Moroccan-Darija-STT-large-turbo-v1.6.6

This model is a fine-tuned version of openai/whisper-large-v3-turbo on an unknown dataset. It achieves the following results on the evaluation set:

• Loss: 0.4219
• Wer: 133.9106
• Cer: 85.5666

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: 4.375e-06
• train_batch_size: 32
• eval_batch_size: 32
• seed: 42
• gradient_accumulation_steps: 4
• total_train_batch_size: 128
• optimizer: Use OptimizerNames.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: 10
• num_epochs: 6

Training results

Training Loss	Epoch	Step	Validation Loss	Wer	Cer
1.0999	0.6135	25	0.5216	181.2082	163.3465
0.9076	1.2209	50	0.4641	124.9749	79.3490
0.8655	1.8344	75	0.4492	110.5171	67.6672
0.8333	2.4417	100	0.4363	144.6369	97.0660
0.8082	3.0491	125	0.4344	140.3196	92.7554
0.7585	3.6626	150	0.4323	136.1278	89.9938
0.7443	4.2699	175	0.4214	119.7289	76.0586
0.7444	4.8834	200	0.4252	130.3715	84.1934
0.7218	5.4908	225	0.4219	133.9106	85.5666

Framework versions

• Transformers 4.48.0
• Pytorch 2.5.1+cu124
• Datasets 3.1.0
• Tokenizers 0.21.0