Modelpublic

jindaznb/torgo_tiny_finetune_F01_frozen_encoder

sourceHugging Faceapache-2.0updated 2y agoView on Hugging Face

0likes3downloads

Model Card

torgotinyfinetuneF01frozen_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: 0.2915
—Wer: 73.9389

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
0.7815	0.83	500	0.2625	39.3888
0.0936	1.66	1000	0.2655	29.4567
0.0711	2.49	1500	0.2517	25.4669
0.0456	3.32	2000	0.2738	28.6927
0.0327	4.15	2500	0.2770	34.8896
0.0258	4.98	3000	0.2653	20.0340
0.0181	5.8	3500	0.2902	27.0798
0.0145	6.63	4000	0.2801	22.3260
0.0114	7.46	4500	0.3174	27.0798
0.0094	8.29	5000	0.2789	47.8778
0.0072	9.12	5500	0.2827	20.7980
0.0058	9.95	6000	0.3011	23.8540
0.0046	10.78	6500	0.2892	23.0051
0.0035	11.61	7000	0.2858	20.5433
0.0034	12.44	7500	0.2876	25.2122
0.0021	13.27	8000	0.2876	23.1749
0.002	14.1	8500	0.3039	41.9355
0.0019	14.93	9000	0.3060	24.7029
0.001	15.75	9500	0.2938	30.4754
0.0009	16.58	10000	0.2998	31.3243
0.0007	17.41	10500	0.2933	37.0968
0.0005	18.24	11000	0.2937	39.7284
0.0004	19.07	11500	0.2921	69.8642
0.0002	19.9	12000	0.2915	73.9389

Framework versions

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