Modelpublic

jindaznb/torgo_tiny_finetune_M02_frozen_encoder

sourceHugging Faceapache-2.0updated 2y agoView on Hugging Face

0likes5downloads

Model Card

torgotinyfinetuneM02frozen_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: 0.2969
—Wer: 44.9915

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
0.7661	0.85	500	0.2672	64.6859
0.0893	1.7	1000	0.2523	24.2784
0.0664	2.55	1500	0.2562	20.3735
0.0439	3.4	2000	0.2674	98.8115
0.0303	4.25	2500	0.2566	22.1562
0.0224	5.1	3000	0.2737	24.7878
0.0164	5.95	3500	0.2761	41.3413
0.0139	6.8	4000	0.2923	31.3243
0.0102	7.65	4500	0.2841	45.5008
0.0082	8.5	5000	0.2913	36.5874
0.0058	9.35	5500	0.3038	22.2411
0.0065	10.2	6000	0.2853	22.6655
0.0052	11.05	6500	0.2806	22.4958
0.0033	11.9	7000	0.2866	30.8149
0.0026	12.76	7500	0.2852	24.3633
0.0027	13.61	8000	0.2956	54.4992
0.0017	14.46	8500	0.2959	31.0696
0.0012	15.31	9000	0.2974	35.9932
0.0012	16.16	9500	0.2993	39.5586
0.0008	17.01	10000	0.2950	44.1426
0.0004	17.86	10500	0.2988	47.0289
0.0002	18.71	11000	0.2948	44.2275
0.0002	19.56	11500	0.2969	44.9915

Framework versions

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