whisper-small-ar-12hrsdarijadata-April21-params1

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: 0.6189
• Wer: 57.5010
• Cer: 22.7156

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: 16
• eval_batch_size: 8
• seed: 42
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• lr_scheduler_warmup_steps: 300
• training_steps: 4000
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer	Cer
1.0603	0.38	250	0.8352	73.3652	36.2817
0.7041	0.75	500	0.6119	66.8656	29.7437
0.4434	1.13	750	0.5534	56.4001	24.9692
0.4344	1.51	1000	0.5341	55.8827	22.2788
0.3683	1.89	1250	0.5171	53.8931	23.3545
0.2964	2.26	1500	0.5245	53.2697	21.4447
0.2126	2.64	1750	0.5267	51.0545	19.9624
0.2095	3.02	2000	0.5428	55.1267	21.9697
0.1452	3.39	2250	0.5636	59.2386	23.4962
0.1123	3.77	2500	0.5598	54.0257	21.7816
0.077	4.15	2750	0.5809	53.5747	23.3614
0.0624	4.52	3000	0.5897	53.6941	22.1417
0.0492	4.9	3250	0.5973	54.0921	20.3922
0.0571	5.28	3500	0.6128	54.3043	21.2704
0.0637	5.66	3750	0.6139	54.6890	22.2114
0.0288	6.03	4000	0.6189	57.5010	22.7156

Framework versions

• Transformers 4.28.0
• Pytorch 2.0.0+cu118
• Datasets 2.11.1.dev0
• Tokenizers 0.13.3