whisper-small-toi

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

• Loss: 3.1668
• Wer: 63.5938

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: 1e-05
• train_batch_size: 4
• eval_batch_size: 4
• seed: 42
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• lr_scheduler_warmup_steps: 500
• training_steps: 10000
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
0.568	1.47	500	2.1883	72.0402
0.2614	2.95	1000	2.1071	67.1034
0.0811	4.42	1500	2.3456	67.5012
0.0383	5.9	2000	2.4961	67.9691
0.021	7.37	2500	2.6259	68.8348
0.0077	8.85	3000	2.6423	66.6823
0.0046	10.32	3500	2.8497	65.9336
0.0005	11.8	4000	2.8305	64.6467
0.0014	13.27	4500	2.9174	66.0739
0.0003	14.75	5000	2.9358	63.2663
0.0002	16.22	5500	2.9820	63.8278
0.0002	17.7	6000	3.0369	64.7403
0.0001	19.17	6500	3.0641	63.3832
0.0005	20.65	7000	3.0512	63.1493
0.0001	22.12	7500	3.0924	63.5002
0.0001	23.6	8000	3.1215	65.0679
0.0001	25.07	8500	3.1336	64.6233
0.0001	26.55	9000	3.1513	63.7108
0.0001	28.02	9500	3.1620	63.5938
0.0001	29.5	10000	3.1668	63.5938

Framework versions

• Transformers 4.26.0.dev0
• Pytorch 1.12.1+cu113
• Datasets 2.7.1
• Tokenizers 0.13.2