Arbi-Houssem/TunLangModel1.4

Whisper Tunisien

This model is a fine-tuned version of openai/whisper-small on the datasetSTT-TTS dataset. It achieves the following results on the evaluation set:

• Loss: 2.1002
• Wer: 69.0947

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-07
• train_batch_size: 8
• eval_batch_size: 8
• seed: 42
• gradient_accumulation_steps: 2
• 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: 500
• training_steps: 7000
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
1.2207	62.5	500	2.2689	90.9469
0.8267	125.0	1000	2.0114	80.4370
0.6297	187.5	1500	1.9396	73.5692
0.5283	250.0	2000	1.9364	70.5515
0.4231	312.5	2500	1.9509	70.4475
0.3683	375.0	3000	1.9729	74.2976
0.319	437.5	3500	1.9950	73.2570
0.2884	500.0	4000	2.0182	72.6327
0.259	562.5	4500	2.0410	72.5286
0.2364	625.0	5000	2.0619	69.0947
0.2181	687.5	5500	2.0780	69.0947
0.2133	750.0	6000	2.0901	68.8866
0.201	812.5	6500	2.0979	68.8866
0.2033	875.0	7000	2.1002	69.0947

Framework versions

• Transformers 4.42.0.dev0
• Pytorch 2.3.0+cu121
• Datasets 2.19.1
• Tokenizers 0.19.1