transformers 4.52.0
—Pytorch 2.9.1+cu128
—Datasets 4.5.0
—Tokenizers 0.21.4