Mohamedshaaban2001/MSDC-whisper-large-v3-55

Whisper whisper-large-v3 ar1 - Mohamed Shaaban

This model is a fine-tuned version of openai/whisper-large-v3 on the Common standard ar Voice 11.0 dataset. It achieves the following results on the evaluation set:

• Loss: 0.1983
• Wer: 50.0

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
• gradient_accumulation_steps: 4
• total_train_batch_size: 16
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• lr_scheduler_warmup_steps: 1
• num_epochs: 20

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
0.6276	1.0	1	1.5308	100.0
0.6286	2.0	2	0.5920	0.0
0.2312	3.0	3	0.1197	0.0
0.0463	4.0	4	0.0939	0.0
0.02	5.0	5	0.0918	50.0
0.0112	6.0	6	0.0955	50.0
0.0046	7.0	7	0.1133	50.0
0.0022	8.0	8	0.1343	50.0
0.0011	9.0	9	0.1518	50.0
0.0005	10.0	10	0.1655	50.0
0.0003	11.0	11	0.1758	50.0
0.0002	12.0	12	0.1835	50.0
0.0002	13.0	13	0.1890	50.0
0.0001	14.0	14	0.1929	50.0
0.0001	15.0	15	0.1954	50.0
0.0001	16.0	16	0.1970	50.0
0.0001	17.0	17	0.1978	50.0
0.0001	18.0	18	0.1982	50.0
0.0001	19.0	19	0.1983	50.0
0.0001	20.0	20	0.1983	50.0

Framework versions

• Transformers 4.39.3
• Pytorch 2.2.2
• Datasets 2.18.0
• Tokenizers 0.15.2