README.md

metadata

library_name: transformers
language:
  - ar
license: apache-2.0
base_model: openai/whisper-tiny
tags:
  - generated_from_trainer
metrics:
  - wer
model-index:
  - name: Whisper tiny AR - BH
    results: []

Whisper tiny AR - BH

This model is a fine-tuned version of openai/whisper-tiny on the quran-ayat-speech-to-text dataset. It achieves the following results on the evaluation set:

• Loss: 0.0244
• Wer: 0.1485
• Cer: 0.0491

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
• train_batch_size: 16
• eval_batch_size: 16
• seed: 42
• gradient_accumulation_steps: 4
• total_train_batch_size: 64
• optimizer: Use adamw_torch with betas=(0.9,0.999) and epsilon=1e-08 and optimizer_args=No additional optimizer arguments
• lr_scheduler_type: linear
• lr_scheduler_warmup_steps: 500
• num_epochs: 20
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer	Cer
0.0508	1.0	157	0.0447	2.1609	0.9033
0.0294	2.0	314	0.0286	1.9660	1.0025
0.0226	3.0	471	0.0254	0.4391	0.1838
0.0147	4.0	628	0.0224	0.3542	0.1358
0.0129	5.0	785	0.0213	0.5243	0.2350
0.0091	6.0	942	0.0199	0.3001	0.1045
0.0065	7.0	1099	0.0196	0.2268	0.0728
0.0043	8.0	1256	0.0196	0.2011	0.0645
0.003	9.0	1413	0.0201	0.2885	0.1164
0.003	10.0	1570	0.0209	0.3251	0.1150
0.0016	11.0	1727	0.0207	0.1828	0.0607
0.0009	12.0	1884	0.0209	0.1747	0.0537
0.0007	13.0	2041	0.0211	0.1680	0.0517
0.0005	14.0	2198	0.0218	0.1652	0.0504
0.0003	15.0	2355	0.0211	0.1580	0.0496
0.0002	16.0	2512	0.0216	0.1566	0.0487
0.0001	17.0	2669	0.0224	0.1562	0.0492
0.0	18.0	2826	0.0227	0.1481	0.0454
0.0	19.0	2983	0.0239	0.1471	0.0489
0.0	19.8768	3120	0.0237	0.1461	0.0462

Framework versions

• Transformers 4.47.0
• Pytorch 2.5.1+cu121
• Datasets 3.2.0
• Tokenizers 0.21.0