Modelpublic

jindaznb/torgo_tiny_finetune_F04_frozen_encoder

sourceHugging Faceapache-2.0updated 2y agoView on Hugging Face

0likes4downloads

Model Card

torgotinyfinetuneF04frozen_encoder

This model is a fine-tuned version of openai/whisper-tiny on the None dataset. It achieves the following results on the evaluation set:

—Loss: 0.2948
—Wer: 46.1800

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: 0.0001
—trainbatchsize: 16
—evalbatchsize: 1
—seed: 42
—optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
—lrschedulertype: linear
—lrschedulerwarmup_steps: 1000
—num_epochs: 20

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
0.7886	0.85	500	0.2527	38.2003
0.0987	1.69	1000	0.2771	51.7827
0.0695	2.54	1500	0.2463	38.6248
0.0479	3.39	2000	0.2699	26.8251
0.0314	4.24	2500	0.2857	23.2598
0.0239	5.08	3000	0.2698	23.6842
0.0173	5.93	3500	0.2771	25.2122
0.0122	6.78	4000	0.2733	26.7402
0.0099	7.63	4500	0.2812	26.5705
0.0091	8.47	5000	0.2773	23.4295
0.0077	9.32	5500	0.2839	30.5603
0.0057	10.17	6000	0.2722	23.7691
0.0043	11.02	6500	0.2959	34.3803
0.0028	11.86	7000	0.2783	33.0221
0.0026	12.71	7500	0.3000	32.7674
0.0025	13.56	8000	0.2865	32.6825
0.0022	14.41	8500	0.2946	38.8795
0.0014	15.25	9000	0.2858	38.3701
0.0012	16.1	9500	0.2953	63.8370
0.0006	16.95	10000	0.2928	42.9542
0.0004	17.8	10500	0.2910	43.7182
0.0004	18.64	11000	0.2947	44.8217
0.0002	19.49	11500	0.2948	46.1800

Framework versions

—Transformers 4.32.0
—Pytorch 2.1.0+cu121
—Datasets 2.14.7
—Tokenizers 0.13.3