Marialab/finetuned-s2t-1000-step

Finetuned-facebook-s2t-for-darija-speech-translation

This model is a fine-tuned version of facebook/s2t-medium-mustc-multilingual-st on the Darija-C dataset. It achieves the following results on the evaluation set:

• Loss: 5.7855
• Bleu: 0.0032

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: 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: 100
• training_steps: 1000
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Bleu
9.1689	12.5	50	8.4431	0.0
7.9984	25.0	100	7.6555	0.0
7.4717	37.5	150	7.2774	0.0
7.2484	50.0	200	7.1061	0.0
7.0982	62.5	250	6.9703	0.0
6.9724	75.0	300	6.8526	0.0011
6.8564	87.5	350	6.7225	0.0034
6.7332	100.0	400	6.6144	0.0034
6.6511	112.5	450	6.5264	0.0034
6.5283	125.0	500	6.4174	0.0034
6.4477	137.5	550	6.3187	0.0034
6.3455	150.0	600	6.2208	0.0031
6.2683	162.5	650	6.0831	0.0034
6.1757	175.0	700	6.0449	0.0032
6.1017	187.5	750	5.9507	0.0034
6.0438	200.0	800	5.8899	0.0032
5.9752	212.5	850	5.8447	0.0034
5.9657	225.0	900	5.8105	0.0032
5.925	237.5	950	5.7858	0.0032
5.9142	250.0	1000	5.7855	0.0032

Framework versions

• Transformers 4.47.1
• Pytorch 2.5.1+cu121
• Datasets 2.19.2
• Tokenizers 0.21.0